Line data Source code
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 : #include <linux/suspend.h>
45 :
46 : #include <asm/tlbflush.h>
47 : #include <linux/swapops.h>
48 : #include <linux/swap_cgroup.h>
49 : #include "swap.h"
50 :
51 : static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
52 : unsigned char);
53 : static void free_swap_count_continuations(struct swap_info_struct *);
54 :
55 : static DEFINE_SPINLOCK(swap_lock);
56 : static unsigned int nr_swapfiles;
57 : atomic_long_t nr_swap_pages;
58 : /*
59 : * Some modules use swappable objects and may try to swap them out under
60 : * memory pressure (via the shrinker). Before doing so, they may wish to
61 : * check to see if any swap space is available.
62 : */
63 : EXPORT_SYMBOL_GPL(nr_swap_pages);
64 : /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
65 : long total_swap_pages;
66 : static int least_priority = -1;
67 : unsigned long swapfile_maximum_size;
68 : #ifdef CONFIG_MIGRATION
69 : bool swap_migration_ad_supported;
70 : #endif /* CONFIG_MIGRATION */
71 :
72 : static const char Bad_file[] = "Bad swap file entry ";
73 : static const char Unused_file[] = "Unused swap file entry ";
74 : static const char Bad_offset[] = "Bad swap offset entry ";
75 : static const char Unused_offset[] = "Unused swap offset entry ";
76 :
77 : /*
78 : * all active swap_info_structs
79 : * protected with swap_lock, and ordered by priority.
80 : */
81 : static PLIST_HEAD(swap_active_head);
82 :
83 : /*
84 : * all available (active, not full) swap_info_structs
85 : * protected with swap_avail_lock, ordered by priority.
86 : * This is used by folio_alloc_swap() instead of swap_active_head
87 : * because swap_active_head includes all swap_info_structs,
88 : * but folio_alloc_swap() doesn't need to look at full ones.
89 : * This uses its own lock instead of swap_lock because when a
90 : * swap_info_struct changes between not-full/full, it needs to
91 : * add/remove itself to/from this list, but the swap_info_struct->lock
92 : * is held and the locking order requires swap_lock to be taken
93 : * before any swap_info_struct->lock.
94 : */
95 : static struct plist_head *swap_avail_heads;
96 : static DEFINE_SPINLOCK(swap_avail_lock);
97 :
98 : struct swap_info_struct *swap_info[MAX_SWAPFILES];
99 :
100 : static DEFINE_MUTEX(swapon_mutex);
101 :
102 : static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
103 : /* Activity counter to indicate that a swapon or swapoff has occurred */
104 : static atomic_t proc_poll_event = ATOMIC_INIT(0);
105 :
106 : atomic_t nr_rotate_swap = ATOMIC_INIT(0);
107 :
108 : static struct swap_info_struct *swap_type_to_swap_info(int type)
109 : {
110 0 : if (type >= MAX_SWAPFILES)
111 : return NULL;
112 :
113 0 : return READ_ONCE(swap_info[type]); /* rcu_dereference() */
114 : }
115 :
116 : static inline unsigned char swap_count(unsigned char ent)
117 : {
118 0 : return ent & ~SWAP_HAS_CACHE; /* may include COUNT_CONTINUED flag */
119 : }
120 :
121 : /* Reclaim the swap entry anyway if possible */
122 : #define TTRS_ANYWAY 0x1
123 : /*
124 : * Reclaim the swap entry if there are no more mappings of the
125 : * corresponding page
126 : */
127 : #define TTRS_UNMAPPED 0x2
128 : /* Reclaim the swap entry if swap is getting full*/
129 : #define TTRS_FULL 0x4
130 :
131 : /* returns 1 if swap entry is freed */
132 0 : static int __try_to_reclaim_swap(struct swap_info_struct *si,
133 : unsigned long offset, unsigned long flags)
134 : {
135 0 : swp_entry_t entry = swp_entry(si->type, offset);
136 : struct folio *folio;
137 0 : int ret = 0;
138 :
139 0 : folio = filemap_get_folio(swap_address_space(entry), offset);
140 0 : if (IS_ERR(folio))
141 : return 0;
142 : /*
143 : * When this function is called from scan_swap_map_slots() and it's
144 : * called by vmscan.c at reclaiming folios. So we hold a folio lock
145 : * here. We have to use trylock for avoiding deadlock. This is a special
146 : * case and you should use folio_free_swap() with explicit folio_lock()
147 : * in usual operations.
148 : */
149 0 : if (folio_trylock(folio)) {
150 0 : if ((flags & TTRS_ANYWAY) ||
151 0 : ((flags & TTRS_UNMAPPED) && !folio_mapped(folio)) ||
152 0 : ((flags & TTRS_FULL) && mem_cgroup_swap_full(folio)))
153 0 : ret = folio_free_swap(folio);
154 0 : folio_unlock(folio);
155 : }
156 : folio_put(folio);
157 : return ret;
158 : }
159 :
160 : static inline struct swap_extent *first_se(struct swap_info_struct *sis)
161 : {
162 0 : struct rb_node *rb = rb_first(&sis->swap_extent_root);
163 0 : return rb_entry(rb, struct swap_extent, rb_node);
164 : }
165 :
166 : static inline struct swap_extent *next_se(struct swap_extent *se)
167 : {
168 0 : struct rb_node *rb = rb_next(&se->rb_node);
169 0 : return rb ? rb_entry(rb, struct swap_extent, rb_node) : NULL;
170 : }
171 :
172 : /*
173 : * swapon tell device that all the old swap contents can be discarded,
174 : * to allow the swap device to optimize its wear-levelling.
175 : */
176 0 : static int discard_swap(struct swap_info_struct *si)
177 : {
178 : struct swap_extent *se;
179 : sector_t start_block;
180 : sector_t nr_blocks;
181 0 : int err = 0;
182 :
183 : /* Do not discard the swap header page! */
184 0 : se = first_se(si);
185 0 : start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
186 0 : nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
187 0 : if (nr_blocks) {
188 0 : err = blkdev_issue_discard(si->bdev, start_block,
189 : nr_blocks, GFP_KERNEL);
190 0 : if (err)
191 : return err;
192 0 : cond_resched();
193 : }
194 :
195 0 : for (se = next_se(se); se; se = next_se(se)) {
196 0 : start_block = se->start_block << (PAGE_SHIFT - 9);
197 0 : nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
198 :
199 0 : err = blkdev_issue_discard(si->bdev, start_block,
200 : nr_blocks, GFP_KERNEL);
201 0 : if (err)
202 : break;
203 :
204 0 : cond_resched();
205 : }
206 : return err; /* That will often be -EOPNOTSUPP */
207 : }
208 :
209 : static struct swap_extent *
210 0 : offset_to_swap_extent(struct swap_info_struct *sis, unsigned long offset)
211 : {
212 : struct swap_extent *se;
213 : struct rb_node *rb;
214 :
215 0 : rb = sis->swap_extent_root.rb_node;
216 0 : while (rb) {
217 0 : se = rb_entry(rb, struct swap_extent, rb_node);
218 0 : if (offset < se->start_page)
219 0 : rb = rb->rb_left;
220 0 : else if (offset >= se->start_page + se->nr_pages)
221 0 : rb = rb->rb_right;
222 : else
223 0 : return se;
224 : }
225 : /* It *must* be present */
226 0 : BUG();
227 : }
228 :
229 0 : sector_t swap_page_sector(struct page *page)
230 : {
231 0 : struct swap_info_struct *sis = page_swap_info(page);
232 : struct swap_extent *se;
233 : sector_t sector;
234 : pgoff_t offset;
235 :
236 0 : offset = __page_file_index(page);
237 0 : se = offset_to_swap_extent(sis, offset);
238 0 : sector = se->start_block + (offset - se->start_page);
239 0 : return sector << (PAGE_SHIFT - 9);
240 : }
241 :
242 : /*
243 : * swap allocation tell device that a cluster of swap can now be discarded,
244 : * to allow the swap device to optimize its wear-levelling.
245 : */
246 0 : static void discard_swap_cluster(struct swap_info_struct *si,
247 : pgoff_t start_page, pgoff_t nr_pages)
248 : {
249 0 : struct swap_extent *se = offset_to_swap_extent(si, start_page);
250 :
251 0 : while (nr_pages) {
252 0 : pgoff_t offset = start_page - se->start_page;
253 0 : sector_t start_block = se->start_block + offset;
254 0 : sector_t nr_blocks = se->nr_pages - offset;
255 :
256 0 : if (nr_blocks > nr_pages)
257 0 : nr_blocks = nr_pages;
258 0 : start_page += nr_blocks;
259 0 : nr_pages -= nr_blocks;
260 :
261 0 : start_block <<= PAGE_SHIFT - 9;
262 0 : nr_blocks <<= PAGE_SHIFT - 9;
263 0 : if (blkdev_issue_discard(si->bdev, start_block,
264 : nr_blocks, GFP_NOIO))
265 : break;
266 :
267 : se = next_se(se);
268 : }
269 0 : }
270 :
271 : #ifdef CONFIG_THP_SWAP
272 : #define SWAPFILE_CLUSTER HPAGE_PMD_NR
273 :
274 : #define swap_entry_size(size) (size)
275 : #else
276 : #define SWAPFILE_CLUSTER 256
277 :
278 : /*
279 : * Define swap_entry_size() as constant to let compiler to optimize
280 : * out some code if !CONFIG_THP_SWAP
281 : */
282 : #define swap_entry_size(size) 1
283 : #endif
284 : #define LATENCY_LIMIT 256
285 :
286 : static inline void cluster_set_flag(struct swap_cluster_info *info,
287 : unsigned int flag)
288 : {
289 0 : info->flags = flag;
290 : }
291 :
292 : static inline unsigned int cluster_count(struct swap_cluster_info *info)
293 : {
294 0 : return info->data;
295 : }
296 :
297 : static inline void cluster_set_count(struct swap_cluster_info *info,
298 : unsigned int c)
299 : {
300 0 : info->data = c;
301 : }
302 :
303 : static inline void cluster_set_count_flag(struct swap_cluster_info *info,
304 : unsigned int c, unsigned int f)
305 : {
306 0 : info->flags = f;
307 0 : info->data = c;
308 : }
309 :
310 : static inline unsigned int cluster_next(struct swap_cluster_info *info)
311 : {
312 0 : return info->data;
313 : }
314 :
315 : static inline void cluster_set_next(struct swap_cluster_info *info,
316 : unsigned int n)
317 : {
318 0 : info->data = n;
319 : }
320 :
321 : static inline void cluster_set_next_flag(struct swap_cluster_info *info,
322 : unsigned int n, unsigned int f)
323 : {
324 0 : info->flags = f;
325 0 : info->data = n;
326 : }
327 :
328 : static inline bool cluster_is_free(struct swap_cluster_info *info)
329 : {
330 0 : return info->flags & CLUSTER_FLAG_FREE;
331 : }
332 :
333 : static inline bool cluster_is_null(struct swap_cluster_info *info)
334 : {
335 0 : return info->flags & CLUSTER_FLAG_NEXT_NULL;
336 : }
337 :
338 : static inline void cluster_set_null(struct swap_cluster_info *info)
339 : {
340 0 : info->flags = CLUSTER_FLAG_NEXT_NULL;
341 0 : info->data = 0;
342 : }
343 :
344 : static inline bool cluster_is_huge(struct swap_cluster_info *info)
345 : {
346 : if (IS_ENABLED(CONFIG_THP_SWAP))
347 : return info->flags & CLUSTER_FLAG_HUGE;
348 : return false;
349 : }
350 :
351 : static inline void cluster_clear_huge(struct swap_cluster_info *info)
352 : {
353 : info->flags &= ~CLUSTER_FLAG_HUGE;
354 : }
355 :
356 : static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si,
357 : unsigned long offset)
358 : {
359 : struct swap_cluster_info *ci;
360 :
361 0 : ci = si->cluster_info;
362 0 : if (ci) {
363 0 : ci += offset / SWAPFILE_CLUSTER;
364 0 : spin_lock(&ci->lock);
365 : }
366 : return ci;
367 : }
368 :
369 : static inline void unlock_cluster(struct swap_cluster_info *ci)
370 : {
371 0 : if (ci)
372 0 : spin_unlock(&ci->lock);
373 : }
374 :
375 : /*
376 : * Determine the locking method in use for this device. Return
377 : * swap_cluster_info if SSD-style cluster-based locking is in place.
378 : */
379 : static inline struct swap_cluster_info *lock_cluster_or_swap_info(
380 : struct swap_info_struct *si, unsigned long offset)
381 : {
382 : struct swap_cluster_info *ci;
383 :
384 : /* Try to use fine-grained SSD-style locking if available: */
385 0 : ci = lock_cluster(si, offset);
386 : /* Otherwise, fall back to traditional, coarse locking: */
387 0 : if (!ci)
388 0 : spin_lock(&si->lock);
389 :
390 : return ci;
391 : }
392 :
393 : static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si,
394 : struct swap_cluster_info *ci)
395 : {
396 0 : if (ci)
397 : unlock_cluster(ci);
398 : else
399 0 : spin_unlock(&si->lock);
400 : }
401 :
402 : static inline bool cluster_list_empty(struct swap_cluster_list *list)
403 : {
404 0 : return cluster_is_null(&list->head);
405 : }
406 :
407 : static inline unsigned int cluster_list_first(struct swap_cluster_list *list)
408 : {
409 0 : return cluster_next(&list->head);
410 : }
411 :
412 : static void cluster_list_init(struct swap_cluster_list *list)
413 : {
414 0 : cluster_set_null(&list->head);
415 0 : cluster_set_null(&list->tail);
416 : }
417 :
418 : static void cluster_list_add_tail(struct swap_cluster_list *list,
419 : struct swap_cluster_info *ci,
420 : unsigned int idx)
421 : {
422 0 : if (cluster_list_empty(list)) {
423 0 : cluster_set_next_flag(&list->head, idx, 0);
424 0 : cluster_set_next_flag(&list->tail, idx, 0);
425 : } else {
426 : struct swap_cluster_info *ci_tail;
427 0 : unsigned int tail = cluster_next(&list->tail);
428 :
429 : /*
430 : * Nested cluster lock, but both cluster locks are
431 : * only acquired when we held swap_info_struct->lock
432 : */
433 0 : ci_tail = ci + tail;
434 0 : spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING);
435 0 : cluster_set_next(ci_tail, idx);
436 0 : spin_unlock(&ci_tail->lock);
437 0 : cluster_set_next_flag(&list->tail, idx, 0);
438 : }
439 : }
440 :
441 : static unsigned int cluster_list_del_first(struct swap_cluster_list *list,
442 : struct swap_cluster_info *ci)
443 : {
444 : unsigned int idx;
445 :
446 0 : idx = cluster_next(&list->head);
447 0 : if (cluster_next(&list->tail) == idx) {
448 0 : cluster_set_null(&list->head);
449 0 : cluster_set_null(&list->tail);
450 : } else
451 0 : cluster_set_next_flag(&list->head,
452 0 : cluster_next(&ci[idx]), 0);
453 :
454 : return idx;
455 : }
456 :
457 : /* Add a cluster to discard list and schedule it to do discard */
458 0 : static void swap_cluster_schedule_discard(struct swap_info_struct *si,
459 : unsigned int idx)
460 : {
461 : /*
462 : * If scan_swap_map_slots() can't find a free cluster, it will check
463 : * si->swap_map directly. To make sure the discarding cluster isn't
464 : * taken by scan_swap_map_slots(), mark the swap entries bad (occupied).
465 : * It will be cleared after discard
466 : */
467 0 : memset(si->swap_map + idx * SWAPFILE_CLUSTER,
468 : SWAP_MAP_BAD, SWAPFILE_CLUSTER);
469 :
470 0 : cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx);
471 :
472 0 : schedule_work(&si->discard_work);
473 0 : }
474 :
475 : static void __free_cluster(struct swap_info_struct *si, unsigned long idx)
476 : {
477 0 : struct swap_cluster_info *ci = si->cluster_info;
478 :
479 0 : cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE);
480 0 : cluster_list_add_tail(&si->free_clusters, ci, idx);
481 : }
482 :
483 : /*
484 : * Doing discard actually. After a cluster discard is finished, the cluster
485 : * will be added to free cluster list. caller should hold si->lock.
486 : */
487 0 : static void swap_do_scheduled_discard(struct swap_info_struct *si)
488 : {
489 : struct swap_cluster_info *info, *ci;
490 : unsigned int idx;
491 :
492 0 : info = si->cluster_info;
493 :
494 0 : while (!cluster_list_empty(&si->discard_clusters)) {
495 0 : idx = cluster_list_del_first(&si->discard_clusters, info);
496 0 : spin_unlock(&si->lock);
497 :
498 0 : discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
499 : SWAPFILE_CLUSTER);
500 :
501 0 : spin_lock(&si->lock);
502 0 : ci = lock_cluster(si, idx * SWAPFILE_CLUSTER);
503 0 : __free_cluster(si, idx);
504 0 : memset(si->swap_map + idx * SWAPFILE_CLUSTER,
505 : 0, SWAPFILE_CLUSTER);
506 : unlock_cluster(ci);
507 : }
508 0 : }
509 :
510 0 : static void swap_discard_work(struct work_struct *work)
511 : {
512 : struct swap_info_struct *si;
513 :
514 0 : si = container_of(work, struct swap_info_struct, discard_work);
515 :
516 0 : spin_lock(&si->lock);
517 0 : swap_do_scheduled_discard(si);
518 0 : spin_unlock(&si->lock);
519 0 : }
520 :
521 0 : static void swap_users_ref_free(struct percpu_ref *ref)
522 : {
523 : struct swap_info_struct *si;
524 :
525 0 : si = container_of(ref, struct swap_info_struct, users);
526 0 : complete(&si->comp);
527 0 : }
528 :
529 : static void alloc_cluster(struct swap_info_struct *si, unsigned long idx)
530 : {
531 0 : struct swap_cluster_info *ci = si->cluster_info;
532 :
533 : VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx);
534 0 : cluster_list_del_first(&si->free_clusters, ci);
535 0 : cluster_set_count_flag(ci + idx, 0, 0);
536 : }
537 :
538 0 : static void free_cluster(struct swap_info_struct *si, unsigned long idx)
539 : {
540 0 : struct swap_cluster_info *ci = si->cluster_info + idx;
541 :
542 : VM_BUG_ON(cluster_count(ci) != 0);
543 : /*
544 : * If the swap is discardable, prepare discard the cluster
545 : * instead of free it immediately. The cluster will be freed
546 : * after discard.
547 : */
548 0 : if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
549 : (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
550 0 : swap_cluster_schedule_discard(si, idx);
551 0 : return;
552 : }
553 :
554 : __free_cluster(si, idx);
555 : }
556 :
557 : /*
558 : * The cluster corresponding to page_nr will be used. The cluster will be
559 : * removed from free cluster list and its usage counter will be increased.
560 : */
561 0 : static void inc_cluster_info_page(struct swap_info_struct *p,
562 : struct swap_cluster_info *cluster_info, unsigned long page_nr)
563 : {
564 0 : unsigned long idx = page_nr / SWAPFILE_CLUSTER;
565 :
566 0 : if (!cluster_info)
567 : return;
568 0 : if (cluster_is_free(&cluster_info[idx]))
569 : alloc_cluster(p, idx);
570 :
571 : VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
572 0 : cluster_set_count(&cluster_info[idx],
573 0 : cluster_count(&cluster_info[idx]) + 1);
574 : }
575 :
576 : /*
577 : * The cluster corresponding to page_nr decreases one usage. If the usage
578 : * counter becomes 0, which means no page in the cluster is in using, we can
579 : * optionally discard the cluster and add it to free cluster list.
580 : */
581 0 : static void dec_cluster_info_page(struct swap_info_struct *p,
582 : struct swap_cluster_info *cluster_info, unsigned long page_nr)
583 : {
584 0 : unsigned long idx = page_nr / SWAPFILE_CLUSTER;
585 :
586 0 : if (!cluster_info)
587 : return;
588 :
589 : VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
590 0 : cluster_set_count(&cluster_info[idx],
591 0 : cluster_count(&cluster_info[idx]) - 1);
592 :
593 0 : if (cluster_count(&cluster_info[idx]) == 0)
594 0 : free_cluster(p, idx);
595 : }
596 :
597 : /*
598 : * It's possible scan_swap_map_slots() uses a free cluster in the middle of free
599 : * cluster list. Avoiding such abuse to avoid list corruption.
600 : */
601 : static bool
602 : scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
603 : unsigned long offset)
604 : {
605 : struct percpu_cluster *percpu_cluster;
606 : bool conflict;
607 :
608 0 : offset /= SWAPFILE_CLUSTER;
609 0 : conflict = !cluster_list_empty(&si->free_clusters) &&
610 0 : offset != cluster_list_first(&si->free_clusters) &&
611 0 : cluster_is_free(&si->cluster_info[offset]);
612 :
613 0 : if (!conflict)
614 : return false;
615 :
616 0 : percpu_cluster = this_cpu_ptr(si->percpu_cluster);
617 0 : cluster_set_null(&percpu_cluster->index);
618 : return true;
619 : }
620 :
621 : /*
622 : * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
623 : * might involve allocating a new cluster for current CPU too.
624 : */
625 0 : static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
626 : unsigned long *offset, unsigned long *scan_base)
627 : {
628 : struct percpu_cluster *cluster;
629 : struct swap_cluster_info *ci;
630 : unsigned long tmp, max;
631 :
632 : new_cluster:
633 0 : cluster = this_cpu_ptr(si->percpu_cluster);
634 0 : if (cluster_is_null(&cluster->index)) {
635 0 : if (!cluster_list_empty(&si->free_clusters)) {
636 0 : cluster->index = si->free_clusters.head;
637 0 : cluster->next = cluster_next(&cluster->index) *
638 : SWAPFILE_CLUSTER;
639 0 : } else if (!cluster_list_empty(&si->discard_clusters)) {
640 : /*
641 : * we don't have free cluster but have some clusters in
642 : * discarding, do discard now and reclaim them, then
643 : * reread cluster_next_cpu since we dropped si->lock
644 : */
645 0 : swap_do_scheduled_discard(si);
646 0 : *scan_base = this_cpu_read(*si->cluster_next_cpu);
647 0 : *offset = *scan_base;
648 0 : goto new_cluster;
649 : } else
650 : return false;
651 : }
652 :
653 : /*
654 : * Other CPUs can use our cluster if they can't find a free cluster,
655 : * check if there is still free entry in the cluster
656 : */
657 0 : tmp = cluster->next;
658 0 : max = min_t(unsigned long, si->max,
659 : (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER);
660 0 : if (tmp < max) {
661 0 : ci = lock_cluster(si, tmp);
662 0 : while (tmp < max) {
663 0 : if (!si->swap_map[tmp])
664 : break;
665 0 : tmp++;
666 : }
667 : unlock_cluster(ci);
668 : }
669 0 : if (tmp >= max) {
670 0 : cluster_set_null(&cluster->index);
671 : goto new_cluster;
672 : }
673 0 : cluster->next = tmp + 1;
674 0 : *offset = tmp;
675 0 : *scan_base = tmp;
676 0 : return true;
677 : }
678 :
679 : static void __del_from_avail_list(struct swap_info_struct *p)
680 : {
681 : int nid;
682 :
683 : assert_spin_locked(&p->lock);
684 0 : for_each_node(nid)
685 0 : plist_del(&p->avail_lists[nid], &swap_avail_heads[nid]);
686 : }
687 :
688 : static void del_from_avail_list(struct swap_info_struct *p)
689 : {
690 : spin_lock(&swap_avail_lock);
691 0 : __del_from_avail_list(p);
692 0 : spin_unlock(&swap_avail_lock);
693 : }
694 :
695 0 : static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset,
696 : unsigned int nr_entries)
697 : {
698 0 : unsigned int end = offset + nr_entries - 1;
699 :
700 0 : if (offset == si->lowest_bit)
701 0 : si->lowest_bit += nr_entries;
702 0 : if (end == si->highest_bit)
703 0 : WRITE_ONCE(si->highest_bit, si->highest_bit - nr_entries);
704 0 : WRITE_ONCE(si->inuse_pages, si->inuse_pages + nr_entries);
705 0 : if (si->inuse_pages == si->pages) {
706 0 : si->lowest_bit = si->max;
707 0 : si->highest_bit = 0;
708 : del_from_avail_list(si);
709 : }
710 0 : }
711 :
712 0 : static void add_to_avail_list(struct swap_info_struct *p)
713 : {
714 : int nid;
715 :
716 0 : spin_lock(&swap_avail_lock);
717 0 : for_each_node(nid) {
718 0 : WARN_ON(!plist_node_empty(&p->avail_lists[nid]));
719 0 : plist_add(&p->avail_lists[nid], &swap_avail_heads[nid]);
720 : }
721 0 : spin_unlock(&swap_avail_lock);
722 0 : }
723 :
724 0 : static void swap_range_free(struct swap_info_struct *si, unsigned long offset,
725 : unsigned int nr_entries)
726 : {
727 0 : unsigned long begin = offset;
728 0 : unsigned long end = offset + nr_entries - 1;
729 : void (*swap_slot_free_notify)(struct block_device *, unsigned long);
730 :
731 0 : if (offset < si->lowest_bit)
732 0 : si->lowest_bit = offset;
733 0 : if (end > si->highest_bit) {
734 0 : bool was_full = !si->highest_bit;
735 :
736 0 : WRITE_ONCE(si->highest_bit, end);
737 0 : if (was_full && (si->flags & SWP_WRITEOK))
738 0 : add_to_avail_list(si);
739 : }
740 0 : atomic_long_add(nr_entries, &nr_swap_pages);
741 0 : WRITE_ONCE(si->inuse_pages, si->inuse_pages - nr_entries);
742 0 : if (si->flags & SWP_BLKDEV)
743 0 : swap_slot_free_notify =
744 0 : si->bdev->bd_disk->fops->swap_slot_free_notify;
745 : else
746 : swap_slot_free_notify = NULL;
747 0 : while (offset <= end) {
748 0 : arch_swap_invalidate_page(si->type, offset);
749 0 : frontswap_invalidate_page(si->type, offset);
750 0 : if (swap_slot_free_notify)
751 0 : swap_slot_free_notify(si->bdev, offset);
752 0 : offset++;
753 : }
754 0 : clear_shadow_from_swap_cache(si->type, begin, end);
755 0 : }
756 :
757 0 : static void set_cluster_next(struct swap_info_struct *si, unsigned long next)
758 : {
759 : unsigned long prev;
760 :
761 0 : if (!(si->flags & SWP_SOLIDSTATE)) {
762 0 : si->cluster_next = next;
763 0 : return;
764 : }
765 :
766 0 : prev = this_cpu_read(*si->cluster_next_cpu);
767 : /*
768 : * Cross the swap address space size aligned trunk, choose
769 : * another trunk randomly to avoid lock contention on swap
770 : * address space if possible.
771 : */
772 0 : if ((prev >> SWAP_ADDRESS_SPACE_SHIFT) !=
773 0 : (next >> SWAP_ADDRESS_SPACE_SHIFT)) {
774 : /* No free swap slots available */
775 0 : if (si->highest_bit <= si->lowest_bit)
776 : return;
777 0 : next = get_random_u32_inclusive(si->lowest_bit, si->highest_bit);
778 0 : next = ALIGN_DOWN(next, SWAP_ADDRESS_SPACE_PAGES);
779 0 : next = max_t(unsigned int, next, si->lowest_bit);
780 : }
781 0 : this_cpu_write(*si->cluster_next_cpu, next);
782 : }
783 :
784 : static bool swap_offset_available_and_locked(struct swap_info_struct *si,
785 : unsigned long offset)
786 : {
787 0 : if (data_race(!si->swap_map[offset])) {
788 0 : spin_lock(&si->lock);
789 : return true;
790 : }
791 :
792 0 : if (vm_swap_full() && READ_ONCE(si->swap_map[offset]) == SWAP_HAS_CACHE) {
793 0 : spin_lock(&si->lock);
794 : return true;
795 : }
796 :
797 : return false;
798 : }
799 :
800 0 : static int scan_swap_map_slots(struct swap_info_struct *si,
801 : unsigned char usage, int nr,
802 : swp_entry_t slots[])
803 : {
804 : struct swap_cluster_info *ci;
805 : unsigned long offset;
806 : unsigned long scan_base;
807 0 : unsigned long last_in_cluster = 0;
808 0 : int latency_ration = LATENCY_LIMIT;
809 0 : int n_ret = 0;
810 0 : bool scanned_many = false;
811 :
812 : /*
813 : * We try to cluster swap pages by allocating them sequentially
814 : * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
815 : * way, however, we resort to first-free allocation, starting
816 : * a new cluster. This prevents us from scattering swap pages
817 : * all over the entire swap partition, so that we reduce
818 : * overall disk seek times between swap pages. -- sct
819 : * But we do now try to find an empty cluster. -Andrea
820 : * And we let swap pages go all over an SSD partition. Hugh
821 : */
822 :
823 0 : si->flags += SWP_SCANNING;
824 : /*
825 : * Use percpu scan base for SSD to reduce lock contention on
826 : * cluster and swap cache. For HDD, sequential access is more
827 : * important.
828 : */
829 0 : if (si->flags & SWP_SOLIDSTATE)
830 0 : scan_base = this_cpu_read(*si->cluster_next_cpu);
831 : else
832 0 : scan_base = si->cluster_next;
833 0 : offset = scan_base;
834 :
835 : /* SSD algorithm */
836 0 : if (si->cluster_info) {
837 0 : if (!scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
838 : goto scan;
839 0 : } else if (unlikely(!si->cluster_nr--)) {
840 0 : if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
841 0 : si->cluster_nr = SWAPFILE_CLUSTER - 1;
842 0 : goto checks;
843 : }
844 :
845 0 : spin_unlock(&si->lock);
846 :
847 : /*
848 : * If seek is expensive, start searching for new cluster from
849 : * start of partition, to minimize the span of allocated swap.
850 : * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
851 : * case, just handled by scan_swap_map_try_ssd_cluster() above.
852 : */
853 0 : scan_base = offset = si->lowest_bit;
854 0 : last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
855 :
856 : /* Locate the first empty (unaligned) cluster */
857 0 : for (; last_in_cluster <= si->highest_bit; offset++) {
858 0 : if (si->swap_map[offset])
859 0 : last_in_cluster = offset + SWAPFILE_CLUSTER;
860 0 : else if (offset == last_in_cluster) {
861 0 : spin_lock(&si->lock);
862 0 : offset -= SWAPFILE_CLUSTER - 1;
863 0 : si->cluster_next = offset;
864 0 : si->cluster_nr = SWAPFILE_CLUSTER - 1;
865 0 : goto checks;
866 : }
867 0 : if (unlikely(--latency_ration < 0)) {
868 0 : cond_resched();
869 0 : latency_ration = LATENCY_LIMIT;
870 : }
871 : }
872 :
873 0 : offset = scan_base;
874 0 : spin_lock(&si->lock);
875 0 : si->cluster_nr = SWAPFILE_CLUSTER - 1;
876 : }
877 :
878 : checks:
879 0 : if (si->cluster_info) {
880 0 : while (scan_swap_map_ssd_cluster_conflict(si, offset)) {
881 : /* take a break if we already got some slots */
882 0 : if (n_ret)
883 : goto done;
884 0 : if (!scan_swap_map_try_ssd_cluster(si, &offset,
885 : &scan_base))
886 : goto scan;
887 : }
888 : }
889 0 : if (!(si->flags & SWP_WRITEOK))
890 : goto no_page;
891 0 : if (!si->highest_bit)
892 : goto no_page;
893 0 : if (offset > si->highest_bit)
894 0 : scan_base = offset = si->lowest_bit;
895 :
896 0 : ci = lock_cluster(si, offset);
897 : /* reuse swap entry of cache-only swap if not busy. */
898 0 : if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
899 : int swap_was_freed;
900 0 : unlock_cluster(ci);
901 0 : spin_unlock(&si->lock);
902 0 : swap_was_freed = __try_to_reclaim_swap(si, offset, TTRS_ANYWAY);
903 0 : spin_lock(&si->lock);
904 : /* entry was freed successfully, try to use this again */
905 0 : if (swap_was_freed)
906 : goto checks;
907 : goto scan; /* check next one */
908 : }
909 :
910 0 : if (si->swap_map[offset]) {
911 0 : unlock_cluster(ci);
912 0 : if (!n_ret)
913 : goto scan;
914 : else
915 : goto done;
916 : }
917 0 : WRITE_ONCE(si->swap_map[offset], usage);
918 0 : inc_cluster_info_page(si, si->cluster_info, offset);
919 0 : unlock_cluster(ci);
920 :
921 0 : swap_range_alloc(si, offset, 1);
922 0 : slots[n_ret++] = swp_entry(si->type, offset);
923 :
924 : /* got enough slots or reach max slots? */
925 0 : if ((n_ret == nr) || (offset >= si->highest_bit))
926 : goto done;
927 :
928 : /* search for next available slot */
929 :
930 : /* time to take a break? */
931 0 : if (unlikely(--latency_ration < 0)) {
932 0 : if (n_ret)
933 : goto done;
934 0 : spin_unlock(&si->lock);
935 0 : cond_resched();
936 0 : spin_lock(&si->lock);
937 0 : latency_ration = LATENCY_LIMIT;
938 : }
939 :
940 : /* try to get more slots in cluster */
941 0 : if (si->cluster_info) {
942 0 : if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
943 : goto checks;
944 0 : } else if (si->cluster_nr && !si->swap_map[++offset]) {
945 : /* non-ssd case, still more slots in cluster? */
946 0 : --si->cluster_nr;
947 0 : goto checks;
948 : }
949 :
950 : /*
951 : * Even if there's no free clusters available (fragmented),
952 : * try to scan a little more quickly with lock held unless we
953 : * have scanned too many slots already.
954 : */
955 0 : if (!scanned_many) {
956 : unsigned long scan_limit;
957 :
958 0 : if (offset < scan_base)
959 : scan_limit = scan_base;
960 : else
961 0 : scan_limit = si->highest_bit;
962 0 : for (; offset <= scan_limit && --latency_ration > 0;
963 0 : offset++) {
964 0 : if (!si->swap_map[offset])
965 : goto checks;
966 : }
967 : }
968 :
969 : done:
970 0 : set_cluster_next(si, offset + 1);
971 0 : si->flags -= SWP_SCANNING;
972 0 : return n_ret;
973 :
974 : scan:
975 0 : spin_unlock(&si->lock);
976 0 : while (++offset <= READ_ONCE(si->highest_bit)) {
977 0 : if (unlikely(--latency_ration < 0)) {
978 0 : cond_resched();
979 0 : latency_ration = LATENCY_LIMIT;
980 0 : scanned_many = true;
981 : }
982 0 : if (swap_offset_available_and_locked(si, offset))
983 : goto checks;
984 : }
985 0 : offset = si->lowest_bit;
986 0 : while (offset < scan_base) {
987 0 : if (unlikely(--latency_ration < 0)) {
988 0 : cond_resched();
989 0 : latency_ration = LATENCY_LIMIT;
990 0 : scanned_many = true;
991 : }
992 0 : if (swap_offset_available_and_locked(si, offset))
993 : goto checks;
994 0 : offset++;
995 : }
996 0 : spin_lock(&si->lock);
997 :
998 : no_page:
999 0 : si->flags -= SWP_SCANNING;
1000 0 : return n_ret;
1001 : }
1002 :
1003 : static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot)
1004 : {
1005 : unsigned long idx;
1006 : struct swap_cluster_info *ci;
1007 : unsigned long offset;
1008 :
1009 : /*
1010 : * Should not even be attempting cluster allocations when huge
1011 : * page swap is disabled. Warn and fail the allocation.
1012 : */
1013 : if (!IS_ENABLED(CONFIG_THP_SWAP)) {
1014 : VM_WARN_ON_ONCE(1);
1015 : return 0;
1016 : }
1017 :
1018 : if (cluster_list_empty(&si->free_clusters))
1019 : return 0;
1020 :
1021 : idx = cluster_list_first(&si->free_clusters);
1022 : offset = idx * SWAPFILE_CLUSTER;
1023 : ci = lock_cluster(si, offset);
1024 : alloc_cluster(si, idx);
1025 : cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE);
1026 :
1027 : memset(si->swap_map + offset, SWAP_HAS_CACHE, SWAPFILE_CLUSTER);
1028 : unlock_cluster(ci);
1029 : swap_range_alloc(si, offset, SWAPFILE_CLUSTER);
1030 : *slot = swp_entry(si->type, offset);
1031 :
1032 : return 1;
1033 : }
1034 :
1035 : static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx)
1036 : {
1037 : unsigned long offset = idx * SWAPFILE_CLUSTER;
1038 : struct swap_cluster_info *ci;
1039 :
1040 : ci = lock_cluster(si, offset);
1041 : memset(si->swap_map + offset, 0, SWAPFILE_CLUSTER);
1042 : cluster_set_count_flag(ci, 0, 0);
1043 : free_cluster(si, idx);
1044 : unlock_cluster(ci);
1045 : swap_range_free(si, offset, SWAPFILE_CLUSTER);
1046 : }
1047 :
1048 0 : int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size)
1049 : {
1050 0 : unsigned long size = swap_entry_size(entry_size);
1051 : struct swap_info_struct *si, *next;
1052 : long avail_pgs;
1053 0 : int n_ret = 0;
1054 : int node;
1055 :
1056 : /* Only single cluster request supported */
1057 0 : WARN_ON_ONCE(n_goal > 1 && size == SWAPFILE_CLUSTER);
1058 :
1059 0 : spin_lock(&swap_avail_lock);
1060 :
1061 0 : avail_pgs = atomic_long_read(&nr_swap_pages) / size;
1062 0 : if (avail_pgs <= 0) {
1063 : spin_unlock(&swap_avail_lock);
1064 : goto noswap;
1065 : }
1066 :
1067 0 : n_goal = min3((long)n_goal, (long)SWAP_BATCH, avail_pgs);
1068 :
1069 0 : atomic_long_sub(n_goal * size, &nr_swap_pages);
1070 :
1071 : start_over:
1072 0 : node = numa_node_id();
1073 0 : plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) {
1074 : /* requeue si to after same-priority siblings */
1075 0 : plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]);
1076 0 : spin_unlock(&swap_avail_lock);
1077 0 : spin_lock(&si->lock);
1078 0 : if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) {
1079 0 : spin_lock(&swap_avail_lock);
1080 0 : if (plist_node_empty(&si->avail_lists[node])) {
1081 0 : spin_unlock(&si->lock);
1082 : goto nextsi;
1083 : }
1084 0 : WARN(!si->highest_bit,
1085 : "swap_info %d in list but !highest_bit\n",
1086 : si->type);
1087 0 : WARN(!(si->flags & SWP_WRITEOK),
1088 : "swap_info %d in list but !SWP_WRITEOK\n",
1089 : si->type);
1090 0 : __del_from_avail_list(si);
1091 0 : spin_unlock(&si->lock);
1092 : goto nextsi;
1093 : }
1094 : if (size == SWAPFILE_CLUSTER) {
1095 : if (si->flags & SWP_BLKDEV)
1096 : n_ret = swap_alloc_cluster(si, swp_entries);
1097 : } else
1098 0 : n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE,
1099 : n_goal, swp_entries);
1100 0 : spin_unlock(&si->lock);
1101 0 : if (n_ret || size == SWAPFILE_CLUSTER)
1102 : goto check_out;
1103 0 : cond_resched();
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 0 : 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 0 : if (n_ret < n_goal)
1126 0 : atomic_long_add((long)(n_goal - n_ret) * size,
1127 : &nr_swap_pages);
1128 : noswap:
1129 0 : return n_ret;
1130 : }
1131 :
1132 0 : 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 0 : if (!entry.val)
1138 : goto out;
1139 0 : p = swp_swap_info(entry);
1140 0 : if (!p)
1141 : goto bad_nofile;
1142 0 : if (data_race(!(p->flags & SWP_USED)))
1143 : goto bad_device;
1144 0 : offset = swp_offset(entry);
1145 0 : if (offset >= p->max)
1146 : goto bad_offset;
1147 0 : if (data_race(!p->swap_map[swp_offset(entry)]))
1148 : goto bad_free;
1149 : return p;
1150 :
1151 : bad_free:
1152 0 : pr_err("%s: %s%08lx\n", __func__, Unused_offset, entry.val);
1153 0 : goto out;
1154 : bad_offset:
1155 0 : pr_err("%s: %s%08lx\n", __func__, Bad_offset, entry.val);
1156 0 : goto out;
1157 : bad_device:
1158 0 : pr_err("%s: %s%08lx\n", __func__, Unused_file, entry.val);
1159 0 : goto out;
1160 : bad_nofile:
1161 0 : pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val);
1162 : out:
1163 : return NULL;
1164 : }
1165 :
1166 : static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry,
1167 : struct swap_info_struct *q)
1168 : {
1169 : struct swap_info_struct *p;
1170 :
1171 0 : p = _swap_info_get(entry);
1172 :
1173 0 : if (p != q) {
1174 0 : if (q != NULL)
1175 0 : spin_unlock(&q->lock);
1176 0 : if (p != NULL)
1177 0 : spin_lock(&p->lock);
1178 : }
1179 : return p;
1180 : }
1181 :
1182 0 : 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 0 : count = p->swap_map[offset];
1190 :
1191 0 : has_cache = count & SWAP_HAS_CACHE;
1192 0 : count &= ~SWAP_HAS_CACHE;
1193 :
1194 0 : if (usage == SWAP_HAS_CACHE) {
1195 : VM_BUG_ON(!has_cache);
1196 : has_cache = 0;
1197 0 : } 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 0 : } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
1204 0 : if (count == COUNT_CONTINUED) {
1205 0 : if (swap_count_continued(p, offset, count))
1206 : count = SWAP_MAP_MAX | COUNT_CONTINUED;
1207 : else
1208 0 : count = SWAP_MAP_MAX;
1209 : } else
1210 0 : count--;
1211 : }
1212 :
1213 0 : usage = count | has_cache;
1214 0 : if (usage)
1215 0 : WRITE_ONCE(p->swap_map[offset], usage);
1216 : else
1217 0 : WRITE_ONCE(p->swap_map[offset], SWAP_HAS_CACHE);
1218 :
1219 0 : return usage;
1220 : }
1221 :
1222 : /*
1223 : * When we get a swap entry, if there aren't some other ways to
1224 : * prevent swapoff, such as the folio in swap cache is locked, page
1225 : * table lock is held, etc., the swap entry may become invalid because
1226 : * of swapoff. Then, we need to enclose all swap related functions
1227 : * with get_swap_device() and put_swap_device(), unless the swap
1228 : * functions call get/put_swap_device() by themselves.
1229 : *
1230 : * Check whether swap entry is valid in the swap device. If so,
1231 : * return pointer to swap_info_struct, and keep the swap entry valid
1232 : * via preventing the swap device from being swapoff, until
1233 : * put_swap_device() is called. Otherwise return NULL.
1234 : *
1235 : * Notice that swapoff or swapoff+swapon can still happen before the
1236 : * percpu_ref_tryget_live() in get_swap_device() or after the
1237 : * percpu_ref_put() in put_swap_device() if there isn't any other way
1238 : * to prevent swapoff. The caller must be prepared for that. For
1239 : * example, the following situation is possible.
1240 : *
1241 : * CPU1 CPU2
1242 : * do_swap_page()
1243 : * ... swapoff+swapon
1244 : * __read_swap_cache_async()
1245 : * swapcache_prepare()
1246 : * __swap_duplicate()
1247 : * // check swap_map
1248 : * // verify PTE not changed
1249 : *
1250 : * In __swap_duplicate(), the swap_map need to be checked before
1251 : * changing partly because the specified swap entry may be for another
1252 : * swap device which has been swapoff. And in do_swap_page(), after
1253 : * the page is read from the swap device, the PTE is verified not
1254 : * changed with the page table locked to check whether the swap device
1255 : * has been swapoff or swapoff+swapon.
1256 : */
1257 0 : struct swap_info_struct *get_swap_device(swp_entry_t entry)
1258 : {
1259 : struct swap_info_struct *si;
1260 : unsigned long offset;
1261 :
1262 0 : if (!entry.val)
1263 : goto out;
1264 0 : si = swp_swap_info(entry);
1265 0 : if (!si)
1266 : goto bad_nofile;
1267 0 : if (!percpu_ref_tryget_live(&si->users))
1268 : goto out;
1269 : /*
1270 : * Guarantee the si->users are checked before accessing other
1271 : * fields of swap_info_struct.
1272 : *
1273 : * Paired with the spin_unlock() after setup_swap_info() in
1274 : * enable_swap_info().
1275 : */
1276 0 : smp_rmb();
1277 0 : offset = swp_offset(entry);
1278 0 : if (offset >= si->max)
1279 : goto put_out;
1280 :
1281 : return si;
1282 : bad_nofile:
1283 0 : pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val);
1284 : out:
1285 : return NULL;
1286 : put_out:
1287 0 : pr_err("%s: %s%08lx\n", __func__, Bad_offset, entry.val);
1288 0 : percpu_ref_put(&si->users);
1289 0 : return NULL;
1290 : }
1291 :
1292 0 : static unsigned char __swap_entry_free(struct swap_info_struct *p,
1293 : swp_entry_t entry)
1294 : {
1295 : struct swap_cluster_info *ci;
1296 0 : unsigned long offset = swp_offset(entry);
1297 : unsigned char usage;
1298 :
1299 0 : ci = lock_cluster_or_swap_info(p, offset);
1300 0 : usage = __swap_entry_free_locked(p, offset, 1);
1301 0 : unlock_cluster_or_swap_info(p, ci);
1302 0 : if (!usage)
1303 0 : free_swap_slot(entry);
1304 :
1305 0 : return usage;
1306 : }
1307 :
1308 0 : static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry)
1309 : {
1310 : struct swap_cluster_info *ci;
1311 0 : unsigned long offset = swp_offset(entry);
1312 : unsigned char count;
1313 :
1314 0 : ci = lock_cluster(p, offset);
1315 0 : count = p->swap_map[offset];
1316 : VM_BUG_ON(count != SWAP_HAS_CACHE);
1317 0 : p->swap_map[offset] = 0;
1318 0 : dec_cluster_info_page(p, p->cluster_info, offset);
1319 0 : unlock_cluster(ci);
1320 :
1321 0 : mem_cgroup_uncharge_swap(entry, 1);
1322 0 : swap_range_free(p, offset, 1);
1323 0 : }
1324 :
1325 : /*
1326 : * Caller has made sure that the swap device corresponding to entry
1327 : * is still around or has not been recycled.
1328 : */
1329 0 : void swap_free(swp_entry_t entry)
1330 : {
1331 : struct swap_info_struct *p;
1332 :
1333 0 : p = _swap_info_get(entry);
1334 0 : if (p)
1335 0 : __swap_entry_free(p, entry);
1336 0 : }
1337 :
1338 : /*
1339 : * Called after dropping swapcache to decrease refcnt to swap entries.
1340 : */
1341 0 : void put_swap_folio(struct folio *folio, swp_entry_t entry)
1342 : {
1343 0 : unsigned long offset = swp_offset(entry);
1344 0 : unsigned long idx = offset / SWAPFILE_CLUSTER;
1345 : struct swap_cluster_info *ci;
1346 : struct swap_info_struct *si;
1347 : unsigned char *map;
1348 0 : unsigned int i, free_entries = 0;
1349 : unsigned char val;
1350 0 : int size = swap_entry_size(folio_nr_pages(folio));
1351 :
1352 0 : si = _swap_info_get(entry);
1353 0 : if (!si)
1354 : return;
1355 :
1356 : ci = lock_cluster_or_swap_info(si, offset);
1357 : if (size == SWAPFILE_CLUSTER) {
1358 : VM_BUG_ON(!cluster_is_huge(ci));
1359 : map = si->swap_map + offset;
1360 : for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1361 : val = map[i];
1362 : VM_BUG_ON(!(val & SWAP_HAS_CACHE));
1363 : if (val == SWAP_HAS_CACHE)
1364 : free_entries++;
1365 : }
1366 : cluster_clear_huge(ci);
1367 : if (free_entries == SWAPFILE_CLUSTER) {
1368 : unlock_cluster_or_swap_info(si, ci);
1369 : spin_lock(&si->lock);
1370 : mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER);
1371 : swap_free_cluster(si, idx);
1372 : spin_unlock(&si->lock);
1373 : return;
1374 : }
1375 : }
1376 0 : for (i = 0; i < size; i++, entry.val++) {
1377 0 : if (!__swap_entry_free_locked(si, offset + i, SWAP_HAS_CACHE)) {
1378 0 : unlock_cluster_or_swap_info(si, ci);
1379 0 : free_swap_slot(entry);
1380 : if (i == size - 1)
1381 : return;
1382 : lock_cluster_or_swap_info(si, offset);
1383 : }
1384 : }
1385 0 : unlock_cluster_or_swap_info(si, ci);
1386 : }
1387 :
1388 : #ifdef CONFIG_THP_SWAP
1389 : int split_swap_cluster(swp_entry_t entry)
1390 : {
1391 : struct swap_info_struct *si;
1392 : struct swap_cluster_info *ci;
1393 : unsigned long offset = swp_offset(entry);
1394 :
1395 : si = _swap_info_get(entry);
1396 : if (!si)
1397 : return -EBUSY;
1398 : ci = lock_cluster(si, offset);
1399 : cluster_clear_huge(ci);
1400 : unlock_cluster(ci);
1401 : return 0;
1402 : }
1403 : #endif
1404 :
1405 0 : static int swp_entry_cmp(const void *ent1, const void *ent2)
1406 : {
1407 0 : const swp_entry_t *e1 = ent1, *e2 = ent2;
1408 :
1409 0 : return (int)swp_type(*e1) - (int)swp_type(*e2);
1410 : }
1411 :
1412 0 : void swapcache_free_entries(swp_entry_t *entries, int n)
1413 : {
1414 : struct swap_info_struct *p, *prev;
1415 : int i;
1416 :
1417 0 : if (n <= 0)
1418 : return;
1419 :
1420 0 : prev = NULL;
1421 0 : p = NULL;
1422 :
1423 : /*
1424 : * Sort swap entries by swap device, so each lock is only taken once.
1425 : * nr_swapfiles isn't absolutely correct, but the overhead of sort() is
1426 : * so low that it isn't necessary to optimize further.
1427 : */
1428 0 : if (nr_swapfiles > 1)
1429 0 : sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL);
1430 0 : for (i = 0; i < n; ++i) {
1431 0 : p = swap_info_get_cont(entries[i], prev);
1432 0 : if (p)
1433 0 : swap_entry_free(p, entries[i]);
1434 0 : prev = p;
1435 : }
1436 0 : if (p)
1437 0 : spin_unlock(&p->lock);
1438 : }
1439 :
1440 0 : int __swap_count(swp_entry_t entry)
1441 : {
1442 0 : struct swap_info_struct *si = swp_swap_info(entry);
1443 0 : pgoff_t offset = swp_offset(entry);
1444 :
1445 0 : return swap_count(si->swap_map[offset]);
1446 : }
1447 :
1448 : /*
1449 : * How many references to @entry are currently swapped out?
1450 : * This does not give an exact answer when swap count is continued,
1451 : * but does include the high COUNT_CONTINUED flag to allow for that.
1452 : */
1453 0 : int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry)
1454 : {
1455 0 : pgoff_t offset = swp_offset(entry);
1456 : struct swap_cluster_info *ci;
1457 : int count;
1458 :
1459 0 : ci = lock_cluster_or_swap_info(si, offset);
1460 0 : count = swap_count(si->swap_map[offset]);
1461 0 : unlock_cluster_or_swap_info(si, ci);
1462 0 : return count;
1463 : }
1464 :
1465 : /*
1466 : * How many references to @entry are currently swapped out?
1467 : * This considers COUNT_CONTINUED so it returns exact answer.
1468 : */
1469 0 : int swp_swapcount(swp_entry_t entry)
1470 : {
1471 : int count, tmp_count, n;
1472 : struct swap_info_struct *p;
1473 : struct swap_cluster_info *ci;
1474 : struct page *page;
1475 : pgoff_t offset;
1476 : unsigned char *map;
1477 :
1478 0 : p = _swap_info_get(entry);
1479 0 : if (!p)
1480 : return 0;
1481 :
1482 0 : offset = swp_offset(entry);
1483 :
1484 0 : ci = lock_cluster_or_swap_info(p, offset);
1485 :
1486 0 : count = swap_count(p->swap_map[offset]);
1487 0 : if (!(count & COUNT_CONTINUED))
1488 : goto out;
1489 :
1490 0 : count &= ~COUNT_CONTINUED;
1491 0 : n = SWAP_MAP_MAX + 1;
1492 :
1493 0 : page = vmalloc_to_page(p->swap_map + offset);
1494 0 : offset &= ~PAGE_MASK;
1495 : VM_BUG_ON(page_private(page) != SWP_CONTINUED);
1496 :
1497 : do {
1498 0 : page = list_next_entry(page, lru);
1499 0 : map = kmap_atomic(page);
1500 0 : tmp_count = map[offset];
1501 0 : kunmap_atomic(map);
1502 :
1503 0 : count += (tmp_count & ~COUNT_CONTINUED) * n;
1504 0 : n *= (SWAP_CONT_MAX + 1);
1505 0 : } while (tmp_count & COUNT_CONTINUED);
1506 : out:
1507 0 : unlock_cluster_or_swap_info(p, ci);
1508 : return count;
1509 : }
1510 :
1511 : static bool swap_page_trans_huge_swapped(struct swap_info_struct *si,
1512 : swp_entry_t entry)
1513 : {
1514 : struct swap_cluster_info *ci;
1515 0 : unsigned char *map = si->swap_map;
1516 0 : unsigned long roffset = swp_offset(entry);
1517 0 : unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER);
1518 : int i;
1519 0 : bool ret = false;
1520 :
1521 0 : ci = lock_cluster_or_swap_info(si, offset);
1522 : if (!ci || !cluster_is_huge(ci)) {
1523 0 : if (swap_count(map[roffset]))
1524 0 : ret = true;
1525 : goto unlock_out;
1526 : }
1527 : for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1528 : if (swap_count(map[offset + i])) {
1529 : ret = true;
1530 : break;
1531 : }
1532 : }
1533 : unlock_out:
1534 0 : unlock_cluster_or_swap_info(si, ci);
1535 : return ret;
1536 : }
1537 :
1538 0 : static bool folio_swapped(struct folio *folio)
1539 : {
1540 0 : swp_entry_t entry = folio_swap_entry(folio);
1541 0 : struct swap_info_struct *si = _swap_info_get(entry);
1542 :
1543 0 : if (!si)
1544 : return false;
1545 :
1546 : if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!folio_test_large(folio)))
1547 0 : return swap_swapcount(si, entry) != 0;
1548 :
1549 : return swap_page_trans_huge_swapped(si, entry);
1550 : }
1551 :
1552 : /**
1553 : * folio_free_swap() - Free the swap space used for this folio.
1554 : * @folio: The folio to remove.
1555 : *
1556 : * If swap is getting full, or if there are no more mappings of this folio,
1557 : * then call folio_free_swap to free its swap space.
1558 : *
1559 : * Return: true if we were able to release the swap space.
1560 : */
1561 0 : bool folio_free_swap(struct folio *folio)
1562 : {
1563 : VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1564 :
1565 0 : if (!folio_test_swapcache(folio))
1566 : return false;
1567 0 : if (folio_test_writeback(folio))
1568 : return false;
1569 0 : if (folio_swapped(folio))
1570 : return false;
1571 :
1572 : /*
1573 : * Once hibernation has begun to create its image of memory,
1574 : * there's a danger that one of the calls to folio_free_swap()
1575 : * - most probably a call from __try_to_reclaim_swap() while
1576 : * hibernation is allocating its own swap pages for the image,
1577 : * but conceivably even a call from memory reclaim - will free
1578 : * the swap from a folio which has already been recorded in the
1579 : * image as a clean swapcache folio, and then reuse its swap for
1580 : * another page of the image. On waking from hibernation, the
1581 : * original folio might be freed under memory pressure, then
1582 : * later read back in from swap, now with the wrong data.
1583 : *
1584 : * Hibernation suspends storage while it is writing the image
1585 : * to disk so check that here.
1586 : */
1587 0 : if (pm_suspended_storage())
1588 : return false;
1589 :
1590 0 : delete_from_swap_cache(folio);
1591 0 : folio_set_dirty(folio);
1592 0 : return true;
1593 : }
1594 :
1595 : /*
1596 : * Free the swap entry like above, but also try to
1597 : * free the page cache entry if it is the last user.
1598 : */
1599 0 : int free_swap_and_cache(swp_entry_t entry)
1600 : {
1601 : struct swap_info_struct *p;
1602 : unsigned char count;
1603 :
1604 0 : if (non_swap_entry(entry))
1605 : return 1;
1606 :
1607 0 : p = _swap_info_get(entry);
1608 0 : if (p) {
1609 0 : count = __swap_entry_free(p, entry);
1610 0 : if (count == SWAP_HAS_CACHE &&
1611 0 : !swap_page_trans_huge_swapped(p, entry))
1612 0 : __try_to_reclaim_swap(p, swp_offset(entry),
1613 : TTRS_UNMAPPED | TTRS_FULL);
1614 : }
1615 0 : return p != NULL;
1616 : }
1617 :
1618 : #ifdef CONFIG_HIBERNATION
1619 :
1620 : swp_entry_t get_swap_page_of_type(int type)
1621 : {
1622 : struct swap_info_struct *si = swap_type_to_swap_info(type);
1623 : swp_entry_t entry = {0};
1624 :
1625 : if (!si)
1626 : goto fail;
1627 :
1628 : /* This is called for allocating swap entry, not cache */
1629 : spin_lock(&si->lock);
1630 : if ((si->flags & SWP_WRITEOK) && scan_swap_map_slots(si, 1, 1, &entry))
1631 : atomic_long_dec(&nr_swap_pages);
1632 : spin_unlock(&si->lock);
1633 : fail:
1634 : return entry;
1635 : }
1636 :
1637 : /*
1638 : * Find the swap type that corresponds to given device (if any).
1639 : *
1640 : * @offset - number of the PAGE_SIZE-sized block of the device, starting
1641 : * from 0, in which the swap header is expected to be located.
1642 : *
1643 : * This is needed for the suspend to disk (aka swsusp).
1644 : */
1645 : int swap_type_of(dev_t device, sector_t offset)
1646 : {
1647 : int type;
1648 :
1649 : if (!device)
1650 : return -1;
1651 :
1652 : spin_lock(&swap_lock);
1653 : for (type = 0; type < nr_swapfiles; type++) {
1654 : struct swap_info_struct *sis = swap_info[type];
1655 :
1656 : if (!(sis->flags & SWP_WRITEOK))
1657 : continue;
1658 :
1659 : if (device == sis->bdev->bd_dev) {
1660 : struct swap_extent *se = first_se(sis);
1661 :
1662 : if (se->start_block == offset) {
1663 : spin_unlock(&swap_lock);
1664 : return type;
1665 : }
1666 : }
1667 : }
1668 : spin_unlock(&swap_lock);
1669 : return -ENODEV;
1670 : }
1671 :
1672 : int find_first_swap(dev_t *device)
1673 : {
1674 : int type;
1675 :
1676 : spin_lock(&swap_lock);
1677 : for (type = 0; type < nr_swapfiles; type++) {
1678 : struct swap_info_struct *sis = swap_info[type];
1679 :
1680 : if (!(sis->flags & SWP_WRITEOK))
1681 : continue;
1682 : *device = sis->bdev->bd_dev;
1683 : spin_unlock(&swap_lock);
1684 : return type;
1685 : }
1686 : spin_unlock(&swap_lock);
1687 : return -ENODEV;
1688 : }
1689 :
1690 : /*
1691 : * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1692 : * corresponding to given index in swap_info (swap type).
1693 : */
1694 : sector_t swapdev_block(int type, pgoff_t offset)
1695 : {
1696 : struct swap_info_struct *si = swap_type_to_swap_info(type);
1697 : struct swap_extent *se;
1698 :
1699 : if (!si || !(si->flags & SWP_WRITEOK))
1700 : return 0;
1701 : se = offset_to_swap_extent(si, offset);
1702 : return se->start_block + (offset - se->start_page);
1703 : }
1704 :
1705 : /*
1706 : * Return either the total number of swap pages of given type, or the number
1707 : * of free pages of that type (depending on @free)
1708 : *
1709 : * This is needed for software suspend
1710 : */
1711 : unsigned int count_swap_pages(int type, int free)
1712 : {
1713 : unsigned int n = 0;
1714 :
1715 : spin_lock(&swap_lock);
1716 : if ((unsigned int)type < nr_swapfiles) {
1717 : struct swap_info_struct *sis = swap_info[type];
1718 :
1719 : spin_lock(&sis->lock);
1720 : if (sis->flags & SWP_WRITEOK) {
1721 : n = sis->pages;
1722 : if (free)
1723 : n -= sis->inuse_pages;
1724 : }
1725 : spin_unlock(&sis->lock);
1726 : }
1727 : spin_unlock(&swap_lock);
1728 : return n;
1729 : }
1730 : #endif /* CONFIG_HIBERNATION */
1731 :
1732 : static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte)
1733 : {
1734 0 : return pte_same(pte_swp_clear_flags(pte), swp_pte);
1735 : }
1736 :
1737 : /*
1738 : * No need to decide whether this PTE shares the swap entry with others,
1739 : * just let do_wp_page work it out if a write is requested later - to
1740 : * force COW, vm_page_prot omits write permission from any private vma.
1741 : */
1742 0 : static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
1743 : unsigned long addr, swp_entry_t entry, struct folio *folio)
1744 : {
1745 0 : struct page *page = folio_file_page(folio, swp_offset(entry));
1746 : struct page *swapcache;
1747 : spinlock_t *ptl;
1748 : pte_t *pte, new_pte, old_pte;
1749 0 : bool hwposioned = false;
1750 0 : int ret = 1;
1751 :
1752 0 : swapcache = page;
1753 0 : page = ksm_might_need_to_copy(page, vma, addr);
1754 0 : if (unlikely(!page))
1755 : return -ENOMEM;
1756 0 : else if (unlikely(PTR_ERR(page) == -EHWPOISON))
1757 0 : hwposioned = true;
1758 :
1759 0 : pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1760 0 : if (unlikely(!pte || !pte_same_as_swp(ptep_get(pte),
1761 : swp_entry_to_pte(entry)))) {
1762 : ret = 0;
1763 : goto out;
1764 : }
1765 :
1766 0 : old_pte = ptep_get(pte);
1767 :
1768 0 : if (unlikely(hwposioned || !PageUptodate(page))) {
1769 : swp_entry_t swp_entry;
1770 :
1771 0 : dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1772 0 : if (hwposioned) {
1773 : swp_entry = make_hwpoison_entry(swapcache);
1774 : page = swapcache;
1775 : } else {
1776 : swp_entry = make_swapin_error_entry();
1777 : }
1778 0 : new_pte = swp_entry_to_pte(swp_entry);
1779 0 : ret = 0;
1780 : goto setpte;
1781 : }
1782 :
1783 : /* See do_swap_page() */
1784 0 : BUG_ON(!PageAnon(page) && PageMappedToDisk(page));
1785 0 : BUG_ON(PageAnon(page) && PageAnonExclusive(page));
1786 :
1787 0 : dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1788 0 : inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
1789 0 : get_page(page);
1790 : if (page == swapcache) {
1791 0 : rmap_t rmap_flags = RMAP_NONE;
1792 :
1793 : /*
1794 : * See do_swap_page(): PageWriteback() would be problematic.
1795 : * However, we do a wait_on_page_writeback() just before this
1796 : * call and have the page locked.
1797 : */
1798 : VM_BUG_ON_PAGE(PageWriteback(page), page);
1799 0 : if (pte_swp_exclusive(old_pte))
1800 0 : rmap_flags |= RMAP_EXCLUSIVE;
1801 :
1802 0 : page_add_anon_rmap(page, vma, addr, rmap_flags);
1803 : } else { /* ksm created a completely new copy */
1804 : page_add_new_anon_rmap(page, vma, addr);
1805 : lru_cache_add_inactive_or_unevictable(page, vma);
1806 : }
1807 0 : new_pte = pte_mkold(mk_pte(page, vma->vm_page_prot));
1808 0 : if (pte_swp_soft_dirty(old_pte))
1809 : new_pte = pte_mksoft_dirty(new_pte);
1810 : if (pte_swp_uffd_wp(old_pte))
1811 : new_pte = pte_mkuffd_wp(new_pte);
1812 : setpte:
1813 0 : set_pte_at(vma->vm_mm, addr, pte, new_pte);
1814 0 : swap_free(entry);
1815 : out:
1816 0 : if (pte)
1817 0 : pte_unmap_unlock(pte, ptl);
1818 : if (page != swapcache) {
1819 : unlock_page(page);
1820 : put_page(page);
1821 : }
1822 : return ret;
1823 : }
1824 :
1825 0 : static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1826 : unsigned long addr, unsigned long end,
1827 : unsigned int type)
1828 : {
1829 0 : pte_t *pte = NULL;
1830 : struct swap_info_struct *si;
1831 :
1832 0 : si = swap_info[type];
1833 : do {
1834 : struct folio *folio;
1835 : unsigned long offset;
1836 : unsigned char swp_count;
1837 : swp_entry_t entry;
1838 : int ret;
1839 : pte_t ptent;
1840 :
1841 0 : if (!pte++) {
1842 0 : pte = pte_offset_map(pmd, addr);
1843 0 : if (!pte)
1844 : break;
1845 : }
1846 :
1847 0 : ptent = ptep_get_lockless(pte);
1848 :
1849 0 : if (!is_swap_pte(ptent))
1850 0 : continue;
1851 :
1852 0 : entry = pte_to_swp_entry(ptent);
1853 0 : if (swp_type(entry) != type)
1854 0 : continue;
1855 :
1856 0 : offset = swp_offset(entry);
1857 0 : pte_unmap(pte);
1858 0 : pte = NULL;
1859 :
1860 0 : folio = swap_cache_get_folio(entry, vma, addr);
1861 0 : if (!folio) {
1862 : struct page *page;
1863 0 : struct vm_fault vmf = {
1864 : .vma = vma,
1865 : .address = addr,
1866 : .real_address = addr,
1867 : .pmd = pmd,
1868 : };
1869 :
1870 0 : page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE,
1871 : &vmf);
1872 0 : if (page)
1873 0 : folio = page_folio(page);
1874 : }
1875 0 : if (!folio) {
1876 0 : swp_count = READ_ONCE(si->swap_map[offset]);
1877 0 : if (swp_count == 0 || swp_count == SWAP_MAP_BAD)
1878 0 : continue;
1879 : return -ENOMEM;
1880 : }
1881 :
1882 0 : folio_lock(folio);
1883 0 : folio_wait_writeback(folio);
1884 0 : ret = unuse_pte(vma, pmd, addr, entry, folio);
1885 0 : if (ret < 0) {
1886 0 : folio_unlock(folio);
1887 : folio_put(folio);
1888 : return ret;
1889 : }
1890 :
1891 0 : folio_free_swap(folio);
1892 0 : folio_unlock(folio);
1893 : folio_put(folio);
1894 0 : } while (addr += PAGE_SIZE, addr != end);
1895 :
1896 : if (pte)
1897 : pte_unmap(pte);
1898 : return 0;
1899 : }
1900 :
1901 0 : static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1902 : unsigned long addr, unsigned long end,
1903 : unsigned int type)
1904 : {
1905 : pmd_t *pmd;
1906 : unsigned long next;
1907 : int ret;
1908 :
1909 0 : pmd = pmd_offset(pud, addr);
1910 : do {
1911 0 : cond_resched();
1912 0 : next = pmd_addr_end(addr, end);
1913 0 : ret = unuse_pte_range(vma, pmd, addr, next, type);
1914 0 : if (ret)
1915 : return ret;
1916 0 : } while (pmd++, addr = next, addr != end);
1917 : return 0;
1918 : }
1919 :
1920 0 : static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d,
1921 : unsigned long addr, unsigned long end,
1922 : unsigned int type)
1923 : {
1924 : pud_t *pud;
1925 : unsigned long next;
1926 : int ret;
1927 :
1928 0 : pud = pud_offset(p4d, addr);
1929 : do {
1930 0 : next = pud_addr_end(addr, end);
1931 0 : if (pud_none_or_clear_bad(pud))
1932 0 : continue;
1933 0 : ret = unuse_pmd_range(vma, pud, addr, next, type);
1934 0 : if (ret)
1935 : return ret;
1936 0 : } while (pud++, addr = next, addr != end);
1937 0 : return 0;
1938 : }
1939 :
1940 : static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd,
1941 : unsigned long addr, unsigned long end,
1942 : unsigned int type)
1943 : {
1944 : p4d_t *p4d;
1945 : unsigned long next;
1946 : int ret;
1947 :
1948 0 : p4d = p4d_offset(pgd, addr);
1949 : do {
1950 0 : next = p4d_addr_end(addr, end);
1951 0 : if (p4d_none_or_clear_bad(p4d))
1952 : continue;
1953 0 : ret = unuse_pud_range(vma, p4d, addr, next, type);
1954 0 : if (ret)
1955 : return ret;
1956 0 : } while (p4d++, addr = next, addr != end);
1957 : return 0;
1958 : }
1959 :
1960 0 : static int unuse_vma(struct vm_area_struct *vma, unsigned int type)
1961 : {
1962 : pgd_t *pgd;
1963 : unsigned long addr, end, next;
1964 : int ret;
1965 :
1966 0 : addr = vma->vm_start;
1967 0 : end = vma->vm_end;
1968 :
1969 0 : pgd = pgd_offset(vma->vm_mm, addr);
1970 : do {
1971 0 : next = pgd_addr_end(addr, end);
1972 0 : if (pgd_none_or_clear_bad(pgd))
1973 : continue;
1974 0 : ret = unuse_p4d_range(vma, pgd, addr, next, type);
1975 0 : if (ret)
1976 : return ret;
1977 0 : } while (pgd++, addr = next, addr != end);
1978 : return 0;
1979 : }
1980 :
1981 0 : static int unuse_mm(struct mm_struct *mm, unsigned int type)
1982 : {
1983 : struct vm_area_struct *vma;
1984 0 : int ret = 0;
1985 0 : VMA_ITERATOR(vmi, mm, 0);
1986 :
1987 : mmap_read_lock(mm);
1988 0 : for_each_vma(vmi, vma) {
1989 0 : if (vma->anon_vma) {
1990 0 : ret = unuse_vma(vma, type);
1991 0 : if (ret)
1992 : break;
1993 : }
1994 :
1995 0 : cond_resched();
1996 : }
1997 0 : mmap_read_unlock(mm);
1998 0 : return ret;
1999 : }
2000 :
2001 : /*
2002 : * Scan swap_map from current position to next entry still in use.
2003 : * Return 0 if there are no inuse entries after prev till end of
2004 : * the map.
2005 : */
2006 0 : static unsigned int find_next_to_unuse(struct swap_info_struct *si,
2007 : unsigned int prev)
2008 : {
2009 : unsigned int i;
2010 : unsigned char count;
2011 :
2012 : /*
2013 : * No need for swap_lock here: we're just looking
2014 : * for whether an entry is in use, not modifying it; false
2015 : * hits are okay, and sys_swapoff() has already prevented new
2016 : * allocations from this area (while holding swap_lock).
2017 : */
2018 0 : for (i = prev + 1; i < si->max; i++) {
2019 0 : count = READ_ONCE(si->swap_map[i]);
2020 0 : if (count && swap_count(count) != SWAP_MAP_BAD)
2021 : break;
2022 0 : if ((i % LATENCY_LIMIT) == 0)
2023 0 : cond_resched();
2024 : }
2025 :
2026 0 : if (i == si->max)
2027 0 : i = 0;
2028 :
2029 0 : return i;
2030 : }
2031 :
2032 0 : static int try_to_unuse(unsigned int type)
2033 : {
2034 : struct mm_struct *prev_mm;
2035 : struct mm_struct *mm;
2036 : struct list_head *p;
2037 0 : int retval = 0;
2038 0 : struct swap_info_struct *si = swap_info[type];
2039 : struct folio *folio;
2040 : swp_entry_t entry;
2041 : unsigned int i;
2042 :
2043 0 : if (!READ_ONCE(si->inuse_pages))
2044 : return 0;
2045 :
2046 : retry:
2047 0 : retval = shmem_unuse(type);
2048 0 : if (retval)
2049 : return retval;
2050 :
2051 0 : prev_mm = &init_mm;
2052 0 : mmget(prev_mm);
2053 :
2054 0 : spin_lock(&mmlist_lock);
2055 0 : p = &init_mm.mmlist;
2056 0 : while (READ_ONCE(si->inuse_pages) &&
2057 0 : !signal_pending(current) &&
2058 0 : (p = p->next) != &init_mm.mmlist) {
2059 :
2060 0 : mm = list_entry(p, struct mm_struct, mmlist);
2061 0 : if (!mmget_not_zero(mm))
2062 0 : continue;
2063 0 : spin_unlock(&mmlist_lock);
2064 0 : mmput(prev_mm);
2065 0 : prev_mm = mm;
2066 0 : retval = unuse_mm(mm, type);
2067 0 : if (retval) {
2068 0 : mmput(prev_mm);
2069 0 : return retval;
2070 : }
2071 :
2072 : /*
2073 : * Make sure that we aren't completely killing
2074 : * interactive performance.
2075 : */
2076 0 : cond_resched();
2077 : spin_lock(&mmlist_lock);
2078 : }
2079 0 : spin_unlock(&mmlist_lock);
2080 :
2081 0 : mmput(prev_mm);
2082 :
2083 0 : i = 0;
2084 0 : while (READ_ONCE(si->inuse_pages) &&
2085 0 : !signal_pending(current) &&
2086 : (i = find_next_to_unuse(si, i)) != 0) {
2087 :
2088 0 : entry = swp_entry(type, i);
2089 0 : folio = filemap_get_folio(swap_address_space(entry), i);
2090 0 : if (IS_ERR(folio))
2091 0 : continue;
2092 :
2093 : /*
2094 : * It is conceivable that a racing task removed this folio from
2095 : * swap cache just before we acquired the page lock. The folio
2096 : * might even be back in swap cache on another swap area. But
2097 : * that is okay, folio_free_swap() only removes stale folios.
2098 : */
2099 0 : folio_lock(folio);
2100 0 : folio_wait_writeback(folio);
2101 0 : folio_free_swap(folio);
2102 0 : folio_unlock(folio);
2103 : folio_put(folio);
2104 : }
2105 :
2106 : /*
2107 : * Lets check again to see if there are still swap entries in the map.
2108 : * If yes, we would need to do retry the unuse logic again.
2109 : * Under global memory pressure, swap entries can be reinserted back
2110 : * into process space after the mmlist loop above passes over them.
2111 : *
2112 : * Limit the number of retries? No: when mmget_not_zero()
2113 : * above fails, that mm is likely to be freeing swap from
2114 : * exit_mmap(), which proceeds at its own independent pace;
2115 : * and even shmem_writepage() could have been preempted after
2116 : * folio_alloc_swap(), temporarily hiding that swap. It's easy
2117 : * and robust (though cpu-intensive) just to keep retrying.
2118 : */
2119 0 : if (READ_ONCE(si->inuse_pages)) {
2120 0 : if (!signal_pending(current))
2121 : goto retry;
2122 : return -EINTR;
2123 : }
2124 :
2125 : return 0;
2126 : }
2127 :
2128 : /*
2129 : * After a successful try_to_unuse, if no swap is now in use, we know
2130 : * we can empty the mmlist. swap_lock must be held on entry and exit.
2131 : * Note that mmlist_lock nests inside swap_lock, and an mm must be
2132 : * added to the mmlist just after page_duplicate - before would be racy.
2133 : */
2134 0 : static void drain_mmlist(void)
2135 : {
2136 : struct list_head *p, *next;
2137 : unsigned int type;
2138 :
2139 0 : for (type = 0; type < nr_swapfiles; type++)
2140 0 : if (swap_info[type]->inuse_pages)
2141 : return;
2142 0 : spin_lock(&mmlist_lock);
2143 0 : list_for_each_safe(p, next, &init_mm.mmlist)
2144 0 : list_del_init(p);
2145 : spin_unlock(&mmlist_lock);
2146 : }
2147 :
2148 : /*
2149 : * Free all of a swapdev's extent information
2150 : */
2151 0 : static void destroy_swap_extents(struct swap_info_struct *sis)
2152 : {
2153 0 : while (!RB_EMPTY_ROOT(&sis->swap_extent_root)) {
2154 0 : struct rb_node *rb = sis->swap_extent_root.rb_node;
2155 0 : struct swap_extent *se = rb_entry(rb, struct swap_extent, rb_node);
2156 :
2157 0 : rb_erase(rb, &sis->swap_extent_root);
2158 0 : kfree(se);
2159 : }
2160 :
2161 0 : if (sis->flags & SWP_ACTIVATED) {
2162 0 : struct file *swap_file = sis->swap_file;
2163 0 : struct address_space *mapping = swap_file->f_mapping;
2164 :
2165 0 : sis->flags &= ~SWP_ACTIVATED;
2166 0 : if (mapping->a_ops->swap_deactivate)
2167 0 : mapping->a_ops->swap_deactivate(swap_file);
2168 : }
2169 0 : }
2170 :
2171 : /*
2172 : * Add a block range (and the corresponding page range) into this swapdev's
2173 : * extent tree.
2174 : *
2175 : * This function rather assumes that it is called in ascending page order.
2176 : */
2177 : int
2178 0 : add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
2179 : unsigned long nr_pages, sector_t start_block)
2180 : {
2181 0 : struct rb_node **link = &sis->swap_extent_root.rb_node, *parent = NULL;
2182 : struct swap_extent *se;
2183 : struct swap_extent *new_se;
2184 :
2185 : /*
2186 : * place the new node at the right most since the
2187 : * function is called in ascending page order.
2188 : */
2189 0 : while (*link) {
2190 0 : parent = *link;
2191 0 : link = &parent->rb_right;
2192 : }
2193 :
2194 0 : if (parent) {
2195 0 : se = rb_entry(parent, struct swap_extent, rb_node);
2196 0 : BUG_ON(se->start_page + se->nr_pages != start_page);
2197 0 : if (se->start_block + se->nr_pages == start_block) {
2198 : /* Merge it */
2199 0 : se->nr_pages += nr_pages;
2200 0 : return 0;
2201 : }
2202 : }
2203 :
2204 : /* No merge, insert a new extent. */
2205 0 : new_se = kmalloc(sizeof(*se), GFP_KERNEL);
2206 0 : if (new_se == NULL)
2207 : return -ENOMEM;
2208 0 : new_se->start_page = start_page;
2209 0 : new_se->nr_pages = nr_pages;
2210 0 : new_se->start_block = start_block;
2211 :
2212 0 : rb_link_node(&new_se->rb_node, parent, link);
2213 0 : rb_insert_color(&new_se->rb_node, &sis->swap_extent_root);
2214 0 : return 1;
2215 : }
2216 : EXPORT_SYMBOL_GPL(add_swap_extent);
2217 :
2218 : /*
2219 : * A `swap extent' is a simple thing which maps a contiguous range of pages
2220 : * onto a contiguous range of disk blocks. A rbtree of swap extents is
2221 : * built at swapon time and is then used at swap_writepage/swap_readpage
2222 : * time for locating where on disk a page belongs.
2223 : *
2224 : * If the swapfile is an S_ISBLK block device, a single extent is installed.
2225 : * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2226 : * swap files identically.
2227 : *
2228 : * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2229 : * extent rbtree operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
2230 : * swapfiles are handled *identically* after swapon time.
2231 : *
2232 : * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2233 : * and will parse them into a rbtree, in PAGE_SIZE chunks. If some stray
2234 : * blocks are found which do not fall within the PAGE_SIZE alignment
2235 : * requirements, they are simply tossed out - we will never use those blocks
2236 : * for swapping.
2237 : *
2238 : * For all swap devices we set S_SWAPFILE across the life of the swapon. This
2239 : * prevents users from writing to the swap device, which will corrupt memory.
2240 : *
2241 : * The amount of disk space which a single swap extent represents varies.
2242 : * Typically it is in the 1-4 megabyte range. So we can have hundreds of
2243 : * extents in the rbtree. - akpm.
2244 : */
2245 0 : static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
2246 : {
2247 0 : struct file *swap_file = sis->swap_file;
2248 0 : struct address_space *mapping = swap_file->f_mapping;
2249 0 : struct inode *inode = mapping->host;
2250 : int ret;
2251 :
2252 0 : if (S_ISBLK(inode->i_mode)) {
2253 0 : ret = add_swap_extent(sis, 0, sis->max, 0);
2254 0 : *span = sis->pages;
2255 0 : return ret;
2256 : }
2257 :
2258 0 : if (mapping->a_ops->swap_activate) {
2259 0 : ret = mapping->a_ops->swap_activate(sis, swap_file, span);
2260 0 : if (ret < 0)
2261 : return ret;
2262 0 : sis->flags |= SWP_ACTIVATED;
2263 0 : if ((sis->flags & SWP_FS_OPS) &&
2264 0 : sio_pool_init() != 0) {
2265 0 : destroy_swap_extents(sis);
2266 0 : return -ENOMEM;
2267 : }
2268 : return ret;
2269 : }
2270 :
2271 0 : return generic_swapfile_activate(sis, swap_file, span);
2272 : }
2273 :
2274 : static int swap_node(struct swap_info_struct *p)
2275 : {
2276 : struct block_device *bdev;
2277 :
2278 0 : if (p->bdev)
2279 : bdev = p->bdev;
2280 : else
2281 0 : bdev = p->swap_file->f_inode->i_sb->s_bdev;
2282 :
2283 0 : return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE;
2284 : }
2285 :
2286 0 : static void setup_swap_info(struct swap_info_struct *p, int prio,
2287 : unsigned char *swap_map,
2288 : struct swap_cluster_info *cluster_info)
2289 : {
2290 : int i;
2291 :
2292 0 : if (prio >= 0)
2293 0 : p->prio = prio;
2294 : else
2295 0 : p->prio = --least_priority;
2296 : /*
2297 : * the plist prio is negated because plist ordering is
2298 : * low-to-high, while swap ordering is high-to-low
2299 : */
2300 0 : p->list.prio = -p->prio;
2301 0 : for_each_node(i) {
2302 0 : if (p->prio >= 0)
2303 0 : p->avail_lists[i].prio = -p->prio;
2304 : else {
2305 0 : if (swap_node(p) == i)
2306 0 : p->avail_lists[i].prio = 1;
2307 : else
2308 0 : p->avail_lists[i].prio = -p->prio;
2309 : }
2310 : }
2311 0 : p->swap_map = swap_map;
2312 0 : p->cluster_info = cluster_info;
2313 0 : }
2314 :
2315 0 : static void _enable_swap_info(struct swap_info_struct *p)
2316 : {
2317 0 : p->flags |= SWP_WRITEOK;
2318 0 : atomic_long_add(p->pages, &nr_swap_pages);
2319 0 : total_swap_pages += p->pages;
2320 :
2321 : assert_spin_locked(&swap_lock);
2322 : /*
2323 : * both lists are plists, and thus priority ordered.
2324 : * swap_active_head needs to be priority ordered for swapoff(),
2325 : * which on removal of any swap_info_struct with an auto-assigned
2326 : * (i.e. negative) priority increments the auto-assigned priority
2327 : * of any lower-priority swap_info_structs.
2328 : * swap_avail_head needs to be priority ordered for folio_alloc_swap(),
2329 : * which allocates swap pages from the highest available priority
2330 : * swap_info_struct.
2331 : */
2332 0 : plist_add(&p->list, &swap_active_head);
2333 0 : add_to_avail_list(p);
2334 0 : }
2335 :
2336 0 : static void enable_swap_info(struct swap_info_struct *p, int prio,
2337 : unsigned char *swap_map,
2338 : struct swap_cluster_info *cluster_info,
2339 : unsigned long *frontswap_map)
2340 : {
2341 : if (IS_ENABLED(CONFIG_FRONTSWAP))
2342 : frontswap_init(p->type, frontswap_map);
2343 0 : spin_lock(&swap_lock);
2344 0 : spin_lock(&p->lock);
2345 0 : setup_swap_info(p, prio, swap_map, cluster_info);
2346 0 : spin_unlock(&p->lock);
2347 0 : spin_unlock(&swap_lock);
2348 : /*
2349 : * Finished initializing swap device, now it's safe to reference it.
2350 : */
2351 0 : percpu_ref_resurrect(&p->users);
2352 0 : spin_lock(&swap_lock);
2353 0 : spin_lock(&p->lock);
2354 0 : _enable_swap_info(p);
2355 0 : spin_unlock(&p->lock);
2356 0 : spin_unlock(&swap_lock);
2357 0 : }
2358 :
2359 0 : static void reinsert_swap_info(struct swap_info_struct *p)
2360 : {
2361 0 : spin_lock(&swap_lock);
2362 0 : spin_lock(&p->lock);
2363 0 : setup_swap_info(p, p->prio, p->swap_map, p->cluster_info);
2364 0 : _enable_swap_info(p);
2365 0 : spin_unlock(&p->lock);
2366 0 : spin_unlock(&swap_lock);
2367 0 : }
2368 :
2369 0 : bool has_usable_swap(void)
2370 : {
2371 0 : bool ret = true;
2372 :
2373 0 : spin_lock(&swap_lock);
2374 0 : if (plist_head_empty(&swap_active_head))
2375 0 : ret = false;
2376 0 : spin_unlock(&swap_lock);
2377 0 : return ret;
2378 : }
2379 :
2380 0 : SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
2381 : {
2382 0 : struct swap_info_struct *p = NULL;
2383 : unsigned char *swap_map;
2384 : struct swap_cluster_info *cluster_info;
2385 : unsigned long *frontswap_map;
2386 : struct file *swap_file, *victim;
2387 : struct address_space *mapping;
2388 : struct inode *inode;
2389 : struct filename *pathname;
2390 0 : int err, found = 0;
2391 : unsigned int old_block_size;
2392 :
2393 0 : if (!capable(CAP_SYS_ADMIN))
2394 : return -EPERM;
2395 :
2396 0 : BUG_ON(!current->mm);
2397 :
2398 0 : pathname = getname(specialfile);
2399 0 : if (IS_ERR(pathname))
2400 0 : return PTR_ERR(pathname);
2401 :
2402 0 : victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
2403 0 : err = PTR_ERR(victim);
2404 0 : if (IS_ERR(victim))
2405 : goto out;
2406 :
2407 0 : mapping = victim->f_mapping;
2408 0 : spin_lock(&swap_lock);
2409 0 : plist_for_each_entry(p, &swap_active_head, list) {
2410 0 : if (p->flags & SWP_WRITEOK) {
2411 0 : if (p->swap_file->f_mapping == mapping) {
2412 : found = 1;
2413 : break;
2414 : }
2415 : }
2416 : }
2417 0 : if (!found) {
2418 0 : err = -EINVAL;
2419 : spin_unlock(&swap_lock);
2420 : goto out_dput;
2421 : }
2422 0 : if (!security_vm_enough_memory_mm(current->mm, p->pages))
2423 0 : vm_unacct_memory(p->pages);
2424 : else {
2425 0 : err = -ENOMEM;
2426 : spin_unlock(&swap_lock);
2427 : goto out_dput;
2428 : }
2429 0 : spin_lock(&p->lock);
2430 0 : del_from_avail_list(p);
2431 0 : if (p->prio < 0) {
2432 0 : struct swap_info_struct *si = p;
2433 : int nid;
2434 :
2435 0 : plist_for_each_entry_continue(si, &swap_active_head, list) {
2436 0 : si->prio++;
2437 0 : si->list.prio--;
2438 0 : for_each_node(nid) {
2439 0 : if (si->avail_lists[nid].prio != 1)
2440 0 : si->avail_lists[nid].prio--;
2441 : }
2442 : }
2443 0 : least_priority++;
2444 : }
2445 0 : plist_del(&p->list, &swap_active_head);
2446 0 : atomic_long_sub(p->pages, &nr_swap_pages);
2447 0 : total_swap_pages -= p->pages;
2448 0 : p->flags &= ~SWP_WRITEOK;
2449 0 : spin_unlock(&p->lock);
2450 0 : spin_unlock(&swap_lock);
2451 :
2452 0 : disable_swap_slots_cache_lock();
2453 :
2454 0 : set_current_oom_origin();
2455 0 : err = try_to_unuse(p->type);
2456 0 : clear_current_oom_origin();
2457 :
2458 0 : if (err) {
2459 : /* re-insert swap space back into swap_list */
2460 0 : reinsert_swap_info(p);
2461 0 : reenable_swap_slots_cache_unlock();
2462 0 : goto out_dput;
2463 : }
2464 :
2465 0 : reenable_swap_slots_cache_unlock();
2466 :
2467 : /*
2468 : * Wait for swap operations protected by get/put_swap_device()
2469 : * to complete.
2470 : *
2471 : * We need synchronize_rcu() here to protect the accessing to
2472 : * the swap cache data structure.
2473 : */
2474 0 : percpu_ref_kill(&p->users);
2475 0 : synchronize_rcu();
2476 0 : wait_for_completion(&p->comp);
2477 :
2478 0 : flush_work(&p->discard_work);
2479 :
2480 0 : destroy_swap_extents(p);
2481 0 : if (p->flags & SWP_CONTINUED)
2482 0 : free_swap_count_continuations(p);
2483 :
2484 0 : if (!p->bdev || !bdev_nonrot(p->bdev))
2485 : atomic_dec(&nr_rotate_swap);
2486 :
2487 0 : mutex_lock(&swapon_mutex);
2488 0 : spin_lock(&swap_lock);
2489 0 : spin_lock(&p->lock);
2490 0 : drain_mmlist();
2491 :
2492 : /* wait for anyone still in scan_swap_map_slots */
2493 0 : p->highest_bit = 0; /* cuts scans short */
2494 0 : while (p->flags >= SWP_SCANNING) {
2495 0 : spin_unlock(&p->lock);
2496 0 : spin_unlock(&swap_lock);
2497 0 : schedule_timeout_uninterruptible(1);
2498 0 : spin_lock(&swap_lock);
2499 0 : spin_lock(&p->lock);
2500 : }
2501 :
2502 0 : swap_file = p->swap_file;
2503 0 : old_block_size = p->old_block_size;
2504 0 : p->swap_file = NULL;
2505 0 : p->max = 0;
2506 0 : swap_map = p->swap_map;
2507 0 : p->swap_map = NULL;
2508 0 : cluster_info = p->cluster_info;
2509 0 : p->cluster_info = NULL;
2510 0 : frontswap_map = frontswap_map_get(p);
2511 0 : spin_unlock(&p->lock);
2512 0 : spin_unlock(&swap_lock);
2513 0 : arch_swap_invalidate_area(p->type);
2514 0 : frontswap_invalidate_area(p->type);
2515 0 : frontswap_map_set(p, NULL);
2516 0 : mutex_unlock(&swapon_mutex);
2517 0 : free_percpu(p->percpu_cluster);
2518 0 : p->percpu_cluster = NULL;
2519 0 : free_percpu(p->cluster_next_cpu);
2520 0 : p->cluster_next_cpu = NULL;
2521 0 : vfree(swap_map);
2522 0 : kvfree(cluster_info);
2523 0 : kvfree(frontswap_map);
2524 : /* Destroy swap account information */
2525 0 : swap_cgroup_swapoff(p->type);
2526 0 : exit_swap_address_space(p->type);
2527 :
2528 0 : inode = mapping->host;
2529 0 : if (S_ISBLK(inode->i_mode)) {
2530 0 : struct block_device *bdev = I_BDEV(inode);
2531 :
2532 0 : set_blocksize(bdev, old_block_size);
2533 0 : blkdev_put(bdev, p);
2534 : }
2535 :
2536 0 : inode_lock(inode);
2537 0 : inode->i_flags &= ~S_SWAPFILE;
2538 0 : inode_unlock(inode);
2539 0 : filp_close(swap_file, NULL);
2540 :
2541 : /*
2542 : * Clear the SWP_USED flag after all resources are freed so that swapon
2543 : * can reuse this swap_info in alloc_swap_info() safely. It is ok to
2544 : * not hold p->lock after we cleared its SWP_WRITEOK.
2545 : */
2546 0 : spin_lock(&swap_lock);
2547 0 : p->flags = 0;
2548 0 : spin_unlock(&swap_lock);
2549 :
2550 0 : err = 0;
2551 0 : atomic_inc(&proc_poll_event);
2552 0 : wake_up_interruptible(&proc_poll_wait);
2553 :
2554 : out_dput:
2555 0 : filp_close(victim, NULL);
2556 : out:
2557 0 : putname(pathname);
2558 0 : return err;
2559 : }
2560 :
2561 : #ifdef CONFIG_PROC_FS
2562 0 : static __poll_t swaps_poll(struct file *file, poll_table *wait)
2563 : {
2564 0 : struct seq_file *seq = file->private_data;
2565 :
2566 0 : poll_wait(file, &proc_poll_wait, wait);
2567 :
2568 0 : if (seq->poll_event != atomic_read(&proc_poll_event)) {
2569 0 : seq->poll_event = atomic_read(&proc_poll_event);
2570 0 : return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
2571 : }
2572 :
2573 : return EPOLLIN | EPOLLRDNORM;
2574 : }
2575 :
2576 : /* iterator */
2577 0 : static void *swap_start(struct seq_file *swap, loff_t *pos)
2578 : {
2579 : struct swap_info_struct *si;
2580 : int type;
2581 0 : loff_t l = *pos;
2582 :
2583 0 : mutex_lock(&swapon_mutex);
2584 :
2585 0 : if (!l)
2586 : return SEQ_START_TOKEN;
2587 :
2588 0 : for (type = 0; (si = swap_type_to_swap_info(type)); type++) {
2589 0 : if (!(si->flags & SWP_USED) || !si->swap_map)
2590 0 : continue;
2591 0 : if (!--l)
2592 : return si;
2593 : }
2594 :
2595 : return NULL;
2596 : }
2597 :
2598 0 : static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2599 : {
2600 0 : struct swap_info_struct *si = v;
2601 : int type;
2602 :
2603 0 : if (v == SEQ_START_TOKEN)
2604 : type = 0;
2605 : else
2606 0 : type = si->type + 1;
2607 :
2608 0 : ++(*pos);
2609 0 : for (; (si = swap_type_to_swap_info(type)); type++) {
2610 0 : if (!(si->flags & SWP_USED) || !si->swap_map)
2611 0 : continue;
2612 : return si;
2613 : }
2614 :
2615 : return NULL;
2616 : }
2617 :
2618 0 : static void swap_stop(struct seq_file *swap, void *v)
2619 : {
2620 0 : mutex_unlock(&swapon_mutex);
2621 0 : }
2622 :
2623 0 : static int swap_show(struct seq_file *swap, void *v)
2624 : {
2625 0 : struct swap_info_struct *si = v;
2626 : struct file *file;
2627 : int len;
2628 : unsigned long bytes, inuse;
2629 :
2630 0 : if (si == SEQ_START_TOKEN) {
2631 0 : seq_puts(swap, "Filename\t\t\t\tType\t\tSize\t\tUsed\t\tPriority\n");
2632 0 : return 0;
2633 : }
2634 :
2635 0 : bytes = si->pages << (PAGE_SHIFT - 10);
2636 0 : inuse = READ_ONCE(si->inuse_pages) << (PAGE_SHIFT - 10);
2637 :
2638 0 : file = si->swap_file;
2639 0 : len = seq_file_path(swap, file, " \t\n\\");
2640 0 : seq_printf(swap, "%*s%s\t%lu\t%s%lu\t%s%d\n",
2641 : len < 40 ? 40 - len : 1, " ",
2642 0 : S_ISBLK(file_inode(file)->i_mode) ?
2643 : "partition" : "file\t",
2644 : bytes, bytes < 10000000 ? "\t" : "",
2645 : inuse, inuse < 10000000 ? "\t" : "",
2646 0 : si->prio);
2647 0 : return 0;
2648 : }
2649 :
2650 : static const struct seq_operations swaps_op = {
2651 : .start = swap_start,
2652 : .next = swap_next,
2653 : .stop = swap_stop,
2654 : .show = swap_show
2655 : };
2656 :
2657 0 : static int swaps_open(struct inode *inode, struct file *file)
2658 : {
2659 : struct seq_file *seq;
2660 : int ret;
2661 :
2662 0 : ret = seq_open(file, &swaps_op);
2663 0 : if (ret)
2664 : return ret;
2665 :
2666 0 : seq = file->private_data;
2667 0 : seq->poll_event = atomic_read(&proc_poll_event);
2668 0 : return 0;
2669 : }
2670 :
2671 : static const struct proc_ops swaps_proc_ops = {
2672 : .proc_flags = PROC_ENTRY_PERMANENT,
2673 : .proc_open = swaps_open,
2674 : .proc_read = seq_read,
2675 : .proc_lseek = seq_lseek,
2676 : .proc_release = seq_release,
2677 : .proc_poll = swaps_poll,
2678 : };
2679 :
2680 1 : static int __init procswaps_init(void)
2681 : {
2682 1 : proc_create("swaps", 0, NULL, &swaps_proc_ops);
2683 1 : return 0;
2684 : }
2685 : __initcall(procswaps_init);
2686 : #endif /* CONFIG_PROC_FS */
2687 :
2688 : #ifdef MAX_SWAPFILES_CHECK
2689 : static int __init max_swapfiles_check(void)
2690 : {
2691 : MAX_SWAPFILES_CHECK();
2692 : return 0;
2693 : }
2694 : late_initcall(max_swapfiles_check);
2695 : #endif
2696 :
2697 0 : static struct swap_info_struct *alloc_swap_info(void)
2698 : {
2699 : struct swap_info_struct *p;
2700 0 : struct swap_info_struct *defer = NULL;
2701 : unsigned int type;
2702 : int i;
2703 :
2704 0 : p = kvzalloc(struct_size(p, avail_lists, nr_node_ids), GFP_KERNEL);
2705 0 : if (!p)
2706 : return ERR_PTR(-ENOMEM);
2707 :
2708 0 : if (percpu_ref_init(&p->users, swap_users_ref_free,
2709 : PERCPU_REF_INIT_DEAD, GFP_KERNEL)) {
2710 0 : kvfree(p);
2711 0 : return ERR_PTR(-ENOMEM);
2712 : }
2713 :
2714 0 : spin_lock(&swap_lock);
2715 0 : for (type = 0; type < nr_swapfiles; type++) {
2716 0 : if (!(swap_info[type]->flags & SWP_USED))
2717 : break;
2718 : }
2719 0 : if (type >= MAX_SWAPFILES) {
2720 0 : spin_unlock(&swap_lock);
2721 0 : percpu_ref_exit(&p->users);
2722 0 : kvfree(p);
2723 0 : return ERR_PTR(-EPERM);
2724 : }
2725 0 : if (type >= nr_swapfiles) {
2726 0 : p->type = type;
2727 : /*
2728 : * Publish the swap_info_struct after initializing it.
2729 : * Note that kvzalloc() above zeroes all its fields.
2730 : */
2731 0 : smp_store_release(&swap_info[type], p); /* rcu_assign_pointer() */
2732 0 : nr_swapfiles++;
2733 : } else {
2734 0 : defer = p;
2735 0 : p = swap_info[type];
2736 : /*
2737 : * Do not memset this entry: a racing procfs swap_next()
2738 : * would be relying on p->type to remain valid.
2739 : */
2740 : }
2741 0 : p->swap_extent_root = RB_ROOT;
2742 0 : plist_node_init(&p->list, 0);
2743 0 : for_each_node(i)
2744 0 : plist_node_init(&p->avail_lists[i], 0);
2745 0 : p->flags = SWP_USED;
2746 0 : spin_unlock(&swap_lock);
2747 0 : if (defer) {
2748 0 : percpu_ref_exit(&defer->users);
2749 0 : kvfree(defer);
2750 : }
2751 0 : spin_lock_init(&p->lock);
2752 0 : spin_lock_init(&p->cont_lock);
2753 0 : init_completion(&p->comp);
2754 :
2755 0 : return p;
2756 : }
2757 :
2758 0 : static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2759 : {
2760 : int error;
2761 :
2762 0 : if (S_ISBLK(inode->i_mode)) {
2763 0 : p->bdev = blkdev_get_by_dev(inode->i_rdev,
2764 : BLK_OPEN_READ | BLK_OPEN_WRITE, p, NULL);
2765 0 : if (IS_ERR(p->bdev)) {
2766 0 : error = PTR_ERR(p->bdev);
2767 0 : p->bdev = NULL;
2768 0 : return error;
2769 : }
2770 0 : p->old_block_size = block_size(p->bdev);
2771 0 : error = set_blocksize(p->bdev, PAGE_SIZE);
2772 0 : if (error < 0)
2773 : return error;
2774 : /*
2775 : * Zoned block devices contain zones that have a sequential
2776 : * write only restriction. Hence zoned block devices are not
2777 : * suitable for swapping. Disallow them here.
2778 : */
2779 0 : if (bdev_is_zoned(p->bdev))
2780 : return -EINVAL;
2781 0 : p->flags |= SWP_BLKDEV;
2782 0 : } else if (S_ISREG(inode->i_mode)) {
2783 0 : p->bdev = inode->i_sb->s_bdev;
2784 : }
2785 :
2786 : return 0;
2787 : }
2788 :
2789 :
2790 : /*
2791 : * Find out how many pages are allowed for a single swap device. There
2792 : * are two limiting factors:
2793 : * 1) the number of bits for the swap offset in the swp_entry_t type, and
2794 : * 2) the number of bits in the swap pte, as defined by the different
2795 : * architectures.
2796 : *
2797 : * In order to find the largest possible bit mask, a swap entry with
2798 : * swap type 0 and swap offset ~0UL is created, encoded to a swap pte,
2799 : * decoded to a swp_entry_t again, and finally the swap offset is
2800 : * extracted.
2801 : *
2802 : * This will mask all the bits from the initial ~0UL mask that can't
2803 : * be encoded in either the swp_entry_t or the architecture definition
2804 : * of a swap pte.
2805 : */
2806 0 : unsigned long generic_max_swapfile_size(void)
2807 : {
2808 3 : return swp_offset(pte_to_swp_entry(
2809 0 : swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2810 : }
2811 :
2812 : /* Can be overridden by an architecture for additional checks. */
2813 1 : __weak unsigned long arch_max_swapfile_size(void)
2814 : {
2815 1 : return generic_max_swapfile_size();
2816 : }
2817 :
2818 0 : static unsigned long read_swap_header(struct swap_info_struct *p,
2819 : union swap_header *swap_header,
2820 : struct inode *inode)
2821 : {
2822 : int i;
2823 : unsigned long maxpages;
2824 : unsigned long swapfilepages;
2825 : unsigned long last_page;
2826 :
2827 0 : if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
2828 0 : pr_err("Unable to find swap-space signature\n");
2829 0 : return 0;
2830 : }
2831 :
2832 : /* swap partition endianness hack... */
2833 0 : if (swab32(swap_header->info.version) == 1) {
2834 0 : swab32s(&swap_header->info.version);
2835 0 : swab32s(&swap_header->info.last_page);
2836 0 : swab32s(&swap_header->info.nr_badpages);
2837 0 : if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2838 : return 0;
2839 0 : for (i = 0; i < swap_header->info.nr_badpages; i++)
2840 0 : swab32s(&swap_header->info.badpages[i]);
2841 : }
2842 : /* Check the swap header's sub-version */
2843 0 : if (swap_header->info.version != 1) {
2844 0 : pr_warn("Unable to handle swap header version %d\n",
2845 : swap_header->info.version);
2846 0 : return 0;
2847 : }
2848 :
2849 0 : p->lowest_bit = 1;
2850 0 : p->cluster_next = 1;
2851 0 : p->cluster_nr = 0;
2852 :
2853 0 : maxpages = swapfile_maximum_size;
2854 0 : last_page = swap_header->info.last_page;
2855 0 : if (!last_page) {
2856 0 : pr_warn("Empty swap-file\n");
2857 0 : return 0;
2858 : }
2859 0 : if (last_page > maxpages) {
2860 0 : pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
2861 : maxpages << (PAGE_SHIFT - 10),
2862 : last_page << (PAGE_SHIFT - 10));
2863 : }
2864 0 : if (maxpages > last_page) {
2865 0 : maxpages = last_page + 1;
2866 : /* p->max is an unsigned int: don't overflow it */
2867 0 : if ((unsigned int)maxpages == 0)
2868 0 : maxpages = UINT_MAX;
2869 : }
2870 0 : p->highest_bit = maxpages - 1;
2871 :
2872 0 : if (!maxpages)
2873 : return 0;
2874 0 : swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2875 0 : if (swapfilepages && maxpages > swapfilepages) {
2876 0 : pr_warn("Swap area shorter than signature indicates\n");
2877 0 : return 0;
2878 : }
2879 0 : if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
2880 : return 0;
2881 0 : if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2882 : return 0;
2883 :
2884 0 : return maxpages;
2885 : }
2886 :
2887 : #define SWAP_CLUSTER_INFO_COLS \
2888 : DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
2889 : #define SWAP_CLUSTER_SPACE_COLS \
2890 : DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
2891 : #define SWAP_CLUSTER_COLS \
2892 : max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
2893 :
2894 0 : static int setup_swap_map_and_extents(struct swap_info_struct *p,
2895 : union swap_header *swap_header,
2896 : unsigned char *swap_map,
2897 : struct swap_cluster_info *cluster_info,
2898 : unsigned long maxpages,
2899 : sector_t *span)
2900 : {
2901 : unsigned int j, k;
2902 : unsigned int nr_good_pages;
2903 : int nr_extents;
2904 0 : unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
2905 0 : unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS;
2906 : unsigned long i, idx;
2907 :
2908 0 : nr_good_pages = maxpages - 1; /* omit header page */
2909 :
2910 0 : cluster_list_init(&p->free_clusters);
2911 0 : cluster_list_init(&p->discard_clusters);
2912 :
2913 0 : for (i = 0; i < swap_header->info.nr_badpages; i++) {
2914 0 : unsigned int page_nr = swap_header->info.badpages[i];
2915 0 : if (page_nr == 0 || page_nr > swap_header->info.last_page)
2916 : return -EINVAL;
2917 0 : if (page_nr < maxpages) {
2918 0 : swap_map[page_nr] = SWAP_MAP_BAD;
2919 0 : nr_good_pages--;
2920 : /*
2921 : * Haven't marked the cluster free yet, no list
2922 : * operation involved
2923 : */
2924 0 : inc_cluster_info_page(p, cluster_info, page_nr);
2925 : }
2926 : }
2927 :
2928 : /* Haven't marked the cluster free yet, no list operation involved */
2929 0 : for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
2930 0 : inc_cluster_info_page(p, cluster_info, i);
2931 :
2932 0 : if (nr_good_pages) {
2933 0 : swap_map[0] = SWAP_MAP_BAD;
2934 : /*
2935 : * Not mark the cluster free yet, no list
2936 : * operation involved
2937 : */
2938 0 : inc_cluster_info_page(p, cluster_info, 0);
2939 0 : p->max = maxpages;
2940 0 : p->pages = nr_good_pages;
2941 0 : nr_extents = setup_swap_extents(p, span);
2942 0 : if (nr_extents < 0)
2943 : return nr_extents;
2944 0 : nr_good_pages = p->pages;
2945 : }
2946 0 : if (!nr_good_pages) {
2947 0 : pr_warn("Empty swap-file\n");
2948 0 : return -EINVAL;
2949 : }
2950 :
2951 0 : if (!cluster_info)
2952 : return nr_extents;
2953 :
2954 :
2955 : /*
2956 : * Reduce false cache line sharing between cluster_info and
2957 : * sharing same address space.
2958 : */
2959 0 : for (k = 0; k < SWAP_CLUSTER_COLS; k++) {
2960 0 : j = (k + col) % SWAP_CLUSTER_COLS;
2961 0 : for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) {
2962 0 : idx = i * SWAP_CLUSTER_COLS + j;
2963 0 : if (idx >= nr_clusters)
2964 0 : continue;
2965 0 : if (cluster_count(&cluster_info[idx]))
2966 0 : continue;
2967 0 : cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
2968 0 : cluster_list_add_tail(&p->free_clusters, cluster_info,
2969 : idx);
2970 : }
2971 : }
2972 : return nr_extents;
2973 : }
2974 :
2975 0 : SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2976 : {
2977 : struct swap_info_struct *p;
2978 : struct filename *name;
2979 0 : struct file *swap_file = NULL;
2980 : struct address_space *mapping;
2981 : struct dentry *dentry;
2982 : int prio;
2983 : int error;
2984 : union swap_header *swap_header;
2985 : int nr_extents;
2986 : sector_t span;
2987 : unsigned long maxpages;
2988 0 : unsigned char *swap_map = NULL;
2989 0 : struct swap_cluster_info *cluster_info = NULL;
2990 0 : unsigned long *frontswap_map = NULL;
2991 0 : struct page *page = NULL;
2992 0 : struct inode *inode = NULL;
2993 0 : bool inced_nr_rotate_swap = false;
2994 :
2995 0 : if (swap_flags & ~SWAP_FLAGS_VALID)
2996 : return -EINVAL;
2997 :
2998 0 : if (!capable(CAP_SYS_ADMIN))
2999 : return -EPERM;
3000 :
3001 0 : if (!swap_avail_heads)
3002 : return -ENOMEM;
3003 :
3004 0 : p = alloc_swap_info();
3005 0 : if (IS_ERR(p))
3006 0 : return PTR_ERR(p);
3007 :
3008 0 : INIT_WORK(&p->discard_work, swap_discard_work);
3009 :
3010 0 : name = getname(specialfile);
3011 0 : if (IS_ERR(name)) {
3012 0 : error = PTR_ERR(name);
3013 0 : name = NULL;
3014 0 : goto bad_swap;
3015 : }
3016 0 : swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
3017 0 : if (IS_ERR(swap_file)) {
3018 0 : error = PTR_ERR(swap_file);
3019 0 : swap_file = NULL;
3020 0 : goto bad_swap;
3021 : }
3022 :
3023 0 : p->swap_file = swap_file;
3024 0 : mapping = swap_file->f_mapping;
3025 0 : dentry = swap_file->f_path.dentry;
3026 0 : inode = mapping->host;
3027 :
3028 0 : error = claim_swapfile(p, inode);
3029 0 : if (unlikely(error))
3030 : goto bad_swap;
3031 :
3032 0 : inode_lock(inode);
3033 0 : if (d_unlinked(dentry) || cant_mount(dentry)) {
3034 : error = -ENOENT;
3035 : goto bad_swap_unlock_inode;
3036 : }
3037 0 : if (IS_SWAPFILE(inode)) {
3038 : error = -EBUSY;
3039 : goto bad_swap_unlock_inode;
3040 : }
3041 :
3042 : /*
3043 : * Read the swap header.
3044 : */
3045 0 : if (!mapping->a_ops->read_folio) {
3046 : error = -EINVAL;
3047 : goto bad_swap_unlock_inode;
3048 : }
3049 0 : page = read_mapping_page(mapping, 0, swap_file);
3050 0 : if (IS_ERR(page)) {
3051 0 : error = PTR_ERR(page);
3052 0 : goto bad_swap_unlock_inode;
3053 : }
3054 0 : swap_header = kmap(page);
3055 :
3056 0 : maxpages = read_swap_header(p, swap_header, inode);
3057 0 : if (unlikely(!maxpages)) {
3058 : error = -EINVAL;
3059 : goto bad_swap_unlock_inode;
3060 : }
3061 :
3062 : /* OK, set up the swap map and apply the bad block list */
3063 0 : swap_map = vzalloc(maxpages);
3064 0 : if (!swap_map) {
3065 : error = -ENOMEM;
3066 : goto bad_swap_unlock_inode;
3067 : }
3068 :
3069 0 : if (p->bdev && bdev_stable_writes(p->bdev))
3070 0 : p->flags |= SWP_STABLE_WRITES;
3071 :
3072 0 : if (p->bdev && bdev_synchronous(p->bdev))
3073 0 : p->flags |= SWP_SYNCHRONOUS_IO;
3074 :
3075 0 : if (p->bdev && bdev_nonrot(p->bdev)) {
3076 : int cpu;
3077 : unsigned long ci, nr_cluster;
3078 :
3079 0 : p->flags |= SWP_SOLIDSTATE;
3080 0 : p->cluster_next_cpu = alloc_percpu(unsigned int);
3081 0 : if (!p->cluster_next_cpu) {
3082 : error = -ENOMEM;
3083 : goto bad_swap_unlock_inode;
3084 : }
3085 : /*
3086 : * select a random position to start with to help wear leveling
3087 : * SSD
3088 : */
3089 0 : for_each_possible_cpu(cpu) {
3090 0 : per_cpu(*p->cluster_next_cpu, cpu) =
3091 0 : get_random_u32_inclusive(1, p->highest_bit);
3092 : }
3093 0 : nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
3094 :
3095 0 : cluster_info = kvcalloc(nr_cluster, sizeof(*cluster_info),
3096 : GFP_KERNEL);
3097 0 : if (!cluster_info) {
3098 : error = -ENOMEM;
3099 : goto bad_swap_unlock_inode;
3100 : }
3101 :
3102 : for (ci = 0; ci < nr_cluster; ci++)
3103 : spin_lock_init(&((cluster_info + ci)->lock));
3104 :
3105 0 : p->percpu_cluster = alloc_percpu(struct percpu_cluster);
3106 0 : if (!p->percpu_cluster) {
3107 : error = -ENOMEM;
3108 : goto bad_swap_unlock_inode;
3109 : }
3110 0 : for_each_possible_cpu(cpu) {
3111 : struct percpu_cluster *cluster;
3112 0 : cluster = per_cpu_ptr(p->percpu_cluster, cpu);
3113 0 : cluster_set_null(&cluster->index);
3114 : }
3115 : } else {
3116 0 : atomic_inc(&nr_rotate_swap);
3117 0 : inced_nr_rotate_swap = true;
3118 : }
3119 :
3120 0 : error = swap_cgroup_swapon(p->type, maxpages);
3121 : if (error)
3122 : goto bad_swap_unlock_inode;
3123 :
3124 0 : nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
3125 : cluster_info, maxpages, &span);
3126 0 : if (unlikely(nr_extents < 0)) {
3127 : error = nr_extents;
3128 : goto bad_swap_unlock_inode;
3129 : }
3130 : /* frontswap enabled? set up bit-per-page map for frontswap */
3131 : if (IS_ENABLED(CONFIG_FRONTSWAP))
3132 : frontswap_map = kvcalloc(BITS_TO_LONGS(maxpages),
3133 : sizeof(long),
3134 : GFP_KERNEL);
3135 :
3136 0 : if ((swap_flags & SWAP_FLAG_DISCARD) &&
3137 0 : p->bdev && bdev_max_discard_sectors(p->bdev)) {
3138 : /*
3139 : * When discard is enabled for swap with no particular
3140 : * policy flagged, we set all swap discard flags here in
3141 : * order to sustain backward compatibility with older
3142 : * swapon(8) releases.
3143 : */
3144 0 : p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
3145 : SWP_PAGE_DISCARD);
3146 :
3147 : /*
3148 : * By flagging sys_swapon, a sysadmin can tell us to
3149 : * either do single-time area discards only, or to just
3150 : * perform discards for released swap page-clusters.
3151 : * Now it's time to adjust the p->flags accordingly.
3152 : */
3153 0 : if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
3154 0 : p->flags &= ~SWP_PAGE_DISCARD;
3155 0 : else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
3156 0 : p->flags &= ~SWP_AREA_DISCARD;
3157 :
3158 : /* issue a swapon-time discard if it's still required */
3159 0 : if (p->flags & SWP_AREA_DISCARD) {
3160 0 : int err = discard_swap(p);
3161 0 : if (unlikely(err))
3162 0 : pr_err("swapon: discard_swap(%p): %d\n",
3163 : p, err);
3164 : }
3165 : }
3166 :
3167 0 : error = init_swap_address_space(p->type, maxpages);
3168 0 : if (error)
3169 : goto bad_swap_unlock_inode;
3170 :
3171 : /*
3172 : * Flush any pending IO and dirty mappings before we start using this
3173 : * swap device.
3174 : */
3175 0 : inode->i_flags |= S_SWAPFILE;
3176 0 : error = inode_drain_writes(inode);
3177 0 : if (error) {
3178 0 : inode->i_flags &= ~S_SWAPFILE;
3179 : goto free_swap_address_space;
3180 : }
3181 :
3182 0 : mutex_lock(&swapon_mutex);
3183 0 : prio = -1;
3184 0 : if (swap_flags & SWAP_FLAG_PREFER)
3185 0 : prio =
3186 : (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
3187 0 : enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
3188 :
3189 0 : pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
3190 : p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
3191 : nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
3192 : (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
3193 : (p->flags & SWP_DISCARDABLE) ? "D" : "",
3194 : (p->flags & SWP_AREA_DISCARD) ? "s" : "",
3195 : (p->flags & SWP_PAGE_DISCARD) ? "c" : "",
3196 : (frontswap_map) ? "FS" : "");
3197 :
3198 0 : mutex_unlock(&swapon_mutex);
3199 0 : atomic_inc(&proc_poll_event);
3200 0 : wake_up_interruptible(&proc_poll_wait);
3201 :
3202 0 : error = 0;
3203 0 : goto out;
3204 : free_swap_address_space:
3205 0 : exit_swap_address_space(p->type);
3206 : bad_swap_unlock_inode:
3207 : inode_unlock(inode);
3208 : bad_swap:
3209 0 : free_percpu(p->percpu_cluster);
3210 0 : p->percpu_cluster = NULL;
3211 0 : free_percpu(p->cluster_next_cpu);
3212 0 : p->cluster_next_cpu = NULL;
3213 0 : if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
3214 0 : set_blocksize(p->bdev, p->old_block_size);
3215 0 : blkdev_put(p->bdev, p);
3216 : }
3217 0 : inode = NULL;
3218 0 : destroy_swap_extents(p);
3219 0 : swap_cgroup_swapoff(p->type);
3220 0 : spin_lock(&swap_lock);
3221 0 : p->swap_file = NULL;
3222 0 : p->flags = 0;
3223 0 : spin_unlock(&swap_lock);
3224 0 : vfree(swap_map);
3225 0 : kvfree(cluster_info);
3226 0 : kvfree(frontswap_map);
3227 0 : if (inced_nr_rotate_swap)
3228 : atomic_dec(&nr_rotate_swap);
3229 0 : if (swap_file)
3230 0 : filp_close(swap_file, NULL);
3231 : out:
3232 0 : if (page && !IS_ERR(page)) {
3233 0 : kunmap(page);
3234 0 : put_page(page);
3235 : }
3236 0 : if (name)
3237 0 : putname(name);
3238 0 : if (inode)
3239 : inode_unlock(inode);
3240 0 : if (!error)
3241 0 : enable_swap_slots_cache();
3242 0 : return error;
3243 : }
3244 :
3245 0 : void si_swapinfo(struct sysinfo *val)
3246 : {
3247 : unsigned int type;
3248 0 : unsigned long nr_to_be_unused = 0;
3249 :
3250 0 : spin_lock(&swap_lock);
3251 0 : for (type = 0; type < nr_swapfiles; type++) {
3252 0 : struct swap_info_struct *si = swap_info[type];
3253 :
3254 0 : if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
3255 0 : nr_to_be_unused += READ_ONCE(si->inuse_pages);
3256 : }
3257 0 : val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
3258 0 : val->totalswap = total_swap_pages + nr_to_be_unused;
3259 0 : spin_unlock(&swap_lock);
3260 0 : }
3261 :
3262 : /*
3263 : * Verify that a swap entry is valid and increment its swap map count.
3264 : *
3265 : * Returns error code in following case.
3266 : * - success -> 0
3267 : * - swp_entry is invalid -> EINVAL
3268 : * - swp_entry is migration entry -> EINVAL
3269 : * - swap-cache reference is requested but there is already one. -> EEXIST
3270 : * - swap-cache reference is requested but the entry is not used. -> ENOENT
3271 : * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
3272 : */
3273 0 : static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
3274 : {
3275 : struct swap_info_struct *p;
3276 : struct swap_cluster_info *ci;
3277 : unsigned long offset;
3278 : unsigned char count;
3279 : unsigned char has_cache;
3280 : int err;
3281 :
3282 0 : p = swp_swap_info(entry);
3283 :
3284 0 : offset = swp_offset(entry);
3285 0 : ci = lock_cluster_or_swap_info(p, offset);
3286 :
3287 0 : count = p->swap_map[offset];
3288 :
3289 : /*
3290 : * swapin_readahead() doesn't check if a swap entry is valid, so the
3291 : * swap entry could be SWAP_MAP_BAD. Check here with lock held.
3292 : */
3293 0 : if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
3294 : err = -ENOENT;
3295 : goto unlock_out;
3296 : }
3297 :
3298 0 : has_cache = count & SWAP_HAS_CACHE;
3299 0 : count &= ~SWAP_HAS_CACHE;
3300 0 : err = 0;
3301 :
3302 0 : if (usage == SWAP_HAS_CACHE) {
3303 :
3304 : /* set SWAP_HAS_CACHE if there is no cache and entry is used */
3305 0 : if (!has_cache && count)
3306 : has_cache = SWAP_HAS_CACHE;
3307 0 : else if (has_cache) /* someone else added cache */
3308 : err = -EEXIST;
3309 : else /* no users remaining */
3310 0 : err = -ENOENT;
3311 :
3312 0 : } else if (count || has_cache) {
3313 :
3314 0 : if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
3315 0 : count += usage;
3316 0 : else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
3317 : err = -EINVAL;
3318 0 : else if (swap_count_continued(p, offset, count))
3319 : count = COUNT_CONTINUED;
3320 : else
3321 0 : err = -ENOMEM;
3322 : } else
3323 : err = -ENOENT; /* unused swap entry */
3324 :
3325 0 : WRITE_ONCE(p->swap_map[offset], count | has_cache);
3326 :
3327 : unlock_out:
3328 0 : unlock_cluster_or_swap_info(p, ci);
3329 0 : return err;
3330 : }
3331 :
3332 : /*
3333 : * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3334 : * (in which case its reference count is never incremented).
3335 : */
3336 0 : void swap_shmem_alloc(swp_entry_t entry)
3337 : {
3338 0 : __swap_duplicate(entry, SWAP_MAP_SHMEM);
3339 0 : }
3340 :
3341 : /*
3342 : * Increase reference count of swap entry by 1.
3343 : * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3344 : * but could not be atomically allocated. Returns 0, just as if it succeeded,
3345 : * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3346 : * might occur if a page table entry has got corrupted.
3347 : */
3348 0 : int swap_duplicate(swp_entry_t entry)
3349 : {
3350 0 : int err = 0;
3351 :
3352 0 : while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
3353 0 : err = add_swap_count_continuation(entry, GFP_ATOMIC);
3354 0 : return err;
3355 : }
3356 :
3357 : /*
3358 : * @entry: swap entry for which we allocate swap cache.
3359 : *
3360 : * Called when allocating swap cache for existing swap entry,
3361 : * This can return error codes. Returns 0 at success.
3362 : * -EEXIST means there is a swap cache.
3363 : * Note: return code is different from swap_duplicate().
3364 : */
3365 0 : int swapcache_prepare(swp_entry_t entry)
3366 : {
3367 0 : return __swap_duplicate(entry, SWAP_HAS_CACHE);
3368 : }
3369 :
3370 0 : struct swap_info_struct *swp_swap_info(swp_entry_t entry)
3371 : {
3372 0 : return swap_type_to_swap_info(swp_type(entry));
3373 : }
3374 :
3375 0 : struct swap_info_struct *page_swap_info(struct page *page)
3376 : {
3377 0 : swp_entry_t entry = { .val = page_private(page) };
3378 0 : return swp_swap_info(entry);
3379 : }
3380 :
3381 : /*
3382 : * out-of-line methods to avoid include hell.
3383 : */
3384 0 : struct address_space *swapcache_mapping(struct folio *folio)
3385 : {
3386 0 : return page_swap_info(&folio->page)->swap_file->f_mapping;
3387 : }
3388 : EXPORT_SYMBOL_GPL(swapcache_mapping);
3389 :
3390 0 : pgoff_t __page_file_index(struct page *page)
3391 : {
3392 0 : swp_entry_t swap = { .val = page_private(page) };
3393 0 : return swp_offset(swap);
3394 : }
3395 : EXPORT_SYMBOL_GPL(__page_file_index);
3396 :
3397 : /*
3398 : * add_swap_count_continuation - called when a swap count is duplicated
3399 : * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3400 : * page of the original vmalloc'ed swap_map, to hold the continuation count
3401 : * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
3402 : * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3403 : *
3404 : * These continuation pages are seldom referenced: the common paths all work
3405 : * on the original swap_map, only referring to a continuation page when the
3406 : * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3407 : *
3408 : * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3409 : * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3410 : * can be called after dropping locks.
3411 : */
3412 0 : int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
3413 : {
3414 : struct swap_info_struct *si;
3415 : struct swap_cluster_info *ci;
3416 : struct page *head;
3417 : struct page *page;
3418 : struct page *list_page;
3419 : pgoff_t offset;
3420 : unsigned char count;
3421 0 : int ret = 0;
3422 :
3423 : /*
3424 : * When debugging, it's easier to use __GFP_ZERO here; but it's better
3425 : * for latency not to zero a page while GFP_ATOMIC and holding locks.
3426 : */
3427 0 : page = alloc_page(gfp_mask | __GFP_HIGHMEM);
3428 :
3429 0 : si = get_swap_device(entry);
3430 0 : if (!si) {
3431 : /*
3432 : * An acceptable race has occurred since the failing
3433 : * __swap_duplicate(): the swap device may be swapoff
3434 : */
3435 : goto outer;
3436 : }
3437 0 : spin_lock(&si->lock);
3438 :
3439 0 : offset = swp_offset(entry);
3440 :
3441 0 : ci = lock_cluster(si, offset);
3442 :
3443 0 : count = swap_count(si->swap_map[offset]);
3444 :
3445 0 : if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
3446 : /*
3447 : * The higher the swap count, the more likely it is that tasks
3448 : * will race to add swap count continuation: we need to avoid
3449 : * over-provisioning.
3450 : */
3451 : goto out;
3452 : }
3453 :
3454 0 : if (!page) {
3455 : ret = -ENOMEM;
3456 : goto out;
3457 : }
3458 :
3459 0 : head = vmalloc_to_page(si->swap_map + offset);
3460 0 : offset &= ~PAGE_MASK;
3461 :
3462 0 : spin_lock(&si->cont_lock);
3463 : /*
3464 : * Page allocation does not initialize the page's lru field,
3465 : * but it does always reset its private field.
3466 : */
3467 0 : if (!page_private(head)) {
3468 0 : BUG_ON(count & COUNT_CONTINUED);
3469 0 : INIT_LIST_HEAD(&head->lru);
3470 0 : set_page_private(head, SWP_CONTINUED);
3471 0 : si->flags |= SWP_CONTINUED;
3472 : }
3473 :
3474 0 : list_for_each_entry(list_page, &head->lru, lru) {
3475 : unsigned char *map;
3476 :
3477 : /*
3478 : * If the previous map said no continuation, but we've found
3479 : * a continuation page, free our allocation and use this one.
3480 : */
3481 0 : if (!(count & COUNT_CONTINUED))
3482 : goto out_unlock_cont;
3483 :
3484 0 : map = kmap_atomic(list_page) + offset;
3485 0 : count = *map;
3486 0 : kunmap_atomic(map);
3487 :
3488 : /*
3489 : * If this continuation count now has some space in it,
3490 : * free our allocation and use this one.
3491 : */
3492 0 : if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
3493 : goto out_unlock_cont;
3494 : }
3495 :
3496 0 : list_add_tail(&page->lru, &head->lru);
3497 0 : page = NULL; /* now it's attached, don't free it */
3498 : out_unlock_cont:
3499 0 : spin_unlock(&si->cont_lock);
3500 : out:
3501 0 : unlock_cluster(ci);
3502 0 : spin_unlock(&si->lock);
3503 : put_swap_device(si);
3504 : outer:
3505 0 : if (page)
3506 0 : __free_page(page);
3507 0 : return ret;
3508 : }
3509 :
3510 : /*
3511 : * swap_count_continued - when the original swap_map count is incremented
3512 : * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3513 : * into, carry if so, or else fail until a new continuation page is allocated;
3514 : * when the original swap_map count is decremented from 0 with continuation,
3515 : * borrow from the continuation and report whether it still holds more.
3516 : * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3517 : * lock.
3518 : */
3519 0 : static bool swap_count_continued(struct swap_info_struct *si,
3520 : pgoff_t offset, unsigned char count)
3521 : {
3522 : struct page *head;
3523 : struct page *page;
3524 : unsigned char *map;
3525 : bool ret;
3526 :
3527 0 : head = vmalloc_to_page(si->swap_map + offset);
3528 0 : if (page_private(head) != SWP_CONTINUED) {
3529 0 : BUG_ON(count & COUNT_CONTINUED);
3530 : return false; /* need to add count continuation */
3531 : }
3532 :
3533 0 : spin_lock(&si->cont_lock);
3534 0 : offset &= ~PAGE_MASK;
3535 0 : page = list_next_entry(head, lru);
3536 0 : map = kmap_atomic(page) + offset;
3537 :
3538 0 : if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
3539 : goto init_map; /* jump over SWAP_CONT_MAX checks */
3540 :
3541 0 : if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
3542 : /*
3543 : * Think of how you add 1 to 999
3544 : */
3545 0 : while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
3546 0 : kunmap_atomic(map);
3547 0 : page = list_next_entry(page, lru);
3548 0 : BUG_ON(page == head);
3549 0 : map = kmap_atomic(page) + offset;
3550 : }
3551 0 : if (*map == SWAP_CONT_MAX) {
3552 0 : kunmap_atomic(map);
3553 0 : page = list_next_entry(page, lru);
3554 0 : if (page == head) {
3555 : ret = false; /* add count continuation */
3556 : goto out;
3557 : }
3558 0 : map = kmap_atomic(page) + offset;
3559 0 : init_map: *map = 0; /* we didn't zero the page */
3560 : }
3561 0 : *map += 1;
3562 0 : kunmap_atomic(map);
3563 0 : while ((page = list_prev_entry(page, lru)) != head) {
3564 0 : map = kmap_atomic(page) + offset;
3565 0 : *map = COUNT_CONTINUED;
3566 0 : kunmap_atomic(map);
3567 : }
3568 : ret = true; /* incremented */
3569 :
3570 : } else { /* decrementing */
3571 : /*
3572 : * Think of how you subtract 1 from 1000
3573 : */
3574 0 : BUG_ON(count != COUNT_CONTINUED);
3575 0 : while (*map == COUNT_CONTINUED) {
3576 0 : kunmap_atomic(map);
3577 0 : page = list_next_entry(page, lru);
3578 0 : BUG_ON(page == head);
3579 0 : map = kmap_atomic(page) + offset;
3580 : }
3581 0 : BUG_ON(*map == 0);
3582 0 : *map -= 1;
3583 0 : if (*map == 0)
3584 0 : count = 0;
3585 0 : kunmap_atomic(map);
3586 0 : while ((page = list_prev_entry(page, lru)) != head) {
3587 0 : map = kmap_atomic(page) + offset;
3588 0 : *map = SWAP_CONT_MAX | count;
3589 0 : count = COUNT_CONTINUED;
3590 0 : kunmap_atomic(map);
3591 : }
3592 0 : ret = count == COUNT_CONTINUED;
3593 : }
3594 : out:
3595 0 : spin_unlock(&si->cont_lock);
3596 0 : return ret;
3597 : }
3598 :
3599 : /*
3600 : * free_swap_count_continuations - swapoff free all the continuation pages
3601 : * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3602 : */
3603 0 : static void free_swap_count_continuations(struct swap_info_struct *si)
3604 : {
3605 : pgoff_t offset;
3606 :
3607 0 : for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
3608 : struct page *head;
3609 0 : head = vmalloc_to_page(si->swap_map + offset);
3610 0 : if (page_private(head)) {
3611 : struct page *page, *next;
3612 :
3613 0 : list_for_each_entry_safe(page, next, &head->lru, lru) {
3614 0 : list_del(&page->lru);
3615 0 : __free_page(page);
3616 : }
3617 : }
3618 : }
3619 0 : }
3620 :
3621 : #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
3622 : void __folio_throttle_swaprate(struct folio *folio, gfp_t gfp)
3623 : {
3624 : struct swap_info_struct *si, *next;
3625 : int nid = folio_nid(folio);
3626 :
3627 : if (!(gfp & __GFP_IO))
3628 : return;
3629 :
3630 : if (!blk_cgroup_congested())
3631 : return;
3632 :
3633 : /*
3634 : * We've already scheduled a throttle, avoid taking the global swap
3635 : * lock.
3636 : */
3637 : if (current->throttle_disk)
3638 : return;
3639 :
3640 : spin_lock(&swap_avail_lock);
3641 : plist_for_each_entry_safe(si, next, &swap_avail_heads[nid],
3642 : avail_lists[nid]) {
3643 : if (si->bdev) {
3644 : blkcg_schedule_throttle(si->bdev->bd_disk, true);
3645 : break;
3646 : }
3647 : }
3648 : spin_unlock(&swap_avail_lock);
3649 : }
3650 : #endif
3651 :
3652 1 : static int __init swapfile_init(void)
3653 : {
3654 : int nid;
3655 :
3656 1 : swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head),
3657 : GFP_KERNEL);
3658 1 : if (!swap_avail_heads) {
3659 0 : pr_emerg("Not enough memory for swap heads, swap is disabled\n");
3660 0 : return -ENOMEM;
3661 : }
3662 :
3663 1 : for_each_node(nid)
3664 2 : plist_head_init(&swap_avail_heads[nid]);
3665 :
3666 1 : swapfile_maximum_size = arch_max_swapfile_size();
3667 :
3668 : #ifdef CONFIG_MIGRATION
3669 1 : if (swapfile_maximum_size >= (1UL << SWP_MIG_TOTAL_BITS))
3670 0 : swap_migration_ad_supported = true;
3671 : #endif /* CONFIG_MIGRATION */
3672 :
3673 : return 0;
3674 : }
3675 : subsys_initcall(swapfile_init);
|
