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 
swap_type_to_swap_info(int type)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 
swap_count(unsigned char ent)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 */
__try_to_reclaim_swap(struct swap_info_struct * si,unsigned long offset,unsigned long flags)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 folio *folio;
136 	int ret = 0;
137 
138 	folio = filemap_get_folio(swap_address_space(entry), offset);
139 	if (!folio)
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 folios. So we hold a folio lock
144 	 * here. We have to use trylock for avoiding deadlock. This is a special
145 	 * case and you should use folio_free_swap() with explicit folio_lock()
146 	 * in usual operations.
147 	 */
148 	if (folio_trylock(folio)) {
149 		if ((flags & TTRS_ANYWAY) ||
150 		    ((flags & TTRS_UNMAPPED) && !folio_mapped(folio)) ||
151 		    ((flags & TTRS_FULL) && mem_cgroup_swap_full(folio)))
152 			ret = folio_free_swap(folio);
153 		folio_unlock(folio);
154 	}
155 	folio_put(folio);
156 	return ret;
157 }
158 
first_se(struct swap_info_struct * sis)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 
next_se(struct swap_extent * se)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  */
discard_swap(struct swap_info_struct * si)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 *
offset_to_swap_extent(struct swap_info_struct * sis,unsigned long offset)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 
swap_page_sector(struct page * page)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  */
discard_swap_cluster(struct swap_info_struct * si,pgoff_t start_page,pgoff_t nr_pages)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 
cluster_set_flag(struct swap_cluster_info * info,unsigned int flag)285 static inline void cluster_set_flag(struct swap_cluster_info *info,
286 	unsigned int flag)
287 {
288 	info->flags = flag;
289 }
290 
cluster_count(struct swap_cluster_info * info)291 static inline unsigned int cluster_count(struct swap_cluster_info *info)
292 {
293 	return info->data;
294 }
295 
cluster_set_count(struct swap_cluster_info * info,unsigned int c)296 static inline void cluster_set_count(struct swap_cluster_info *info,
297 				     unsigned int c)
298 {
299 	info->data = c;
300 }
301 
cluster_set_count_flag(struct swap_cluster_info * info,unsigned int c,unsigned int f)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 
cluster_next(struct swap_cluster_info * info)309 static inline unsigned int cluster_next(struct swap_cluster_info *info)
310 {
311 	return info->data;
312 }
313 
cluster_set_next(struct swap_cluster_info * info,unsigned int n)314 static inline void cluster_set_next(struct swap_cluster_info *info,
315 				    unsigned int n)
316 {
317 	info->data = n;
318 }
319 
cluster_set_next_flag(struct swap_cluster_info * info,unsigned int n,unsigned int f)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 
cluster_is_free(struct swap_cluster_info * info)327 static inline bool cluster_is_free(struct swap_cluster_info *info)
328 {
329 	return info->flags & CLUSTER_FLAG_FREE;
330 }
331 
cluster_is_null(struct swap_cluster_info * info)332 static inline bool cluster_is_null(struct swap_cluster_info *info)
333 {
334 	return info->flags & CLUSTER_FLAG_NEXT_NULL;
335 }
336 
cluster_set_null(struct swap_cluster_info * info)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 
cluster_is_huge(struct swap_cluster_info * info)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 
cluster_clear_huge(struct swap_cluster_info * info)350 static inline void cluster_clear_huge(struct swap_cluster_info *info)
351 {
352 	info->flags &= ~CLUSTER_FLAG_HUGE;
353 }
354 
lock_cluster(struct swap_info_struct * si,unsigned long offset)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 
unlock_cluster(struct swap_cluster_info * ci)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  */
lock_cluster_or_swap_info(struct swap_info_struct * si,unsigned long offset)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 
unlock_cluster_or_swap_info(struct swap_info_struct * si,struct swap_cluster_info * ci)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 
cluster_list_empty(struct swap_cluster_list * list)401 static inline bool cluster_list_empty(struct swap_cluster_list *list)
402 {
403 	return cluster_is_null(&list->head);
404 }
405 
cluster_list_first(struct swap_cluster_list * list)406 static inline unsigned int cluster_list_first(struct swap_cluster_list *list)
407 {
408 	return cluster_next(&list->head);
409 }
410 
cluster_list_init(struct swap_cluster_list * list)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 
cluster_list_add_tail(struct swap_cluster_list * list,struct swap_cluster_info * ci,unsigned int idx)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 
cluster_list_del_first(struct swap_cluster_list * list,struct swap_cluster_info * ci)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 */
swap_cluster_schedule_discard(struct swap_info_struct * si,unsigned int idx)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 
__free_cluster(struct swap_info_struct * si,unsigned long idx)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 */
swap_do_scheduled_discard(struct swap_info_struct * si)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 
swap_discard_work(struct work_struct * work)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 
swap_users_ref_free(struct percpu_ref * ref)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 
alloc_cluster(struct swap_info_struct * si,unsigned long idx)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 
free_cluster(struct swap_info_struct * si,unsigned long idx)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  */
inc_cluster_info_page(struct swap_info_struct * p,struct swap_cluster_info * cluster_info,unsigned long page_nr)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  */
dec_cluster_info_page(struct swap_info_struct * p,struct swap_cluster_info * cluster_info,unsigned long page_nr)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
scan_swap_map_ssd_cluster_conflict(struct swap_info_struct * si,unsigned long offset)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  */
scan_swap_map_try_ssd_cluster(struct swap_info_struct * si,unsigned long * offset,unsigned long * scan_base)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 
__del_from_avail_list(struct swap_info_struct * p)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 
del_from_avail_list(struct swap_info_struct * p)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 
swap_range_alloc(struct swap_info_struct * si,unsigned long offset,unsigned int nr_entries)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 
add_to_avail_list(struct swap_info_struct * p)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 
swap_range_free(struct swap_info_struct * si,unsigned long offset,unsigned int nr_entries)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 
set_cluster_next(struct swap_info_struct * si,unsigned long next)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 
swap_offset_available_and_locked(struct swap_info_struct * si,unsigned long offset)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 
scan_swap_map_slots(struct swap_info_struct * si,unsigned char usage,int nr,swp_entry_t slots[])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 (unlikely(--latency_ration < 0)) {
977 			cond_resched();
978 			latency_ration = LATENCY_LIMIT;
979 			scanned_many = true;
980 		}
981 		if (swap_offset_available_and_locked(si, offset))
982 			goto checks;
983 	}
984 	offset = si->lowest_bit;
985 	while (offset < scan_base) {
986 		if (unlikely(--latency_ration < 0)) {
987 			cond_resched();
988 			latency_ration = LATENCY_LIMIT;
989 			scanned_many = true;
990 		}
991 		if (swap_offset_available_and_locked(si, offset))
992 			goto checks;
993 		offset++;
994 	}
995 	spin_lock(&si->lock);
996 
997 no_page:
998 	si->flags -= SWP_SCANNING;
999 	return n_ret;
1000 }
1001 
swap_alloc_cluster(struct swap_info_struct * si,swp_entry_t * slot)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 
swap_free_cluster(struct swap_info_struct * si,unsigned long idx)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 
get_swap_pages(int n_goal,swp_entry_t swp_entries[],int entry_size)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 
_swap_info_get(swp_entry_t entry)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 
swap_info_get_cont(swp_entry_t entry,struct swap_info_struct * q)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 
__swap_entry_free_locked(struct swap_info_struct * p,unsigned long offset,unsigned char usage)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  */
get_swap_device(swp_entry_t entry)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 
__swap_entry_free(struct swap_info_struct * p,swp_entry_t entry)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 
swap_entry_free(struct swap_info_struct * p,swp_entry_t entry)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  */
swap_free(swp_entry_t entry)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  */
put_swap_folio(struct folio * folio,swp_entry_t entry)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
split_swap_cluster(swp_entry_t entry)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 
swp_entry_cmp(const void * ent1,const void * ent2)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 
swapcache_free_entries(swp_entry_t * entries,int n)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 
__swap_count(swp_entry_t entry)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  */
swap_swapcount(struct swap_info_struct * si,swp_entry_t entry)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  */
__swp_swapcount(swp_entry_t entry)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  */
swp_swapcount(swp_entry_t entry)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 
swap_page_trans_huge_swapped(struct swap_info_struct * si,swp_entry_t entry)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 
folio_swapped(struct folio * folio)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  */
folio_free_swap(struct folio * folio)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  */
free_swap_and_cache(swp_entry_t entry)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 
get_swap_page_of_type(int type)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  */
swap_type_of(dev_t device,sector_t offset)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 
find_first_swap(dev_t * device)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  */
swapdev_block(int type,pgoff_t offset)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  */
count_swap_pages(int type,int free)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 
pte_same_as_swp(pte_t pte,pte_t swp_pte)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  */
unuse_pte(struct vm_area_struct * vma,pmd_t * pmd,unsigned long addr,swp_entry_t entry,struct folio * folio)1760 static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
1761 		unsigned long addr, swp_entry_t entry, struct folio *folio)
1762 {
1763 	struct page *page = folio_file_page(folio, swp_offset(entry));
1764 	struct page *swapcache;
1765 	spinlock_t *ptl;
1766 	pte_t *pte, new_pte;
1767 	int ret = 1;
1768 
1769 	swapcache = page;
1770 	page = ksm_might_need_to_copy(page, vma, addr);
1771 	if (unlikely(!page))
1772 		return -ENOMEM;
1773 
1774 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1775 	if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) {
1776 		ret = 0;
1777 		goto out;
1778 	}
1779 
1780 	if (unlikely(!PageUptodate(page))) {
1781 		pte_t pteval;
1782 
1783 		dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1784 		pteval = swp_entry_to_pte(make_swapin_error_entry(page));
1785 		set_pte_at(vma->vm_mm, addr, pte, pteval);
1786 		swap_free(entry);
1787 		ret = 0;
1788 		goto out;
1789 	}
1790 
1791 	/* See do_swap_page() */
1792 	BUG_ON(!PageAnon(page) && PageMappedToDisk(page));
1793 	BUG_ON(PageAnon(page) && PageAnonExclusive(page));
1794 
1795 	dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1796 	inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
1797 	get_page(page);
1798 	if (page == swapcache) {
1799 		rmap_t rmap_flags = RMAP_NONE;
1800 
1801 		/*
1802 		 * See do_swap_page(): PageWriteback() would be problematic.
1803 		 * However, we do a wait_on_page_writeback() just before this
1804 		 * call and have the page locked.
1805 		 */
1806 		VM_BUG_ON_PAGE(PageWriteback(page), page);
1807 		if (pte_swp_exclusive(*pte))
1808 			rmap_flags |= RMAP_EXCLUSIVE;
1809 
1810 		page_add_anon_rmap(page, vma, addr, rmap_flags);
1811 	} else { /* ksm created a completely new copy */
1812 		page_add_new_anon_rmap(page, vma, addr);
1813 		lru_cache_add_inactive_or_unevictable(page, vma);
1814 	}
1815 	new_pte = pte_mkold(mk_pte(page, vma->vm_page_prot));
1816 	if (pte_swp_soft_dirty(*pte))
1817 		new_pte = pte_mksoft_dirty(new_pte);
1818 	if (pte_swp_uffd_wp(*pte))
1819 		new_pte = pte_mkuffd_wp(new_pte);
1820 	set_pte_at(vma->vm_mm, addr, pte, new_pte);
1821 	swap_free(entry);
1822 out:
1823 	pte_unmap_unlock(pte, ptl);
1824 	if (page != swapcache) {
1825 		unlock_page(page);
1826 		put_page(page);
1827 	}
1828 	return ret;
1829 }
1830 
unuse_pte_range(struct vm_area_struct * vma,pmd_t * pmd,unsigned long addr,unsigned long end,unsigned int type)1831 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1832 			unsigned long addr, unsigned long end,
1833 			unsigned int type)
1834 {
1835 	swp_entry_t entry;
1836 	pte_t *pte;
1837 	struct swap_info_struct *si;
1838 	int ret = 0;
1839 	volatile unsigned char *swap_map;
1840 
1841 	si = swap_info[type];
1842 	pte = pte_offset_map(pmd, addr);
1843 	do {
1844 		struct folio *folio;
1845 		unsigned long offset;
1846 
1847 		if (!is_swap_pte(*pte))
1848 			continue;
1849 
1850 		entry = pte_to_swp_entry(*pte);
1851 		if (swp_type(entry) != type)
1852 			continue;
1853 
1854 		offset = swp_offset(entry);
1855 		pte_unmap(pte);
1856 		swap_map = &si->swap_map[offset];
1857 		folio = swap_cache_get_folio(entry, vma, addr);
1858 		if (!folio) {
1859 			struct page *page;
1860 			struct vm_fault vmf = {
1861 				.vma = vma,
1862 				.address = addr,
1863 				.real_address = addr,
1864 				.pmd = pmd,
1865 			};
1866 
1867 			page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE,
1868 						&vmf);
1869 			if (page)
1870 				folio = page_folio(page);
1871 		}
1872 		if (!folio) {
1873 			if (*swap_map == 0 || *swap_map == SWAP_MAP_BAD)
1874 				goto try_next;
1875 			return -ENOMEM;
1876 		}
1877 
1878 		folio_lock(folio);
1879 		folio_wait_writeback(folio);
1880 		ret = unuse_pte(vma, pmd, addr, entry, folio);
1881 		if (ret < 0) {
1882 			folio_unlock(folio);
1883 			folio_put(folio);
1884 			goto out;
1885 		}
1886 
1887 		folio_free_swap(folio);
1888 		folio_unlock(folio);
1889 		folio_put(folio);
1890 try_next:
1891 		pte = pte_offset_map(pmd, addr);
1892 	} while (pte++, addr += PAGE_SIZE, addr != end);
1893 	pte_unmap(pte - 1);
1894 
1895 	ret = 0;
1896 out:
1897 	return ret;
1898 }
1899 
unuse_pmd_range(struct vm_area_struct * vma,pud_t * pud,unsigned long addr,unsigned long end,unsigned int type)1900 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1901 				unsigned long addr, unsigned long end,
1902 				unsigned int type)
1903 {
1904 	pmd_t *pmd;
1905 	unsigned long next;
1906 	int ret;
1907 
1908 	pmd = pmd_offset(pud, addr);
1909 	do {
1910 		cond_resched();
1911 		next = pmd_addr_end(addr, end);
1912 		if (pmd_none_or_trans_huge_or_clear_bad(pmd))
1913 			continue;
1914 		ret = unuse_pte_range(vma, pmd, addr, next, type);
1915 		if (ret)
1916 			return ret;
1917 	} while (pmd++, addr = next, addr != end);
1918 	return 0;
1919 }
1920 
unuse_pud_range(struct vm_area_struct * vma,p4d_t * p4d,unsigned long addr,unsigned long end,unsigned int type)1921 static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d,
1922 				unsigned long addr, unsigned long end,
1923 				unsigned int type)
1924 {
1925 	pud_t *pud;
1926 	unsigned long next;
1927 	int ret;
1928 
1929 	pud = pud_offset(p4d, addr);
1930 	do {
1931 		next = pud_addr_end(addr, end);
1932 		if (pud_none_or_clear_bad(pud))
1933 			continue;
1934 		ret = unuse_pmd_range(vma, pud, addr, next, type);
1935 		if (ret)
1936 			return ret;
1937 	} while (pud++, addr = next, addr != end);
1938 	return 0;
1939 }
1940 
unuse_p4d_range(struct vm_area_struct * vma,pgd_t * pgd,unsigned long addr,unsigned long end,unsigned int type)1941 static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd,
1942 				unsigned long addr, unsigned long end,
1943 				unsigned int type)
1944 {
1945 	p4d_t *p4d;
1946 	unsigned long next;
1947 	int ret;
1948 
1949 	p4d = p4d_offset(pgd, addr);
1950 	do {
1951 		next = p4d_addr_end(addr, end);
1952 		if (p4d_none_or_clear_bad(p4d))
1953 			continue;
1954 		ret = unuse_pud_range(vma, p4d, addr, next, type);
1955 		if (ret)
1956 			return ret;
1957 	} while (p4d++, addr = next, addr != end);
1958 	return 0;
1959 }
1960 
unuse_vma(struct vm_area_struct * vma,unsigned int type)1961 static int unuse_vma(struct vm_area_struct *vma, unsigned int type)
1962 {
1963 	pgd_t *pgd;
1964 	unsigned long addr, end, next;
1965 	int ret;
1966 
1967 	addr = vma->vm_start;
1968 	end = vma->vm_end;
1969 
1970 	pgd = pgd_offset(vma->vm_mm, addr);
1971 	do {
1972 		next = pgd_addr_end(addr, end);
1973 		if (pgd_none_or_clear_bad(pgd))
1974 			continue;
1975 		ret = unuse_p4d_range(vma, pgd, addr, next, type);
1976 		if (ret)
1977 			return ret;
1978 	} while (pgd++, addr = next, addr != end);
1979 	return 0;
1980 }
1981 
unuse_mm(struct mm_struct * mm,unsigned int type)1982 static int unuse_mm(struct mm_struct *mm, unsigned int type)
1983 {
1984 	struct vm_area_struct *vma;
1985 	int ret = 0;
1986 	VMA_ITERATOR(vmi, mm, 0);
1987 
1988 	mmap_read_lock(mm);
1989 	for_each_vma(vmi, vma) {
1990 		if (vma->anon_vma) {
1991 			ret = unuse_vma(vma, type);
1992 			if (ret)
1993 				break;
1994 		}
1995 
1996 		cond_resched();
1997 	}
1998 	mmap_read_unlock(mm);
1999 	return ret;
2000 }
2001 
2002 /*
2003  * Scan swap_map from current position to next entry still in use.
2004  * Return 0 if there are no inuse entries after prev till end of
2005  * the map.
2006  */
find_next_to_unuse(struct swap_info_struct * si,unsigned int prev)2007 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
2008 					unsigned int prev)
2009 {
2010 	unsigned int i;
2011 	unsigned char count;
2012 
2013 	/*
2014 	 * No need for swap_lock here: we're just looking
2015 	 * for whether an entry is in use, not modifying it; false
2016 	 * hits are okay, and sys_swapoff() has already prevented new
2017 	 * allocations from this area (while holding swap_lock).
2018 	 */
2019 	for (i = prev + 1; i < si->max; i++) {
2020 		count = READ_ONCE(si->swap_map[i]);
2021 		if (count && swap_count(count) != SWAP_MAP_BAD)
2022 			break;
2023 		if ((i % LATENCY_LIMIT) == 0)
2024 			cond_resched();
2025 	}
2026 
2027 	if (i == si->max)
2028 		i = 0;
2029 
2030 	return i;
2031 }
2032 
try_to_unuse(unsigned int type)2033 static int try_to_unuse(unsigned int type)
2034 {
2035 	struct mm_struct *prev_mm;
2036 	struct mm_struct *mm;
2037 	struct list_head *p;
2038 	int retval = 0;
2039 	struct swap_info_struct *si = swap_info[type];
2040 	struct folio *folio;
2041 	swp_entry_t entry;
2042 	unsigned int i;
2043 
2044 	if (!READ_ONCE(si->inuse_pages))
2045 		return 0;
2046 
2047 retry:
2048 	retval = shmem_unuse(type);
2049 	if (retval)
2050 		return retval;
2051 
2052 	prev_mm = &init_mm;
2053 	mmget(prev_mm);
2054 
2055 	spin_lock(&mmlist_lock);
2056 	p = &init_mm.mmlist;
2057 	while (READ_ONCE(si->inuse_pages) &&
2058 	       !signal_pending(current) &&
2059 	       (p = p->next) != &init_mm.mmlist) {
2060 
2061 		mm = list_entry(p, struct mm_struct, mmlist);
2062 		if (!mmget_not_zero(mm))
2063 			continue;
2064 		spin_unlock(&mmlist_lock);
2065 		mmput(prev_mm);
2066 		prev_mm = mm;
2067 		retval = unuse_mm(mm, type);
2068 		if (retval) {
2069 			mmput(prev_mm);
2070 			return retval;
2071 		}
2072 
2073 		/*
2074 		 * Make sure that we aren't completely killing
2075 		 * interactive performance.
2076 		 */
2077 		cond_resched();
2078 		spin_lock(&mmlist_lock);
2079 	}
2080 	spin_unlock(&mmlist_lock);
2081 
2082 	mmput(prev_mm);
2083 
2084 	i = 0;
2085 	while (READ_ONCE(si->inuse_pages) &&
2086 	       !signal_pending(current) &&
2087 	       (i = find_next_to_unuse(si, i)) != 0) {
2088 
2089 		entry = swp_entry(type, i);
2090 		folio = filemap_get_folio(swap_address_space(entry), i);
2091 		if (!folio)
2092 			continue;
2093 
2094 		/*
2095 		 * It is conceivable that a racing task removed this folio from
2096 		 * swap cache just before we acquired the page lock. The folio
2097 		 * might even be back in swap cache on another swap area. But
2098 		 * that is okay, folio_free_swap() only removes stale folios.
2099 		 */
2100 		folio_lock(folio);
2101 		folio_wait_writeback(folio);
2102 		folio_free_swap(folio);
2103 		folio_unlock(folio);
2104 		folio_put(folio);
2105 	}
2106 
2107 	/*
2108 	 * Lets check again to see if there are still swap entries in the map.
2109 	 * If yes, we would need to do retry the unuse logic again.
2110 	 * Under global memory pressure, swap entries can be reinserted back
2111 	 * into process space after the mmlist loop above passes over them.
2112 	 *
2113 	 * Limit the number of retries? No: when mmget_not_zero()
2114 	 * above fails, that mm is likely to be freeing swap from
2115 	 * exit_mmap(), which proceeds at its own independent pace;
2116 	 * and even shmem_writepage() could have been preempted after
2117 	 * folio_alloc_swap(), temporarily hiding that swap.  It's easy
2118 	 * and robust (though cpu-intensive) just to keep retrying.
2119 	 */
2120 	if (READ_ONCE(si->inuse_pages)) {
2121 		if (!signal_pending(current))
2122 			goto retry;
2123 		return -EINTR;
2124 	}
2125 
2126 	return 0;
2127 }
2128 
2129 /*
2130  * After a successful try_to_unuse, if no swap is now in use, we know
2131  * we can empty the mmlist.  swap_lock must be held on entry and exit.
2132  * Note that mmlist_lock nests inside swap_lock, and an mm must be
2133  * added to the mmlist just after page_duplicate - before would be racy.
2134  */
drain_mmlist(void)2135 static void drain_mmlist(void)
2136 {
2137 	struct list_head *p, *next;
2138 	unsigned int type;
2139 
2140 	for (type = 0; type < nr_swapfiles; type++)
2141 		if (swap_info[type]->inuse_pages)
2142 			return;
2143 	spin_lock(&mmlist_lock);
2144 	list_for_each_safe(p, next, &init_mm.mmlist)
2145 		list_del_init(p);
2146 	spin_unlock(&mmlist_lock);
2147 }
2148 
2149 /*
2150  * Free all of a swapdev's extent information
2151  */
destroy_swap_extents(struct swap_info_struct * sis)2152 static void destroy_swap_extents(struct swap_info_struct *sis)
2153 {
2154 	while (!RB_EMPTY_ROOT(&sis->swap_extent_root)) {
2155 		struct rb_node *rb = sis->swap_extent_root.rb_node;
2156 		struct swap_extent *se = rb_entry(rb, struct swap_extent, rb_node);
2157 
2158 		rb_erase(rb, &sis->swap_extent_root);
2159 		kfree(se);
2160 	}
2161 
2162 	if (sis->flags & SWP_ACTIVATED) {
2163 		struct file *swap_file = sis->swap_file;
2164 		struct address_space *mapping = swap_file->f_mapping;
2165 
2166 		sis->flags &= ~SWP_ACTIVATED;
2167 		if (mapping->a_ops->swap_deactivate)
2168 			mapping->a_ops->swap_deactivate(swap_file);
2169 	}
2170 }
2171 
2172 /*
2173  * Add a block range (and the corresponding page range) into this swapdev's
2174  * extent tree.
2175  *
2176  * This function rather assumes that it is called in ascending page order.
2177  */
2178 int
add_swap_extent(struct swap_info_struct * sis,unsigned long start_page,unsigned long nr_pages,sector_t start_block)2179 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
2180 		unsigned long nr_pages, sector_t start_block)
2181 {
2182 	struct rb_node **link = &sis->swap_extent_root.rb_node, *parent = NULL;
2183 	struct swap_extent *se;
2184 	struct swap_extent *new_se;
2185 
2186 	/*
2187 	 * place the new node at the right most since the
2188 	 * function is called in ascending page order.
2189 	 */
2190 	while (*link) {
2191 		parent = *link;
2192 		link = &parent->rb_right;
2193 	}
2194 
2195 	if (parent) {
2196 		se = rb_entry(parent, struct swap_extent, rb_node);
2197 		BUG_ON(se->start_page + se->nr_pages != start_page);
2198 		if (se->start_block + se->nr_pages == start_block) {
2199 			/* Merge it */
2200 			se->nr_pages += nr_pages;
2201 			return 0;
2202 		}
2203 	}
2204 
2205 	/* No merge, insert a new extent. */
2206 	new_se = kmalloc(sizeof(*se), GFP_KERNEL);
2207 	if (new_se == NULL)
2208 		return -ENOMEM;
2209 	new_se->start_page = start_page;
2210 	new_se->nr_pages = nr_pages;
2211 	new_se->start_block = start_block;
2212 
2213 	rb_link_node(&new_se->rb_node, parent, link);
2214 	rb_insert_color(&new_se->rb_node, &sis->swap_extent_root);
2215 	return 1;
2216 }
2217 EXPORT_SYMBOL_GPL(add_swap_extent);
2218 
2219 /*
2220  * A `swap extent' is a simple thing which maps a contiguous range of pages
2221  * onto a contiguous range of disk blocks.  A rbtree of swap extents is
2222  * built at swapon time and is then used at swap_writepage/swap_readpage
2223  * time for locating where on disk a page belongs.
2224  *
2225  * If the swapfile is an S_ISBLK block device, a single extent is installed.
2226  * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2227  * swap files identically.
2228  *
2229  * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2230  * extent rbtree operates in PAGE_SIZE disk blocks.  Both S_ISREG and S_ISBLK
2231  * swapfiles are handled *identically* after swapon time.
2232  *
2233  * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2234  * and will parse them into a rbtree, in PAGE_SIZE chunks.  If some stray
2235  * blocks are found which do not fall within the PAGE_SIZE alignment
2236  * requirements, they are simply tossed out - we will never use those blocks
2237  * for swapping.
2238  *
2239  * For all swap devices we set S_SWAPFILE across the life of the swapon.  This
2240  * prevents users from writing to the swap device, which will corrupt memory.
2241  *
2242  * The amount of disk space which a single swap extent represents varies.
2243  * Typically it is in the 1-4 megabyte range.  So we can have hundreds of
2244  * extents in the rbtree. - akpm.
2245  */
setup_swap_extents(struct swap_info_struct * sis,sector_t * span)2246 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
2247 {
2248 	struct file *swap_file = sis->swap_file;
2249 	struct address_space *mapping = swap_file->f_mapping;
2250 	struct inode *inode = mapping->host;
2251 	int ret;
2252 
2253 	if (S_ISBLK(inode->i_mode)) {
2254 		ret = add_swap_extent(sis, 0, sis->max, 0);
2255 		*span = sis->pages;
2256 		return ret;
2257 	}
2258 
2259 	if (mapping->a_ops->swap_activate) {
2260 		ret = mapping->a_ops->swap_activate(sis, swap_file, span);
2261 		if (ret < 0)
2262 			return ret;
2263 		sis->flags |= SWP_ACTIVATED;
2264 		if ((sis->flags & SWP_FS_OPS) &&
2265 		    sio_pool_init() != 0) {
2266 			destroy_swap_extents(sis);
2267 			return -ENOMEM;
2268 		}
2269 		return ret;
2270 	}
2271 
2272 	return generic_swapfile_activate(sis, swap_file, span);
2273 }
2274 
swap_node(struct swap_info_struct * p)2275 static int swap_node(struct swap_info_struct *p)
2276 {
2277 	struct block_device *bdev;
2278 
2279 	if (p->bdev)
2280 		bdev = p->bdev;
2281 	else
2282 		bdev = p->swap_file->f_inode->i_sb->s_bdev;
2283 
2284 	return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE;
2285 }
2286 
setup_swap_info(struct swap_info_struct * p,int prio,unsigned char * swap_map,struct swap_cluster_info * cluster_info)2287 static void setup_swap_info(struct swap_info_struct *p, int prio,
2288 			    unsigned char *swap_map,
2289 			    struct swap_cluster_info *cluster_info)
2290 {
2291 	int i;
2292 
2293 	if (prio >= 0)
2294 		p->prio = prio;
2295 	else
2296 		p->prio = --least_priority;
2297 	/*
2298 	 * the plist prio is negated because plist ordering is
2299 	 * low-to-high, while swap ordering is high-to-low
2300 	 */
2301 	p->list.prio = -p->prio;
2302 	for_each_node(i) {
2303 		if (p->prio >= 0)
2304 			p->avail_lists[i].prio = -p->prio;
2305 		else {
2306 			if (swap_node(p) == i)
2307 				p->avail_lists[i].prio = 1;
2308 			else
2309 				p->avail_lists[i].prio = -p->prio;
2310 		}
2311 	}
2312 	p->swap_map = swap_map;
2313 	p->cluster_info = cluster_info;
2314 }
2315 
_enable_swap_info(struct swap_info_struct * p)2316 static void _enable_swap_info(struct swap_info_struct *p)
2317 {
2318 	p->flags |= SWP_WRITEOK;
2319 	atomic_long_add(p->pages, &nr_swap_pages);
2320 	total_swap_pages += p->pages;
2321 
2322 	assert_spin_locked(&swap_lock);
2323 	/*
2324 	 * both lists are plists, and thus priority ordered.
2325 	 * swap_active_head needs to be priority ordered for swapoff(),
2326 	 * which on removal of any swap_info_struct with an auto-assigned
2327 	 * (i.e. negative) priority increments the auto-assigned priority
2328 	 * of any lower-priority swap_info_structs.
2329 	 * swap_avail_head needs to be priority ordered for folio_alloc_swap(),
2330 	 * which allocates swap pages from the highest available priority
2331 	 * swap_info_struct.
2332 	 */
2333 	plist_add(&p->list, &swap_active_head);
2334 	add_to_avail_list(p);
2335 }
2336 
enable_swap_info(struct swap_info_struct * p,int prio,unsigned char * swap_map,struct swap_cluster_info * cluster_info,unsigned long * frontswap_map)2337 static void enable_swap_info(struct swap_info_struct *p, int prio,
2338 				unsigned char *swap_map,
2339 				struct swap_cluster_info *cluster_info,
2340 				unsigned long *frontswap_map)
2341 {
2342 	if (IS_ENABLED(CONFIG_FRONTSWAP))
2343 		frontswap_init(p->type, frontswap_map);
2344 	spin_lock(&swap_lock);
2345 	spin_lock(&p->lock);
2346 	setup_swap_info(p, prio, swap_map, cluster_info);
2347 	spin_unlock(&p->lock);
2348 	spin_unlock(&swap_lock);
2349 	/*
2350 	 * Finished initializing swap device, now it's safe to reference it.
2351 	 */
2352 	percpu_ref_resurrect(&p->users);
2353 	spin_lock(&swap_lock);
2354 	spin_lock(&p->lock);
2355 	_enable_swap_info(p);
2356 	spin_unlock(&p->lock);
2357 	spin_unlock(&swap_lock);
2358 }
2359 
reinsert_swap_info(struct swap_info_struct * p)2360 static void reinsert_swap_info(struct swap_info_struct *p)
2361 {
2362 	spin_lock(&swap_lock);
2363 	spin_lock(&p->lock);
2364 	setup_swap_info(p, p->prio, p->swap_map, p->cluster_info);
2365 	_enable_swap_info(p);
2366 	spin_unlock(&p->lock);
2367 	spin_unlock(&swap_lock);
2368 }
2369 
has_usable_swap(void)2370 bool has_usable_swap(void)
2371 {
2372 	bool ret = true;
2373 
2374 	spin_lock(&swap_lock);
2375 	if (plist_head_empty(&swap_active_head))
2376 		ret = false;
2377 	spin_unlock(&swap_lock);
2378 	return ret;
2379 }
2380 
SYSCALL_DEFINE1(swapoff,const char __user *,specialfile)2381 SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
2382 {
2383 	struct swap_info_struct *p = NULL;
2384 	unsigned char *swap_map;
2385 	struct swap_cluster_info *cluster_info;
2386 	unsigned long *frontswap_map;
2387 	struct file *swap_file, *victim;
2388 	struct address_space *mapping;
2389 	struct inode *inode;
2390 	struct filename *pathname;
2391 	int err, found = 0;
2392 	unsigned int old_block_size;
2393 
2394 	if (!capable(CAP_SYS_ADMIN))
2395 		return -EPERM;
2396 
2397 	BUG_ON(!current->mm);
2398 
2399 	pathname = getname(specialfile);
2400 	if (IS_ERR(pathname))
2401 		return PTR_ERR(pathname);
2402 
2403 	victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
2404 	err = PTR_ERR(victim);
2405 	if (IS_ERR(victim))
2406 		goto out;
2407 
2408 	mapping = victim->f_mapping;
2409 	spin_lock(&swap_lock);
2410 	plist_for_each_entry(p, &swap_active_head, list) {
2411 		if (p->flags & SWP_WRITEOK) {
2412 			if (p->swap_file->f_mapping == mapping) {
2413 				found = 1;
2414 				break;
2415 			}
2416 		}
2417 	}
2418 	if (!found) {
2419 		err = -EINVAL;
2420 		spin_unlock(&swap_lock);
2421 		goto out_dput;
2422 	}
2423 	if (!security_vm_enough_memory_mm(current->mm, p->pages))
2424 		vm_unacct_memory(p->pages);
2425 	else {
2426 		err = -ENOMEM;
2427 		spin_unlock(&swap_lock);
2428 		goto out_dput;
2429 	}
2430 	del_from_avail_list(p);
2431 	spin_lock(&p->lock);
2432 	if (p->prio < 0) {
2433 		struct swap_info_struct *si = p;
2434 		int nid;
2435 
2436 		plist_for_each_entry_continue(si, &swap_active_head, list) {
2437 			si->prio++;
2438 			si->list.prio--;
2439 			for_each_node(nid) {
2440 				if (si->avail_lists[nid].prio != 1)
2441 					si->avail_lists[nid].prio--;
2442 			}
2443 		}
2444 		least_priority++;
2445 	}
2446 	plist_del(&p->list, &swap_active_head);
2447 	atomic_long_sub(p->pages, &nr_swap_pages);
2448 	total_swap_pages -= p->pages;
2449 	p->flags &= ~SWP_WRITEOK;
2450 	spin_unlock(&p->lock);
2451 	spin_unlock(&swap_lock);
2452 
2453 	disable_swap_slots_cache_lock();
2454 
2455 	set_current_oom_origin();
2456 	err = try_to_unuse(p->type);
2457 	clear_current_oom_origin();
2458 
2459 	if (err) {
2460 		/* re-insert swap space back into swap_list */
2461 		reinsert_swap_info(p);
2462 		reenable_swap_slots_cache_unlock();
2463 		goto out_dput;
2464 	}
2465 
2466 	reenable_swap_slots_cache_unlock();
2467 
2468 	/*
2469 	 * Wait for swap operations protected by get/put_swap_device()
2470 	 * to complete.
2471 	 *
2472 	 * We need synchronize_rcu() here to protect the accessing to
2473 	 * the swap cache data structure.
2474 	 */
2475 	percpu_ref_kill(&p->users);
2476 	synchronize_rcu();
2477 	wait_for_completion(&p->comp);
2478 
2479 	flush_work(&p->discard_work);
2480 
2481 	destroy_swap_extents(p);
2482 	if (p->flags & SWP_CONTINUED)
2483 		free_swap_count_continuations(p);
2484 
2485 	if (!p->bdev || !bdev_nonrot(p->bdev))
2486 		atomic_dec(&nr_rotate_swap);
2487 
2488 	mutex_lock(&swapon_mutex);
2489 	spin_lock(&swap_lock);
2490 	spin_lock(&p->lock);
2491 	drain_mmlist();
2492 
2493 	/* wait for anyone still in scan_swap_map_slots */
2494 	p->highest_bit = 0;		/* cuts scans short */
2495 	while (p->flags >= SWP_SCANNING) {
2496 		spin_unlock(&p->lock);
2497 		spin_unlock(&swap_lock);
2498 		schedule_timeout_uninterruptible(1);
2499 		spin_lock(&swap_lock);
2500 		spin_lock(&p->lock);
2501 	}
2502 
2503 	swap_file = p->swap_file;
2504 	old_block_size = p->old_block_size;
2505 	p->swap_file = NULL;
2506 	p->max = 0;
2507 	swap_map = p->swap_map;
2508 	p->swap_map = NULL;
2509 	cluster_info = p->cluster_info;
2510 	p->cluster_info = NULL;
2511 	frontswap_map = frontswap_map_get(p);
2512 	spin_unlock(&p->lock);
2513 	spin_unlock(&swap_lock);
2514 	arch_swap_invalidate_area(p->type);
2515 	frontswap_invalidate_area(p->type);
2516 	frontswap_map_set(p, NULL);
2517 	mutex_unlock(&swapon_mutex);
2518 	free_percpu(p->percpu_cluster);
2519 	p->percpu_cluster = NULL;
2520 	free_percpu(p->cluster_next_cpu);
2521 	p->cluster_next_cpu = NULL;
2522 	vfree(swap_map);
2523 	kvfree(cluster_info);
2524 	kvfree(frontswap_map);
2525 	/* Destroy swap account information */
2526 	swap_cgroup_swapoff(p->type);
2527 	exit_swap_address_space(p->type);
2528 
2529 	inode = mapping->host;
2530 	if (S_ISBLK(inode->i_mode)) {
2531 		struct block_device *bdev = I_BDEV(inode);
2532 
2533 		set_blocksize(bdev, old_block_size);
2534 		blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2535 	}
2536 
2537 	inode_lock(inode);
2538 	inode->i_flags &= ~S_SWAPFILE;
2539 	inode_unlock(inode);
2540 	filp_close(swap_file, NULL);
2541 
2542 	/*
2543 	 * Clear the SWP_USED flag after all resources are freed so that swapon
2544 	 * can reuse this swap_info in alloc_swap_info() safely.  It is ok to
2545 	 * not hold p->lock after we cleared its SWP_WRITEOK.
2546 	 */
2547 	spin_lock(&swap_lock);
2548 	p->flags = 0;
2549 	spin_unlock(&swap_lock);
2550 
2551 	err = 0;
2552 	atomic_inc(&proc_poll_event);
2553 	wake_up_interruptible(&proc_poll_wait);
2554 
2555 out_dput:
2556 	filp_close(victim, NULL);
2557 out:
2558 	putname(pathname);
2559 	return err;
2560 }
2561 
2562 #ifdef CONFIG_PROC_FS
swaps_poll(struct file * file,poll_table * wait)2563 static __poll_t swaps_poll(struct file *file, poll_table *wait)
2564 {
2565 	struct seq_file *seq = file->private_data;
2566 
2567 	poll_wait(file, &proc_poll_wait, wait);
2568 
2569 	if (seq->poll_event != atomic_read(&proc_poll_event)) {
2570 		seq->poll_event = atomic_read(&proc_poll_event);
2571 		return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
2572 	}
2573 
2574 	return EPOLLIN | EPOLLRDNORM;
2575 }
2576 
2577 /* iterator */
swap_start(struct seq_file * swap,loff_t * pos)2578 static void *swap_start(struct seq_file *swap, loff_t *pos)
2579 {
2580 	struct swap_info_struct *si;
2581 	int type;
2582 	loff_t l = *pos;
2583 
2584 	mutex_lock(&swapon_mutex);
2585 
2586 	if (!l)
2587 		return SEQ_START_TOKEN;
2588 
2589 	for (type = 0; (si = swap_type_to_swap_info(type)); type++) {
2590 		if (!(si->flags & SWP_USED) || !si->swap_map)
2591 			continue;
2592 		if (!--l)
2593 			return si;
2594 	}
2595 
2596 	return NULL;
2597 }
2598 
swap_next(struct seq_file * swap,void * v,loff_t * pos)2599 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2600 {
2601 	struct swap_info_struct *si = v;
2602 	int type;
2603 
2604 	if (v == SEQ_START_TOKEN)
2605 		type = 0;
2606 	else
2607 		type = si->type + 1;
2608 
2609 	++(*pos);
2610 	for (; (si = swap_type_to_swap_info(type)); type++) {
2611 		if (!(si->flags & SWP_USED) || !si->swap_map)
2612 			continue;
2613 		return si;
2614 	}
2615 
2616 	return NULL;
2617 }
2618 
swap_stop(struct seq_file * swap,void * v)2619 static void swap_stop(struct seq_file *swap, void *v)
2620 {
2621 	mutex_unlock(&swapon_mutex);
2622 }
2623 
swap_show(struct seq_file * swap,void * v)2624 static int swap_show(struct seq_file *swap, void *v)
2625 {
2626 	struct swap_info_struct *si = v;
2627 	struct file *file;
2628 	int len;
2629 	unsigned long bytes, inuse;
2630 
2631 	if (si == SEQ_START_TOKEN) {
2632 		seq_puts(swap, "Filename\t\t\t\tType\t\tSize\t\tUsed\t\tPriority\n");
2633 		return 0;
2634 	}
2635 
2636 	bytes = si->pages << (PAGE_SHIFT - 10);
2637 	inuse = READ_ONCE(si->inuse_pages) << (PAGE_SHIFT - 10);
2638 
2639 	file = si->swap_file;
2640 	len = seq_file_path(swap, file, " \t\n\\");
2641 	seq_printf(swap, "%*s%s\t%lu\t%s%lu\t%s%d\n",
2642 			len < 40 ? 40 - len : 1, " ",
2643 			S_ISBLK(file_inode(file)->i_mode) ?
2644 				"partition" : "file\t",
2645 			bytes, bytes < 10000000 ? "\t" : "",
2646 			inuse, inuse < 10000000 ? "\t" : "",
2647 			si->prio);
2648 	return 0;
2649 }
2650 
2651 static const struct seq_operations swaps_op = {
2652 	.start =	swap_start,
2653 	.next =		swap_next,
2654 	.stop =		swap_stop,
2655 	.show =		swap_show
2656 };
2657 
swaps_open(struct inode * inode,struct file * file)2658 static int swaps_open(struct inode *inode, struct file *file)
2659 {
2660 	struct seq_file *seq;
2661 	int ret;
2662 
2663 	ret = seq_open(file, &swaps_op);
2664 	if (ret)
2665 		return ret;
2666 
2667 	seq = file->private_data;
2668 	seq->poll_event = atomic_read(&proc_poll_event);
2669 	return 0;
2670 }
2671 
2672 static const struct proc_ops swaps_proc_ops = {
2673 	.proc_flags	= PROC_ENTRY_PERMANENT,
2674 	.proc_open	= swaps_open,
2675 	.proc_read	= seq_read,
2676 	.proc_lseek	= seq_lseek,
2677 	.proc_release	= seq_release,
2678 	.proc_poll	= swaps_poll,
2679 };
2680 
procswaps_init(void)2681 static int __init procswaps_init(void)
2682 {
2683 	proc_create("swaps", 0, NULL, &swaps_proc_ops);
2684 	return 0;
2685 }
2686 __initcall(procswaps_init);
2687 #endif /* CONFIG_PROC_FS */
2688 
2689 #ifdef MAX_SWAPFILES_CHECK
max_swapfiles_check(void)2690 static int __init max_swapfiles_check(void)
2691 {
2692 	MAX_SWAPFILES_CHECK();
2693 	return 0;
2694 }
2695 late_initcall(max_swapfiles_check);
2696 #endif
2697 
alloc_swap_info(void)2698 static struct swap_info_struct *alloc_swap_info(void)
2699 {
2700 	struct swap_info_struct *p;
2701 	struct swap_info_struct *defer = NULL;
2702 	unsigned int type;
2703 	int i;
2704 
2705 	p = kvzalloc(struct_size(p, avail_lists, nr_node_ids), GFP_KERNEL);
2706 	if (!p)
2707 		return ERR_PTR(-ENOMEM);
2708 
2709 	if (percpu_ref_init(&p->users, swap_users_ref_free,
2710 			    PERCPU_REF_INIT_DEAD, GFP_KERNEL)) {
2711 		kvfree(p);
2712 		return ERR_PTR(-ENOMEM);
2713 	}
2714 
2715 	spin_lock(&swap_lock);
2716 	for (type = 0; type < nr_swapfiles; type++) {
2717 		if (!(swap_info[type]->flags & SWP_USED))
2718 			break;
2719 	}
2720 	if (type >= MAX_SWAPFILES) {
2721 		spin_unlock(&swap_lock);
2722 		percpu_ref_exit(&p->users);
2723 		kvfree(p);
2724 		return ERR_PTR(-EPERM);
2725 	}
2726 	if (type >= nr_swapfiles) {
2727 		p->type = type;
2728 		/*
2729 		 * Publish the swap_info_struct after initializing it.
2730 		 * Note that kvzalloc() above zeroes all its fields.
2731 		 */
2732 		smp_store_release(&swap_info[type], p); /* rcu_assign_pointer() */
2733 		nr_swapfiles++;
2734 	} else {
2735 		defer = p;
2736 		p = swap_info[type];
2737 		/*
2738 		 * Do not memset this entry: a racing procfs swap_next()
2739 		 * would be relying on p->type to remain valid.
2740 		 */
2741 	}
2742 	p->swap_extent_root = RB_ROOT;
2743 	plist_node_init(&p->list, 0);
2744 	for_each_node(i)
2745 		plist_node_init(&p->avail_lists[i], 0);
2746 	p->flags = SWP_USED;
2747 	spin_unlock(&swap_lock);
2748 	if (defer) {
2749 		percpu_ref_exit(&defer->users);
2750 		kvfree(defer);
2751 	}
2752 	spin_lock_init(&p->lock);
2753 	spin_lock_init(&p->cont_lock);
2754 	init_completion(&p->comp);
2755 
2756 	return p;
2757 }
2758 
claim_swapfile(struct swap_info_struct * p,struct inode * inode)2759 static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2760 {
2761 	int error;
2762 
2763 	if (S_ISBLK(inode->i_mode)) {
2764 		p->bdev = blkdev_get_by_dev(inode->i_rdev,
2765 				   FMODE_READ | FMODE_WRITE | FMODE_EXCL, p);
2766 		if (IS_ERR(p->bdev)) {
2767 			error = PTR_ERR(p->bdev);
2768 			p->bdev = NULL;
2769 			return error;
2770 		}
2771 		p->old_block_size = block_size(p->bdev);
2772 		error = set_blocksize(p->bdev, PAGE_SIZE);
2773 		if (error < 0)
2774 			return error;
2775 		/*
2776 		 * Zoned block devices contain zones that have a sequential
2777 		 * write only restriction.  Hence zoned block devices are not
2778 		 * suitable for swapping.  Disallow them here.
2779 		 */
2780 		if (bdev_is_zoned(p->bdev))
2781 			return -EINVAL;
2782 		p->flags |= SWP_BLKDEV;
2783 	} else if (S_ISREG(inode->i_mode)) {
2784 		p->bdev = inode->i_sb->s_bdev;
2785 	}
2786 
2787 	return 0;
2788 }
2789 
2790 
2791 /*
2792  * Find out how many pages are allowed for a single swap device. There
2793  * are two limiting factors:
2794  * 1) the number of bits for the swap offset in the swp_entry_t type, and
2795  * 2) the number of bits in the swap pte, as defined by the different
2796  * architectures.
2797  *
2798  * In order to find the largest possible bit mask, a swap entry with
2799  * swap type 0 and swap offset ~0UL is created, encoded to a swap pte,
2800  * decoded to a swp_entry_t again, and finally the swap offset is
2801  * extracted.
2802  *
2803  * This will mask all the bits from the initial ~0UL mask that can't
2804  * be encoded in either the swp_entry_t or the architecture definition
2805  * of a swap pte.
2806  */
generic_max_swapfile_size(void)2807 unsigned long generic_max_swapfile_size(void)
2808 {
2809 	return swp_offset(pte_to_swp_entry(
2810 			swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2811 }
2812 
2813 /* Can be overridden by an architecture for additional checks. */
arch_max_swapfile_size(void)2814 __weak unsigned long arch_max_swapfile_size(void)
2815 {
2816 	return generic_max_swapfile_size();
2817 }
2818 
read_swap_header(struct swap_info_struct * p,union swap_header * swap_header,struct inode * inode)2819 static unsigned long read_swap_header(struct swap_info_struct *p,
2820 					union swap_header *swap_header,
2821 					struct inode *inode)
2822 {
2823 	int i;
2824 	unsigned long maxpages;
2825 	unsigned long swapfilepages;
2826 	unsigned long last_page;
2827 
2828 	if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
2829 		pr_err("Unable to find swap-space signature\n");
2830 		return 0;
2831 	}
2832 
2833 	/* swap partition endianness hack... */
2834 	if (swab32(swap_header->info.version) == 1) {
2835 		swab32s(&swap_header->info.version);
2836 		swab32s(&swap_header->info.last_page);
2837 		swab32s(&swap_header->info.nr_badpages);
2838 		if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2839 			return 0;
2840 		for (i = 0; i < swap_header->info.nr_badpages; i++)
2841 			swab32s(&swap_header->info.badpages[i]);
2842 	}
2843 	/* Check the swap header's sub-version */
2844 	if (swap_header->info.version != 1) {
2845 		pr_warn("Unable to handle swap header version %d\n",
2846 			swap_header->info.version);
2847 		return 0;
2848 	}
2849 
2850 	p->lowest_bit  = 1;
2851 	p->cluster_next = 1;
2852 	p->cluster_nr = 0;
2853 
2854 	maxpages = swapfile_maximum_size;
2855 	last_page = swap_header->info.last_page;
2856 	if (!last_page) {
2857 		pr_warn("Empty swap-file\n");
2858 		return 0;
2859 	}
2860 	if (last_page > maxpages) {
2861 		pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
2862 			maxpages << (PAGE_SHIFT - 10),
2863 			last_page << (PAGE_SHIFT - 10));
2864 	}
2865 	if (maxpages > last_page) {
2866 		maxpages = last_page + 1;
2867 		/* p->max is an unsigned int: don't overflow it */
2868 		if ((unsigned int)maxpages == 0)
2869 			maxpages = UINT_MAX;
2870 	}
2871 	p->highest_bit = maxpages - 1;
2872 
2873 	if (!maxpages)
2874 		return 0;
2875 	swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2876 	if (swapfilepages && maxpages > swapfilepages) {
2877 		pr_warn("Swap area shorter than signature indicates\n");
2878 		return 0;
2879 	}
2880 	if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
2881 		return 0;
2882 	if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2883 		return 0;
2884 
2885 	return maxpages;
2886 }
2887 
2888 #define SWAP_CLUSTER_INFO_COLS						\
2889 	DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
2890 #define SWAP_CLUSTER_SPACE_COLS						\
2891 	DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
2892 #define SWAP_CLUSTER_COLS						\
2893 	max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
2894 
setup_swap_map_and_extents(struct swap_info_struct * p,union swap_header * swap_header,unsigned char * swap_map,struct swap_cluster_info * cluster_info,unsigned long maxpages,sector_t * span)2895 static int setup_swap_map_and_extents(struct swap_info_struct *p,
2896 					union swap_header *swap_header,
2897 					unsigned char *swap_map,
2898 					struct swap_cluster_info *cluster_info,
2899 					unsigned long maxpages,
2900 					sector_t *span)
2901 {
2902 	unsigned int j, k;
2903 	unsigned int nr_good_pages;
2904 	int nr_extents;
2905 	unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
2906 	unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS;
2907 	unsigned long i, idx;
2908 
2909 	nr_good_pages = maxpages - 1;	/* omit header page */
2910 
2911 	cluster_list_init(&p->free_clusters);
2912 	cluster_list_init(&p->discard_clusters);
2913 
2914 	for (i = 0; i < swap_header->info.nr_badpages; i++) {
2915 		unsigned int page_nr = swap_header->info.badpages[i];
2916 		if (page_nr == 0 || page_nr > swap_header->info.last_page)
2917 			return -EINVAL;
2918 		if (page_nr < maxpages) {
2919 			swap_map[page_nr] = SWAP_MAP_BAD;
2920 			nr_good_pages--;
2921 			/*
2922 			 * Haven't marked the cluster free yet, no list
2923 			 * operation involved
2924 			 */
2925 			inc_cluster_info_page(p, cluster_info, page_nr);
2926 		}
2927 	}
2928 
2929 	/* Haven't marked the cluster free yet, no list operation involved */
2930 	for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
2931 		inc_cluster_info_page(p, cluster_info, i);
2932 
2933 	if (nr_good_pages) {
2934 		swap_map[0] = SWAP_MAP_BAD;
2935 		/*
2936 		 * Not mark the cluster free yet, no list
2937 		 * operation involved
2938 		 */
2939 		inc_cluster_info_page(p, cluster_info, 0);
2940 		p->max = maxpages;
2941 		p->pages = nr_good_pages;
2942 		nr_extents = setup_swap_extents(p, span);
2943 		if (nr_extents < 0)
2944 			return nr_extents;
2945 		nr_good_pages = p->pages;
2946 	}
2947 	if (!nr_good_pages) {
2948 		pr_warn("Empty swap-file\n");
2949 		return -EINVAL;
2950 	}
2951 
2952 	if (!cluster_info)
2953 		return nr_extents;
2954 
2955 
2956 	/*
2957 	 * Reduce false cache line sharing between cluster_info and
2958 	 * sharing same address space.
2959 	 */
2960 	for (k = 0; k < SWAP_CLUSTER_COLS; k++) {
2961 		j = (k + col) % SWAP_CLUSTER_COLS;
2962 		for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) {
2963 			idx = i * SWAP_CLUSTER_COLS + j;
2964 			if (idx >= nr_clusters)
2965 				continue;
2966 			if (cluster_count(&cluster_info[idx]))
2967 				continue;
2968 			cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
2969 			cluster_list_add_tail(&p->free_clusters, cluster_info,
2970 					      idx);
2971 		}
2972 	}
2973 	return nr_extents;
2974 }
2975 
SYSCALL_DEFINE2(swapon,const char __user *,specialfile,int,swap_flags)2976 SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2977 {
2978 	struct swap_info_struct *p;
2979 	struct filename *name;
2980 	struct file *swap_file = NULL;
2981 	struct address_space *mapping;
2982 	struct dentry *dentry;
2983 	int prio;
2984 	int error;
2985 	union swap_header *swap_header;
2986 	int nr_extents;
2987 	sector_t span;
2988 	unsigned long maxpages;
2989 	unsigned char *swap_map = NULL;
2990 	struct swap_cluster_info *cluster_info = NULL;
2991 	unsigned long *frontswap_map = NULL;
2992 	struct page *page = NULL;
2993 	struct inode *inode = NULL;
2994 	bool inced_nr_rotate_swap = false;
2995 
2996 	if (swap_flags & ~SWAP_FLAGS_VALID)
2997 		return -EINVAL;
2998 
2999 	if (!capable(CAP_SYS_ADMIN))
3000 		return -EPERM;
3001 
3002 	if (!swap_avail_heads)
3003 		return -ENOMEM;
3004 
3005 	p = alloc_swap_info();
3006 	if (IS_ERR(p))
3007 		return PTR_ERR(p);
3008 
3009 	INIT_WORK(&p->discard_work, swap_discard_work);
3010 
3011 	name = getname(specialfile);
3012 	if (IS_ERR(name)) {
3013 		error = PTR_ERR(name);
3014 		name = NULL;
3015 		goto bad_swap;
3016 	}
3017 	swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
3018 	if (IS_ERR(swap_file)) {
3019 		error = PTR_ERR(swap_file);
3020 		swap_file = NULL;
3021 		goto bad_swap;
3022 	}
3023 
3024 	p->swap_file = swap_file;
3025 	mapping = swap_file->f_mapping;
3026 	dentry = swap_file->f_path.dentry;
3027 	inode = mapping->host;
3028 
3029 	error = claim_swapfile(p, inode);
3030 	if (unlikely(error))
3031 		goto bad_swap;
3032 
3033 	inode_lock(inode);
3034 	if (d_unlinked(dentry) || cant_mount(dentry)) {
3035 		error = -ENOENT;
3036 		goto bad_swap_unlock_inode;
3037 	}
3038 	if (IS_SWAPFILE(inode)) {
3039 		error = -EBUSY;
3040 		goto bad_swap_unlock_inode;
3041 	}
3042 
3043 	/*
3044 	 * Read the swap header.
3045 	 */
3046 	if (!mapping->a_ops->read_folio) {
3047 		error = -EINVAL;
3048 		goto bad_swap_unlock_inode;
3049 	}
3050 	page = read_mapping_page(mapping, 0, swap_file);
3051 	if (IS_ERR(page)) {
3052 		error = PTR_ERR(page);
3053 		goto bad_swap_unlock_inode;
3054 	}
3055 	swap_header = kmap(page);
3056 
3057 	maxpages = read_swap_header(p, swap_header, inode);
3058 	if (unlikely(!maxpages)) {
3059 		error = -EINVAL;
3060 		goto bad_swap_unlock_inode;
3061 	}
3062 
3063 	/* OK, set up the swap map and apply the bad block list */
3064 	swap_map = vzalloc(maxpages);
3065 	if (!swap_map) {
3066 		error = -ENOMEM;
3067 		goto bad_swap_unlock_inode;
3068 	}
3069 
3070 	if (p->bdev && bdev_stable_writes(p->bdev))
3071 		p->flags |= SWP_STABLE_WRITES;
3072 
3073 	if (p->bdev && p->bdev->bd_disk->fops->rw_page)
3074 		p->flags |= SWP_SYNCHRONOUS_IO;
3075 
3076 	if (p->bdev && bdev_nonrot(p->bdev)) {
3077 		int cpu;
3078 		unsigned long ci, nr_cluster;
3079 
3080 		p->flags |= SWP_SOLIDSTATE;
3081 		p->cluster_next_cpu = alloc_percpu(unsigned int);
3082 		if (!p->cluster_next_cpu) {
3083 			error = -ENOMEM;
3084 			goto bad_swap_unlock_inode;
3085 		}
3086 		/*
3087 		 * select a random position to start with to help wear leveling
3088 		 * SSD
3089 		 */
3090 		for_each_possible_cpu(cpu) {
3091 			per_cpu(*p->cluster_next_cpu, cpu) =
3092 				1 + prandom_u32_max(p->highest_bit);
3093 		}
3094 		nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
3095 
3096 		cluster_info = kvcalloc(nr_cluster, sizeof(*cluster_info),
3097 					GFP_KERNEL);
3098 		if (!cluster_info) {
3099 			error = -ENOMEM;
3100 			goto bad_swap_unlock_inode;
3101 		}
3102 
3103 		for (ci = 0; ci < nr_cluster; ci++)
3104 			spin_lock_init(&((cluster_info + ci)->lock));
3105 
3106 		p->percpu_cluster = alloc_percpu(struct percpu_cluster);
3107 		if (!p->percpu_cluster) {
3108 			error = -ENOMEM;
3109 			goto bad_swap_unlock_inode;
3110 		}
3111 		for_each_possible_cpu(cpu) {
3112 			struct percpu_cluster *cluster;
3113 			cluster = per_cpu_ptr(p->percpu_cluster, cpu);
3114 			cluster_set_null(&cluster->index);
3115 		}
3116 	} else {
3117 		atomic_inc(&nr_rotate_swap);
3118 		inced_nr_rotate_swap = true;
3119 	}
3120 
3121 	error = swap_cgroup_swapon(p->type, maxpages);
3122 	if (error)
3123 		goto bad_swap_unlock_inode;
3124 
3125 	nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
3126 		cluster_info, maxpages, &span);
3127 	if (unlikely(nr_extents < 0)) {
3128 		error = nr_extents;
3129 		goto bad_swap_unlock_inode;
3130 	}
3131 	/* frontswap enabled? set up bit-per-page map for frontswap */
3132 	if (IS_ENABLED(CONFIG_FRONTSWAP))
3133 		frontswap_map = kvcalloc(BITS_TO_LONGS(maxpages),
3134 					 sizeof(long),
3135 					 GFP_KERNEL);
3136 
3137 	if ((swap_flags & SWAP_FLAG_DISCARD) &&
3138 	    p->bdev && bdev_max_discard_sectors(p->bdev)) {
3139 		/*
3140 		 * When discard is enabled for swap with no particular
3141 		 * policy flagged, we set all swap discard flags here in
3142 		 * order to sustain backward compatibility with older
3143 		 * swapon(8) releases.
3144 		 */
3145 		p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
3146 			     SWP_PAGE_DISCARD);
3147 
3148 		/*
3149 		 * By flagging sys_swapon, a sysadmin can tell us to
3150 		 * either do single-time area discards only, or to just
3151 		 * perform discards for released swap page-clusters.
3152 		 * Now it's time to adjust the p->flags accordingly.
3153 		 */
3154 		if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
3155 			p->flags &= ~SWP_PAGE_DISCARD;
3156 		else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
3157 			p->flags &= ~SWP_AREA_DISCARD;
3158 
3159 		/* issue a swapon-time discard if it's still required */
3160 		if (p->flags & SWP_AREA_DISCARD) {
3161 			int err = discard_swap(p);
3162 			if (unlikely(err))
3163 				pr_err("swapon: discard_swap(%p): %d\n",
3164 					p, err);
3165 		}
3166 	}
3167 
3168 	error = init_swap_address_space(p->type, maxpages);
3169 	if (error)
3170 		goto bad_swap_unlock_inode;
3171 
3172 	/*
3173 	 * Flush any pending IO and dirty mappings before we start using this
3174 	 * swap device.
3175 	 */
3176 	inode->i_flags |= S_SWAPFILE;
3177 	error = inode_drain_writes(inode);
3178 	if (error) {
3179 		inode->i_flags &= ~S_SWAPFILE;
3180 		goto free_swap_address_space;
3181 	}
3182 
3183 	mutex_lock(&swapon_mutex);
3184 	prio = -1;
3185 	if (swap_flags & SWAP_FLAG_PREFER)
3186 		prio =
3187 		  (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
3188 	enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
3189 
3190 	pr_info("Adding %uk swap on %s.  Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
3191 		p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
3192 		nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
3193 		(p->flags & SWP_SOLIDSTATE) ? "SS" : "",
3194 		(p->flags & SWP_DISCARDABLE) ? "D" : "",
3195 		(p->flags & SWP_AREA_DISCARD) ? "s" : "",
3196 		(p->flags & SWP_PAGE_DISCARD) ? "c" : "",
3197 		(frontswap_map) ? "FS" : "");
3198 
3199 	mutex_unlock(&swapon_mutex);
3200 	atomic_inc(&proc_poll_event);
3201 	wake_up_interruptible(&proc_poll_wait);
3202 
3203 	error = 0;
3204 	goto out;
3205 free_swap_address_space:
3206 	exit_swap_address_space(p->type);
3207 bad_swap_unlock_inode:
3208 	inode_unlock(inode);
3209 bad_swap:
3210 	free_percpu(p->percpu_cluster);
3211 	p->percpu_cluster = NULL;
3212 	free_percpu(p->cluster_next_cpu);
3213 	p->cluster_next_cpu = NULL;
3214 	if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
3215 		set_blocksize(p->bdev, p->old_block_size);
3216 		blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
3217 	}
3218 	inode = NULL;
3219 	destroy_swap_extents(p);
3220 	swap_cgroup_swapoff(p->type);
3221 	spin_lock(&swap_lock);
3222 	p->swap_file = NULL;
3223 	p->flags = 0;
3224 	spin_unlock(&swap_lock);
3225 	vfree(swap_map);
3226 	kvfree(cluster_info);
3227 	kvfree(frontswap_map);
3228 	if (inced_nr_rotate_swap)
3229 		atomic_dec(&nr_rotate_swap);
3230 	if (swap_file)
3231 		filp_close(swap_file, NULL);
3232 out:
3233 	if (page && !IS_ERR(page)) {
3234 		kunmap(page);
3235 		put_page(page);
3236 	}
3237 	if (name)
3238 		putname(name);
3239 	if (inode)
3240 		inode_unlock(inode);
3241 	if (!error)
3242 		enable_swap_slots_cache();
3243 	return error;
3244 }
3245 
si_swapinfo(struct sysinfo * val)3246 void si_swapinfo(struct sysinfo *val)
3247 {
3248 	unsigned int type;
3249 	unsigned long nr_to_be_unused = 0;
3250 
3251 	spin_lock(&swap_lock);
3252 	for (type = 0; type < nr_swapfiles; type++) {
3253 		struct swap_info_struct *si = swap_info[type];
3254 
3255 		if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
3256 			nr_to_be_unused += READ_ONCE(si->inuse_pages);
3257 	}
3258 	val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
3259 	val->totalswap = total_swap_pages + nr_to_be_unused;
3260 	spin_unlock(&swap_lock);
3261 }
3262 
3263 /*
3264  * Verify that a swap entry is valid and increment its swap map count.
3265  *
3266  * Returns error code in following case.
3267  * - success -> 0
3268  * - swp_entry is invalid -> EINVAL
3269  * - swp_entry is migration entry -> EINVAL
3270  * - swap-cache reference is requested but there is already one. -> EEXIST
3271  * - swap-cache reference is requested but the entry is not used. -> ENOENT
3272  * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
3273  */
__swap_duplicate(swp_entry_t entry,unsigned char usage)3274 static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
3275 {
3276 	struct swap_info_struct *p;
3277 	struct swap_cluster_info *ci;
3278 	unsigned long offset;
3279 	unsigned char count;
3280 	unsigned char has_cache;
3281 	int err;
3282 
3283 	p = get_swap_device(entry);
3284 	if (!p)
3285 		return -EINVAL;
3286 
3287 	offset = swp_offset(entry);
3288 	ci = lock_cluster_or_swap_info(p, offset);
3289 
3290 	count = p->swap_map[offset];
3291 
3292 	/*
3293 	 * swapin_readahead() doesn't check if a swap entry is valid, so the
3294 	 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
3295 	 */
3296 	if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
3297 		err = -ENOENT;
3298 		goto unlock_out;
3299 	}
3300 
3301 	has_cache = count & SWAP_HAS_CACHE;
3302 	count &= ~SWAP_HAS_CACHE;
3303 	err = 0;
3304 
3305 	if (usage == SWAP_HAS_CACHE) {
3306 
3307 		/* set SWAP_HAS_CACHE if there is no cache and entry is used */
3308 		if (!has_cache && count)
3309 			has_cache = SWAP_HAS_CACHE;
3310 		else if (has_cache)		/* someone else added cache */
3311 			err = -EEXIST;
3312 		else				/* no users remaining */
3313 			err = -ENOENT;
3314 
3315 	} else if (count || has_cache) {
3316 
3317 		if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
3318 			count += usage;
3319 		else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
3320 			err = -EINVAL;
3321 		else if (swap_count_continued(p, offset, count))
3322 			count = COUNT_CONTINUED;
3323 		else
3324 			err = -ENOMEM;
3325 	} else
3326 		err = -ENOENT;			/* unused swap entry */
3327 
3328 	WRITE_ONCE(p->swap_map[offset], count | has_cache);
3329 
3330 unlock_out:
3331 	unlock_cluster_or_swap_info(p, ci);
3332 	put_swap_device(p);
3333 	return err;
3334 }
3335 
3336 /*
3337  * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3338  * (in which case its reference count is never incremented).
3339  */
swap_shmem_alloc(swp_entry_t entry)3340 void swap_shmem_alloc(swp_entry_t entry)
3341 {
3342 	__swap_duplicate(entry, SWAP_MAP_SHMEM);
3343 }
3344 
3345 /*
3346  * Increase reference count of swap entry by 1.
3347  * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3348  * but could not be atomically allocated.  Returns 0, just as if it succeeded,
3349  * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3350  * might occur if a page table entry has got corrupted.
3351  */
swap_duplicate(swp_entry_t entry)3352 int swap_duplicate(swp_entry_t entry)
3353 {
3354 	int err = 0;
3355 
3356 	while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
3357 		err = add_swap_count_continuation(entry, GFP_ATOMIC);
3358 	return err;
3359 }
3360 
3361 /*
3362  * @entry: swap entry for which we allocate swap cache.
3363  *
3364  * Called when allocating swap cache for existing swap entry,
3365  * This can return error codes. Returns 0 at success.
3366  * -EEXIST means there is a swap cache.
3367  * Note: return code is different from swap_duplicate().
3368  */
swapcache_prepare(swp_entry_t entry)3369 int swapcache_prepare(swp_entry_t entry)
3370 {
3371 	return __swap_duplicate(entry, SWAP_HAS_CACHE);
3372 }
3373 
swp_swap_info(swp_entry_t entry)3374 struct swap_info_struct *swp_swap_info(swp_entry_t entry)
3375 {
3376 	return swap_type_to_swap_info(swp_type(entry));
3377 }
3378 
page_swap_info(struct page * page)3379 struct swap_info_struct *page_swap_info(struct page *page)
3380 {
3381 	swp_entry_t entry = { .val = page_private(page) };
3382 	return swp_swap_info(entry);
3383 }
3384 
3385 /*
3386  * out-of-line methods to avoid include hell.
3387  */
swapcache_mapping(struct folio * folio)3388 struct address_space *swapcache_mapping(struct folio *folio)
3389 {
3390 	return page_swap_info(&folio->page)->swap_file->f_mapping;
3391 }
3392 EXPORT_SYMBOL_GPL(swapcache_mapping);
3393 
__page_file_index(struct page * page)3394 pgoff_t __page_file_index(struct page *page)
3395 {
3396 	swp_entry_t swap = { .val = page_private(page) };
3397 	return swp_offset(swap);
3398 }
3399 EXPORT_SYMBOL_GPL(__page_file_index);
3400 
3401 /*
3402  * add_swap_count_continuation - called when a swap count is duplicated
3403  * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3404  * page of the original vmalloc'ed swap_map, to hold the continuation count
3405  * (for that entry and for its neighbouring PAGE_SIZE swap entries).  Called
3406  * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3407  *
3408  * These continuation pages are seldom referenced: the common paths all work
3409  * on the original swap_map, only referring to a continuation page when the
3410  * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3411  *
3412  * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3413  * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3414  * can be called after dropping locks.
3415  */
add_swap_count_continuation(swp_entry_t entry,gfp_t gfp_mask)3416 int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
3417 {
3418 	struct swap_info_struct *si;
3419 	struct swap_cluster_info *ci;
3420 	struct page *head;
3421 	struct page *page;
3422 	struct page *list_page;
3423 	pgoff_t offset;
3424 	unsigned char count;
3425 	int ret = 0;
3426 
3427 	/*
3428 	 * When debugging, it's easier to use __GFP_ZERO here; but it's better
3429 	 * for latency not to zero a page while GFP_ATOMIC and holding locks.
3430 	 */
3431 	page = alloc_page(gfp_mask | __GFP_HIGHMEM);
3432 
3433 	si = get_swap_device(entry);
3434 	if (!si) {
3435 		/*
3436 		 * An acceptable race has occurred since the failing
3437 		 * __swap_duplicate(): the swap device may be swapoff
3438 		 */
3439 		goto outer;
3440 	}
3441 	spin_lock(&si->lock);
3442 
3443 	offset = swp_offset(entry);
3444 
3445 	ci = lock_cluster(si, offset);
3446 
3447 	count = swap_count(si->swap_map[offset]);
3448 
3449 	if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
3450 		/*
3451 		 * The higher the swap count, the more likely it is that tasks
3452 		 * will race to add swap count continuation: we need to avoid
3453 		 * over-provisioning.
3454 		 */
3455 		goto out;
3456 	}
3457 
3458 	if (!page) {
3459 		ret = -ENOMEM;
3460 		goto out;
3461 	}
3462 
3463 	/*
3464 	 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
3465 	 * no architecture is using highmem pages for kernel page tables: so it
3466 	 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
3467 	 */
3468 	head = vmalloc_to_page(si->swap_map + offset);
3469 	offset &= ~PAGE_MASK;
3470 
3471 	spin_lock(&si->cont_lock);
3472 	/*
3473 	 * Page allocation does not initialize the page's lru field,
3474 	 * but it does always reset its private field.
3475 	 */
3476 	if (!page_private(head)) {
3477 		BUG_ON(count & COUNT_CONTINUED);
3478 		INIT_LIST_HEAD(&head->lru);
3479 		set_page_private(head, SWP_CONTINUED);
3480 		si->flags |= SWP_CONTINUED;
3481 	}
3482 
3483 	list_for_each_entry(list_page, &head->lru, lru) {
3484 		unsigned char *map;
3485 
3486 		/*
3487 		 * If the previous map said no continuation, but we've found
3488 		 * a continuation page, free our allocation and use this one.
3489 		 */
3490 		if (!(count & COUNT_CONTINUED))
3491 			goto out_unlock_cont;
3492 
3493 		map = kmap_atomic(list_page) + offset;
3494 		count = *map;
3495 		kunmap_atomic(map);
3496 
3497 		/*
3498 		 * If this continuation count now has some space in it,
3499 		 * free our allocation and use this one.
3500 		 */
3501 		if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
3502 			goto out_unlock_cont;
3503 	}
3504 
3505 	list_add_tail(&page->lru, &head->lru);
3506 	page = NULL;			/* now it's attached, don't free it */
3507 out_unlock_cont:
3508 	spin_unlock(&si->cont_lock);
3509 out:
3510 	unlock_cluster(ci);
3511 	spin_unlock(&si->lock);
3512 	put_swap_device(si);
3513 outer:
3514 	if (page)
3515 		__free_page(page);
3516 	return ret;
3517 }
3518 
3519 /*
3520  * swap_count_continued - when the original swap_map count is incremented
3521  * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3522  * into, carry if so, or else fail until a new continuation page is allocated;
3523  * when the original swap_map count is decremented from 0 with continuation,
3524  * borrow from the continuation and report whether it still holds more.
3525  * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3526  * lock.
3527  */
swap_count_continued(struct swap_info_struct * si,pgoff_t offset,unsigned char count)3528 static bool swap_count_continued(struct swap_info_struct *si,
3529 				 pgoff_t offset, unsigned char count)
3530 {
3531 	struct page *head;
3532 	struct page *page;
3533 	unsigned char *map;
3534 	bool ret;
3535 
3536 	head = vmalloc_to_page(si->swap_map + offset);
3537 	if (page_private(head) != SWP_CONTINUED) {
3538 		BUG_ON(count & COUNT_CONTINUED);
3539 		return false;		/* need to add count continuation */
3540 	}
3541 
3542 	spin_lock(&si->cont_lock);
3543 	offset &= ~PAGE_MASK;
3544 	page = list_next_entry(head, lru);
3545 	map = kmap_atomic(page) + offset;
3546 
3547 	if (count == SWAP_MAP_MAX)	/* initial increment from swap_map */
3548 		goto init_map;		/* jump over SWAP_CONT_MAX checks */
3549 
3550 	if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
3551 		/*
3552 		 * Think of how you add 1 to 999
3553 		 */
3554 		while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
3555 			kunmap_atomic(map);
3556 			page = list_next_entry(page, lru);
3557 			BUG_ON(page == head);
3558 			map = kmap_atomic(page) + offset;
3559 		}
3560 		if (*map == SWAP_CONT_MAX) {
3561 			kunmap_atomic(map);
3562 			page = list_next_entry(page, lru);
3563 			if (page == head) {
3564 				ret = false;	/* add count continuation */
3565 				goto out;
3566 			}
3567 			map = kmap_atomic(page) + offset;
3568 init_map:		*map = 0;		/* we didn't zero the page */
3569 		}
3570 		*map += 1;
3571 		kunmap_atomic(map);
3572 		while ((page = list_prev_entry(page, lru)) != head) {
3573 			map = kmap_atomic(page) + offset;
3574 			*map = COUNT_CONTINUED;
3575 			kunmap_atomic(map);
3576 		}
3577 		ret = true;			/* incremented */
3578 
3579 	} else {				/* decrementing */
3580 		/*
3581 		 * Think of how you subtract 1 from 1000
3582 		 */
3583 		BUG_ON(count != COUNT_CONTINUED);
3584 		while (*map == COUNT_CONTINUED) {
3585 			kunmap_atomic(map);
3586 			page = list_next_entry(page, lru);
3587 			BUG_ON(page == head);
3588 			map = kmap_atomic(page) + offset;
3589 		}
3590 		BUG_ON(*map == 0);
3591 		*map -= 1;
3592 		if (*map == 0)
3593 			count = 0;
3594 		kunmap_atomic(map);
3595 		while ((page = list_prev_entry(page, lru)) != head) {
3596 			map = kmap_atomic(page) + offset;
3597 			*map = SWAP_CONT_MAX | count;
3598 			count = COUNT_CONTINUED;
3599 			kunmap_atomic(map);
3600 		}
3601 		ret = count == COUNT_CONTINUED;
3602 	}
3603 out:
3604 	spin_unlock(&si->cont_lock);
3605 	return ret;
3606 }
3607 
3608 /*
3609  * free_swap_count_continuations - swapoff free all the continuation pages
3610  * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3611  */
free_swap_count_continuations(struct swap_info_struct * si)3612 static void free_swap_count_continuations(struct swap_info_struct *si)
3613 {
3614 	pgoff_t offset;
3615 
3616 	for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
3617 		struct page *head;
3618 		head = vmalloc_to_page(si->swap_map + offset);
3619 		if (page_private(head)) {
3620 			struct page *page, *next;
3621 
3622 			list_for_each_entry_safe(page, next, &head->lru, lru) {
3623 				list_del(&page->lru);
3624 				__free_page(page);
3625 			}
3626 		}
3627 	}
3628 }
3629 
3630 #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
__cgroup_throttle_swaprate(struct page * page,gfp_t gfp_mask)3631 void __cgroup_throttle_swaprate(struct page *page, gfp_t gfp_mask)
3632 {
3633 	struct swap_info_struct *si, *next;
3634 	int nid = page_to_nid(page);
3635 
3636 	if (!(gfp_mask & __GFP_IO))
3637 		return;
3638 
3639 	if (!blk_cgroup_congested())
3640 		return;
3641 
3642 	/*
3643 	 * We've already scheduled a throttle, avoid taking the global swap
3644 	 * lock.
3645 	 */
3646 	if (current->throttle_queue)
3647 		return;
3648 
3649 	spin_lock(&swap_avail_lock);
3650 	plist_for_each_entry_safe(si, next, &swap_avail_heads[nid],
3651 				  avail_lists[nid]) {
3652 		if (si->bdev) {
3653 			blkcg_schedule_throttle(si->bdev->bd_disk, true);
3654 			break;
3655 		}
3656 	}
3657 	spin_unlock(&swap_avail_lock);
3658 }
3659 #endif
3660 
swapfile_init(void)3661 static int __init swapfile_init(void)
3662 {
3663 	int nid;
3664 
3665 	swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head),
3666 					 GFP_KERNEL);
3667 	if (!swap_avail_heads) {
3668 		pr_emerg("Not enough memory for swap heads, swap is disabled\n");
3669 		return -ENOMEM;
3670 	}
3671 
3672 	for_each_node(nid)
3673 		plist_head_init(&swap_avail_heads[nid]);
3674 
3675 	swapfile_maximum_size = arch_max_swapfile_size();
3676 
3677 #ifdef CONFIG_MIGRATION
3678 	if (swapfile_maximum_size >= (1UL << SWP_MIG_TOTAL_BITS))
3679 		swap_migration_ad_supported = true;
3680 #endif	/* CONFIG_MIGRATION */
3681 
3682 	return 0;
3683 }
3684 subsys_initcall(swapfile_init);
3685