1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
6 * Swap reorganised 29.12.95, Stephen Tweedie
10 #include <linux/sched/mm.h>
11 #include <linux/sched/task.h>
12 #include <linux/hugetlb.h>
13 #include <linux/mman.h>
14 #include <linux/slab.h>
15 #include <linux/kernel_stat.h>
16 #include <linux/swap.h>
17 #include <linux/vmalloc.h>
18 #include <linux/pagemap.h>
19 #include <linux/namei.h>
20 #include <linux/shmem_fs.h>
21 #include <linux/blkdev.h>
22 #include <linux/random.h>
23 #include <linux/writeback.h>
24 #include <linux/proc_fs.h>
25 #include <linux/seq_file.h>
26 #include <linux/init.h>
27 #include <linux/ksm.h>
28 #include <linux/rmap.h>
29 #include <linux/security.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mutex.h>
32 #include <linux/capability.h>
33 #include <linux/syscalls.h>
34 #include <linux/memcontrol.h>
35 #include <linux/poll.h>
36 #include <linux/oom.h>
37 #include <linux/frontswap.h>
38 #include <linux/swapfile.h>
39 #include <linux/export.h>
40 #include <linux/swap_slots.h>
41 #include <linux/sort.h>
43 #include <asm/pgtable.h>
44 #include <asm/tlbflush.h>
45 #include <linux/swapops.h>
46 #include <linux/swap_cgroup.h>
48 static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
50 static void free_swap_count_continuations(struct swap_info_struct *);
51 static sector_t map_swap_entry(swp_entry_t, struct block_device**);
53 DEFINE_SPINLOCK(swap_lock);
54 static unsigned int nr_swapfiles;
55 atomic_long_t nr_swap_pages;
57 * Some modules use swappable objects and may try to swap them out under
58 * memory pressure (via the shrinker). Before doing so, they may wish to
59 * check to see if any swap space is available.
61 EXPORT_SYMBOL_GPL(nr_swap_pages);
62 /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
63 long total_swap_pages;
64 static int least_priority = -1;
66 static const char Bad_file[] = "Bad swap file entry ";
67 static const char Unused_file[] = "Unused swap file entry ";
68 static const char Bad_offset[] = "Bad swap offset entry ";
69 static const char Unused_offset[] = "Unused swap offset entry ";
72 * all active swap_info_structs
73 * protected with swap_lock, and ordered by priority.
75 PLIST_HEAD(swap_active_head);
78 * all available (active, not full) swap_info_structs
79 * protected with swap_avail_lock, ordered by priority.
80 * This is used by get_swap_page() instead of swap_active_head
81 * because swap_active_head includes all swap_info_structs,
82 * but get_swap_page() doesn't need to look at full ones.
83 * This uses its own lock instead of swap_lock because when a
84 * swap_info_struct changes between not-full/full, it needs to
85 * add/remove itself to/from this list, but the swap_info_struct->lock
86 * is held and the locking order requires swap_lock to be taken
87 * before any swap_info_struct->lock.
89 static struct plist_head *swap_avail_heads;
90 static DEFINE_SPINLOCK(swap_avail_lock);
92 struct swap_info_struct *swap_info[MAX_SWAPFILES];
94 static DEFINE_MUTEX(swapon_mutex);
96 static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
97 /* Activity counter to indicate that a swapon or swapoff has occurred */
98 static atomic_t proc_poll_event = ATOMIC_INIT(0);
100 atomic_t nr_rotate_swap = ATOMIC_INIT(0);
102 static struct swap_info_struct *swap_type_to_swap_info(int type)
104 if (type >= READ_ONCE(nr_swapfiles))
107 smp_rmb(); /* Pairs with smp_wmb in alloc_swap_info. */
108 return READ_ONCE(swap_info[type]);
111 static inline unsigned char swap_count(unsigned char ent)
113 return ent & ~SWAP_HAS_CACHE; /* may include COUNT_CONTINUED flag */
116 /* Reclaim the swap entry anyway if possible */
117 #define TTRS_ANYWAY 0x1
119 * Reclaim the swap entry if there are no more mappings of the
122 #define TTRS_UNMAPPED 0x2
123 /* Reclaim the swap entry if swap is getting full*/
124 #define TTRS_FULL 0x4
126 /* returns 1 if swap entry is freed */
127 static int __try_to_reclaim_swap(struct swap_info_struct *si,
128 unsigned long offset, unsigned long flags)
130 swp_entry_t entry = swp_entry(si->type, offset);
134 page = find_get_page(swap_address_space(entry), offset);
138 * When this function is called from scan_swap_map_slots() and it's
139 * called by vmscan.c at reclaiming pages. So, we hold a lock on a page,
140 * here. We have to use trylock for avoiding deadlock. This is a special
141 * case and you should use try_to_free_swap() with explicit lock_page()
142 * in usual operations.
144 if (trylock_page(page)) {
145 if ((flags & TTRS_ANYWAY) ||
146 ((flags & TTRS_UNMAPPED) && !page_mapped(page)) ||
147 ((flags & TTRS_FULL) && mem_cgroup_swap_full(page)))
148 ret = try_to_free_swap(page);
155 static inline struct swap_extent *first_se(struct swap_info_struct *sis)
157 struct rb_node *rb = rb_first(&sis->swap_extent_root);
158 return rb_entry(rb, struct swap_extent, rb_node);
161 static inline struct swap_extent *next_se(struct swap_extent *se)
163 struct rb_node *rb = rb_next(&se->rb_node);
164 return rb ? rb_entry(rb, struct swap_extent, rb_node) : NULL;
168 * swapon tell device that all the old swap contents can be discarded,
169 * to allow the swap device to optimize its wear-levelling.
171 static int discard_swap(struct swap_info_struct *si)
173 struct swap_extent *se;
174 sector_t start_block;
178 /* Do not discard the swap header page! */
180 start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
181 nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
183 err = blkdev_issue_discard(si->bdev, start_block,
184 nr_blocks, GFP_KERNEL, 0);
190 for (se = next_se(se); se; se = next_se(se)) {
191 start_block = se->start_block << (PAGE_SHIFT - 9);
192 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
194 err = blkdev_issue_discard(si->bdev, start_block,
195 nr_blocks, GFP_KERNEL, 0);
201 return err; /* That will often be -EOPNOTSUPP */
204 static struct swap_extent *
205 offset_to_swap_extent(struct swap_info_struct *sis, unsigned long offset)
207 struct swap_extent *se;
210 rb = sis->swap_extent_root.rb_node;
212 se = rb_entry(rb, struct swap_extent, rb_node);
213 if (offset < se->start_page)
215 else if (offset >= se->start_page + se->nr_pages)
220 /* It *must* be present */
225 * swap allocation tell device that a cluster of swap can now be discarded,
226 * to allow the swap device to optimize its wear-levelling.
228 static void discard_swap_cluster(struct swap_info_struct *si,
229 pgoff_t start_page, pgoff_t nr_pages)
231 struct swap_extent *se = offset_to_swap_extent(si, start_page);
234 pgoff_t offset = start_page - se->start_page;
235 sector_t start_block = se->start_block + offset;
236 sector_t nr_blocks = se->nr_pages - offset;
238 if (nr_blocks > nr_pages)
239 nr_blocks = nr_pages;
240 start_page += nr_blocks;
241 nr_pages -= nr_blocks;
243 start_block <<= PAGE_SHIFT - 9;
244 nr_blocks <<= PAGE_SHIFT - 9;
245 if (blkdev_issue_discard(si->bdev, start_block,
246 nr_blocks, GFP_NOIO, 0))
253 #ifdef CONFIG_THP_SWAP
254 #define SWAPFILE_CLUSTER HPAGE_PMD_NR
256 #define swap_entry_size(size) (size)
258 #define SWAPFILE_CLUSTER 256
261 * Define swap_entry_size() as constant to let compiler to optimize
262 * out some code if !CONFIG_THP_SWAP
264 #define swap_entry_size(size) 1
266 #define LATENCY_LIMIT 256
268 static inline void cluster_set_flag(struct swap_cluster_info *info,
274 static inline unsigned int cluster_count(struct swap_cluster_info *info)
279 static inline void cluster_set_count(struct swap_cluster_info *info,
285 static inline void cluster_set_count_flag(struct swap_cluster_info *info,
286 unsigned int c, unsigned int f)
292 static inline unsigned int cluster_next(struct swap_cluster_info *info)
297 static inline void cluster_set_next(struct swap_cluster_info *info,
303 static inline void cluster_set_next_flag(struct swap_cluster_info *info,
304 unsigned int n, unsigned int f)
310 static inline bool cluster_is_free(struct swap_cluster_info *info)
312 return info->flags & CLUSTER_FLAG_FREE;
315 static inline bool cluster_is_null(struct swap_cluster_info *info)
317 return info->flags & CLUSTER_FLAG_NEXT_NULL;
320 static inline void cluster_set_null(struct swap_cluster_info *info)
322 info->flags = CLUSTER_FLAG_NEXT_NULL;
326 static inline bool cluster_is_huge(struct swap_cluster_info *info)
328 if (IS_ENABLED(CONFIG_THP_SWAP))
329 return info->flags & CLUSTER_FLAG_HUGE;
333 static inline void cluster_clear_huge(struct swap_cluster_info *info)
335 info->flags &= ~CLUSTER_FLAG_HUGE;
338 static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si,
339 unsigned long offset)
341 struct swap_cluster_info *ci;
343 ci = si->cluster_info;
345 ci += offset / SWAPFILE_CLUSTER;
346 spin_lock(&ci->lock);
351 static inline void unlock_cluster(struct swap_cluster_info *ci)
354 spin_unlock(&ci->lock);
358 * Determine the locking method in use for this device. Return
359 * swap_cluster_info if SSD-style cluster-based locking is in place.
361 static inline struct swap_cluster_info *lock_cluster_or_swap_info(
362 struct swap_info_struct *si, unsigned long offset)
364 struct swap_cluster_info *ci;
366 /* Try to use fine-grained SSD-style locking if available: */
367 ci = lock_cluster(si, offset);
368 /* Otherwise, fall back to traditional, coarse locking: */
370 spin_lock(&si->lock);
375 static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si,
376 struct swap_cluster_info *ci)
381 spin_unlock(&si->lock);
384 static inline bool cluster_list_empty(struct swap_cluster_list *list)
386 return cluster_is_null(&list->head);
389 static inline unsigned int cluster_list_first(struct swap_cluster_list *list)
391 return cluster_next(&list->head);
394 static void cluster_list_init(struct swap_cluster_list *list)
396 cluster_set_null(&list->head);
397 cluster_set_null(&list->tail);
400 static void cluster_list_add_tail(struct swap_cluster_list *list,
401 struct swap_cluster_info *ci,
404 if (cluster_list_empty(list)) {
405 cluster_set_next_flag(&list->head, idx, 0);
406 cluster_set_next_flag(&list->tail, idx, 0);
408 struct swap_cluster_info *ci_tail;
409 unsigned int tail = cluster_next(&list->tail);
412 * Nested cluster lock, but both cluster locks are
413 * only acquired when we held swap_info_struct->lock
416 spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING);
417 cluster_set_next(ci_tail, idx);
418 spin_unlock(&ci_tail->lock);
419 cluster_set_next_flag(&list->tail, idx, 0);
423 static unsigned int cluster_list_del_first(struct swap_cluster_list *list,
424 struct swap_cluster_info *ci)
428 idx = cluster_next(&list->head);
429 if (cluster_next(&list->tail) == idx) {
430 cluster_set_null(&list->head);
431 cluster_set_null(&list->tail);
433 cluster_set_next_flag(&list->head,
434 cluster_next(&ci[idx]), 0);
439 /* Add a cluster to discard list and schedule it to do discard */
440 static void swap_cluster_schedule_discard(struct swap_info_struct *si,
444 * If scan_swap_map() can't find a free cluster, it will check
445 * si->swap_map directly. To make sure the discarding cluster isn't
446 * taken by scan_swap_map(), mark the swap entries bad (occupied). It
447 * will be cleared after discard
449 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
450 SWAP_MAP_BAD, SWAPFILE_CLUSTER);
452 cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx);
454 schedule_work(&si->discard_work);
457 static void __free_cluster(struct swap_info_struct *si, unsigned long idx)
459 struct swap_cluster_info *ci = si->cluster_info;
461 cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE);
462 cluster_list_add_tail(&si->free_clusters, ci, idx);
466 * Doing discard actually. After a cluster discard is finished, the cluster
467 * will be added to free cluster list. caller should hold si->lock.
469 static void swap_do_scheduled_discard(struct swap_info_struct *si)
471 struct swap_cluster_info *info, *ci;
474 info = si->cluster_info;
476 while (!cluster_list_empty(&si->discard_clusters)) {
477 idx = cluster_list_del_first(&si->discard_clusters, info);
478 spin_unlock(&si->lock);
480 discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
483 spin_lock(&si->lock);
484 ci = lock_cluster(si, idx * SWAPFILE_CLUSTER);
485 __free_cluster(si, idx);
486 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
487 0, SWAPFILE_CLUSTER);
492 static void swap_discard_work(struct work_struct *work)
494 struct swap_info_struct *si;
496 si = container_of(work, struct swap_info_struct, discard_work);
498 spin_lock(&si->lock);
499 swap_do_scheduled_discard(si);
500 spin_unlock(&si->lock);
503 static void alloc_cluster(struct swap_info_struct *si, unsigned long idx)
505 struct swap_cluster_info *ci = si->cluster_info;
507 VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx);
508 cluster_list_del_first(&si->free_clusters, ci);
509 cluster_set_count_flag(ci + idx, 0, 0);
512 static void free_cluster(struct swap_info_struct *si, unsigned long idx)
514 struct swap_cluster_info *ci = si->cluster_info + idx;
516 VM_BUG_ON(cluster_count(ci) != 0);
518 * If the swap is discardable, prepare discard the cluster
519 * instead of free it immediately. The cluster will be freed
522 if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
523 (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
524 swap_cluster_schedule_discard(si, idx);
528 __free_cluster(si, idx);
532 * The cluster corresponding to page_nr will be used. The cluster will be
533 * removed from free cluster list and its usage counter will be increased.
535 static void inc_cluster_info_page(struct swap_info_struct *p,
536 struct swap_cluster_info *cluster_info, unsigned long page_nr)
538 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
542 if (cluster_is_free(&cluster_info[idx]))
543 alloc_cluster(p, idx);
545 VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
546 cluster_set_count(&cluster_info[idx],
547 cluster_count(&cluster_info[idx]) + 1);
551 * The cluster corresponding to page_nr decreases one usage. If the usage
552 * counter becomes 0, which means no page in the cluster is in using, we can
553 * optionally discard the cluster and add it to free cluster list.
555 static void dec_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;
563 VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
564 cluster_set_count(&cluster_info[idx],
565 cluster_count(&cluster_info[idx]) - 1);
567 if (cluster_count(&cluster_info[idx]) == 0)
568 free_cluster(p, idx);
572 * It's possible scan_swap_map() uses a free cluster in the middle of free
573 * cluster list. Avoiding such abuse to avoid list corruption.
576 scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
577 unsigned long offset)
579 struct percpu_cluster *percpu_cluster;
582 offset /= SWAPFILE_CLUSTER;
583 conflict = !cluster_list_empty(&si->free_clusters) &&
584 offset != cluster_list_first(&si->free_clusters) &&
585 cluster_is_free(&si->cluster_info[offset]);
590 percpu_cluster = this_cpu_ptr(si->percpu_cluster);
591 cluster_set_null(&percpu_cluster->index);
596 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
597 * might involve allocating a new cluster for current CPU too.
599 static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
600 unsigned long *offset, unsigned long *scan_base)
602 struct percpu_cluster *cluster;
603 struct swap_cluster_info *ci;
605 unsigned long tmp, max;
608 cluster = this_cpu_ptr(si->percpu_cluster);
609 if (cluster_is_null(&cluster->index)) {
610 if (!cluster_list_empty(&si->free_clusters)) {
611 cluster->index = si->free_clusters.head;
612 cluster->next = cluster_next(&cluster->index) *
614 } else if (!cluster_list_empty(&si->discard_clusters)) {
616 * we don't have free cluster but have some clusters in
617 * discarding, do discard now and reclaim them
619 swap_do_scheduled_discard(si);
620 *scan_base = *offset = si->cluster_next;
629 * Other CPUs can use our cluster if they can't find a free cluster,
630 * check if there is still free entry in the cluster
633 max = min_t(unsigned long, si->max,
634 (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER);
636 cluster_set_null(&cluster->index);
639 ci = lock_cluster(si, tmp);
641 if (!si->swap_map[tmp]) {
649 cluster_set_null(&cluster->index);
652 cluster->next = tmp + 1;
658 static void __del_from_avail_list(struct swap_info_struct *p)
663 plist_del(&p->avail_lists[nid], &swap_avail_heads[nid]);
666 static void del_from_avail_list(struct swap_info_struct *p)
668 spin_lock(&swap_avail_lock);
669 __del_from_avail_list(p);
670 spin_unlock(&swap_avail_lock);
673 static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset,
674 unsigned int nr_entries)
676 unsigned int end = offset + nr_entries - 1;
678 if (offset == si->lowest_bit)
679 si->lowest_bit += nr_entries;
680 if (end == si->highest_bit)
681 si->highest_bit -= nr_entries;
682 si->inuse_pages += nr_entries;
683 if (si->inuse_pages == si->pages) {
684 si->lowest_bit = si->max;
686 del_from_avail_list(si);
690 static void add_to_avail_list(struct swap_info_struct *p)
694 spin_lock(&swap_avail_lock);
696 WARN_ON(!plist_node_empty(&p->avail_lists[nid]));
697 plist_add(&p->avail_lists[nid], &swap_avail_heads[nid]);
699 spin_unlock(&swap_avail_lock);
702 static void swap_range_free(struct swap_info_struct *si, unsigned long offset,
703 unsigned int nr_entries)
705 unsigned long end = offset + nr_entries - 1;
706 void (*swap_slot_free_notify)(struct block_device *, unsigned long);
708 if (offset < si->lowest_bit)
709 si->lowest_bit = offset;
710 if (end > si->highest_bit) {
711 bool was_full = !si->highest_bit;
713 si->highest_bit = end;
714 if (was_full && (si->flags & SWP_WRITEOK))
715 add_to_avail_list(si);
717 atomic_long_add(nr_entries, &nr_swap_pages);
718 si->inuse_pages -= nr_entries;
719 if (si->flags & SWP_BLKDEV)
720 swap_slot_free_notify =
721 si->bdev->bd_disk->fops->swap_slot_free_notify;
723 swap_slot_free_notify = NULL;
724 while (offset <= end) {
725 frontswap_invalidate_page(si->type, offset);
726 if (swap_slot_free_notify)
727 swap_slot_free_notify(si->bdev, offset);
732 static int scan_swap_map_slots(struct swap_info_struct *si,
733 unsigned char usage, int nr,
736 struct swap_cluster_info *ci;
737 unsigned long offset;
738 unsigned long scan_base;
739 unsigned long last_in_cluster = 0;
740 int latency_ration = LATENCY_LIMIT;
747 * We try to cluster swap pages by allocating them sequentially
748 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
749 * way, however, we resort to first-free allocation, starting
750 * a new cluster. This prevents us from scattering swap pages
751 * all over the entire swap partition, so that we reduce
752 * overall disk seek times between swap pages. -- sct
753 * But we do now try to find an empty cluster. -Andrea
754 * And we let swap pages go all over an SSD partition. Hugh
757 si->flags += SWP_SCANNING;
758 scan_base = offset = si->cluster_next;
761 if (si->cluster_info) {
762 if (!scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
764 } else if (unlikely(!si->cluster_nr--)) {
765 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
766 si->cluster_nr = SWAPFILE_CLUSTER - 1;
770 spin_unlock(&si->lock);
773 * If seek is expensive, start searching for new cluster from
774 * start of partition, to minimize the span of allocated swap.
775 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
776 * case, just handled by scan_swap_map_try_ssd_cluster() above.
778 scan_base = offset = si->lowest_bit;
779 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
781 /* Locate the first empty (unaligned) cluster */
782 for (; last_in_cluster <= si->highest_bit; offset++) {
783 if (si->swap_map[offset])
784 last_in_cluster = offset + SWAPFILE_CLUSTER;
785 else if (offset == last_in_cluster) {
786 spin_lock(&si->lock);
787 offset -= SWAPFILE_CLUSTER - 1;
788 si->cluster_next = offset;
789 si->cluster_nr = SWAPFILE_CLUSTER - 1;
792 if (unlikely(--latency_ration < 0)) {
794 latency_ration = LATENCY_LIMIT;
799 spin_lock(&si->lock);
800 si->cluster_nr = SWAPFILE_CLUSTER - 1;
804 if (si->cluster_info) {
805 while (scan_swap_map_ssd_cluster_conflict(si, offset)) {
806 /* take a break if we already got some slots */
809 if (!scan_swap_map_try_ssd_cluster(si, &offset,
814 if (!(si->flags & SWP_WRITEOK))
816 if (!si->highest_bit)
818 if (offset > si->highest_bit)
819 scan_base = offset = si->lowest_bit;
821 ci = lock_cluster(si, offset);
822 /* reuse swap entry of cache-only swap if not busy. */
823 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
826 spin_unlock(&si->lock);
827 swap_was_freed = __try_to_reclaim_swap(si, offset, TTRS_ANYWAY);
828 spin_lock(&si->lock);
829 /* entry was freed successfully, try to use this again */
832 goto scan; /* check next one */
835 if (si->swap_map[offset]) {
842 si->swap_map[offset] = usage;
843 inc_cluster_info_page(si, si->cluster_info, offset);
846 swap_range_alloc(si, offset, 1);
847 si->cluster_next = offset + 1;
848 slots[n_ret++] = swp_entry(si->type, offset);
850 /* got enough slots or reach max slots? */
851 if ((n_ret == nr) || (offset >= si->highest_bit))
854 /* search for next available slot */
856 /* time to take a break? */
857 if (unlikely(--latency_ration < 0)) {
860 spin_unlock(&si->lock);
862 spin_lock(&si->lock);
863 latency_ration = LATENCY_LIMIT;
866 /* try to get more slots in cluster */
867 if (si->cluster_info) {
868 if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
870 } else if (si->cluster_nr && !si->swap_map[++offset]) {
871 /* non-ssd case, still more slots in cluster? */
877 si->flags -= SWP_SCANNING;
881 spin_unlock(&si->lock);
882 while (++offset <= si->highest_bit) {
883 if (!si->swap_map[offset]) {
884 spin_lock(&si->lock);
887 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
888 spin_lock(&si->lock);
891 if (unlikely(--latency_ration < 0)) {
893 latency_ration = LATENCY_LIMIT;
896 offset = si->lowest_bit;
897 while (offset < scan_base) {
898 if (!si->swap_map[offset]) {
899 spin_lock(&si->lock);
902 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
903 spin_lock(&si->lock);
906 if (unlikely(--latency_ration < 0)) {
908 latency_ration = LATENCY_LIMIT;
912 spin_lock(&si->lock);
915 si->flags -= SWP_SCANNING;
919 static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot)
922 struct swap_cluster_info *ci;
923 unsigned long offset, i;
927 * Should not even be attempting cluster allocations when huge
928 * page swap is disabled. Warn and fail the allocation.
930 if (!IS_ENABLED(CONFIG_THP_SWAP)) {
935 if (cluster_list_empty(&si->free_clusters))
938 idx = cluster_list_first(&si->free_clusters);
939 offset = idx * SWAPFILE_CLUSTER;
940 ci = lock_cluster(si, offset);
941 alloc_cluster(si, idx);
942 cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE);
944 map = si->swap_map + offset;
945 for (i = 0; i < SWAPFILE_CLUSTER; i++)
946 map[i] = SWAP_HAS_CACHE;
948 swap_range_alloc(si, offset, SWAPFILE_CLUSTER);
949 *slot = swp_entry(si->type, offset);
954 static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx)
956 unsigned long offset = idx * SWAPFILE_CLUSTER;
957 struct swap_cluster_info *ci;
959 ci = lock_cluster(si, offset);
960 memset(si->swap_map + offset, 0, SWAPFILE_CLUSTER);
961 cluster_set_count_flag(ci, 0, 0);
962 free_cluster(si, idx);
964 swap_range_free(si, offset, SWAPFILE_CLUSTER);
967 static unsigned long scan_swap_map(struct swap_info_struct *si,
973 n_ret = scan_swap_map_slots(si, usage, 1, &entry);
976 return swp_offset(entry);
982 int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size)
984 unsigned long size = swap_entry_size(entry_size);
985 struct swap_info_struct *si, *next;
990 /* Only single cluster request supported */
991 WARN_ON_ONCE(n_goal > 1 && size == SWAPFILE_CLUSTER);
993 avail_pgs = atomic_long_read(&nr_swap_pages) / size;
997 n_goal = min3((long)n_goal, (long)SWAP_BATCH, avail_pgs);
999 atomic_long_sub(n_goal * size, &nr_swap_pages);
1001 spin_lock(&swap_avail_lock);
1004 node = numa_node_id();
1005 plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) {
1006 /* requeue si to after same-priority siblings */
1007 plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]);
1008 spin_unlock(&swap_avail_lock);
1009 spin_lock(&si->lock);
1010 if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) {
1011 spin_lock(&swap_avail_lock);
1012 if (plist_node_empty(&si->avail_lists[node])) {
1013 spin_unlock(&si->lock);
1016 WARN(!si->highest_bit,
1017 "swap_info %d in list but !highest_bit\n",
1019 WARN(!(si->flags & SWP_WRITEOK),
1020 "swap_info %d in list but !SWP_WRITEOK\n",
1022 __del_from_avail_list(si);
1023 spin_unlock(&si->lock);
1026 if (size == SWAPFILE_CLUSTER) {
1027 if (!(si->flags & SWP_FS))
1028 n_ret = swap_alloc_cluster(si, swp_entries);
1030 n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE,
1031 n_goal, swp_entries);
1032 spin_unlock(&si->lock);
1033 if (n_ret || size == SWAPFILE_CLUSTER)
1035 pr_debug("scan_swap_map of si %d failed to find offset\n",
1038 spin_lock(&swap_avail_lock);
1041 * if we got here, it's likely that si was almost full before,
1042 * and since scan_swap_map() can drop the si->lock, multiple
1043 * callers probably all tried to get a page from the same si
1044 * and it filled up before we could get one; or, the si filled
1045 * up between us dropping swap_avail_lock and taking si->lock.
1046 * Since we dropped the swap_avail_lock, the swap_avail_head
1047 * list may have been modified; so if next is still in the
1048 * swap_avail_head list then try it, otherwise start over
1049 * if we have not gotten any slots.
1051 if (plist_node_empty(&next->avail_lists[node]))
1055 spin_unlock(&swap_avail_lock);
1059 atomic_long_add((long)(n_goal - n_ret) * size,
1065 /* The only caller of this function is now suspend routine */
1066 swp_entry_t get_swap_page_of_type(int type)
1068 struct swap_info_struct *si = swap_type_to_swap_info(type);
1074 spin_lock(&si->lock);
1075 if (si->flags & SWP_WRITEOK) {
1076 atomic_long_dec(&nr_swap_pages);
1077 /* This is called for allocating swap entry, not cache */
1078 offset = scan_swap_map(si, 1);
1080 spin_unlock(&si->lock);
1081 return swp_entry(type, offset);
1083 atomic_long_inc(&nr_swap_pages);
1085 spin_unlock(&si->lock);
1087 return (swp_entry_t) {0};
1090 static struct swap_info_struct *__swap_info_get(swp_entry_t entry)
1092 struct swap_info_struct *p;
1093 unsigned long offset;
1097 p = swp_swap_info(entry);
1100 if (!(p->flags & SWP_USED))
1102 offset = swp_offset(entry);
1103 if (offset >= p->max)
1108 pr_err("swap_info_get: %s%08lx\n", Bad_offset, entry.val);
1111 pr_err("swap_info_get: %s%08lx\n", Unused_file, entry.val);
1114 pr_err("swap_info_get: %s%08lx\n", Bad_file, entry.val);
1119 static struct swap_info_struct *_swap_info_get(swp_entry_t entry)
1121 struct swap_info_struct *p;
1123 p = __swap_info_get(entry);
1126 if (!p->swap_map[swp_offset(entry)])
1131 pr_err("swap_info_get: %s%08lx\n", Unused_offset, entry.val);
1137 static struct swap_info_struct *swap_info_get(swp_entry_t entry)
1139 struct swap_info_struct *p;
1141 p = _swap_info_get(entry);
1143 spin_lock(&p->lock);
1147 static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry,
1148 struct swap_info_struct *q)
1150 struct swap_info_struct *p;
1152 p = _swap_info_get(entry);
1156 spin_unlock(&q->lock);
1158 spin_lock(&p->lock);
1163 static unsigned char __swap_entry_free_locked(struct swap_info_struct *p,
1164 unsigned long offset,
1165 unsigned char usage)
1167 unsigned char count;
1168 unsigned char has_cache;
1170 count = p->swap_map[offset];
1172 has_cache = count & SWAP_HAS_CACHE;
1173 count &= ~SWAP_HAS_CACHE;
1175 if (usage == SWAP_HAS_CACHE) {
1176 VM_BUG_ON(!has_cache);
1178 } else if (count == SWAP_MAP_SHMEM) {
1180 * Or we could insist on shmem.c using a special
1181 * swap_shmem_free() and free_shmem_swap_and_cache()...
1184 } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
1185 if (count == COUNT_CONTINUED) {
1186 if (swap_count_continued(p, offset, count))
1187 count = SWAP_MAP_MAX | COUNT_CONTINUED;
1189 count = SWAP_MAP_MAX;
1194 usage = count | has_cache;
1195 p->swap_map[offset] = usage ? : SWAP_HAS_CACHE;
1201 * Check whether swap entry is valid in the swap device. If so,
1202 * return pointer to swap_info_struct, and keep the swap entry valid
1203 * via preventing the swap device from being swapoff, until
1204 * put_swap_device() is called. Otherwise return NULL.
1206 * The entirety of the RCU read critical section must come before the
1207 * return from or after the call to synchronize_rcu() in
1208 * enable_swap_info() or swapoff(). So if "si->flags & SWP_VALID" is
1209 * true, the si->map, si->cluster_info, etc. must be valid in the
1212 * Notice that swapoff or swapoff+swapon can still happen before the
1213 * rcu_read_lock() in get_swap_device() or after the rcu_read_unlock()
1214 * in put_swap_device() if there isn't any other way to prevent
1215 * swapoff, such as page lock, page table lock, etc. The caller must
1216 * be prepared for that. For example, the following situation is
1221 * ... swapoff+swapon
1222 * __read_swap_cache_async()
1223 * swapcache_prepare()
1224 * __swap_duplicate()
1226 * // verify PTE not changed
1228 * In __swap_duplicate(), the swap_map need to be checked before
1229 * changing partly because the specified swap entry may be for another
1230 * swap device which has been swapoff. And in do_swap_page(), after
1231 * the page is read from the swap device, the PTE is verified not
1232 * changed with the page table locked to check whether the swap device
1233 * has been swapoff or swapoff+swapon.
1235 struct swap_info_struct *get_swap_device(swp_entry_t entry)
1237 struct swap_info_struct *si;
1238 unsigned long offset;
1242 si = swp_swap_info(entry);
1247 if (!(si->flags & SWP_VALID))
1249 offset = swp_offset(entry);
1250 if (offset >= si->max)
1255 pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val);
1263 static unsigned char __swap_entry_free(struct swap_info_struct *p,
1264 swp_entry_t entry, unsigned char usage)
1266 struct swap_cluster_info *ci;
1267 unsigned long offset = swp_offset(entry);
1269 ci = lock_cluster_or_swap_info(p, offset);
1270 usage = __swap_entry_free_locked(p, offset, usage);
1271 unlock_cluster_or_swap_info(p, ci);
1273 free_swap_slot(entry);
1278 static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry)
1280 struct swap_cluster_info *ci;
1281 unsigned long offset = swp_offset(entry);
1282 unsigned char count;
1284 ci = lock_cluster(p, offset);
1285 count = p->swap_map[offset];
1286 VM_BUG_ON(count != SWAP_HAS_CACHE);
1287 p->swap_map[offset] = 0;
1288 dec_cluster_info_page(p, p->cluster_info, offset);
1291 mem_cgroup_uncharge_swap(entry, 1);
1292 swap_range_free(p, offset, 1);
1296 * Caller has made sure that the swap device corresponding to entry
1297 * is still around or has not been recycled.
1299 void swap_free(swp_entry_t entry)
1301 struct swap_info_struct *p;
1303 p = _swap_info_get(entry);
1305 __swap_entry_free(p, entry, 1);
1309 * Called after dropping swapcache to decrease refcnt to swap entries.
1311 void put_swap_page(struct page *page, swp_entry_t entry)
1313 unsigned long offset = swp_offset(entry);
1314 unsigned long idx = offset / SWAPFILE_CLUSTER;
1315 struct swap_cluster_info *ci;
1316 struct swap_info_struct *si;
1318 unsigned int i, free_entries = 0;
1320 int size = swap_entry_size(hpage_nr_pages(page));
1322 si = _swap_info_get(entry);
1326 ci = lock_cluster_or_swap_info(si, offset);
1327 if (size == SWAPFILE_CLUSTER) {
1328 VM_BUG_ON(!cluster_is_huge(ci));
1329 map = si->swap_map + offset;
1330 for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1332 VM_BUG_ON(!(val & SWAP_HAS_CACHE));
1333 if (val == SWAP_HAS_CACHE)
1336 cluster_clear_huge(ci);
1337 if (free_entries == SWAPFILE_CLUSTER) {
1338 unlock_cluster_or_swap_info(si, ci);
1339 spin_lock(&si->lock);
1340 mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER);
1341 swap_free_cluster(si, idx);
1342 spin_unlock(&si->lock);
1346 for (i = 0; i < size; i++, entry.val++) {
1347 if (!__swap_entry_free_locked(si, offset + i, SWAP_HAS_CACHE)) {
1348 unlock_cluster_or_swap_info(si, ci);
1349 free_swap_slot(entry);
1352 lock_cluster_or_swap_info(si, offset);
1355 unlock_cluster_or_swap_info(si, ci);
1358 #ifdef CONFIG_THP_SWAP
1359 int split_swap_cluster(swp_entry_t entry)
1361 struct swap_info_struct *si;
1362 struct swap_cluster_info *ci;
1363 unsigned long offset = swp_offset(entry);
1365 si = _swap_info_get(entry);
1368 ci = lock_cluster(si, offset);
1369 cluster_clear_huge(ci);
1375 static int swp_entry_cmp(const void *ent1, const void *ent2)
1377 const swp_entry_t *e1 = ent1, *e2 = ent2;
1379 return (int)swp_type(*e1) - (int)swp_type(*e2);
1382 void swapcache_free_entries(swp_entry_t *entries, int n)
1384 struct swap_info_struct *p, *prev;
1394 * Sort swap entries by swap device, so each lock is only taken once.
1395 * nr_swapfiles isn't absolutely correct, but the overhead of sort() is
1396 * so low that it isn't necessary to optimize further.
1398 if (nr_swapfiles > 1)
1399 sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL);
1400 for (i = 0; i < n; ++i) {
1401 p = swap_info_get_cont(entries[i], prev);
1403 swap_entry_free(p, entries[i]);
1407 spin_unlock(&p->lock);
1411 * How many references to page are currently swapped out?
1412 * This does not give an exact answer when swap count is continued,
1413 * but does include the high COUNT_CONTINUED flag to allow for that.
1415 int page_swapcount(struct page *page)
1418 struct swap_info_struct *p;
1419 struct swap_cluster_info *ci;
1421 unsigned long offset;
1423 entry.val = page_private(page);
1424 p = _swap_info_get(entry);
1426 offset = swp_offset(entry);
1427 ci = lock_cluster_or_swap_info(p, offset);
1428 count = swap_count(p->swap_map[offset]);
1429 unlock_cluster_or_swap_info(p, ci);
1434 int __swap_count(swp_entry_t entry)
1436 struct swap_info_struct *si;
1437 pgoff_t offset = swp_offset(entry);
1440 si = get_swap_device(entry);
1442 count = swap_count(si->swap_map[offset]);
1443 put_swap_device(si);
1448 static int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry)
1451 pgoff_t offset = swp_offset(entry);
1452 struct swap_cluster_info *ci;
1454 ci = lock_cluster_or_swap_info(si, offset);
1455 count = swap_count(si->swap_map[offset]);
1456 unlock_cluster_or_swap_info(si, ci);
1461 * How many references to @entry are currently swapped out?
1462 * This does not give an exact answer when swap count is continued,
1463 * but does include the high COUNT_CONTINUED flag to allow for that.
1465 int __swp_swapcount(swp_entry_t entry)
1468 struct swap_info_struct *si;
1470 si = get_swap_device(entry);
1472 count = swap_swapcount(si, entry);
1473 put_swap_device(si);
1479 * How many references to @entry are currently swapped out?
1480 * This considers COUNT_CONTINUED so it returns exact answer.
1482 int swp_swapcount(swp_entry_t entry)
1484 int count, tmp_count, n;
1485 struct swap_info_struct *p;
1486 struct swap_cluster_info *ci;
1491 p = _swap_info_get(entry);
1495 offset = swp_offset(entry);
1497 ci = lock_cluster_or_swap_info(p, offset);
1499 count = swap_count(p->swap_map[offset]);
1500 if (!(count & COUNT_CONTINUED))
1503 count &= ~COUNT_CONTINUED;
1504 n = SWAP_MAP_MAX + 1;
1506 page = vmalloc_to_page(p->swap_map + offset);
1507 offset &= ~PAGE_MASK;
1508 VM_BUG_ON(page_private(page) != SWP_CONTINUED);
1511 page = list_next_entry(page, lru);
1512 map = kmap_atomic(page);
1513 tmp_count = map[offset];
1516 count += (tmp_count & ~COUNT_CONTINUED) * n;
1517 n *= (SWAP_CONT_MAX + 1);
1518 } while (tmp_count & COUNT_CONTINUED);
1520 unlock_cluster_or_swap_info(p, ci);
1524 static bool swap_page_trans_huge_swapped(struct swap_info_struct *si,
1527 struct swap_cluster_info *ci;
1528 unsigned char *map = si->swap_map;
1529 unsigned long roffset = swp_offset(entry);
1530 unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER);
1534 ci = lock_cluster_or_swap_info(si, offset);
1535 if (!ci || !cluster_is_huge(ci)) {
1536 if (swap_count(map[roffset]))
1540 for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1541 if (swap_count(map[offset + i])) {
1547 unlock_cluster_or_swap_info(si, ci);
1551 static bool page_swapped(struct page *page)
1554 struct swap_info_struct *si;
1556 if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page)))
1557 return page_swapcount(page) != 0;
1559 page = compound_head(page);
1560 entry.val = page_private(page);
1561 si = _swap_info_get(entry);
1563 return swap_page_trans_huge_swapped(si, entry);
1567 static int page_trans_huge_map_swapcount(struct page *page, int *total_mapcount,
1568 int *total_swapcount)
1570 int i, map_swapcount, _total_mapcount, _total_swapcount;
1571 unsigned long offset = 0;
1572 struct swap_info_struct *si;
1573 struct swap_cluster_info *ci = NULL;
1574 unsigned char *map = NULL;
1575 int mapcount, swapcount = 0;
1577 /* hugetlbfs shouldn't call it */
1578 VM_BUG_ON_PAGE(PageHuge(page), page);
1580 if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page))) {
1581 mapcount = page_trans_huge_mapcount(page, total_mapcount);
1582 if (PageSwapCache(page))
1583 swapcount = page_swapcount(page);
1584 if (total_swapcount)
1585 *total_swapcount = swapcount;
1586 return mapcount + swapcount;
1589 page = compound_head(page);
1591 _total_mapcount = _total_swapcount = map_swapcount = 0;
1592 if (PageSwapCache(page)) {
1595 entry.val = page_private(page);
1596 si = _swap_info_get(entry);
1599 offset = swp_offset(entry);
1603 ci = lock_cluster(si, offset);
1604 for (i = 0; i < HPAGE_PMD_NR; i++) {
1605 mapcount = atomic_read(&page[i]._mapcount) + 1;
1606 _total_mapcount += mapcount;
1608 swapcount = swap_count(map[offset + i]);
1609 _total_swapcount += swapcount;
1611 map_swapcount = max(map_swapcount, mapcount + swapcount);
1614 if (PageDoubleMap(page)) {
1616 _total_mapcount -= HPAGE_PMD_NR;
1618 mapcount = compound_mapcount(page);
1619 map_swapcount += mapcount;
1620 _total_mapcount += mapcount;
1622 *total_mapcount = _total_mapcount;
1623 if (total_swapcount)
1624 *total_swapcount = _total_swapcount;
1626 return map_swapcount;
1630 * We can write to an anon page without COW if there are no other references
1631 * to it. And as a side-effect, free up its swap: because the old content
1632 * on disk will never be read, and seeking back there to write new content
1633 * later would only waste time away from clustering.
1635 * NOTE: total_map_swapcount should not be relied upon by the caller if
1636 * reuse_swap_page() returns false, but it may be always overwritten
1637 * (see the other implementation for CONFIG_SWAP=n).
1639 bool reuse_swap_page(struct page *page, int *total_map_swapcount)
1641 int count, total_mapcount, total_swapcount;
1643 VM_BUG_ON_PAGE(!PageLocked(page), page);
1644 if (unlikely(PageKsm(page)))
1646 count = page_trans_huge_map_swapcount(page, &total_mapcount,
1648 if (total_map_swapcount)
1649 *total_map_swapcount = total_mapcount + total_swapcount;
1650 if (count == 1 && PageSwapCache(page) &&
1651 (likely(!PageTransCompound(page)) ||
1652 /* The remaining swap count will be freed soon */
1653 total_swapcount == page_swapcount(page))) {
1654 if (!PageWriteback(page)) {
1655 page = compound_head(page);
1656 delete_from_swap_cache(page);
1660 struct swap_info_struct *p;
1662 entry.val = page_private(page);
1663 p = swap_info_get(entry);
1664 if (p->flags & SWP_STABLE_WRITES) {
1665 spin_unlock(&p->lock);
1668 spin_unlock(&p->lock);
1676 * If swap is getting full, or if there are no more mappings of this page,
1677 * then try_to_free_swap is called to free its swap space.
1679 int try_to_free_swap(struct page *page)
1681 VM_BUG_ON_PAGE(!PageLocked(page), page);
1683 if (!PageSwapCache(page))
1685 if (PageWriteback(page))
1687 if (page_swapped(page))
1691 * Once hibernation has begun to create its image of memory,
1692 * there's a danger that one of the calls to try_to_free_swap()
1693 * - most probably a call from __try_to_reclaim_swap() while
1694 * hibernation is allocating its own swap pages for the image,
1695 * but conceivably even a call from memory reclaim - will free
1696 * the swap from a page which has already been recorded in the
1697 * image as a clean swapcache page, and then reuse its swap for
1698 * another page of the image. On waking from hibernation, the
1699 * original page might be freed under memory pressure, then
1700 * later read back in from swap, now with the wrong data.
1702 * Hibernation suspends storage while it is writing the image
1703 * to disk so check that here.
1705 if (pm_suspended_storage())
1708 page = compound_head(page);
1709 delete_from_swap_cache(page);
1715 * Free the swap entry like above, but also try to
1716 * free the page cache entry if it is the last user.
1718 int free_swap_and_cache(swp_entry_t entry)
1720 struct swap_info_struct *p;
1721 unsigned char count;
1723 if (non_swap_entry(entry))
1726 p = _swap_info_get(entry);
1728 count = __swap_entry_free(p, entry, 1);
1729 if (count == SWAP_HAS_CACHE &&
1730 !swap_page_trans_huge_swapped(p, entry))
1731 __try_to_reclaim_swap(p, swp_offset(entry),
1732 TTRS_UNMAPPED | TTRS_FULL);
1737 #ifdef CONFIG_HIBERNATION
1739 * Find the swap type that corresponds to given device (if any).
1741 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1742 * from 0, in which the swap header is expected to be located.
1744 * This is needed for the suspend to disk (aka swsusp).
1746 int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
1748 struct block_device *bdev = NULL;
1752 bdev = bdget(device);
1754 spin_lock(&swap_lock);
1755 for (type = 0; type < nr_swapfiles; type++) {
1756 struct swap_info_struct *sis = swap_info[type];
1758 if (!(sis->flags & SWP_WRITEOK))
1763 *bdev_p = bdgrab(sis->bdev);
1765 spin_unlock(&swap_lock);
1768 if (bdev == sis->bdev) {
1769 struct swap_extent *se = first_se(sis);
1771 if (se->start_block == offset) {
1773 *bdev_p = bdgrab(sis->bdev);
1775 spin_unlock(&swap_lock);
1781 spin_unlock(&swap_lock);
1789 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1790 * corresponding to given index in swap_info (swap type).
1792 sector_t swapdev_block(int type, pgoff_t offset)
1794 struct block_device *bdev;
1795 struct swap_info_struct *si = swap_type_to_swap_info(type);
1797 if (!si || !(si->flags & SWP_WRITEOK))
1799 return map_swap_entry(swp_entry(type, offset), &bdev);
1803 * Return either the total number of swap pages of given type, or the number
1804 * of free pages of that type (depending on @free)
1806 * This is needed for software suspend
1808 unsigned int count_swap_pages(int type, int free)
1812 spin_lock(&swap_lock);
1813 if ((unsigned int)type < nr_swapfiles) {
1814 struct swap_info_struct *sis = swap_info[type];
1816 spin_lock(&sis->lock);
1817 if (sis->flags & SWP_WRITEOK) {
1820 n -= sis->inuse_pages;
1822 spin_unlock(&sis->lock);
1824 spin_unlock(&swap_lock);
1827 #endif /* CONFIG_HIBERNATION */
1829 static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte)
1831 return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte);
1835 * No need to decide whether this PTE shares the swap entry with others,
1836 * just let do_wp_page work it out if a write is requested later - to
1837 * force COW, vm_page_prot omits write permission from any private vma.
1839 static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
1840 unsigned long addr, swp_entry_t entry, struct page *page)
1842 struct page *swapcache;
1843 struct mem_cgroup *memcg;
1849 page = ksm_might_need_to_copy(page, vma, addr);
1850 if (unlikely(!page))
1853 if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL,
1859 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1860 if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) {
1861 mem_cgroup_cancel_charge(page, memcg, false);
1866 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1867 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
1869 set_pte_at(vma->vm_mm, addr, pte,
1870 pte_mkold(mk_pte(page, vma->vm_page_prot)));
1871 if (page == swapcache) {
1872 page_add_anon_rmap(page, vma, addr, false);
1873 mem_cgroup_commit_charge(page, memcg, true, false);
1874 } else { /* ksm created a completely new copy */
1875 page_add_new_anon_rmap(page, vma, addr, false);
1876 mem_cgroup_commit_charge(page, memcg, false, false);
1877 lru_cache_add_active_or_unevictable(page, vma);
1881 * Move the page to the active list so it is not
1882 * immediately swapped out again after swapon.
1884 activate_page(page);
1886 pte_unmap_unlock(pte, ptl);
1888 if (page != swapcache) {
1895 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1896 unsigned long addr, unsigned long end,
1897 unsigned int type, bool frontswap,
1898 unsigned long *fs_pages_to_unuse)
1903 struct swap_info_struct *si;
1904 unsigned long offset;
1906 volatile unsigned char *swap_map;
1908 si = swap_info[type];
1909 pte = pte_offset_map(pmd, addr);
1911 struct vm_fault vmf;
1913 if (!is_swap_pte(*pte))
1916 entry = pte_to_swp_entry(*pte);
1917 if (swp_type(entry) != type)
1920 offset = swp_offset(entry);
1921 if (frontswap && !frontswap_test(si, offset))
1925 swap_map = &si->swap_map[offset];
1926 page = lookup_swap_cache(entry, vma, addr);
1931 page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE,
1935 if (*swap_map == 0 || *swap_map == SWAP_MAP_BAD)
1941 wait_on_page_writeback(page);
1942 ret = unuse_pte(vma, pmd, addr, entry, page);
1949 try_to_free_swap(page);
1953 if (*fs_pages_to_unuse && !--(*fs_pages_to_unuse)) {
1954 ret = FRONTSWAP_PAGES_UNUSED;
1958 pte = pte_offset_map(pmd, addr);
1959 } while (pte++, addr += PAGE_SIZE, addr != end);
1967 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1968 unsigned long addr, unsigned long end,
1969 unsigned int type, bool frontswap,
1970 unsigned long *fs_pages_to_unuse)
1976 pmd = pmd_offset(pud, addr);
1979 next = pmd_addr_end(addr, end);
1980 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
1982 ret = unuse_pte_range(vma, pmd, addr, next, type,
1983 frontswap, fs_pages_to_unuse);
1986 } while (pmd++, addr = next, addr != end);
1990 static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d,
1991 unsigned long addr, unsigned long end,
1992 unsigned int type, bool frontswap,
1993 unsigned long *fs_pages_to_unuse)
1999 pud = pud_offset(p4d, addr);
2001 next = pud_addr_end(addr, end);
2002 if (pud_none_or_clear_bad(pud))
2004 ret = unuse_pmd_range(vma, pud, addr, next, type,
2005 frontswap, fs_pages_to_unuse);
2008 } while (pud++, addr = next, addr != end);
2012 static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd,
2013 unsigned long addr, unsigned long end,
2014 unsigned int type, bool frontswap,
2015 unsigned long *fs_pages_to_unuse)
2021 p4d = p4d_offset(pgd, addr);
2023 next = p4d_addr_end(addr, end);
2024 if (p4d_none_or_clear_bad(p4d))
2026 ret = unuse_pud_range(vma, p4d, addr, next, type,
2027 frontswap, fs_pages_to_unuse);
2030 } while (p4d++, addr = next, addr != end);
2034 static int unuse_vma(struct vm_area_struct *vma, unsigned int type,
2035 bool frontswap, unsigned long *fs_pages_to_unuse)
2038 unsigned long addr, end, next;
2041 addr = vma->vm_start;
2044 pgd = pgd_offset(vma->vm_mm, addr);
2046 next = pgd_addr_end(addr, end);
2047 if (pgd_none_or_clear_bad(pgd))
2049 ret = unuse_p4d_range(vma, pgd, addr, next, type,
2050 frontswap, fs_pages_to_unuse);
2053 } while (pgd++, addr = next, addr != end);
2057 static int unuse_mm(struct mm_struct *mm, unsigned int type,
2058 bool frontswap, unsigned long *fs_pages_to_unuse)
2060 struct vm_area_struct *vma;
2063 down_read(&mm->mmap_sem);
2064 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2065 if (vma->anon_vma) {
2066 ret = unuse_vma(vma, type, frontswap,
2073 up_read(&mm->mmap_sem);
2078 * Scan swap_map (or frontswap_map if frontswap parameter is true)
2079 * from current position to next entry still in use. Return 0
2080 * if there are no inuse entries after prev till end of the map.
2082 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
2083 unsigned int prev, bool frontswap)
2086 unsigned char count;
2089 * No need for swap_lock here: we're just looking
2090 * for whether an entry is in use, not modifying it; false
2091 * hits are okay, and sys_swapoff() has already prevented new
2092 * allocations from this area (while holding swap_lock).
2094 for (i = prev + 1; i < si->max; i++) {
2095 count = READ_ONCE(si->swap_map[i]);
2096 if (count && swap_count(count) != SWAP_MAP_BAD)
2097 if (!frontswap || frontswap_test(si, i))
2099 if ((i % LATENCY_LIMIT) == 0)
2110 * If the boolean frontswap is true, only unuse pages_to_unuse pages;
2111 * pages_to_unuse==0 means all pages; ignored if frontswap is false
2113 int try_to_unuse(unsigned int type, bool frontswap,
2114 unsigned long pages_to_unuse)
2116 struct mm_struct *prev_mm;
2117 struct mm_struct *mm;
2118 struct list_head *p;
2120 struct swap_info_struct *si = swap_info[type];
2125 if (!READ_ONCE(si->inuse_pages))
2132 retval = shmem_unuse(type, frontswap, &pages_to_unuse);
2139 spin_lock(&mmlist_lock);
2140 p = &init_mm.mmlist;
2141 while (READ_ONCE(si->inuse_pages) &&
2142 !signal_pending(current) &&
2143 (p = p->next) != &init_mm.mmlist) {
2145 mm = list_entry(p, struct mm_struct, mmlist);
2146 if (!mmget_not_zero(mm))
2148 spin_unlock(&mmlist_lock);
2151 retval = unuse_mm(mm, type, frontswap, &pages_to_unuse);
2159 * Make sure that we aren't completely killing
2160 * interactive performance.
2163 spin_lock(&mmlist_lock);
2165 spin_unlock(&mmlist_lock);
2170 while (READ_ONCE(si->inuse_pages) &&
2171 !signal_pending(current) &&
2172 (i = find_next_to_unuse(si, i, frontswap)) != 0) {
2174 entry = swp_entry(type, i);
2175 page = find_get_page(swap_address_space(entry), i);
2180 * It is conceivable that a racing task removed this page from
2181 * swap cache just before we acquired the page lock. The page
2182 * might even be back in swap cache on another swap area. But
2183 * that is okay, try_to_free_swap() only removes stale pages.
2186 wait_on_page_writeback(page);
2187 try_to_free_swap(page);
2192 * For frontswap, we just need to unuse pages_to_unuse, if
2193 * it was specified. Need not check frontswap again here as
2194 * we already zeroed out pages_to_unuse if not frontswap.
2196 if (pages_to_unuse && --pages_to_unuse == 0)
2201 * Lets check again to see if there are still swap entries in the map.
2202 * If yes, we would need to do retry the unuse logic again.
2203 * Under global memory pressure, swap entries can be reinserted back
2204 * into process space after the mmlist loop above passes over them.
2206 * Limit the number of retries? No: when mmget_not_zero() above fails,
2207 * that mm is likely to be freeing swap from exit_mmap(), which proceeds
2208 * at its own independent pace; and even shmem_writepage() could have
2209 * been preempted after get_swap_page(), temporarily hiding that swap.
2210 * It's easy and robust (though cpu-intensive) just to keep retrying.
2212 if (READ_ONCE(si->inuse_pages)) {
2213 if (!signal_pending(current))
2218 return (retval == FRONTSWAP_PAGES_UNUSED) ? 0 : retval;
2222 * After a successful try_to_unuse, if no swap is now in use, we know
2223 * we can empty the mmlist. swap_lock must be held on entry and exit.
2224 * Note that mmlist_lock nests inside swap_lock, and an mm must be
2225 * added to the mmlist just after page_duplicate - before would be racy.
2227 static void drain_mmlist(void)
2229 struct list_head *p, *next;
2232 for (type = 0; type < nr_swapfiles; type++)
2233 if (swap_info[type]->inuse_pages)
2235 spin_lock(&mmlist_lock);
2236 list_for_each_safe(p, next, &init_mm.mmlist)
2238 spin_unlock(&mmlist_lock);
2242 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
2243 * corresponds to page offset for the specified swap entry.
2244 * Note that the type of this function is sector_t, but it returns page offset
2245 * into the bdev, not sector offset.
2247 static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
2249 struct swap_info_struct *sis;
2250 struct swap_extent *se;
2253 sis = swp_swap_info(entry);
2256 offset = swp_offset(entry);
2257 se = offset_to_swap_extent(sis, offset);
2258 return se->start_block + (offset - se->start_page);
2262 * Returns the page offset into bdev for the specified page's swap entry.
2264 sector_t map_swap_page(struct page *page, struct block_device **bdev)
2267 entry.val = page_private(page);
2268 return map_swap_entry(entry, bdev);
2272 * Free all of a swapdev's extent information
2274 static void destroy_swap_extents(struct swap_info_struct *sis)
2276 while (!RB_EMPTY_ROOT(&sis->swap_extent_root)) {
2277 struct rb_node *rb = sis->swap_extent_root.rb_node;
2278 struct swap_extent *se = rb_entry(rb, struct swap_extent, rb_node);
2280 rb_erase(rb, &sis->swap_extent_root);
2284 if (sis->flags & SWP_ACTIVATED) {
2285 struct file *swap_file = sis->swap_file;
2286 struct address_space *mapping = swap_file->f_mapping;
2288 sis->flags &= ~SWP_ACTIVATED;
2289 if (mapping->a_ops->swap_deactivate)
2290 mapping->a_ops->swap_deactivate(swap_file);
2295 * Add a block range (and the corresponding page range) into this swapdev's
2298 * This function rather assumes that it is called in ascending page order.
2301 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
2302 unsigned long nr_pages, sector_t start_block)
2304 struct rb_node **link = &sis->swap_extent_root.rb_node, *parent = NULL;
2305 struct swap_extent *se;
2306 struct swap_extent *new_se;
2309 * place the new node at the right most since the
2310 * function is called in ascending page order.
2314 link = &parent->rb_right;
2318 se = rb_entry(parent, struct swap_extent, rb_node);
2319 BUG_ON(se->start_page + se->nr_pages != start_page);
2320 if (se->start_block + se->nr_pages == start_block) {
2322 se->nr_pages += nr_pages;
2327 /* No merge, insert a new extent. */
2328 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
2331 new_se->start_page = start_page;
2332 new_se->nr_pages = nr_pages;
2333 new_se->start_block = start_block;
2335 rb_link_node(&new_se->rb_node, parent, link);
2336 rb_insert_color(&new_se->rb_node, &sis->swap_extent_root);
2339 EXPORT_SYMBOL_GPL(add_swap_extent);
2342 * A `swap extent' is a simple thing which maps a contiguous range of pages
2343 * onto a contiguous range of disk blocks. An ordered list of swap extents
2344 * is built at swapon time and is then used at swap_writepage/swap_readpage
2345 * time for locating where on disk a page belongs.
2347 * If the swapfile is an S_ISBLK block device, a single extent is installed.
2348 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2349 * swap files identically.
2351 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2352 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
2353 * swapfiles are handled *identically* after swapon time.
2355 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2356 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
2357 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
2358 * requirements, they are simply tossed out - we will never use those blocks
2361 * For all swap devices we set S_SWAPFILE across the life of the swapon. This
2362 * prevents users from writing to the swap device, which will corrupt memory.
2364 * The amount of disk space which a single swap extent represents varies.
2365 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
2366 * extents in the list. To avoid much list walking, we cache the previous
2367 * search location in `curr_swap_extent', and start new searches from there.
2368 * This is extremely effective. The average number of iterations in
2369 * map_swap_page() has been measured at about 0.3 per page. - akpm.
2371 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
2373 struct file *swap_file = sis->swap_file;
2374 struct address_space *mapping = swap_file->f_mapping;
2375 struct inode *inode = mapping->host;
2378 if (S_ISBLK(inode->i_mode)) {
2379 ret = add_swap_extent(sis, 0, sis->max, 0);
2384 if (mapping->a_ops->swap_activate) {
2385 ret = mapping->a_ops->swap_activate(sis, swap_file, span);
2387 sis->flags |= SWP_ACTIVATED;
2389 sis->flags |= SWP_FS;
2390 ret = add_swap_extent(sis, 0, sis->max, 0);
2396 return generic_swapfile_activate(sis, swap_file, span);
2399 static int swap_node(struct swap_info_struct *p)
2401 struct block_device *bdev;
2406 bdev = p->swap_file->f_inode->i_sb->s_bdev;
2408 return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE;
2411 static void setup_swap_info(struct swap_info_struct *p, int prio,
2412 unsigned char *swap_map,
2413 struct swap_cluster_info *cluster_info)
2420 p->prio = --least_priority;
2422 * the plist prio is negated because plist ordering is
2423 * low-to-high, while swap ordering is high-to-low
2425 p->list.prio = -p->prio;
2428 p->avail_lists[i].prio = -p->prio;
2430 if (swap_node(p) == i)
2431 p->avail_lists[i].prio = 1;
2433 p->avail_lists[i].prio = -p->prio;
2436 p->swap_map = swap_map;
2437 p->cluster_info = cluster_info;
2440 static void _enable_swap_info(struct swap_info_struct *p)
2442 p->flags |= SWP_WRITEOK | SWP_VALID;
2443 atomic_long_add(p->pages, &nr_swap_pages);
2444 total_swap_pages += p->pages;
2446 assert_spin_locked(&swap_lock);
2448 * both lists are plists, and thus priority ordered.
2449 * swap_active_head needs to be priority ordered for swapoff(),
2450 * which on removal of any swap_info_struct with an auto-assigned
2451 * (i.e. negative) priority increments the auto-assigned priority
2452 * of any lower-priority swap_info_structs.
2453 * swap_avail_head needs to be priority ordered for get_swap_page(),
2454 * which allocates swap pages from the highest available priority
2457 plist_add(&p->list, &swap_active_head);
2458 add_to_avail_list(p);
2461 static void enable_swap_info(struct swap_info_struct *p, int prio,
2462 unsigned char *swap_map,
2463 struct swap_cluster_info *cluster_info,
2464 unsigned long *frontswap_map)
2466 frontswap_init(p->type, frontswap_map);
2467 spin_lock(&swap_lock);
2468 spin_lock(&p->lock);
2469 setup_swap_info(p, prio, swap_map, cluster_info);
2470 spin_unlock(&p->lock);
2471 spin_unlock(&swap_lock);
2473 * Guarantee swap_map, cluster_info, etc. fields are valid
2474 * between get/put_swap_device() if SWP_VALID bit is set
2477 spin_lock(&swap_lock);
2478 spin_lock(&p->lock);
2479 _enable_swap_info(p);
2480 spin_unlock(&p->lock);
2481 spin_unlock(&swap_lock);
2484 static void reinsert_swap_info(struct swap_info_struct *p)
2486 spin_lock(&swap_lock);
2487 spin_lock(&p->lock);
2488 setup_swap_info(p, p->prio, p->swap_map, p->cluster_info);
2489 _enable_swap_info(p);
2490 spin_unlock(&p->lock);
2491 spin_unlock(&swap_lock);
2494 bool has_usable_swap(void)
2498 spin_lock(&swap_lock);
2499 if (plist_head_empty(&swap_active_head))
2501 spin_unlock(&swap_lock);
2505 SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
2507 struct swap_info_struct *p = NULL;
2508 unsigned char *swap_map;
2509 struct swap_cluster_info *cluster_info;
2510 unsigned long *frontswap_map;
2511 struct file *swap_file, *victim;
2512 struct address_space *mapping;
2513 struct inode *inode;
2514 struct filename *pathname;
2516 unsigned int old_block_size;
2518 if (!capable(CAP_SYS_ADMIN))
2521 BUG_ON(!current->mm);
2523 pathname = getname(specialfile);
2524 if (IS_ERR(pathname))
2525 return PTR_ERR(pathname);
2527 victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
2528 err = PTR_ERR(victim);
2532 mapping = victim->f_mapping;
2533 spin_lock(&swap_lock);
2534 plist_for_each_entry(p, &swap_active_head, list) {
2535 if (p->flags & SWP_WRITEOK) {
2536 if (p->swap_file->f_mapping == mapping) {
2544 spin_unlock(&swap_lock);
2547 if (!security_vm_enough_memory_mm(current->mm, p->pages))
2548 vm_unacct_memory(p->pages);
2551 spin_unlock(&swap_lock);
2554 del_from_avail_list(p);
2555 spin_lock(&p->lock);
2557 struct swap_info_struct *si = p;
2560 plist_for_each_entry_continue(si, &swap_active_head, list) {
2563 for_each_node(nid) {
2564 if (si->avail_lists[nid].prio != 1)
2565 si->avail_lists[nid].prio--;
2570 plist_del(&p->list, &swap_active_head);
2571 atomic_long_sub(p->pages, &nr_swap_pages);
2572 total_swap_pages -= p->pages;
2573 p->flags &= ~SWP_WRITEOK;
2574 spin_unlock(&p->lock);
2575 spin_unlock(&swap_lock);
2577 disable_swap_slots_cache_lock();
2579 set_current_oom_origin();
2580 err = try_to_unuse(p->type, false, 0); /* force unuse all pages */
2581 clear_current_oom_origin();
2584 /* re-insert swap space back into swap_list */
2585 reinsert_swap_info(p);
2586 reenable_swap_slots_cache_unlock();
2590 reenable_swap_slots_cache_unlock();
2592 spin_lock(&swap_lock);
2593 spin_lock(&p->lock);
2594 p->flags &= ~SWP_VALID; /* mark swap device as invalid */
2595 spin_unlock(&p->lock);
2596 spin_unlock(&swap_lock);
2598 * wait for swap operations protected by get/put_swap_device()
2603 flush_work(&p->discard_work);
2605 destroy_swap_extents(p);
2606 if (p->flags & SWP_CONTINUED)
2607 free_swap_count_continuations(p);
2609 if (!p->bdev || !blk_queue_nonrot(bdev_get_queue(p->bdev)))
2610 atomic_dec(&nr_rotate_swap);
2612 mutex_lock(&swapon_mutex);
2613 spin_lock(&swap_lock);
2614 spin_lock(&p->lock);
2617 /* wait for anyone still in scan_swap_map */
2618 p->highest_bit = 0; /* cuts scans short */
2619 while (p->flags >= SWP_SCANNING) {
2620 spin_unlock(&p->lock);
2621 spin_unlock(&swap_lock);
2622 schedule_timeout_uninterruptible(1);
2623 spin_lock(&swap_lock);
2624 spin_lock(&p->lock);
2627 swap_file = p->swap_file;
2628 old_block_size = p->old_block_size;
2629 p->swap_file = NULL;
2631 swap_map = p->swap_map;
2633 cluster_info = p->cluster_info;
2634 p->cluster_info = NULL;
2635 frontswap_map = frontswap_map_get(p);
2636 spin_unlock(&p->lock);
2637 spin_unlock(&swap_lock);
2638 frontswap_invalidate_area(p->type);
2639 frontswap_map_set(p, NULL);
2640 mutex_unlock(&swapon_mutex);
2641 free_percpu(p->percpu_cluster);
2642 p->percpu_cluster = NULL;
2644 kvfree(cluster_info);
2645 kvfree(frontswap_map);
2646 /* Destroy swap account information */
2647 swap_cgroup_swapoff(p->type);
2648 exit_swap_address_space(p->type);
2650 inode = mapping->host;
2651 if (S_ISBLK(inode->i_mode)) {
2652 struct block_device *bdev = I_BDEV(inode);
2654 set_blocksize(bdev, old_block_size);
2655 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2659 inode->i_flags &= ~S_SWAPFILE;
2660 inode_unlock(inode);
2661 filp_close(swap_file, NULL);
2664 * Clear the SWP_USED flag after all resources are freed so that swapon
2665 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
2666 * not hold p->lock after we cleared its SWP_WRITEOK.
2668 spin_lock(&swap_lock);
2670 spin_unlock(&swap_lock);
2673 atomic_inc(&proc_poll_event);
2674 wake_up_interruptible(&proc_poll_wait);
2677 filp_close(victim, NULL);
2683 #ifdef CONFIG_PROC_FS
2684 static __poll_t swaps_poll(struct file *file, poll_table *wait)
2686 struct seq_file *seq = file->private_data;
2688 poll_wait(file, &proc_poll_wait, wait);
2690 if (seq->poll_event != atomic_read(&proc_poll_event)) {
2691 seq->poll_event = atomic_read(&proc_poll_event);
2692 return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
2695 return EPOLLIN | EPOLLRDNORM;
2699 static void *swap_start(struct seq_file *swap, loff_t *pos)
2701 struct swap_info_struct *si;
2705 mutex_lock(&swapon_mutex);
2708 return SEQ_START_TOKEN;
2710 for (type = 0; (si = swap_type_to_swap_info(type)); type++) {
2711 if (!(si->flags & SWP_USED) || !si->swap_map)
2720 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2722 struct swap_info_struct *si = v;
2725 if (v == SEQ_START_TOKEN)
2728 type = si->type + 1;
2731 for (; (si = swap_type_to_swap_info(type)); type++) {
2732 if (!(si->flags & SWP_USED) || !si->swap_map)
2740 static void swap_stop(struct seq_file *swap, void *v)
2742 mutex_unlock(&swapon_mutex);
2745 static int swap_show(struct seq_file *swap, void *v)
2747 struct swap_info_struct *si = v;
2751 if (si == SEQ_START_TOKEN) {
2752 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2756 file = si->swap_file;
2757 len = seq_file_path(swap, file, " \t\n\\");
2758 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
2759 len < 40 ? 40 - len : 1, " ",
2760 S_ISBLK(file_inode(file)->i_mode) ?
2761 "partition" : "file\t",
2762 si->pages << (PAGE_SHIFT - 10),
2763 si->inuse_pages << (PAGE_SHIFT - 10),
2768 static const struct seq_operations swaps_op = {
2769 .start = swap_start,
2775 static int swaps_open(struct inode *inode, struct file *file)
2777 struct seq_file *seq;
2780 ret = seq_open(file, &swaps_op);
2784 seq = file->private_data;
2785 seq->poll_event = atomic_read(&proc_poll_event);
2789 static const struct proc_ops swaps_proc_ops = {
2790 .proc_flags = PROC_ENTRY_PERMANENT,
2791 .proc_open = swaps_open,
2792 .proc_read = seq_read,
2793 .proc_lseek = seq_lseek,
2794 .proc_release = seq_release,
2795 .proc_poll = swaps_poll,
2798 static int __init procswaps_init(void)
2800 proc_create("swaps", 0, NULL, &swaps_proc_ops);
2803 __initcall(procswaps_init);
2804 #endif /* CONFIG_PROC_FS */
2806 #ifdef MAX_SWAPFILES_CHECK
2807 static int __init max_swapfiles_check(void)
2809 MAX_SWAPFILES_CHECK();
2812 late_initcall(max_swapfiles_check);
2815 static struct swap_info_struct *alloc_swap_info(void)
2817 struct swap_info_struct *p;
2821 p = kvzalloc(struct_size(p, avail_lists, nr_node_ids), GFP_KERNEL);
2823 return ERR_PTR(-ENOMEM);
2825 spin_lock(&swap_lock);
2826 for (type = 0; type < nr_swapfiles; type++) {
2827 if (!(swap_info[type]->flags & SWP_USED))
2830 if (type >= MAX_SWAPFILES) {
2831 spin_unlock(&swap_lock);
2833 return ERR_PTR(-EPERM);
2835 if (type >= nr_swapfiles) {
2837 WRITE_ONCE(swap_info[type], p);
2839 * Write swap_info[type] before nr_swapfiles, in case a
2840 * racing procfs swap_start() or swap_next() is reading them.
2841 * (We never shrink nr_swapfiles, we never free this entry.)
2844 WRITE_ONCE(nr_swapfiles, nr_swapfiles + 1);
2847 p = swap_info[type];
2849 * Do not memset this entry: a racing procfs swap_next()
2850 * would be relying on p->type to remain valid.
2853 p->swap_extent_root = RB_ROOT;
2854 plist_node_init(&p->list, 0);
2856 plist_node_init(&p->avail_lists[i], 0);
2857 p->flags = SWP_USED;
2858 spin_unlock(&swap_lock);
2859 spin_lock_init(&p->lock);
2860 spin_lock_init(&p->cont_lock);
2865 static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2869 if (S_ISBLK(inode->i_mode)) {
2870 p->bdev = bdgrab(I_BDEV(inode));
2871 error = blkdev_get(p->bdev,
2872 FMODE_READ | FMODE_WRITE | FMODE_EXCL, p);
2877 p->old_block_size = block_size(p->bdev);
2878 error = set_blocksize(p->bdev, PAGE_SIZE);
2882 * Zoned block devices contain zones that have a sequential
2883 * write only restriction. Hence zoned block devices are not
2884 * suitable for swapping. Disallow them here.
2886 if (blk_queue_is_zoned(p->bdev->bd_queue))
2888 p->flags |= SWP_BLKDEV;
2889 } else if (S_ISREG(inode->i_mode)) {
2890 p->bdev = inode->i_sb->s_bdev;
2898 * Find out how many pages are allowed for a single swap device. There
2899 * are two limiting factors:
2900 * 1) the number of bits for the swap offset in the swp_entry_t type, and
2901 * 2) the number of bits in the swap pte, as defined by the different
2904 * In order to find the largest possible bit mask, a swap entry with
2905 * swap type 0 and swap offset ~0UL is created, encoded to a swap pte,
2906 * decoded to a swp_entry_t again, and finally the swap offset is
2909 * This will mask all the bits from the initial ~0UL mask that can't
2910 * be encoded in either the swp_entry_t or the architecture definition
2913 unsigned long generic_max_swapfile_size(void)
2915 return swp_offset(pte_to_swp_entry(
2916 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2919 /* Can be overridden by an architecture for additional checks. */
2920 __weak unsigned long max_swapfile_size(void)
2922 return generic_max_swapfile_size();
2925 static unsigned long read_swap_header(struct swap_info_struct *p,
2926 union swap_header *swap_header,
2927 struct inode *inode)
2930 unsigned long maxpages;
2931 unsigned long swapfilepages;
2932 unsigned long last_page;
2934 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
2935 pr_err("Unable to find swap-space signature\n");
2939 /* swap partition endianess hack... */
2940 if (swab32(swap_header->info.version) == 1) {
2941 swab32s(&swap_header->info.version);
2942 swab32s(&swap_header->info.last_page);
2943 swab32s(&swap_header->info.nr_badpages);
2944 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2946 for (i = 0; i < swap_header->info.nr_badpages; i++)
2947 swab32s(&swap_header->info.badpages[i]);
2949 /* Check the swap header's sub-version */
2950 if (swap_header->info.version != 1) {
2951 pr_warn("Unable to handle swap header version %d\n",
2952 swap_header->info.version);
2957 p->cluster_next = 1;
2960 maxpages = max_swapfile_size();
2961 last_page = swap_header->info.last_page;
2963 pr_warn("Empty swap-file\n");
2966 if (last_page > maxpages) {
2967 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
2968 maxpages << (PAGE_SHIFT - 10),
2969 last_page << (PAGE_SHIFT - 10));
2971 if (maxpages > last_page) {
2972 maxpages = last_page + 1;
2973 /* p->max is an unsigned int: don't overflow it */
2974 if ((unsigned int)maxpages == 0)
2975 maxpages = UINT_MAX;
2977 p->highest_bit = maxpages - 1;
2981 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2982 if (swapfilepages && maxpages > swapfilepages) {
2983 pr_warn("Swap area shorter than signature indicates\n");
2986 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
2988 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2994 #define SWAP_CLUSTER_INFO_COLS \
2995 DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
2996 #define SWAP_CLUSTER_SPACE_COLS \
2997 DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
2998 #define SWAP_CLUSTER_COLS \
2999 max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
3001 static int setup_swap_map_and_extents(struct swap_info_struct *p,
3002 union swap_header *swap_header,
3003 unsigned char *swap_map,
3004 struct swap_cluster_info *cluster_info,
3005 unsigned long maxpages,
3009 unsigned int nr_good_pages;
3011 unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
3012 unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS;
3013 unsigned long i, idx;
3015 nr_good_pages = maxpages - 1; /* omit header page */
3017 cluster_list_init(&p->free_clusters);
3018 cluster_list_init(&p->discard_clusters);
3020 for (i = 0; i < swap_header->info.nr_badpages; i++) {
3021 unsigned int page_nr = swap_header->info.badpages[i];
3022 if (page_nr == 0 || page_nr > swap_header->info.last_page)
3024 if (page_nr < maxpages) {
3025 swap_map[page_nr] = SWAP_MAP_BAD;
3028 * Haven't marked the cluster free yet, no list
3029 * operation involved
3031 inc_cluster_info_page(p, cluster_info, page_nr);
3035 /* Haven't marked the cluster free yet, no list operation involved */
3036 for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
3037 inc_cluster_info_page(p, cluster_info, i);
3039 if (nr_good_pages) {
3040 swap_map[0] = SWAP_MAP_BAD;
3042 * Not mark the cluster free yet, no list
3043 * operation involved
3045 inc_cluster_info_page(p, cluster_info, 0);
3047 p->pages = nr_good_pages;
3048 nr_extents = setup_swap_extents(p, span);
3051 nr_good_pages = p->pages;
3053 if (!nr_good_pages) {
3054 pr_warn("Empty swap-file\n");
3063 * Reduce false cache line sharing between cluster_info and
3064 * sharing same address space.
3066 for (k = 0; k < SWAP_CLUSTER_COLS; k++) {
3067 j = (k + col) % SWAP_CLUSTER_COLS;
3068 for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) {
3069 idx = i * SWAP_CLUSTER_COLS + j;
3070 if (idx >= nr_clusters)
3072 if (cluster_count(&cluster_info[idx]))
3074 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
3075 cluster_list_add_tail(&p->free_clusters, cluster_info,
3083 * Helper to sys_swapon determining if a given swap
3084 * backing device queue supports DISCARD operations.
3086 static bool swap_discardable(struct swap_info_struct *si)
3088 struct request_queue *q = bdev_get_queue(si->bdev);
3090 if (!q || !blk_queue_discard(q))
3096 SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
3098 struct swap_info_struct *p;
3099 struct filename *name;
3100 struct file *swap_file = NULL;
3101 struct address_space *mapping;
3104 union swap_header *swap_header;
3107 unsigned long maxpages;
3108 unsigned char *swap_map = NULL;
3109 struct swap_cluster_info *cluster_info = NULL;
3110 unsigned long *frontswap_map = NULL;
3111 struct page *page = NULL;
3112 struct inode *inode = NULL;
3113 bool inced_nr_rotate_swap = false;
3115 if (swap_flags & ~SWAP_FLAGS_VALID)
3118 if (!capable(CAP_SYS_ADMIN))
3121 if (!swap_avail_heads)
3124 p = alloc_swap_info();
3128 INIT_WORK(&p->discard_work, swap_discard_work);
3130 name = getname(specialfile);
3132 error = PTR_ERR(name);
3136 swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
3137 if (IS_ERR(swap_file)) {
3138 error = PTR_ERR(swap_file);
3143 p->swap_file = swap_file;
3144 mapping = swap_file->f_mapping;
3145 inode = mapping->host;
3147 error = claim_swapfile(p, inode);
3148 if (unlikely(error))
3152 if (IS_SWAPFILE(inode)) {
3154 goto bad_swap_unlock_inode;
3158 * Read the swap header.
3160 if (!mapping->a_ops->readpage) {
3162 goto bad_swap_unlock_inode;
3164 page = read_mapping_page(mapping, 0, swap_file);
3166 error = PTR_ERR(page);
3167 goto bad_swap_unlock_inode;
3169 swap_header = kmap(page);
3171 maxpages = read_swap_header(p, swap_header, inode);
3172 if (unlikely(!maxpages)) {
3174 goto bad_swap_unlock_inode;
3177 /* OK, set up the swap map and apply the bad block list */
3178 swap_map = vzalloc(maxpages);
3181 goto bad_swap_unlock_inode;
3184 if (bdi_cap_stable_pages_required(inode_to_bdi(inode)))
3185 p->flags |= SWP_STABLE_WRITES;
3187 if (bdi_cap_synchronous_io(inode_to_bdi(inode)))
3188 p->flags |= SWP_SYNCHRONOUS_IO;
3190 if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) {
3192 unsigned long ci, nr_cluster;
3194 p->flags |= SWP_SOLIDSTATE;
3196 * select a random position to start with to help wear leveling
3199 p->cluster_next = 1 + (prandom_u32() % p->highest_bit);
3200 nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
3202 cluster_info = kvcalloc(nr_cluster, sizeof(*cluster_info),
3204 if (!cluster_info) {
3206 goto bad_swap_unlock_inode;
3209 for (ci = 0; ci < nr_cluster; ci++)
3210 spin_lock_init(&((cluster_info + ci)->lock));
3212 p->percpu_cluster = alloc_percpu(struct percpu_cluster);
3213 if (!p->percpu_cluster) {
3215 goto bad_swap_unlock_inode;
3217 for_each_possible_cpu(cpu) {
3218 struct percpu_cluster *cluster;
3219 cluster = per_cpu_ptr(p->percpu_cluster, cpu);
3220 cluster_set_null(&cluster->index);
3223 atomic_inc(&nr_rotate_swap);
3224 inced_nr_rotate_swap = true;
3227 error = swap_cgroup_swapon(p->type, maxpages);
3229 goto bad_swap_unlock_inode;
3231 nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
3232 cluster_info, maxpages, &span);
3233 if (unlikely(nr_extents < 0)) {
3235 goto bad_swap_unlock_inode;
3237 /* frontswap enabled? set up bit-per-page map for frontswap */
3238 if (IS_ENABLED(CONFIG_FRONTSWAP))
3239 frontswap_map = kvcalloc(BITS_TO_LONGS(maxpages),
3243 if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) {
3245 * When discard is enabled for swap with no particular
3246 * policy flagged, we set all swap discard flags here in
3247 * order to sustain backward compatibility with older
3248 * swapon(8) releases.
3250 p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
3254 * By flagging sys_swapon, a sysadmin can tell us to
3255 * either do single-time area discards only, or to just
3256 * perform discards for released swap page-clusters.
3257 * Now it's time to adjust the p->flags accordingly.
3259 if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
3260 p->flags &= ~SWP_PAGE_DISCARD;
3261 else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
3262 p->flags &= ~SWP_AREA_DISCARD;
3264 /* issue a swapon-time discard if it's still required */
3265 if (p->flags & SWP_AREA_DISCARD) {
3266 int err = discard_swap(p);
3268 pr_err("swapon: discard_swap(%p): %d\n",
3273 error = init_swap_address_space(p->type, maxpages);
3275 goto bad_swap_unlock_inode;
3278 * Flush any pending IO and dirty mappings before we start using this
3281 inode->i_flags |= S_SWAPFILE;
3282 error = inode_drain_writes(inode);
3284 inode->i_flags &= ~S_SWAPFILE;
3285 goto bad_swap_unlock_inode;
3288 mutex_lock(&swapon_mutex);
3290 if (swap_flags & SWAP_FLAG_PREFER)
3292 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
3293 enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
3295 pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
3296 p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
3297 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
3298 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
3299 (p->flags & SWP_DISCARDABLE) ? "D" : "",
3300 (p->flags & SWP_AREA_DISCARD) ? "s" : "",
3301 (p->flags & SWP_PAGE_DISCARD) ? "c" : "",
3302 (frontswap_map) ? "FS" : "");
3304 mutex_unlock(&swapon_mutex);
3305 atomic_inc(&proc_poll_event);
3306 wake_up_interruptible(&proc_poll_wait);
3310 bad_swap_unlock_inode:
3311 inode_unlock(inode);
3313 free_percpu(p->percpu_cluster);
3314 p->percpu_cluster = NULL;
3315 if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
3316 set_blocksize(p->bdev, p->old_block_size);
3317 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
3320 destroy_swap_extents(p);
3321 swap_cgroup_swapoff(p->type);
3322 spin_lock(&swap_lock);
3323 p->swap_file = NULL;
3325 spin_unlock(&swap_lock);
3327 kvfree(cluster_info);
3328 kvfree(frontswap_map);
3329 if (inced_nr_rotate_swap)
3330 atomic_dec(&nr_rotate_swap);
3332 filp_close(swap_file, NULL);
3334 if (page && !IS_ERR(page)) {
3341 inode_unlock(inode);
3343 enable_swap_slots_cache();
3347 void si_swapinfo(struct sysinfo *val)
3350 unsigned long nr_to_be_unused = 0;
3352 spin_lock(&swap_lock);
3353 for (type = 0; type < nr_swapfiles; type++) {
3354 struct swap_info_struct *si = swap_info[type];
3356 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
3357 nr_to_be_unused += si->inuse_pages;
3359 val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
3360 val->totalswap = total_swap_pages + nr_to_be_unused;
3361 spin_unlock(&swap_lock);
3365 * Verify that a swap entry is valid and increment its swap map count.
3367 * Returns error code in following case.
3369 * - swp_entry is invalid -> EINVAL
3370 * - swp_entry is migration entry -> EINVAL
3371 * - swap-cache reference is requested but there is already one. -> EEXIST
3372 * - swap-cache reference is requested but the entry is not used. -> ENOENT
3373 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
3375 static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
3377 struct swap_info_struct *p;
3378 struct swap_cluster_info *ci;
3379 unsigned long offset;
3380 unsigned char count;
3381 unsigned char has_cache;
3384 p = get_swap_device(entry);
3388 offset = swp_offset(entry);
3389 ci = lock_cluster_or_swap_info(p, offset);
3391 count = p->swap_map[offset];
3394 * swapin_readahead() doesn't check if a swap entry is valid, so the
3395 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
3397 if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
3402 has_cache = count & SWAP_HAS_CACHE;
3403 count &= ~SWAP_HAS_CACHE;
3406 if (usage == SWAP_HAS_CACHE) {
3408 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
3409 if (!has_cache && count)
3410 has_cache = SWAP_HAS_CACHE;
3411 else if (has_cache) /* someone else added cache */
3413 else /* no users remaining */
3416 } else if (count || has_cache) {
3418 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
3420 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
3422 else if (swap_count_continued(p, offset, count))
3423 count = COUNT_CONTINUED;
3427 err = -ENOENT; /* unused swap entry */
3429 p->swap_map[offset] = count | has_cache;
3432 unlock_cluster_or_swap_info(p, ci);
3440 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3441 * (in which case its reference count is never incremented).
3443 void swap_shmem_alloc(swp_entry_t entry)
3445 __swap_duplicate(entry, SWAP_MAP_SHMEM);
3449 * Increase reference count of swap entry by 1.
3450 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3451 * but could not be atomically allocated. Returns 0, just as if it succeeded,
3452 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3453 * might occur if a page table entry has got corrupted.
3455 int swap_duplicate(swp_entry_t entry)
3459 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
3460 err = add_swap_count_continuation(entry, GFP_ATOMIC);
3465 * @entry: swap entry for which we allocate swap cache.
3467 * Called when allocating swap cache for existing swap entry,
3468 * This can return error codes. Returns 0 at success.
3469 * -EEXIST means there is a swap cache.
3470 * Note: return code is different from swap_duplicate().
3472 int swapcache_prepare(swp_entry_t entry)
3474 return __swap_duplicate(entry, SWAP_HAS_CACHE);
3477 struct swap_info_struct *swp_swap_info(swp_entry_t entry)
3479 return swap_type_to_swap_info(swp_type(entry));
3482 struct swap_info_struct *page_swap_info(struct page *page)
3484 swp_entry_t entry = { .val = page_private(page) };
3485 return swp_swap_info(entry);
3489 * out-of-line __page_file_ methods to avoid include hell.
3491 struct address_space *__page_file_mapping(struct page *page)
3493 return page_swap_info(page)->swap_file->f_mapping;
3495 EXPORT_SYMBOL_GPL(__page_file_mapping);
3497 pgoff_t __page_file_index(struct page *page)
3499 swp_entry_t swap = { .val = page_private(page) };
3500 return swp_offset(swap);
3502 EXPORT_SYMBOL_GPL(__page_file_index);
3505 * add_swap_count_continuation - called when a swap count is duplicated
3506 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3507 * page of the original vmalloc'ed swap_map, to hold the continuation count
3508 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
3509 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3511 * These continuation pages are seldom referenced: the common paths all work
3512 * on the original swap_map, only referring to a continuation page when the
3513 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3515 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3516 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3517 * can be called after dropping locks.
3519 int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
3521 struct swap_info_struct *si;
3522 struct swap_cluster_info *ci;
3525 struct page *list_page;
3527 unsigned char count;
3531 * When debugging, it's easier to use __GFP_ZERO here; but it's better
3532 * for latency not to zero a page while GFP_ATOMIC and holding locks.
3534 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
3536 si = get_swap_device(entry);
3539 * An acceptable race has occurred since the failing
3540 * __swap_duplicate(): the swap device may be swapoff
3544 spin_lock(&si->lock);
3546 offset = swp_offset(entry);
3548 ci = lock_cluster(si, offset);
3550 count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
3552 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
3554 * The higher the swap count, the more likely it is that tasks
3555 * will race to add swap count continuation: we need to avoid
3556 * over-provisioning.
3567 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
3568 * no architecture is using highmem pages for kernel page tables: so it
3569 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
3571 head = vmalloc_to_page(si->swap_map + offset);
3572 offset &= ~PAGE_MASK;
3574 spin_lock(&si->cont_lock);
3576 * Page allocation does not initialize the page's lru field,
3577 * but it does always reset its private field.
3579 if (!page_private(head)) {
3580 BUG_ON(count & COUNT_CONTINUED);
3581 INIT_LIST_HEAD(&head->lru);
3582 set_page_private(head, SWP_CONTINUED);
3583 si->flags |= SWP_CONTINUED;
3586 list_for_each_entry(list_page, &head->lru, lru) {
3590 * If the previous map said no continuation, but we've found
3591 * a continuation page, free our allocation and use this one.
3593 if (!(count & COUNT_CONTINUED))
3594 goto out_unlock_cont;
3596 map = kmap_atomic(list_page) + offset;
3601 * If this continuation count now has some space in it,
3602 * free our allocation and use this one.
3604 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
3605 goto out_unlock_cont;
3608 list_add_tail(&page->lru, &head->lru);
3609 page = NULL; /* now it's attached, don't free it */
3611 spin_unlock(&si->cont_lock);
3614 spin_unlock(&si->lock);
3615 put_swap_device(si);
3623 * swap_count_continued - when the original swap_map count is incremented
3624 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3625 * into, carry if so, or else fail until a new continuation page is allocated;
3626 * when the original swap_map count is decremented from 0 with continuation,
3627 * borrow from the continuation and report whether it still holds more.
3628 * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3631 static bool swap_count_continued(struct swap_info_struct *si,
3632 pgoff_t offset, unsigned char count)
3639 head = vmalloc_to_page(si->swap_map + offset);
3640 if (page_private(head) != SWP_CONTINUED) {
3641 BUG_ON(count & COUNT_CONTINUED);
3642 return false; /* need to add count continuation */
3645 spin_lock(&si->cont_lock);
3646 offset &= ~PAGE_MASK;
3647 page = list_next_entry(head, lru);
3648 map = kmap_atomic(page) + offset;
3650 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
3651 goto init_map; /* jump over SWAP_CONT_MAX checks */
3653 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
3655 * Think of how you add 1 to 999
3657 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
3659 page = list_next_entry(page, lru);
3660 BUG_ON(page == head);
3661 map = kmap_atomic(page) + offset;
3663 if (*map == SWAP_CONT_MAX) {
3665 page = list_next_entry(page, lru);
3667 ret = false; /* add count continuation */
3670 map = kmap_atomic(page) + offset;
3671 init_map: *map = 0; /* we didn't zero the page */
3675 while ((page = list_prev_entry(page, lru)) != head) {
3676 map = kmap_atomic(page) + offset;
3677 *map = COUNT_CONTINUED;
3680 ret = true; /* incremented */
3682 } else { /* decrementing */
3684 * Think of how you subtract 1 from 1000
3686 BUG_ON(count != COUNT_CONTINUED);
3687 while (*map == COUNT_CONTINUED) {
3689 page = list_next_entry(page, lru);
3690 BUG_ON(page == head);
3691 map = kmap_atomic(page) + offset;
3698 while ((page = list_prev_entry(page, lru)) != head) {
3699 map = kmap_atomic(page) + offset;
3700 *map = SWAP_CONT_MAX | count;
3701 count = COUNT_CONTINUED;
3704 ret = count == COUNT_CONTINUED;
3707 spin_unlock(&si->cont_lock);
3712 * free_swap_count_continuations - swapoff free all the continuation pages
3713 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3715 static void free_swap_count_continuations(struct swap_info_struct *si)
3719 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
3721 head = vmalloc_to_page(si->swap_map + offset);
3722 if (page_private(head)) {
3723 struct page *page, *next;
3725 list_for_each_entry_safe(page, next, &head->lru, lru) {
3726 list_del(&page->lru);
3733 #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
3734 void mem_cgroup_throttle_swaprate(struct mem_cgroup *memcg, int node,
3737 struct swap_info_struct *si, *next;
3738 if (!(gfp_mask & __GFP_IO) || !memcg)
3741 if (!blk_cgroup_congested())
3745 * We've already scheduled a throttle, avoid taking the global swap
3748 if (current->throttle_queue)
3751 spin_lock(&swap_avail_lock);
3752 plist_for_each_entry_safe(si, next, &swap_avail_heads[node],
3753 avail_lists[node]) {
3755 blkcg_schedule_throttle(bdev_get_queue(si->bdev),
3760 spin_unlock(&swap_avail_lock);
3764 static int __init swapfile_init(void)
3768 swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head),
3770 if (!swap_avail_heads) {
3771 pr_emerg("Not enough memory for swap heads, swap is disabled\n");
3776 plist_head_init(&swap_avail_heads[nid]);
3780 subsys_initcall(swapfile_init);