xref: /openbmc/linux/mm/swapfile.c (revision 4081f7446d95a9d3ced12dc04ff02c187a761e90)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/mm/swapfile.c
4  *
5  *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
6  *  Swap reorganised 29.12.95, Stephen Tweedie
7  */
8 
9 #include <linux/blkdev.h>
10 #include <linux/mm.h>
11 #include <linux/sched/mm.h>
12 #include <linux/sched/task.h>
13 #include <linux/hugetlb.h>
14 #include <linux/mman.h>
15 #include <linux/slab.h>
16 #include <linux/kernel_stat.h>
17 #include <linux/swap.h>
18 #include <linux/vmalloc.h>
19 #include <linux/pagemap.h>
20 #include <linux/namei.h>
21 #include <linux/shmem_fs.h>
22 #include <linux/blk-cgroup.h>
23 #include <linux/random.h>
24 #include <linux/writeback.h>
25 #include <linux/proc_fs.h>
26 #include <linux/seq_file.h>
27 #include <linux/init.h>
28 #include <linux/ksm.h>
29 #include <linux/rmap.h>
30 #include <linux/security.h>
31 #include <linux/backing-dev.h>
32 #include <linux/mutex.h>
33 #include <linux/capability.h>
34 #include <linux/syscalls.h>
35 #include <linux/memcontrol.h>
36 #include <linux/poll.h>
37 #include <linux/oom.h>
38 #include <linux/frontswap.h>
39 #include <linux/swapfile.h>
40 #include <linux/export.h>
41 #include <linux/swap_slots.h>
42 #include <linux/sort.h>
43 #include <linux/completion.h>
44 
45 #include <asm/tlbflush.h>
46 #include <linux/swapops.h>
47 #include <linux/swap_cgroup.h>
48 #include "swap.h"
49 
50 static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
51 				 unsigned char);
52 static void free_swap_count_continuations(struct swap_info_struct *);
53 
54 static DEFINE_SPINLOCK(swap_lock);
55 static unsigned int nr_swapfiles;
56 atomic_long_t nr_swap_pages;
57 /*
58  * Some modules use swappable objects and may try to swap them out under
59  * memory pressure (via the shrinker). Before doing so, they may wish to
60  * check to see if any swap space is available.
61  */
62 EXPORT_SYMBOL_GPL(nr_swap_pages);
63 /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
64 long total_swap_pages;
65 static int least_priority = -1;
66 unsigned long swapfile_maximum_size;
67 #ifdef CONFIG_MIGRATION
68 bool swap_migration_ad_supported;
69 #endif	/* CONFIG_MIGRATION */
70 
71 static const char Bad_file[] = "Bad swap file entry ";
72 static const char Unused_file[] = "Unused swap file entry ";
73 static const char Bad_offset[] = "Bad swap offset entry ";
74 static const char Unused_offset[] = "Unused swap offset entry ";
75 
76 /*
77  * all active swap_info_structs
78  * protected with swap_lock, and ordered by priority.
79  */
80 static PLIST_HEAD(swap_active_head);
81 
82 /*
83  * all available (active, not full) swap_info_structs
84  * protected with swap_avail_lock, ordered by priority.
85  * This is used by folio_alloc_swap() instead of swap_active_head
86  * because swap_active_head includes all swap_info_structs,
87  * but folio_alloc_swap() doesn't need to look at full ones.
88  * This uses its own lock instead of swap_lock because when a
89  * swap_info_struct changes between not-full/full, it needs to
90  * add/remove itself to/from this list, but the swap_info_struct->lock
91  * is held and the locking order requires swap_lock to be taken
92  * before any swap_info_struct->lock.
93  */
94 static struct plist_head *swap_avail_heads;
95 static DEFINE_SPINLOCK(swap_avail_lock);
96 
97 struct swap_info_struct *swap_info[MAX_SWAPFILES];
98 
99 static DEFINE_MUTEX(swapon_mutex);
100 
101 static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
102 /* Activity counter to indicate that a swapon or swapoff has occurred */
103 static atomic_t proc_poll_event = ATOMIC_INIT(0);
104 
105 atomic_t nr_rotate_swap = ATOMIC_INIT(0);
106 
107 static struct swap_info_struct *swap_type_to_swap_info(int type)
108 {
109 	if (type >= MAX_SWAPFILES)
110 		return NULL;
111 
112 	return READ_ONCE(swap_info[type]); /* rcu_dereference() */
113 }
114 
115 static inline unsigned char swap_count(unsigned char ent)
116 {
117 	return ent & ~SWAP_HAS_CACHE;	/* may include COUNT_CONTINUED flag */
118 }
119 
120 /* Reclaim the swap entry anyway if possible */
121 #define TTRS_ANYWAY		0x1
122 /*
123  * Reclaim the swap entry if there are no more mappings of the
124  * corresponding page
125  */
126 #define TTRS_UNMAPPED		0x2
127 /* Reclaim the swap entry if swap is getting full*/
128 #define TTRS_FULL		0x4
129 
130 /* returns 1 if swap entry is freed */
131 static int __try_to_reclaim_swap(struct swap_info_struct *si,
132 				 unsigned long offset, unsigned long flags)
133 {
134 	swp_entry_t entry = swp_entry(si->type, offset);
135 	struct page *page;
136 	int ret = 0;
137 
138 	page = find_get_page(swap_address_space(entry), offset);
139 	if (!page)
140 		return 0;
141 	/*
142 	 * When this function is called from scan_swap_map_slots() and it's
143 	 * called by vmscan.c at reclaiming pages. So, we hold a lock on a page,
144 	 * here. We have to use trylock for avoiding deadlock. This is a special
145 	 * case and you should use try_to_free_swap() with explicit lock_page()
146 	 * in usual operations.
147 	 */
148 	if (trylock_page(page)) {
149 		if ((flags & TTRS_ANYWAY) ||
150 		    ((flags & TTRS_UNMAPPED) && !page_mapped(page)) ||
151 		    ((flags & TTRS_FULL) && mem_cgroup_swap_full(page)))
152 			ret = try_to_free_swap(page);
153 		unlock_page(page);
154 	}
155 	put_page(page);
156 	return ret;
157 }
158 
159 static inline struct swap_extent *first_se(struct swap_info_struct *sis)
160 {
161 	struct rb_node *rb = rb_first(&sis->swap_extent_root);
162 	return rb_entry(rb, struct swap_extent, rb_node);
163 }
164 
165 static inline struct swap_extent *next_se(struct swap_extent *se)
166 {
167 	struct rb_node *rb = rb_next(&se->rb_node);
168 	return rb ? rb_entry(rb, struct swap_extent, rb_node) : NULL;
169 }
170 
171 /*
172  * swapon tell device that all the old swap contents can be discarded,
173  * to allow the swap device to optimize its wear-levelling.
174  */
175 static int discard_swap(struct swap_info_struct *si)
176 {
177 	struct swap_extent *se;
178 	sector_t start_block;
179 	sector_t nr_blocks;
180 	int err = 0;
181 
182 	/* Do not discard the swap header page! */
183 	se = first_se(si);
184 	start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
185 	nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
186 	if (nr_blocks) {
187 		err = blkdev_issue_discard(si->bdev, start_block,
188 				nr_blocks, GFP_KERNEL);
189 		if (err)
190 			return err;
191 		cond_resched();
192 	}
193 
194 	for (se = next_se(se); se; se = next_se(se)) {
195 		start_block = se->start_block << (PAGE_SHIFT - 9);
196 		nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
197 
198 		err = blkdev_issue_discard(si->bdev, start_block,
199 				nr_blocks, GFP_KERNEL);
200 		if (err)
201 			break;
202 
203 		cond_resched();
204 	}
205 	return err;		/* That will often be -EOPNOTSUPP */
206 }
207 
208 static struct swap_extent *
209 offset_to_swap_extent(struct swap_info_struct *sis, unsigned long offset)
210 {
211 	struct swap_extent *se;
212 	struct rb_node *rb;
213 
214 	rb = sis->swap_extent_root.rb_node;
215 	while (rb) {
216 		se = rb_entry(rb, struct swap_extent, rb_node);
217 		if (offset < se->start_page)
218 			rb = rb->rb_left;
219 		else if (offset >= se->start_page + se->nr_pages)
220 			rb = rb->rb_right;
221 		else
222 			return se;
223 	}
224 	/* It *must* be present */
225 	BUG();
226 }
227 
228 sector_t swap_page_sector(struct page *page)
229 {
230 	struct swap_info_struct *sis = page_swap_info(page);
231 	struct swap_extent *se;
232 	sector_t sector;
233 	pgoff_t offset;
234 
235 	offset = __page_file_index(page);
236 	se = offset_to_swap_extent(sis, offset);
237 	sector = se->start_block + (offset - se->start_page);
238 	return sector << (PAGE_SHIFT - 9);
239 }
240 
241 /*
242  * swap allocation tell device that a cluster of swap can now be discarded,
243  * to allow the swap device to optimize its wear-levelling.
244  */
245 static void discard_swap_cluster(struct swap_info_struct *si,
246 				 pgoff_t start_page, pgoff_t nr_pages)
247 {
248 	struct swap_extent *se = offset_to_swap_extent(si, start_page);
249 
250 	while (nr_pages) {
251 		pgoff_t offset = start_page - se->start_page;
252 		sector_t start_block = se->start_block + offset;
253 		sector_t nr_blocks = se->nr_pages - offset;
254 
255 		if (nr_blocks > nr_pages)
256 			nr_blocks = nr_pages;
257 		start_page += nr_blocks;
258 		nr_pages -= nr_blocks;
259 
260 		start_block <<= PAGE_SHIFT - 9;
261 		nr_blocks <<= PAGE_SHIFT - 9;
262 		if (blkdev_issue_discard(si->bdev, start_block,
263 					nr_blocks, GFP_NOIO))
264 			break;
265 
266 		se = next_se(se);
267 	}
268 }
269 
270 #ifdef CONFIG_THP_SWAP
271 #define SWAPFILE_CLUSTER	HPAGE_PMD_NR
272 
273 #define swap_entry_size(size)	(size)
274 #else
275 #define SWAPFILE_CLUSTER	256
276 
277 /*
278  * Define swap_entry_size() as constant to let compiler to optimize
279  * out some code if !CONFIG_THP_SWAP
280  */
281 #define swap_entry_size(size)	1
282 #endif
283 #define LATENCY_LIMIT		256
284 
285 static inline void cluster_set_flag(struct swap_cluster_info *info,
286 	unsigned int flag)
287 {
288 	info->flags = flag;
289 }
290 
291 static inline unsigned int cluster_count(struct swap_cluster_info *info)
292 {
293 	return info->data;
294 }
295 
296 static inline void cluster_set_count(struct swap_cluster_info *info,
297 				     unsigned int c)
298 {
299 	info->data = c;
300 }
301 
302 static inline void cluster_set_count_flag(struct swap_cluster_info *info,
303 					 unsigned int c, unsigned int f)
304 {
305 	info->flags = f;
306 	info->data = c;
307 }
308 
309 static inline unsigned int cluster_next(struct swap_cluster_info *info)
310 {
311 	return info->data;
312 }
313 
314 static inline void cluster_set_next(struct swap_cluster_info *info,
315 				    unsigned int n)
316 {
317 	info->data = n;
318 }
319 
320 static inline void cluster_set_next_flag(struct swap_cluster_info *info,
321 					 unsigned int n, unsigned int f)
322 {
323 	info->flags = f;
324 	info->data = n;
325 }
326 
327 static inline bool cluster_is_free(struct swap_cluster_info *info)
328 {
329 	return info->flags & CLUSTER_FLAG_FREE;
330 }
331 
332 static inline bool cluster_is_null(struct swap_cluster_info *info)
333 {
334 	return info->flags & CLUSTER_FLAG_NEXT_NULL;
335 }
336 
337 static inline void cluster_set_null(struct swap_cluster_info *info)
338 {
339 	info->flags = CLUSTER_FLAG_NEXT_NULL;
340 	info->data = 0;
341 }
342 
343 static inline bool cluster_is_huge(struct swap_cluster_info *info)
344 {
345 	if (IS_ENABLED(CONFIG_THP_SWAP))
346 		return info->flags & CLUSTER_FLAG_HUGE;
347 	return false;
348 }
349 
350 static inline void cluster_clear_huge(struct swap_cluster_info *info)
351 {
352 	info->flags &= ~CLUSTER_FLAG_HUGE;
353 }
354 
355 static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si,
356 						     unsigned long offset)
357 {
358 	struct swap_cluster_info *ci;
359 
360 	ci = si->cluster_info;
361 	if (ci) {
362 		ci += offset / SWAPFILE_CLUSTER;
363 		spin_lock(&ci->lock);
364 	}
365 	return ci;
366 }
367 
368 static inline void unlock_cluster(struct swap_cluster_info *ci)
369 {
370 	if (ci)
371 		spin_unlock(&ci->lock);
372 }
373 
374 /*
375  * Determine the locking method in use for this device.  Return
376  * swap_cluster_info if SSD-style cluster-based locking is in place.
377  */
378 static inline struct swap_cluster_info *lock_cluster_or_swap_info(
379 		struct swap_info_struct *si, unsigned long offset)
380 {
381 	struct swap_cluster_info *ci;
382 
383 	/* Try to use fine-grained SSD-style locking if available: */
384 	ci = lock_cluster(si, offset);
385 	/* Otherwise, fall back to traditional, coarse locking: */
386 	if (!ci)
387 		spin_lock(&si->lock);
388 
389 	return ci;
390 }
391 
392 static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si,
393 					       struct swap_cluster_info *ci)
394 {
395 	if (ci)
396 		unlock_cluster(ci);
397 	else
398 		spin_unlock(&si->lock);
399 }
400 
401 static inline bool cluster_list_empty(struct swap_cluster_list *list)
402 {
403 	return cluster_is_null(&list->head);
404 }
405 
406 static inline unsigned int cluster_list_first(struct swap_cluster_list *list)
407 {
408 	return cluster_next(&list->head);
409 }
410 
411 static void cluster_list_init(struct swap_cluster_list *list)
412 {
413 	cluster_set_null(&list->head);
414 	cluster_set_null(&list->tail);
415 }
416 
417 static void cluster_list_add_tail(struct swap_cluster_list *list,
418 				  struct swap_cluster_info *ci,
419 				  unsigned int idx)
420 {
421 	if (cluster_list_empty(list)) {
422 		cluster_set_next_flag(&list->head, idx, 0);
423 		cluster_set_next_flag(&list->tail, idx, 0);
424 	} else {
425 		struct swap_cluster_info *ci_tail;
426 		unsigned int tail = cluster_next(&list->tail);
427 
428 		/*
429 		 * Nested cluster lock, but both cluster locks are
430 		 * only acquired when we held swap_info_struct->lock
431 		 */
432 		ci_tail = ci + tail;
433 		spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING);
434 		cluster_set_next(ci_tail, idx);
435 		spin_unlock(&ci_tail->lock);
436 		cluster_set_next_flag(&list->tail, idx, 0);
437 	}
438 }
439 
440 static unsigned int cluster_list_del_first(struct swap_cluster_list *list,
441 					   struct swap_cluster_info *ci)
442 {
443 	unsigned int idx;
444 
445 	idx = cluster_next(&list->head);
446 	if (cluster_next(&list->tail) == idx) {
447 		cluster_set_null(&list->head);
448 		cluster_set_null(&list->tail);
449 	} else
450 		cluster_set_next_flag(&list->head,
451 				      cluster_next(&ci[idx]), 0);
452 
453 	return idx;
454 }
455 
456 /* Add a cluster to discard list and schedule it to do discard */
457 static void swap_cluster_schedule_discard(struct swap_info_struct *si,
458 		unsigned int idx)
459 {
460 	/*
461 	 * If scan_swap_map_slots() can't find a free cluster, it will check
462 	 * si->swap_map directly. To make sure the discarding cluster isn't
463 	 * taken by scan_swap_map_slots(), mark the swap entries bad (occupied).
464 	 * It will be cleared after discard
465 	 */
466 	memset(si->swap_map + idx * SWAPFILE_CLUSTER,
467 			SWAP_MAP_BAD, SWAPFILE_CLUSTER);
468 
469 	cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx);
470 
471 	schedule_work(&si->discard_work);
472 }
473 
474 static void __free_cluster(struct swap_info_struct *si, unsigned long idx)
475 {
476 	struct swap_cluster_info *ci = si->cluster_info;
477 
478 	cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE);
479 	cluster_list_add_tail(&si->free_clusters, ci, idx);
480 }
481 
482 /*
483  * Doing discard actually. After a cluster discard is finished, the cluster
484  * will be added to free cluster list. caller should hold si->lock.
485 */
486 static void swap_do_scheduled_discard(struct swap_info_struct *si)
487 {
488 	struct swap_cluster_info *info, *ci;
489 	unsigned int idx;
490 
491 	info = si->cluster_info;
492 
493 	while (!cluster_list_empty(&si->discard_clusters)) {
494 		idx = cluster_list_del_first(&si->discard_clusters, info);
495 		spin_unlock(&si->lock);
496 
497 		discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
498 				SWAPFILE_CLUSTER);
499 
500 		spin_lock(&si->lock);
501 		ci = lock_cluster(si, idx * SWAPFILE_CLUSTER);
502 		__free_cluster(si, idx);
503 		memset(si->swap_map + idx * SWAPFILE_CLUSTER,
504 				0, SWAPFILE_CLUSTER);
505 		unlock_cluster(ci);
506 	}
507 }
508 
509 static void swap_discard_work(struct work_struct *work)
510 {
511 	struct swap_info_struct *si;
512 
513 	si = container_of(work, struct swap_info_struct, discard_work);
514 
515 	spin_lock(&si->lock);
516 	swap_do_scheduled_discard(si);
517 	spin_unlock(&si->lock);
518 }
519 
520 static void swap_users_ref_free(struct percpu_ref *ref)
521 {
522 	struct swap_info_struct *si;
523 
524 	si = container_of(ref, struct swap_info_struct, users);
525 	complete(&si->comp);
526 }
527 
528 static void alloc_cluster(struct swap_info_struct *si, unsigned long idx)
529 {
530 	struct swap_cluster_info *ci = si->cluster_info;
531 
532 	VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx);
533 	cluster_list_del_first(&si->free_clusters, ci);
534 	cluster_set_count_flag(ci + idx, 0, 0);
535 }
536 
537 static void free_cluster(struct swap_info_struct *si, unsigned long idx)
538 {
539 	struct swap_cluster_info *ci = si->cluster_info + idx;
540 
541 	VM_BUG_ON(cluster_count(ci) != 0);
542 	/*
543 	 * If the swap is discardable, prepare discard the cluster
544 	 * instead of free it immediately. The cluster will be freed
545 	 * after discard.
546 	 */
547 	if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
548 	    (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
549 		swap_cluster_schedule_discard(si, idx);
550 		return;
551 	}
552 
553 	__free_cluster(si, idx);
554 }
555 
556 /*
557  * The cluster corresponding to page_nr will be used. The cluster will be
558  * removed from free cluster list and its usage counter will be increased.
559  */
560 static void inc_cluster_info_page(struct swap_info_struct *p,
561 	struct swap_cluster_info *cluster_info, unsigned long page_nr)
562 {
563 	unsigned long idx = page_nr / SWAPFILE_CLUSTER;
564 
565 	if (!cluster_info)
566 		return;
567 	if (cluster_is_free(&cluster_info[idx]))
568 		alloc_cluster(p, idx);
569 
570 	VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
571 	cluster_set_count(&cluster_info[idx],
572 		cluster_count(&cluster_info[idx]) + 1);
573 }
574 
575 /*
576  * The cluster corresponding to page_nr decreases one usage. If the usage
577  * counter becomes 0, which means no page in the cluster is in using, we can
578  * optionally discard the cluster and add it to free cluster list.
579  */
580 static void dec_cluster_info_page(struct swap_info_struct *p,
581 	struct swap_cluster_info *cluster_info, unsigned long page_nr)
582 {
583 	unsigned long idx = page_nr / SWAPFILE_CLUSTER;
584 
585 	if (!cluster_info)
586 		return;
587 
588 	VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
589 	cluster_set_count(&cluster_info[idx],
590 		cluster_count(&cluster_info[idx]) - 1);
591 
592 	if (cluster_count(&cluster_info[idx]) == 0)
593 		free_cluster(p, idx);
594 }
595 
596 /*
597  * It's possible scan_swap_map_slots() uses a free cluster in the middle of free
598  * cluster list. Avoiding such abuse to avoid list corruption.
599  */
600 static bool
601 scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
602 	unsigned long offset)
603 {
604 	struct percpu_cluster *percpu_cluster;
605 	bool conflict;
606 
607 	offset /= SWAPFILE_CLUSTER;
608 	conflict = !cluster_list_empty(&si->free_clusters) &&
609 		offset != cluster_list_first(&si->free_clusters) &&
610 		cluster_is_free(&si->cluster_info[offset]);
611 
612 	if (!conflict)
613 		return false;
614 
615 	percpu_cluster = this_cpu_ptr(si->percpu_cluster);
616 	cluster_set_null(&percpu_cluster->index);
617 	return true;
618 }
619 
620 /*
621  * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
622  * might involve allocating a new cluster for current CPU too.
623  */
624 static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
625 	unsigned long *offset, unsigned long *scan_base)
626 {
627 	struct percpu_cluster *cluster;
628 	struct swap_cluster_info *ci;
629 	unsigned long tmp, max;
630 
631 new_cluster:
632 	cluster = this_cpu_ptr(si->percpu_cluster);
633 	if (cluster_is_null(&cluster->index)) {
634 		if (!cluster_list_empty(&si->free_clusters)) {
635 			cluster->index = si->free_clusters.head;
636 			cluster->next = cluster_next(&cluster->index) *
637 					SWAPFILE_CLUSTER;
638 		} else if (!cluster_list_empty(&si->discard_clusters)) {
639 			/*
640 			 * we don't have free cluster but have some clusters in
641 			 * discarding, do discard now and reclaim them, then
642 			 * reread cluster_next_cpu since we dropped si->lock
643 			 */
644 			swap_do_scheduled_discard(si);
645 			*scan_base = this_cpu_read(*si->cluster_next_cpu);
646 			*offset = *scan_base;
647 			goto new_cluster;
648 		} else
649 			return false;
650 	}
651 
652 	/*
653 	 * Other CPUs can use our cluster if they can't find a free cluster,
654 	 * check if there is still free entry in the cluster
655 	 */
656 	tmp = cluster->next;
657 	max = min_t(unsigned long, si->max,
658 		    (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER);
659 	if (tmp < max) {
660 		ci = lock_cluster(si, tmp);
661 		while (tmp < max) {
662 			if (!si->swap_map[tmp])
663 				break;
664 			tmp++;
665 		}
666 		unlock_cluster(ci);
667 	}
668 	if (tmp >= max) {
669 		cluster_set_null(&cluster->index);
670 		goto new_cluster;
671 	}
672 	cluster->next = tmp + 1;
673 	*offset = tmp;
674 	*scan_base = tmp;
675 	return true;
676 }
677 
678 static void __del_from_avail_list(struct swap_info_struct *p)
679 {
680 	int nid;
681 
682 	for_each_node(nid)
683 		plist_del(&p->avail_lists[nid], &swap_avail_heads[nid]);
684 }
685 
686 static void del_from_avail_list(struct swap_info_struct *p)
687 {
688 	spin_lock(&swap_avail_lock);
689 	__del_from_avail_list(p);
690 	spin_unlock(&swap_avail_lock);
691 }
692 
693 static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset,
694 			     unsigned int nr_entries)
695 {
696 	unsigned int end = offset + nr_entries - 1;
697 
698 	if (offset == si->lowest_bit)
699 		si->lowest_bit += nr_entries;
700 	if (end == si->highest_bit)
701 		WRITE_ONCE(si->highest_bit, si->highest_bit - nr_entries);
702 	WRITE_ONCE(si->inuse_pages, si->inuse_pages + nr_entries);
703 	if (si->inuse_pages == si->pages) {
704 		si->lowest_bit = si->max;
705 		si->highest_bit = 0;
706 		del_from_avail_list(si);
707 	}
708 }
709 
710 static void add_to_avail_list(struct swap_info_struct *p)
711 {
712 	int nid;
713 
714 	spin_lock(&swap_avail_lock);
715 	for_each_node(nid) {
716 		WARN_ON(!plist_node_empty(&p->avail_lists[nid]));
717 		plist_add(&p->avail_lists[nid], &swap_avail_heads[nid]);
718 	}
719 	spin_unlock(&swap_avail_lock);
720 }
721 
722 static void swap_range_free(struct swap_info_struct *si, unsigned long offset,
723 			    unsigned int nr_entries)
724 {
725 	unsigned long begin = offset;
726 	unsigned long end = offset + nr_entries - 1;
727 	void (*swap_slot_free_notify)(struct block_device *, unsigned long);
728 
729 	if (offset < si->lowest_bit)
730 		si->lowest_bit = offset;
731 	if (end > si->highest_bit) {
732 		bool was_full = !si->highest_bit;
733 
734 		WRITE_ONCE(si->highest_bit, end);
735 		if (was_full && (si->flags & SWP_WRITEOK))
736 			add_to_avail_list(si);
737 	}
738 	atomic_long_add(nr_entries, &nr_swap_pages);
739 	WRITE_ONCE(si->inuse_pages, si->inuse_pages - nr_entries);
740 	if (si->flags & SWP_BLKDEV)
741 		swap_slot_free_notify =
742 			si->bdev->bd_disk->fops->swap_slot_free_notify;
743 	else
744 		swap_slot_free_notify = NULL;
745 	while (offset <= end) {
746 		arch_swap_invalidate_page(si->type, offset);
747 		frontswap_invalidate_page(si->type, offset);
748 		if (swap_slot_free_notify)
749 			swap_slot_free_notify(si->bdev, offset);
750 		offset++;
751 	}
752 	clear_shadow_from_swap_cache(si->type, begin, end);
753 }
754 
755 static void set_cluster_next(struct swap_info_struct *si, unsigned long next)
756 {
757 	unsigned long prev;
758 
759 	if (!(si->flags & SWP_SOLIDSTATE)) {
760 		si->cluster_next = next;
761 		return;
762 	}
763 
764 	prev = this_cpu_read(*si->cluster_next_cpu);
765 	/*
766 	 * Cross the swap address space size aligned trunk, choose
767 	 * another trunk randomly to avoid lock contention on swap
768 	 * address space if possible.
769 	 */
770 	if ((prev >> SWAP_ADDRESS_SPACE_SHIFT) !=
771 	    (next >> SWAP_ADDRESS_SPACE_SHIFT)) {
772 		/* No free swap slots available */
773 		if (si->highest_bit <= si->lowest_bit)
774 			return;
775 		next = si->lowest_bit +
776 			prandom_u32_max(si->highest_bit - si->lowest_bit + 1);
777 		next = ALIGN_DOWN(next, SWAP_ADDRESS_SPACE_PAGES);
778 		next = max_t(unsigned int, next, si->lowest_bit);
779 	}
780 	this_cpu_write(*si->cluster_next_cpu, next);
781 }
782 
783 static bool swap_offset_available_and_locked(struct swap_info_struct *si,
784 					     unsigned long offset)
785 {
786 	if (data_race(!si->swap_map[offset])) {
787 		spin_lock(&si->lock);
788 		return true;
789 	}
790 
791 	if (vm_swap_full() && READ_ONCE(si->swap_map[offset]) == SWAP_HAS_CACHE) {
792 		spin_lock(&si->lock);
793 		return true;
794 	}
795 
796 	return false;
797 }
798 
799 static int scan_swap_map_slots(struct swap_info_struct *si,
800 			       unsigned char usage, int nr,
801 			       swp_entry_t slots[])
802 {
803 	struct swap_cluster_info *ci;
804 	unsigned long offset;
805 	unsigned long scan_base;
806 	unsigned long last_in_cluster = 0;
807 	int latency_ration = LATENCY_LIMIT;
808 	int n_ret = 0;
809 	bool scanned_many = false;
810 
811 	/*
812 	 * We try to cluster swap pages by allocating them sequentially
813 	 * in swap.  Once we've allocated SWAPFILE_CLUSTER pages this
814 	 * way, however, we resort to first-free allocation, starting
815 	 * a new cluster.  This prevents us from scattering swap pages
816 	 * all over the entire swap partition, so that we reduce
817 	 * overall disk seek times between swap pages.  -- sct
818 	 * But we do now try to find an empty cluster.  -Andrea
819 	 * And we let swap pages go all over an SSD partition.  Hugh
820 	 */
821 
822 	si->flags += SWP_SCANNING;
823 	/*
824 	 * Use percpu scan base for SSD to reduce lock contention on
825 	 * cluster and swap cache.  For HDD, sequential access is more
826 	 * important.
827 	 */
828 	if (si->flags & SWP_SOLIDSTATE)
829 		scan_base = this_cpu_read(*si->cluster_next_cpu);
830 	else
831 		scan_base = si->cluster_next;
832 	offset = scan_base;
833 
834 	/* SSD algorithm */
835 	if (si->cluster_info) {
836 		if (!scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
837 			goto scan;
838 	} else if (unlikely(!si->cluster_nr--)) {
839 		if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
840 			si->cluster_nr = SWAPFILE_CLUSTER - 1;
841 			goto checks;
842 		}
843 
844 		spin_unlock(&si->lock);
845 
846 		/*
847 		 * If seek is expensive, start searching for new cluster from
848 		 * start of partition, to minimize the span of allocated swap.
849 		 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
850 		 * case, just handled by scan_swap_map_try_ssd_cluster() above.
851 		 */
852 		scan_base = offset = si->lowest_bit;
853 		last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
854 
855 		/* Locate the first empty (unaligned) cluster */
856 		for (; last_in_cluster <= si->highest_bit; offset++) {
857 			if (si->swap_map[offset])
858 				last_in_cluster = offset + SWAPFILE_CLUSTER;
859 			else if (offset == last_in_cluster) {
860 				spin_lock(&si->lock);
861 				offset -= SWAPFILE_CLUSTER - 1;
862 				si->cluster_next = offset;
863 				si->cluster_nr = SWAPFILE_CLUSTER - 1;
864 				goto checks;
865 			}
866 			if (unlikely(--latency_ration < 0)) {
867 				cond_resched();
868 				latency_ration = LATENCY_LIMIT;
869 			}
870 		}
871 
872 		offset = scan_base;
873 		spin_lock(&si->lock);
874 		si->cluster_nr = SWAPFILE_CLUSTER - 1;
875 	}
876 
877 checks:
878 	if (si->cluster_info) {
879 		while (scan_swap_map_ssd_cluster_conflict(si, offset)) {
880 		/* take a break if we already got some slots */
881 			if (n_ret)
882 				goto done;
883 			if (!scan_swap_map_try_ssd_cluster(si, &offset,
884 							&scan_base))
885 				goto scan;
886 		}
887 	}
888 	if (!(si->flags & SWP_WRITEOK))
889 		goto no_page;
890 	if (!si->highest_bit)
891 		goto no_page;
892 	if (offset > si->highest_bit)
893 		scan_base = offset = si->lowest_bit;
894 
895 	ci = lock_cluster(si, offset);
896 	/* reuse swap entry of cache-only swap if not busy. */
897 	if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
898 		int swap_was_freed;
899 		unlock_cluster(ci);
900 		spin_unlock(&si->lock);
901 		swap_was_freed = __try_to_reclaim_swap(si, offset, TTRS_ANYWAY);
902 		spin_lock(&si->lock);
903 		/* entry was freed successfully, try to use this again */
904 		if (swap_was_freed)
905 			goto checks;
906 		goto scan; /* check next one */
907 	}
908 
909 	if (si->swap_map[offset]) {
910 		unlock_cluster(ci);
911 		if (!n_ret)
912 			goto scan;
913 		else
914 			goto done;
915 	}
916 	WRITE_ONCE(si->swap_map[offset], usage);
917 	inc_cluster_info_page(si, si->cluster_info, offset);
918 	unlock_cluster(ci);
919 
920 	swap_range_alloc(si, offset, 1);
921 	slots[n_ret++] = swp_entry(si->type, offset);
922 
923 	/* got enough slots or reach max slots? */
924 	if ((n_ret == nr) || (offset >= si->highest_bit))
925 		goto done;
926 
927 	/* search for next available slot */
928 
929 	/* time to take a break? */
930 	if (unlikely(--latency_ration < 0)) {
931 		if (n_ret)
932 			goto done;
933 		spin_unlock(&si->lock);
934 		cond_resched();
935 		spin_lock(&si->lock);
936 		latency_ration = LATENCY_LIMIT;
937 	}
938 
939 	/* try to get more slots in cluster */
940 	if (si->cluster_info) {
941 		if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
942 			goto checks;
943 	} else if (si->cluster_nr && !si->swap_map[++offset]) {
944 		/* non-ssd case, still more slots in cluster? */
945 		--si->cluster_nr;
946 		goto checks;
947 	}
948 
949 	/*
950 	 * Even if there's no free clusters available (fragmented),
951 	 * try to scan a little more quickly with lock held unless we
952 	 * have scanned too many slots already.
953 	 */
954 	if (!scanned_many) {
955 		unsigned long scan_limit;
956 
957 		if (offset < scan_base)
958 			scan_limit = scan_base;
959 		else
960 			scan_limit = si->highest_bit;
961 		for (; offset <= scan_limit && --latency_ration > 0;
962 		     offset++) {
963 			if (!si->swap_map[offset])
964 				goto checks;
965 		}
966 	}
967 
968 done:
969 	set_cluster_next(si, offset + 1);
970 	si->flags -= SWP_SCANNING;
971 	return n_ret;
972 
973 scan:
974 	spin_unlock(&si->lock);
975 	while (++offset <= READ_ONCE(si->highest_bit)) {
976 		if (swap_offset_available_and_locked(si, offset))
977 			goto checks;
978 		if (unlikely(--latency_ration < 0)) {
979 			cond_resched();
980 			latency_ration = LATENCY_LIMIT;
981 			scanned_many = true;
982 		}
983 	}
984 	offset = si->lowest_bit;
985 	while (offset < scan_base) {
986 		if (swap_offset_available_and_locked(si, offset))
987 			goto checks;
988 		if (unlikely(--latency_ration < 0)) {
989 			cond_resched();
990 			latency_ration = LATENCY_LIMIT;
991 			scanned_many = true;
992 		}
993 		offset++;
994 	}
995 	spin_lock(&si->lock);
996 
997 no_page:
998 	si->flags -= SWP_SCANNING;
999 	return n_ret;
1000 }
1001 
1002 static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot)
1003 {
1004 	unsigned long idx;
1005 	struct swap_cluster_info *ci;
1006 	unsigned long offset;
1007 
1008 	/*
1009 	 * Should not even be attempting cluster allocations when huge
1010 	 * page swap is disabled.  Warn and fail the allocation.
1011 	 */
1012 	if (!IS_ENABLED(CONFIG_THP_SWAP)) {
1013 		VM_WARN_ON_ONCE(1);
1014 		return 0;
1015 	}
1016 
1017 	if (cluster_list_empty(&si->free_clusters))
1018 		return 0;
1019 
1020 	idx = cluster_list_first(&si->free_clusters);
1021 	offset = idx * SWAPFILE_CLUSTER;
1022 	ci = lock_cluster(si, offset);
1023 	alloc_cluster(si, idx);
1024 	cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE);
1025 
1026 	memset(si->swap_map + offset, SWAP_HAS_CACHE, SWAPFILE_CLUSTER);
1027 	unlock_cluster(ci);
1028 	swap_range_alloc(si, offset, SWAPFILE_CLUSTER);
1029 	*slot = swp_entry(si->type, offset);
1030 
1031 	return 1;
1032 }
1033 
1034 static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx)
1035 {
1036 	unsigned long offset = idx * SWAPFILE_CLUSTER;
1037 	struct swap_cluster_info *ci;
1038 
1039 	ci = lock_cluster(si, offset);
1040 	memset(si->swap_map + offset, 0, SWAPFILE_CLUSTER);
1041 	cluster_set_count_flag(ci, 0, 0);
1042 	free_cluster(si, idx);
1043 	unlock_cluster(ci);
1044 	swap_range_free(si, offset, SWAPFILE_CLUSTER);
1045 }
1046 
1047 int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size)
1048 {
1049 	unsigned long size = swap_entry_size(entry_size);
1050 	struct swap_info_struct *si, *next;
1051 	long avail_pgs;
1052 	int n_ret = 0;
1053 	int node;
1054 
1055 	/* Only single cluster request supported */
1056 	WARN_ON_ONCE(n_goal > 1 && size == SWAPFILE_CLUSTER);
1057 
1058 	spin_lock(&swap_avail_lock);
1059 
1060 	avail_pgs = atomic_long_read(&nr_swap_pages) / size;
1061 	if (avail_pgs <= 0) {
1062 		spin_unlock(&swap_avail_lock);
1063 		goto noswap;
1064 	}
1065 
1066 	n_goal = min3((long)n_goal, (long)SWAP_BATCH, avail_pgs);
1067 
1068 	atomic_long_sub(n_goal * size, &nr_swap_pages);
1069 
1070 start_over:
1071 	node = numa_node_id();
1072 	plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) {
1073 		/* requeue si to after same-priority siblings */
1074 		plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]);
1075 		spin_unlock(&swap_avail_lock);
1076 		spin_lock(&si->lock);
1077 		if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) {
1078 			spin_lock(&swap_avail_lock);
1079 			if (plist_node_empty(&si->avail_lists[node])) {
1080 				spin_unlock(&si->lock);
1081 				goto nextsi;
1082 			}
1083 			WARN(!si->highest_bit,
1084 			     "swap_info %d in list but !highest_bit\n",
1085 			     si->type);
1086 			WARN(!(si->flags & SWP_WRITEOK),
1087 			     "swap_info %d in list but !SWP_WRITEOK\n",
1088 			     si->type);
1089 			__del_from_avail_list(si);
1090 			spin_unlock(&si->lock);
1091 			goto nextsi;
1092 		}
1093 		if (size == SWAPFILE_CLUSTER) {
1094 			if (si->flags & SWP_BLKDEV)
1095 				n_ret = swap_alloc_cluster(si, swp_entries);
1096 		} else
1097 			n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE,
1098 						    n_goal, swp_entries);
1099 		spin_unlock(&si->lock);
1100 		if (n_ret || size == SWAPFILE_CLUSTER)
1101 			goto check_out;
1102 		pr_debug("scan_swap_map of si %d failed to find offset\n",
1103 			si->type);
1104 
1105 		spin_lock(&swap_avail_lock);
1106 nextsi:
1107 		/*
1108 		 * if we got here, it's likely that si was almost full before,
1109 		 * and since scan_swap_map_slots() can drop the si->lock,
1110 		 * multiple callers probably all tried to get a page from the
1111 		 * same si and it filled up before we could get one; or, the si
1112 		 * filled up between us dropping swap_avail_lock and taking
1113 		 * si->lock. Since we dropped the swap_avail_lock, the
1114 		 * swap_avail_head list may have been modified; so if next is
1115 		 * still in the swap_avail_head list then try it, otherwise
1116 		 * start over if we have not gotten any slots.
1117 		 */
1118 		if (plist_node_empty(&next->avail_lists[node]))
1119 			goto start_over;
1120 	}
1121 
1122 	spin_unlock(&swap_avail_lock);
1123 
1124 check_out:
1125 	if (n_ret < n_goal)
1126 		atomic_long_add((long)(n_goal - n_ret) * size,
1127 				&nr_swap_pages);
1128 noswap:
1129 	return n_ret;
1130 }
1131 
1132 static struct swap_info_struct *_swap_info_get(swp_entry_t entry)
1133 {
1134 	struct swap_info_struct *p;
1135 	unsigned long offset;
1136 
1137 	if (!entry.val)
1138 		goto out;
1139 	p = swp_swap_info(entry);
1140 	if (!p)
1141 		goto bad_nofile;
1142 	if (data_race(!(p->flags & SWP_USED)))
1143 		goto bad_device;
1144 	offset = swp_offset(entry);
1145 	if (offset >= p->max)
1146 		goto bad_offset;
1147 	if (data_race(!p->swap_map[swp_offset(entry)]))
1148 		goto bad_free;
1149 	return p;
1150 
1151 bad_free:
1152 	pr_err("%s: %s%08lx\n", __func__, Unused_offset, entry.val);
1153 	goto out;
1154 bad_offset:
1155 	pr_err("%s: %s%08lx\n", __func__, Bad_offset, entry.val);
1156 	goto out;
1157 bad_device:
1158 	pr_err("%s: %s%08lx\n", __func__, Unused_file, entry.val);
1159 	goto out;
1160 bad_nofile:
1161 	pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val);
1162 out:
1163 	return NULL;
1164 }
1165 
1166 static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry,
1167 					struct swap_info_struct *q)
1168 {
1169 	struct swap_info_struct *p;
1170 
1171 	p = _swap_info_get(entry);
1172 
1173 	if (p != q) {
1174 		if (q != NULL)
1175 			spin_unlock(&q->lock);
1176 		if (p != NULL)
1177 			spin_lock(&p->lock);
1178 	}
1179 	return p;
1180 }
1181 
1182 static unsigned char __swap_entry_free_locked(struct swap_info_struct *p,
1183 					      unsigned long offset,
1184 					      unsigned char usage)
1185 {
1186 	unsigned char count;
1187 	unsigned char has_cache;
1188 
1189 	count = p->swap_map[offset];
1190 
1191 	has_cache = count & SWAP_HAS_CACHE;
1192 	count &= ~SWAP_HAS_CACHE;
1193 
1194 	if (usage == SWAP_HAS_CACHE) {
1195 		VM_BUG_ON(!has_cache);
1196 		has_cache = 0;
1197 	} else if (count == SWAP_MAP_SHMEM) {
1198 		/*
1199 		 * Or we could insist on shmem.c using a special
1200 		 * swap_shmem_free() and free_shmem_swap_and_cache()...
1201 		 */
1202 		count = 0;
1203 	} else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
1204 		if (count == COUNT_CONTINUED) {
1205 			if (swap_count_continued(p, offset, count))
1206 				count = SWAP_MAP_MAX | COUNT_CONTINUED;
1207 			else
1208 				count = SWAP_MAP_MAX;
1209 		} else
1210 			count--;
1211 	}
1212 
1213 	usage = count | has_cache;
1214 	if (usage)
1215 		WRITE_ONCE(p->swap_map[offset], usage);
1216 	else
1217 		WRITE_ONCE(p->swap_map[offset], SWAP_HAS_CACHE);
1218 
1219 	return usage;
1220 }
1221 
1222 /*
1223  * Check whether swap entry is valid in the swap device.  If so,
1224  * return pointer to swap_info_struct, and keep the swap entry valid
1225  * via preventing the swap device from being swapoff, until
1226  * put_swap_device() is called.  Otherwise return NULL.
1227  *
1228  * Notice that swapoff or swapoff+swapon can still happen before the
1229  * percpu_ref_tryget_live() in get_swap_device() or after the
1230  * percpu_ref_put() in put_swap_device() if there isn't any other way
1231  * to prevent swapoff, such as page lock, page table lock, etc.  The
1232  * caller must be prepared for that.  For example, the following
1233  * situation is possible.
1234  *
1235  *   CPU1				CPU2
1236  *   do_swap_page()
1237  *     ...				swapoff+swapon
1238  *     __read_swap_cache_async()
1239  *       swapcache_prepare()
1240  *         __swap_duplicate()
1241  *           // check swap_map
1242  *     // verify PTE not changed
1243  *
1244  * In __swap_duplicate(), the swap_map need to be checked before
1245  * changing partly because the specified swap entry may be for another
1246  * swap device which has been swapoff.  And in do_swap_page(), after
1247  * the page is read from the swap device, the PTE is verified not
1248  * changed with the page table locked to check whether the swap device
1249  * has been swapoff or swapoff+swapon.
1250  */
1251 struct swap_info_struct *get_swap_device(swp_entry_t entry)
1252 {
1253 	struct swap_info_struct *si;
1254 	unsigned long offset;
1255 
1256 	if (!entry.val)
1257 		goto out;
1258 	si = swp_swap_info(entry);
1259 	if (!si)
1260 		goto bad_nofile;
1261 	if (!percpu_ref_tryget_live(&si->users))
1262 		goto out;
1263 	/*
1264 	 * Guarantee the si->users are checked before accessing other
1265 	 * fields of swap_info_struct.
1266 	 *
1267 	 * Paired with the spin_unlock() after setup_swap_info() in
1268 	 * enable_swap_info().
1269 	 */
1270 	smp_rmb();
1271 	offset = swp_offset(entry);
1272 	if (offset >= si->max)
1273 		goto put_out;
1274 
1275 	return si;
1276 bad_nofile:
1277 	pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val);
1278 out:
1279 	return NULL;
1280 put_out:
1281 	pr_err("%s: %s%08lx\n", __func__, Bad_offset, entry.val);
1282 	percpu_ref_put(&si->users);
1283 	return NULL;
1284 }
1285 
1286 static unsigned char __swap_entry_free(struct swap_info_struct *p,
1287 				       swp_entry_t entry)
1288 {
1289 	struct swap_cluster_info *ci;
1290 	unsigned long offset = swp_offset(entry);
1291 	unsigned char usage;
1292 
1293 	ci = lock_cluster_or_swap_info(p, offset);
1294 	usage = __swap_entry_free_locked(p, offset, 1);
1295 	unlock_cluster_or_swap_info(p, ci);
1296 	if (!usage)
1297 		free_swap_slot(entry);
1298 
1299 	return usage;
1300 }
1301 
1302 static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry)
1303 {
1304 	struct swap_cluster_info *ci;
1305 	unsigned long offset = swp_offset(entry);
1306 	unsigned char count;
1307 
1308 	ci = lock_cluster(p, offset);
1309 	count = p->swap_map[offset];
1310 	VM_BUG_ON(count != SWAP_HAS_CACHE);
1311 	p->swap_map[offset] = 0;
1312 	dec_cluster_info_page(p, p->cluster_info, offset);
1313 	unlock_cluster(ci);
1314 
1315 	mem_cgroup_uncharge_swap(entry, 1);
1316 	swap_range_free(p, offset, 1);
1317 }
1318 
1319 /*
1320  * Caller has made sure that the swap device corresponding to entry
1321  * is still around or has not been recycled.
1322  */
1323 void swap_free(swp_entry_t entry)
1324 {
1325 	struct swap_info_struct *p;
1326 
1327 	p = _swap_info_get(entry);
1328 	if (p)
1329 		__swap_entry_free(p, entry);
1330 }
1331 
1332 /*
1333  * Called after dropping swapcache to decrease refcnt to swap entries.
1334  */
1335 void put_swap_folio(struct folio *folio, swp_entry_t entry)
1336 {
1337 	unsigned long offset = swp_offset(entry);
1338 	unsigned long idx = offset / SWAPFILE_CLUSTER;
1339 	struct swap_cluster_info *ci;
1340 	struct swap_info_struct *si;
1341 	unsigned char *map;
1342 	unsigned int i, free_entries = 0;
1343 	unsigned char val;
1344 	int size = swap_entry_size(folio_nr_pages(folio));
1345 
1346 	si = _swap_info_get(entry);
1347 	if (!si)
1348 		return;
1349 
1350 	ci = lock_cluster_or_swap_info(si, offset);
1351 	if (size == SWAPFILE_CLUSTER) {
1352 		VM_BUG_ON(!cluster_is_huge(ci));
1353 		map = si->swap_map + offset;
1354 		for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1355 			val = map[i];
1356 			VM_BUG_ON(!(val & SWAP_HAS_CACHE));
1357 			if (val == SWAP_HAS_CACHE)
1358 				free_entries++;
1359 		}
1360 		cluster_clear_huge(ci);
1361 		if (free_entries == SWAPFILE_CLUSTER) {
1362 			unlock_cluster_or_swap_info(si, ci);
1363 			spin_lock(&si->lock);
1364 			mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER);
1365 			swap_free_cluster(si, idx);
1366 			spin_unlock(&si->lock);
1367 			return;
1368 		}
1369 	}
1370 	for (i = 0; i < size; i++, entry.val++) {
1371 		if (!__swap_entry_free_locked(si, offset + i, SWAP_HAS_CACHE)) {
1372 			unlock_cluster_or_swap_info(si, ci);
1373 			free_swap_slot(entry);
1374 			if (i == size - 1)
1375 				return;
1376 			lock_cluster_or_swap_info(si, offset);
1377 		}
1378 	}
1379 	unlock_cluster_or_swap_info(si, ci);
1380 }
1381 
1382 #ifdef CONFIG_THP_SWAP
1383 int split_swap_cluster(swp_entry_t entry)
1384 {
1385 	struct swap_info_struct *si;
1386 	struct swap_cluster_info *ci;
1387 	unsigned long offset = swp_offset(entry);
1388 
1389 	si = _swap_info_get(entry);
1390 	if (!si)
1391 		return -EBUSY;
1392 	ci = lock_cluster(si, offset);
1393 	cluster_clear_huge(ci);
1394 	unlock_cluster(ci);
1395 	return 0;
1396 }
1397 #endif
1398 
1399 static int swp_entry_cmp(const void *ent1, const void *ent2)
1400 {
1401 	const swp_entry_t *e1 = ent1, *e2 = ent2;
1402 
1403 	return (int)swp_type(*e1) - (int)swp_type(*e2);
1404 }
1405 
1406 void swapcache_free_entries(swp_entry_t *entries, int n)
1407 {
1408 	struct swap_info_struct *p, *prev;
1409 	int i;
1410 
1411 	if (n <= 0)
1412 		return;
1413 
1414 	prev = NULL;
1415 	p = NULL;
1416 
1417 	/*
1418 	 * Sort swap entries by swap device, so each lock is only taken once.
1419 	 * nr_swapfiles isn't absolutely correct, but the overhead of sort() is
1420 	 * so low that it isn't necessary to optimize further.
1421 	 */
1422 	if (nr_swapfiles > 1)
1423 		sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL);
1424 	for (i = 0; i < n; ++i) {
1425 		p = swap_info_get_cont(entries[i], prev);
1426 		if (p)
1427 			swap_entry_free(p, entries[i]);
1428 		prev = p;
1429 	}
1430 	if (p)
1431 		spin_unlock(&p->lock);
1432 }
1433 
1434 int __swap_count(swp_entry_t entry)
1435 {
1436 	struct swap_info_struct *si;
1437 	pgoff_t offset = swp_offset(entry);
1438 	int count = 0;
1439 
1440 	si = get_swap_device(entry);
1441 	if (si) {
1442 		count = swap_count(si->swap_map[offset]);
1443 		put_swap_device(si);
1444 	}
1445 	return count;
1446 }
1447 
1448 /*
1449  * How many references to @entry are currently swapped out?
1450  * This does not give an exact answer when swap count is continued,
1451  * but does include the high COUNT_CONTINUED flag to allow for that.
1452  */
1453 static int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry)
1454 {
1455 	pgoff_t offset = swp_offset(entry);
1456 	struct swap_cluster_info *ci;
1457 	int count;
1458 
1459 	ci = lock_cluster_or_swap_info(si, offset);
1460 	count = swap_count(si->swap_map[offset]);
1461 	unlock_cluster_or_swap_info(si, ci);
1462 	return count;
1463 }
1464 
1465 /*
1466  * How many references to @entry are currently swapped out?
1467  * This does not give an exact answer when swap count is continued,
1468  * but does include the high COUNT_CONTINUED flag to allow for that.
1469  */
1470 int __swp_swapcount(swp_entry_t entry)
1471 {
1472 	int count = 0;
1473 	struct swap_info_struct *si;
1474 
1475 	si = get_swap_device(entry);
1476 	if (si) {
1477 		count = swap_swapcount(si, entry);
1478 		put_swap_device(si);
1479 	}
1480 	return count;
1481 }
1482 
1483 /*
1484  * How many references to @entry are currently swapped out?
1485  * This considers COUNT_CONTINUED so it returns exact answer.
1486  */
1487 int swp_swapcount(swp_entry_t entry)
1488 {
1489 	int count, tmp_count, n;
1490 	struct swap_info_struct *p;
1491 	struct swap_cluster_info *ci;
1492 	struct page *page;
1493 	pgoff_t offset;
1494 	unsigned char *map;
1495 
1496 	p = _swap_info_get(entry);
1497 	if (!p)
1498 		return 0;
1499 
1500 	offset = swp_offset(entry);
1501 
1502 	ci = lock_cluster_or_swap_info(p, offset);
1503 
1504 	count = swap_count(p->swap_map[offset]);
1505 	if (!(count & COUNT_CONTINUED))
1506 		goto out;
1507 
1508 	count &= ~COUNT_CONTINUED;
1509 	n = SWAP_MAP_MAX + 1;
1510 
1511 	page = vmalloc_to_page(p->swap_map + offset);
1512 	offset &= ~PAGE_MASK;
1513 	VM_BUG_ON(page_private(page) != SWP_CONTINUED);
1514 
1515 	do {
1516 		page = list_next_entry(page, lru);
1517 		map = kmap_atomic(page);
1518 		tmp_count = map[offset];
1519 		kunmap_atomic(map);
1520 
1521 		count += (tmp_count & ~COUNT_CONTINUED) * n;
1522 		n *= (SWAP_CONT_MAX + 1);
1523 	} while (tmp_count & COUNT_CONTINUED);
1524 out:
1525 	unlock_cluster_or_swap_info(p, ci);
1526 	return count;
1527 }
1528 
1529 static bool swap_page_trans_huge_swapped(struct swap_info_struct *si,
1530 					 swp_entry_t entry)
1531 {
1532 	struct swap_cluster_info *ci;
1533 	unsigned char *map = si->swap_map;
1534 	unsigned long roffset = swp_offset(entry);
1535 	unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER);
1536 	int i;
1537 	bool ret = false;
1538 
1539 	ci = lock_cluster_or_swap_info(si, offset);
1540 	if (!ci || !cluster_is_huge(ci)) {
1541 		if (swap_count(map[roffset]))
1542 			ret = true;
1543 		goto unlock_out;
1544 	}
1545 	for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1546 		if (swap_count(map[offset + i])) {
1547 			ret = true;
1548 			break;
1549 		}
1550 	}
1551 unlock_out:
1552 	unlock_cluster_or_swap_info(si, ci);
1553 	return ret;
1554 }
1555 
1556 static bool folio_swapped(struct folio *folio)
1557 {
1558 	swp_entry_t entry = folio_swap_entry(folio);
1559 	struct swap_info_struct *si = _swap_info_get(entry);
1560 
1561 	if (!si)
1562 		return false;
1563 
1564 	if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!folio_test_large(folio)))
1565 		return swap_swapcount(si, entry) != 0;
1566 
1567 	return swap_page_trans_huge_swapped(si, entry);
1568 }
1569 
1570 /**
1571  * folio_free_swap() - Free the swap space used for this folio.
1572  * @folio: The folio to remove.
1573  *
1574  * If swap is getting full, or if there are no more mappings of this folio,
1575  * then call folio_free_swap to free its swap space.
1576  *
1577  * Return: true if we were able to release the swap space.
1578  */
1579 bool folio_free_swap(struct folio *folio)
1580 {
1581 	VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1582 
1583 	if (!folio_test_swapcache(folio))
1584 		return false;
1585 	if (folio_test_writeback(folio))
1586 		return false;
1587 	if (folio_swapped(folio))
1588 		return false;
1589 
1590 	/*
1591 	 * Once hibernation has begun to create its image of memory,
1592 	 * there's a danger that one of the calls to folio_free_swap()
1593 	 * - most probably a call from __try_to_reclaim_swap() while
1594 	 * hibernation is allocating its own swap pages for the image,
1595 	 * but conceivably even a call from memory reclaim - will free
1596 	 * the swap from a folio which has already been recorded in the
1597 	 * image as a clean swapcache folio, and then reuse its swap for
1598 	 * another page of the image.  On waking from hibernation, the
1599 	 * original folio might be freed under memory pressure, then
1600 	 * later read back in from swap, now with the wrong data.
1601 	 *
1602 	 * Hibernation suspends storage while it is writing the image
1603 	 * to disk so check that here.
1604 	 */
1605 	if (pm_suspended_storage())
1606 		return false;
1607 
1608 	delete_from_swap_cache(folio);
1609 	folio_set_dirty(folio);
1610 	return true;
1611 }
1612 
1613 /*
1614  * Free the swap entry like above, but also try to
1615  * free the page cache entry if it is the last user.
1616  */
1617 int free_swap_and_cache(swp_entry_t entry)
1618 {
1619 	struct swap_info_struct *p;
1620 	unsigned char count;
1621 
1622 	if (non_swap_entry(entry))
1623 		return 1;
1624 
1625 	p = _swap_info_get(entry);
1626 	if (p) {
1627 		count = __swap_entry_free(p, entry);
1628 		if (count == SWAP_HAS_CACHE &&
1629 		    !swap_page_trans_huge_swapped(p, entry))
1630 			__try_to_reclaim_swap(p, swp_offset(entry),
1631 					      TTRS_UNMAPPED | TTRS_FULL);
1632 	}
1633 	return p != NULL;
1634 }
1635 
1636 #ifdef CONFIG_HIBERNATION
1637 
1638 swp_entry_t get_swap_page_of_type(int type)
1639 {
1640 	struct swap_info_struct *si = swap_type_to_swap_info(type);
1641 	swp_entry_t entry = {0};
1642 
1643 	if (!si)
1644 		goto fail;
1645 
1646 	/* This is called for allocating swap entry, not cache */
1647 	spin_lock(&si->lock);
1648 	if ((si->flags & SWP_WRITEOK) && scan_swap_map_slots(si, 1, 1, &entry))
1649 		atomic_long_dec(&nr_swap_pages);
1650 	spin_unlock(&si->lock);
1651 fail:
1652 	return entry;
1653 }
1654 
1655 /*
1656  * Find the swap type that corresponds to given device (if any).
1657  *
1658  * @offset - number of the PAGE_SIZE-sized block of the device, starting
1659  * from 0, in which the swap header is expected to be located.
1660  *
1661  * This is needed for the suspend to disk (aka swsusp).
1662  */
1663 int swap_type_of(dev_t device, sector_t offset)
1664 {
1665 	int type;
1666 
1667 	if (!device)
1668 		return -1;
1669 
1670 	spin_lock(&swap_lock);
1671 	for (type = 0; type < nr_swapfiles; type++) {
1672 		struct swap_info_struct *sis = swap_info[type];
1673 
1674 		if (!(sis->flags & SWP_WRITEOK))
1675 			continue;
1676 
1677 		if (device == sis->bdev->bd_dev) {
1678 			struct swap_extent *se = first_se(sis);
1679 
1680 			if (se->start_block == offset) {
1681 				spin_unlock(&swap_lock);
1682 				return type;
1683 			}
1684 		}
1685 	}
1686 	spin_unlock(&swap_lock);
1687 	return -ENODEV;
1688 }
1689 
1690 int find_first_swap(dev_t *device)
1691 {
1692 	int type;
1693 
1694 	spin_lock(&swap_lock);
1695 	for (type = 0; type < nr_swapfiles; type++) {
1696 		struct swap_info_struct *sis = swap_info[type];
1697 
1698 		if (!(sis->flags & SWP_WRITEOK))
1699 			continue;
1700 		*device = sis->bdev->bd_dev;
1701 		spin_unlock(&swap_lock);
1702 		return type;
1703 	}
1704 	spin_unlock(&swap_lock);
1705 	return -ENODEV;
1706 }
1707 
1708 /*
1709  * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1710  * corresponding to given index in swap_info (swap type).
1711  */
1712 sector_t swapdev_block(int type, pgoff_t offset)
1713 {
1714 	struct swap_info_struct *si = swap_type_to_swap_info(type);
1715 	struct swap_extent *se;
1716 
1717 	if (!si || !(si->flags & SWP_WRITEOK))
1718 		return 0;
1719 	se = offset_to_swap_extent(si, offset);
1720 	return se->start_block + (offset - se->start_page);
1721 }
1722 
1723 /*
1724  * Return either the total number of swap pages of given type, or the number
1725  * of free pages of that type (depending on @free)
1726  *
1727  * This is needed for software suspend
1728  */
1729 unsigned int count_swap_pages(int type, int free)
1730 {
1731 	unsigned int n = 0;
1732 
1733 	spin_lock(&swap_lock);
1734 	if ((unsigned int)type < nr_swapfiles) {
1735 		struct swap_info_struct *sis = swap_info[type];
1736 
1737 		spin_lock(&sis->lock);
1738 		if (sis->flags & SWP_WRITEOK) {
1739 			n = sis->pages;
1740 			if (free)
1741 				n -= sis->inuse_pages;
1742 		}
1743 		spin_unlock(&sis->lock);
1744 	}
1745 	spin_unlock(&swap_lock);
1746 	return n;
1747 }
1748 #endif /* CONFIG_HIBERNATION */
1749 
1750 static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte)
1751 {
1752 	return pte_same(pte_swp_clear_flags(pte), swp_pte);
1753 }
1754 
1755 /*
1756  * No need to decide whether this PTE shares the swap entry with others,
1757  * just let do_wp_page work it out if a write is requested later - to
1758  * force COW, vm_page_prot omits write permission from any private vma.
1759  */
1760 static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
1761 		unsigned long addr, swp_entry_t entry, struct page *page)
1762 {
1763 	struct page *swapcache;
1764 	spinlock_t *ptl;
1765 	pte_t *pte, new_pte;
1766 	int ret = 1;
1767 
1768 	swapcache = page;
1769 	page = ksm_might_need_to_copy(page, vma, addr);
1770 	if (unlikely(!page))
1771 		return -ENOMEM;
1772 
1773 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1774 	if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) {
1775 		ret = 0;
1776 		goto out;
1777 	}
1778 
1779 	if (unlikely(!PageUptodate(page))) {
1780 		pte_t pteval;
1781 
1782 		dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1783 		pteval = swp_entry_to_pte(make_swapin_error_entry(page));
1784 		set_pte_at(vma->vm_mm, addr, pte, pteval);
1785 		swap_free(entry);
1786 		ret = 0;
1787 		goto out;
1788 	}
1789 
1790 	/* See do_swap_page() */
1791 	BUG_ON(!PageAnon(page) && PageMappedToDisk(page));
1792 	BUG_ON(PageAnon(page) && PageAnonExclusive(page));
1793 
1794 	dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1795 	inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
1796 	get_page(page);
1797 	if (page == swapcache) {
1798 		rmap_t rmap_flags = RMAP_NONE;
1799 
1800 		/*
1801 		 * See do_swap_page(): PageWriteback() would be problematic.
1802 		 * However, we do a wait_on_page_writeback() just before this
1803 		 * call and have the page locked.
1804 		 */
1805 		VM_BUG_ON_PAGE(PageWriteback(page), page);
1806 		if (pte_swp_exclusive(*pte))
1807 			rmap_flags |= RMAP_EXCLUSIVE;
1808 
1809 		page_add_anon_rmap(page, vma, addr, rmap_flags);
1810 	} else { /* ksm created a completely new copy */
1811 		page_add_new_anon_rmap(page, vma, addr);
1812 		lru_cache_add_inactive_or_unevictable(page, vma);
1813 	}
1814 	new_pte = pte_mkold(mk_pte(page, vma->vm_page_prot));
1815 	if (pte_swp_soft_dirty(*pte))
1816 		new_pte = pte_mksoft_dirty(new_pte);
1817 	if (pte_swp_uffd_wp(*pte))
1818 		new_pte = pte_mkuffd_wp(new_pte);
1819 	set_pte_at(vma->vm_mm, addr, pte, new_pte);
1820 	swap_free(entry);
1821 out:
1822 	pte_unmap_unlock(pte, ptl);
1823 	if (page != swapcache) {
1824 		unlock_page(page);
1825 		put_page(page);
1826 	}
1827 	return ret;
1828 }
1829 
1830 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1831 			unsigned long addr, unsigned long end,
1832 			unsigned int type)
1833 {
1834 	struct page *page;
1835 	swp_entry_t entry;
1836 	pte_t *pte;
1837 	struct swap_info_struct *si;
1838 	unsigned long offset;
1839 	int ret = 0;
1840 	volatile unsigned char *swap_map;
1841 
1842 	si = swap_info[type];
1843 	pte = pte_offset_map(pmd, addr);
1844 	do {
1845 		if (!is_swap_pte(*pte))
1846 			continue;
1847 
1848 		entry = pte_to_swp_entry(*pte);
1849 		if (swp_type(entry) != type)
1850 			continue;
1851 
1852 		offset = swp_offset(entry);
1853 		pte_unmap(pte);
1854 		swap_map = &si->swap_map[offset];
1855 		page = lookup_swap_cache(entry, vma, addr);
1856 		if (!page) {
1857 			struct vm_fault vmf = {
1858 				.vma = vma,
1859 				.address = addr,
1860 				.real_address = addr,
1861 				.pmd = pmd,
1862 			};
1863 
1864 			page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE,
1865 						&vmf);
1866 		}
1867 		if (!page) {
1868 			if (*swap_map == 0 || *swap_map == SWAP_MAP_BAD)
1869 				goto try_next;
1870 			return -ENOMEM;
1871 		}
1872 
1873 		lock_page(page);
1874 		wait_on_page_writeback(page);
1875 		ret = unuse_pte(vma, pmd, addr, entry, page);
1876 		if (ret < 0) {
1877 			unlock_page(page);
1878 			put_page(page);
1879 			goto out;
1880 		}
1881 
1882 		try_to_free_swap(page);
1883 		unlock_page(page);
1884 		put_page(page);
1885 try_next:
1886 		pte = pte_offset_map(pmd, addr);
1887 	} while (pte++, addr += PAGE_SIZE, addr != end);
1888 	pte_unmap(pte - 1);
1889 
1890 	ret = 0;
1891 out:
1892 	return ret;
1893 }
1894 
1895 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1896 				unsigned long addr, unsigned long end,
1897 				unsigned int type)
1898 {
1899 	pmd_t *pmd;
1900 	unsigned long next;
1901 	int ret;
1902 
1903 	pmd = pmd_offset(pud, addr);
1904 	do {
1905 		cond_resched();
1906 		next = pmd_addr_end(addr, end);
1907 		if (pmd_none_or_trans_huge_or_clear_bad(pmd))
1908 			continue;
1909 		ret = unuse_pte_range(vma, pmd, addr, next, type);
1910 		if (ret)
1911 			return ret;
1912 	} while (pmd++, addr = next, addr != end);
1913 	return 0;
1914 }
1915 
1916 static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d,
1917 				unsigned long addr, unsigned long end,
1918 				unsigned int type)
1919 {
1920 	pud_t *pud;
1921 	unsigned long next;
1922 	int ret;
1923 
1924 	pud = pud_offset(p4d, addr);
1925 	do {
1926 		next = pud_addr_end(addr, end);
1927 		if (pud_none_or_clear_bad(pud))
1928 			continue;
1929 		ret = unuse_pmd_range(vma, pud, addr, next, type);
1930 		if (ret)
1931 			return ret;
1932 	} while (pud++, addr = next, addr != end);
1933 	return 0;
1934 }
1935 
1936 static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd,
1937 				unsigned long addr, unsigned long end,
1938 				unsigned int type)
1939 {
1940 	p4d_t *p4d;
1941 	unsigned long next;
1942 	int ret;
1943 
1944 	p4d = p4d_offset(pgd, addr);
1945 	do {
1946 		next = p4d_addr_end(addr, end);
1947 		if (p4d_none_or_clear_bad(p4d))
1948 			continue;
1949 		ret = unuse_pud_range(vma, p4d, addr, next, type);
1950 		if (ret)
1951 			return ret;
1952 	} while (p4d++, addr = next, addr != end);
1953 	return 0;
1954 }
1955 
1956 static int unuse_vma(struct vm_area_struct *vma, unsigned int type)
1957 {
1958 	pgd_t *pgd;
1959 	unsigned long addr, end, next;
1960 	int ret;
1961 
1962 	addr = vma->vm_start;
1963 	end = vma->vm_end;
1964 
1965 	pgd = pgd_offset(vma->vm_mm, addr);
1966 	do {
1967 		next = pgd_addr_end(addr, end);
1968 		if (pgd_none_or_clear_bad(pgd))
1969 			continue;
1970 		ret = unuse_p4d_range(vma, pgd, addr, next, type);
1971 		if (ret)
1972 			return ret;
1973 	} while (pgd++, addr = next, addr != end);
1974 	return 0;
1975 }
1976 
1977 static int unuse_mm(struct mm_struct *mm, unsigned int type)
1978 {
1979 	struct vm_area_struct *vma;
1980 	int ret = 0;
1981 	VMA_ITERATOR(vmi, mm, 0);
1982 
1983 	mmap_read_lock(mm);
1984 	for_each_vma(vmi, vma) {
1985 		if (vma->anon_vma) {
1986 			ret = unuse_vma(vma, type);
1987 			if (ret)
1988 				break;
1989 		}
1990 
1991 		cond_resched();
1992 	}
1993 	mmap_read_unlock(mm);
1994 	return ret;
1995 }
1996 
1997 /*
1998  * Scan swap_map from current position to next entry still in use.
1999  * Return 0 if there are no inuse entries after prev till end of
2000  * the map.
2001  */
2002 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
2003 					unsigned int prev)
2004 {
2005 	unsigned int i;
2006 	unsigned char count;
2007 
2008 	/*
2009 	 * No need for swap_lock here: we're just looking
2010 	 * for whether an entry is in use, not modifying it; false
2011 	 * hits are okay, and sys_swapoff() has already prevented new
2012 	 * allocations from this area (while holding swap_lock).
2013 	 */
2014 	for (i = prev + 1; i < si->max; i++) {
2015 		count = READ_ONCE(si->swap_map[i]);
2016 		if (count && swap_count(count) != SWAP_MAP_BAD)
2017 			break;
2018 		if ((i % LATENCY_LIMIT) == 0)
2019 			cond_resched();
2020 	}
2021 
2022 	if (i == si->max)
2023 		i = 0;
2024 
2025 	return i;
2026 }
2027 
2028 static int try_to_unuse(unsigned int type)
2029 {
2030 	struct mm_struct *prev_mm;
2031 	struct mm_struct *mm;
2032 	struct list_head *p;
2033 	int retval = 0;
2034 	struct swap_info_struct *si = swap_info[type];
2035 	struct page *page;
2036 	swp_entry_t entry;
2037 	unsigned int i;
2038 
2039 	if (!READ_ONCE(si->inuse_pages))
2040 		return 0;
2041 
2042 retry:
2043 	retval = shmem_unuse(type);
2044 	if (retval)
2045 		return retval;
2046 
2047 	prev_mm = &init_mm;
2048 	mmget(prev_mm);
2049 
2050 	spin_lock(&mmlist_lock);
2051 	p = &init_mm.mmlist;
2052 	while (READ_ONCE(si->inuse_pages) &&
2053 	       !signal_pending(current) &&
2054 	       (p = p->next) != &init_mm.mmlist) {
2055 
2056 		mm = list_entry(p, struct mm_struct, mmlist);
2057 		if (!mmget_not_zero(mm))
2058 			continue;
2059 		spin_unlock(&mmlist_lock);
2060 		mmput(prev_mm);
2061 		prev_mm = mm;
2062 		retval = unuse_mm(mm, type);
2063 		if (retval) {
2064 			mmput(prev_mm);
2065 			return retval;
2066 		}
2067 
2068 		/*
2069 		 * Make sure that we aren't completely killing
2070 		 * interactive performance.
2071 		 */
2072 		cond_resched();
2073 		spin_lock(&mmlist_lock);
2074 	}
2075 	spin_unlock(&mmlist_lock);
2076 
2077 	mmput(prev_mm);
2078 
2079 	i = 0;
2080 	while (READ_ONCE(si->inuse_pages) &&
2081 	       !signal_pending(current) &&
2082 	       (i = find_next_to_unuse(si, i)) != 0) {
2083 
2084 		entry = swp_entry(type, i);
2085 		page = find_get_page(swap_address_space(entry), i);
2086 		if (!page)
2087 			continue;
2088 
2089 		/*
2090 		 * It is conceivable that a racing task removed this page from
2091 		 * swap cache just before we acquired the page lock. The page
2092 		 * might even be back in swap cache on another swap area. But
2093 		 * that is okay, try_to_free_swap() only removes stale pages.
2094 		 */
2095 		lock_page(page);
2096 		wait_on_page_writeback(page);
2097 		try_to_free_swap(page);
2098 		unlock_page(page);
2099 		put_page(page);
2100 	}
2101 
2102 	/*
2103 	 * Lets check again to see if there are still swap entries in the map.
2104 	 * If yes, we would need to do retry the unuse logic again.
2105 	 * Under global memory pressure, swap entries can be reinserted back
2106 	 * into process space after the mmlist loop above passes over them.
2107 	 *
2108 	 * Limit the number of retries? No: when mmget_not_zero()
2109 	 * above fails, that mm is likely to be freeing swap from
2110 	 * exit_mmap(), which proceeds at its own independent pace;
2111 	 * and even shmem_writepage() could have been preempted after
2112 	 * folio_alloc_swap(), temporarily hiding that swap.  It's easy
2113 	 * and robust (though cpu-intensive) just to keep retrying.
2114 	 */
2115 	if (READ_ONCE(si->inuse_pages)) {
2116 		if (!signal_pending(current))
2117 			goto retry;
2118 		return -EINTR;
2119 	}
2120 
2121 	return 0;
2122 }
2123 
2124 /*
2125  * After a successful try_to_unuse, if no swap is now in use, we know
2126  * we can empty the mmlist.  swap_lock must be held on entry and exit.
2127  * Note that mmlist_lock nests inside swap_lock, and an mm must be
2128  * added to the mmlist just after page_duplicate - before would be racy.
2129  */
2130 static void drain_mmlist(void)
2131 {
2132 	struct list_head *p, *next;
2133 	unsigned int type;
2134 
2135 	for (type = 0; type < nr_swapfiles; type++)
2136 		if (swap_info[type]->inuse_pages)
2137 			return;
2138 	spin_lock(&mmlist_lock);
2139 	list_for_each_safe(p, next, &init_mm.mmlist)
2140 		list_del_init(p);
2141 	spin_unlock(&mmlist_lock);
2142 }
2143 
2144 /*
2145  * Free all of a swapdev's extent information
2146  */
2147 static void destroy_swap_extents(struct swap_info_struct *sis)
2148 {
2149 	while (!RB_EMPTY_ROOT(&sis->swap_extent_root)) {
2150 		struct rb_node *rb = sis->swap_extent_root.rb_node;
2151 		struct swap_extent *se = rb_entry(rb, struct swap_extent, rb_node);
2152 
2153 		rb_erase(rb, &sis->swap_extent_root);
2154 		kfree(se);
2155 	}
2156 
2157 	if (sis->flags & SWP_ACTIVATED) {
2158 		struct file *swap_file = sis->swap_file;
2159 		struct address_space *mapping = swap_file->f_mapping;
2160 
2161 		sis->flags &= ~SWP_ACTIVATED;
2162 		if (mapping->a_ops->swap_deactivate)
2163 			mapping->a_ops->swap_deactivate(swap_file);
2164 	}
2165 }
2166 
2167 /*
2168  * Add a block range (and the corresponding page range) into this swapdev's
2169  * extent tree.
2170  *
2171  * This function rather assumes that it is called in ascending page order.
2172  */
2173 int
2174 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
2175 		unsigned long nr_pages, sector_t start_block)
2176 {
2177 	struct rb_node **link = &sis->swap_extent_root.rb_node, *parent = NULL;
2178 	struct swap_extent *se;
2179 	struct swap_extent *new_se;
2180 
2181 	/*
2182 	 * place the new node at the right most since the
2183 	 * function is called in ascending page order.
2184 	 */
2185 	while (*link) {
2186 		parent = *link;
2187 		link = &parent->rb_right;
2188 	}
2189 
2190 	if (parent) {
2191 		se = rb_entry(parent, struct swap_extent, rb_node);
2192 		BUG_ON(se->start_page + se->nr_pages != start_page);
2193 		if (se->start_block + se->nr_pages == start_block) {
2194 			/* Merge it */
2195 			se->nr_pages += nr_pages;
2196 			return 0;
2197 		}
2198 	}
2199 
2200 	/* No merge, insert a new extent. */
2201 	new_se = kmalloc(sizeof(*se), GFP_KERNEL);
2202 	if (new_se == NULL)
2203 		return -ENOMEM;
2204 	new_se->start_page = start_page;
2205 	new_se->nr_pages = nr_pages;
2206 	new_se->start_block = start_block;
2207 
2208 	rb_link_node(&new_se->rb_node, parent, link);
2209 	rb_insert_color(&new_se->rb_node, &sis->swap_extent_root);
2210 	return 1;
2211 }
2212 EXPORT_SYMBOL_GPL(add_swap_extent);
2213 
2214 /*
2215  * A `swap extent' is a simple thing which maps a contiguous range of pages
2216  * onto a contiguous range of disk blocks.  A rbtree of swap extents is
2217  * built at swapon time and is then used at swap_writepage/swap_readpage
2218  * time for locating where on disk a page belongs.
2219  *
2220  * If the swapfile is an S_ISBLK block device, a single extent is installed.
2221  * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2222  * swap files identically.
2223  *
2224  * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2225  * extent rbtree operates in PAGE_SIZE disk blocks.  Both S_ISREG and S_ISBLK
2226  * swapfiles are handled *identically* after swapon time.
2227  *
2228  * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2229  * and will parse them into a rbtree, in PAGE_SIZE chunks.  If some stray
2230  * blocks are found which do not fall within the PAGE_SIZE alignment
2231  * requirements, they are simply tossed out - we will never use those blocks
2232  * for swapping.
2233  *
2234  * For all swap devices we set S_SWAPFILE across the life of the swapon.  This
2235  * prevents users from writing to the swap device, which will corrupt memory.
2236  *
2237  * The amount of disk space which a single swap extent represents varies.
2238  * Typically it is in the 1-4 megabyte range.  So we can have hundreds of
2239  * extents in the rbtree. - akpm.
2240  */
2241 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
2242 {
2243 	struct file *swap_file = sis->swap_file;
2244 	struct address_space *mapping = swap_file->f_mapping;
2245 	struct inode *inode = mapping->host;
2246 	int ret;
2247 
2248 	if (S_ISBLK(inode->i_mode)) {
2249 		ret = add_swap_extent(sis, 0, sis->max, 0);
2250 		*span = sis->pages;
2251 		return ret;
2252 	}
2253 
2254 	if (mapping->a_ops->swap_activate) {
2255 		ret = mapping->a_ops->swap_activate(sis, swap_file, span);
2256 		if (ret < 0)
2257 			return ret;
2258 		sis->flags |= SWP_ACTIVATED;
2259 		if ((sis->flags & SWP_FS_OPS) &&
2260 		    sio_pool_init() != 0) {
2261 			destroy_swap_extents(sis);
2262 			return -ENOMEM;
2263 		}
2264 		return ret;
2265 	}
2266 
2267 	return generic_swapfile_activate(sis, swap_file, span);
2268 }
2269 
2270 static int swap_node(struct swap_info_struct *p)
2271 {
2272 	struct block_device *bdev;
2273 
2274 	if (p->bdev)
2275 		bdev = p->bdev;
2276 	else
2277 		bdev = p->swap_file->f_inode->i_sb->s_bdev;
2278 
2279 	return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE;
2280 }
2281 
2282 static void setup_swap_info(struct swap_info_struct *p, int prio,
2283 			    unsigned char *swap_map,
2284 			    struct swap_cluster_info *cluster_info)
2285 {
2286 	int i;
2287 
2288 	if (prio >= 0)
2289 		p->prio = prio;
2290 	else
2291 		p->prio = --least_priority;
2292 	/*
2293 	 * the plist prio is negated because plist ordering is
2294 	 * low-to-high, while swap ordering is high-to-low
2295 	 */
2296 	p->list.prio = -p->prio;
2297 	for_each_node(i) {
2298 		if (p->prio >= 0)
2299 			p->avail_lists[i].prio = -p->prio;
2300 		else {
2301 			if (swap_node(p) == i)
2302 				p->avail_lists[i].prio = 1;
2303 			else
2304 				p->avail_lists[i].prio = -p->prio;
2305 		}
2306 	}
2307 	p->swap_map = swap_map;
2308 	p->cluster_info = cluster_info;
2309 }
2310 
2311 static void _enable_swap_info(struct swap_info_struct *p)
2312 {
2313 	p->flags |= SWP_WRITEOK;
2314 	atomic_long_add(p->pages, &nr_swap_pages);
2315 	total_swap_pages += p->pages;
2316 
2317 	assert_spin_locked(&swap_lock);
2318 	/*
2319 	 * both lists are plists, and thus priority ordered.
2320 	 * swap_active_head needs to be priority ordered for swapoff(),
2321 	 * which on removal of any swap_info_struct with an auto-assigned
2322 	 * (i.e. negative) priority increments the auto-assigned priority
2323 	 * of any lower-priority swap_info_structs.
2324 	 * swap_avail_head needs to be priority ordered for folio_alloc_swap(),
2325 	 * which allocates swap pages from the highest available priority
2326 	 * swap_info_struct.
2327 	 */
2328 	plist_add(&p->list, &swap_active_head);
2329 	add_to_avail_list(p);
2330 }
2331 
2332 static void enable_swap_info(struct swap_info_struct *p, int prio,
2333 				unsigned char *swap_map,
2334 				struct swap_cluster_info *cluster_info,
2335 				unsigned long *frontswap_map)
2336 {
2337 	if (IS_ENABLED(CONFIG_FRONTSWAP))
2338 		frontswap_init(p->type, frontswap_map);
2339 	spin_lock(&swap_lock);
2340 	spin_lock(&p->lock);
2341 	setup_swap_info(p, prio, swap_map, cluster_info);
2342 	spin_unlock(&p->lock);
2343 	spin_unlock(&swap_lock);
2344 	/*
2345 	 * Finished initializing swap device, now it's safe to reference it.
2346 	 */
2347 	percpu_ref_resurrect(&p->users);
2348 	spin_lock(&swap_lock);
2349 	spin_lock(&p->lock);
2350 	_enable_swap_info(p);
2351 	spin_unlock(&p->lock);
2352 	spin_unlock(&swap_lock);
2353 }
2354 
2355 static void reinsert_swap_info(struct swap_info_struct *p)
2356 {
2357 	spin_lock(&swap_lock);
2358 	spin_lock(&p->lock);
2359 	setup_swap_info(p, p->prio, p->swap_map, p->cluster_info);
2360 	_enable_swap_info(p);
2361 	spin_unlock(&p->lock);
2362 	spin_unlock(&swap_lock);
2363 }
2364 
2365 bool has_usable_swap(void)
2366 {
2367 	bool ret = true;
2368 
2369 	spin_lock(&swap_lock);
2370 	if (plist_head_empty(&swap_active_head))
2371 		ret = false;
2372 	spin_unlock(&swap_lock);
2373 	return ret;
2374 }
2375 
2376 SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
2377 {
2378 	struct swap_info_struct *p = NULL;
2379 	unsigned char *swap_map;
2380 	struct swap_cluster_info *cluster_info;
2381 	unsigned long *frontswap_map;
2382 	struct file *swap_file, *victim;
2383 	struct address_space *mapping;
2384 	struct inode *inode;
2385 	struct filename *pathname;
2386 	int err, found = 0;
2387 	unsigned int old_block_size;
2388 
2389 	if (!capable(CAP_SYS_ADMIN))
2390 		return -EPERM;
2391 
2392 	BUG_ON(!current->mm);
2393 
2394 	pathname = getname(specialfile);
2395 	if (IS_ERR(pathname))
2396 		return PTR_ERR(pathname);
2397 
2398 	victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
2399 	err = PTR_ERR(victim);
2400 	if (IS_ERR(victim))
2401 		goto out;
2402 
2403 	mapping = victim->f_mapping;
2404 	spin_lock(&swap_lock);
2405 	plist_for_each_entry(p, &swap_active_head, list) {
2406 		if (p->flags & SWP_WRITEOK) {
2407 			if (p->swap_file->f_mapping == mapping) {
2408 				found = 1;
2409 				break;
2410 			}
2411 		}
2412 	}
2413 	if (!found) {
2414 		err = -EINVAL;
2415 		spin_unlock(&swap_lock);
2416 		goto out_dput;
2417 	}
2418 	if (!security_vm_enough_memory_mm(current->mm, p->pages))
2419 		vm_unacct_memory(p->pages);
2420 	else {
2421 		err = -ENOMEM;
2422 		spin_unlock(&swap_lock);
2423 		goto out_dput;
2424 	}
2425 	del_from_avail_list(p);
2426 	spin_lock(&p->lock);
2427 	if (p->prio < 0) {
2428 		struct swap_info_struct *si = p;
2429 		int nid;
2430 
2431 		plist_for_each_entry_continue(si, &swap_active_head, list) {
2432 			si->prio++;
2433 			si->list.prio--;
2434 			for_each_node(nid) {
2435 				if (si->avail_lists[nid].prio != 1)
2436 					si->avail_lists[nid].prio--;
2437 			}
2438 		}
2439 		least_priority++;
2440 	}
2441 	plist_del(&p->list, &swap_active_head);
2442 	atomic_long_sub(p->pages, &nr_swap_pages);
2443 	total_swap_pages -= p->pages;
2444 	p->flags &= ~SWP_WRITEOK;
2445 	spin_unlock(&p->lock);
2446 	spin_unlock(&swap_lock);
2447 
2448 	disable_swap_slots_cache_lock();
2449 
2450 	set_current_oom_origin();
2451 	err = try_to_unuse(p->type);
2452 	clear_current_oom_origin();
2453 
2454 	if (err) {
2455 		/* re-insert swap space back into swap_list */
2456 		reinsert_swap_info(p);
2457 		reenable_swap_slots_cache_unlock();
2458 		goto out_dput;
2459 	}
2460 
2461 	reenable_swap_slots_cache_unlock();
2462 
2463 	/*
2464 	 * Wait for swap operations protected by get/put_swap_device()
2465 	 * to complete.
2466 	 *
2467 	 * We need synchronize_rcu() here to protect the accessing to
2468 	 * the swap cache data structure.
2469 	 */
2470 	percpu_ref_kill(&p->users);
2471 	synchronize_rcu();
2472 	wait_for_completion(&p->comp);
2473 
2474 	flush_work(&p->discard_work);
2475 
2476 	destroy_swap_extents(p);
2477 	if (p->flags & SWP_CONTINUED)
2478 		free_swap_count_continuations(p);
2479 
2480 	if (!p->bdev || !bdev_nonrot(p->bdev))
2481 		atomic_dec(&nr_rotate_swap);
2482 
2483 	mutex_lock(&swapon_mutex);
2484 	spin_lock(&swap_lock);
2485 	spin_lock(&p->lock);
2486 	drain_mmlist();
2487 
2488 	/* wait for anyone still in scan_swap_map_slots */
2489 	p->highest_bit = 0;		/* cuts scans short */
2490 	while (p->flags >= SWP_SCANNING) {
2491 		spin_unlock(&p->lock);
2492 		spin_unlock(&swap_lock);
2493 		schedule_timeout_uninterruptible(1);
2494 		spin_lock(&swap_lock);
2495 		spin_lock(&p->lock);
2496 	}
2497 
2498 	swap_file = p->swap_file;
2499 	old_block_size = p->old_block_size;
2500 	p->swap_file = NULL;
2501 	p->max = 0;
2502 	swap_map = p->swap_map;
2503 	p->swap_map = NULL;
2504 	cluster_info = p->cluster_info;
2505 	p->cluster_info = NULL;
2506 	frontswap_map = frontswap_map_get(p);
2507 	spin_unlock(&p->lock);
2508 	spin_unlock(&swap_lock);
2509 	arch_swap_invalidate_area(p->type);
2510 	frontswap_invalidate_area(p->type);
2511 	frontswap_map_set(p, NULL);
2512 	mutex_unlock(&swapon_mutex);
2513 	free_percpu(p->percpu_cluster);
2514 	p->percpu_cluster = NULL;
2515 	free_percpu(p->cluster_next_cpu);
2516 	p->cluster_next_cpu = NULL;
2517 	vfree(swap_map);
2518 	kvfree(cluster_info);
2519 	kvfree(frontswap_map);
2520 	/* Destroy swap account information */
2521 	swap_cgroup_swapoff(p->type);
2522 	exit_swap_address_space(p->type);
2523 
2524 	inode = mapping->host;
2525 	if (S_ISBLK(inode->i_mode)) {
2526 		struct block_device *bdev = I_BDEV(inode);
2527 
2528 		set_blocksize(bdev, old_block_size);
2529 		blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2530 	}
2531 
2532 	inode_lock(inode);
2533 	inode->i_flags &= ~S_SWAPFILE;
2534 	inode_unlock(inode);
2535 	filp_close(swap_file, NULL);
2536 
2537 	/*
2538 	 * Clear the SWP_USED flag after all resources are freed so that swapon
2539 	 * can reuse this swap_info in alloc_swap_info() safely.  It is ok to
2540 	 * not hold p->lock after we cleared its SWP_WRITEOK.
2541 	 */
2542 	spin_lock(&swap_lock);
2543 	p->flags = 0;
2544 	spin_unlock(&swap_lock);
2545 
2546 	err = 0;
2547 	atomic_inc(&proc_poll_event);
2548 	wake_up_interruptible(&proc_poll_wait);
2549 
2550 out_dput:
2551 	filp_close(victim, NULL);
2552 out:
2553 	putname(pathname);
2554 	return err;
2555 }
2556 
2557 #ifdef CONFIG_PROC_FS
2558 static __poll_t swaps_poll(struct file *file, poll_table *wait)
2559 {
2560 	struct seq_file *seq = file->private_data;
2561 
2562 	poll_wait(file, &proc_poll_wait, wait);
2563 
2564 	if (seq->poll_event != atomic_read(&proc_poll_event)) {
2565 		seq->poll_event = atomic_read(&proc_poll_event);
2566 		return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
2567 	}
2568 
2569 	return EPOLLIN | EPOLLRDNORM;
2570 }
2571 
2572 /* iterator */
2573 static void *swap_start(struct seq_file *swap, loff_t *pos)
2574 {
2575 	struct swap_info_struct *si;
2576 	int type;
2577 	loff_t l = *pos;
2578 
2579 	mutex_lock(&swapon_mutex);
2580 
2581 	if (!l)
2582 		return SEQ_START_TOKEN;
2583 
2584 	for (type = 0; (si = swap_type_to_swap_info(type)); type++) {
2585 		if (!(si->flags & SWP_USED) || !si->swap_map)
2586 			continue;
2587 		if (!--l)
2588 			return si;
2589 	}
2590 
2591 	return NULL;
2592 }
2593 
2594 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2595 {
2596 	struct swap_info_struct *si = v;
2597 	int type;
2598 
2599 	if (v == SEQ_START_TOKEN)
2600 		type = 0;
2601 	else
2602 		type = si->type + 1;
2603 
2604 	++(*pos);
2605 	for (; (si = swap_type_to_swap_info(type)); type++) {
2606 		if (!(si->flags & SWP_USED) || !si->swap_map)
2607 			continue;
2608 		return si;
2609 	}
2610 
2611 	return NULL;
2612 }
2613 
2614 static void swap_stop(struct seq_file *swap, void *v)
2615 {
2616 	mutex_unlock(&swapon_mutex);
2617 }
2618 
2619 static int swap_show(struct seq_file *swap, void *v)
2620 {
2621 	struct swap_info_struct *si = v;
2622 	struct file *file;
2623 	int len;
2624 	unsigned long bytes, inuse;
2625 
2626 	if (si == SEQ_START_TOKEN) {
2627 		seq_puts(swap, "Filename\t\t\t\tType\t\tSize\t\tUsed\t\tPriority\n");
2628 		return 0;
2629 	}
2630 
2631 	bytes = si->pages << (PAGE_SHIFT - 10);
2632 	inuse = READ_ONCE(si->inuse_pages) << (PAGE_SHIFT - 10);
2633 
2634 	file = si->swap_file;
2635 	len = seq_file_path(swap, file, " \t\n\\");
2636 	seq_printf(swap, "%*s%s\t%lu\t%s%lu\t%s%d\n",
2637 			len < 40 ? 40 - len : 1, " ",
2638 			S_ISBLK(file_inode(file)->i_mode) ?
2639 				"partition" : "file\t",
2640 			bytes, bytes < 10000000 ? "\t" : "",
2641 			inuse, inuse < 10000000 ? "\t" : "",
2642 			si->prio);
2643 	return 0;
2644 }
2645 
2646 static const struct seq_operations swaps_op = {
2647 	.start =	swap_start,
2648 	.next =		swap_next,
2649 	.stop =		swap_stop,
2650 	.show =		swap_show
2651 };
2652 
2653 static int swaps_open(struct inode *inode, struct file *file)
2654 {
2655 	struct seq_file *seq;
2656 	int ret;
2657 
2658 	ret = seq_open(file, &swaps_op);
2659 	if (ret)
2660 		return ret;
2661 
2662 	seq = file->private_data;
2663 	seq->poll_event = atomic_read(&proc_poll_event);
2664 	return 0;
2665 }
2666 
2667 static const struct proc_ops swaps_proc_ops = {
2668 	.proc_flags	= PROC_ENTRY_PERMANENT,
2669 	.proc_open	= swaps_open,
2670 	.proc_read	= seq_read,
2671 	.proc_lseek	= seq_lseek,
2672 	.proc_release	= seq_release,
2673 	.proc_poll	= swaps_poll,
2674 };
2675 
2676 static int __init procswaps_init(void)
2677 {
2678 	proc_create("swaps", 0, NULL, &swaps_proc_ops);
2679 	return 0;
2680 }
2681 __initcall(procswaps_init);
2682 #endif /* CONFIG_PROC_FS */
2683 
2684 #ifdef MAX_SWAPFILES_CHECK
2685 static int __init max_swapfiles_check(void)
2686 {
2687 	MAX_SWAPFILES_CHECK();
2688 	return 0;
2689 }
2690 late_initcall(max_swapfiles_check);
2691 #endif
2692 
2693 static struct swap_info_struct *alloc_swap_info(void)
2694 {
2695 	struct swap_info_struct *p;
2696 	struct swap_info_struct *defer = NULL;
2697 	unsigned int type;
2698 	int i;
2699 
2700 	p = kvzalloc(struct_size(p, avail_lists, nr_node_ids), GFP_KERNEL);
2701 	if (!p)
2702 		return ERR_PTR(-ENOMEM);
2703 
2704 	if (percpu_ref_init(&p->users, swap_users_ref_free,
2705 			    PERCPU_REF_INIT_DEAD, GFP_KERNEL)) {
2706 		kvfree(p);
2707 		return ERR_PTR(-ENOMEM);
2708 	}
2709 
2710 	spin_lock(&swap_lock);
2711 	for (type = 0; type < nr_swapfiles; type++) {
2712 		if (!(swap_info[type]->flags & SWP_USED))
2713 			break;
2714 	}
2715 	if (type >= MAX_SWAPFILES) {
2716 		spin_unlock(&swap_lock);
2717 		percpu_ref_exit(&p->users);
2718 		kvfree(p);
2719 		return ERR_PTR(-EPERM);
2720 	}
2721 	if (type >= nr_swapfiles) {
2722 		p->type = type;
2723 		/*
2724 		 * Publish the swap_info_struct after initializing it.
2725 		 * Note that kvzalloc() above zeroes all its fields.
2726 		 */
2727 		smp_store_release(&swap_info[type], p); /* rcu_assign_pointer() */
2728 		nr_swapfiles++;
2729 	} else {
2730 		defer = p;
2731 		p = swap_info[type];
2732 		/*
2733 		 * Do not memset this entry: a racing procfs swap_next()
2734 		 * would be relying on p->type to remain valid.
2735 		 */
2736 	}
2737 	p->swap_extent_root = RB_ROOT;
2738 	plist_node_init(&p->list, 0);
2739 	for_each_node(i)
2740 		plist_node_init(&p->avail_lists[i], 0);
2741 	p->flags = SWP_USED;
2742 	spin_unlock(&swap_lock);
2743 	if (defer) {
2744 		percpu_ref_exit(&defer->users);
2745 		kvfree(defer);
2746 	}
2747 	spin_lock_init(&p->lock);
2748 	spin_lock_init(&p->cont_lock);
2749 	init_completion(&p->comp);
2750 
2751 	return p;
2752 }
2753 
2754 static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2755 {
2756 	int error;
2757 
2758 	if (S_ISBLK(inode->i_mode)) {
2759 		p->bdev = blkdev_get_by_dev(inode->i_rdev,
2760 				   FMODE_READ | FMODE_WRITE | FMODE_EXCL, p);
2761 		if (IS_ERR(p->bdev)) {
2762 			error = PTR_ERR(p->bdev);
2763 			p->bdev = NULL;
2764 			return error;
2765 		}
2766 		p->old_block_size = block_size(p->bdev);
2767 		error = set_blocksize(p->bdev, PAGE_SIZE);
2768 		if (error < 0)
2769 			return error;
2770 		/*
2771 		 * Zoned block devices contain zones that have a sequential
2772 		 * write only restriction.  Hence zoned block devices are not
2773 		 * suitable for swapping.  Disallow them here.
2774 		 */
2775 		if (bdev_is_zoned(p->bdev))
2776 			return -EINVAL;
2777 		p->flags |= SWP_BLKDEV;
2778 	} else if (S_ISREG(inode->i_mode)) {
2779 		p->bdev = inode->i_sb->s_bdev;
2780 	}
2781 
2782 	return 0;
2783 }
2784 
2785 
2786 /*
2787  * Find out how many pages are allowed for a single swap device. There
2788  * are two limiting factors:
2789  * 1) the number of bits for the swap offset in the swp_entry_t type, and
2790  * 2) the number of bits in the swap pte, as defined by the different
2791  * architectures.
2792  *
2793  * In order to find the largest possible bit mask, a swap entry with
2794  * swap type 0 and swap offset ~0UL is created, encoded to a swap pte,
2795  * decoded to a swp_entry_t again, and finally the swap offset is
2796  * extracted.
2797  *
2798  * This will mask all the bits from the initial ~0UL mask that can't
2799  * be encoded in either the swp_entry_t or the architecture definition
2800  * of a swap pte.
2801  */
2802 unsigned long generic_max_swapfile_size(void)
2803 {
2804 	return swp_offset(pte_to_swp_entry(
2805 			swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2806 }
2807 
2808 /* Can be overridden by an architecture for additional checks. */
2809 __weak unsigned long arch_max_swapfile_size(void)
2810 {
2811 	return generic_max_swapfile_size();
2812 }
2813 
2814 static unsigned long read_swap_header(struct swap_info_struct *p,
2815 					union swap_header *swap_header,
2816 					struct inode *inode)
2817 {
2818 	int i;
2819 	unsigned long maxpages;
2820 	unsigned long swapfilepages;
2821 	unsigned long last_page;
2822 
2823 	if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
2824 		pr_err("Unable to find swap-space signature\n");
2825 		return 0;
2826 	}
2827 
2828 	/* swap partition endianness hack... */
2829 	if (swab32(swap_header->info.version) == 1) {
2830 		swab32s(&swap_header->info.version);
2831 		swab32s(&swap_header->info.last_page);
2832 		swab32s(&swap_header->info.nr_badpages);
2833 		if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2834 			return 0;
2835 		for (i = 0; i < swap_header->info.nr_badpages; i++)
2836 			swab32s(&swap_header->info.badpages[i]);
2837 	}
2838 	/* Check the swap header's sub-version */
2839 	if (swap_header->info.version != 1) {
2840 		pr_warn("Unable to handle swap header version %d\n",
2841 			swap_header->info.version);
2842 		return 0;
2843 	}
2844 
2845 	p->lowest_bit  = 1;
2846 	p->cluster_next = 1;
2847 	p->cluster_nr = 0;
2848 
2849 	maxpages = swapfile_maximum_size;
2850 	last_page = swap_header->info.last_page;
2851 	if (!last_page) {
2852 		pr_warn("Empty swap-file\n");
2853 		return 0;
2854 	}
2855 	if (last_page > maxpages) {
2856 		pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
2857 			maxpages << (PAGE_SHIFT - 10),
2858 			last_page << (PAGE_SHIFT - 10));
2859 	}
2860 	if (maxpages > last_page) {
2861 		maxpages = last_page + 1;
2862 		/* p->max is an unsigned int: don't overflow it */
2863 		if ((unsigned int)maxpages == 0)
2864 			maxpages = UINT_MAX;
2865 	}
2866 	p->highest_bit = maxpages - 1;
2867 
2868 	if (!maxpages)
2869 		return 0;
2870 	swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2871 	if (swapfilepages && maxpages > swapfilepages) {
2872 		pr_warn("Swap area shorter than signature indicates\n");
2873 		return 0;
2874 	}
2875 	if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
2876 		return 0;
2877 	if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2878 		return 0;
2879 
2880 	return maxpages;
2881 }
2882 
2883 #define SWAP_CLUSTER_INFO_COLS						\
2884 	DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
2885 #define SWAP_CLUSTER_SPACE_COLS						\
2886 	DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
2887 #define SWAP_CLUSTER_COLS						\
2888 	max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
2889 
2890 static int setup_swap_map_and_extents(struct swap_info_struct *p,
2891 					union swap_header *swap_header,
2892 					unsigned char *swap_map,
2893 					struct swap_cluster_info *cluster_info,
2894 					unsigned long maxpages,
2895 					sector_t *span)
2896 {
2897 	unsigned int j, k;
2898 	unsigned int nr_good_pages;
2899 	int nr_extents;
2900 	unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
2901 	unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS;
2902 	unsigned long i, idx;
2903 
2904 	nr_good_pages = maxpages - 1;	/* omit header page */
2905 
2906 	cluster_list_init(&p->free_clusters);
2907 	cluster_list_init(&p->discard_clusters);
2908 
2909 	for (i = 0; i < swap_header->info.nr_badpages; i++) {
2910 		unsigned int page_nr = swap_header->info.badpages[i];
2911 		if (page_nr == 0 || page_nr > swap_header->info.last_page)
2912 			return -EINVAL;
2913 		if (page_nr < maxpages) {
2914 			swap_map[page_nr] = SWAP_MAP_BAD;
2915 			nr_good_pages--;
2916 			/*
2917 			 * Haven't marked the cluster free yet, no list
2918 			 * operation involved
2919 			 */
2920 			inc_cluster_info_page(p, cluster_info, page_nr);
2921 		}
2922 	}
2923 
2924 	/* Haven't marked the cluster free yet, no list operation involved */
2925 	for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
2926 		inc_cluster_info_page(p, cluster_info, i);
2927 
2928 	if (nr_good_pages) {
2929 		swap_map[0] = SWAP_MAP_BAD;
2930 		/*
2931 		 * Not mark the cluster free yet, no list
2932 		 * operation involved
2933 		 */
2934 		inc_cluster_info_page(p, cluster_info, 0);
2935 		p->max = maxpages;
2936 		p->pages = nr_good_pages;
2937 		nr_extents = setup_swap_extents(p, span);
2938 		if (nr_extents < 0)
2939 			return nr_extents;
2940 		nr_good_pages = p->pages;
2941 	}
2942 	if (!nr_good_pages) {
2943 		pr_warn("Empty swap-file\n");
2944 		return -EINVAL;
2945 	}
2946 
2947 	if (!cluster_info)
2948 		return nr_extents;
2949 
2950 
2951 	/*
2952 	 * Reduce false cache line sharing between cluster_info and
2953 	 * sharing same address space.
2954 	 */
2955 	for (k = 0; k < SWAP_CLUSTER_COLS; k++) {
2956 		j = (k + col) % SWAP_CLUSTER_COLS;
2957 		for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) {
2958 			idx = i * SWAP_CLUSTER_COLS + j;
2959 			if (idx >= nr_clusters)
2960 				continue;
2961 			if (cluster_count(&cluster_info[idx]))
2962 				continue;
2963 			cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
2964 			cluster_list_add_tail(&p->free_clusters, cluster_info,
2965 					      idx);
2966 		}
2967 	}
2968 	return nr_extents;
2969 }
2970 
2971 SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2972 {
2973 	struct swap_info_struct *p;
2974 	struct filename *name;
2975 	struct file *swap_file = NULL;
2976 	struct address_space *mapping;
2977 	struct dentry *dentry;
2978 	int prio;
2979 	int error;
2980 	union swap_header *swap_header;
2981 	int nr_extents;
2982 	sector_t span;
2983 	unsigned long maxpages;
2984 	unsigned char *swap_map = NULL;
2985 	struct swap_cluster_info *cluster_info = NULL;
2986 	unsigned long *frontswap_map = NULL;
2987 	struct page *page = NULL;
2988 	struct inode *inode = NULL;
2989 	bool inced_nr_rotate_swap = false;
2990 
2991 	if (swap_flags & ~SWAP_FLAGS_VALID)
2992 		return -EINVAL;
2993 
2994 	if (!capable(CAP_SYS_ADMIN))
2995 		return -EPERM;
2996 
2997 	if (!swap_avail_heads)
2998 		return -ENOMEM;
2999 
3000 	p = alloc_swap_info();
3001 	if (IS_ERR(p))
3002 		return PTR_ERR(p);
3003 
3004 	INIT_WORK(&p->discard_work, swap_discard_work);
3005 
3006 	name = getname(specialfile);
3007 	if (IS_ERR(name)) {
3008 		error = PTR_ERR(name);
3009 		name = NULL;
3010 		goto bad_swap;
3011 	}
3012 	swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
3013 	if (IS_ERR(swap_file)) {
3014 		error = PTR_ERR(swap_file);
3015 		swap_file = NULL;
3016 		goto bad_swap;
3017 	}
3018 
3019 	p->swap_file = swap_file;
3020 	mapping = swap_file->f_mapping;
3021 	dentry = swap_file->f_path.dentry;
3022 	inode = mapping->host;
3023 
3024 	error = claim_swapfile(p, inode);
3025 	if (unlikely(error))
3026 		goto bad_swap;
3027 
3028 	inode_lock(inode);
3029 	if (d_unlinked(dentry) || cant_mount(dentry)) {
3030 		error = -ENOENT;
3031 		goto bad_swap_unlock_inode;
3032 	}
3033 	if (IS_SWAPFILE(inode)) {
3034 		error = -EBUSY;
3035 		goto bad_swap_unlock_inode;
3036 	}
3037 
3038 	/*
3039 	 * Read the swap header.
3040 	 */
3041 	if (!mapping->a_ops->read_folio) {
3042 		error = -EINVAL;
3043 		goto bad_swap_unlock_inode;
3044 	}
3045 	page = read_mapping_page(mapping, 0, swap_file);
3046 	if (IS_ERR(page)) {
3047 		error = PTR_ERR(page);
3048 		goto bad_swap_unlock_inode;
3049 	}
3050 	swap_header = kmap(page);
3051 
3052 	maxpages = read_swap_header(p, swap_header, inode);
3053 	if (unlikely(!maxpages)) {
3054 		error = -EINVAL;
3055 		goto bad_swap_unlock_inode;
3056 	}
3057 
3058 	/* OK, set up the swap map and apply the bad block list */
3059 	swap_map = vzalloc(maxpages);
3060 	if (!swap_map) {
3061 		error = -ENOMEM;
3062 		goto bad_swap_unlock_inode;
3063 	}
3064 
3065 	if (p->bdev && bdev_stable_writes(p->bdev))
3066 		p->flags |= SWP_STABLE_WRITES;
3067 
3068 	if (p->bdev && p->bdev->bd_disk->fops->rw_page)
3069 		p->flags |= SWP_SYNCHRONOUS_IO;
3070 
3071 	if (p->bdev && bdev_nonrot(p->bdev)) {
3072 		int cpu;
3073 		unsigned long ci, nr_cluster;
3074 
3075 		p->flags |= SWP_SOLIDSTATE;
3076 		p->cluster_next_cpu = alloc_percpu(unsigned int);
3077 		if (!p->cluster_next_cpu) {
3078 			error = -ENOMEM;
3079 			goto bad_swap_unlock_inode;
3080 		}
3081 		/*
3082 		 * select a random position to start with to help wear leveling
3083 		 * SSD
3084 		 */
3085 		for_each_possible_cpu(cpu) {
3086 			per_cpu(*p->cluster_next_cpu, cpu) =
3087 				1 + prandom_u32_max(p->highest_bit);
3088 		}
3089 		nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
3090 
3091 		cluster_info = kvcalloc(nr_cluster, sizeof(*cluster_info),
3092 					GFP_KERNEL);
3093 		if (!cluster_info) {
3094 			error = -ENOMEM;
3095 			goto bad_swap_unlock_inode;
3096 		}
3097 
3098 		for (ci = 0; ci < nr_cluster; ci++)
3099 			spin_lock_init(&((cluster_info + ci)->lock));
3100 
3101 		p->percpu_cluster = alloc_percpu(struct percpu_cluster);
3102 		if (!p->percpu_cluster) {
3103 			error = -ENOMEM;
3104 			goto bad_swap_unlock_inode;
3105 		}
3106 		for_each_possible_cpu(cpu) {
3107 			struct percpu_cluster *cluster;
3108 			cluster = per_cpu_ptr(p->percpu_cluster, cpu);
3109 			cluster_set_null(&cluster->index);
3110 		}
3111 	} else {
3112 		atomic_inc(&nr_rotate_swap);
3113 		inced_nr_rotate_swap = true;
3114 	}
3115 
3116 	error = swap_cgroup_swapon(p->type, maxpages);
3117 	if (error)
3118 		goto bad_swap_unlock_inode;
3119 
3120 	nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
3121 		cluster_info, maxpages, &span);
3122 	if (unlikely(nr_extents < 0)) {
3123 		error = nr_extents;
3124 		goto bad_swap_unlock_inode;
3125 	}
3126 	/* frontswap enabled? set up bit-per-page map for frontswap */
3127 	if (IS_ENABLED(CONFIG_FRONTSWAP))
3128 		frontswap_map = kvcalloc(BITS_TO_LONGS(maxpages),
3129 					 sizeof(long),
3130 					 GFP_KERNEL);
3131 
3132 	if ((swap_flags & SWAP_FLAG_DISCARD) &&
3133 	    p->bdev && bdev_max_discard_sectors(p->bdev)) {
3134 		/*
3135 		 * When discard is enabled for swap with no particular
3136 		 * policy flagged, we set all swap discard flags here in
3137 		 * order to sustain backward compatibility with older
3138 		 * swapon(8) releases.
3139 		 */
3140 		p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
3141 			     SWP_PAGE_DISCARD);
3142 
3143 		/*
3144 		 * By flagging sys_swapon, a sysadmin can tell us to
3145 		 * either do single-time area discards only, or to just
3146 		 * perform discards for released swap page-clusters.
3147 		 * Now it's time to adjust the p->flags accordingly.
3148 		 */
3149 		if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
3150 			p->flags &= ~SWP_PAGE_DISCARD;
3151 		else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
3152 			p->flags &= ~SWP_AREA_DISCARD;
3153 
3154 		/* issue a swapon-time discard if it's still required */
3155 		if (p->flags & SWP_AREA_DISCARD) {
3156 			int err = discard_swap(p);
3157 			if (unlikely(err))
3158 				pr_err("swapon: discard_swap(%p): %d\n",
3159 					p, err);
3160 		}
3161 	}
3162 
3163 	error = init_swap_address_space(p->type, maxpages);
3164 	if (error)
3165 		goto bad_swap_unlock_inode;
3166 
3167 	/*
3168 	 * Flush any pending IO and dirty mappings before we start using this
3169 	 * swap device.
3170 	 */
3171 	inode->i_flags |= S_SWAPFILE;
3172 	error = inode_drain_writes(inode);
3173 	if (error) {
3174 		inode->i_flags &= ~S_SWAPFILE;
3175 		goto free_swap_address_space;
3176 	}
3177 
3178 	mutex_lock(&swapon_mutex);
3179 	prio = -1;
3180 	if (swap_flags & SWAP_FLAG_PREFER)
3181 		prio =
3182 		  (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
3183 	enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
3184 
3185 	pr_info("Adding %uk swap on %s.  Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
3186 		p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
3187 		nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
3188 		(p->flags & SWP_SOLIDSTATE) ? "SS" : "",
3189 		(p->flags & SWP_DISCARDABLE) ? "D" : "",
3190 		(p->flags & SWP_AREA_DISCARD) ? "s" : "",
3191 		(p->flags & SWP_PAGE_DISCARD) ? "c" : "",
3192 		(frontswap_map) ? "FS" : "");
3193 
3194 	mutex_unlock(&swapon_mutex);
3195 	atomic_inc(&proc_poll_event);
3196 	wake_up_interruptible(&proc_poll_wait);
3197 
3198 	error = 0;
3199 	goto out;
3200 free_swap_address_space:
3201 	exit_swap_address_space(p->type);
3202 bad_swap_unlock_inode:
3203 	inode_unlock(inode);
3204 bad_swap:
3205 	free_percpu(p->percpu_cluster);
3206 	p->percpu_cluster = NULL;
3207 	free_percpu(p->cluster_next_cpu);
3208 	p->cluster_next_cpu = NULL;
3209 	if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
3210 		set_blocksize(p->bdev, p->old_block_size);
3211 		blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
3212 	}
3213 	inode = NULL;
3214 	destroy_swap_extents(p);
3215 	swap_cgroup_swapoff(p->type);
3216 	spin_lock(&swap_lock);
3217 	p->swap_file = NULL;
3218 	p->flags = 0;
3219 	spin_unlock(&swap_lock);
3220 	vfree(swap_map);
3221 	kvfree(cluster_info);
3222 	kvfree(frontswap_map);
3223 	if (inced_nr_rotate_swap)
3224 		atomic_dec(&nr_rotate_swap);
3225 	if (swap_file)
3226 		filp_close(swap_file, NULL);
3227 out:
3228 	if (page && !IS_ERR(page)) {
3229 		kunmap(page);
3230 		put_page(page);
3231 	}
3232 	if (name)
3233 		putname(name);
3234 	if (inode)
3235 		inode_unlock(inode);
3236 	if (!error)
3237 		enable_swap_slots_cache();
3238 	return error;
3239 }
3240 
3241 void si_swapinfo(struct sysinfo *val)
3242 {
3243 	unsigned int type;
3244 	unsigned long nr_to_be_unused = 0;
3245 
3246 	spin_lock(&swap_lock);
3247 	for (type = 0; type < nr_swapfiles; type++) {
3248 		struct swap_info_struct *si = swap_info[type];
3249 
3250 		if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
3251 			nr_to_be_unused += READ_ONCE(si->inuse_pages);
3252 	}
3253 	val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
3254 	val->totalswap = total_swap_pages + nr_to_be_unused;
3255 	spin_unlock(&swap_lock);
3256 }
3257 
3258 /*
3259  * Verify that a swap entry is valid and increment its swap map count.
3260  *
3261  * Returns error code in following case.
3262  * - success -> 0
3263  * - swp_entry is invalid -> EINVAL
3264  * - swp_entry is migration entry -> EINVAL
3265  * - swap-cache reference is requested but there is already one. -> EEXIST
3266  * - swap-cache reference is requested but the entry is not used. -> ENOENT
3267  * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
3268  */
3269 static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
3270 {
3271 	struct swap_info_struct *p;
3272 	struct swap_cluster_info *ci;
3273 	unsigned long offset;
3274 	unsigned char count;
3275 	unsigned char has_cache;
3276 	int err;
3277 
3278 	p = get_swap_device(entry);
3279 	if (!p)
3280 		return -EINVAL;
3281 
3282 	offset = swp_offset(entry);
3283 	ci = lock_cluster_or_swap_info(p, offset);
3284 
3285 	count = p->swap_map[offset];
3286 
3287 	/*
3288 	 * swapin_readahead() doesn't check if a swap entry is valid, so the
3289 	 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
3290 	 */
3291 	if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
3292 		err = -ENOENT;
3293 		goto unlock_out;
3294 	}
3295 
3296 	has_cache = count & SWAP_HAS_CACHE;
3297 	count &= ~SWAP_HAS_CACHE;
3298 	err = 0;
3299 
3300 	if (usage == SWAP_HAS_CACHE) {
3301 
3302 		/* set SWAP_HAS_CACHE if there is no cache and entry is used */
3303 		if (!has_cache && count)
3304 			has_cache = SWAP_HAS_CACHE;
3305 		else if (has_cache)		/* someone else added cache */
3306 			err = -EEXIST;
3307 		else				/* no users remaining */
3308 			err = -ENOENT;
3309 
3310 	} else if (count || has_cache) {
3311 
3312 		if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
3313 			count += usage;
3314 		else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
3315 			err = -EINVAL;
3316 		else if (swap_count_continued(p, offset, count))
3317 			count = COUNT_CONTINUED;
3318 		else
3319 			err = -ENOMEM;
3320 	} else
3321 		err = -ENOENT;			/* unused swap entry */
3322 
3323 	WRITE_ONCE(p->swap_map[offset], count | has_cache);
3324 
3325 unlock_out:
3326 	unlock_cluster_or_swap_info(p, ci);
3327 	put_swap_device(p);
3328 	return err;
3329 }
3330 
3331 /*
3332  * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3333  * (in which case its reference count is never incremented).
3334  */
3335 void swap_shmem_alloc(swp_entry_t entry)
3336 {
3337 	__swap_duplicate(entry, SWAP_MAP_SHMEM);
3338 }
3339 
3340 /*
3341  * Increase reference count of swap entry by 1.
3342  * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3343  * but could not be atomically allocated.  Returns 0, just as if it succeeded,
3344  * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3345  * might occur if a page table entry has got corrupted.
3346  */
3347 int swap_duplicate(swp_entry_t entry)
3348 {
3349 	int err = 0;
3350 
3351 	while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
3352 		err = add_swap_count_continuation(entry, GFP_ATOMIC);
3353 	return err;
3354 }
3355 
3356 /*
3357  * @entry: swap entry for which we allocate swap cache.
3358  *
3359  * Called when allocating swap cache for existing swap entry,
3360  * This can return error codes. Returns 0 at success.
3361  * -EEXIST means there is a swap cache.
3362  * Note: return code is different from swap_duplicate().
3363  */
3364 int swapcache_prepare(swp_entry_t entry)
3365 {
3366 	return __swap_duplicate(entry, SWAP_HAS_CACHE);
3367 }
3368 
3369 struct swap_info_struct *swp_swap_info(swp_entry_t entry)
3370 {
3371 	return swap_type_to_swap_info(swp_type(entry));
3372 }
3373 
3374 struct swap_info_struct *page_swap_info(struct page *page)
3375 {
3376 	swp_entry_t entry = { .val = page_private(page) };
3377 	return swp_swap_info(entry);
3378 }
3379 
3380 /*
3381  * out-of-line methods to avoid include hell.
3382  */
3383 struct address_space *swapcache_mapping(struct folio *folio)
3384 {
3385 	return page_swap_info(&folio->page)->swap_file->f_mapping;
3386 }
3387 EXPORT_SYMBOL_GPL(swapcache_mapping);
3388 
3389 pgoff_t __page_file_index(struct page *page)
3390 {
3391 	swp_entry_t swap = { .val = page_private(page) };
3392 	return swp_offset(swap);
3393 }
3394 EXPORT_SYMBOL_GPL(__page_file_index);
3395 
3396 /*
3397  * add_swap_count_continuation - called when a swap count is duplicated
3398  * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3399  * page of the original vmalloc'ed swap_map, to hold the continuation count
3400  * (for that entry and for its neighbouring PAGE_SIZE swap entries).  Called
3401  * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3402  *
3403  * These continuation pages are seldom referenced: the common paths all work
3404  * on the original swap_map, only referring to a continuation page when the
3405  * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3406  *
3407  * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3408  * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3409  * can be called after dropping locks.
3410  */
3411 int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
3412 {
3413 	struct swap_info_struct *si;
3414 	struct swap_cluster_info *ci;
3415 	struct page *head;
3416 	struct page *page;
3417 	struct page *list_page;
3418 	pgoff_t offset;
3419 	unsigned char count;
3420 	int ret = 0;
3421 
3422 	/*
3423 	 * When debugging, it's easier to use __GFP_ZERO here; but it's better
3424 	 * for latency not to zero a page while GFP_ATOMIC and holding locks.
3425 	 */
3426 	page = alloc_page(gfp_mask | __GFP_HIGHMEM);
3427 
3428 	si = get_swap_device(entry);
3429 	if (!si) {
3430 		/*
3431 		 * An acceptable race has occurred since the failing
3432 		 * __swap_duplicate(): the swap device may be swapoff
3433 		 */
3434 		goto outer;
3435 	}
3436 	spin_lock(&si->lock);
3437 
3438 	offset = swp_offset(entry);
3439 
3440 	ci = lock_cluster(si, offset);
3441 
3442 	count = swap_count(si->swap_map[offset]);
3443 
3444 	if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
3445 		/*
3446 		 * The higher the swap count, the more likely it is that tasks
3447 		 * will race to add swap count continuation: we need to avoid
3448 		 * over-provisioning.
3449 		 */
3450 		goto out;
3451 	}
3452 
3453 	if (!page) {
3454 		ret = -ENOMEM;
3455 		goto out;
3456 	}
3457 
3458 	/*
3459 	 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
3460 	 * no architecture is using highmem pages for kernel page tables: so it
3461 	 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
3462 	 */
3463 	head = vmalloc_to_page(si->swap_map + offset);
3464 	offset &= ~PAGE_MASK;
3465 
3466 	spin_lock(&si->cont_lock);
3467 	/*
3468 	 * Page allocation does not initialize the page's lru field,
3469 	 * but it does always reset its private field.
3470 	 */
3471 	if (!page_private(head)) {
3472 		BUG_ON(count & COUNT_CONTINUED);
3473 		INIT_LIST_HEAD(&head->lru);
3474 		set_page_private(head, SWP_CONTINUED);
3475 		si->flags |= SWP_CONTINUED;
3476 	}
3477 
3478 	list_for_each_entry(list_page, &head->lru, lru) {
3479 		unsigned char *map;
3480 
3481 		/*
3482 		 * If the previous map said no continuation, but we've found
3483 		 * a continuation page, free our allocation and use this one.
3484 		 */
3485 		if (!(count & COUNT_CONTINUED))
3486 			goto out_unlock_cont;
3487 
3488 		map = kmap_atomic(list_page) + offset;
3489 		count = *map;
3490 		kunmap_atomic(map);
3491 
3492 		/*
3493 		 * If this continuation count now has some space in it,
3494 		 * free our allocation and use this one.
3495 		 */
3496 		if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
3497 			goto out_unlock_cont;
3498 	}
3499 
3500 	list_add_tail(&page->lru, &head->lru);
3501 	page = NULL;			/* now it's attached, don't free it */
3502 out_unlock_cont:
3503 	spin_unlock(&si->cont_lock);
3504 out:
3505 	unlock_cluster(ci);
3506 	spin_unlock(&si->lock);
3507 	put_swap_device(si);
3508 outer:
3509 	if (page)
3510 		__free_page(page);
3511 	return ret;
3512 }
3513 
3514 /*
3515  * swap_count_continued - when the original swap_map count is incremented
3516  * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3517  * into, carry if so, or else fail until a new continuation page is allocated;
3518  * when the original swap_map count is decremented from 0 with continuation,
3519  * borrow from the continuation and report whether it still holds more.
3520  * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3521  * lock.
3522  */
3523 static bool swap_count_continued(struct swap_info_struct *si,
3524 				 pgoff_t offset, unsigned char count)
3525 {
3526 	struct page *head;
3527 	struct page *page;
3528 	unsigned char *map;
3529 	bool ret;
3530 
3531 	head = vmalloc_to_page(si->swap_map + offset);
3532 	if (page_private(head) != SWP_CONTINUED) {
3533 		BUG_ON(count & COUNT_CONTINUED);
3534 		return false;		/* need to add count continuation */
3535 	}
3536 
3537 	spin_lock(&si->cont_lock);
3538 	offset &= ~PAGE_MASK;
3539 	page = list_next_entry(head, lru);
3540 	map = kmap_atomic(page) + offset;
3541 
3542 	if (count == SWAP_MAP_MAX)	/* initial increment from swap_map */
3543 		goto init_map;		/* jump over SWAP_CONT_MAX checks */
3544 
3545 	if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
3546 		/*
3547 		 * Think of how you add 1 to 999
3548 		 */
3549 		while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
3550 			kunmap_atomic(map);
3551 			page = list_next_entry(page, lru);
3552 			BUG_ON(page == head);
3553 			map = kmap_atomic(page) + offset;
3554 		}
3555 		if (*map == SWAP_CONT_MAX) {
3556 			kunmap_atomic(map);
3557 			page = list_next_entry(page, lru);
3558 			if (page == head) {
3559 				ret = false;	/* add count continuation */
3560 				goto out;
3561 			}
3562 			map = kmap_atomic(page) + offset;
3563 init_map:		*map = 0;		/* we didn't zero the page */
3564 		}
3565 		*map += 1;
3566 		kunmap_atomic(map);
3567 		while ((page = list_prev_entry(page, lru)) != head) {
3568 			map = kmap_atomic(page) + offset;
3569 			*map = COUNT_CONTINUED;
3570 			kunmap_atomic(map);
3571 		}
3572 		ret = true;			/* incremented */
3573 
3574 	} else {				/* decrementing */
3575 		/*
3576 		 * Think of how you subtract 1 from 1000
3577 		 */
3578 		BUG_ON(count != COUNT_CONTINUED);
3579 		while (*map == COUNT_CONTINUED) {
3580 			kunmap_atomic(map);
3581 			page = list_next_entry(page, lru);
3582 			BUG_ON(page == head);
3583 			map = kmap_atomic(page) + offset;
3584 		}
3585 		BUG_ON(*map == 0);
3586 		*map -= 1;
3587 		if (*map == 0)
3588 			count = 0;
3589 		kunmap_atomic(map);
3590 		while ((page = list_prev_entry(page, lru)) != head) {
3591 			map = kmap_atomic(page) + offset;
3592 			*map = SWAP_CONT_MAX | count;
3593 			count = COUNT_CONTINUED;
3594 			kunmap_atomic(map);
3595 		}
3596 		ret = count == COUNT_CONTINUED;
3597 	}
3598 out:
3599 	spin_unlock(&si->cont_lock);
3600 	return ret;
3601 }
3602 
3603 /*
3604  * free_swap_count_continuations - swapoff free all the continuation pages
3605  * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3606  */
3607 static void free_swap_count_continuations(struct swap_info_struct *si)
3608 {
3609 	pgoff_t offset;
3610 
3611 	for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
3612 		struct page *head;
3613 		head = vmalloc_to_page(si->swap_map + offset);
3614 		if (page_private(head)) {
3615 			struct page *page, *next;
3616 
3617 			list_for_each_entry_safe(page, next, &head->lru, lru) {
3618 				list_del(&page->lru);
3619 				__free_page(page);
3620 			}
3621 		}
3622 	}
3623 }
3624 
3625 #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
3626 void __cgroup_throttle_swaprate(struct page *page, gfp_t gfp_mask)
3627 {
3628 	struct swap_info_struct *si, *next;
3629 	int nid = page_to_nid(page);
3630 
3631 	if (!(gfp_mask & __GFP_IO))
3632 		return;
3633 
3634 	if (!blk_cgroup_congested())
3635 		return;
3636 
3637 	/*
3638 	 * We've already scheduled a throttle, avoid taking the global swap
3639 	 * lock.
3640 	 */
3641 	if (current->throttle_queue)
3642 		return;
3643 
3644 	spin_lock(&swap_avail_lock);
3645 	plist_for_each_entry_safe(si, next, &swap_avail_heads[nid],
3646 				  avail_lists[nid]) {
3647 		if (si->bdev) {
3648 			blkcg_schedule_throttle(bdev_get_queue(si->bdev), true);
3649 			break;
3650 		}
3651 	}
3652 	spin_unlock(&swap_avail_lock);
3653 }
3654 #endif
3655 
3656 static int __init swapfile_init(void)
3657 {
3658 	int nid;
3659 
3660 	swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head),
3661 					 GFP_KERNEL);
3662 	if (!swap_avail_heads) {
3663 		pr_emerg("Not enough memory for swap heads, swap is disabled\n");
3664 		return -ENOMEM;
3665 	}
3666 
3667 	for_each_node(nid)
3668 		plist_head_init(&swap_avail_heads[nid]);
3669 
3670 	swapfile_maximum_size = arch_max_swapfile_size();
3671 
3672 #ifdef CONFIG_MIGRATION
3673 	if (swapfile_maximum_size >= (1UL << SWP_MIG_TOTAL_BITS))
3674 		swap_migration_ad_supported = true;
3675 #endif	/* CONFIG_MIGRATION */
3676 
3677 	return 0;
3678 }
3679 subsys_initcall(swapfile_init);
3680