4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 * Swap reorganised 29.12.95, Stephen Tweedie
9 #include <linux/sched/mm.h>
10 #include <linux/sched/task.h>
11 #include <linux/hugetlb.h>
12 #include <linux/mman.h>
13 #include <linux/slab.h>
14 #include <linux/kernel_stat.h>
15 #include <linux/swap.h>
16 #include <linux/vmalloc.h>
17 #include <linux/pagemap.h>
18 #include <linux/namei.h>
19 #include <linux/shmem_fs.h>
20 #include <linux/blkdev.h>
21 #include <linux/random.h>
22 #include <linux/writeback.h>
23 #include <linux/proc_fs.h>
24 #include <linux/seq_file.h>
25 #include <linux/init.h>
26 #include <linux/ksm.h>
27 #include <linux/rmap.h>
28 #include <linux/security.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mutex.h>
31 #include <linux/capability.h>
32 #include <linux/syscalls.h>
33 #include <linux/memcontrol.h>
34 #include <linux/poll.h>
35 #include <linux/oom.h>
36 #include <linux/frontswap.h>
37 #include <linux/swapfile.h>
38 #include <linux/export.h>
39 #include <linux/swap_slots.h>
41 #include <asm/pgtable.h>
42 #include <asm/tlbflush.h>
43 #include <linux/swapops.h>
44 #include <linux/swap_cgroup.h>
46 static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
48 static void free_swap_count_continuations(struct swap_info_struct *);
49 static sector_t map_swap_entry(swp_entry_t, struct block_device**);
51 DEFINE_SPINLOCK(swap_lock);
52 static unsigned int nr_swapfiles;
53 atomic_long_t nr_swap_pages;
55 * Some modules use swappable objects and may try to swap them out under
56 * memory pressure (via the shrinker). Before doing so, they may wish to
57 * check to see if any swap space is available.
59 EXPORT_SYMBOL_GPL(nr_swap_pages);
60 /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
61 long total_swap_pages;
62 static int least_priority;
64 static const char Bad_file[] = "Bad swap file entry ";
65 static const char Unused_file[] = "Unused swap file entry ";
66 static const char Bad_offset[] = "Bad swap offset entry ";
67 static const char Unused_offset[] = "Unused swap offset entry ";
70 * all active swap_info_structs
71 * protected with swap_lock, and ordered by priority.
73 PLIST_HEAD(swap_active_head);
76 * all available (active, not full) swap_info_structs
77 * protected with swap_avail_lock, ordered by priority.
78 * This is used by get_swap_page() instead of swap_active_head
79 * because swap_active_head includes all swap_info_structs,
80 * but get_swap_page() doesn't need to look at full ones.
81 * This uses its own lock instead of swap_lock because when a
82 * swap_info_struct changes between not-full/full, it needs to
83 * add/remove itself to/from this list, but the swap_info_struct->lock
84 * is held and the locking order requires swap_lock to be taken
85 * before any swap_info_struct->lock.
87 static PLIST_HEAD(swap_avail_head);
88 static DEFINE_SPINLOCK(swap_avail_lock);
90 struct swap_info_struct *swap_info[MAX_SWAPFILES];
92 static DEFINE_MUTEX(swapon_mutex);
94 static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
95 /* Activity counter to indicate that a swapon or swapoff has occurred */
96 static atomic_t proc_poll_event = ATOMIC_INIT(0);
98 static inline unsigned char swap_count(unsigned char ent)
100 return ent & ~SWAP_HAS_CACHE; /* may include SWAP_HAS_CONT flag */
103 /* returns 1 if swap entry is freed */
105 __try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset)
107 swp_entry_t entry = swp_entry(si->type, offset);
111 page = find_get_page(swap_address_space(entry), swp_offset(entry));
115 * This function is called from scan_swap_map() and it's called
116 * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
117 * We have to use trylock for avoiding deadlock. This is a special
118 * case and you should use try_to_free_swap() with explicit lock_page()
119 * in usual operations.
121 if (trylock_page(page)) {
122 ret = try_to_free_swap(page);
130 * swapon tell device that all the old swap contents can be discarded,
131 * to allow the swap device to optimize its wear-levelling.
133 static int discard_swap(struct swap_info_struct *si)
135 struct swap_extent *se;
136 sector_t start_block;
140 /* Do not discard the swap header page! */
141 se = &si->first_swap_extent;
142 start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
143 nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
145 err = blkdev_issue_discard(si->bdev, start_block,
146 nr_blocks, GFP_KERNEL, 0);
152 list_for_each_entry(se, &si->first_swap_extent.list, list) {
153 start_block = se->start_block << (PAGE_SHIFT - 9);
154 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
156 err = blkdev_issue_discard(si->bdev, start_block,
157 nr_blocks, GFP_KERNEL, 0);
163 return err; /* That will often be -EOPNOTSUPP */
167 * swap allocation tell device that a cluster of swap can now be discarded,
168 * to allow the swap device to optimize its wear-levelling.
170 static void discard_swap_cluster(struct swap_info_struct *si,
171 pgoff_t start_page, pgoff_t nr_pages)
173 struct swap_extent *se = si->curr_swap_extent;
174 int found_extent = 0;
177 if (se->start_page <= start_page &&
178 start_page < se->start_page + se->nr_pages) {
179 pgoff_t offset = start_page - se->start_page;
180 sector_t start_block = se->start_block + offset;
181 sector_t nr_blocks = se->nr_pages - offset;
183 if (nr_blocks > nr_pages)
184 nr_blocks = nr_pages;
185 start_page += nr_blocks;
186 nr_pages -= nr_blocks;
189 si->curr_swap_extent = se;
191 start_block <<= PAGE_SHIFT - 9;
192 nr_blocks <<= PAGE_SHIFT - 9;
193 if (blkdev_issue_discard(si->bdev, start_block,
194 nr_blocks, GFP_NOIO, 0))
198 se = list_next_entry(se, list);
202 #define SWAPFILE_CLUSTER 256
203 #define LATENCY_LIMIT 256
205 static inline void cluster_set_flag(struct swap_cluster_info *info,
211 static inline unsigned int cluster_count(struct swap_cluster_info *info)
216 static inline void cluster_set_count(struct swap_cluster_info *info,
222 static inline void cluster_set_count_flag(struct swap_cluster_info *info,
223 unsigned int c, unsigned int f)
229 static inline unsigned int cluster_next(struct swap_cluster_info *info)
234 static inline void cluster_set_next(struct swap_cluster_info *info,
240 static inline void cluster_set_next_flag(struct swap_cluster_info *info,
241 unsigned int n, unsigned int f)
247 static inline bool cluster_is_free(struct swap_cluster_info *info)
249 return info->flags & CLUSTER_FLAG_FREE;
252 static inline bool cluster_is_null(struct swap_cluster_info *info)
254 return info->flags & CLUSTER_FLAG_NEXT_NULL;
257 static inline void cluster_set_null(struct swap_cluster_info *info)
259 info->flags = CLUSTER_FLAG_NEXT_NULL;
263 static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si,
264 unsigned long offset)
266 struct swap_cluster_info *ci;
268 ci = si->cluster_info;
270 ci += offset / SWAPFILE_CLUSTER;
271 spin_lock(&ci->lock);
276 static inline void unlock_cluster(struct swap_cluster_info *ci)
279 spin_unlock(&ci->lock);
282 static inline struct swap_cluster_info *lock_cluster_or_swap_info(
283 struct swap_info_struct *si,
284 unsigned long offset)
286 struct swap_cluster_info *ci;
288 ci = lock_cluster(si, offset);
290 spin_lock(&si->lock);
295 static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si,
296 struct swap_cluster_info *ci)
301 spin_unlock(&si->lock);
304 static inline bool cluster_list_empty(struct swap_cluster_list *list)
306 return cluster_is_null(&list->head);
309 static inline unsigned int cluster_list_first(struct swap_cluster_list *list)
311 return cluster_next(&list->head);
314 static void cluster_list_init(struct swap_cluster_list *list)
316 cluster_set_null(&list->head);
317 cluster_set_null(&list->tail);
320 static void cluster_list_add_tail(struct swap_cluster_list *list,
321 struct swap_cluster_info *ci,
324 if (cluster_list_empty(list)) {
325 cluster_set_next_flag(&list->head, idx, 0);
326 cluster_set_next_flag(&list->tail, idx, 0);
328 struct swap_cluster_info *ci_tail;
329 unsigned int tail = cluster_next(&list->tail);
332 * Nested cluster lock, but both cluster locks are
333 * only acquired when we held swap_info_struct->lock
336 spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING);
337 cluster_set_next(ci_tail, idx);
338 unlock_cluster(ci_tail);
339 cluster_set_next_flag(&list->tail, idx, 0);
343 static unsigned int cluster_list_del_first(struct swap_cluster_list *list,
344 struct swap_cluster_info *ci)
348 idx = cluster_next(&list->head);
349 if (cluster_next(&list->tail) == idx) {
350 cluster_set_null(&list->head);
351 cluster_set_null(&list->tail);
353 cluster_set_next_flag(&list->head,
354 cluster_next(&ci[idx]), 0);
359 /* Add a cluster to discard list and schedule it to do discard */
360 static void swap_cluster_schedule_discard(struct swap_info_struct *si,
364 * If scan_swap_map() can't find a free cluster, it will check
365 * si->swap_map directly. To make sure the discarding cluster isn't
366 * taken by scan_swap_map(), mark the swap entries bad (occupied). It
367 * will be cleared after discard
369 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
370 SWAP_MAP_BAD, SWAPFILE_CLUSTER);
372 cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx);
374 schedule_work(&si->discard_work);
378 * Doing discard actually. After a cluster discard is finished, the cluster
379 * will be added to free cluster list. caller should hold si->lock.
381 static void swap_do_scheduled_discard(struct swap_info_struct *si)
383 struct swap_cluster_info *info, *ci;
386 info = si->cluster_info;
388 while (!cluster_list_empty(&si->discard_clusters)) {
389 idx = cluster_list_del_first(&si->discard_clusters, info);
390 spin_unlock(&si->lock);
392 discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
395 spin_lock(&si->lock);
396 ci = lock_cluster(si, idx * SWAPFILE_CLUSTER);
397 cluster_set_flag(ci, CLUSTER_FLAG_FREE);
399 cluster_list_add_tail(&si->free_clusters, info, idx);
400 ci = lock_cluster(si, idx * SWAPFILE_CLUSTER);
401 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
402 0, SWAPFILE_CLUSTER);
407 static void swap_discard_work(struct work_struct *work)
409 struct swap_info_struct *si;
411 si = container_of(work, struct swap_info_struct, discard_work);
413 spin_lock(&si->lock);
414 swap_do_scheduled_discard(si);
415 spin_unlock(&si->lock);
419 * The cluster corresponding to page_nr will be used. The cluster will be
420 * removed from free cluster list and its usage counter will be increased.
422 static void inc_cluster_info_page(struct swap_info_struct *p,
423 struct swap_cluster_info *cluster_info, unsigned long page_nr)
425 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
429 if (cluster_is_free(&cluster_info[idx])) {
430 VM_BUG_ON(cluster_list_first(&p->free_clusters) != idx);
431 cluster_list_del_first(&p->free_clusters, cluster_info);
432 cluster_set_count_flag(&cluster_info[idx], 0, 0);
435 VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
436 cluster_set_count(&cluster_info[idx],
437 cluster_count(&cluster_info[idx]) + 1);
441 * The cluster corresponding to page_nr decreases one usage. If the usage
442 * counter becomes 0, which means no page in the cluster is in using, we can
443 * optionally discard the cluster and add it to free cluster list.
445 static void dec_cluster_info_page(struct swap_info_struct *p,
446 struct swap_cluster_info *cluster_info, unsigned long page_nr)
448 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
453 VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
454 cluster_set_count(&cluster_info[idx],
455 cluster_count(&cluster_info[idx]) - 1);
457 if (cluster_count(&cluster_info[idx]) == 0) {
459 * If the swap is discardable, prepare discard the cluster
460 * instead of free it immediately. The cluster will be freed
463 if ((p->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
464 (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
465 swap_cluster_schedule_discard(p, idx);
469 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
470 cluster_list_add_tail(&p->free_clusters, cluster_info, idx);
475 * It's possible scan_swap_map() uses a free cluster in the middle of free
476 * cluster list. Avoiding such abuse to avoid list corruption.
479 scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
480 unsigned long offset)
482 struct percpu_cluster *percpu_cluster;
485 offset /= SWAPFILE_CLUSTER;
486 conflict = !cluster_list_empty(&si->free_clusters) &&
487 offset != cluster_list_first(&si->free_clusters) &&
488 cluster_is_free(&si->cluster_info[offset]);
493 percpu_cluster = this_cpu_ptr(si->percpu_cluster);
494 cluster_set_null(&percpu_cluster->index);
499 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
500 * might involve allocating a new cluster for current CPU too.
502 static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
503 unsigned long *offset, unsigned long *scan_base)
505 struct percpu_cluster *cluster;
506 struct swap_cluster_info *ci;
508 unsigned long tmp, max;
511 cluster = this_cpu_ptr(si->percpu_cluster);
512 if (cluster_is_null(&cluster->index)) {
513 if (!cluster_list_empty(&si->free_clusters)) {
514 cluster->index = si->free_clusters.head;
515 cluster->next = cluster_next(&cluster->index) *
517 } else if (!cluster_list_empty(&si->discard_clusters)) {
519 * we don't have free cluster but have some clusters in
520 * discarding, do discard now and reclaim them
522 swap_do_scheduled_discard(si);
523 *scan_base = *offset = si->cluster_next;
532 * Other CPUs can use our cluster if they can't find a free cluster,
533 * check if there is still free entry in the cluster
536 max = min_t(unsigned long, si->max,
537 (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER);
539 cluster_set_null(&cluster->index);
542 ci = lock_cluster(si, tmp);
544 if (!si->swap_map[tmp]) {
552 cluster_set_null(&cluster->index);
555 cluster->next = tmp + 1;
561 static int scan_swap_map_slots(struct swap_info_struct *si,
562 unsigned char usage, int nr,
565 struct swap_cluster_info *ci;
566 unsigned long offset;
567 unsigned long scan_base;
568 unsigned long last_in_cluster = 0;
569 int latency_ration = LATENCY_LIMIT;
576 * We try to cluster swap pages by allocating them sequentially
577 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
578 * way, however, we resort to first-free allocation, starting
579 * a new cluster. This prevents us from scattering swap pages
580 * all over the entire swap partition, so that we reduce
581 * overall disk seek times between swap pages. -- sct
582 * But we do now try to find an empty cluster. -Andrea
583 * And we let swap pages go all over an SSD partition. Hugh
586 si->flags += SWP_SCANNING;
587 scan_base = offset = si->cluster_next;
590 if (si->cluster_info) {
591 if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
597 if (unlikely(!si->cluster_nr--)) {
598 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
599 si->cluster_nr = SWAPFILE_CLUSTER - 1;
603 spin_unlock(&si->lock);
606 * If seek is expensive, start searching for new cluster from
607 * start of partition, to minimize the span of allocated swap.
608 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
609 * case, just handled by scan_swap_map_try_ssd_cluster() above.
611 scan_base = offset = si->lowest_bit;
612 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
614 /* Locate the first empty (unaligned) cluster */
615 for (; last_in_cluster <= si->highest_bit; offset++) {
616 if (si->swap_map[offset])
617 last_in_cluster = offset + SWAPFILE_CLUSTER;
618 else if (offset == last_in_cluster) {
619 spin_lock(&si->lock);
620 offset -= SWAPFILE_CLUSTER - 1;
621 si->cluster_next = offset;
622 si->cluster_nr = SWAPFILE_CLUSTER - 1;
625 if (unlikely(--latency_ration < 0)) {
627 latency_ration = LATENCY_LIMIT;
632 spin_lock(&si->lock);
633 si->cluster_nr = SWAPFILE_CLUSTER - 1;
637 if (si->cluster_info) {
638 while (scan_swap_map_ssd_cluster_conflict(si, offset)) {
639 /* take a break if we already got some slots */
642 if (!scan_swap_map_try_ssd_cluster(si, &offset,
647 if (!(si->flags & SWP_WRITEOK))
649 if (!si->highest_bit)
651 if (offset > si->highest_bit)
652 scan_base = offset = si->lowest_bit;
654 ci = lock_cluster(si, offset);
655 /* reuse swap entry of cache-only swap if not busy. */
656 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
659 spin_unlock(&si->lock);
660 swap_was_freed = __try_to_reclaim_swap(si, offset);
661 spin_lock(&si->lock);
662 /* entry was freed successfully, try to use this again */
665 goto scan; /* check next one */
668 if (si->swap_map[offset]) {
676 if (offset == si->lowest_bit)
678 if (offset == si->highest_bit)
681 if (si->inuse_pages == si->pages) {
682 si->lowest_bit = si->max;
684 spin_lock(&swap_avail_lock);
685 plist_del(&si->avail_list, &swap_avail_head);
686 spin_unlock(&swap_avail_lock);
688 si->swap_map[offset] = usage;
689 inc_cluster_info_page(si, si->cluster_info, offset);
691 si->cluster_next = offset + 1;
692 slots[n_ret++] = swp_entry(si->type, offset);
694 /* got enough slots or reach max slots? */
695 if ((n_ret == nr) || (offset >= si->highest_bit))
698 /* search for next available slot */
700 /* time to take a break? */
701 if (unlikely(--latency_ration < 0)) {
704 spin_unlock(&si->lock);
706 spin_lock(&si->lock);
707 latency_ration = LATENCY_LIMIT;
710 /* try to get more slots in cluster */
711 if (si->cluster_info) {
712 if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
720 /* non-ssd case, still more slots in cluster? */
721 if (si->cluster_nr && !si->swap_map[offset]) {
727 si->flags -= SWP_SCANNING;
731 spin_unlock(&si->lock);
732 while (++offset <= si->highest_bit) {
733 if (!si->swap_map[offset]) {
734 spin_lock(&si->lock);
737 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
738 spin_lock(&si->lock);
741 if (unlikely(--latency_ration < 0)) {
743 latency_ration = LATENCY_LIMIT;
746 offset = si->lowest_bit;
747 while (offset < scan_base) {
748 if (!si->swap_map[offset]) {
749 spin_lock(&si->lock);
752 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
753 spin_lock(&si->lock);
756 if (unlikely(--latency_ration < 0)) {
758 latency_ration = LATENCY_LIMIT;
762 spin_lock(&si->lock);
765 si->flags -= SWP_SCANNING;
769 static unsigned long scan_swap_map(struct swap_info_struct *si,
775 n_ret = scan_swap_map_slots(si, usage, 1, &entry);
778 return swp_offset(entry);
784 int get_swap_pages(int n_goal, swp_entry_t swp_entries[])
786 struct swap_info_struct *si, *next;
790 avail_pgs = atomic_long_read(&nr_swap_pages);
794 if (n_goal > SWAP_BATCH)
797 if (n_goal > avail_pgs)
800 atomic_long_sub(n_goal, &nr_swap_pages);
802 spin_lock(&swap_avail_lock);
805 plist_for_each_entry_safe(si, next, &swap_avail_head, avail_list) {
806 /* requeue si to after same-priority siblings */
807 plist_requeue(&si->avail_list, &swap_avail_head);
808 spin_unlock(&swap_avail_lock);
809 spin_lock(&si->lock);
810 if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) {
811 spin_lock(&swap_avail_lock);
812 if (plist_node_empty(&si->avail_list)) {
813 spin_unlock(&si->lock);
816 WARN(!si->highest_bit,
817 "swap_info %d in list but !highest_bit\n",
819 WARN(!(si->flags & SWP_WRITEOK),
820 "swap_info %d in list but !SWP_WRITEOK\n",
822 plist_del(&si->avail_list, &swap_avail_head);
823 spin_unlock(&si->lock);
826 n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE,
827 n_goal, swp_entries);
828 spin_unlock(&si->lock);
831 pr_debug("scan_swap_map of si %d failed to find offset\n",
834 spin_lock(&swap_avail_lock);
837 * if we got here, it's likely that si was almost full before,
838 * and since scan_swap_map() can drop the si->lock, multiple
839 * callers probably all tried to get a page from the same si
840 * and it filled up before we could get one; or, the si filled
841 * up between us dropping swap_avail_lock and taking si->lock.
842 * Since we dropped the swap_avail_lock, the swap_avail_head
843 * list may have been modified; so if next is still in the
844 * swap_avail_head list then try it, otherwise start over
845 * if we have not gotten any slots.
847 if (plist_node_empty(&next->avail_list))
851 spin_unlock(&swap_avail_lock);
855 atomic_long_add((long) (n_goal-n_ret), &nr_swap_pages);
860 /* The only caller of this function is now suspend routine */
861 swp_entry_t get_swap_page_of_type(int type)
863 struct swap_info_struct *si;
866 si = swap_info[type];
867 spin_lock(&si->lock);
868 if (si && (si->flags & SWP_WRITEOK)) {
869 atomic_long_dec(&nr_swap_pages);
870 /* This is called for allocating swap entry, not cache */
871 offset = scan_swap_map(si, 1);
873 spin_unlock(&si->lock);
874 return swp_entry(type, offset);
876 atomic_long_inc(&nr_swap_pages);
878 spin_unlock(&si->lock);
879 return (swp_entry_t) {0};
882 static struct swap_info_struct *__swap_info_get(swp_entry_t entry)
884 struct swap_info_struct *p;
885 unsigned long offset, type;
889 type = swp_type(entry);
890 if (type >= nr_swapfiles)
893 if (!(p->flags & SWP_USED))
895 offset = swp_offset(entry);
896 if (offset >= p->max)
901 pr_err("swap_info_get: %s%08lx\n", Bad_offset, entry.val);
904 pr_err("swap_info_get: %s%08lx\n", Unused_file, entry.val);
907 pr_err("swap_info_get: %s%08lx\n", Bad_file, entry.val);
912 static struct swap_info_struct *_swap_info_get(swp_entry_t entry)
914 struct swap_info_struct *p;
916 p = __swap_info_get(entry);
919 if (!p->swap_map[swp_offset(entry)])
924 pr_err("swap_info_get: %s%08lx\n", Unused_offset, entry.val);
930 static struct swap_info_struct *swap_info_get(swp_entry_t entry)
932 struct swap_info_struct *p;
934 p = _swap_info_get(entry);
940 static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry,
941 struct swap_info_struct *q)
943 struct swap_info_struct *p;
945 p = _swap_info_get(entry);
949 spin_unlock(&q->lock);
956 static unsigned char __swap_entry_free(struct swap_info_struct *p,
957 swp_entry_t entry, unsigned char usage)
959 struct swap_cluster_info *ci;
960 unsigned long offset = swp_offset(entry);
962 unsigned char has_cache;
964 ci = lock_cluster_or_swap_info(p, offset);
966 count = p->swap_map[offset];
968 has_cache = count & SWAP_HAS_CACHE;
969 count &= ~SWAP_HAS_CACHE;
971 if (usage == SWAP_HAS_CACHE) {
972 VM_BUG_ON(!has_cache);
974 } else if (count == SWAP_MAP_SHMEM) {
976 * Or we could insist on shmem.c using a special
977 * swap_shmem_free() and free_shmem_swap_and_cache()...
980 } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
981 if (count == COUNT_CONTINUED) {
982 if (swap_count_continued(p, offset, count))
983 count = SWAP_MAP_MAX | COUNT_CONTINUED;
985 count = SWAP_MAP_MAX;
990 usage = count | has_cache;
991 p->swap_map[offset] = usage ? : SWAP_HAS_CACHE;
993 unlock_cluster_or_swap_info(p, ci);
998 static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry)
1000 struct swap_cluster_info *ci;
1001 unsigned long offset = swp_offset(entry);
1002 unsigned char count;
1004 ci = lock_cluster(p, offset);
1005 count = p->swap_map[offset];
1006 VM_BUG_ON(count != SWAP_HAS_CACHE);
1007 p->swap_map[offset] = 0;
1008 dec_cluster_info_page(p, p->cluster_info, offset);
1011 mem_cgroup_uncharge_swap(entry);
1012 if (offset < p->lowest_bit)
1013 p->lowest_bit = offset;
1014 if (offset > p->highest_bit) {
1015 bool was_full = !p->highest_bit;
1017 p->highest_bit = offset;
1018 if (was_full && (p->flags & SWP_WRITEOK)) {
1019 spin_lock(&swap_avail_lock);
1020 WARN_ON(!plist_node_empty(&p->avail_list));
1021 if (plist_node_empty(&p->avail_list))
1022 plist_add(&p->avail_list,
1024 spin_unlock(&swap_avail_lock);
1027 atomic_long_inc(&nr_swap_pages);
1029 frontswap_invalidate_page(p->type, offset);
1030 if (p->flags & SWP_BLKDEV) {
1031 struct gendisk *disk = p->bdev->bd_disk;
1033 if (disk->fops->swap_slot_free_notify)
1034 disk->fops->swap_slot_free_notify(p->bdev,
1040 * Caller has made sure that the swap device corresponding to entry
1041 * is still around or has not been recycled.
1043 void swap_free(swp_entry_t entry)
1045 struct swap_info_struct *p;
1047 p = _swap_info_get(entry);
1049 if (!__swap_entry_free(p, entry, 1))
1050 free_swap_slot(entry);
1055 * Called after dropping swapcache to decrease refcnt to swap entries.
1057 void swapcache_free(swp_entry_t entry)
1059 struct swap_info_struct *p;
1061 p = _swap_info_get(entry);
1063 if (!__swap_entry_free(p, entry, SWAP_HAS_CACHE))
1064 free_swap_slot(entry);
1068 void swapcache_free_entries(swp_entry_t *entries, int n)
1070 struct swap_info_struct *p, *prev;
1078 for (i = 0; i < n; ++i) {
1079 p = swap_info_get_cont(entries[i], prev);
1081 swap_entry_free(p, entries[i]);
1087 spin_unlock(&p->lock);
1091 * How many references to page are currently swapped out?
1092 * This does not give an exact answer when swap count is continued,
1093 * but does include the high COUNT_CONTINUED flag to allow for that.
1095 int page_swapcount(struct page *page)
1098 struct swap_info_struct *p;
1099 struct swap_cluster_info *ci;
1101 unsigned long offset;
1103 entry.val = page_private(page);
1104 p = _swap_info_get(entry);
1106 offset = swp_offset(entry);
1107 ci = lock_cluster_or_swap_info(p, offset);
1108 count = swap_count(p->swap_map[offset]);
1109 unlock_cluster_or_swap_info(p, ci);
1114 static int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry)
1117 pgoff_t offset = swp_offset(entry);
1118 struct swap_cluster_info *ci;
1120 ci = lock_cluster_or_swap_info(si, offset);
1121 count = swap_count(si->swap_map[offset]);
1122 unlock_cluster_or_swap_info(si, ci);
1127 * How many references to @entry are currently swapped out?
1128 * This does not give an exact answer when swap count is continued,
1129 * but does include the high COUNT_CONTINUED flag to allow for that.
1131 int __swp_swapcount(swp_entry_t entry)
1134 struct swap_info_struct *si;
1136 si = __swap_info_get(entry);
1138 count = swap_swapcount(si, entry);
1143 * How many references to @entry are currently swapped out?
1144 * This considers COUNT_CONTINUED so it returns exact answer.
1146 int swp_swapcount(swp_entry_t entry)
1148 int count, tmp_count, n;
1149 struct swap_info_struct *p;
1150 struct swap_cluster_info *ci;
1155 p = _swap_info_get(entry);
1159 offset = swp_offset(entry);
1161 ci = lock_cluster_or_swap_info(p, offset);
1163 count = swap_count(p->swap_map[offset]);
1164 if (!(count & COUNT_CONTINUED))
1167 count &= ~COUNT_CONTINUED;
1168 n = SWAP_MAP_MAX + 1;
1170 page = vmalloc_to_page(p->swap_map + offset);
1171 offset &= ~PAGE_MASK;
1172 VM_BUG_ON(page_private(page) != SWP_CONTINUED);
1175 page = list_next_entry(page, lru);
1176 map = kmap_atomic(page);
1177 tmp_count = map[offset];
1180 count += (tmp_count & ~COUNT_CONTINUED) * n;
1181 n *= (SWAP_CONT_MAX + 1);
1182 } while (tmp_count & COUNT_CONTINUED);
1184 unlock_cluster_or_swap_info(p, ci);
1189 * We can write to an anon page without COW if there are no other references
1190 * to it. And as a side-effect, free up its swap: because the old content
1191 * on disk will never be read, and seeking back there to write new content
1192 * later would only waste time away from clustering.
1194 * NOTE: total_mapcount should not be relied upon by the caller if
1195 * reuse_swap_page() returns false, but it may be always overwritten
1196 * (see the other implementation for CONFIG_SWAP=n).
1198 bool reuse_swap_page(struct page *page, int *total_mapcount)
1202 VM_BUG_ON_PAGE(!PageLocked(page), page);
1203 if (unlikely(PageKsm(page)))
1205 count = page_trans_huge_mapcount(page, total_mapcount);
1206 if (count <= 1 && PageSwapCache(page)) {
1207 count += page_swapcount(page);
1210 if (!PageWriteback(page)) {
1211 delete_from_swap_cache(page);
1215 struct swap_info_struct *p;
1217 entry.val = page_private(page);
1218 p = swap_info_get(entry);
1219 if (p->flags & SWP_STABLE_WRITES) {
1220 spin_unlock(&p->lock);
1223 spin_unlock(&p->lock);
1231 * If swap is getting full, or if there are no more mappings of this page,
1232 * then try_to_free_swap is called to free its swap space.
1234 int try_to_free_swap(struct page *page)
1236 VM_BUG_ON_PAGE(!PageLocked(page), page);
1238 if (!PageSwapCache(page))
1240 if (PageWriteback(page))
1242 if (page_swapcount(page))
1246 * Once hibernation has begun to create its image of memory,
1247 * there's a danger that one of the calls to try_to_free_swap()
1248 * - most probably a call from __try_to_reclaim_swap() while
1249 * hibernation is allocating its own swap pages for the image,
1250 * but conceivably even a call from memory reclaim - will free
1251 * the swap from a page which has already been recorded in the
1252 * image as a clean swapcache page, and then reuse its swap for
1253 * another page of the image. On waking from hibernation, the
1254 * original page might be freed under memory pressure, then
1255 * later read back in from swap, now with the wrong data.
1257 * Hibernation suspends storage while it is writing the image
1258 * to disk so check that here.
1260 if (pm_suspended_storage())
1263 delete_from_swap_cache(page);
1269 * Free the swap entry like above, but also try to
1270 * free the page cache entry if it is the last user.
1272 int free_swap_and_cache(swp_entry_t entry)
1274 struct swap_info_struct *p;
1275 struct page *page = NULL;
1276 unsigned char count;
1278 if (non_swap_entry(entry))
1281 p = _swap_info_get(entry);
1283 count = __swap_entry_free(p, entry, 1);
1284 if (count == SWAP_HAS_CACHE) {
1285 page = find_get_page(swap_address_space(entry),
1287 if (page && !trylock_page(page)) {
1292 free_swap_slot(entry);
1296 * Not mapped elsewhere, or swap space full? Free it!
1297 * Also recheck PageSwapCache now page is locked (above).
1299 if (PageSwapCache(page) && !PageWriteback(page) &&
1300 (!page_mapped(page) || mem_cgroup_swap_full(page)) &&
1301 !swap_swapcount(p, entry)) {
1302 delete_from_swap_cache(page);
1311 #ifdef CONFIG_HIBERNATION
1313 * Find the swap type that corresponds to given device (if any).
1315 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1316 * from 0, in which the swap header is expected to be located.
1318 * This is needed for the suspend to disk (aka swsusp).
1320 int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
1322 struct block_device *bdev = NULL;
1326 bdev = bdget(device);
1328 spin_lock(&swap_lock);
1329 for (type = 0; type < nr_swapfiles; type++) {
1330 struct swap_info_struct *sis = swap_info[type];
1332 if (!(sis->flags & SWP_WRITEOK))
1337 *bdev_p = bdgrab(sis->bdev);
1339 spin_unlock(&swap_lock);
1342 if (bdev == sis->bdev) {
1343 struct swap_extent *se = &sis->first_swap_extent;
1345 if (se->start_block == offset) {
1347 *bdev_p = bdgrab(sis->bdev);
1349 spin_unlock(&swap_lock);
1355 spin_unlock(&swap_lock);
1363 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1364 * corresponding to given index in swap_info (swap type).
1366 sector_t swapdev_block(int type, pgoff_t offset)
1368 struct block_device *bdev;
1370 if ((unsigned int)type >= nr_swapfiles)
1372 if (!(swap_info[type]->flags & SWP_WRITEOK))
1374 return map_swap_entry(swp_entry(type, offset), &bdev);
1378 * Return either the total number of swap pages of given type, or the number
1379 * of free pages of that type (depending on @free)
1381 * This is needed for software suspend
1383 unsigned int count_swap_pages(int type, int free)
1387 spin_lock(&swap_lock);
1388 if ((unsigned int)type < nr_swapfiles) {
1389 struct swap_info_struct *sis = swap_info[type];
1391 spin_lock(&sis->lock);
1392 if (sis->flags & SWP_WRITEOK) {
1395 n -= sis->inuse_pages;
1397 spin_unlock(&sis->lock);
1399 spin_unlock(&swap_lock);
1402 #endif /* CONFIG_HIBERNATION */
1404 static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte)
1406 return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte);
1410 * No need to decide whether this PTE shares the swap entry with others,
1411 * just let do_wp_page work it out if a write is requested later - to
1412 * force COW, vm_page_prot omits write permission from any private vma.
1414 static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
1415 unsigned long addr, swp_entry_t entry, struct page *page)
1417 struct page *swapcache;
1418 struct mem_cgroup *memcg;
1424 page = ksm_might_need_to_copy(page, vma, addr);
1425 if (unlikely(!page))
1428 if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL,
1434 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1435 if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) {
1436 mem_cgroup_cancel_charge(page, memcg, false);
1441 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1442 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
1444 set_pte_at(vma->vm_mm, addr, pte,
1445 pte_mkold(mk_pte(page, vma->vm_page_prot)));
1446 if (page == swapcache) {
1447 page_add_anon_rmap(page, vma, addr, false);
1448 mem_cgroup_commit_charge(page, memcg, true, false);
1449 } else { /* ksm created a completely new copy */
1450 page_add_new_anon_rmap(page, vma, addr, false);
1451 mem_cgroup_commit_charge(page, memcg, false, false);
1452 lru_cache_add_active_or_unevictable(page, vma);
1456 * Move the page to the active list so it is not
1457 * immediately swapped out again after swapon.
1459 activate_page(page);
1461 pte_unmap_unlock(pte, ptl);
1463 if (page != swapcache) {
1470 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1471 unsigned long addr, unsigned long end,
1472 swp_entry_t entry, struct page *page)
1474 pte_t swp_pte = swp_entry_to_pte(entry);
1479 * We don't actually need pte lock while scanning for swp_pte: since
1480 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
1481 * page table while we're scanning; though it could get zapped, and on
1482 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
1483 * of unmatched parts which look like swp_pte, so unuse_pte must
1484 * recheck under pte lock. Scanning without pte lock lets it be
1485 * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
1487 pte = pte_offset_map(pmd, addr);
1490 * swapoff spends a _lot_ of time in this loop!
1491 * Test inline before going to call unuse_pte.
1493 if (unlikely(pte_same_as_swp(*pte, swp_pte))) {
1495 ret = unuse_pte(vma, pmd, addr, entry, page);
1498 pte = pte_offset_map(pmd, addr);
1500 } while (pte++, addr += PAGE_SIZE, addr != end);
1506 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1507 unsigned long addr, unsigned long end,
1508 swp_entry_t entry, struct page *page)
1514 pmd = pmd_offset(pud, addr);
1517 next = pmd_addr_end(addr, end);
1518 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
1520 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
1523 } while (pmd++, addr = next, addr != end);
1527 static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d,
1528 unsigned long addr, unsigned long end,
1529 swp_entry_t entry, struct page *page)
1535 pud = pud_offset(p4d, addr);
1537 next = pud_addr_end(addr, end);
1538 if (pud_none_or_clear_bad(pud))
1540 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
1543 } while (pud++, addr = next, addr != end);
1547 static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd,
1548 unsigned long addr, unsigned long end,
1549 swp_entry_t entry, struct page *page)
1555 p4d = p4d_offset(pgd, addr);
1557 next = p4d_addr_end(addr, end);
1558 if (p4d_none_or_clear_bad(p4d))
1560 ret = unuse_pud_range(vma, p4d, addr, next, entry, page);
1563 } while (p4d++, addr = next, addr != end);
1567 static int unuse_vma(struct vm_area_struct *vma,
1568 swp_entry_t entry, struct page *page)
1571 unsigned long addr, end, next;
1574 if (page_anon_vma(page)) {
1575 addr = page_address_in_vma(page, vma);
1576 if (addr == -EFAULT)
1579 end = addr + PAGE_SIZE;
1581 addr = vma->vm_start;
1585 pgd = pgd_offset(vma->vm_mm, addr);
1587 next = pgd_addr_end(addr, end);
1588 if (pgd_none_or_clear_bad(pgd))
1590 ret = unuse_p4d_range(vma, pgd, addr, next, entry, page);
1593 } while (pgd++, addr = next, addr != end);
1597 static int unuse_mm(struct mm_struct *mm,
1598 swp_entry_t entry, struct page *page)
1600 struct vm_area_struct *vma;
1603 if (!down_read_trylock(&mm->mmap_sem)) {
1605 * Activate page so shrink_inactive_list is unlikely to unmap
1606 * its ptes while lock is dropped, so swapoff can make progress.
1608 activate_page(page);
1610 down_read(&mm->mmap_sem);
1613 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1614 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
1618 up_read(&mm->mmap_sem);
1619 return (ret < 0)? ret: 0;
1623 * Scan swap_map (or frontswap_map if frontswap parameter is true)
1624 * from current position to next entry still in use.
1625 * Recycle to start on reaching the end, returning 0 when empty.
1627 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
1628 unsigned int prev, bool frontswap)
1630 unsigned int max = si->max;
1631 unsigned int i = prev;
1632 unsigned char count;
1635 * No need for swap_lock here: we're just looking
1636 * for whether an entry is in use, not modifying it; false
1637 * hits are okay, and sys_swapoff() has already prevented new
1638 * allocations from this area (while holding swap_lock).
1647 * No entries in use at top of swap_map,
1648 * loop back to start and recheck there.
1654 count = READ_ONCE(si->swap_map[i]);
1655 if (count && swap_count(count) != SWAP_MAP_BAD)
1656 if (!frontswap || frontswap_test(si, i))
1658 if ((i % LATENCY_LIMIT) == 0)
1665 * We completely avoid races by reading each swap page in advance,
1666 * and then search for the process using it. All the necessary
1667 * page table adjustments can then be made atomically.
1669 * if the boolean frontswap is true, only unuse pages_to_unuse pages;
1670 * pages_to_unuse==0 means all pages; ignored if frontswap is false
1672 int try_to_unuse(unsigned int type, bool frontswap,
1673 unsigned long pages_to_unuse)
1675 struct swap_info_struct *si = swap_info[type];
1676 struct mm_struct *start_mm;
1677 volatile unsigned char *swap_map; /* swap_map is accessed without
1678 * locking. Mark it as volatile
1679 * to prevent compiler doing
1682 unsigned char swcount;
1689 * When searching mms for an entry, a good strategy is to
1690 * start at the first mm we freed the previous entry from
1691 * (though actually we don't notice whether we or coincidence
1692 * freed the entry). Initialize this start_mm with a hold.
1694 * A simpler strategy would be to start at the last mm we
1695 * freed the previous entry from; but that would take less
1696 * advantage of mmlist ordering, which clusters forked mms
1697 * together, child after parent. If we race with dup_mmap(), we
1698 * prefer to resolve parent before child, lest we miss entries
1699 * duplicated after we scanned child: using last mm would invert
1702 start_mm = &init_mm;
1706 * Keep on scanning until all entries have gone. Usually,
1707 * one pass through swap_map is enough, but not necessarily:
1708 * there are races when an instance of an entry might be missed.
1710 while ((i = find_next_to_unuse(si, i, frontswap)) != 0) {
1711 if (signal_pending(current)) {
1717 * Get a page for the entry, using the existing swap
1718 * cache page if there is one. Otherwise, get a clean
1719 * page and read the swap into it.
1721 swap_map = &si->swap_map[i];
1722 entry = swp_entry(type, i);
1723 page = read_swap_cache_async(entry,
1724 GFP_HIGHUSER_MOVABLE, NULL, 0);
1727 * Either swap_duplicate() failed because entry
1728 * has been freed independently, and will not be
1729 * reused since sys_swapoff() already disabled
1730 * allocation from here, or alloc_page() failed.
1732 swcount = *swap_map;
1734 * We don't hold lock here, so the swap entry could be
1735 * SWAP_MAP_BAD (when the cluster is discarding).
1736 * Instead of fail out, We can just skip the swap
1737 * entry because swapoff will wait for discarding
1740 if (!swcount || swcount == SWAP_MAP_BAD)
1747 * Don't hold on to start_mm if it looks like exiting.
1749 if (atomic_read(&start_mm->mm_users) == 1) {
1751 start_mm = &init_mm;
1756 * Wait for and lock page. When do_swap_page races with
1757 * try_to_unuse, do_swap_page can handle the fault much
1758 * faster than try_to_unuse can locate the entry. This
1759 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1760 * defer to do_swap_page in such a case - in some tests,
1761 * do_swap_page and try_to_unuse repeatedly compete.
1763 wait_on_page_locked(page);
1764 wait_on_page_writeback(page);
1766 wait_on_page_writeback(page);
1769 * Remove all references to entry.
1771 swcount = *swap_map;
1772 if (swap_count(swcount) == SWAP_MAP_SHMEM) {
1773 retval = shmem_unuse(entry, page);
1774 /* page has already been unlocked and released */
1779 if (swap_count(swcount) && start_mm != &init_mm)
1780 retval = unuse_mm(start_mm, entry, page);
1782 if (swap_count(*swap_map)) {
1783 int set_start_mm = (*swap_map >= swcount);
1784 struct list_head *p = &start_mm->mmlist;
1785 struct mm_struct *new_start_mm = start_mm;
1786 struct mm_struct *prev_mm = start_mm;
1787 struct mm_struct *mm;
1789 mmget(new_start_mm);
1791 spin_lock(&mmlist_lock);
1792 while (swap_count(*swap_map) && !retval &&
1793 (p = p->next) != &start_mm->mmlist) {
1794 mm = list_entry(p, struct mm_struct, mmlist);
1795 if (!mmget_not_zero(mm))
1797 spin_unlock(&mmlist_lock);
1803 swcount = *swap_map;
1804 if (!swap_count(swcount)) /* any usage ? */
1806 else if (mm == &init_mm)
1809 retval = unuse_mm(mm, entry, page);
1811 if (set_start_mm && *swap_map < swcount) {
1812 mmput(new_start_mm);
1817 spin_lock(&mmlist_lock);
1819 spin_unlock(&mmlist_lock);
1822 start_mm = new_start_mm;
1831 * If a reference remains (rare), we would like to leave
1832 * the page in the swap cache; but try_to_unmap could
1833 * then re-duplicate the entry once we drop page lock,
1834 * so we might loop indefinitely; also, that page could
1835 * not be swapped out to other storage meanwhile. So:
1836 * delete from cache even if there's another reference,
1837 * after ensuring that the data has been saved to disk -
1838 * since if the reference remains (rarer), it will be
1839 * read from disk into another page. Splitting into two
1840 * pages would be incorrect if swap supported "shared
1841 * private" pages, but they are handled by tmpfs files.
1843 * Given how unuse_vma() targets one particular offset
1844 * in an anon_vma, once the anon_vma has been determined,
1845 * this splitting happens to be just what is needed to
1846 * handle where KSM pages have been swapped out: re-reading
1847 * is unnecessarily slow, but we can fix that later on.
1849 if (swap_count(*swap_map) &&
1850 PageDirty(page) && PageSwapCache(page)) {
1851 struct writeback_control wbc = {
1852 .sync_mode = WB_SYNC_NONE,
1855 swap_writepage(page, &wbc);
1857 wait_on_page_writeback(page);
1861 * It is conceivable that a racing task removed this page from
1862 * swap cache just before we acquired the page lock at the top,
1863 * or while we dropped it in unuse_mm(). The page might even
1864 * be back in swap cache on another swap area: that we must not
1865 * delete, since it may not have been written out to swap yet.
1867 if (PageSwapCache(page) &&
1868 likely(page_private(page) == entry.val))
1869 delete_from_swap_cache(page);
1872 * So we could skip searching mms once swap count went
1873 * to 1, we did not mark any present ptes as dirty: must
1874 * mark page dirty so shrink_page_list will preserve it.
1881 * Make sure that we aren't completely killing
1882 * interactive performance.
1885 if (frontswap && pages_to_unuse > 0) {
1886 if (!--pages_to_unuse)
1896 * After a successful try_to_unuse, if no swap is now in use, we know
1897 * we can empty the mmlist. swap_lock must be held on entry and exit.
1898 * Note that mmlist_lock nests inside swap_lock, and an mm must be
1899 * added to the mmlist just after page_duplicate - before would be racy.
1901 static void drain_mmlist(void)
1903 struct list_head *p, *next;
1906 for (type = 0; type < nr_swapfiles; type++)
1907 if (swap_info[type]->inuse_pages)
1909 spin_lock(&mmlist_lock);
1910 list_for_each_safe(p, next, &init_mm.mmlist)
1912 spin_unlock(&mmlist_lock);
1916 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
1917 * corresponds to page offset for the specified swap entry.
1918 * Note that the type of this function is sector_t, but it returns page offset
1919 * into the bdev, not sector offset.
1921 static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
1923 struct swap_info_struct *sis;
1924 struct swap_extent *start_se;
1925 struct swap_extent *se;
1928 sis = swap_info[swp_type(entry)];
1931 offset = swp_offset(entry);
1932 start_se = sis->curr_swap_extent;
1936 if (se->start_page <= offset &&
1937 offset < (se->start_page + se->nr_pages)) {
1938 return se->start_block + (offset - se->start_page);
1940 se = list_next_entry(se, list);
1941 sis->curr_swap_extent = se;
1942 BUG_ON(se == start_se); /* It *must* be present */
1947 * Returns the page offset into bdev for the specified page's swap entry.
1949 sector_t map_swap_page(struct page *page, struct block_device **bdev)
1952 entry.val = page_private(page);
1953 return map_swap_entry(entry, bdev);
1957 * Free all of a swapdev's extent information
1959 static void destroy_swap_extents(struct swap_info_struct *sis)
1961 while (!list_empty(&sis->first_swap_extent.list)) {
1962 struct swap_extent *se;
1964 se = list_first_entry(&sis->first_swap_extent.list,
1965 struct swap_extent, list);
1966 list_del(&se->list);
1970 if (sis->flags & SWP_FILE) {
1971 struct file *swap_file = sis->swap_file;
1972 struct address_space *mapping = swap_file->f_mapping;
1974 sis->flags &= ~SWP_FILE;
1975 mapping->a_ops->swap_deactivate(swap_file);
1980 * Add a block range (and the corresponding page range) into this swapdev's
1981 * extent list. The extent list is kept sorted in page order.
1983 * This function rather assumes that it is called in ascending page order.
1986 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1987 unsigned long nr_pages, sector_t start_block)
1989 struct swap_extent *se;
1990 struct swap_extent *new_se;
1991 struct list_head *lh;
1993 if (start_page == 0) {
1994 se = &sis->first_swap_extent;
1995 sis->curr_swap_extent = se;
1997 se->nr_pages = nr_pages;
1998 se->start_block = start_block;
2001 lh = sis->first_swap_extent.list.prev; /* Highest extent */
2002 se = list_entry(lh, struct swap_extent, list);
2003 BUG_ON(se->start_page + se->nr_pages != start_page);
2004 if (se->start_block + se->nr_pages == start_block) {
2006 se->nr_pages += nr_pages;
2012 * No merge. Insert a new extent, preserving ordering.
2014 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
2017 new_se->start_page = start_page;
2018 new_se->nr_pages = nr_pages;
2019 new_se->start_block = start_block;
2021 list_add_tail(&new_se->list, &sis->first_swap_extent.list);
2026 * A `swap extent' is a simple thing which maps a contiguous range of pages
2027 * onto a contiguous range of disk blocks. An ordered list of swap extents
2028 * is built at swapon time and is then used at swap_writepage/swap_readpage
2029 * time for locating where on disk a page belongs.
2031 * If the swapfile is an S_ISBLK block device, a single extent is installed.
2032 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2033 * swap files identically.
2035 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2036 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
2037 * swapfiles are handled *identically* after swapon time.
2039 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2040 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
2041 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
2042 * requirements, they are simply tossed out - we will never use those blocks
2045 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
2046 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
2047 * which will scribble on the fs.
2049 * The amount of disk space which a single swap extent represents varies.
2050 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
2051 * extents in the list. To avoid much list walking, we cache the previous
2052 * search location in `curr_swap_extent', and start new searches from there.
2053 * This is extremely effective. The average number of iterations in
2054 * map_swap_page() has been measured at about 0.3 per page. - akpm.
2056 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
2058 struct file *swap_file = sis->swap_file;
2059 struct address_space *mapping = swap_file->f_mapping;
2060 struct inode *inode = mapping->host;
2063 if (S_ISBLK(inode->i_mode)) {
2064 ret = add_swap_extent(sis, 0, sis->max, 0);
2069 if (mapping->a_ops->swap_activate) {
2070 ret = mapping->a_ops->swap_activate(sis, swap_file, span);
2072 sis->flags |= SWP_FILE;
2073 ret = add_swap_extent(sis, 0, sis->max, 0);
2079 return generic_swapfile_activate(sis, swap_file, span);
2082 static void _enable_swap_info(struct swap_info_struct *p, int prio,
2083 unsigned char *swap_map,
2084 struct swap_cluster_info *cluster_info)
2089 p->prio = --least_priority;
2091 * the plist prio is negated because plist ordering is
2092 * low-to-high, while swap ordering is high-to-low
2094 p->list.prio = -p->prio;
2095 p->avail_list.prio = -p->prio;
2096 p->swap_map = swap_map;
2097 p->cluster_info = cluster_info;
2098 p->flags |= SWP_WRITEOK;
2099 atomic_long_add(p->pages, &nr_swap_pages);
2100 total_swap_pages += p->pages;
2102 assert_spin_locked(&swap_lock);
2104 * both lists are plists, and thus priority ordered.
2105 * swap_active_head needs to be priority ordered for swapoff(),
2106 * which on removal of any swap_info_struct with an auto-assigned
2107 * (i.e. negative) priority increments the auto-assigned priority
2108 * of any lower-priority swap_info_structs.
2109 * swap_avail_head needs to be priority ordered for get_swap_page(),
2110 * which allocates swap pages from the highest available priority
2113 plist_add(&p->list, &swap_active_head);
2114 spin_lock(&swap_avail_lock);
2115 plist_add(&p->avail_list, &swap_avail_head);
2116 spin_unlock(&swap_avail_lock);
2119 static void enable_swap_info(struct swap_info_struct *p, int prio,
2120 unsigned char *swap_map,
2121 struct swap_cluster_info *cluster_info,
2122 unsigned long *frontswap_map)
2124 frontswap_init(p->type, frontswap_map);
2125 spin_lock(&swap_lock);
2126 spin_lock(&p->lock);
2127 _enable_swap_info(p, prio, swap_map, cluster_info);
2128 spin_unlock(&p->lock);
2129 spin_unlock(&swap_lock);
2132 static void reinsert_swap_info(struct swap_info_struct *p)
2134 spin_lock(&swap_lock);
2135 spin_lock(&p->lock);
2136 _enable_swap_info(p, p->prio, p->swap_map, p->cluster_info);
2137 spin_unlock(&p->lock);
2138 spin_unlock(&swap_lock);
2141 bool has_usable_swap(void)
2145 spin_lock(&swap_lock);
2146 if (plist_head_empty(&swap_active_head))
2148 spin_unlock(&swap_lock);
2152 SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
2154 struct swap_info_struct *p = NULL;
2155 unsigned char *swap_map;
2156 struct swap_cluster_info *cluster_info;
2157 unsigned long *frontswap_map;
2158 struct file *swap_file, *victim;
2159 struct address_space *mapping;
2160 struct inode *inode;
2161 struct filename *pathname;
2163 unsigned int old_block_size;
2165 if (!capable(CAP_SYS_ADMIN))
2168 BUG_ON(!current->mm);
2170 pathname = getname(specialfile);
2171 if (IS_ERR(pathname))
2172 return PTR_ERR(pathname);
2174 victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
2175 err = PTR_ERR(victim);
2179 mapping = victim->f_mapping;
2180 spin_lock(&swap_lock);
2181 plist_for_each_entry(p, &swap_active_head, list) {
2182 if (p->flags & SWP_WRITEOK) {
2183 if (p->swap_file->f_mapping == mapping) {
2191 spin_unlock(&swap_lock);
2194 if (!security_vm_enough_memory_mm(current->mm, p->pages))
2195 vm_unacct_memory(p->pages);
2198 spin_unlock(&swap_lock);
2201 spin_lock(&swap_avail_lock);
2202 plist_del(&p->avail_list, &swap_avail_head);
2203 spin_unlock(&swap_avail_lock);
2204 spin_lock(&p->lock);
2206 struct swap_info_struct *si = p;
2208 plist_for_each_entry_continue(si, &swap_active_head, list) {
2211 si->avail_list.prio--;
2215 plist_del(&p->list, &swap_active_head);
2216 atomic_long_sub(p->pages, &nr_swap_pages);
2217 total_swap_pages -= p->pages;
2218 p->flags &= ~SWP_WRITEOK;
2219 spin_unlock(&p->lock);
2220 spin_unlock(&swap_lock);
2222 disable_swap_slots_cache_lock();
2224 set_current_oom_origin();
2225 err = try_to_unuse(p->type, false, 0); /* force unuse all pages */
2226 clear_current_oom_origin();
2229 /* re-insert swap space back into swap_list */
2230 reinsert_swap_info(p);
2231 reenable_swap_slots_cache_unlock();
2235 reenable_swap_slots_cache_unlock();
2237 flush_work(&p->discard_work);
2239 destroy_swap_extents(p);
2240 if (p->flags & SWP_CONTINUED)
2241 free_swap_count_continuations(p);
2243 mutex_lock(&swapon_mutex);
2244 spin_lock(&swap_lock);
2245 spin_lock(&p->lock);
2248 /* wait for anyone still in scan_swap_map */
2249 p->highest_bit = 0; /* cuts scans short */
2250 while (p->flags >= SWP_SCANNING) {
2251 spin_unlock(&p->lock);
2252 spin_unlock(&swap_lock);
2253 schedule_timeout_uninterruptible(1);
2254 spin_lock(&swap_lock);
2255 spin_lock(&p->lock);
2258 swap_file = p->swap_file;
2259 old_block_size = p->old_block_size;
2260 p->swap_file = NULL;
2262 swap_map = p->swap_map;
2264 cluster_info = p->cluster_info;
2265 p->cluster_info = NULL;
2266 frontswap_map = frontswap_map_get(p);
2267 spin_unlock(&p->lock);
2268 spin_unlock(&swap_lock);
2269 frontswap_invalidate_area(p->type);
2270 frontswap_map_set(p, NULL);
2271 mutex_unlock(&swapon_mutex);
2272 free_percpu(p->percpu_cluster);
2273 p->percpu_cluster = NULL;
2275 vfree(cluster_info);
2276 vfree(frontswap_map);
2277 /* Destroy swap account information */
2278 swap_cgroup_swapoff(p->type);
2279 exit_swap_address_space(p->type);
2281 inode = mapping->host;
2282 if (S_ISBLK(inode->i_mode)) {
2283 struct block_device *bdev = I_BDEV(inode);
2284 set_blocksize(bdev, old_block_size);
2285 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2288 inode->i_flags &= ~S_SWAPFILE;
2289 inode_unlock(inode);
2291 filp_close(swap_file, NULL);
2294 * Clear the SWP_USED flag after all resources are freed so that swapon
2295 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
2296 * not hold p->lock after we cleared its SWP_WRITEOK.
2298 spin_lock(&swap_lock);
2300 spin_unlock(&swap_lock);
2303 atomic_inc(&proc_poll_event);
2304 wake_up_interruptible(&proc_poll_wait);
2307 filp_close(victim, NULL);
2313 #ifdef CONFIG_PROC_FS
2314 static unsigned swaps_poll(struct file *file, poll_table *wait)
2316 struct seq_file *seq = file->private_data;
2318 poll_wait(file, &proc_poll_wait, wait);
2320 if (seq->poll_event != atomic_read(&proc_poll_event)) {
2321 seq->poll_event = atomic_read(&proc_poll_event);
2322 return POLLIN | POLLRDNORM | POLLERR | POLLPRI;
2325 return POLLIN | POLLRDNORM;
2329 static void *swap_start(struct seq_file *swap, loff_t *pos)
2331 struct swap_info_struct *si;
2335 mutex_lock(&swapon_mutex);
2338 return SEQ_START_TOKEN;
2340 for (type = 0; type < nr_swapfiles; type++) {
2341 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2342 si = swap_info[type];
2343 if (!(si->flags & SWP_USED) || !si->swap_map)
2352 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2354 struct swap_info_struct *si = v;
2357 if (v == SEQ_START_TOKEN)
2360 type = si->type + 1;
2362 for (; type < nr_swapfiles; type++) {
2363 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2364 si = swap_info[type];
2365 if (!(si->flags & SWP_USED) || !si->swap_map)
2374 static void swap_stop(struct seq_file *swap, void *v)
2376 mutex_unlock(&swapon_mutex);
2379 static int swap_show(struct seq_file *swap, void *v)
2381 struct swap_info_struct *si = v;
2385 if (si == SEQ_START_TOKEN) {
2386 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2390 file = si->swap_file;
2391 len = seq_file_path(swap, file, " \t\n\\");
2392 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
2393 len < 40 ? 40 - len : 1, " ",
2394 S_ISBLK(file_inode(file)->i_mode) ?
2395 "partition" : "file\t",
2396 si->pages << (PAGE_SHIFT - 10),
2397 si->inuse_pages << (PAGE_SHIFT - 10),
2402 static const struct seq_operations swaps_op = {
2403 .start = swap_start,
2409 static int swaps_open(struct inode *inode, struct file *file)
2411 struct seq_file *seq;
2414 ret = seq_open(file, &swaps_op);
2418 seq = file->private_data;
2419 seq->poll_event = atomic_read(&proc_poll_event);
2423 static const struct file_operations proc_swaps_operations = {
2426 .llseek = seq_lseek,
2427 .release = seq_release,
2431 static int __init procswaps_init(void)
2433 proc_create("swaps", 0, NULL, &proc_swaps_operations);
2436 __initcall(procswaps_init);
2437 #endif /* CONFIG_PROC_FS */
2439 #ifdef MAX_SWAPFILES_CHECK
2440 static int __init max_swapfiles_check(void)
2442 MAX_SWAPFILES_CHECK();
2445 late_initcall(max_swapfiles_check);
2448 static struct swap_info_struct *alloc_swap_info(void)
2450 struct swap_info_struct *p;
2453 p = kzalloc(sizeof(*p), GFP_KERNEL);
2455 return ERR_PTR(-ENOMEM);
2457 spin_lock(&swap_lock);
2458 for (type = 0; type < nr_swapfiles; type++) {
2459 if (!(swap_info[type]->flags & SWP_USED))
2462 if (type >= MAX_SWAPFILES) {
2463 spin_unlock(&swap_lock);
2465 return ERR_PTR(-EPERM);
2467 if (type >= nr_swapfiles) {
2469 swap_info[type] = p;
2471 * Write swap_info[type] before nr_swapfiles, in case a
2472 * racing procfs swap_start() or swap_next() is reading them.
2473 * (We never shrink nr_swapfiles, we never free this entry.)
2479 p = swap_info[type];
2481 * Do not memset this entry: a racing procfs swap_next()
2482 * would be relying on p->type to remain valid.
2485 INIT_LIST_HEAD(&p->first_swap_extent.list);
2486 plist_node_init(&p->list, 0);
2487 plist_node_init(&p->avail_list, 0);
2488 p->flags = SWP_USED;
2489 spin_unlock(&swap_lock);
2490 spin_lock_init(&p->lock);
2495 static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2499 if (S_ISBLK(inode->i_mode)) {
2500 p->bdev = bdgrab(I_BDEV(inode));
2501 error = blkdev_get(p->bdev,
2502 FMODE_READ | FMODE_WRITE | FMODE_EXCL, p);
2507 p->old_block_size = block_size(p->bdev);
2508 error = set_blocksize(p->bdev, PAGE_SIZE);
2511 p->flags |= SWP_BLKDEV;
2512 } else if (S_ISREG(inode->i_mode)) {
2513 p->bdev = inode->i_sb->s_bdev;
2515 if (IS_SWAPFILE(inode))
2523 static unsigned long read_swap_header(struct swap_info_struct *p,
2524 union swap_header *swap_header,
2525 struct inode *inode)
2528 unsigned long maxpages;
2529 unsigned long swapfilepages;
2530 unsigned long last_page;
2532 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
2533 pr_err("Unable to find swap-space signature\n");
2537 /* swap partition endianess hack... */
2538 if (swab32(swap_header->info.version) == 1) {
2539 swab32s(&swap_header->info.version);
2540 swab32s(&swap_header->info.last_page);
2541 swab32s(&swap_header->info.nr_badpages);
2542 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2544 for (i = 0; i < swap_header->info.nr_badpages; i++)
2545 swab32s(&swap_header->info.badpages[i]);
2547 /* Check the swap header's sub-version */
2548 if (swap_header->info.version != 1) {
2549 pr_warn("Unable to handle swap header version %d\n",
2550 swap_header->info.version);
2555 p->cluster_next = 1;
2559 * Find out how many pages are allowed for a single swap
2560 * device. There are two limiting factors: 1) the number
2561 * of bits for the swap offset in the swp_entry_t type, and
2562 * 2) the number of bits in the swap pte as defined by the
2563 * different architectures. In order to find the
2564 * largest possible bit mask, a swap entry with swap type 0
2565 * and swap offset ~0UL is created, encoded to a swap pte,
2566 * decoded to a swp_entry_t again, and finally the swap
2567 * offset is extracted. This will mask all the bits from
2568 * the initial ~0UL mask that can't be encoded in either
2569 * the swp_entry_t or the architecture definition of a
2572 maxpages = swp_offset(pte_to_swp_entry(
2573 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2574 last_page = swap_header->info.last_page;
2575 if (last_page > maxpages) {
2576 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
2577 maxpages << (PAGE_SHIFT - 10),
2578 last_page << (PAGE_SHIFT - 10));
2580 if (maxpages > last_page) {
2581 maxpages = last_page + 1;
2582 /* p->max is an unsigned int: don't overflow it */
2583 if ((unsigned int)maxpages == 0)
2584 maxpages = UINT_MAX;
2586 p->highest_bit = maxpages - 1;
2590 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2591 if (swapfilepages && maxpages > swapfilepages) {
2592 pr_warn("Swap area shorter than signature indicates\n");
2595 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
2597 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2603 #define SWAP_CLUSTER_INFO_COLS \
2604 DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
2605 #define SWAP_CLUSTER_SPACE_COLS \
2606 DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
2607 #define SWAP_CLUSTER_COLS \
2608 max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
2610 static int setup_swap_map_and_extents(struct swap_info_struct *p,
2611 union swap_header *swap_header,
2612 unsigned char *swap_map,
2613 struct swap_cluster_info *cluster_info,
2614 unsigned long maxpages,
2618 unsigned int nr_good_pages;
2620 unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
2621 unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS;
2622 unsigned long i, idx;
2624 nr_good_pages = maxpages - 1; /* omit header page */
2626 cluster_list_init(&p->free_clusters);
2627 cluster_list_init(&p->discard_clusters);
2629 for (i = 0; i < swap_header->info.nr_badpages; i++) {
2630 unsigned int page_nr = swap_header->info.badpages[i];
2631 if (page_nr == 0 || page_nr > swap_header->info.last_page)
2633 if (page_nr < maxpages) {
2634 swap_map[page_nr] = SWAP_MAP_BAD;
2637 * Haven't marked the cluster free yet, no list
2638 * operation involved
2640 inc_cluster_info_page(p, cluster_info, page_nr);
2644 /* Haven't marked the cluster free yet, no list operation involved */
2645 for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
2646 inc_cluster_info_page(p, cluster_info, i);
2648 if (nr_good_pages) {
2649 swap_map[0] = SWAP_MAP_BAD;
2651 * Not mark the cluster free yet, no list
2652 * operation involved
2654 inc_cluster_info_page(p, cluster_info, 0);
2656 p->pages = nr_good_pages;
2657 nr_extents = setup_swap_extents(p, span);
2660 nr_good_pages = p->pages;
2662 if (!nr_good_pages) {
2663 pr_warn("Empty swap-file\n");
2672 * Reduce false cache line sharing between cluster_info and
2673 * sharing same address space.
2675 for (k = 0; k < SWAP_CLUSTER_COLS; k++) {
2676 j = (k + col) % SWAP_CLUSTER_COLS;
2677 for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) {
2678 idx = i * SWAP_CLUSTER_COLS + j;
2679 if (idx >= nr_clusters)
2681 if (cluster_count(&cluster_info[idx]))
2683 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
2684 cluster_list_add_tail(&p->free_clusters, cluster_info,
2692 * Helper to sys_swapon determining if a given swap
2693 * backing device queue supports DISCARD operations.
2695 static bool swap_discardable(struct swap_info_struct *si)
2697 struct request_queue *q = bdev_get_queue(si->bdev);
2699 if (!q || !blk_queue_discard(q))
2705 SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2707 struct swap_info_struct *p;
2708 struct filename *name;
2709 struct file *swap_file = NULL;
2710 struct address_space *mapping;
2713 union swap_header *swap_header;
2716 unsigned long maxpages;
2717 unsigned char *swap_map = NULL;
2718 struct swap_cluster_info *cluster_info = NULL;
2719 unsigned long *frontswap_map = NULL;
2720 struct page *page = NULL;
2721 struct inode *inode = NULL;
2723 if (swap_flags & ~SWAP_FLAGS_VALID)
2726 if (!capable(CAP_SYS_ADMIN))
2729 p = alloc_swap_info();
2733 INIT_WORK(&p->discard_work, swap_discard_work);
2735 name = getname(specialfile);
2737 error = PTR_ERR(name);
2741 swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
2742 if (IS_ERR(swap_file)) {
2743 error = PTR_ERR(swap_file);
2748 p->swap_file = swap_file;
2749 mapping = swap_file->f_mapping;
2750 inode = mapping->host;
2752 /* If S_ISREG(inode->i_mode) will do inode_lock(inode); */
2753 error = claim_swapfile(p, inode);
2754 if (unlikely(error))
2758 * Read the swap header.
2760 if (!mapping->a_ops->readpage) {
2764 page = read_mapping_page(mapping, 0, swap_file);
2766 error = PTR_ERR(page);
2769 swap_header = kmap(page);
2771 maxpages = read_swap_header(p, swap_header, inode);
2772 if (unlikely(!maxpages)) {
2777 /* OK, set up the swap map and apply the bad block list */
2778 swap_map = vzalloc(maxpages);
2784 if (bdi_cap_stable_pages_required(inode_to_bdi(inode)))
2785 p->flags |= SWP_STABLE_WRITES;
2787 if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) {
2789 unsigned long ci, nr_cluster;
2791 p->flags |= SWP_SOLIDSTATE;
2793 * select a random position to start with to help wear leveling
2796 p->cluster_next = 1 + (prandom_u32() % p->highest_bit);
2797 nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
2799 cluster_info = vzalloc(nr_cluster * sizeof(*cluster_info));
2800 if (!cluster_info) {
2805 for (ci = 0; ci < nr_cluster; ci++)
2806 spin_lock_init(&((cluster_info + ci)->lock));
2808 p->percpu_cluster = alloc_percpu(struct percpu_cluster);
2809 if (!p->percpu_cluster) {
2813 for_each_possible_cpu(cpu) {
2814 struct percpu_cluster *cluster;
2815 cluster = per_cpu_ptr(p->percpu_cluster, cpu);
2816 cluster_set_null(&cluster->index);
2820 error = swap_cgroup_swapon(p->type, maxpages);
2824 nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
2825 cluster_info, maxpages, &span);
2826 if (unlikely(nr_extents < 0)) {
2830 /* frontswap enabled? set up bit-per-page map for frontswap */
2831 if (IS_ENABLED(CONFIG_FRONTSWAP))
2832 frontswap_map = vzalloc(BITS_TO_LONGS(maxpages) * sizeof(long));
2834 if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) {
2836 * When discard is enabled for swap with no particular
2837 * policy flagged, we set all swap discard flags here in
2838 * order to sustain backward compatibility with older
2839 * swapon(8) releases.
2841 p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
2845 * By flagging sys_swapon, a sysadmin can tell us to
2846 * either do single-time area discards only, or to just
2847 * perform discards for released swap page-clusters.
2848 * Now it's time to adjust the p->flags accordingly.
2850 if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
2851 p->flags &= ~SWP_PAGE_DISCARD;
2852 else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
2853 p->flags &= ~SWP_AREA_DISCARD;
2855 /* issue a swapon-time discard if it's still required */
2856 if (p->flags & SWP_AREA_DISCARD) {
2857 int err = discard_swap(p);
2859 pr_err("swapon: discard_swap(%p): %d\n",
2864 error = init_swap_address_space(p->type, maxpages);
2868 mutex_lock(&swapon_mutex);
2870 if (swap_flags & SWAP_FLAG_PREFER)
2872 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
2873 enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
2875 pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
2876 p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
2877 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
2878 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
2879 (p->flags & SWP_DISCARDABLE) ? "D" : "",
2880 (p->flags & SWP_AREA_DISCARD) ? "s" : "",
2881 (p->flags & SWP_PAGE_DISCARD) ? "c" : "",
2882 (frontswap_map) ? "FS" : "");
2884 mutex_unlock(&swapon_mutex);
2885 atomic_inc(&proc_poll_event);
2886 wake_up_interruptible(&proc_poll_wait);
2888 if (S_ISREG(inode->i_mode))
2889 inode->i_flags |= S_SWAPFILE;
2893 free_percpu(p->percpu_cluster);
2894 p->percpu_cluster = NULL;
2895 if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
2896 set_blocksize(p->bdev, p->old_block_size);
2897 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2899 destroy_swap_extents(p);
2900 swap_cgroup_swapoff(p->type);
2901 spin_lock(&swap_lock);
2902 p->swap_file = NULL;
2904 spin_unlock(&swap_lock);
2906 vfree(cluster_info);
2908 if (inode && S_ISREG(inode->i_mode)) {
2909 inode_unlock(inode);
2912 filp_close(swap_file, NULL);
2915 if (page && !IS_ERR(page)) {
2921 if (inode && S_ISREG(inode->i_mode))
2922 inode_unlock(inode);
2924 enable_swap_slots_cache();
2928 void si_swapinfo(struct sysinfo *val)
2931 unsigned long nr_to_be_unused = 0;
2933 spin_lock(&swap_lock);
2934 for (type = 0; type < nr_swapfiles; type++) {
2935 struct swap_info_struct *si = swap_info[type];
2937 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
2938 nr_to_be_unused += si->inuse_pages;
2940 val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
2941 val->totalswap = total_swap_pages + nr_to_be_unused;
2942 spin_unlock(&swap_lock);
2946 * Verify that a swap entry is valid and increment its swap map count.
2948 * Returns error code in following case.
2950 * - swp_entry is invalid -> EINVAL
2951 * - swp_entry is migration entry -> EINVAL
2952 * - swap-cache reference is requested but there is already one. -> EEXIST
2953 * - swap-cache reference is requested but the entry is not used. -> ENOENT
2954 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
2956 static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
2958 struct swap_info_struct *p;
2959 struct swap_cluster_info *ci;
2960 unsigned long offset, type;
2961 unsigned char count;
2962 unsigned char has_cache;
2965 if (non_swap_entry(entry))
2968 type = swp_type(entry);
2969 if (type >= nr_swapfiles)
2971 p = swap_info[type];
2972 offset = swp_offset(entry);
2973 if (unlikely(offset >= p->max))
2976 ci = lock_cluster_or_swap_info(p, offset);
2978 count = p->swap_map[offset];
2981 * swapin_readahead() doesn't check if a swap entry is valid, so the
2982 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
2984 if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
2989 has_cache = count & SWAP_HAS_CACHE;
2990 count &= ~SWAP_HAS_CACHE;
2993 if (usage == SWAP_HAS_CACHE) {
2995 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
2996 if (!has_cache && count)
2997 has_cache = SWAP_HAS_CACHE;
2998 else if (has_cache) /* someone else added cache */
3000 else /* no users remaining */
3003 } else if (count || has_cache) {
3005 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
3007 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
3009 else if (swap_count_continued(p, offset, count))
3010 count = COUNT_CONTINUED;
3014 err = -ENOENT; /* unused swap entry */
3016 p->swap_map[offset] = count | has_cache;
3019 unlock_cluster_or_swap_info(p, ci);
3024 pr_err("swap_dup: %s%08lx\n", Bad_file, entry.val);
3029 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3030 * (in which case its reference count is never incremented).
3032 void swap_shmem_alloc(swp_entry_t entry)
3034 __swap_duplicate(entry, SWAP_MAP_SHMEM);
3038 * Increase reference count of swap entry by 1.
3039 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3040 * but could not be atomically allocated. Returns 0, just as if it succeeded,
3041 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3042 * might occur if a page table entry has got corrupted.
3044 int swap_duplicate(swp_entry_t entry)
3048 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
3049 err = add_swap_count_continuation(entry, GFP_ATOMIC);
3054 * @entry: swap entry for which we allocate swap cache.
3056 * Called when allocating swap cache for existing swap entry,
3057 * This can return error codes. Returns 0 at success.
3058 * -EBUSY means there is a swap cache.
3059 * Note: return code is different from swap_duplicate().
3061 int swapcache_prepare(swp_entry_t entry)
3063 return __swap_duplicate(entry, SWAP_HAS_CACHE);
3066 struct swap_info_struct *page_swap_info(struct page *page)
3068 swp_entry_t swap = { .val = page_private(page) };
3069 return swap_info[swp_type(swap)];
3073 * out-of-line __page_file_ methods to avoid include hell.
3075 struct address_space *__page_file_mapping(struct page *page)
3077 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
3078 return page_swap_info(page)->swap_file->f_mapping;
3080 EXPORT_SYMBOL_GPL(__page_file_mapping);
3082 pgoff_t __page_file_index(struct page *page)
3084 swp_entry_t swap = { .val = page_private(page) };
3085 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
3086 return swp_offset(swap);
3088 EXPORT_SYMBOL_GPL(__page_file_index);
3091 * add_swap_count_continuation - called when a swap count is duplicated
3092 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3093 * page of the original vmalloc'ed swap_map, to hold the continuation count
3094 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
3095 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3097 * These continuation pages are seldom referenced: the common paths all work
3098 * on the original swap_map, only referring to a continuation page when the
3099 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3101 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3102 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3103 * can be called after dropping locks.
3105 int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
3107 struct swap_info_struct *si;
3108 struct swap_cluster_info *ci;
3111 struct page *list_page;
3113 unsigned char count;
3116 * When debugging, it's easier to use __GFP_ZERO here; but it's better
3117 * for latency not to zero a page while GFP_ATOMIC and holding locks.
3119 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
3121 si = swap_info_get(entry);
3124 * An acceptable race has occurred since the failing
3125 * __swap_duplicate(): the swap entry has been freed,
3126 * perhaps even the whole swap_map cleared for swapoff.
3131 offset = swp_offset(entry);
3133 ci = lock_cluster(si, offset);
3135 count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
3137 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
3139 * The higher the swap count, the more likely it is that tasks
3140 * will race to add swap count continuation: we need to avoid
3141 * over-provisioning.
3148 spin_unlock(&si->lock);
3153 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
3154 * no architecture is using highmem pages for kernel page tables: so it
3155 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
3157 head = vmalloc_to_page(si->swap_map + offset);
3158 offset &= ~PAGE_MASK;
3161 * Page allocation does not initialize the page's lru field,
3162 * but it does always reset its private field.
3164 if (!page_private(head)) {
3165 BUG_ON(count & COUNT_CONTINUED);
3166 INIT_LIST_HEAD(&head->lru);
3167 set_page_private(head, SWP_CONTINUED);
3168 si->flags |= SWP_CONTINUED;
3171 list_for_each_entry(list_page, &head->lru, lru) {
3175 * If the previous map said no continuation, but we've found
3176 * a continuation page, free our allocation and use this one.
3178 if (!(count & COUNT_CONTINUED))
3181 map = kmap_atomic(list_page) + offset;
3186 * If this continuation count now has some space in it,
3187 * free our allocation and use this one.
3189 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
3193 list_add_tail(&page->lru, &head->lru);
3194 page = NULL; /* now it's attached, don't free it */
3197 spin_unlock(&si->lock);
3205 * swap_count_continued - when the original swap_map count is incremented
3206 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3207 * into, carry if so, or else fail until a new continuation page is allocated;
3208 * when the original swap_map count is decremented from 0 with continuation,
3209 * borrow from the continuation and report whether it still holds more.
3210 * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3213 static bool swap_count_continued(struct swap_info_struct *si,
3214 pgoff_t offset, unsigned char count)
3220 head = vmalloc_to_page(si->swap_map + offset);
3221 if (page_private(head) != SWP_CONTINUED) {
3222 BUG_ON(count & COUNT_CONTINUED);
3223 return false; /* need to add count continuation */
3226 offset &= ~PAGE_MASK;
3227 page = list_entry(head->lru.next, struct page, lru);
3228 map = kmap_atomic(page) + offset;
3230 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
3231 goto init_map; /* jump over SWAP_CONT_MAX checks */
3233 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
3235 * Think of how you add 1 to 999
3237 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
3239 page = list_entry(page->lru.next, struct page, lru);
3240 BUG_ON(page == head);
3241 map = kmap_atomic(page) + offset;
3243 if (*map == SWAP_CONT_MAX) {
3245 page = list_entry(page->lru.next, struct page, lru);
3247 return false; /* add count continuation */
3248 map = kmap_atomic(page) + offset;
3249 init_map: *map = 0; /* we didn't zero the page */
3253 page = list_entry(page->lru.prev, struct page, lru);
3254 while (page != head) {
3255 map = kmap_atomic(page) + offset;
3256 *map = COUNT_CONTINUED;
3258 page = list_entry(page->lru.prev, struct page, lru);
3260 return true; /* incremented */
3262 } else { /* decrementing */
3264 * Think of how you subtract 1 from 1000
3266 BUG_ON(count != COUNT_CONTINUED);
3267 while (*map == COUNT_CONTINUED) {
3269 page = list_entry(page->lru.next, struct page, lru);
3270 BUG_ON(page == head);
3271 map = kmap_atomic(page) + offset;
3278 page = list_entry(page->lru.prev, struct page, lru);
3279 while (page != head) {
3280 map = kmap_atomic(page) + offset;
3281 *map = SWAP_CONT_MAX | count;
3282 count = COUNT_CONTINUED;
3284 page = list_entry(page->lru.prev, struct page, lru);
3286 return count == COUNT_CONTINUED;
3291 * free_swap_count_continuations - swapoff free all the continuation pages
3292 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3294 static void free_swap_count_continuations(struct swap_info_struct *si)
3298 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
3300 head = vmalloc_to_page(si->swap_map + offset);
3301 if (page_private(head)) {
3302 struct page *page, *next;
3304 list_for_each_entry_safe(page, next, &head->lru, lru) {
3305 list_del(&page->lru);