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))
768 if (unlikely(!si->cluster_nr--)) {
769 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
770 si->cluster_nr = SWAPFILE_CLUSTER - 1;
774 spin_unlock(&si->lock);
777 * If seek is expensive, start searching for new cluster from
778 * start of partition, to minimize the span of allocated swap.
779 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
780 * case, just handled by scan_swap_map_try_ssd_cluster() above.
782 scan_base = offset = si->lowest_bit;
783 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
785 /* Locate the first empty (unaligned) cluster */
786 for (; last_in_cluster <= si->highest_bit; offset++) {
787 if (si->swap_map[offset])
788 last_in_cluster = offset + SWAPFILE_CLUSTER;
789 else if (offset == last_in_cluster) {
790 spin_lock(&si->lock);
791 offset -= SWAPFILE_CLUSTER - 1;
792 si->cluster_next = offset;
793 si->cluster_nr = SWAPFILE_CLUSTER - 1;
796 if (unlikely(--latency_ration < 0)) {
798 latency_ration = LATENCY_LIMIT;
803 spin_lock(&si->lock);
804 si->cluster_nr = SWAPFILE_CLUSTER - 1;
808 if (si->cluster_info) {
809 while (scan_swap_map_ssd_cluster_conflict(si, offset)) {
810 /* take a break if we already got some slots */
813 if (!scan_swap_map_try_ssd_cluster(si, &offset,
818 if (!(si->flags & SWP_WRITEOK))
820 if (!si->highest_bit)
822 if (offset > si->highest_bit)
823 scan_base = offset = si->lowest_bit;
825 ci = lock_cluster(si, offset);
826 /* reuse swap entry of cache-only swap if not busy. */
827 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
830 spin_unlock(&si->lock);
831 swap_was_freed = __try_to_reclaim_swap(si, offset, TTRS_ANYWAY);
832 spin_lock(&si->lock);
833 /* entry was freed successfully, try to use this again */
836 goto scan; /* check next one */
839 if (si->swap_map[offset]) {
846 si->swap_map[offset] = usage;
847 inc_cluster_info_page(si, si->cluster_info, offset);
850 swap_range_alloc(si, offset, 1);
851 si->cluster_next = offset + 1;
852 slots[n_ret++] = swp_entry(si->type, offset);
854 /* got enough slots or reach max slots? */
855 if ((n_ret == nr) || (offset >= si->highest_bit))
858 /* search for next available slot */
860 /* time to take a break? */
861 if (unlikely(--latency_ration < 0)) {
864 spin_unlock(&si->lock);
866 spin_lock(&si->lock);
867 latency_ration = LATENCY_LIMIT;
870 /* try to get more slots in cluster */
871 if (si->cluster_info) {
872 if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
878 /* non-ssd case, still more slots in cluster? */
879 if (si->cluster_nr && !si->swap_map[++offset]) {
885 si->flags -= SWP_SCANNING;
889 spin_unlock(&si->lock);
890 while (++offset <= si->highest_bit) {
891 if (!si->swap_map[offset]) {
892 spin_lock(&si->lock);
895 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
896 spin_lock(&si->lock);
899 if (unlikely(--latency_ration < 0)) {
901 latency_ration = LATENCY_LIMIT;
904 offset = si->lowest_bit;
905 while (offset < scan_base) {
906 if (!si->swap_map[offset]) {
907 spin_lock(&si->lock);
910 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
911 spin_lock(&si->lock);
914 if (unlikely(--latency_ration < 0)) {
916 latency_ration = LATENCY_LIMIT;
920 spin_lock(&si->lock);
923 si->flags -= SWP_SCANNING;
927 static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot)
930 struct swap_cluster_info *ci;
931 unsigned long offset, i;
935 * Should not even be attempting cluster allocations when huge
936 * page swap is disabled. Warn and fail the allocation.
938 if (!IS_ENABLED(CONFIG_THP_SWAP)) {
943 if (cluster_list_empty(&si->free_clusters))
946 idx = cluster_list_first(&si->free_clusters);
947 offset = idx * SWAPFILE_CLUSTER;
948 ci = lock_cluster(si, offset);
949 alloc_cluster(si, idx);
950 cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE);
952 map = si->swap_map + offset;
953 for (i = 0; i < SWAPFILE_CLUSTER; i++)
954 map[i] = SWAP_HAS_CACHE;
956 swap_range_alloc(si, offset, SWAPFILE_CLUSTER);
957 *slot = swp_entry(si->type, offset);
962 static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx)
964 unsigned long offset = idx * SWAPFILE_CLUSTER;
965 struct swap_cluster_info *ci;
967 ci = lock_cluster(si, offset);
968 memset(si->swap_map + offset, 0, SWAPFILE_CLUSTER);
969 cluster_set_count_flag(ci, 0, 0);
970 free_cluster(si, idx);
972 swap_range_free(si, offset, SWAPFILE_CLUSTER);
975 static unsigned long scan_swap_map(struct swap_info_struct *si,
981 n_ret = scan_swap_map_slots(si, usage, 1, &entry);
984 return swp_offset(entry);
990 int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size)
992 unsigned long size = swap_entry_size(entry_size);
993 struct swap_info_struct *si, *next;
998 /* Only single cluster request supported */
999 WARN_ON_ONCE(n_goal > 1 && size == SWAPFILE_CLUSTER);
1001 avail_pgs = atomic_long_read(&nr_swap_pages) / size;
1005 if (n_goal > SWAP_BATCH)
1006 n_goal = SWAP_BATCH;
1008 if (n_goal > avail_pgs)
1011 atomic_long_sub(n_goal * size, &nr_swap_pages);
1013 spin_lock(&swap_avail_lock);
1016 node = numa_node_id();
1017 plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) {
1018 /* requeue si to after same-priority siblings */
1019 plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]);
1020 spin_unlock(&swap_avail_lock);
1021 spin_lock(&si->lock);
1022 if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) {
1023 spin_lock(&swap_avail_lock);
1024 if (plist_node_empty(&si->avail_lists[node])) {
1025 spin_unlock(&si->lock);
1028 WARN(!si->highest_bit,
1029 "swap_info %d in list but !highest_bit\n",
1031 WARN(!(si->flags & SWP_WRITEOK),
1032 "swap_info %d in list but !SWP_WRITEOK\n",
1034 __del_from_avail_list(si);
1035 spin_unlock(&si->lock);
1038 if (size == SWAPFILE_CLUSTER) {
1039 if (!(si->flags & SWP_FS))
1040 n_ret = swap_alloc_cluster(si, swp_entries);
1042 n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE,
1043 n_goal, swp_entries);
1044 spin_unlock(&si->lock);
1045 if (n_ret || size == SWAPFILE_CLUSTER)
1047 pr_debug("scan_swap_map of si %d failed to find offset\n",
1050 spin_lock(&swap_avail_lock);
1053 * if we got here, it's likely that si was almost full before,
1054 * and since scan_swap_map() can drop the si->lock, multiple
1055 * callers probably all tried to get a page from the same si
1056 * and it filled up before we could get one; or, the si filled
1057 * up between us dropping swap_avail_lock and taking si->lock.
1058 * Since we dropped the swap_avail_lock, the swap_avail_head
1059 * list may have been modified; so if next is still in the
1060 * swap_avail_head list then try it, otherwise start over
1061 * if we have not gotten any slots.
1063 if (plist_node_empty(&next->avail_lists[node]))
1067 spin_unlock(&swap_avail_lock);
1071 atomic_long_add((long)(n_goal - n_ret) * size,
1077 /* The only caller of this function is now suspend routine */
1078 swp_entry_t get_swap_page_of_type(int type)
1080 struct swap_info_struct *si = swap_type_to_swap_info(type);
1086 spin_lock(&si->lock);
1087 if (si->flags & SWP_WRITEOK) {
1088 atomic_long_dec(&nr_swap_pages);
1089 /* This is called for allocating swap entry, not cache */
1090 offset = scan_swap_map(si, 1);
1092 spin_unlock(&si->lock);
1093 return swp_entry(type, offset);
1095 atomic_long_inc(&nr_swap_pages);
1097 spin_unlock(&si->lock);
1099 return (swp_entry_t) {0};
1102 static struct swap_info_struct *__swap_info_get(swp_entry_t entry)
1104 struct swap_info_struct *p;
1105 unsigned long offset;
1109 p = swp_swap_info(entry);
1112 if (!(p->flags & SWP_USED))
1114 offset = swp_offset(entry);
1115 if (offset >= p->max)
1120 pr_err("swap_info_get: %s%08lx\n", Bad_offset, entry.val);
1123 pr_err("swap_info_get: %s%08lx\n", Unused_file, entry.val);
1126 pr_err("swap_info_get: %s%08lx\n", Bad_file, entry.val);
1131 static struct swap_info_struct *_swap_info_get(swp_entry_t entry)
1133 struct swap_info_struct *p;
1135 p = __swap_info_get(entry);
1138 if (!p->swap_map[swp_offset(entry)])
1143 pr_err("swap_info_get: %s%08lx\n", Unused_offset, entry.val);
1149 static struct swap_info_struct *swap_info_get(swp_entry_t entry)
1151 struct swap_info_struct *p;
1153 p = _swap_info_get(entry);
1155 spin_lock(&p->lock);
1159 static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry,
1160 struct swap_info_struct *q)
1162 struct swap_info_struct *p;
1164 p = _swap_info_get(entry);
1168 spin_unlock(&q->lock);
1170 spin_lock(&p->lock);
1175 static unsigned char __swap_entry_free_locked(struct swap_info_struct *p,
1176 unsigned long offset,
1177 unsigned char usage)
1179 unsigned char count;
1180 unsigned char has_cache;
1182 count = p->swap_map[offset];
1184 has_cache = count & SWAP_HAS_CACHE;
1185 count &= ~SWAP_HAS_CACHE;
1187 if (usage == SWAP_HAS_CACHE) {
1188 VM_BUG_ON(!has_cache);
1190 } else if (count == SWAP_MAP_SHMEM) {
1192 * Or we could insist on shmem.c using a special
1193 * swap_shmem_free() and free_shmem_swap_and_cache()...
1196 } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
1197 if (count == COUNT_CONTINUED) {
1198 if (swap_count_continued(p, offset, count))
1199 count = SWAP_MAP_MAX | COUNT_CONTINUED;
1201 count = SWAP_MAP_MAX;
1206 usage = count | has_cache;
1207 p->swap_map[offset] = usage ? : SWAP_HAS_CACHE;
1213 * Check whether swap entry is valid in the swap device. If so,
1214 * return pointer to swap_info_struct, and keep the swap entry valid
1215 * via preventing the swap device from being swapoff, until
1216 * put_swap_device() is called. Otherwise return NULL.
1218 * The entirety of the RCU read critical section must come before the
1219 * return from or after the call to synchronize_rcu() in
1220 * enable_swap_info() or swapoff(). So if "si->flags & SWP_VALID" is
1221 * true, the si->map, si->cluster_info, etc. must be valid in the
1224 * Notice that swapoff or swapoff+swapon can still happen before the
1225 * rcu_read_lock() in get_swap_device() or after the rcu_read_unlock()
1226 * in put_swap_device() if there isn't any other way to prevent
1227 * swapoff, such as page lock, page table lock, etc. The caller must
1228 * be prepared for that. For example, the following situation is
1233 * ... swapoff+swapon
1234 * __read_swap_cache_async()
1235 * swapcache_prepare()
1236 * __swap_duplicate()
1238 * // verify PTE not changed
1240 * In __swap_duplicate(), the swap_map need to be checked before
1241 * changing partly because the specified swap entry may be for another
1242 * swap device which has been swapoff. And in do_swap_page(), after
1243 * the page is read from the swap device, the PTE is verified not
1244 * changed with the page table locked to check whether the swap device
1245 * has been swapoff or swapoff+swapon.
1247 struct swap_info_struct *get_swap_device(swp_entry_t entry)
1249 struct swap_info_struct *si;
1250 unsigned long offset;
1254 si = swp_swap_info(entry);
1259 if (!(si->flags & SWP_VALID))
1261 offset = swp_offset(entry);
1262 if (offset >= si->max)
1267 pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val);
1275 static unsigned char __swap_entry_free(struct swap_info_struct *p,
1276 swp_entry_t entry, unsigned char usage)
1278 struct swap_cluster_info *ci;
1279 unsigned long offset = swp_offset(entry);
1281 ci = lock_cluster_or_swap_info(p, offset);
1282 usage = __swap_entry_free_locked(p, offset, usage);
1283 unlock_cluster_or_swap_info(p, ci);
1285 free_swap_slot(entry);
1290 static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry)
1292 struct swap_cluster_info *ci;
1293 unsigned long offset = swp_offset(entry);
1294 unsigned char count;
1296 ci = lock_cluster(p, offset);
1297 count = p->swap_map[offset];
1298 VM_BUG_ON(count != SWAP_HAS_CACHE);
1299 p->swap_map[offset] = 0;
1300 dec_cluster_info_page(p, p->cluster_info, offset);
1303 mem_cgroup_uncharge_swap(entry, 1);
1304 swap_range_free(p, offset, 1);
1308 * Caller has made sure that the swap device corresponding to entry
1309 * is still around or has not been recycled.
1311 void swap_free(swp_entry_t entry)
1313 struct swap_info_struct *p;
1315 p = _swap_info_get(entry);
1317 __swap_entry_free(p, entry, 1);
1321 * Called after dropping swapcache to decrease refcnt to swap entries.
1323 void put_swap_page(struct page *page, swp_entry_t entry)
1325 unsigned long offset = swp_offset(entry);
1326 unsigned long idx = offset / SWAPFILE_CLUSTER;
1327 struct swap_cluster_info *ci;
1328 struct swap_info_struct *si;
1330 unsigned int i, free_entries = 0;
1332 int size = swap_entry_size(hpage_nr_pages(page));
1334 si = _swap_info_get(entry);
1338 ci = lock_cluster_or_swap_info(si, offset);
1339 if (size == SWAPFILE_CLUSTER) {
1340 VM_BUG_ON(!cluster_is_huge(ci));
1341 map = si->swap_map + offset;
1342 for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1344 VM_BUG_ON(!(val & SWAP_HAS_CACHE));
1345 if (val == SWAP_HAS_CACHE)
1348 cluster_clear_huge(ci);
1349 if (free_entries == SWAPFILE_CLUSTER) {
1350 unlock_cluster_or_swap_info(si, ci);
1351 spin_lock(&si->lock);
1352 mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER);
1353 swap_free_cluster(si, idx);
1354 spin_unlock(&si->lock);
1358 for (i = 0; i < size; i++, entry.val++) {
1359 if (!__swap_entry_free_locked(si, offset + i, SWAP_HAS_CACHE)) {
1360 unlock_cluster_or_swap_info(si, ci);
1361 free_swap_slot(entry);
1364 lock_cluster_or_swap_info(si, offset);
1367 unlock_cluster_or_swap_info(si, ci);
1370 #ifdef CONFIG_THP_SWAP
1371 int split_swap_cluster(swp_entry_t entry)
1373 struct swap_info_struct *si;
1374 struct swap_cluster_info *ci;
1375 unsigned long offset = swp_offset(entry);
1377 si = _swap_info_get(entry);
1380 ci = lock_cluster(si, offset);
1381 cluster_clear_huge(ci);
1387 static int swp_entry_cmp(const void *ent1, const void *ent2)
1389 const swp_entry_t *e1 = ent1, *e2 = ent2;
1391 return (int)swp_type(*e1) - (int)swp_type(*e2);
1394 void swapcache_free_entries(swp_entry_t *entries, int n)
1396 struct swap_info_struct *p, *prev;
1406 * Sort swap entries by swap device, so each lock is only taken once.
1407 * nr_swapfiles isn't absolutely correct, but the overhead of sort() is
1408 * so low that it isn't necessary to optimize further.
1410 if (nr_swapfiles > 1)
1411 sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL);
1412 for (i = 0; i < n; ++i) {
1413 p = swap_info_get_cont(entries[i], prev);
1415 swap_entry_free(p, entries[i]);
1419 spin_unlock(&p->lock);
1423 * How many references to page are currently swapped out?
1424 * This does not give an exact answer when swap count is continued,
1425 * but does include the high COUNT_CONTINUED flag to allow for that.
1427 int page_swapcount(struct page *page)
1430 struct swap_info_struct *p;
1431 struct swap_cluster_info *ci;
1433 unsigned long offset;
1435 entry.val = page_private(page);
1436 p = _swap_info_get(entry);
1438 offset = swp_offset(entry);
1439 ci = lock_cluster_or_swap_info(p, offset);
1440 count = swap_count(p->swap_map[offset]);
1441 unlock_cluster_or_swap_info(p, ci);
1446 int __swap_count(swp_entry_t entry)
1448 struct swap_info_struct *si;
1449 pgoff_t offset = swp_offset(entry);
1452 si = get_swap_device(entry);
1454 count = swap_count(si->swap_map[offset]);
1455 put_swap_device(si);
1460 static int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry)
1463 pgoff_t offset = swp_offset(entry);
1464 struct swap_cluster_info *ci;
1466 ci = lock_cluster_or_swap_info(si, offset);
1467 count = swap_count(si->swap_map[offset]);
1468 unlock_cluster_or_swap_info(si, ci);
1473 * How many references to @entry are currently swapped out?
1474 * This does not give an exact answer when swap count is continued,
1475 * but does include the high COUNT_CONTINUED flag to allow for that.
1477 int __swp_swapcount(swp_entry_t entry)
1480 struct swap_info_struct *si;
1482 si = get_swap_device(entry);
1484 count = swap_swapcount(si, entry);
1485 put_swap_device(si);
1491 * How many references to @entry are currently swapped out?
1492 * This considers COUNT_CONTINUED so it returns exact answer.
1494 int swp_swapcount(swp_entry_t entry)
1496 int count, tmp_count, n;
1497 struct swap_info_struct *p;
1498 struct swap_cluster_info *ci;
1503 p = _swap_info_get(entry);
1507 offset = swp_offset(entry);
1509 ci = lock_cluster_or_swap_info(p, offset);
1511 count = swap_count(p->swap_map[offset]);
1512 if (!(count & COUNT_CONTINUED))
1515 count &= ~COUNT_CONTINUED;
1516 n = SWAP_MAP_MAX + 1;
1518 page = vmalloc_to_page(p->swap_map + offset);
1519 offset &= ~PAGE_MASK;
1520 VM_BUG_ON(page_private(page) != SWP_CONTINUED);
1523 page = list_next_entry(page, lru);
1524 map = kmap_atomic(page);
1525 tmp_count = map[offset];
1528 count += (tmp_count & ~COUNT_CONTINUED) * n;
1529 n *= (SWAP_CONT_MAX + 1);
1530 } while (tmp_count & COUNT_CONTINUED);
1532 unlock_cluster_or_swap_info(p, ci);
1536 static bool swap_page_trans_huge_swapped(struct swap_info_struct *si,
1539 struct swap_cluster_info *ci;
1540 unsigned char *map = si->swap_map;
1541 unsigned long roffset = swp_offset(entry);
1542 unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER);
1546 ci = lock_cluster_or_swap_info(si, offset);
1547 if (!ci || !cluster_is_huge(ci)) {
1548 if (swap_count(map[roffset]))
1552 for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1553 if (swap_count(map[offset + i])) {
1559 unlock_cluster_or_swap_info(si, ci);
1563 static bool page_swapped(struct page *page)
1566 struct swap_info_struct *si;
1568 if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page)))
1569 return page_swapcount(page) != 0;
1571 page = compound_head(page);
1572 entry.val = page_private(page);
1573 si = _swap_info_get(entry);
1575 return swap_page_trans_huge_swapped(si, entry);
1579 static int page_trans_huge_map_swapcount(struct page *page, int *total_mapcount,
1580 int *total_swapcount)
1582 int i, map_swapcount, _total_mapcount, _total_swapcount;
1583 unsigned long offset = 0;
1584 struct swap_info_struct *si;
1585 struct swap_cluster_info *ci = NULL;
1586 unsigned char *map = NULL;
1587 int mapcount, swapcount = 0;
1589 /* hugetlbfs shouldn't call it */
1590 VM_BUG_ON_PAGE(PageHuge(page), page);
1592 if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page))) {
1593 mapcount = page_trans_huge_mapcount(page, total_mapcount);
1594 if (PageSwapCache(page))
1595 swapcount = page_swapcount(page);
1596 if (total_swapcount)
1597 *total_swapcount = swapcount;
1598 return mapcount + swapcount;
1601 page = compound_head(page);
1603 _total_mapcount = _total_swapcount = map_swapcount = 0;
1604 if (PageSwapCache(page)) {
1607 entry.val = page_private(page);
1608 si = _swap_info_get(entry);
1611 offset = swp_offset(entry);
1615 ci = lock_cluster(si, offset);
1616 for (i = 0; i < HPAGE_PMD_NR; i++) {
1617 mapcount = atomic_read(&page[i]._mapcount) + 1;
1618 _total_mapcount += mapcount;
1620 swapcount = swap_count(map[offset + i]);
1621 _total_swapcount += swapcount;
1623 map_swapcount = max(map_swapcount, mapcount + swapcount);
1626 if (PageDoubleMap(page)) {
1628 _total_mapcount -= HPAGE_PMD_NR;
1630 mapcount = compound_mapcount(page);
1631 map_swapcount += mapcount;
1632 _total_mapcount += mapcount;
1634 *total_mapcount = _total_mapcount;
1635 if (total_swapcount)
1636 *total_swapcount = _total_swapcount;
1638 return map_swapcount;
1642 * We can write to an anon page without COW if there are no other references
1643 * to it. And as a side-effect, free up its swap: because the old content
1644 * on disk will never be read, and seeking back there to write new content
1645 * later would only waste time away from clustering.
1647 * NOTE: total_map_swapcount should not be relied upon by the caller if
1648 * reuse_swap_page() returns false, but it may be always overwritten
1649 * (see the other implementation for CONFIG_SWAP=n).
1651 bool reuse_swap_page(struct page *page, int *total_map_swapcount)
1653 int count, total_mapcount, total_swapcount;
1655 VM_BUG_ON_PAGE(!PageLocked(page), page);
1656 if (unlikely(PageKsm(page)))
1658 count = page_trans_huge_map_swapcount(page, &total_mapcount,
1660 if (total_map_swapcount)
1661 *total_map_swapcount = total_mapcount + total_swapcount;
1662 if (count == 1 && PageSwapCache(page) &&
1663 (likely(!PageTransCompound(page)) ||
1664 /* The remaining swap count will be freed soon */
1665 total_swapcount == page_swapcount(page))) {
1666 if (!PageWriteback(page)) {
1667 page = compound_head(page);
1668 delete_from_swap_cache(page);
1672 struct swap_info_struct *p;
1674 entry.val = page_private(page);
1675 p = swap_info_get(entry);
1676 if (p->flags & SWP_STABLE_WRITES) {
1677 spin_unlock(&p->lock);
1680 spin_unlock(&p->lock);
1688 * If swap is getting full, or if there are no more mappings of this page,
1689 * then try_to_free_swap is called to free its swap space.
1691 int try_to_free_swap(struct page *page)
1693 VM_BUG_ON_PAGE(!PageLocked(page), page);
1695 if (!PageSwapCache(page))
1697 if (PageWriteback(page))
1699 if (page_swapped(page))
1703 * Once hibernation has begun to create its image of memory,
1704 * there's a danger that one of the calls to try_to_free_swap()
1705 * - most probably a call from __try_to_reclaim_swap() while
1706 * hibernation is allocating its own swap pages for the image,
1707 * but conceivably even a call from memory reclaim - will free
1708 * the swap from a page which has already been recorded in the
1709 * image as a clean swapcache page, and then reuse its swap for
1710 * another page of the image. On waking from hibernation, the
1711 * original page might be freed under memory pressure, then
1712 * later read back in from swap, now with the wrong data.
1714 * Hibernation suspends storage while it is writing the image
1715 * to disk so check that here.
1717 if (pm_suspended_storage())
1720 page = compound_head(page);
1721 delete_from_swap_cache(page);
1727 * Free the swap entry like above, but also try to
1728 * free the page cache entry if it is the last user.
1730 int free_swap_and_cache(swp_entry_t entry)
1732 struct swap_info_struct *p;
1733 unsigned char count;
1735 if (non_swap_entry(entry))
1738 p = _swap_info_get(entry);
1740 count = __swap_entry_free(p, entry, 1);
1741 if (count == SWAP_HAS_CACHE &&
1742 !swap_page_trans_huge_swapped(p, entry))
1743 __try_to_reclaim_swap(p, swp_offset(entry),
1744 TTRS_UNMAPPED | TTRS_FULL);
1749 #ifdef CONFIG_HIBERNATION
1751 * Find the swap type that corresponds to given device (if any).
1753 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1754 * from 0, in which the swap header is expected to be located.
1756 * This is needed for the suspend to disk (aka swsusp).
1758 int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
1760 struct block_device *bdev = NULL;
1764 bdev = bdget(device);
1766 spin_lock(&swap_lock);
1767 for (type = 0; type < nr_swapfiles; type++) {
1768 struct swap_info_struct *sis = swap_info[type];
1770 if (!(sis->flags & SWP_WRITEOK))
1775 *bdev_p = bdgrab(sis->bdev);
1777 spin_unlock(&swap_lock);
1780 if (bdev == sis->bdev) {
1781 struct swap_extent *se = first_se(sis);
1783 if (se->start_block == offset) {
1785 *bdev_p = bdgrab(sis->bdev);
1787 spin_unlock(&swap_lock);
1793 spin_unlock(&swap_lock);
1801 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1802 * corresponding to given index in swap_info (swap type).
1804 sector_t swapdev_block(int type, pgoff_t offset)
1806 struct block_device *bdev;
1807 struct swap_info_struct *si = swap_type_to_swap_info(type);
1809 if (!si || !(si->flags & SWP_WRITEOK))
1811 return map_swap_entry(swp_entry(type, offset), &bdev);
1815 * Return either the total number of swap pages of given type, or the number
1816 * of free pages of that type (depending on @free)
1818 * This is needed for software suspend
1820 unsigned int count_swap_pages(int type, int free)
1824 spin_lock(&swap_lock);
1825 if ((unsigned int)type < nr_swapfiles) {
1826 struct swap_info_struct *sis = swap_info[type];
1828 spin_lock(&sis->lock);
1829 if (sis->flags & SWP_WRITEOK) {
1832 n -= sis->inuse_pages;
1834 spin_unlock(&sis->lock);
1836 spin_unlock(&swap_lock);
1839 #endif /* CONFIG_HIBERNATION */
1841 static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte)
1843 return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte);
1847 * No need to decide whether this PTE shares the swap entry with others,
1848 * just let do_wp_page work it out if a write is requested later - to
1849 * force COW, vm_page_prot omits write permission from any private vma.
1851 static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
1852 unsigned long addr, swp_entry_t entry, struct page *page)
1854 struct page *swapcache;
1855 struct mem_cgroup *memcg;
1861 page = ksm_might_need_to_copy(page, vma, addr);
1862 if (unlikely(!page))
1865 if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL,
1871 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1872 if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) {
1873 mem_cgroup_cancel_charge(page, memcg, false);
1878 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1879 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
1881 set_pte_at(vma->vm_mm, addr, pte,
1882 pte_mkold(mk_pte(page, vma->vm_page_prot)));
1883 if (page == swapcache) {
1884 page_add_anon_rmap(page, vma, addr, false);
1885 mem_cgroup_commit_charge(page, memcg, true, false);
1886 } else { /* ksm created a completely new copy */
1887 page_add_new_anon_rmap(page, vma, addr, false);
1888 mem_cgroup_commit_charge(page, memcg, false, false);
1889 lru_cache_add_active_or_unevictable(page, vma);
1893 * Move the page to the active list so it is not
1894 * immediately swapped out again after swapon.
1896 activate_page(page);
1898 pte_unmap_unlock(pte, ptl);
1900 if (page != swapcache) {
1907 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1908 unsigned long addr, unsigned long end,
1909 unsigned int type, bool frontswap,
1910 unsigned long *fs_pages_to_unuse)
1915 struct swap_info_struct *si;
1916 unsigned long offset;
1918 volatile unsigned char *swap_map;
1920 si = swap_info[type];
1921 pte = pte_offset_map(pmd, addr);
1923 struct vm_fault vmf;
1925 if (!is_swap_pte(*pte))
1928 entry = pte_to_swp_entry(*pte);
1929 if (swp_type(entry) != type)
1932 offset = swp_offset(entry);
1933 if (frontswap && !frontswap_test(si, offset))
1937 swap_map = &si->swap_map[offset];
1938 page = lookup_swap_cache(entry, vma, addr);
1943 page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE,
1947 if (*swap_map == 0 || *swap_map == SWAP_MAP_BAD)
1953 wait_on_page_writeback(page);
1954 ret = unuse_pte(vma, pmd, addr, entry, page);
1961 try_to_free_swap(page);
1965 if (*fs_pages_to_unuse && !--(*fs_pages_to_unuse)) {
1966 ret = FRONTSWAP_PAGES_UNUSED;
1970 pte = pte_offset_map(pmd, addr);
1971 } while (pte++, addr += PAGE_SIZE, addr != end);
1979 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1980 unsigned long addr, unsigned long end,
1981 unsigned int type, bool frontswap,
1982 unsigned long *fs_pages_to_unuse)
1988 pmd = pmd_offset(pud, addr);
1991 next = pmd_addr_end(addr, end);
1992 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
1994 ret = unuse_pte_range(vma, pmd, addr, next, type,
1995 frontswap, fs_pages_to_unuse);
1998 } while (pmd++, addr = next, addr != end);
2002 static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d,
2003 unsigned long addr, unsigned long end,
2004 unsigned int type, bool frontswap,
2005 unsigned long *fs_pages_to_unuse)
2011 pud = pud_offset(p4d, addr);
2013 next = pud_addr_end(addr, end);
2014 if (pud_none_or_clear_bad(pud))
2016 ret = unuse_pmd_range(vma, pud, addr, next, type,
2017 frontswap, fs_pages_to_unuse);
2020 } while (pud++, addr = next, addr != end);
2024 static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd,
2025 unsigned long addr, unsigned long end,
2026 unsigned int type, bool frontswap,
2027 unsigned long *fs_pages_to_unuse)
2033 p4d = p4d_offset(pgd, addr);
2035 next = p4d_addr_end(addr, end);
2036 if (p4d_none_or_clear_bad(p4d))
2038 ret = unuse_pud_range(vma, p4d, addr, next, type,
2039 frontswap, fs_pages_to_unuse);
2042 } while (p4d++, addr = next, addr != end);
2046 static int unuse_vma(struct vm_area_struct *vma, unsigned int type,
2047 bool frontswap, unsigned long *fs_pages_to_unuse)
2050 unsigned long addr, end, next;
2053 addr = vma->vm_start;
2056 pgd = pgd_offset(vma->vm_mm, addr);
2058 next = pgd_addr_end(addr, end);
2059 if (pgd_none_or_clear_bad(pgd))
2061 ret = unuse_p4d_range(vma, pgd, addr, next, type,
2062 frontswap, fs_pages_to_unuse);
2065 } while (pgd++, addr = next, addr != end);
2069 static int unuse_mm(struct mm_struct *mm, unsigned int type,
2070 bool frontswap, unsigned long *fs_pages_to_unuse)
2072 struct vm_area_struct *vma;
2075 down_read(&mm->mmap_sem);
2076 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2077 if (vma->anon_vma) {
2078 ret = unuse_vma(vma, type, frontswap,
2085 up_read(&mm->mmap_sem);
2090 * Scan swap_map (or frontswap_map if frontswap parameter is true)
2091 * from current position to next entry still in use. Return 0
2092 * if there are no inuse entries after prev till end of the map.
2094 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
2095 unsigned int prev, bool frontswap)
2098 unsigned char count;
2101 * No need for swap_lock here: we're just looking
2102 * for whether an entry is in use, not modifying it; false
2103 * hits are okay, and sys_swapoff() has already prevented new
2104 * allocations from this area (while holding swap_lock).
2106 for (i = prev + 1; i < si->max; i++) {
2107 count = READ_ONCE(si->swap_map[i]);
2108 if (count && swap_count(count) != SWAP_MAP_BAD)
2109 if (!frontswap || frontswap_test(si, i))
2111 if ((i % LATENCY_LIMIT) == 0)
2122 * If the boolean frontswap is true, only unuse pages_to_unuse pages;
2123 * pages_to_unuse==0 means all pages; ignored if frontswap is false
2125 int try_to_unuse(unsigned int type, bool frontswap,
2126 unsigned long pages_to_unuse)
2128 struct mm_struct *prev_mm;
2129 struct mm_struct *mm;
2130 struct list_head *p;
2132 struct swap_info_struct *si = swap_info[type];
2137 if (!READ_ONCE(si->inuse_pages))
2144 retval = shmem_unuse(type, frontswap, &pages_to_unuse);
2151 spin_lock(&mmlist_lock);
2152 p = &init_mm.mmlist;
2153 while (READ_ONCE(si->inuse_pages) &&
2154 !signal_pending(current) &&
2155 (p = p->next) != &init_mm.mmlist) {
2157 mm = list_entry(p, struct mm_struct, mmlist);
2158 if (!mmget_not_zero(mm))
2160 spin_unlock(&mmlist_lock);
2163 retval = unuse_mm(mm, type, frontswap, &pages_to_unuse);
2171 * Make sure that we aren't completely killing
2172 * interactive performance.
2175 spin_lock(&mmlist_lock);
2177 spin_unlock(&mmlist_lock);
2182 while (READ_ONCE(si->inuse_pages) &&
2183 !signal_pending(current) &&
2184 (i = find_next_to_unuse(si, i, frontswap)) != 0) {
2186 entry = swp_entry(type, i);
2187 page = find_get_page(swap_address_space(entry), i);
2192 * It is conceivable that a racing task removed this page from
2193 * swap cache just before we acquired the page lock. The page
2194 * might even be back in swap cache on another swap area. But
2195 * that is okay, try_to_free_swap() only removes stale pages.
2198 wait_on_page_writeback(page);
2199 try_to_free_swap(page);
2204 * For frontswap, we just need to unuse pages_to_unuse, if
2205 * it was specified. Need not check frontswap again here as
2206 * we already zeroed out pages_to_unuse if not frontswap.
2208 if (pages_to_unuse && --pages_to_unuse == 0)
2213 * Lets check again to see if there are still swap entries in the map.
2214 * If yes, we would need to do retry the unuse logic again.
2215 * Under global memory pressure, swap entries can be reinserted back
2216 * into process space after the mmlist loop above passes over them.
2218 * Limit the number of retries? No: when mmget_not_zero() above fails,
2219 * that mm is likely to be freeing swap from exit_mmap(), which proceeds
2220 * at its own independent pace; and even shmem_writepage() could have
2221 * been preempted after get_swap_page(), temporarily hiding that swap.
2222 * It's easy and robust (though cpu-intensive) just to keep retrying.
2224 if (READ_ONCE(si->inuse_pages)) {
2225 if (!signal_pending(current))
2230 return (retval == FRONTSWAP_PAGES_UNUSED) ? 0 : retval;
2234 * After a successful try_to_unuse, if no swap is now in use, we know
2235 * we can empty the mmlist. swap_lock must be held on entry and exit.
2236 * Note that mmlist_lock nests inside swap_lock, and an mm must be
2237 * added to the mmlist just after page_duplicate - before would be racy.
2239 static void drain_mmlist(void)
2241 struct list_head *p, *next;
2244 for (type = 0; type < nr_swapfiles; type++)
2245 if (swap_info[type]->inuse_pages)
2247 spin_lock(&mmlist_lock);
2248 list_for_each_safe(p, next, &init_mm.mmlist)
2250 spin_unlock(&mmlist_lock);
2254 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
2255 * corresponds to page offset for the specified swap entry.
2256 * Note that the type of this function is sector_t, but it returns page offset
2257 * into the bdev, not sector offset.
2259 static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
2261 struct swap_info_struct *sis;
2262 struct swap_extent *se;
2265 sis = swp_swap_info(entry);
2268 offset = swp_offset(entry);
2269 se = offset_to_swap_extent(sis, offset);
2270 return se->start_block + (offset - se->start_page);
2274 * Returns the page offset into bdev for the specified page's swap entry.
2276 sector_t map_swap_page(struct page *page, struct block_device **bdev)
2279 entry.val = page_private(page);
2280 return map_swap_entry(entry, bdev);
2284 * Free all of a swapdev's extent information
2286 static void destroy_swap_extents(struct swap_info_struct *sis)
2288 while (!RB_EMPTY_ROOT(&sis->swap_extent_root)) {
2289 struct rb_node *rb = sis->swap_extent_root.rb_node;
2290 struct swap_extent *se = rb_entry(rb, struct swap_extent, rb_node);
2292 rb_erase(rb, &sis->swap_extent_root);
2296 if (sis->flags & SWP_ACTIVATED) {
2297 struct file *swap_file = sis->swap_file;
2298 struct address_space *mapping = swap_file->f_mapping;
2300 sis->flags &= ~SWP_ACTIVATED;
2301 if (mapping->a_ops->swap_deactivate)
2302 mapping->a_ops->swap_deactivate(swap_file);
2307 * Add a block range (and the corresponding page range) into this swapdev's
2310 * This function rather assumes that it is called in ascending page order.
2313 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
2314 unsigned long nr_pages, sector_t start_block)
2316 struct rb_node **link = &sis->swap_extent_root.rb_node, *parent = NULL;
2317 struct swap_extent *se;
2318 struct swap_extent *new_se;
2321 * place the new node at the right most since the
2322 * function is called in ascending page order.
2326 link = &parent->rb_right;
2330 se = rb_entry(parent, struct swap_extent, rb_node);
2331 BUG_ON(se->start_page + se->nr_pages != start_page);
2332 if (se->start_block + se->nr_pages == start_block) {
2334 se->nr_pages += nr_pages;
2339 /* No merge, insert a new extent. */
2340 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
2343 new_se->start_page = start_page;
2344 new_se->nr_pages = nr_pages;
2345 new_se->start_block = start_block;
2347 rb_link_node(&new_se->rb_node, parent, link);
2348 rb_insert_color(&new_se->rb_node, &sis->swap_extent_root);
2351 EXPORT_SYMBOL_GPL(add_swap_extent);
2354 * A `swap extent' is a simple thing which maps a contiguous range of pages
2355 * onto a contiguous range of disk blocks. An ordered list of swap extents
2356 * is built at swapon time and is then used at swap_writepage/swap_readpage
2357 * time for locating where on disk a page belongs.
2359 * If the swapfile is an S_ISBLK block device, a single extent is installed.
2360 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2361 * swap files identically.
2363 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2364 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
2365 * swapfiles are handled *identically* after swapon time.
2367 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2368 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
2369 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
2370 * requirements, they are simply tossed out - we will never use those blocks
2373 * For all swap devices we set S_SWAPFILE across the life of the swapon. This
2374 * prevents users from writing to the swap device, which will corrupt memory.
2376 * The amount of disk space which a single swap extent represents varies.
2377 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
2378 * extents in the list. To avoid much list walking, we cache the previous
2379 * search location in `curr_swap_extent', and start new searches from there.
2380 * This is extremely effective. The average number of iterations in
2381 * map_swap_page() has been measured at about 0.3 per page. - akpm.
2383 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
2385 struct file *swap_file = sis->swap_file;
2386 struct address_space *mapping = swap_file->f_mapping;
2387 struct inode *inode = mapping->host;
2390 if (S_ISBLK(inode->i_mode)) {
2391 ret = add_swap_extent(sis, 0, sis->max, 0);
2396 if (mapping->a_ops->swap_activate) {
2397 ret = mapping->a_ops->swap_activate(sis, swap_file, span);
2399 sis->flags |= SWP_ACTIVATED;
2401 sis->flags |= SWP_FS;
2402 ret = add_swap_extent(sis, 0, sis->max, 0);
2408 return generic_swapfile_activate(sis, swap_file, span);
2411 static int swap_node(struct swap_info_struct *p)
2413 struct block_device *bdev;
2418 bdev = p->swap_file->f_inode->i_sb->s_bdev;
2420 return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE;
2423 static void setup_swap_info(struct swap_info_struct *p, int prio,
2424 unsigned char *swap_map,
2425 struct swap_cluster_info *cluster_info)
2432 p->prio = --least_priority;
2434 * the plist prio is negated because plist ordering is
2435 * low-to-high, while swap ordering is high-to-low
2437 p->list.prio = -p->prio;
2440 p->avail_lists[i].prio = -p->prio;
2442 if (swap_node(p) == i)
2443 p->avail_lists[i].prio = 1;
2445 p->avail_lists[i].prio = -p->prio;
2448 p->swap_map = swap_map;
2449 p->cluster_info = cluster_info;
2452 static void _enable_swap_info(struct swap_info_struct *p)
2454 p->flags |= SWP_WRITEOK | SWP_VALID;
2455 atomic_long_add(p->pages, &nr_swap_pages);
2456 total_swap_pages += p->pages;
2458 assert_spin_locked(&swap_lock);
2460 * both lists are plists, and thus priority ordered.
2461 * swap_active_head needs to be priority ordered for swapoff(),
2462 * which on removal of any swap_info_struct with an auto-assigned
2463 * (i.e. negative) priority increments the auto-assigned priority
2464 * of any lower-priority swap_info_structs.
2465 * swap_avail_head needs to be priority ordered for get_swap_page(),
2466 * which allocates swap pages from the highest available priority
2469 plist_add(&p->list, &swap_active_head);
2470 add_to_avail_list(p);
2473 static void enable_swap_info(struct swap_info_struct *p, int prio,
2474 unsigned char *swap_map,
2475 struct swap_cluster_info *cluster_info,
2476 unsigned long *frontswap_map)
2478 frontswap_init(p->type, frontswap_map);
2479 spin_lock(&swap_lock);
2480 spin_lock(&p->lock);
2481 setup_swap_info(p, prio, swap_map, cluster_info);
2482 spin_unlock(&p->lock);
2483 spin_unlock(&swap_lock);
2485 * Guarantee swap_map, cluster_info, etc. fields are valid
2486 * between get/put_swap_device() if SWP_VALID bit is set
2489 spin_lock(&swap_lock);
2490 spin_lock(&p->lock);
2491 _enable_swap_info(p);
2492 spin_unlock(&p->lock);
2493 spin_unlock(&swap_lock);
2496 static void reinsert_swap_info(struct swap_info_struct *p)
2498 spin_lock(&swap_lock);
2499 spin_lock(&p->lock);
2500 setup_swap_info(p, p->prio, p->swap_map, p->cluster_info);
2501 _enable_swap_info(p);
2502 spin_unlock(&p->lock);
2503 spin_unlock(&swap_lock);
2506 bool has_usable_swap(void)
2510 spin_lock(&swap_lock);
2511 if (plist_head_empty(&swap_active_head))
2513 spin_unlock(&swap_lock);
2517 SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
2519 struct swap_info_struct *p = NULL;
2520 unsigned char *swap_map;
2521 struct swap_cluster_info *cluster_info;
2522 unsigned long *frontswap_map;
2523 struct file *swap_file, *victim;
2524 struct address_space *mapping;
2525 struct inode *inode;
2526 struct filename *pathname;
2528 unsigned int old_block_size;
2530 if (!capable(CAP_SYS_ADMIN))
2533 BUG_ON(!current->mm);
2535 pathname = getname(specialfile);
2536 if (IS_ERR(pathname))
2537 return PTR_ERR(pathname);
2539 victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
2540 err = PTR_ERR(victim);
2544 mapping = victim->f_mapping;
2545 spin_lock(&swap_lock);
2546 plist_for_each_entry(p, &swap_active_head, list) {
2547 if (p->flags & SWP_WRITEOK) {
2548 if (p->swap_file->f_mapping == mapping) {
2556 spin_unlock(&swap_lock);
2559 if (!security_vm_enough_memory_mm(current->mm, p->pages))
2560 vm_unacct_memory(p->pages);
2563 spin_unlock(&swap_lock);
2566 del_from_avail_list(p);
2567 spin_lock(&p->lock);
2569 struct swap_info_struct *si = p;
2572 plist_for_each_entry_continue(si, &swap_active_head, list) {
2575 for_each_node(nid) {
2576 if (si->avail_lists[nid].prio != 1)
2577 si->avail_lists[nid].prio--;
2582 plist_del(&p->list, &swap_active_head);
2583 atomic_long_sub(p->pages, &nr_swap_pages);
2584 total_swap_pages -= p->pages;
2585 p->flags &= ~SWP_WRITEOK;
2586 spin_unlock(&p->lock);
2587 spin_unlock(&swap_lock);
2589 disable_swap_slots_cache_lock();
2591 set_current_oom_origin();
2592 err = try_to_unuse(p->type, false, 0); /* force unuse all pages */
2593 clear_current_oom_origin();
2596 /* re-insert swap space back into swap_list */
2597 reinsert_swap_info(p);
2598 reenable_swap_slots_cache_unlock();
2602 reenable_swap_slots_cache_unlock();
2604 spin_lock(&swap_lock);
2605 spin_lock(&p->lock);
2606 p->flags &= ~SWP_VALID; /* mark swap device as invalid */
2607 spin_unlock(&p->lock);
2608 spin_unlock(&swap_lock);
2610 * wait for swap operations protected by get/put_swap_device()
2615 flush_work(&p->discard_work);
2617 destroy_swap_extents(p);
2618 if (p->flags & SWP_CONTINUED)
2619 free_swap_count_continuations(p);
2621 if (!p->bdev || !blk_queue_nonrot(bdev_get_queue(p->bdev)))
2622 atomic_dec(&nr_rotate_swap);
2624 mutex_lock(&swapon_mutex);
2625 spin_lock(&swap_lock);
2626 spin_lock(&p->lock);
2629 /* wait for anyone still in scan_swap_map */
2630 p->highest_bit = 0; /* cuts scans short */
2631 while (p->flags >= SWP_SCANNING) {
2632 spin_unlock(&p->lock);
2633 spin_unlock(&swap_lock);
2634 schedule_timeout_uninterruptible(1);
2635 spin_lock(&swap_lock);
2636 spin_lock(&p->lock);
2639 swap_file = p->swap_file;
2640 old_block_size = p->old_block_size;
2641 p->swap_file = NULL;
2643 swap_map = p->swap_map;
2645 cluster_info = p->cluster_info;
2646 p->cluster_info = NULL;
2647 frontswap_map = frontswap_map_get(p);
2648 spin_unlock(&p->lock);
2649 spin_unlock(&swap_lock);
2650 frontswap_invalidate_area(p->type);
2651 frontswap_map_set(p, NULL);
2652 mutex_unlock(&swapon_mutex);
2653 free_percpu(p->percpu_cluster);
2654 p->percpu_cluster = NULL;
2656 kvfree(cluster_info);
2657 kvfree(frontswap_map);
2658 /* Destroy swap account information */
2659 swap_cgroup_swapoff(p->type);
2660 exit_swap_address_space(p->type);
2662 inode = mapping->host;
2663 if (S_ISBLK(inode->i_mode)) {
2664 struct block_device *bdev = I_BDEV(inode);
2666 set_blocksize(bdev, old_block_size);
2667 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2671 inode->i_flags &= ~S_SWAPFILE;
2672 inode_unlock(inode);
2673 filp_close(swap_file, NULL);
2676 * Clear the SWP_USED flag after all resources are freed so that swapon
2677 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
2678 * not hold p->lock after we cleared its SWP_WRITEOK.
2680 spin_lock(&swap_lock);
2682 spin_unlock(&swap_lock);
2685 atomic_inc(&proc_poll_event);
2686 wake_up_interruptible(&proc_poll_wait);
2689 filp_close(victim, NULL);
2695 #ifdef CONFIG_PROC_FS
2696 static __poll_t swaps_poll(struct file *file, poll_table *wait)
2698 struct seq_file *seq = file->private_data;
2700 poll_wait(file, &proc_poll_wait, wait);
2702 if (seq->poll_event != atomic_read(&proc_poll_event)) {
2703 seq->poll_event = atomic_read(&proc_poll_event);
2704 return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
2707 return EPOLLIN | EPOLLRDNORM;
2711 static void *swap_start(struct seq_file *swap, loff_t *pos)
2713 struct swap_info_struct *si;
2717 mutex_lock(&swapon_mutex);
2720 return SEQ_START_TOKEN;
2722 for (type = 0; (si = swap_type_to_swap_info(type)); type++) {
2723 if (!(si->flags & SWP_USED) || !si->swap_map)
2732 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2734 struct swap_info_struct *si = v;
2737 if (v == SEQ_START_TOKEN)
2740 type = si->type + 1;
2743 for (; (si = swap_type_to_swap_info(type)); type++) {
2744 if (!(si->flags & SWP_USED) || !si->swap_map)
2752 static void swap_stop(struct seq_file *swap, void *v)
2754 mutex_unlock(&swapon_mutex);
2757 static int swap_show(struct seq_file *swap, void *v)
2759 struct swap_info_struct *si = v;
2763 if (si == SEQ_START_TOKEN) {
2764 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2768 file = si->swap_file;
2769 len = seq_file_path(swap, file, " \t\n\\");
2770 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
2771 len < 40 ? 40 - len : 1, " ",
2772 S_ISBLK(file_inode(file)->i_mode) ?
2773 "partition" : "file\t",
2774 si->pages << (PAGE_SHIFT - 10),
2775 si->inuse_pages << (PAGE_SHIFT - 10),
2780 static const struct seq_operations swaps_op = {
2781 .start = swap_start,
2787 static int swaps_open(struct inode *inode, struct file *file)
2789 struct seq_file *seq;
2792 ret = seq_open(file, &swaps_op);
2796 seq = file->private_data;
2797 seq->poll_event = atomic_read(&proc_poll_event);
2801 static const struct proc_ops swaps_proc_ops = {
2802 .proc_flags = PROC_ENTRY_PERMANENT,
2803 .proc_open = swaps_open,
2804 .proc_read = seq_read,
2805 .proc_lseek = seq_lseek,
2806 .proc_release = seq_release,
2807 .proc_poll = swaps_poll,
2810 static int __init procswaps_init(void)
2812 proc_create("swaps", 0, NULL, &swaps_proc_ops);
2815 __initcall(procswaps_init);
2816 #endif /* CONFIG_PROC_FS */
2818 #ifdef MAX_SWAPFILES_CHECK
2819 static int __init max_swapfiles_check(void)
2821 MAX_SWAPFILES_CHECK();
2824 late_initcall(max_swapfiles_check);
2827 static struct swap_info_struct *alloc_swap_info(void)
2829 struct swap_info_struct *p;
2833 p = kvzalloc(struct_size(p, avail_lists, nr_node_ids), GFP_KERNEL);
2835 return ERR_PTR(-ENOMEM);
2837 spin_lock(&swap_lock);
2838 for (type = 0; type < nr_swapfiles; type++) {
2839 if (!(swap_info[type]->flags & SWP_USED))
2842 if (type >= MAX_SWAPFILES) {
2843 spin_unlock(&swap_lock);
2845 return ERR_PTR(-EPERM);
2847 if (type >= nr_swapfiles) {
2849 WRITE_ONCE(swap_info[type], p);
2851 * Write swap_info[type] before nr_swapfiles, in case a
2852 * racing procfs swap_start() or swap_next() is reading them.
2853 * (We never shrink nr_swapfiles, we never free this entry.)
2856 WRITE_ONCE(nr_swapfiles, nr_swapfiles + 1);
2859 p = swap_info[type];
2861 * Do not memset this entry: a racing procfs swap_next()
2862 * would be relying on p->type to remain valid.
2865 p->swap_extent_root = RB_ROOT;
2866 plist_node_init(&p->list, 0);
2868 plist_node_init(&p->avail_lists[i], 0);
2869 p->flags = SWP_USED;
2870 spin_unlock(&swap_lock);
2871 spin_lock_init(&p->lock);
2872 spin_lock_init(&p->cont_lock);
2877 static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2881 if (S_ISBLK(inode->i_mode)) {
2882 p->bdev = bdgrab(I_BDEV(inode));
2883 error = blkdev_get(p->bdev,
2884 FMODE_READ | FMODE_WRITE | FMODE_EXCL, p);
2889 p->old_block_size = block_size(p->bdev);
2890 error = set_blocksize(p->bdev, PAGE_SIZE);
2894 * Zoned block devices contain zones that have a sequential
2895 * write only restriction. Hence zoned block devices are not
2896 * suitable for swapping. Disallow them here.
2898 if (blk_queue_is_zoned(p->bdev->bd_queue))
2900 p->flags |= SWP_BLKDEV;
2901 } else if (S_ISREG(inode->i_mode)) {
2902 p->bdev = inode->i_sb->s_bdev;
2910 * Find out how many pages are allowed for a single swap device. There
2911 * are two limiting factors:
2912 * 1) the number of bits for the swap offset in the swp_entry_t type, and
2913 * 2) the number of bits in the swap pte, as defined by the different
2916 * In order to find the largest possible bit mask, a swap entry with
2917 * swap type 0 and swap offset ~0UL is created, encoded to a swap pte,
2918 * decoded to a swp_entry_t again, and finally the swap offset is
2921 * This will mask all the bits from the initial ~0UL mask that can't
2922 * be encoded in either the swp_entry_t or the architecture definition
2925 unsigned long generic_max_swapfile_size(void)
2927 return swp_offset(pte_to_swp_entry(
2928 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2931 /* Can be overridden by an architecture for additional checks. */
2932 __weak unsigned long max_swapfile_size(void)
2934 return generic_max_swapfile_size();
2937 static unsigned long read_swap_header(struct swap_info_struct *p,
2938 union swap_header *swap_header,
2939 struct inode *inode)
2942 unsigned long maxpages;
2943 unsigned long swapfilepages;
2944 unsigned long last_page;
2946 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
2947 pr_err("Unable to find swap-space signature\n");
2951 /* swap partition endianess hack... */
2952 if (swab32(swap_header->info.version) == 1) {
2953 swab32s(&swap_header->info.version);
2954 swab32s(&swap_header->info.last_page);
2955 swab32s(&swap_header->info.nr_badpages);
2956 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2958 for (i = 0; i < swap_header->info.nr_badpages; i++)
2959 swab32s(&swap_header->info.badpages[i]);
2961 /* Check the swap header's sub-version */
2962 if (swap_header->info.version != 1) {
2963 pr_warn("Unable to handle swap header version %d\n",
2964 swap_header->info.version);
2969 p->cluster_next = 1;
2972 maxpages = max_swapfile_size();
2973 last_page = swap_header->info.last_page;
2975 pr_warn("Empty swap-file\n");
2978 if (last_page > maxpages) {
2979 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
2980 maxpages << (PAGE_SHIFT - 10),
2981 last_page << (PAGE_SHIFT - 10));
2983 if (maxpages > last_page) {
2984 maxpages = last_page + 1;
2985 /* p->max is an unsigned int: don't overflow it */
2986 if ((unsigned int)maxpages == 0)
2987 maxpages = UINT_MAX;
2989 p->highest_bit = maxpages - 1;
2993 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2994 if (swapfilepages && maxpages > swapfilepages) {
2995 pr_warn("Swap area shorter than signature indicates\n");
2998 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
3000 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
3006 #define SWAP_CLUSTER_INFO_COLS \
3007 DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
3008 #define SWAP_CLUSTER_SPACE_COLS \
3009 DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
3010 #define SWAP_CLUSTER_COLS \
3011 max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
3013 static int setup_swap_map_and_extents(struct swap_info_struct *p,
3014 union swap_header *swap_header,
3015 unsigned char *swap_map,
3016 struct swap_cluster_info *cluster_info,
3017 unsigned long maxpages,
3021 unsigned int nr_good_pages;
3023 unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
3024 unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS;
3025 unsigned long i, idx;
3027 nr_good_pages = maxpages - 1; /* omit header page */
3029 cluster_list_init(&p->free_clusters);
3030 cluster_list_init(&p->discard_clusters);
3032 for (i = 0; i < swap_header->info.nr_badpages; i++) {
3033 unsigned int page_nr = swap_header->info.badpages[i];
3034 if (page_nr == 0 || page_nr > swap_header->info.last_page)
3036 if (page_nr < maxpages) {
3037 swap_map[page_nr] = SWAP_MAP_BAD;
3040 * Haven't marked the cluster free yet, no list
3041 * operation involved
3043 inc_cluster_info_page(p, cluster_info, page_nr);
3047 /* Haven't marked the cluster free yet, no list operation involved */
3048 for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
3049 inc_cluster_info_page(p, cluster_info, i);
3051 if (nr_good_pages) {
3052 swap_map[0] = SWAP_MAP_BAD;
3054 * Not mark the cluster free yet, no list
3055 * operation involved
3057 inc_cluster_info_page(p, cluster_info, 0);
3059 p->pages = nr_good_pages;
3060 nr_extents = setup_swap_extents(p, span);
3063 nr_good_pages = p->pages;
3065 if (!nr_good_pages) {
3066 pr_warn("Empty swap-file\n");
3075 * Reduce false cache line sharing between cluster_info and
3076 * sharing same address space.
3078 for (k = 0; k < SWAP_CLUSTER_COLS; k++) {
3079 j = (k + col) % SWAP_CLUSTER_COLS;
3080 for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) {
3081 idx = i * SWAP_CLUSTER_COLS + j;
3082 if (idx >= nr_clusters)
3084 if (cluster_count(&cluster_info[idx]))
3086 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
3087 cluster_list_add_tail(&p->free_clusters, cluster_info,
3095 * Helper to sys_swapon determining if a given swap
3096 * backing device queue supports DISCARD operations.
3098 static bool swap_discardable(struct swap_info_struct *si)
3100 struct request_queue *q = bdev_get_queue(si->bdev);
3102 if (!q || !blk_queue_discard(q))
3108 SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
3110 struct swap_info_struct *p;
3111 struct filename *name;
3112 struct file *swap_file = NULL;
3113 struct address_space *mapping;
3116 union swap_header *swap_header;
3119 unsigned long maxpages;
3120 unsigned char *swap_map = NULL;
3121 struct swap_cluster_info *cluster_info = NULL;
3122 unsigned long *frontswap_map = NULL;
3123 struct page *page = NULL;
3124 struct inode *inode = NULL;
3125 bool inced_nr_rotate_swap = false;
3127 if (swap_flags & ~SWAP_FLAGS_VALID)
3130 if (!capable(CAP_SYS_ADMIN))
3133 if (!swap_avail_heads)
3136 p = alloc_swap_info();
3140 INIT_WORK(&p->discard_work, swap_discard_work);
3142 name = getname(specialfile);
3144 error = PTR_ERR(name);
3148 swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
3149 if (IS_ERR(swap_file)) {
3150 error = PTR_ERR(swap_file);
3155 p->swap_file = swap_file;
3156 mapping = swap_file->f_mapping;
3157 inode = mapping->host;
3159 error = claim_swapfile(p, inode);
3160 if (unlikely(error))
3164 if (IS_SWAPFILE(inode)) {
3166 goto bad_swap_unlock_inode;
3170 * Read the swap header.
3172 if (!mapping->a_ops->readpage) {
3174 goto bad_swap_unlock_inode;
3176 page = read_mapping_page(mapping, 0, swap_file);
3178 error = PTR_ERR(page);
3179 goto bad_swap_unlock_inode;
3181 swap_header = kmap(page);
3183 maxpages = read_swap_header(p, swap_header, inode);
3184 if (unlikely(!maxpages)) {
3186 goto bad_swap_unlock_inode;
3189 /* OK, set up the swap map and apply the bad block list */
3190 swap_map = vzalloc(maxpages);
3193 goto bad_swap_unlock_inode;
3196 if (bdi_cap_stable_pages_required(inode_to_bdi(inode)))
3197 p->flags |= SWP_STABLE_WRITES;
3199 if (bdi_cap_synchronous_io(inode_to_bdi(inode)))
3200 p->flags |= SWP_SYNCHRONOUS_IO;
3202 if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) {
3204 unsigned long ci, nr_cluster;
3206 p->flags |= SWP_SOLIDSTATE;
3208 * select a random position to start with to help wear leveling
3211 p->cluster_next = 1 + (prandom_u32() % p->highest_bit);
3212 nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
3214 cluster_info = kvcalloc(nr_cluster, sizeof(*cluster_info),
3216 if (!cluster_info) {
3218 goto bad_swap_unlock_inode;
3221 for (ci = 0; ci < nr_cluster; ci++)
3222 spin_lock_init(&((cluster_info + ci)->lock));
3224 p->percpu_cluster = alloc_percpu(struct percpu_cluster);
3225 if (!p->percpu_cluster) {
3227 goto bad_swap_unlock_inode;
3229 for_each_possible_cpu(cpu) {
3230 struct percpu_cluster *cluster;
3231 cluster = per_cpu_ptr(p->percpu_cluster, cpu);
3232 cluster_set_null(&cluster->index);
3235 atomic_inc(&nr_rotate_swap);
3236 inced_nr_rotate_swap = true;
3239 error = swap_cgroup_swapon(p->type, maxpages);
3241 goto bad_swap_unlock_inode;
3243 nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
3244 cluster_info, maxpages, &span);
3245 if (unlikely(nr_extents < 0)) {
3247 goto bad_swap_unlock_inode;
3249 /* frontswap enabled? set up bit-per-page map for frontswap */
3250 if (IS_ENABLED(CONFIG_FRONTSWAP))
3251 frontswap_map = kvcalloc(BITS_TO_LONGS(maxpages),
3255 if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) {
3257 * When discard is enabled for swap with no particular
3258 * policy flagged, we set all swap discard flags here in
3259 * order to sustain backward compatibility with older
3260 * swapon(8) releases.
3262 p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
3266 * By flagging sys_swapon, a sysadmin can tell us to
3267 * either do single-time area discards only, or to just
3268 * perform discards for released swap page-clusters.
3269 * Now it's time to adjust the p->flags accordingly.
3271 if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
3272 p->flags &= ~SWP_PAGE_DISCARD;
3273 else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
3274 p->flags &= ~SWP_AREA_DISCARD;
3276 /* issue a swapon-time discard if it's still required */
3277 if (p->flags & SWP_AREA_DISCARD) {
3278 int err = discard_swap(p);
3280 pr_err("swapon: discard_swap(%p): %d\n",
3285 error = init_swap_address_space(p->type, maxpages);
3287 goto bad_swap_unlock_inode;
3290 * Flush any pending IO and dirty mappings before we start using this
3293 inode->i_flags |= S_SWAPFILE;
3294 error = inode_drain_writes(inode);
3296 inode->i_flags &= ~S_SWAPFILE;
3297 goto bad_swap_unlock_inode;
3300 mutex_lock(&swapon_mutex);
3302 if (swap_flags & SWAP_FLAG_PREFER)
3304 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
3305 enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
3307 pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
3308 p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
3309 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
3310 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
3311 (p->flags & SWP_DISCARDABLE) ? "D" : "",
3312 (p->flags & SWP_AREA_DISCARD) ? "s" : "",
3313 (p->flags & SWP_PAGE_DISCARD) ? "c" : "",
3314 (frontswap_map) ? "FS" : "");
3316 mutex_unlock(&swapon_mutex);
3317 atomic_inc(&proc_poll_event);
3318 wake_up_interruptible(&proc_poll_wait);
3322 bad_swap_unlock_inode:
3323 inode_unlock(inode);
3325 free_percpu(p->percpu_cluster);
3326 p->percpu_cluster = NULL;
3327 if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
3328 set_blocksize(p->bdev, p->old_block_size);
3329 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
3332 destroy_swap_extents(p);
3333 swap_cgroup_swapoff(p->type);
3334 spin_lock(&swap_lock);
3335 p->swap_file = NULL;
3337 spin_unlock(&swap_lock);
3339 kvfree(cluster_info);
3340 kvfree(frontswap_map);
3341 if (inced_nr_rotate_swap)
3342 atomic_dec(&nr_rotate_swap);
3344 filp_close(swap_file, NULL);
3346 if (page && !IS_ERR(page)) {
3353 inode_unlock(inode);
3355 enable_swap_slots_cache();
3359 void si_swapinfo(struct sysinfo *val)
3362 unsigned long nr_to_be_unused = 0;
3364 spin_lock(&swap_lock);
3365 for (type = 0; type < nr_swapfiles; type++) {
3366 struct swap_info_struct *si = swap_info[type];
3368 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
3369 nr_to_be_unused += si->inuse_pages;
3371 val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
3372 val->totalswap = total_swap_pages + nr_to_be_unused;
3373 spin_unlock(&swap_lock);
3377 * Verify that a swap entry is valid and increment its swap map count.
3379 * Returns error code in following case.
3381 * - swp_entry is invalid -> EINVAL
3382 * - swp_entry is migration entry -> EINVAL
3383 * - swap-cache reference is requested but there is already one. -> EEXIST
3384 * - swap-cache reference is requested but the entry is not used. -> ENOENT
3385 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
3387 static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
3389 struct swap_info_struct *p;
3390 struct swap_cluster_info *ci;
3391 unsigned long offset;
3392 unsigned char count;
3393 unsigned char has_cache;
3396 p = get_swap_device(entry);
3400 offset = swp_offset(entry);
3401 ci = lock_cluster_or_swap_info(p, offset);
3403 count = p->swap_map[offset];
3406 * swapin_readahead() doesn't check if a swap entry is valid, so the
3407 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
3409 if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
3414 has_cache = count & SWAP_HAS_CACHE;
3415 count &= ~SWAP_HAS_CACHE;
3418 if (usage == SWAP_HAS_CACHE) {
3420 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
3421 if (!has_cache && count)
3422 has_cache = SWAP_HAS_CACHE;
3423 else if (has_cache) /* someone else added cache */
3425 else /* no users remaining */
3428 } else if (count || has_cache) {
3430 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
3432 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
3434 else if (swap_count_continued(p, offset, count))
3435 count = COUNT_CONTINUED;
3439 err = -ENOENT; /* unused swap entry */
3441 p->swap_map[offset] = count | has_cache;
3444 unlock_cluster_or_swap_info(p, ci);
3452 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3453 * (in which case its reference count is never incremented).
3455 void swap_shmem_alloc(swp_entry_t entry)
3457 __swap_duplicate(entry, SWAP_MAP_SHMEM);
3461 * Increase reference count of swap entry by 1.
3462 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3463 * but could not be atomically allocated. Returns 0, just as if it succeeded,
3464 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3465 * might occur if a page table entry has got corrupted.
3467 int swap_duplicate(swp_entry_t entry)
3471 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
3472 err = add_swap_count_continuation(entry, GFP_ATOMIC);
3477 * @entry: swap entry for which we allocate swap cache.
3479 * Called when allocating swap cache for existing swap entry,
3480 * This can return error codes. Returns 0 at success.
3481 * -EEXIST means there is a swap cache.
3482 * Note: return code is different from swap_duplicate().
3484 int swapcache_prepare(swp_entry_t entry)
3486 return __swap_duplicate(entry, SWAP_HAS_CACHE);
3489 struct swap_info_struct *swp_swap_info(swp_entry_t entry)
3491 return swap_type_to_swap_info(swp_type(entry));
3494 struct swap_info_struct *page_swap_info(struct page *page)
3496 swp_entry_t entry = { .val = page_private(page) };
3497 return swp_swap_info(entry);
3501 * out-of-line __page_file_ methods to avoid include hell.
3503 struct address_space *__page_file_mapping(struct page *page)
3505 return page_swap_info(page)->swap_file->f_mapping;
3507 EXPORT_SYMBOL_GPL(__page_file_mapping);
3509 pgoff_t __page_file_index(struct page *page)
3511 swp_entry_t swap = { .val = page_private(page) };
3512 return swp_offset(swap);
3514 EXPORT_SYMBOL_GPL(__page_file_index);
3517 * add_swap_count_continuation - called when a swap count is duplicated
3518 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3519 * page of the original vmalloc'ed swap_map, to hold the continuation count
3520 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
3521 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3523 * These continuation pages are seldom referenced: the common paths all work
3524 * on the original swap_map, only referring to a continuation page when the
3525 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3527 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3528 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3529 * can be called after dropping locks.
3531 int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
3533 struct swap_info_struct *si;
3534 struct swap_cluster_info *ci;
3537 struct page *list_page;
3539 unsigned char count;
3543 * When debugging, it's easier to use __GFP_ZERO here; but it's better
3544 * for latency not to zero a page while GFP_ATOMIC and holding locks.
3546 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
3548 si = get_swap_device(entry);
3551 * An acceptable race has occurred since the failing
3552 * __swap_duplicate(): the swap device may be swapoff
3556 spin_lock(&si->lock);
3558 offset = swp_offset(entry);
3560 ci = lock_cluster(si, offset);
3562 count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
3564 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
3566 * The higher the swap count, the more likely it is that tasks
3567 * will race to add swap count continuation: we need to avoid
3568 * over-provisioning.
3579 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
3580 * no architecture is using highmem pages for kernel page tables: so it
3581 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
3583 head = vmalloc_to_page(si->swap_map + offset);
3584 offset &= ~PAGE_MASK;
3586 spin_lock(&si->cont_lock);
3588 * Page allocation does not initialize the page's lru field,
3589 * but it does always reset its private field.
3591 if (!page_private(head)) {
3592 BUG_ON(count & COUNT_CONTINUED);
3593 INIT_LIST_HEAD(&head->lru);
3594 set_page_private(head, SWP_CONTINUED);
3595 si->flags |= SWP_CONTINUED;
3598 list_for_each_entry(list_page, &head->lru, lru) {
3602 * If the previous map said no continuation, but we've found
3603 * a continuation page, free our allocation and use this one.
3605 if (!(count & COUNT_CONTINUED))
3606 goto out_unlock_cont;
3608 map = kmap_atomic(list_page) + offset;
3613 * If this continuation count now has some space in it,
3614 * free our allocation and use this one.
3616 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
3617 goto out_unlock_cont;
3620 list_add_tail(&page->lru, &head->lru);
3621 page = NULL; /* now it's attached, don't free it */
3623 spin_unlock(&si->cont_lock);
3626 spin_unlock(&si->lock);
3627 put_swap_device(si);
3635 * swap_count_continued - when the original swap_map count is incremented
3636 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3637 * into, carry if so, or else fail until a new continuation page is allocated;
3638 * when the original swap_map count is decremented from 0 with continuation,
3639 * borrow from the continuation and report whether it still holds more.
3640 * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3643 static bool swap_count_continued(struct swap_info_struct *si,
3644 pgoff_t offset, unsigned char count)
3651 head = vmalloc_to_page(si->swap_map + offset);
3652 if (page_private(head) != SWP_CONTINUED) {
3653 BUG_ON(count & COUNT_CONTINUED);
3654 return false; /* need to add count continuation */
3657 spin_lock(&si->cont_lock);
3658 offset &= ~PAGE_MASK;
3659 page = list_next_entry(head, lru);
3660 map = kmap_atomic(page) + offset;
3662 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
3663 goto init_map; /* jump over SWAP_CONT_MAX checks */
3665 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
3667 * Think of how you add 1 to 999
3669 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
3671 page = list_next_entry(page, lru);
3672 BUG_ON(page == head);
3673 map = kmap_atomic(page) + offset;
3675 if (*map == SWAP_CONT_MAX) {
3677 page = list_next_entry(page, lru);
3679 ret = false; /* add count continuation */
3682 map = kmap_atomic(page) + offset;
3683 init_map: *map = 0; /* we didn't zero the page */
3687 while ((page = list_prev_entry(page, lru)) != head) {
3688 map = kmap_atomic(page) + offset;
3689 *map = COUNT_CONTINUED;
3692 ret = true; /* incremented */
3694 } else { /* decrementing */
3696 * Think of how you subtract 1 from 1000
3698 BUG_ON(count != COUNT_CONTINUED);
3699 while (*map == COUNT_CONTINUED) {
3701 page = list_next_entry(page, lru);
3702 BUG_ON(page == head);
3703 map = kmap_atomic(page) + offset;
3710 while ((page = list_prev_entry(page, lru)) != head) {
3711 map = kmap_atomic(page) + offset;
3712 *map = SWAP_CONT_MAX | count;
3713 count = COUNT_CONTINUED;
3716 ret = count == COUNT_CONTINUED;
3719 spin_unlock(&si->cont_lock);
3724 * free_swap_count_continuations - swapoff free all the continuation pages
3725 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3727 static void free_swap_count_continuations(struct swap_info_struct *si)
3731 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
3733 head = vmalloc_to_page(si->swap_map + offset);
3734 if (page_private(head)) {
3735 struct page *page, *next;
3737 list_for_each_entry_safe(page, next, &head->lru, lru) {
3738 list_del(&page->lru);
3745 #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
3746 void mem_cgroup_throttle_swaprate(struct mem_cgroup *memcg, int node,
3749 struct swap_info_struct *si, *next;
3750 if (!(gfp_mask & __GFP_IO) || !memcg)
3753 if (!blk_cgroup_congested())
3757 * We've already scheduled a throttle, avoid taking the global swap
3760 if (current->throttle_queue)
3763 spin_lock(&swap_avail_lock);
3764 plist_for_each_entry_safe(si, next, &swap_avail_heads[node],
3765 avail_lists[node]) {
3767 blkcg_schedule_throttle(bdev_get_queue(si->bdev),
3772 spin_unlock(&swap_avail_lock);
3776 static int __init swapfile_init(void)
3780 swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head),
3782 if (!swap_avail_heads) {
3783 pr_emerg("Not enough memory for swap heads, swap is disabled\n");
3788 plist_head_init(&swap_avail_heads[nid]);
3792 subsys_initcall(swapfile_init);