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 :
45 : #include <asm/tlbflush.h>
46 : #include <linux/swapops.h>
47 : #include <linux/swap_cgroup.h>
48 : #include "swap.h"
49 :
50 : static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
51 : unsigned char);
52 : static void free_swap_count_continuations(struct swap_info_struct *);
53 :
54 : static DEFINE_SPINLOCK(swap_lock);
55 : static unsigned int nr_swapfiles;
56 : atomic_long_t nr_swap_pages;
57 : /*
58 : * Some modules use swappable objects and may try to swap them out under
59 : * memory pressure (via the shrinker). Before doing so, they may wish to
60 : * check to see if any swap space is available.
61 : */
62 : EXPORT_SYMBOL_GPL(nr_swap_pages);
63 : /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
64 : long total_swap_pages;
65 : static int least_priority = -1;
66 : unsigned long swapfile_maximum_size;
67 : #ifdef CONFIG_MIGRATION
68 : bool swap_migration_ad_supported;
69 : #endif /* CONFIG_MIGRATION */
70 :
71 : static const char Bad_file[] = "Bad swap file entry ";
72 : static const char Unused_file[] = "Unused swap file entry ";
73 : static const char Bad_offset[] = "Bad swap offset entry ";
74 : static const char Unused_offset[] = "Unused swap offset entry ";
75 :
76 : /*
77 : * all active swap_info_structs
78 : * protected with swap_lock, and ordered by priority.
79 : */
80 : static PLIST_HEAD(swap_active_head);
81 :
82 : /*
83 : * all available (active, not full) swap_info_structs
84 : * protected with swap_avail_lock, ordered by priority.
85 : * This is used by folio_alloc_swap() instead of swap_active_head
86 : * because swap_active_head includes all swap_info_structs,
87 : * but folio_alloc_swap() doesn't need to look at full ones.
88 : * This uses its own lock instead of swap_lock because when a
89 : * swap_info_struct changes between not-full/full, it needs to
90 : * add/remove itself to/from this list, but the swap_info_struct->lock
91 : * is held and the locking order requires swap_lock to be taken
92 : * before any swap_info_struct->lock.
93 : */
94 : static struct plist_head *swap_avail_heads;
95 : static DEFINE_SPINLOCK(swap_avail_lock);
96 :
97 : struct swap_info_struct *swap_info[MAX_SWAPFILES];
98 :
99 : static DEFINE_MUTEX(swapon_mutex);
100 :
101 : static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
102 : /* Activity counter to indicate that a swapon or swapoff has occurred */
103 : static atomic_t proc_poll_event = ATOMIC_INIT(0);
104 :
105 : atomic_t nr_rotate_swap = ATOMIC_INIT(0);
106 :
107 : static struct swap_info_struct *swap_type_to_swap_info(int type)
108 : {
109 0 : if (type >= MAX_SWAPFILES)
110 : return NULL;
111 :
112 0 : return READ_ONCE(swap_info[type]); /* rcu_dereference() */
113 : }
114 :
115 : static inline unsigned char swap_count(unsigned char ent)
116 : {
117 0 : return ent & ~SWAP_HAS_CACHE; /* may include COUNT_CONTINUED flag */
118 : }
119 :
120 : /* Reclaim the swap entry anyway if possible */
121 : #define TTRS_ANYWAY 0x1
122 : /*
123 : * Reclaim the swap entry if there are no more mappings of the
124 : * corresponding page
125 : */
126 : #define TTRS_UNMAPPED 0x2
127 : /* Reclaim the swap entry if swap is getting full*/
128 : #define TTRS_FULL 0x4
129 :
130 : /* returns 1 if swap entry is freed */
131 0 : static int __try_to_reclaim_swap(struct swap_info_struct *si,
132 : unsigned long offset, unsigned long flags)
133 : {
134 0 : swp_entry_t entry = swp_entry(si->type, offset);
135 : struct folio *folio;
136 0 : int ret = 0;
137 :
138 0 : folio = filemap_get_folio(swap_address_space(entry), offset);
139 0 : if (IS_ERR(folio))
140 : return 0;
141 : /*
142 : * When this function is called from scan_swap_map_slots() and it's
143 : * called by vmscan.c at reclaiming folios. So we hold a folio lock
144 : * here. We have to use trylock for avoiding deadlock. This is a special
145 : * case and you should use folio_free_swap() with explicit folio_lock()
146 : * in usual operations.
147 : */
148 0 : if (folio_trylock(folio)) {
149 0 : if ((flags & TTRS_ANYWAY) ||
150 0 : ((flags & TTRS_UNMAPPED) && !folio_mapped(folio)) ||
151 0 : ((flags & TTRS_FULL) && mem_cgroup_swap_full(folio)))
152 0 : ret = folio_free_swap(folio);
153 0 : folio_unlock(folio);
154 : }
155 : folio_put(folio);
156 : return ret;
157 : }
158 :
159 : static inline struct swap_extent *first_se(struct swap_info_struct *sis)
160 : {
161 0 : struct rb_node *rb = rb_first(&sis->swap_extent_root);
162 0 : return rb_entry(rb, struct swap_extent, rb_node);
163 : }
164 :
165 : static inline struct swap_extent *next_se(struct swap_extent *se)
166 : {
167 0 : struct rb_node *rb = rb_next(&se->rb_node);
168 0 : return rb ? rb_entry(rb, struct swap_extent, rb_node) : NULL;
169 : }
170 :
171 : /*
172 : * swapon tell device that all the old swap contents can be discarded,
173 : * to allow the swap device to optimize its wear-levelling.
174 : */
175 0 : static int discard_swap(struct swap_info_struct *si)
176 : {
177 : struct swap_extent *se;
178 : sector_t start_block;
179 : sector_t nr_blocks;
180 0 : int err = 0;
181 :
182 : /* Do not discard the swap header page! */
183 0 : se = first_se(si);
184 0 : start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
185 0 : nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
186 0 : if (nr_blocks) {
187 0 : err = blkdev_issue_discard(si->bdev, start_block,
188 : nr_blocks, GFP_KERNEL);
189 0 : if (err)
190 : return err;
191 0 : cond_resched();
192 : }
193 :
194 0 : for (se = next_se(se); se; se = next_se(se)) {
195 0 : start_block = se->start_block << (PAGE_SHIFT - 9);
196 0 : nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
197 :
198 0 : err = blkdev_issue_discard(si->bdev, start_block,
199 : nr_blocks, GFP_KERNEL);
200 0 : if (err)
201 : break;
202 :
203 0 : cond_resched();
204 : }
205 : return err; /* That will often be -EOPNOTSUPP */
206 : }
207 :
208 : static struct swap_extent *
209 0 : offset_to_swap_extent(struct swap_info_struct *sis, unsigned long offset)
210 : {
211 : struct swap_extent *se;
212 : struct rb_node *rb;
213 :
214 0 : rb = sis->swap_extent_root.rb_node;
215 0 : while (rb) {
216 0 : se = rb_entry(rb, struct swap_extent, rb_node);
217 0 : if (offset < se->start_page)
218 0 : rb = rb->rb_left;
219 0 : else if (offset >= se->start_page + se->nr_pages)
220 0 : rb = rb->rb_right;
221 : else
222 0 : return se;
223 : }
224 : /* It *must* be present */
225 0 : BUG();
226 : }
227 :
228 0 : sector_t swap_page_sector(struct page *page)
229 : {
230 0 : struct swap_info_struct *sis = page_swap_info(page);
231 : struct swap_extent *se;
232 : sector_t sector;
233 : pgoff_t offset;
234 :
235 0 : offset = __page_file_index(page);
236 0 : se = offset_to_swap_extent(sis, offset);
237 0 : sector = se->start_block + (offset - se->start_page);
238 0 : return sector << (PAGE_SHIFT - 9);
239 : }
240 :
241 : /*
242 : * swap allocation tell device that a cluster of swap can now be discarded,
243 : * to allow the swap device to optimize its wear-levelling.
244 : */
245 0 : static void discard_swap_cluster(struct swap_info_struct *si,
246 : pgoff_t start_page, pgoff_t nr_pages)
247 : {
248 0 : struct swap_extent *se = offset_to_swap_extent(si, start_page);
249 :
250 0 : while (nr_pages) {
251 0 : pgoff_t offset = start_page - se->start_page;
252 0 : sector_t start_block = se->start_block + offset;
253 0 : sector_t nr_blocks = se->nr_pages - offset;
254 :
255 0 : if (nr_blocks > nr_pages)
256 0 : nr_blocks = nr_pages;
257 0 : start_page += nr_blocks;
258 0 : nr_pages -= nr_blocks;
259 :
260 0 : start_block <<= PAGE_SHIFT - 9;
261 0 : nr_blocks <<= PAGE_SHIFT - 9;
262 0 : if (blkdev_issue_discard(si->bdev, start_block,
263 : nr_blocks, GFP_NOIO))
264 : break;
265 :
266 : se = next_se(se);
267 : }
268 0 : }
269 :
270 : #ifdef CONFIG_THP_SWAP
271 : #define SWAPFILE_CLUSTER HPAGE_PMD_NR
272 :
273 : #define swap_entry_size(size) (size)
274 : #else
275 : #define SWAPFILE_CLUSTER 256
276 :
277 : /*
278 : * Define swap_entry_size() as constant to let compiler to optimize
279 : * out some code if !CONFIG_THP_SWAP
280 : */
281 : #define swap_entry_size(size) 1
282 : #endif
283 : #define LATENCY_LIMIT 256
284 :
285 : static inline void cluster_set_flag(struct swap_cluster_info *info,
286 : unsigned int flag)
287 : {
288 0 : info->flags = flag;
289 : }
290 :
291 : static inline unsigned int cluster_count(struct swap_cluster_info *info)
292 : {
293 0 : return info->data;
294 : }
295 :
296 : static inline void cluster_set_count(struct swap_cluster_info *info,
297 : unsigned int c)
298 : {
299 0 : info->data = c;
300 : }
301 :
302 : static inline void cluster_set_count_flag(struct swap_cluster_info *info,
303 : unsigned int c, unsigned int f)
304 : {
305 0 : info->flags = f;
306 0 : info->data = c;
307 : }
308 :
309 : static inline unsigned int cluster_next(struct swap_cluster_info *info)
310 : {
311 0 : return info->data;
312 : }
313 :
314 : static inline void cluster_set_next(struct swap_cluster_info *info,
315 : unsigned int n)
316 : {
317 0 : info->data = n;
318 : }
319 :
320 : static inline void cluster_set_next_flag(struct swap_cluster_info *info,
321 : unsigned int n, unsigned int f)
322 : {
323 0 : info->flags = f;
324 0 : info->data = n;
325 : }
326 :
327 : static inline bool cluster_is_free(struct swap_cluster_info *info)
328 : {
329 0 : return info->flags & CLUSTER_FLAG_FREE;
330 : }
331 :
332 : static inline bool cluster_is_null(struct swap_cluster_info *info)
333 : {
334 0 : return info->flags & CLUSTER_FLAG_NEXT_NULL;
335 : }
336 :
337 : static inline void cluster_set_null(struct swap_cluster_info *info)
338 : {
339 0 : info->flags = CLUSTER_FLAG_NEXT_NULL;
340 0 : info->data = 0;
341 : }
342 :
343 : static inline bool cluster_is_huge(struct swap_cluster_info *info)
344 : {
345 : if (IS_ENABLED(CONFIG_THP_SWAP))
346 : return info->flags & CLUSTER_FLAG_HUGE;
347 : return false;
348 : }
349 :
350 : static inline void cluster_clear_huge(struct swap_cluster_info *info)
351 : {
352 : info->flags &= ~CLUSTER_FLAG_HUGE;
353 : }
354 :
355 : static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si,
356 : unsigned long offset)
357 : {
358 : struct swap_cluster_info *ci;
359 :
360 0 : ci = si->cluster_info;
361 0 : if (ci) {
362 0 : ci += offset / SWAPFILE_CLUSTER;
363 0 : spin_lock(&ci->lock);
364 : }
365 : return ci;
366 : }
367 :
368 : static inline void unlock_cluster(struct swap_cluster_info *ci)
369 : {
370 0 : if (ci)
371 0 : spin_unlock(&ci->lock);
372 : }
373 :
374 : /*
375 : * Determine the locking method in use for this device. Return
376 : * swap_cluster_info if SSD-style cluster-based locking is in place.
377 : */
378 : static inline struct swap_cluster_info *lock_cluster_or_swap_info(
379 : struct swap_info_struct *si, unsigned long offset)
380 : {
381 : struct swap_cluster_info *ci;
382 :
383 : /* Try to use fine-grained SSD-style locking if available: */
384 0 : ci = lock_cluster(si, offset);
385 : /* Otherwise, fall back to traditional, coarse locking: */
386 0 : if (!ci)
387 0 : spin_lock(&si->lock);
388 :
389 : return ci;
390 : }
391 :
392 : static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si,
393 : struct swap_cluster_info *ci)
394 : {
395 0 : if (ci)
396 : unlock_cluster(ci);
397 : else
398 0 : spin_unlock(&si->lock);
399 : }
400 :
401 : static inline bool cluster_list_empty(struct swap_cluster_list *list)
402 : {
403 0 : return cluster_is_null(&list->head);
404 : }
405 :
406 : static inline unsigned int cluster_list_first(struct swap_cluster_list *list)
407 : {
408 0 : return cluster_next(&list->head);
409 : }
410 :
411 : static void cluster_list_init(struct swap_cluster_list *list)
412 : {
413 0 : cluster_set_null(&list->head);
414 0 : cluster_set_null(&list->tail);
415 : }
416 :
417 : static void cluster_list_add_tail(struct swap_cluster_list *list,
418 : struct swap_cluster_info *ci,
419 : unsigned int idx)
420 : {
421 0 : if (cluster_list_empty(list)) {
422 0 : cluster_set_next_flag(&list->head, idx, 0);
423 0 : cluster_set_next_flag(&list->tail, idx, 0);
424 : } else {
425 : struct swap_cluster_info *ci_tail;
426 0 : unsigned int tail = cluster_next(&list->tail);
427 :
428 : /*
429 : * Nested cluster lock, but both cluster locks are
430 : * only acquired when we held swap_info_struct->lock
431 : */
432 0 : ci_tail = ci + tail;
433 0 : spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING);
434 0 : cluster_set_next(ci_tail, idx);
435 0 : spin_unlock(&ci_tail->lock);
436 0 : cluster_set_next_flag(&list->tail, idx, 0);
437 : }
438 : }
439 :
440 : static unsigned int cluster_list_del_first(struct swap_cluster_list *list,
441 : struct swap_cluster_info *ci)
442 : {
443 : unsigned int idx;
444 :
445 0 : idx = cluster_next(&list->head);
446 0 : if (cluster_next(&list->tail) == idx) {
447 0 : cluster_set_null(&list->head);
448 0 : cluster_set_null(&list->tail);
449 : } else
450 0 : cluster_set_next_flag(&list->head,
451 0 : cluster_next(&ci[idx]), 0);
452 :
453 : return idx;
454 : }
455 :
456 : /* Add a cluster to discard list and schedule it to do discard */
457 0 : static void swap_cluster_schedule_discard(struct swap_info_struct *si,
458 : unsigned int idx)
459 : {
460 : /*
461 : * If scan_swap_map_slots() can't find a free cluster, it will check
462 : * si->swap_map directly. To make sure the discarding cluster isn't
463 : * taken by scan_swap_map_slots(), mark the swap entries bad (occupied).
464 : * It will be cleared after discard
465 : */
466 0 : memset(si->swap_map + idx * SWAPFILE_CLUSTER,
467 : SWAP_MAP_BAD, SWAPFILE_CLUSTER);
468 :
469 0 : cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx);
470 :
471 0 : schedule_work(&si->discard_work);
472 0 : }
473 :
474 : static void __free_cluster(struct swap_info_struct *si, unsigned long idx)
475 : {
476 0 : struct swap_cluster_info *ci = si->cluster_info;
477 :
478 0 : cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE);
479 0 : cluster_list_add_tail(&si->free_clusters, ci, idx);
480 : }
481 :
482 : /*
483 : * Doing discard actually. After a cluster discard is finished, the cluster
484 : * will be added to free cluster list. caller should hold si->lock.
485 : */
486 0 : static void swap_do_scheduled_discard(struct swap_info_struct *si)
487 : {
488 : struct swap_cluster_info *info, *ci;
489 : unsigned int idx;
490 :
491 0 : info = si->cluster_info;
492 :
493 0 : while (!cluster_list_empty(&si->discard_clusters)) {
494 0 : idx = cluster_list_del_first(&si->discard_clusters, info);
495 0 : spin_unlock(&si->lock);
496 :
497 0 : discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
498 : SWAPFILE_CLUSTER);
499 :
500 0 : spin_lock(&si->lock);
501 0 : ci = lock_cluster(si, idx * SWAPFILE_CLUSTER);
502 0 : __free_cluster(si, idx);
503 0 : memset(si->swap_map + idx * SWAPFILE_CLUSTER,
504 : 0, SWAPFILE_CLUSTER);
505 : unlock_cluster(ci);
506 : }
507 0 : }
508 :
509 0 : static void swap_discard_work(struct work_struct *work)
510 : {
511 : struct swap_info_struct *si;
512 :
513 0 : si = container_of(work, struct swap_info_struct, discard_work);
514 :
515 0 : spin_lock(&si->lock);
516 0 : swap_do_scheduled_discard(si);
517 0 : spin_unlock(&si->lock);
518 0 : }
519 :
520 0 : static void swap_users_ref_free(struct percpu_ref *ref)
521 : {
522 : struct swap_info_struct *si;
523 :
524 0 : si = container_of(ref, struct swap_info_struct, users);
525 0 : complete(&si->comp);
526 0 : }
527 :
528 : static void alloc_cluster(struct swap_info_struct *si, unsigned long idx)
529 : {
530 0 : struct swap_cluster_info *ci = si->cluster_info;
531 :
532 : VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx);
533 0 : cluster_list_del_first(&si->free_clusters, ci);
534 0 : cluster_set_count_flag(ci + idx, 0, 0);
535 : }
536 :
537 0 : static void free_cluster(struct swap_info_struct *si, unsigned long idx)
538 : {
539 0 : struct swap_cluster_info *ci = si->cluster_info + idx;
540 :
541 : VM_BUG_ON(cluster_count(ci) != 0);
542 : /*
543 : * If the swap is discardable, prepare discard the cluster
544 : * instead of free it immediately. The cluster will be freed
545 : * after discard.
546 : */
547 0 : if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
548 : (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
549 0 : swap_cluster_schedule_discard(si, idx);
550 0 : return;
551 : }
552 :
553 : __free_cluster(si, idx);
554 : }
555 :
556 : /*
557 : * The cluster corresponding to page_nr will be used. The cluster will be
558 : * removed from free cluster list and its usage counter will be increased.
559 : */
560 0 : static void inc_cluster_info_page(struct swap_info_struct *p,
561 : struct swap_cluster_info *cluster_info, unsigned long page_nr)
562 : {
563 0 : unsigned long idx = page_nr / SWAPFILE_CLUSTER;
564 :
565 0 : if (!cluster_info)
566 : return;
567 0 : if (cluster_is_free(&cluster_info[idx]))
568 : alloc_cluster(p, idx);
569 :
570 : VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
571 0 : cluster_set_count(&cluster_info[idx],
572 0 : cluster_count(&cluster_info[idx]) + 1);
573 : }
574 :
575 : /*
576 : * The cluster corresponding to page_nr decreases one usage. If the usage
577 : * counter becomes 0, which means no page in the cluster is in using, we can
578 : * optionally discard the cluster and add it to free cluster list.
579 : */
580 0 : static void dec_cluster_info_page(struct swap_info_struct *p,
581 : struct swap_cluster_info *cluster_info, unsigned long page_nr)
582 : {
583 0 : unsigned long idx = page_nr / SWAPFILE_CLUSTER;
584 :
585 0 : if (!cluster_info)
586 : return;
587 :
588 : VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
589 0 : cluster_set_count(&cluster_info[idx],
590 0 : cluster_count(&cluster_info[idx]) - 1);
591 :
592 0 : if (cluster_count(&cluster_info[idx]) == 0)
593 0 : free_cluster(p, idx);
594 : }
595 :
596 : /*
597 : * It's possible scan_swap_map_slots() uses a free cluster in the middle of free
598 : * cluster list. Avoiding such abuse to avoid list corruption.
599 : */
600 : static bool
601 : scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
602 : unsigned long offset)
603 : {
604 : struct percpu_cluster *percpu_cluster;
605 : bool conflict;
606 :
607 0 : offset /= SWAPFILE_CLUSTER;
608 0 : conflict = !cluster_list_empty(&si->free_clusters) &&
609 0 : offset != cluster_list_first(&si->free_clusters) &&
610 0 : cluster_is_free(&si->cluster_info[offset]);
611 :
612 0 : if (!conflict)
613 : return false;
614 :
615 0 : percpu_cluster = this_cpu_ptr(si->percpu_cluster);
616 0 : cluster_set_null(&percpu_cluster->index);
617 : return true;
618 : }
619 :
620 : /*
621 : * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
622 : * might involve allocating a new cluster for current CPU too.
623 : */
624 0 : static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
625 : unsigned long *offset, unsigned long *scan_base)
626 : {
627 : struct percpu_cluster *cluster;
628 : struct swap_cluster_info *ci;
629 : unsigned long tmp, max;
630 :
631 : new_cluster:
632 0 : cluster = this_cpu_ptr(si->percpu_cluster);
633 0 : if (cluster_is_null(&cluster->index)) {
634 0 : if (!cluster_list_empty(&si->free_clusters)) {
635 0 : cluster->index = si->free_clusters.head;
636 0 : cluster->next = cluster_next(&cluster->index) *
637 : SWAPFILE_CLUSTER;
638 0 : } else if (!cluster_list_empty(&si->discard_clusters)) {
639 : /*
640 : * we don't have free cluster but have some clusters in
641 : * discarding, do discard now and reclaim them, then
642 : * reread cluster_next_cpu since we dropped si->lock
643 : */
644 0 : swap_do_scheduled_discard(si);
645 0 : *scan_base = this_cpu_read(*si->cluster_next_cpu);
646 0 : *offset = *scan_base;
647 0 : goto new_cluster;
648 : } else
649 : return false;
650 : }
651 :
652 : /*
653 : * Other CPUs can use our cluster if they can't find a free cluster,
654 : * check if there is still free entry in the cluster
655 : */
656 0 : tmp = cluster->next;
657 0 : max = min_t(unsigned long, si->max,
658 : (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER);
659 0 : if (tmp < max) {
660 0 : ci = lock_cluster(si, tmp);
661 0 : while (tmp < max) {
662 0 : if (!si->swap_map[tmp])
663 : break;
664 0 : tmp++;
665 : }
666 : unlock_cluster(ci);
667 : }
668 0 : if (tmp >= max) {
669 0 : cluster_set_null(&cluster->index);
670 : goto new_cluster;
671 : }
672 0 : cluster->next = tmp + 1;
673 0 : *offset = tmp;
674 0 : *scan_base = tmp;
675 0 : return true;
676 : }
677 :
678 : static void __del_from_avail_list(struct swap_info_struct *p)
679 : {
680 : int nid;
681 :
682 : assert_spin_locked(&p->lock);
683 0 : for_each_node(nid)
684 0 : plist_del(&p->avail_lists[nid], &swap_avail_heads[nid]);
685 : }
686 :
687 : static void del_from_avail_list(struct swap_info_struct *p)
688 : {
689 : spin_lock(&swap_avail_lock);
690 0 : __del_from_avail_list(p);
691 0 : spin_unlock(&swap_avail_lock);
692 : }
693 :
694 0 : static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset,
695 : unsigned int nr_entries)
696 : {
697 0 : unsigned int end = offset + nr_entries - 1;
698 :
699 0 : if (offset == si->lowest_bit)
700 0 : si->lowest_bit += nr_entries;
701 0 : if (end == si->highest_bit)
702 0 : WRITE_ONCE(si->highest_bit, si->highest_bit - nr_entries);
703 0 : WRITE_ONCE(si->inuse_pages, si->inuse_pages + nr_entries);
704 0 : if (si->inuse_pages == si->pages) {
705 0 : si->lowest_bit = si->max;
706 0 : si->highest_bit = 0;
707 : del_from_avail_list(si);
708 : }
709 0 : }
710 :
711 0 : static void add_to_avail_list(struct swap_info_struct *p)
712 : {
713 : int nid;
714 :
715 0 : spin_lock(&swap_avail_lock);
716 0 : for_each_node(nid) {
717 0 : WARN_ON(!plist_node_empty(&p->avail_lists[nid]));
718 0 : plist_add(&p->avail_lists[nid], &swap_avail_heads[nid]);
719 : }
720 0 : spin_unlock(&swap_avail_lock);
721 0 : }
722 :
723 0 : static void swap_range_free(struct swap_info_struct *si, unsigned long offset,
724 : unsigned int nr_entries)
725 : {
726 0 : unsigned long begin = offset;
727 0 : unsigned long end = offset + nr_entries - 1;
728 : void (*swap_slot_free_notify)(struct block_device *, unsigned long);
729 :
730 0 : if (offset < si->lowest_bit)
731 0 : si->lowest_bit = offset;
732 0 : if (end > si->highest_bit) {
733 0 : bool was_full = !si->highest_bit;
734 :
735 0 : WRITE_ONCE(si->highest_bit, end);
736 0 : if (was_full && (si->flags & SWP_WRITEOK))
737 0 : add_to_avail_list(si);
738 : }
739 0 : atomic_long_add(nr_entries, &nr_swap_pages);
740 0 : WRITE_ONCE(si->inuse_pages, si->inuse_pages - nr_entries);
741 0 : if (si->flags & SWP_BLKDEV)
742 0 : swap_slot_free_notify =
743 0 : si->bdev->bd_disk->fops->swap_slot_free_notify;
744 : else
745 : swap_slot_free_notify = NULL;
746 0 : while (offset <= end) {
747 0 : arch_swap_invalidate_page(si->type, offset);
748 0 : frontswap_invalidate_page(si->type, offset);
749 0 : if (swap_slot_free_notify)
750 0 : swap_slot_free_notify(si->bdev, offset);
751 0 : offset++;
752 : }
753 0 : clear_shadow_from_swap_cache(si->type, begin, end);
754 0 : }
755 :
756 0 : static void set_cluster_next(struct swap_info_struct *si, unsigned long next)
757 : {
758 : unsigned long prev;
759 :
760 0 : if (!(si->flags & SWP_SOLIDSTATE)) {
761 0 : si->cluster_next = next;
762 0 : return;
763 : }
764 :
765 0 : prev = this_cpu_read(*si->cluster_next_cpu);
766 : /*
767 : * Cross the swap address space size aligned trunk, choose
768 : * another trunk randomly to avoid lock contention on swap
769 : * address space if possible.
770 : */
771 0 : if ((prev >> SWAP_ADDRESS_SPACE_SHIFT) !=
772 0 : (next >> SWAP_ADDRESS_SPACE_SHIFT)) {
773 : /* No free swap slots available */
774 0 : if (si->highest_bit <= si->lowest_bit)
775 : return;
776 0 : next = get_random_u32_inclusive(si->lowest_bit, si->highest_bit);
777 0 : next = ALIGN_DOWN(next, SWAP_ADDRESS_SPACE_PAGES);
778 0 : next = max_t(unsigned int, next, si->lowest_bit);
779 : }
780 0 : this_cpu_write(*si->cluster_next_cpu, next);
781 : }
782 :
783 : static bool swap_offset_available_and_locked(struct swap_info_struct *si,
784 : unsigned long offset)
785 : {
786 0 : if (data_race(!si->swap_map[offset])) {
787 0 : spin_lock(&si->lock);
788 : return true;
789 : }
790 :
791 0 : if (vm_swap_full() && READ_ONCE(si->swap_map[offset]) == SWAP_HAS_CACHE) {
792 0 : spin_lock(&si->lock);
793 : return true;
794 : }
795 :
796 : return false;
797 : }
798 :
799 0 : static int scan_swap_map_slots(struct swap_info_struct *si,
800 : unsigned char usage, int nr,
801 : swp_entry_t slots[])
802 : {
803 : struct swap_cluster_info *ci;
804 : unsigned long offset;
805 : unsigned long scan_base;
806 0 : unsigned long last_in_cluster = 0;
807 0 : int latency_ration = LATENCY_LIMIT;
808 0 : int n_ret = 0;
809 0 : bool scanned_many = false;
810 :
811 : /*
812 : * We try to cluster swap pages by allocating them sequentially
813 : * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
814 : * way, however, we resort to first-free allocation, starting
815 : * a new cluster. This prevents us from scattering swap pages
816 : * all over the entire swap partition, so that we reduce
817 : * overall disk seek times between swap pages. -- sct
818 : * But we do now try to find an empty cluster. -Andrea
819 : * And we let swap pages go all over an SSD partition. Hugh
820 : */
821 :
822 0 : si->flags += SWP_SCANNING;
823 : /*
824 : * Use percpu scan base for SSD to reduce lock contention on
825 : * cluster and swap cache. For HDD, sequential access is more
826 : * important.
827 : */
828 0 : if (si->flags & SWP_SOLIDSTATE)
829 0 : scan_base = this_cpu_read(*si->cluster_next_cpu);
830 : else
831 0 : scan_base = si->cluster_next;
832 0 : offset = scan_base;
833 :
834 : /* SSD algorithm */
835 0 : if (si->cluster_info) {
836 0 : if (!scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
837 : goto scan;
838 0 : } else if (unlikely(!si->cluster_nr--)) {
839 0 : if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
840 0 : si->cluster_nr = SWAPFILE_CLUSTER - 1;
841 0 : goto checks;
842 : }
843 :
844 0 : spin_unlock(&si->lock);
845 :
846 : /*
847 : * If seek is expensive, start searching for new cluster from
848 : * start of partition, to minimize the span of allocated swap.
849 : * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
850 : * case, just handled by scan_swap_map_try_ssd_cluster() above.
851 : */
852 0 : scan_base = offset = si->lowest_bit;
853 0 : last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
854 :
855 : /* Locate the first empty (unaligned) cluster */
856 0 : for (; last_in_cluster <= si->highest_bit; offset++) {
857 0 : if (si->swap_map[offset])
858 0 : last_in_cluster = offset + SWAPFILE_CLUSTER;
859 0 : else if (offset == last_in_cluster) {
860 0 : spin_lock(&si->lock);
861 0 : offset -= SWAPFILE_CLUSTER - 1;
862 0 : si->cluster_next = offset;
863 0 : si->cluster_nr = SWAPFILE_CLUSTER - 1;
864 0 : goto checks;
865 : }
866 0 : if (unlikely(--latency_ration < 0)) {
867 0 : cond_resched();
868 0 : latency_ration = LATENCY_LIMIT;
869 : }
870 : }
871 :
872 0 : offset = scan_base;
873 0 : spin_lock(&si->lock);
874 0 : si->cluster_nr = SWAPFILE_CLUSTER - 1;
875 : }
876 :
877 : checks:
878 0 : if (si->cluster_info) {
879 0 : while (scan_swap_map_ssd_cluster_conflict(si, offset)) {
880 : /* take a break if we already got some slots */
881 0 : if (n_ret)
882 : goto done;
883 0 : if (!scan_swap_map_try_ssd_cluster(si, &offset,
884 : &scan_base))
885 : goto scan;
886 : }
887 : }
888 0 : if (!(si->flags & SWP_WRITEOK))
889 : goto no_page;
890 0 : if (!si->highest_bit)
891 : goto no_page;
892 0 : if (offset > si->highest_bit)
893 0 : scan_base = offset = si->lowest_bit;
894 :
895 0 : ci = lock_cluster(si, offset);
896 : /* reuse swap entry of cache-only swap if not busy. */
897 0 : if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
898 : int swap_was_freed;
899 0 : unlock_cluster(ci);
900 0 : spin_unlock(&si->lock);
901 0 : swap_was_freed = __try_to_reclaim_swap(si, offset, TTRS_ANYWAY);
902 0 : spin_lock(&si->lock);
903 : /* entry was freed successfully, try to use this again */
904 0 : if (swap_was_freed)
905 : goto checks;
906 : goto scan; /* check next one */
907 : }
908 :
909 0 : if (si->swap_map[offset]) {
910 0 : unlock_cluster(ci);
911 0 : if (!n_ret)
912 : goto scan;
913 : else
914 : goto done;
915 : }
916 0 : WRITE_ONCE(si->swap_map[offset], usage);
917 0 : inc_cluster_info_page(si, si->cluster_info, offset);
918 0 : unlock_cluster(ci);
919 :
920 0 : swap_range_alloc(si, offset, 1);
921 0 : slots[n_ret++] = swp_entry(si->type, offset);
922 :
923 : /* got enough slots or reach max slots? */
924 0 : if ((n_ret == nr) || (offset >= si->highest_bit))
925 : goto done;
926 :
927 : /* search for next available slot */
928 :
929 : /* time to take a break? */
930 0 : if (unlikely(--latency_ration < 0)) {
931 0 : if (n_ret)
932 : goto done;
933 0 : spin_unlock(&si->lock);
934 0 : cond_resched();
935 0 : spin_lock(&si->lock);
936 0 : latency_ration = LATENCY_LIMIT;
937 : }
938 :
939 : /* try to get more slots in cluster */
940 0 : if (si->cluster_info) {
941 0 : if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
942 : goto checks;
943 0 : } else if (si->cluster_nr && !si->swap_map[++offset]) {
944 : /* non-ssd case, still more slots in cluster? */
945 0 : --si->cluster_nr;
946 0 : goto checks;
947 : }
948 :
949 : /*
950 : * Even if there's no free clusters available (fragmented),
951 : * try to scan a little more quickly with lock held unless we
952 : * have scanned too many slots already.
953 : */
954 0 : if (!scanned_many) {
955 : unsigned long scan_limit;
956 :
957 0 : if (offset < scan_base)
958 : scan_limit = scan_base;
959 : else
960 0 : scan_limit = si->highest_bit;
961 0 : for (; offset <= scan_limit && --latency_ration > 0;
962 0 : offset++) {
963 0 : if (!si->swap_map[offset])
964 : goto checks;
965 : }
966 : }
967 :
968 : done:
969 0 : set_cluster_next(si, offset + 1);
970 0 : si->flags -= SWP_SCANNING;
971 0 : return n_ret;
972 :
973 : scan:
974 0 : spin_unlock(&si->lock);
975 0 : while (++offset <= READ_ONCE(si->highest_bit)) {
976 0 : if (unlikely(--latency_ration < 0)) {
977 0 : cond_resched();
978 0 : latency_ration = LATENCY_LIMIT;
979 0 : scanned_many = true;
980 : }
981 0 : if (swap_offset_available_and_locked(si, offset))
982 : goto checks;
983 : }
984 0 : offset = si->lowest_bit;
985 0 : while (offset < scan_base) {
986 0 : if (unlikely(--latency_ration < 0)) {
987 0 : cond_resched();
988 0 : latency_ration = LATENCY_LIMIT;
989 0 : scanned_many = true;
990 : }
991 0 : if (swap_offset_available_and_locked(si, offset))
992 : goto checks;
993 0 : offset++;
994 : }
995 0 : spin_lock(&si->lock);
996 :
997 : no_page:
998 0 : si->flags -= SWP_SCANNING;
999 0 : return n_ret;
1000 : }
1001 :
1002 : static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot)
1003 : {
1004 : unsigned long idx;
1005 : struct swap_cluster_info *ci;
1006 : unsigned long offset;
1007 :
1008 : /*
1009 : * Should not even be attempting cluster allocations when huge
1010 : * page swap is disabled. Warn and fail the allocation.
1011 : */
1012 : if (!IS_ENABLED(CONFIG_THP_SWAP)) {
1013 : VM_WARN_ON_ONCE(1);
1014 : return 0;
1015 : }
1016 :
1017 : if (cluster_list_empty(&si->free_clusters))
1018 : return 0;
1019 :
1020 : idx = cluster_list_first(&si->free_clusters);
1021 : offset = idx * SWAPFILE_CLUSTER;
1022 : ci = lock_cluster(si, offset);
1023 : alloc_cluster(si, idx);
1024 : cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE);
1025 :
1026 : memset(si->swap_map + offset, SWAP_HAS_CACHE, SWAPFILE_CLUSTER);
1027 : unlock_cluster(ci);
1028 : swap_range_alloc(si, offset, SWAPFILE_CLUSTER);
1029 : *slot = swp_entry(si->type, offset);
1030 :
1031 : return 1;
1032 : }
1033 :
1034 : static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx)
1035 : {
1036 : unsigned long offset = idx * SWAPFILE_CLUSTER;
1037 : struct swap_cluster_info *ci;
1038 :
1039 : ci = lock_cluster(si, offset);
1040 : memset(si->swap_map + offset, 0, SWAPFILE_CLUSTER);
1041 : cluster_set_count_flag(ci, 0, 0);
1042 : free_cluster(si, idx);
1043 : unlock_cluster(ci);
1044 : swap_range_free(si, offset, SWAPFILE_CLUSTER);
1045 : }
1046 :
1047 0 : int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size)
1048 : {
1049 0 : unsigned long size = swap_entry_size(entry_size);
1050 : struct swap_info_struct *si, *next;
1051 : long avail_pgs;
1052 0 : int n_ret = 0;
1053 : int node;
1054 :
1055 : /* Only single cluster request supported */
1056 0 : WARN_ON_ONCE(n_goal > 1 && size == SWAPFILE_CLUSTER);
1057 :
1058 0 : spin_lock(&swap_avail_lock);
1059 :
1060 0 : avail_pgs = atomic_long_read(&nr_swap_pages) / size;
1061 0 : if (avail_pgs <= 0) {
1062 : spin_unlock(&swap_avail_lock);
1063 : goto noswap;
1064 : }
1065 :
1066 0 : n_goal = min3((long)n_goal, (long)SWAP_BATCH, avail_pgs);
1067 :
1068 0 : atomic_long_sub(n_goal * size, &nr_swap_pages);
1069 :
1070 : start_over:
1071 0 : node = numa_node_id();
1072 0 : plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) {
1073 : /* requeue si to after same-priority siblings */
1074 0 : plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]);
1075 0 : spin_unlock(&swap_avail_lock);
1076 0 : spin_lock(&si->lock);
1077 0 : if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) {
1078 0 : spin_lock(&swap_avail_lock);
1079 0 : if (plist_node_empty(&si->avail_lists[node])) {
1080 0 : spin_unlock(&si->lock);
1081 : goto nextsi;
1082 : }
1083 0 : WARN(!si->highest_bit,
1084 : "swap_info %d in list but !highest_bit\n",
1085 : si->type);
1086 0 : WARN(!(si->flags & SWP_WRITEOK),
1087 : "swap_info %d in list but !SWP_WRITEOK\n",
1088 : si->type);
1089 0 : __del_from_avail_list(si);
1090 0 : spin_unlock(&si->lock);
1091 : goto nextsi;
1092 : }
1093 : if (size == SWAPFILE_CLUSTER) {
1094 : if (si->flags & SWP_BLKDEV)
1095 : n_ret = swap_alloc_cluster(si, swp_entries);
1096 : } else
1097 0 : n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE,
1098 : n_goal, swp_entries);
1099 0 : spin_unlock(&si->lock);
1100 0 : if (n_ret || size == SWAPFILE_CLUSTER)
1101 : goto check_out;
1102 0 : cond_resched();
1103 :
1104 : spin_lock(&swap_avail_lock);
1105 : nextsi:
1106 : /*
1107 : * if we got here, it's likely that si was almost full before,
1108 : * and since scan_swap_map_slots() can drop the si->lock,
1109 : * multiple callers probably all tried to get a page from the
1110 : * same si and it filled up before we could get one; or, the si
1111 : * filled up between us dropping swap_avail_lock and taking
1112 : * si->lock. Since we dropped the swap_avail_lock, the
1113 : * swap_avail_head list may have been modified; so if next is
1114 : * still in the swap_avail_head list then try it, otherwise
1115 : * start over if we have not gotten any slots.
1116 : */
1117 0 : if (plist_node_empty(&next->avail_lists[node]))
1118 : goto start_over;
1119 : }
1120 :
1121 : spin_unlock(&swap_avail_lock);
1122 :
1123 : check_out:
1124 0 : if (n_ret < n_goal)
1125 0 : atomic_long_add((long)(n_goal - n_ret) * size,
1126 : &nr_swap_pages);
1127 : noswap:
1128 0 : return n_ret;
1129 : }
1130 :
1131 0 : static struct swap_info_struct *_swap_info_get(swp_entry_t entry)
1132 : {
1133 : struct swap_info_struct *p;
1134 : unsigned long offset;
1135 :
1136 0 : if (!entry.val)
1137 : goto out;
1138 0 : p = swp_swap_info(entry);
1139 0 : if (!p)
1140 : goto bad_nofile;
1141 0 : if (data_race(!(p->flags & SWP_USED)))
1142 : goto bad_device;
1143 0 : offset = swp_offset(entry);
1144 0 : if (offset >= p->max)
1145 : goto bad_offset;
1146 0 : if (data_race(!p->swap_map[swp_offset(entry)]))
1147 : goto bad_free;
1148 : return p;
1149 :
1150 : bad_free:
1151 0 : pr_err("%s: %s%08lx\n", __func__, Unused_offset, entry.val);
1152 0 : goto out;
1153 : bad_offset:
1154 0 : pr_err("%s: %s%08lx\n", __func__, Bad_offset, entry.val);
1155 0 : goto out;
1156 : bad_device:
1157 0 : pr_err("%s: %s%08lx\n", __func__, Unused_file, entry.val);
1158 0 : goto out;
1159 : bad_nofile:
1160 0 : pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val);
1161 : out:
1162 : return NULL;
1163 : }
1164 :
1165 : static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry,
1166 : struct swap_info_struct *q)
1167 : {
1168 : struct swap_info_struct *p;
1169 :
1170 0 : p = _swap_info_get(entry);
1171 :
1172 0 : if (p != q) {
1173 0 : if (q != NULL)
1174 0 : spin_unlock(&q->lock);
1175 0 : if (p != NULL)
1176 0 : spin_lock(&p->lock);
1177 : }
1178 : return p;
1179 : }
1180 :
1181 0 : static unsigned char __swap_entry_free_locked(struct swap_info_struct *p,
1182 : unsigned long offset,
1183 : unsigned char usage)
1184 : {
1185 : unsigned char count;
1186 : unsigned char has_cache;
1187 :
1188 0 : count = p->swap_map[offset];
1189 :
1190 0 : has_cache = count & SWAP_HAS_CACHE;
1191 0 : count &= ~SWAP_HAS_CACHE;
1192 :
1193 0 : if (usage == SWAP_HAS_CACHE) {
1194 : VM_BUG_ON(!has_cache);
1195 : has_cache = 0;
1196 0 : } else if (count == SWAP_MAP_SHMEM) {
1197 : /*
1198 : * Or we could insist on shmem.c using a special
1199 : * swap_shmem_free() and free_shmem_swap_and_cache()...
1200 : */
1201 : count = 0;
1202 0 : } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
1203 0 : if (count == COUNT_CONTINUED) {
1204 0 : if (swap_count_continued(p, offset, count))
1205 : count = SWAP_MAP_MAX | COUNT_CONTINUED;
1206 : else
1207 0 : count = SWAP_MAP_MAX;
1208 : } else
1209 0 : count--;
1210 : }
1211 :
1212 0 : usage = count | has_cache;
1213 0 : if (usage)
1214 0 : WRITE_ONCE(p->swap_map[offset], usage);
1215 : else
1216 0 : WRITE_ONCE(p->swap_map[offset], SWAP_HAS_CACHE);
1217 :
1218 0 : return usage;
1219 : }
1220 :
1221 : /*
1222 : * Check whether swap entry is valid in the swap device. If so,
1223 : * return pointer to swap_info_struct, and keep the swap entry valid
1224 : * via preventing the swap device from being swapoff, until
1225 : * put_swap_device() is called. Otherwise return NULL.
1226 : *
1227 : * Notice that swapoff or swapoff+swapon can still happen before the
1228 : * percpu_ref_tryget_live() in get_swap_device() or after the
1229 : * percpu_ref_put() in put_swap_device() if there isn't any other way
1230 : * to prevent swapoff, such as page lock, page table lock, etc. The
1231 : * caller must be prepared for that. For example, the following
1232 : * situation is possible.
1233 : *
1234 : * CPU1 CPU2
1235 : * do_swap_page()
1236 : * ... swapoff+swapon
1237 : * __read_swap_cache_async()
1238 : * swapcache_prepare()
1239 : * __swap_duplicate()
1240 : * // check swap_map
1241 : * // verify PTE not changed
1242 : *
1243 : * In __swap_duplicate(), the swap_map need to be checked before
1244 : * changing partly because the specified swap entry may be for another
1245 : * swap device which has been swapoff. And in do_swap_page(), after
1246 : * the page is read from the swap device, the PTE is verified not
1247 : * changed with the page table locked to check whether the swap device
1248 : * has been swapoff or swapoff+swapon.
1249 : */
1250 0 : struct swap_info_struct *get_swap_device(swp_entry_t entry)
1251 : {
1252 : struct swap_info_struct *si;
1253 : unsigned long offset;
1254 :
1255 0 : if (!entry.val)
1256 : goto out;
1257 0 : si = swp_swap_info(entry);
1258 0 : if (!si)
1259 : goto bad_nofile;
1260 0 : if (!percpu_ref_tryget_live(&si->users))
1261 : goto out;
1262 : /*
1263 : * Guarantee the si->users are checked before accessing other
1264 : * fields of swap_info_struct.
1265 : *
1266 : * Paired with the spin_unlock() after setup_swap_info() in
1267 : * enable_swap_info().
1268 : */
1269 0 : smp_rmb();
1270 0 : offset = swp_offset(entry);
1271 0 : if (offset >= si->max)
1272 : goto put_out;
1273 :
1274 : return si;
1275 : bad_nofile:
1276 0 : pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val);
1277 : out:
1278 : return NULL;
1279 : put_out:
1280 0 : pr_err("%s: %s%08lx\n", __func__, Bad_offset, entry.val);
1281 0 : percpu_ref_put(&si->users);
1282 0 : return NULL;
1283 : }
1284 :
1285 0 : static unsigned char __swap_entry_free(struct swap_info_struct *p,
1286 : swp_entry_t entry)
1287 : {
1288 : struct swap_cluster_info *ci;
1289 0 : unsigned long offset = swp_offset(entry);
1290 : unsigned char usage;
1291 :
1292 0 : ci = lock_cluster_or_swap_info(p, offset);
1293 0 : usage = __swap_entry_free_locked(p, offset, 1);
1294 0 : unlock_cluster_or_swap_info(p, ci);
1295 0 : if (!usage)
1296 0 : free_swap_slot(entry);
1297 :
1298 0 : return usage;
1299 : }
1300 :
1301 0 : static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry)
1302 : {
1303 : struct swap_cluster_info *ci;
1304 0 : unsigned long offset = swp_offset(entry);
1305 : unsigned char count;
1306 :
1307 0 : ci = lock_cluster(p, offset);
1308 0 : count = p->swap_map[offset];
1309 : VM_BUG_ON(count != SWAP_HAS_CACHE);
1310 0 : p->swap_map[offset] = 0;
1311 0 : dec_cluster_info_page(p, p->cluster_info, offset);
1312 0 : unlock_cluster(ci);
1313 :
1314 0 : mem_cgroup_uncharge_swap(entry, 1);
1315 0 : swap_range_free(p, offset, 1);
1316 0 : }
1317 :
1318 : /*
1319 : * Caller has made sure that the swap device corresponding to entry
1320 : * is still around or has not been recycled.
1321 : */
1322 0 : void swap_free(swp_entry_t entry)
1323 : {
1324 : struct swap_info_struct *p;
1325 :
1326 0 : p = _swap_info_get(entry);
1327 0 : if (p)
1328 0 : __swap_entry_free(p, entry);
1329 0 : }
1330 :
1331 : /*
1332 : * Called after dropping swapcache to decrease refcnt to swap entries.
1333 : */
1334 0 : void put_swap_folio(struct folio *folio, swp_entry_t entry)
1335 : {
1336 0 : unsigned long offset = swp_offset(entry);
1337 0 : unsigned long idx = offset / SWAPFILE_CLUSTER;
1338 : struct swap_cluster_info *ci;
1339 : struct swap_info_struct *si;
1340 : unsigned char *map;
1341 0 : unsigned int i, free_entries = 0;
1342 : unsigned char val;
1343 0 : int size = swap_entry_size(folio_nr_pages(folio));
1344 :
1345 0 : si = _swap_info_get(entry);
1346 0 : if (!si)
1347 : return;
1348 :
1349 : ci = lock_cluster_or_swap_info(si, offset);
1350 : if (size == SWAPFILE_CLUSTER) {
1351 : VM_BUG_ON(!cluster_is_huge(ci));
1352 : map = si->swap_map + offset;
1353 : for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1354 : val = map[i];
1355 : VM_BUG_ON(!(val & SWAP_HAS_CACHE));
1356 : if (val == SWAP_HAS_CACHE)
1357 : free_entries++;
1358 : }
1359 : cluster_clear_huge(ci);
1360 : if (free_entries == SWAPFILE_CLUSTER) {
1361 : unlock_cluster_or_swap_info(si, ci);
1362 : spin_lock(&si->lock);
1363 : mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER);
1364 : swap_free_cluster(si, idx);
1365 : spin_unlock(&si->lock);
1366 : return;
1367 : }
1368 : }
1369 0 : for (i = 0; i < size; i++, entry.val++) {
1370 0 : if (!__swap_entry_free_locked(si, offset + i, SWAP_HAS_CACHE)) {
1371 0 : unlock_cluster_or_swap_info(si, ci);
1372 0 : free_swap_slot(entry);
1373 : if (i == size - 1)
1374 : return;
1375 : lock_cluster_or_swap_info(si, offset);
1376 : }
1377 : }
1378 0 : unlock_cluster_or_swap_info(si, ci);
1379 : }
1380 :
1381 : #ifdef CONFIG_THP_SWAP
1382 : int split_swap_cluster(swp_entry_t entry)
1383 : {
1384 : struct swap_info_struct *si;
1385 : struct swap_cluster_info *ci;
1386 : unsigned long offset = swp_offset(entry);
1387 :
1388 : si = _swap_info_get(entry);
1389 : if (!si)
1390 : return -EBUSY;
1391 : ci = lock_cluster(si, offset);
1392 : cluster_clear_huge(ci);
1393 : unlock_cluster(ci);
1394 : return 0;
1395 : }
1396 : #endif
1397 :
1398 0 : static int swp_entry_cmp(const void *ent1, const void *ent2)
1399 : {
1400 0 : const swp_entry_t *e1 = ent1, *e2 = ent2;
1401 :
1402 0 : return (int)swp_type(*e1) - (int)swp_type(*e2);
1403 : }
1404 :
1405 0 : void swapcache_free_entries(swp_entry_t *entries, int n)
1406 : {
1407 : struct swap_info_struct *p, *prev;
1408 : int i;
1409 :
1410 0 : if (n <= 0)
1411 : return;
1412 :
1413 0 : prev = NULL;
1414 0 : p = NULL;
1415 :
1416 : /*
1417 : * Sort swap entries by swap device, so each lock is only taken once.
1418 : * nr_swapfiles isn't absolutely correct, but the overhead of sort() is
1419 : * so low that it isn't necessary to optimize further.
1420 : */
1421 0 : if (nr_swapfiles > 1)
1422 0 : sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL);
1423 0 : for (i = 0; i < n; ++i) {
1424 0 : p = swap_info_get_cont(entries[i], prev);
1425 0 : if (p)
1426 0 : swap_entry_free(p, entries[i]);
1427 0 : prev = p;
1428 : }
1429 0 : if (p)
1430 0 : spin_unlock(&p->lock);
1431 : }
1432 :
1433 0 : int __swap_count(swp_entry_t entry)
1434 : {
1435 : struct swap_info_struct *si;
1436 0 : pgoff_t offset = swp_offset(entry);
1437 0 : int count = 0;
1438 :
1439 0 : si = get_swap_device(entry);
1440 0 : if (si) {
1441 0 : count = swap_count(si->swap_map[offset]);
1442 : put_swap_device(si);
1443 : }
1444 0 : return count;
1445 : }
1446 :
1447 : /*
1448 : * How many references to @entry are currently swapped out?
1449 : * This does not give an exact answer when swap count is continued,
1450 : * but does include the high COUNT_CONTINUED flag to allow for that.
1451 : */
1452 : static int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry)
1453 : {
1454 0 : pgoff_t offset = swp_offset(entry);
1455 : struct swap_cluster_info *ci;
1456 : int count;
1457 :
1458 0 : ci = lock_cluster_or_swap_info(si, offset);
1459 0 : count = swap_count(si->swap_map[offset]);
1460 0 : unlock_cluster_or_swap_info(si, ci);
1461 : return count;
1462 : }
1463 :
1464 : /*
1465 : * How many references to @entry are currently swapped out?
1466 : * This does not give an exact answer when swap count is continued,
1467 : * but does include the high COUNT_CONTINUED flag to allow for that.
1468 : */
1469 0 : int __swp_swapcount(swp_entry_t entry)
1470 : {
1471 0 : int count = 0;
1472 : struct swap_info_struct *si;
1473 :
1474 0 : si = get_swap_device(entry);
1475 0 : if (si) {
1476 0 : count = swap_swapcount(si, entry);
1477 : put_swap_device(si);
1478 : }
1479 0 : return count;
1480 : }
1481 :
1482 : /*
1483 : * How many references to @entry are currently swapped out?
1484 : * This considers COUNT_CONTINUED so it returns exact answer.
1485 : */
1486 0 : int swp_swapcount(swp_entry_t entry)
1487 : {
1488 : int count, tmp_count, n;
1489 : struct swap_info_struct *p;
1490 : struct swap_cluster_info *ci;
1491 : struct page *page;
1492 : pgoff_t offset;
1493 : unsigned char *map;
1494 :
1495 0 : p = _swap_info_get(entry);
1496 0 : if (!p)
1497 : return 0;
1498 :
1499 0 : offset = swp_offset(entry);
1500 :
1501 0 : ci = lock_cluster_or_swap_info(p, offset);
1502 :
1503 0 : count = swap_count(p->swap_map[offset]);
1504 0 : if (!(count & COUNT_CONTINUED))
1505 : goto out;
1506 :
1507 0 : count &= ~COUNT_CONTINUED;
1508 0 : n = SWAP_MAP_MAX + 1;
1509 :
1510 0 : page = vmalloc_to_page(p->swap_map + offset);
1511 0 : offset &= ~PAGE_MASK;
1512 : VM_BUG_ON(page_private(page) != SWP_CONTINUED);
1513 :
1514 : do {
1515 0 : page = list_next_entry(page, lru);
1516 0 : map = kmap_atomic(page);
1517 0 : tmp_count = map[offset];
1518 0 : kunmap_atomic(map);
1519 :
1520 0 : count += (tmp_count & ~COUNT_CONTINUED) * n;
1521 0 : n *= (SWAP_CONT_MAX + 1);
1522 0 : } while (tmp_count & COUNT_CONTINUED);
1523 : out:
1524 0 : unlock_cluster_or_swap_info(p, ci);
1525 : return count;
1526 : }
1527 :
1528 : static bool swap_page_trans_huge_swapped(struct swap_info_struct *si,
1529 : swp_entry_t entry)
1530 : {
1531 : struct swap_cluster_info *ci;
1532 0 : unsigned char *map = si->swap_map;
1533 0 : unsigned long roffset = swp_offset(entry);
1534 0 : unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER);
1535 : int i;
1536 0 : bool ret = false;
1537 :
1538 0 : ci = lock_cluster_or_swap_info(si, offset);
1539 : if (!ci || !cluster_is_huge(ci)) {
1540 0 : if (swap_count(map[roffset]))
1541 0 : ret = true;
1542 : goto unlock_out;
1543 : }
1544 : for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1545 : if (swap_count(map[offset + i])) {
1546 : ret = true;
1547 : break;
1548 : }
1549 : }
1550 : unlock_out:
1551 0 : unlock_cluster_or_swap_info(si, ci);
1552 : return ret;
1553 : }
1554 :
1555 0 : static bool folio_swapped(struct folio *folio)
1556 : {
1557 0 : swp_entry_t entry = folio_swap_entry(folio);
1558 0 : struct swap_info_struct *si = _swap_info_get(entry);
1559 :
1560 0 : if (!si)
1561 : return false;
1562 :
1563 : if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!folio_test_large(folio)))
1564 0 : return swap_swapcount(si, entry) != 0;
1565 :
1566 : return swap_page_trans_huge_swapped(si, entry);
1567 : }
1568 :
1569 : /**
1570 : * folio_free_swap() - Free the swap space used for this folio.
1571 : * @folio: The folio to remove.
1572 : *
1573 : * If swap is getting full, or if there are no more mappings of this folio,
1574 : * then call folio_free_swap to free its swap space.
1575 : *
1576 : * Return: true if we were able to release the swap space.
1577 : */
1578 0 : bool folio_free_swap(struct folio *folio)
1579 : {
1580 : VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1581 :
1582 0 : if (!folio_test_swapcache(folio))
1583 : return false;
1584 0 : if (folio_test_writeback(folio))
1585 : return false;
1586 0 : if (folio_swapped(folio))
1587 : return false;
1588 :
1589 : /*
1590 : * Once hibernation has begun to create its image of memory,
1591 : * there's a danger that one of the calls to folio_free_swap()
1592 : * - most probably a call from __try_to_reclaim_swap() while
1593 : * hibernation is allocating its own swap pages for the image,
1594 : * but conceivably even a call from memory reclaim - will free
1595 : * the swap from a folio which has already been recorded in the
1596 : * image as a clean swapcache folio, and then reuse its swap for
1597 : * another page of the image. On waking from hibernation, the
1598 : * original folio might be freed under memory pressure, then
1599 : * later read back in from swap, now with the wrong data.
1600 : *
1601 : * Hibernation suspends storage while it is writing the image
1602 : * to disk so check that here.
1603 : */
1604 0 : if (pm_suspended_storage())
1605 : return false;
1606 :
1607 0 : delete_from_swap_cache(folio);
1608 0 : folio_set_dirty(folio);
1609 0 : return true;
1610 : }
1611 :
1612 : /*
1613 : * Free the swap entry like above, but also try to
1614 : * free the page cache entry if it is the last user.
1615 : */
1616 0 : int free_swap_and_cache(swp_entry_t entry)
1617 : {
1618 : struct swap_info_struct *p;
1619 : unsigned char count;
1620 :
1621 0 : if (non_swap_entry(entry))
1622 : return 1;
1623 :
1624 0 : p = _swap_info_get(entry);
1625 0 : if (p) {
1626 0 : count = __swap_entry_free(p, entry);
1627 0 : if (count == SWAP_HAS_CACHE &&
1628 0 : !swap_page_trans_huge_swapped(p, entry))
1629 0 : __try_to_reclaim_swap(p, swp_offset(entry),
1630 : TTRS_UNMAPPED | TTRS_FULL);
1631 : }
1632 0 : return p != NULL;
1633 : }
1634 :
1635 : #ifdef CONFIG_HIBERNATION
1636 :
1637 : swp_entry_t get_swap_page_of_type(int type)
1638 : {
1639 : struct swap_info_struct *si = swap_type_to_swap_info(type);
1640 : swp_entry_t entry = {0};
1641 :
1642 : if (!si)
1643 : goto fail;
1644 :
1645 : /* This is called for allocating swap entry, not cache */
1646 : spin_lock(&si->lock);
1647 : if ((si->flags & SWP_WRITEOK) && scan_swap_map_slots(si, 1, 1, &entry))
1648 : atomic_long_dec(&nr_swap_pages);
1649 : spin_unlock(&si->lock);
1650 : fail:
1651 : return entry;
1652 : }
1653 :
1654 : /*
1655 : * Find the swap type that corresponds to given device (if any).
1656 : *
1657 : * @offset - number of the PAGE_SIZE-sized block of the device, starting
1658 : * from 0, in which the swap header is expected to be located.
1659 : *
1660 : * This is needed for the suspend to disk (aka swsusp).
1661 : */
1662 : int swap_type_of(dev_t device, sector_t offset)
1663 : {
1664 : int type;
1665 :
1666 : if (!device)
1667 : return -1;
1668 :
1669 : spin_lock(&swap_lock);
1670 : for (type = 0; type < nr_swapfiles; type++) {
1671 : struct swap_info_struct *sis = swap_info[type];
1672 :
1673 : if (!(sis->flags & SWP_WRITEOK))
1674 : continue;
1675 :
1676 : if (device == sis->bdev->bd_dev) {
1677 : struct swap_extent *se = first_se(sis);
1678 :
1679 : if (se->start_block == offset) {
1680 : spin_unlock(&swap_lock);
1681 : return type;
1682 : }
1683 : }
1684 : }
1685 : spin_unlock(&swap_lock);
1686 : return -ENODEV;
1687 : }
1688 :
1689 : int find_first_swap(dev_t *device)
1690 : {
1691 : int type;
1692 :
1693 : spin_lock(&swap_lock);
1694 : for (type = 0; type < nr_swapfiles; type++) {
1695 : struct swap_info_struct *sis = swap_info[type];
1696 :
1697 : if (!(sis->flags & SWP_WRITEOK))
1698 : continue;
1699 : *device = sis->bdev->bd_dev;
1700 : spin_unlock(&swap_lock);
1701 : return type;
1702 : }
1703 : spin_unlock(&swap_lock);
1704 : return -ENODEV;
1705 : }
1706 :
1707 : /*
1708 : * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1709 : * corresponding to given index in swap_info (swap type).
1710 : */
1711 : sector_t swapdev_block(int type, pgoff_t offset)
1712 : {
1713 : struct swap_info_struct *si = swap_type_to_swap_info(type);
1714 : struct swap_extent *se;
1715 :
1716 : if (!si || !(si->flags & SWP_WRITEOK))
1717 : return 0;
1718 : se = offset_to_swap_extent(si, offset);
1719 : return se->start_block + (offset - se->start_page);
1720 : }
1721 :
1722 : /*
1723 : * Return either the total number of swap pages of given type, or the number
1724 : * of free pages of that type (depending on @free)
1725 : *
1726 : * This is needed for software suspend
1727 : */
1728 : unsigned int count_swap_pages(int type, int free)
1729 : {
1730 : unsigned int n = 0;
1731 :
1732 : spin_lock(&swap_lock);
1733 : if ((unsigned int)type < nr_swapfiles) {
1734 : struct swap_info_struct *sis = swap_info[type];
1735 :
1736 : spin_lock(&sis->lock);
1737 : if (sis->flags & SWP_WRITEOK) {
1738 : n = sis->pages;
1739 : if (free)
1740 : n -= sis->inuse_pages;
1741 : }
1742 : spin_unlock(&sis->lock);
1743 : }
1744 : spin_unlock(&swap_lock);
1745 : return n;
1746 : }
1747 : #endif /* CONFIG_HIBERNATION */
1748 :
1749 : static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte)
1750 : {
1751 0 : return pte_same(pte_swp_clear_flags(pte), swp_pte);
1752 : }
1753 :
1754 : /*
1755 : * No need to decide whether this PTE shares the swap entry with others,
1756 : * just let do_wp_page work it out if a write is requested later - to
1757 : * force COW, vm_page_prot omits write permission from any private vma.
1758 : */
1759 0 : static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
1760 : unsigned long addr, swp_entry_t entry, struct folio *folio)
1761 : {
1762 0 : struct page *page = folio_file_page(folio, swp_offset(entry));
1763 : struct page *swapcache;
1764 : spinlock_t *ptl;
1765 : pte_t *pte, new_pte;
1766 0 : bool hwposioned = false;
1767 0 : int ret = 1;
1768 :
1769 0 : swapcache = page;
1770 0 : page = ksm_might_need_to_copy(page, vma, addr);
1771 0 : if (unlikely(!page))
1772 : return -ENOMEM;
1773 0 : else if (unlikely(PTR_ERR(page) == -EHWPOISON))
1774 0 : hwposioned = true;
1775 :
1776 0 : pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1777 0 : if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) {
1778 : ret = 0;
1779 : goto out;
1780 : }
1781 :
1782 0 : if (unlikely(hwposioned || !PageUptodate(page))) {
1783 : swp_entry_t swp_entry;
1784 :
1785 0 : dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1786 0 : if (hwposioned) {
1787 : swp_entry = make_hwpoison_entry(swapcache);
1788 : page = swapcache;
1789 : } else {
1790 : swp_entry = make_swapin_error_entry();
1791 : }
1792 0 : new_pte = swp_entry_to_pte(swp_entry);
1793 0 : ret = 0;
1794 : goto setpte;
1795 : }
1796 :
1797 : /* See do_swap_page() */
1798 0 : BUG_ON(!PageAnon(page) && PageMappedToDisk(page));
1799 0 : BUG_ON(PageAnon(page) && PageAnonExclusive(page));
1800 :
1801 0 : dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1802 0 : inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
1803 0 : get_page(page);
1804 : if (page == swapcache) {
1805 0 : rmap_t rmap_flags = RMAP_NONE;
1806 :
1807 : /*
1808 : * See do_swap_page(): PageWriteback() would be problematic.
1809 : * However, we do a wait_on_page_writeback() just before this
1810 : * call and have the page locked.
1811 : */
1812 : VM_BUG_ON_PAGE(PageWriteback(page), page);
1813 0 : if (pte_swp_exclusive(*pte))
1814 0 : rmap_flags |= RMAP_EXCLUSIVE;
1815 :
1816 0 : page_add_anon_rmap(page, vma, addr, rmap_flags);
1817 : } else { /* ksm created a completely new copy */
1818 : page_add_new_anon_rmap(page, vma, addr);
1819 : lru_cache_add_inactive_or_unevictable(page, vma);
1820 : }
1821 0 : new_pte = pte_mkold(mk_pte(page, vma->vm_page_prot));
1822 0 : if (pte_swp_soft_dirty(*pte))
1823 : new_pte = pte_mksoft_dirty(new_pte);
1824 : if (pte_swp_uffd_wp(*pte))
1825 : new_pte = pte_mkuffd_wp(new_pte);
1826 : setpte:
1827 0 : set_pte_at(vma->vm_mm, addr, pte, new_pte);
1828 0 : swap_free(entry);
1829 : out:
1830 0 : pte_unmap_unlock(pte, ptl);
1831 : if (page != swapcache) {
1832 : unlock_page(page);
1833 : put_page(page);
1834 : }
1835 0 : return ret;
1836 : }
1837 :
1838 0 : static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1839 : unsigned long addr, unsigned long end,
1840 : unsigned int type)
1841 : {
1842 : swp_entry_t entry;
1843 : pte_t *pte;
1844 : struct swap_info_struct *si;
1845 0 : int ret = 0;
1846 :
1847 0 : si = swap_info[type];
1848 0 : pte = pte_offset_map(pmd, addr);
1849 : do {
1850 : struct folio *folio;
1851 : unsigned long offset;
1852 : unsigned char swp_count;
1853 :
1854 0 : if (!is_swap_pte(*pte))
1855 0 : continue;
1856 :
1857 0 : entry = pte_to_swp_entry(*pte);
1858 0 : if (swp_type(entry) != type)
1859 0 : continue;
1860 :
1861 0 : offset = swp_offset(entry);
1862 : pte_unmap(pte);
1863 0 : folio = swap_cache_get_folio(entry, vma, addr);
1864 0 : if (!folio) {
1865 : struct page *page;
1866 0 : struct vm_fault vmf = {
1867 : .vma = vma,
1868 : .address = addr,
1869 : .real_address = addr,
1870 : .pmd = pmd,
1871 : };
1872 :
1873 0 : page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE,
1874 : &vmf);
1875 0 : if (page)
1876 0 : folio = page_folio(page);
1877 : }
1878 0 : if (!folio) {
1879 0 : swp_count = READ_ONCE(si->swap_map[offset]);
1880 0 : if (swp_count == 0 || swp_count == SWAP_MAP_BAD)
1881 : goto try_next;
1882 :
1883 : return -ENOMEM;
1884 : }
1885 :
1886 0 : folio_lock(folio);
1887 0 : folio_wait_writeback(folio);
1888 0 : ret = unuse_pte(vma, pmd, addr, entry, folio);
1889 0 : if (ret < 0) {
1890 0 : folio_unlock(folio);
1891 : folio_put(folio);
1892 : goto out;
1893 : }
1894 :
1895 0 : folio_free_swap(folio);
1896 0 : folio_unlock(folio);
1897 : folio_put(folio);
1898 : try_next:
1899 0 : pte = pte_offset_map(pmd, addr);
1900 0 : } while (pte++, addr += PAGE_SIZE, addr != end);
1901 : pte_unmap(pte - 1);
1902 :
1903 : ret = 0;
1904 : out:
1905 : return ret;
1906 : }
1907 :
1908 0 : static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1909 : unsigned long addr, unsigned long end,
1910 : unsigned int type)
1911 : {
1912 : pmd_t *pmd;
1913 : unsigned long next;
1914 : int ret;
1915 :
1916 0 : pmd = pmd_offset(pud, addr);
1917 : do {
1918 0 : cond_resched();
1919 0 : next = pmd_addr_end(addr, end);
1920 0 : if (pmd_none_or_trans_huge_or_clear_bad(pmd))
1921 0 : continue;
1922 0 : ret = unuse_pte_range(vma, pmd, addr, next, type);
1923 0 : if (ret)
1924 : return ret;
1925 0 : } while (pmd++, addr = next, addr != end);
1926 : return 0;
1927 : }
1928 :
1929 0 : static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d,
1930 : unsigned long addr, unsigned long end,
1931 : unsigned int type)
1932 : {
1933 : pud_t *pud;
1934 : unsigned long next;
1935 : int ret;
1936 :
1937 0 : pud = pud_offset(p4d, addr);
1938 : do {
1939 0 : next = pud_addr_end(addr, end);
1940 0 : if (pud_none_or_clear_bad(pud))
1941 0 : continue;
1942 0 : ret = unuse_pmd_range(vma, pud, addr, next, type);
1943 0 : if (ret)
1944 : return ret;
1945 0 : } while (pud++, addr = next, addr != end);
1946 0 : return 0;
1947 : }
1948 :
1949 : static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd,
1950 : unsigned long addr, unsigned long end,
1951 : unsigned int type)
1952 : {
1953 : p4d_t *p4d;
1954 : unsigned long next;
1955 : int ret;
1956 :
1957 0 : p4d = p4d_offset(pgd, addr);
1958 : do {
1959 0 : next = p4d_addr_end(addr, end);
1960 0 : if (p4d_none_or_clear_bad(p4d))
1961 : continue;
1962 0 : ret = unuse_pud_range(vma, p4d, addr, next, type);
1963 0 : if (ret)
1964 : return ret;
1965 0 : } while (p4d++, addr = next, addr != end);
1966 : return 0;
1967 : }
1968 :
1969 0 : static int unuse_vma(struct vm_area_struct *vma, unsigned int type)
1970 : {
1971 : pgd_t *pgd;
1972 : unsigned long addr, end, next;
1973 : int ret;
1974 :
1975 0 : addr = vma->vm_start;
1976 0 : end = vma->vm_end;
1977 :
1978 0 : pgd = pgd_offset(vma->vm_mm, addr);
1979 : do {
1980 0 : next = pgd_addr_end(addr, end);
1981 0 : if (pgd_none_or_clear_bad(pgd))
1982 : continue;
1983 0 : ret = unuse_p4d_range(vma, pgd, addr, next, type);
1984 0 : if (ret)
1985 : return ret;
1986 0 : } while (pgd++, addr = next, addr != end);
1987 : return 0;
1988 : }
1989 :
1990 0 : static int unuse_mm(struct mm_struct *mm, unsigned int type)
1991 : {
1992 : struct vm_area_struct *vma;
1993 0 : int ret = 0;
1994 0 : VMA_ITERATOR(vmi, mm, 0);
1995 :
1996 : mmap_read_lock(mm);
1997 0 : for_each_vma(vmi, vma) {
1998 0 : if (vma->anon_vma) {
1999 0 : ret = unuse_vma(vma, type);
2000 0 : if (ret)
2001 : break;
2002 : }
2003 :
2004 0 : cond_resched();
2005 : }
2006 0 : mmap_read_unlock(mm);
2007 0 : return ret;
2008 : }
2009 :
2010 : /*
2011 : * Scan swap_map from current position to next entry still in use.
2012 : * Return 0 if there are no inuse entries after prev till end of
2013 : * the map.
2014 : */
2015 0 : static unsigned int find_next_to_unuse(struct swap_info_struct *si,
2016 : unsigned int prev)
2017 : {
2018 : unsigned int i;
2019 : unsigned char count;
2020 :
2021 : /*
2022 : * No need for swap_lock here: we're just looking
2023 : * for whether an entry is in use, not modifying it; false
2024 : * hits are okay, and sys_swapoff() has already prevented new
2025 : * allocations from this area (while holding swap_lock).
2026 : */
2027 0 : for (i = prev + 1; i < si->max; i++) {
2028 0 : count = READ_ONCE(si->swap_map[i]);
2029 0 : if (count && swap_count(count) != SWAP_MAP_BAD)
2030 : break;
2031 0 : if ((i % LATENCY_LIMIT) == 0)
2032 0 : cond_resched();
2033 : }
2034 :
2035 0 : if (i == si->max)
2036 0 : i = 0;
2037 :
2038 0 : return i;
2039 : }
2040 :
2041 0 : static int try_to_unuse(unsigned int type)
2042 : {
2043 : struct mm_struct *prev_mm;
2044 : struct mm_struct *mm;
2045 : struct list_head *p;
2046 0 : int retval = 0;
2047 0 : struct swap_info_struct *si = swap_info[type];
2048 : struct folio *folio;
2049 : swp_entry_t entry;
2050 : unsigned int i;
2051 :
2052 0 : if (!READ_ONCE(si->inuse_pages))
2053 : return 0;
2054 :
2055 : retry:
2056 0 : retval = shmem_unuse(type);
2057 0 : if (retval)
2058 : return retval;
2059 :
2060 0 : prev_mm = &init_mm;
2061 0 : mmget(prev_mm);
2062 :
2063 0 : spin_lock(&mmlist_lock);
2064 0 : p = &init_mm.mmlist;
2065 0 : while (READ_ONCE(si->inuse_pages) &&
2066 0 : !signal_pending(current) &&
2067 0 : (p = p->next) != &init_mm.mmlist) {
2068 :
2069 0 : mm = list_entry(p, struct mm_struct, mmlist);
2070 0 : if (!mmget_not_zero(mm))
2071 0 : continue;
2072 0 : spin_unlock(&mmlist_lock);
2073 0 : mmput(prev_mm);
2074 0 : prev_mm = mm;
2075 0 : retval = unuse_mm(mm, type);
2076 0 : if (retval) {
2077 0 : mmput(prev_mm);
2078 0 : return retval;
2079 : }
2080 :
2081 : /*
2082 : * Make sure that we aren't completely killing
2083 : * interactive performance.
2084 : */
2085 0 : cond_resched();
2086 : spin_lock(&mmlist_lock);
2087 : }
2088 0 : spin_unlock(&mmlist_lock);
2089 :
2090 0 : mmput(prev_mm);
2091 :
2092 0 : i = 0;
2093 0 : while (READ_ONCE(si->inuse_pages) &&
2094 0 : !signal_pending(current) &&
2095 : (i = find_next_to_unuse(si, i)) != 0) {
2096 :
2097 0 : entry = swp_entry(type, i);
2098 0 : folio = filemap_get_folio(swap_address_space(entry), i);
2099 0 : if (IS_ERR(folio))
2100 0 : continue;
2101 :
2102 : /*
2103 : * It is conceivable that a racing task removed this folio from
2104 : * swap cache just before we acquired the page lock. The folio
2105 : * might even be back in swap cache on another swap area. But
2106 : * that is okay, folio_free_swap() only removes stale folios.
2107 : */
2108 0 : folio_lock(folio);
2109 0 : folio_wait_writeback(folio);
2110 0 : folio_free_swap(folio);
2111 0 : folio_unlock(folio);
2112 : folio_put(folio);
2113 : }
2114 :
2115 : /*
2116 : * Lets check again to see if there are still swap entries in the map.
2117 : * If yes, we would need to do retry the unuse logic again.
2118 : * Under global memory pressure, swap entries can be reinserted back
2119 : * into process space after the mmlist loop above passes over them.
2120 : *
2121 : * Limit the number of retries? No: when mmget_not_zero()
2122 : * above fails, that mm is likely to be freeing swap from
2123 : * exit_mmap(), which proceeds at its own independent pace;
2124 : * and even shmem_writepage() could have been preempted after
2125 : * folio_alloc_swap(), temporarily hiding that swap. It's easy
2126 : * and robust (though cpu-intensive) just to keep retrying.
2127 : */
2128 0 : if (READ_ONCE(si->inuse_pages)) {
2129 0 : if (!signal_pending(current))
2130 : goto retry;
2131 : return -EINTR;
2132 : }
2133 :
2134 : return 0;
2135 : }
2136 :
2137 : /*
2138 : * After a successful try_to_unuse, if no swap is now in use, we know
2139 : * we can empty the mmlist. swap_lock must be held on entry and exit.
2140 : * Note that mmlist_lock nests inside swap_lock, and an mm must be
2141 : * added to the mmlist just after page_duplicate - before would be racy.
2142 : */
2143 0 : static void drain_mmlist(void)
2144 : {
2145 : struct list_head *p, *next;
2146 : unsigned int type;
2147 :
2148 0 : for (type = 0; type < nr_swapfiles; type++)
2149 0 : if (swap_info[type]->inuse_pages)
2150 : return;
2151 0 : spin_lock(&mmlist_lock);
2152 0 : list_for_each_safe(p, next, &init_mm.mmlist)
2153 0 : list_del_init(p);
2154 : spin_unlock(&mmlist_lock);
2155 : }
2156 :
2157 : /*
2158 : * Free all of a swapdev's extent information
2159 : */
2160 0 : static void destroy_swap_extents(struct swap_info_struct *sis)
2161 : {
2162 0 : while (!RB_EMPTY_ROOT(&sis->swap_extent_root)) {
2163 0 : struct rb_node *rb = sis->swap_extent_root.rb_node;
2164 0 : struct swap_extent *se = rb_entry(rb, struct swap_extent, rb_node);
2165 :
2166 0 : rb_erase(rb, &sis->swap_extent_root);
2167 0 : kfree(se);
2168 : }
2169 :
2170 0 : if (sis->flags & SWP_ACTIVATED) {
2171 0 : struct file *swap_file = sis->swap_file;
2172 0 : struct address_space *mapping = swap_file->f_mapping;
2173 :
2174 0 : sis->flags &= ~SWP_ACTIVATED;
2175 0 : if (mapping->a_ops->swap_deactivate)
2176 0 : mapping->a_ops->swap_deactivate(swap_file);
2177 : }
2178 0 : }
2179 :
2180 : /*
2181 : * Add a block range (and the corresponding page range) into this swapdev's
2182 : * extent tree.
2183 : *
2184 : * This function rather assumes that it is called in ascending page order.
2185 : */
2186 : int
2187 0 : add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
2188 : unsigned long nr_pages, sector_t start_block)
2189 : {
2190 0 : struct rb_node **link = &sis->swap_extent_root.rb_node, *parent = NULL;
2191 : struct swap_extent *se;
2192 : struct swap_extent *new_se;
2193 :
2194 : /*
2195 : * place the new node at the right most since the
2196 : * function is called in ascending page order.
2197 : */
2198 0 : while (*link) {
2199 0 : parent = *link;
2200 0 : link = &parent->rb_right;
2201 : }
2202 :
2203 0 : if (parent) {
2204 0 : se = rb_entry(parent, struct swap_extent, rb_node);
2205 0 : BUG_ON(se->start_page + se->nr_pages != start_page);
2206 0 : if (se->start_block + se->nr_pages == start_block) {
2207 : /* Merge it */
2208 0 : se->nr_pages += nr_pages;
2209 0 : return 0;
2210 : }
2211 : }
2212 :
2213 : /* No merge, insert a new extent. */
2214 0 : new_se = kmalloc(sizeof(*se), GFP_KERNEL);
2215 0 : if (new_se == NULL)
2216 : return -ENOMEM;
2217 0 : new_se->start_page = start_page;
2218 0 : new_se->nr_pages = nr_pages;
2219 0 : new_se->start_block = start_block;
2220 :
2221 0 : rb_link_node(&new_se->rb_node, parent, link);
2222 0 : rb_insert_color(&new_se->rb_node, &sis->swap_extent_root);
2223 0 : return 1;
2224 : }
2225 : EXPORT_SYMBOL_GPL(add_swap_extent);
2226 :
2227 : /*
2228 : * A `swap extent' is a simple thing which maps a contiguous range of pages
2229 : * onto a contiguous range of disk blocks. A rbtree of swap extents is
2230 : * built at swapon time and is then used at swap_writepage/swap_readpage
2231 : * time for locating where on disk a page belongs.
2232 : *
2233 : * If the swapfile is an S_ISBLK block device, a single extent is installed.
2234 : * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2235 : * swap files identically.
2236 : *
2237 : * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2238 : * extent rbtree operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
2239 : * swapfiles are handled *identically* after swapon time.
2240 : *
2241 : * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2242 : * and will parse them into a rbtree, in PAGE_SIZE chunks. If some stray
2243 : * blocks are found which do not fall within the PAGE_SIZE alignment
2244 : * requirements, they are simply tossed out - we will never use those blocks
2245 : * for swapping.
2246 : *
2247 : * For all swap devices we set S_SWAPFILE across the life of the swapon. This
2248 : * prevents users from writing to the swap device, which will corrupt memory.
2249 : *
2250 : * The amount of disk space which a single swap extent represents varies.
2251 : * Typically it is in the 1-4 megabyte range. So we can have hundreds of
2252 : * extents in the rbtree. - akpm.
2253 : */
2254 0 : static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
2255 : {
2256 0 : struct file *swap_file = sis->swap_file;
2257 0 : struct address_space *mapping = swap_file->f_mapping;
2258 0 : struct inode *inode = mapping->host;
2259 : int ret;
2260 :
2261 0 : if (S_ISBLK(inode->i_mode)) {
2262 0 : ret = add_swap_extent(sis, 0, sis->max, 0);
2263 0 : *span = sis->pages;
2264 0 : return ret;
2265 : }
2266 :
2267 0 : if (mapping->a_ops->swap_activate) {
2268 0 : ret = mapping->a_ops->swap_activate(sis, swap_file, span);
2269 0 : if (ret < 0)
2270 : return ret;
2271 0 : sis->flags |= SWP_ACTIVATED;
2272 0 : if ((sis->flags & SWP_FS_OPS) &&
2273 0 : sio_pool_init() != 0) {
2274 0 : destroy_swap_extents(sis);
2275 0 : return -ENOMEM;
2276 : }
2277 : return ret;
2278 : }
2279 :
2280 0 : return generic_swapfile_activate(sis, swap_file, span);
2281 : }
2282 :
2283 : static int swap_node(struct swap_info_struct *p)
2284 : {
2285 : struct block_device *bdev;
2286 :
2287 0 : if (p->bdev)
2288 : bdev = p->bdev;
2289 : else
2290 0 : bdev = p->swap_file->f_inode->i_sb->s_bdev;
2291 :
2292 0 : return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE;
2293 : }
2294 :
2295 0 : static void setup_swap_info(struct swap_info_struct *p, int prio,
2296 : unsigned char *swap_map,
2297 : struct swap_cluster_info *cluster_info)
2298 : {
2299 : int i;
2300 :
2301 0 : if (prio >= 0)
2302 0 : p->prio = prio;
2303 : else
2304 0 : p->prio = --least_priority;
2305 : /*
2306 : * the plist prio is negated because plist ordering is
2307 : * low-to-high, while swap ordering is high-to-low
2308 : */
2309 0 : p->list.prio = -p->prio;
2310 0 : for_each_node(i) {
2311 0 : if (p->prio >= 0)
2312 0 : p->avail_lists[i].prio = -p->prio;
2313 : else {
2314 0 : if (swap_node(p) == i)
2315 0 : p->avail_lists[i].prio = 1;
2316 : else
2317 0 : p->avail_lists[i].prio = -p->prio;
2318 : }
2319 : }
2320 0 : p->swap_map = swap_map;
2321 0 : p->cluster_info = cluster_info;
2322 0 : }
2323 :
2324 0 : static void _enable_swap_info(struct swap_info_struct *p)
2325 : {
2326 0 : p->flags |= SWP_WRITEOK;
2327 0 : atomic_long_add(p->pages, &nr_swap_pages);
2328 0 : total_swap_pages += p->pages;
2329 :
2330 : assert_spin_locked(&swap_lock);
2331 : /*
2332 : * both lists are plists, and thus priority ordered.
2333 : * swap_active_head needs to be priority ordered for swapoff(),
2334 : * which on removal of any swap_info_struct with an auto-assigned
2335 : * (i.e. negative) priority increments the auto-assigned priority
2336 : * of any lower-priority swap_info_structs.
2337 : * swap_avail_head needs to be priority ordered for folio_alloc_swap(),
2338 : * which allocates swap pages from the highest available priority
2339 : * swap_info_struct.
2340 : */
2341 0 : plist_add(&p->list, &swap_active_head);
2342 0 : add_to_avail_list(p);
2343 0 : }
2344 :
2345 0 : static void enable_swap_info(struct swap_info_struct *p, int prio,
2346 : unsigned char *swap_map,
2347 : struct swap_cluster_info *cluster_info,
2348 : unsigned long *frontswap_map)
2349 : {
2350 : if (IS_ENABLED(CONFIG_FRONTSWAP))
2351 : frontswap_init(p->type, frontswap_map);
2352 0 : spin_lock(&swap_lock);
2353 0 : spin_lock(&p->lock);
2354 0 : setup_swap_info(p, prio, swap_map, cluster_info);
2355 0 : spin_unlock(&p->lock);
2356 0 : spin_unlock(&swap_lock);
2357 : /*
2358 : * Finished initializing swap device, now it's safe to reference it.
2359 : */
2360 0 : percpu_ref_resurrect(&p->users);
2361 0 : spin_lock(&swap_lock);
2362 0 : spin_lock(&p->lock);
2363 0 : _enable_swap_info(p);
2364 0 : spin_unlock(&p->lock);
2365 0 : spin_unlock(&swap_lock);
2366 0 : }
2367 :
2368 0 : static void reinsert_swap_info(struct swap_info_struct *p)
2369 : {
2370 0 : spin_lock(&swap_lock);
2371 0 : spin_lock(&p->lock);
2372 0 : setup_swap_info(p, p->prio, p->swap_map, p->cluster_info);
2373 0 : _enable_swap_info(p);
2374 0 : spin_unlock(&p->lock);
2375 0 : spin_unlock(&swap_lock);
2376 0 : }
2377 :
2378 0 : bool has_usable_swap(void)
2379 : {
2380 0 : bool ret = true;
2381 :
2382 0 : spin_lock(&swap_lock);
2383 0 : if (plist_head_empty(&swap_active_head))
2384 0 : ret = false;
2385 0 : spin_unlock(&swap_lock);
2386 0 : return ret;
2387 : }
2388 :
2389 0 : SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
2390 : {
2391 0 : struct swap_info_struct *p = NULL;
2392 : unsigned char *swap_map;
2393 : struct swap_cluster_info *cluster_info;
2394 : unsigned long *frontswap_map;
2395 : struct file *swap_file, *victim;
2396 : struct address_space *mapping;
2397 : struct inode *inode;
2398 : struct filename *pathname;
2399 0 : int err, found = 0;
2400 : unsigned int old_block_size;
2401 :
2402 0 : if (!capable(CAP_SYS_ADMIN))
2403 : return -EPERM;
2404 :
2405 0 : BUG_ON(!current->mm);
2406 :
2407 0 : pathname = getname(specialfile);
2408 0 : if (IS_ERR(pathname))
2409 0 : return PTR_ERR(pathname);
2410 :
2411 0 : victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
2412 0 : err = PTR_ERR(victim);
2413 0 : if (IS_ERR(victim))
2414 : goto out;
2415 :
2416 0 : mapping = victim->f_mapping;
2417 0 : spin_lock(&swap_lock);
2418 0 : plist_for_each_entry(p, &swap_active_head, list) {
2419 0 : if (p->flags & SWP_WRITEOK) {
2420 0 : if (p->swap_file->f_mapping == mapping) {
2421 : found = 1;
2422 : break;
2423 : }
2424 : }
2425 : }
2426 0 : if (!found) {
2427 0 : err = -EINVAL;
2428 : spin_unlock(&swap_lock);
2429 : goto out_dput;
2430 : }
2431 0 : if (!security_vm_enough_memory_mm(current->mm, p->pages))
2432 0 : vm_unacct_memory(p->pages);
2433 : else {
2434 0 : err = -ENOMEM;
2435 : spin_unlock(&swap_lock);
2436 : goto out_dput;
2437 : }
2438 0 : spin_lock(&p->lock);
2439 0 : del_from_avail_list(p);
2440 0 : if (p->prio < 0) {
2441 0 : struct swap_info_struct *si = p;
2442 : int nid;
2443 :
2444 0 : plist_for_each_entry_continue(si, &swap_active_head, list) {
2445 0 : si->prio++;
2446 0 : si->list.prio--;
2447 0 : for_each_node(nid) {
2448 0 : if (si->avail_lists[nid].prio != 1)
2449 0 : si->avail_lists[nid].prio--;
2450 : }
2451 : }
2452 0 : least_priority++;
2453 : }
2454 0 : plist_del(&p->list, &swap_active_head);
2455 0 : atomic_long_sub(p->pages, &nr_swap_pages);
2456 0 : total_swap_pages -= p->pages;
2457 0 : p->flags &= ~SWP_WRITEOK;
2458 0 : spin_unlock(&p->lock);
2459 0 : spin_unlock(&swap_lock);
2460 :
2461 0 : disable_swap_slots_cache_lock();
2462 :
2463 0 : set_current_oom_origin();
2464 0 : err = try_to_unuse(p->type);
2465 0 : clear_current_oom_origin();
2466 :
2467 0 : if (err) {
2468 : /* re-insert swap space back into swap_list */
2469 0 : reinsert_swap_info(p);
2470 0 : reenable_swap_slots_cache_unlock();
2471 0 : goto out_dput;
2472 : }
2473 :
2474 0 : reenable_swap_slots_cache_unlock();
2475 :
2476 : /*
2477 : * Wait for swap operations protected by get/put_swap_device()
2478 : * to complete.
2479 : *
2480 : * We need synchronize_rcu() here to protect the accessing to
2481 : * the swap cache data structure.
2482 : */
2483 0 : percpu_ref_kill(&p->users);
2484 0 : synchronize_rcu();
2485 0 : wait_for_completion(&p->comp);
2486 :
2487 0 : flush_work(&p->discard_work);
2488 :
2489 0 : destroy_swap_extents(p);
2490 0 : if (p->flags & SWP_CONTINUED)
2491 0 : free_swap_count_continuations(p);
2492 :
2493 0 : if (!p->bdev || !bdev_nonrot(p->bdev))
2494 : atomic_dec(&nr_rotate_swap);
2495 :
2496 0 : mutex_lock(&swapon_mutex);
2497 0 : spin_lock(&swap_lock);
2498 0 : spin_lock(&p->lock);
2499 0 : drain_mmlist();
2500 :
2501 : /* wait for anyone still in scan_swap_map_slots */
2502 0 : p->highest_bit = 0; /* cuts scans short */
2503 0 : while (p->flags >= SWP_SCANNING) {
2504 0 : spin_unlock(&p->lock);
2505 0 : spin_unlock(&swap_lock);
2506 0 : schedule_timeout_uninterruptible(1);
2507 0 : spin_lock(&swap_lock);
2508 0 : spin_lock(&p->lock);
2509 : }
2510 :
2511 0 : swap_file = p->swap_file;
2512 0 : old_block_size = p->old_block_size;
2513 0 : p->swap_file = NULL;
2514 0 : p->max = 0;
2515 0 : swap_map = p->swap_map;
2516 0 : p->swap_map = NULL;
2517 0 : cluster_info = p->cluster_info;
2518 0 : p->cluster_info = NULL;
2519 0 : frontswap_map = frontswap_map_get(p);
2520 0 : spin_unlock(&p->lock);
2521 0 : spin_unlock(&swap_lock);
2522 0 : arch_swap_invalidate_area(p->type);
2523 0 : frontswap_invalidate_area(p->type);
2524 0 : frontswap_map_set(p, NULL);
2525 0 : mutex_unlock(&swapon_mutex);
2526 0 : free_percpu(p->percpu_cluster);
2527 0 : p->percpu_cluster = NULL;
2528 0 : free_percpu(p->cluster_next_cpu);
2529 0 : p->cluster_next_cpu = NULL;
2530 0 : vfree(swap_map);
2531 0 : kvfree(cluster_info);
2532 0 : kvfree(frontswap_map);
2533 : /* Destroy swap account information */
2534 0 : swap_cgroup_swapoff(p->type);
2535 0 : exit_swap_address_space(p->type);
2536 :
2537 0 : inode = mapping->host;
2538 0 : if (S_ISBLK(inode->i_mode)) {
2539 0 : struct block_device *bdev = I_BDEV(inode);
2540 :
2541 0 : set_blocksize(bdev, old_block_size);
2542 0 : blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2543 : }
2544 :
2545 0 : inode_lock(inode);
2546 0 : inode->i_flags &= ~S_SWAPFILE;
2547 0 : inode_unlock(inode);
2548 0 : filp_close(swap_file, NULL);
2549 :
2550 : /*
2551 : * Clear the SWP_USED flag after all resources are freed so that swapon
2552 : * can reuse this swap_info in alloc_swap_info() safely. It is ok to
2553 : * not hold p->lock after we cleared its SWP_WRITEOK.
2554 : */
2555 0 : spin_lock(&swap_lock);
2556 0 : p->flags = 0;
2557 0 : spin_unlock(&swap_lock);
2558 :
2559 0 : err = 0;
2560 0 : atomic_inc(&proc_poll_event);
2561 0 : wake_up_interruptible(&proc_poll_wait);
2562 :
2563 : out_dput:
2564 0 : filp_close(victim, NULL);
2565 : out:
2566 0 : putname(pathname);
2567 0 : return err;
2568 : }
2569 :
2570 : #ifdef CONFIG_PROC_FS
2571 0 : static __poll_t swaps_poll(struct file *file, poll_table *wait)
2572 : {
2573 0 : struct seq_file *seq = file->private_data;
2574 :
2575 0 : poll_wait(file, &proc_poll_wait, wait);
2576 :
2577 0 : if (seq->poll_event != atomic_read(&proc_poll_event)) {
2578 0 : seq->poll_event = atomic_read(&proc_poll_event);
2579 0 : return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
2580 : }
2581 :
2582 : return EPOLLIN | EPOLLRDNORM;
2583 : }
2584 :
2585 : /* iterator */
2586 0 : static void *swap_start(struct seq_file *swap, loff_t *pos)
2587 : {
2588 : struct swap_info_struct *si;
2589 : int type;
2590 0 : loff_t l = *pos;
2591 :
2592 0 : mutex_lock(&swapon_mutex);
2593 :
2594 0 : if (!l)
2595 : return SEQ_START_TOKEN;
2596 :
2597 0 : for (type = 0; (si = swap_type_to_swap_info(type)); type++) {
2598 0 : if (!(si->flags & SWP_USED) || !si->swap_map)
2599 0 : continue;
2600 0 : if (!--l)
2601 : return si;
2602 : }
2603 :
2604 : return NULL;
2605 : }
2606 :
2607 0 : static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2608 : {
2609 0 : struct swap_info_struct *si = v;
2610 : int type;
2611 :
2612 0 : if (v == SEQ_START_TOKEN)
2613 : type = 0;
2614 : else
2615 0 : type = si->type + 1;
2616 :
2617 0 : ++(*pos);
2618 0 : for (; (si = swap_type_to_swap_info(type)); type++) {
2619 0 : if (!(si->flags & SWP_USED) || !si->swap_map)
2620 0 : continue;
2621 : return si;
2622 : }
2623 :
2624 : return NULL;
2625 : }
2626 :
2627 0 : static void swap_stop(struct seq_file *swap, void *v)
2628 : {
2629 0 : mutex_unlock(&swapon_mutex);
2630 0 : }
2631 :
2632 0 : static int swap_show(struct seq_file *swap, void *v)
2633 : {
2634 0 : struct swap_info_struct *si = v;
2635 : struct file *file;
2636 : int len;
2637 : unsigned long bytes, inuse;
2638 :
2639 0 : if (si == SEQ_START_TOKEN) {
2640 0 : seq_puts(swap, "Filename\t\t\t\tType\t\tSize\t\tUsed\t\tPriority\n");
2641 0 : return 0;
2642 : }
2643 :
2644 0 : bytes = si->pages << (PAGE_SHIFT - 10);
2645 0 : inuse = READ_ONCE(si->inuse_pages) << (PAGE_SHIFT - 10);
2646 :
2647 0 : file = si->swap_file;
2648 0 : len = seq_file_path(swap, file, " \t\n\\");
2649 0 : seq_printf(swap, "%*s%s\t%lu\t%s%lu\t%s%d\n",
2650 : len < 40 ? 40 - len : 1, " ",
2651 0 : S_ISBLK(file_inode(file)->i_mode) ?
2652 : "partition" : "file\t",
2653 : bytes, bytes < 10000000 ? "\t" : "",
2654 : inuse, inuse < 10000000 ? "\t" : "",
2655 0 : si->prio);
2656 0 : return 0;
2657 : }
2658 :
2659 : static const struct seq_operations swaps_op = {
2660 : .start = swap_start,
2661 : .next = swap_next,
2662 : .stop = swap_stop,
2663 : .show = swap_show
2664 : };
2665 :
2666 0 : static int swaps_open(struct inode *inode, struct file *file)
2667 : {
2668 : struct seq_file *seq;
2669 : int ret;
2670 :
2671 0 : ret = seq_open(file, &swaps_op);
2672 0 : if (ret)
2673 : return ret;
2674 :
2675 0 : seq = file->private_data;
2676 0 : seq->poll_event = atomic_read(&proc_poll_event);
2677 0 : return 0;
2678 : }
2679 :
2680 : static const struct proc_ops swaps_proc_ops = {
2681 : .proc_flags = PROC_ENTRY_PERMANENT,
2682 : .proc_open = swaps_open,
2683 : .proc_read = seq_read,
2684 : .proc_lseek = seq_lseek,
2685 : .proc_release = seq_release,
2686 : .proc_poll = swaps_poll,
2687 : };
2688 :
2689 1 : static int __init procswaps_init(void)
2690 : {
2691 1 : proc_create("swaps", 0, NULL, &swaps_proc_ops);
2692 1 : return 0;
2693 : }
2694 : __initcall(procswaps_init);
2695 : #endif /* CONFIG_PROC_FS */
2696 :
2697 : #ifdef MAX_SWAPFILES_CHECK
2698 : static int __init max_swapfiles_check(void)
2699 : {
2700 : MAX_SWAPFILES_CHECK();
2701 : return 0;
2702 : }
2703 : late_initcall(max_swapfiles_check);
2704 : #endif
2705 :
2706 0 : static struct swap_info_struct *alloc_swap_info(void)
2707 : {
2708 : struct swap_info_struct *p;
2709 0 : struct swap_info_struct *defer = NULL;
2710 : unsigned int type;
2711 : int i;
2712 :
2713 0 : p = kvzalloc(struct_size(p, avail_lists, nr_node_ids), GFP_KERNEL);
2714 0 : if (!p)
2715 : return ERR_PTR(-ENOMEM);
2716 :
2717 0 : if (percpu_ref_init(&p->users, swap_users_ref_free,
2718 : PERCPU_REF_INIT_DEAD, GFP_KERNEL)) {
2719 0 : kvfree(p);
2720 0 : return ERR_PTR(-ENOMEM);
2721 : }
2722 :
2723 0 : spin_lock(&swap_lock);
2724 0 : for (type = 0; type < nr_swapfiles; type++) {
2725 0 : if (!(swap_info[type]->flags & SWP_USED))
2726 : break;
2727 : }
2728 0 : if (type >= MAX_SWAPFILES) {
2729 0 : spin_unlock(&swap_lock);
2730 0 : percpu_ref_exit(&p->users);
2731 0 : kvfree(p);
2732 0 : return ERR_PTR(-EPERM);
2733 : }
2734 0 : if (type >= nr_swapfiles) {
2735 0 : p->type = type;
2736 : /*
2737 : * Publish the swap_info_struct after initializing it.
2738 : * Note that kvzalloc() above zeroes all its fields.
2739 : */
2740 0 : smp_store_release(&swap_info[type], p); /* rcu_assign_pointer() */
2741 0 : nr_swapfiles++;
2742 : } else {
2743 0 : defer = p;
2744 0 : p = swap_info[type];
2745 : /*
2746 : * Do not memset this entry: a racing procfs swap_next()
2747 : * would be relying on p->type to remain valid.
2748 : */
2749 : }
2750 0 : p->swap_extent_root = RB_ROOT;
2751 0 : plist_node_init(&p->list, 0);
2752 0 : for_each_node(i)
2753 0 : plist_node_init(&p->avail_lists[i], 0);
2754 0 : p->flags = SWP_USED;
2755 0 : spin_unlock(&swap_lock);
2756 0 : if (defer) {
2757 0 : percpu_ref_exit(&defer->users);
2758 0 : kvfree(defer);
2759 : }
2760 0 : spin_lock_init(&p->lock);
2761 0 : spin_lock_init(&p->cont_lock);
2762 0 : init_completion(&p->comp);
2763 :
2764 0 : return p;
2765 : }
2766 :
2767 0 : static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2768 : {
2769 : int error;
2770 :
2771 0 : if (S_ISBLK(inode->i_mode)) {
2772 0 : p->bdev = blkdev_get_by_dev(inode->i_rdev,
2773 : FMODE_READ | FMODE_WRITE | FMODE_EXCL, p);
2774 0 : if (IS_ERR(p->bdev)) {
2775 0 : error = PTR_ERR(p->bdev);
2776 0 : p->bdev = NULL;
2777 0 : return error;
2778 : }
2779 0 : p->old_block_size = block_size(p->bdev);
2780 0 : error = set_blocksize(p->bdev, PAGE_SIZE);
2781 0 : if (error < 0)
2782 : return error;
2783 : /*
2784 : * Zoned block devices contain zones that have a sequential
2785 : * write only restriction. Hence zoned block devices are not
2786 : * suitable for swapping. Disallow them here.
2787 : */
2788 0 : if (bdev_is_zoned(p->bdev))
2789 : return -EINVAL;
2790 0 : p->flags |= SWP_BLKDEV;
2791 0 : } else if (S_ISREG(inode->i_mode)) {
2792 0 : p->bdev = inode->i_sb->s_bdev;
2793 : }
2794 :
2795 : return 0;
2796 : }
2797 :
2798 :
2799 : /*
2800 : * Find out how many pages are allowed for a single swap device. There
2801 : * are two limiting factors:
2802 : * 1) the number of bits for the swap offset in the swp_entry_t type, and
2803 : * 2) the number of bits in the swap pte, as defined by the different
2804 : * architectures.
2805 : *
2806 : * In order to find the largest possible bit mask, a swap entry with
2807 : * swap type 0 and swap offset ~0UL is created, encoded to a swap pte,
2808 : * decoded to a swp_entry_t again, and finally the swap offset is
2809 : * extracted.
2810 : *
2811 : * This will mask all the bits from the initial ~0UL mask that can't
2812 : * be encoded in either the swp_entry_t or the architecture definition
2813 : * of a swap pte.
2814 : */
2815 0 : unsigned long generic_max_swapfile_size(void)
2816 : {
2817 3 : return swp_offset(pte_to_swp_entry(
2818 0 : swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2819 : }
2820 :
2821 : /* Can be overridden by an architecture for additional checks. */
2822 1 : __weak unsigned long arch_max_swapfile_size(void)
2823 : {
2824 1 : return generic_max_swapfile_size();
2825 : }
2826 :
2827 0 : static unsigned long read_swap_header(struct swap_info_struct *p,
2828 : union swap_header *swap_header,
2829 : struct inode *inode)
2830 : {
2831 : int i;
2832 : unsigned long maxpages;
2833 : unsigned long swapfilepages;
2834 : unsigned long last_page;
2835 :
2836 0 : if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
2837 0 : pr_err("Unable to find swap-space signature\n");
2838 0 : return 0;
2839 : }
2840 :
2841 : /* swap partition endianness hack... */
2842 0 : if (swab32(swap_header->info.version) == 1) {
2843 0 : swab32s(&swap_header->info.version);
2844 0 : swab32s(&swap_header->info.last_page);
2845 0 : swab32s(&swap_header->info.nr_badpages);
2846 0 : if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2847 : return 0;
2848 0 : for (i = 0; i < swap_header->info.nr_badpages; i++)
2849 0 : swab32s(&swap_header->info.badpages[i]);
2850 : }
2851 : /* Check the swap header's sub-version */
2852 0 : if (swap_header->info.version != 1) {
2853 0 : pr_warn("Unable to handle swap header version %d\n",
2854 : swap_header->info.version);
2855 0 : return 0;
2856 : }
2857 :
2858 0 : p->lowest_bit = 1;
2859 0 : p->cluster_next = 1;
2860 0 : p->cluster_nr = 0;
2861 :
2862 0 : maxpages = swapfile_maximum_size;
2863 0 : last_page = swap_header->info.last_page;
2864 0 : if (!last_page) {
2865 0 : pr_warn("Empty swap-file\n");
2866 0 : return 0;
2867 : }
2868 0 : if (last_page > maxpages) {
2869 0 : pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
2870 : maxpages << (PAGE_SHIFT - 10),
2871 : last_page << (PAGE_SHIFT - 10));
2872 : }
2873 0 : if (maxpages > last_page) {
2874 0 : maxpages = last_page + 1;
2875 : /* p->max is an unsigned int: don't overflow it */
2876 0 : if ((unsigned int)maxpages == 0)
2877 0 : maxpages = UINT_MAX;
2878 : }
2879 0 : p->highest_bit = maxpages - 1;
2880 :
2881 0 : if (!maxpages)
2882 : return 0;
2883 0 : swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2884 0 : if (swapfilepages && maxpages > swapfilepages) {
2885 0 : pr_warn("Swap area shorter than signature indicates\n");
2886 0 : return 0;
2887 : }
2888 0 : if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
2889 : return 0;
2890 0 : if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2891 : return 0;
2892 :
2893 0 : return maxpages;
2894 : }
2895 :
2896 : #define SWAP_CLUSTER_INFO_COLS \
2897 : DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
2898 : #define SWAP_CLUSTER_SPACE_COLS \
2899 : DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
2900 : #define SWAP_CLUSTER_COLS \
2901 : max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
2902 :
2903 0 : static int setup_swap_map_and_extents(struct swap_info_struct *p,
2904 : union swap_header *swap_header,
2905 : unsigned char *swap_map,
2906 : struct swap_cluster_info *cluster_info,
2907 : unsigned long maxpages,
2908 : sector_t *span)
2909 : {
2910 : unsigned int j, k;
2911 : unsigned int nr_good_pages;
2912 : int nr_extents;
2913 0 : unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
2914 0 : unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS;
2915 : unsigned long i, idx;
2916 :
2917 0 : nr_good_pages = maxpages - 1; /* omit header page */
2918 :
2919 0 : cluster_list_init(&p->free_clusters);
2920 0 : cluster_list_init(&p->discard_clusters);
2921 :
2922 0 : for (i = 0; i < swap_header->info.nr_badpages; i++) {
2923 0 : unsigned int page_nr = swap_header->info.badpages[i];
2924 0 : if (page_nr == 0 || page_nr > swap_header->info.last_page)
2925 : return -EINVAL;
2926 0 : if (page_nr < maxpages) {
2927 0 : swap_map[page_nr] = SWAP_MAP_BAD;
2928 0 : nr_good_pages--;
2929 : /*
2930 : * Haven't marked the cluster free yet, no list
2931 : * operation involved
2932 : */
2933 0 : inc_cluster_info_page(p, cluster_info, page_nr);
2934 : }
2935 : }
2936 :
2937 : /* Haven't marked the cluster free yet, no list operation involved */
2938 0 : for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
2939 0 : inc_cluster_info_page(p, cluster_info, i);
2940 :
2941 0 : if (nr_good_pages) {
2942 0 : swap_map[0] = SWAP_MAP_BAD;
2943 : /*
2944 : * Not mark the cluster free yet, no list
2945 : * operation involved
2946 : */
2947 0 : inc_cluster_info_page(p, cluster_info, 0);
2948 0 : p->max = maxpages;
2949 0 : p->pages = nr_good_pages;
2950 0 : nr_extents = setup_swap_extents(p, span);
2951 0 : if (nr_extents < 0)
2952 : return nr_extents;
2953 0 : nr_good_pages = p->pages;
2954 : }
2955 0 : if (!nr_good_pages) {
2956 0 : pr_warn("Empty swap-file\n");
2957 0 : return -EINVAL;
2958 : }
2959 :
2960 0 : if (!cluster_info)
2961 : return nr_extents;
2962 :
2963 :
2964 : /*
2965 : * Reduce false cache line sharing between cluster_info and
2966 : * sharing same address space.
2967 : */
2968 0 : for (k = 0; k < SWAP_CLUSTER_COLS; k++) {
2969 0 : j = (k + col) % SWAP_CLUSTER_COLS;
2970 0 : for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) {
2971 0 : idx = i * SWAP_CLUSTER_COLS + j;
2972 0 : if (idx >= nr_clusters)
2973 0 : continue;
2974 0 : if (cluster_count(&cluster_info[idx]))
2975 0 : continue;
2976 0 : cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
2977 0 : cluster_list_add_tail(&p->free_clusters, cluster_info,
2978 : idx);
2979 : }
2980 : }
2981 : return nr_extents;
2982 : }
2983 :
2984 0 : SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2985 : {
2986 : struct swap_info_struct *p;
2987 : struct filename *name;
2988 0 : struct file *swap_file = NULL;
2989 : struct address_space *mapping;
2990 : struct dentry *dentry;
2991 : int prio;
2992 : int error;
2993 : union swap_header *swap_header;
2994 : int nr_extents;
2995 : sector_t span;
2996 : unsigned long maxpages;
2997 0 : unsigned char *swap_map = NULL;
2998 0 : struct swap_cluster_info *cluster_info = NULL;
2999 0 : unsigned long *frontswap_map = NULL;
3000 0 : struct page *page = NULL;
3001 0 : struct inode *inode = NULL;
3002 0 : bool inced_nr_rotate_swap = false;
3003 :
3004 0 : if (swap_flags & ~SWAP_FLAGS_VALID)
3005 : return -EINVAL;
3006 :
3007 0 : if (!capable(CAP_SYS_ADMIN))
3008 : return -EPERM;
3009 :
3010 0 : if (!swap_avail_heads)
3011 : return -ENOMEM;
3012 :
3013 0 : p = alloc_swap_info();
3014 0 : if (IS_ERR(p))
3015 0 : return PTR_ERR(p);
3016 :
3017 0 : INIT_WORK(&p->discard_work, swap_discard_work);
3018 :
3019 0 : name = getname(specialfile);
3020 0 : if (IS_ERR(name)) {
3021 0 : error = PTR_ERR(name);
3022 0 : name = NULL;
3023 0 : goto bad_swap;
3024 : }
3025 0 : swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
3026 0 : if (IS_ERR(swap_file)) {
3027 0 : error = PTR_ERR(swap_file);
3028 0 : swap_file = NULL;
3029 0 : goto bad_swap;
3030 : }
3031 :
3032 0 : p->swap_file = swap_file;
3033 0 : mapping = swap_file->f_mapping;
3034 0 : dentry = swap_file->f_path.dentry;
3035 0 : inode = mapping->host;
3036 :
3037 0 : error = claim_swapfile(p, inode);
3038 0 : if (unlikely(error))
3039 : goto bad_swap;
3040 :
3041 0 : inode_lock(inode);
3042 0 : if (d_unlinked(dentry) || cant_mount(dentry)) {
3043 : error = -ENOENT;
3044 : goto bad_swap_unlock_inode;
3045 : }
3046 0 : if (IS_SWAPFILE(inode)) {
3047 : error = -EBUSY;
3048 : goto bad_swap_unlock_inode;
3049 : }
3050 :
3051 : /*
3052 : * Read the swap header.
3053 : */
3054 0 : if (!mapping->a_ops->read_folio) {
3055 : error = -EINVAL;
3056 : goto bad_swap_unlock_inode;
3057 : }
3058 0 : page = read_mapping_page(mapping, 0, swap_file);
3059 0 : if (IS_ERR(page)) {
3060 0 : error = PTR_ERR(page);
3061 0 : goto bad_swap_unlock_inode;
3062 : }
3063 0 : swap_header = kmap(page);
3064 :
3065 0 : maxpages = read_swap_header(p, swap_header, inode);
3066 0 : if (unlikely(!maxpages)) {
3067 : error = -EINVAL;
3068 : goto bad_swap_unlock_inode;
3069 : }
3070 :
3071 : /* OK, set up the swap map and apply the bad block list */
3072 0 : swap_map = vzalloc(maxpages);
3073 0 : if (!swap_map) {
3074 : error = -ENOMEM;
3075 : goto bad_swap_unlock_inode;
3076 : }
3077 :
3078 0 : if (p->bdev && bdev_stable_writes(p->bdev))
3079 0 : p->flags |= SWP_STABLE_WRITES;
3080 :
3081 0 : if (p->bdev && bdev_synchronous(p->bdev))
3082 0 : p->flags |= SWP_SYNCHRONOUS_IO;
3083 :
3084 0 : if (p->bdev && bdev_nonrot(p->bdev)) {
3085 : int cpu;
3086 : unsigned long ci, nr_cluster;
3087 :
3088 0 : p->flags |= SWP_SOLIDSTATE;
3089 0 : p->cluster_next_cpu = alloc_percpu(unsigned int);
3090 0 : if (!p->cluster_next_cpu) {
3091 : error = -ENOMEM;
3092 : goto bad_swap_unlock_inode;
3093 : }
3094 : /*
3095 : * select a random position to start with to help wear leveling
3096 : * SSD
3097 : */
3098 0 : for_each_possible_cpu(cpu) {
3099 0 : per_cpu(*p->cluster_next_cpu, cpu) =
3100 0 : get_random_u32_inclusive(1, p->highest_bit);
3101 : }
3102 0 : nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
3103 :
3104 0 : cluster_info = kvcalloc(nr_cluster, sizeof(*cluster_info),
3105 : GFP_KERNEL);
3106 0 : if (!cluster_info) {
3107 : error = -ENOMEM;
3108 : goto bad_swap_unlock_inode;
3109 : }
3110 :
3111 : for (ci = 0; ci < nr_cluster; ci++)
3112 : spin_lock_init(&((cluster_info + ci)->lock));
3113 :
3114 0 : p->percpu_cluster = alloc_percpu(struct percpu_cluster);
3115 0 : if (!p->percpu_cluster) {
3116 : error = -ENOMEM;
3117 : goto bad_swap_unlock_inode;
3118 : }
3119 0 : for_each_possible_cpu(cpu) {
3120 : struct percpu_cluster *cluster;
3121 0 : cluster = per_cpu_ptr(p->percpu_cluster, cpu);
3122 0 : cluster_set_null(&cluster->index);
3123 : }
3124 : } else {
3125 0 : atomic_inc(&nr_rotate_swap);
3126 0 : inced_nr_rotate_swap = true;
3127 : }
3128 :
3129 0 : error = swap_cgroup_swapon(p->type, maxpages);
3130 : if (error)
3131 : goto bad_swap_unlock_inode;
3132 :
3133 0 : nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
3134 : cluster_info, maxpages, &span);
3135 0 : if (unlikely(nr_extents < 0)) {
3136 : error = nr_extents;
3137 : goto bad_swap_unlock_inode;
3138 : }
3139 : /* frontswap enabled? set up bit-per-page map for frontswap */
3140 : if (IS_ENABLED(CONFIG_FRONTSWAP))
3141 : frontswap_map = kvcalloc(BITS_TO_LONGS(maxpages),
3142 : sizeof(long),
3143 : GFP_KERNEL);
3144 :
3145 0 : if ((swap_flags & SWAP_FLAG_DISCARD) &&
3146 0 : p->bdev && bdev_max_discard_sectors(p->bdev)) {
3147 : /*
3148 : * When discard is enabled for swap with no particular
3149 : * policy flagged, we set all swap discard flags here in
3150 : * order to sustain backward compatibility with older
3151 : * swapon(8) releases.
3152 : */
3153 0 : p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
3154 : SWP_PAGE_DISCARD);
3155 :
3156 : /*
3157 : * By flagging sys_swapon, a sysadmin can tell us to
3158 : * either do single-time area discards only, or to just
3159 : * perform discards for released swap page-clusters.
3160 : * Now it's time to adjust the p->flags accordingly.
3161 : */
3162 0 : if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
3163 0 : p->flags &= ~SWP_PAGE_DISCARD;
3164 0 : else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
3165 0 : p->flags &= ~SWP_AREA_DISCARD;
3166 :
3167 : /* issue a swapon-time discard if it's still required */
3168 0 : if (p->flags & SWP_AREA_DISCARD) {
3169 0 : int err = discard_swap(p);
3170 0 : if (unlikely(err))
3171 0 : pr_err("swapon: discard_swap(%p): %d\n",
3172 : p, err);
3173 : }
3174 : }
3175 :
3176 0 : error = init_swap_address_space(p->type, maxpages);
3177 0 : if (error)
3178 : goto bad_swap_unlock_inode;
3179 :
3180 : /*
3181 : * Flush any pending IO and dirty mappings before we start using this
3182 : * swap device.
3183 : */
3184 0 : inode->i_flags |= S_SWAPFILE;
3185 0 : error = inode_drain_writes(inode);
3186 0 : if (error) {
3187 0 : inode->i_flags &= ~S_SWAPFILE;
3188 : goto free_swap_address_space;
3189 : }
3190 :
3191 0 : mutex_lock(&swapon_mutex);
3192 0 : prio = -1;
3193 0 : if (swap_flags & SWAP_FLAG_PREFER)
3194 0 : prio =
3195 : (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
3196 0 : enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
3197 :
3198 0 : pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
3199 : p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
3200 : nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
3201 : (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
3202 : (p->flags & SWP_DISCARDABLE) ? "D" : "",
3203 : (p->flags & SWP_AREA_DISCARD) ? "s" : "",
3204 : (p->flags & SWP_PAGE_DISCARD) ? "c" : "",
3205 : (frontswap_map) ? "FS" : "");
3206 :
3207 0 : mutex_unlock(&swapon_mutex);
3208 0 : atomic_inc(&proc_poll_event);
3209 0 : wake_up_interruptible(&proc_poll_wait);
3210 :
3211 0 : error = 0;
3212 0 : goto out;
3213 : free_swap_address_space:
3214 0 : exit_swap_address_space(p->type);
3215 : bad_swap_unlock_inode:
3216 : inode_unlock(inode);
3217 : bad_swap:
3218 0 : free_percpu(p->percpu_cluster);
3219 0 : p->percpu_cluster = NULL;
3220 0 : free_percpu(p->cluster_next_cpu);
3221 0 : p->cluster_next_cpu = NULL;
3222 0 : if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
3223 0 : set_blocksize(p->bdev, p->old_block_size);
3224 0 : blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
3225 : }
3226 0 : inode = NULL;
3227 0 : destroy_swap_extents(p);
3228 0 : swap_cgroup_swapoff(p->type);
3229 0 : spin_lock(&swap_lock);
3230 0 : p->swap_file = NULL;
3231 0 : p->flags = 0;
3232 0 : spin_unlock(&swap_lock);
3233 0 : vfree(swap_map);
3234 0 : kvfree(cluster_info);
3235 0 : kvfree(frontswap_map);
3236 0 : if (inced_nr_rotate_swap)
3237 : atomic_dec(&nr_rotate_swap);
3238 0 : if (swap_file)
3239 0 : filp_close(swap_file, NULL);
3240 : out:
3241 0 : if (page && !IS_ERR(page)) {
3242 0 : kunmap(page);
3243 0 : put_page(page);
3244 : }
3245 0 : if (name)
3246 0 : putname(name);
3247 0 : if (inode)
3248 : inode_unlock(inode);
3249 0 : if (!error)
3250 0 : enable_swap_slots_cache();
3251 0 : return error;
3252 : }
3253 :
3254 0 : void si_swapinfo(struct sysinfo *val)
3255 : {
3256 : unsigned int type;
3257 0 : unsigned long nr_to_be_unused = 0;
3258 :
3259 0 : spin_lock(&swap_lock);
3260 0 : for (type = 0; type < nr_swapfiles; type++) {
3261 0 : struct swap_info_struct *si = swap_info[type];
3262 :
3263 0 : if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
3264 0 : nr_to_be_unused += READ_ONCE(si->inuse_pages);
3265 : }
3266 0 : val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
3267 0 : val->totalswap = total_swap_pages + nr_to_be_unused;
3268 0 : spin_unlock(&swap_lock);
3269 0 : }
3270 :
3271 : /*
3272 : * Verify that a swap entry is valid and increment its swap map count.
3273 : *
3274 : * Returns error code in following case.
3275 : * - success -> 0
3276 : * - swp_entry is invalid -> EINVAL
3277 : * - swp_entry is migration entry -> EINVAL
3278 : * - swap-cache reference is requested but there is already one. -> EEXIST
3279 : * - swap-cache reference is requested but the entry is not used. -> ENOENT
3280 : * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
3281 : */
3282 0 : static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
3283 : {
3284 : struct swap_info_struct *p;
3285 : struct swap_cluster_info *ci;
3286 : unsigned long offset;
3287 : unsigned char count;
3288 : unsigned char has_cache;
3289 : int err;
3290 :
3291 0 : p = get_swap_device(entry);
3292 0 : if (!p)
3293 : return -EINVAL;
3294 :
3295 0 : offset = swp_offset(entry);
3296 0 : ci = lock_cluster_or_swap_info(p, offset);
3297 :
3298 0 : count = p->swap_map[offset];
3299 :
3300 : /*
3301 : * swapin_readahead() doesn't check if a swap entry is valid, so the
3302 : * swap entry could be SWAP_MAP_BAD. Check here with lock held.
3303 : */
3304 0 : if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
3305 : err = -ENOENT;
3306 : goto unlock_out;
3307 : }
3308 :
3309 0 : has_cache = count & SWAP_HAS_CACHE;
3310 0 : count &= ~SWAP_HAS_CACHE;
3311 0 : err = 0;
3312 :
3313 0 : if (usage == SWAP_HAS_CACHE) {
3314 :
3315 : /* set SWAP_HAS_CACHE if there is no cache and entry is used */
3316 0 : if (!has_cache && count)
3317 : has_cache = SWAP_HAS_CACHE;
3318 0 : else if (has_cache) /* someone else added cache */
3319 : err = -EEXIST;
3320 : else /* no users remaining */
3321 0 : err = -ENOENT;
3322 :
3323 0 : } else if (count || has_cache) {
3324 :
3325 0 : if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
3326 0 : count += usage;
3327 0 : else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
3328 : err = -EINVAL;
3329 0 : else if (swap_count_continued(p, offset, count))
3330 : count = COUNT_CONTINUED;
3331 : else
3332 0 : err = -ENOMEM;
3333 : } else
3334 : err = -ENOENT; /* unused swap entry */
3335 :
3336 0 : WRITE_ONCE(p->swap_map[offset], count | has_cache);
3337 :
3338 : unlock_out:
3339 0 : unlock_cluster_or_swap_info(p, ci);
3340 0 : put_swap_device(p);
3341 0 : return err;
3342 : }
3343 :
3344 : /*
3345 : * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3346 : * (in which case its reference count is never incremented).
3347 : */
3348 0 : void swap_shmem_alloc(swp_entry_t entry)
3349 : {
3350 0 : __swap_duplicate(entry, SWAP_MAP_SHMEM);
3351 0 : }
3352 :
3353 : /*
3354 : * Increase reference count of swap entry by 1.
3355 : * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3356 : * but could not be atomically allocated. Returns 0, just as if it succeeded,
3357 : * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3358 : * might occur if a page table entry has got corrupted.
3359 : */
3360 0 : int swap_duplicate(swp_entry_t entry)
3361 : {
3362 0 : int err = 0;
3363 :
3364 0 : while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
3365 0 : err = add_swap_count_continuation(entry, GFP_ATOMIC);
3366 0 : return err;
3367 : }
3368 :
3369 : /*
3370 : * @entry: swap entry for which we allocate swap cache.
3371 : *
3372 : * Called when allocating swap cache for existing swap entry,
3373 : * This can return error codes. Returns 0 at success.
3374 : * -EEXIST means there is a swap cache.
3375 : * Note: return code is different from swap_duplicate().
3376 : */
3377 0 : int swapcache_prepare(swp_entry_t entry)
3378 : {
3379 0 : return __swap_duplicate(entry, SWAP_HAS_CACHE);
3380 : }
3381 :
3382 0 : struct swap_info_struct *swp_swap_info(swp_entry_t entry)
3383 : {
3384 0 : return swap_type_to_swap_info(swp_type(entry));
3385 : }
3386 :
3387 0 : struct swap_info_struct *page_swap_info(struct page *page)
3388 : {
3389 0 : swp_entry_t entry = { .val = page_private(page) };
3390 0 : return swp_swap_info(entry);
3391 : }
3392 :
3393 : /*
3394 : * out-of-line methods to avoid include hell.
3395 : */
3396 0 : struct address_space *swapcache_mapping(struct folio *folio)
3397 : {
3398 0 : return page_swap_info(&folio->page)->swap_file->f_mapping;
3399 : }
3400 : EXPORT_SYMBOL_GPL(swapcache_mapping);
3401 :
3402 0 : pgoff_t __page_file_index(struct page *page)
3403 : {
3404 0 : swp_entry_t swap = { .val = page_private(page) };
3405 0 : return swp_offset(swap);
3406 : }
3407 : EXPORT_SYMBOL_GPL(__page_file_index);
3408 :
3409 : /*
3410 : * add_swap_count_continuation - called when a swap count is duplicated
3411 : * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3412 : * page of the original vmalloc'ed swap_map, to hold the continuation count
3413 : * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
3414 : * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3415 : *
3416 : * These continuation pages are seldom referenced: the common paths all work
3417 : * on the original swap_map, only referring to a continuation page when the
3418 : * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3419 : *
3420 : * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3421 : * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3422 : * can be called after dropping locks.
3423 : */
3424 0 : int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
3425 : {
3426 : struct swap_info_struct *si;
3427 : struct swap_cluster_info *ci;
3428 : struct page *head;
3429 : struct page *page;
3430 : struct page *list_page;
3431 : pgoff_t offset;
3432 : unsigned char count;
3433 0 : int ret = 0;
3434 :
3435 : /*
3436 : * When debugging, it's easier to use __GFP_ZERO here; but it's better
3437 : * for latency not to zero a page while GFP_ATOMIC and holding locks.
3438 : */
3439 0 : page = alloc_page(gfp_mask | __GFP_HIGHMEM);
3440 :
3441 0 : si = get_swap_device(entry);
3442 0 : if (!si) {
3443 : /*
3444 : * An acceptable race has occurred since the failing
3445 : * __swap_duplicate(): the swap device may be swapoff
3446 : */
3447 : goto outer;
3448 : }
3449 0 : spin_lock(&si->lock);
3450 :
3451 0 : offset = swp_offset(entry);
3452 :
3453 0 : ci = lock_cluster(si, offset);
3454 :
3455 0 : count = swap_count(si->swap_map[offset]);
3456 :
3457 0 : if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
3458 : /*
3459 : * The higher the swap count, the more likely it is that tasks
3460 : * will race to add swap count continuation: we need to avoid
3461 : * over-provisioning.
3462 : */
3463 : goto out;
3464 : }
3465 :
3466 0 : if (!page) {
3467 : ret = -ENOMEM;
3468 : goto out;
3469 : }
3470 :
3471 : /*
3472 : * We are fortunate that although vmalloc_to_page uses pte_offset_map,
3473 : * no architecture is using highmem pages for kernel page tables: so it
3474 : * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
3475 : */
3476 0 : head = vmalloc_to_page(si->swap_map + offset);
3477 0 : offset &= ~PAGE_MASK;
3478 :
3479 0 : spin_lock(&si->cont_lock);
3480 : /*
3481 : * Page allocation does not initialize the page's lru field,
3482 : * but it does always reset its private field.
3483 : */
3484 0 : if (!page_private(head)) {
3485 0 : BUG_ON(count & COUNT_CONTINUED);
3486 0 : INIT_LIST_HEAD(&head->lru);
3487 0 : set_page_private(head, SWP_CONTINUED);
3488 0 : si->flags |= SWP_CONTINUED;
3489 : }
3490 :
3491 0 : list_for_each_entry(list_page, &head->lru, lru) {
3492 : unsigned char *map;
3493 :
3494 : /*
3495 : * If the previous map said no continuation, but we've found
3496 : * a continuation page, free our allocation and use this one.
3497 : */
3498 0 : if (!(count & COUNT_CONTINUED))
3499 : goto out_unlock_cont;
3500 :
3501 0 : map = kmap_atomic(list_page) + offset;
3502 0 : count = *map;
3503 0 : kunmap_atomic(map);
3504 :
3505 : /*
3506 : * If this continuation count now has some space in it,
3507 : * free our allocation and use this one.
3508 : */
3509 0 : if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
3510 : goto out_unlock_cont;
3511 : }
3512 :
3513 0 : list_add_tail(&page->lru, &head->lru);
3514 0 : page = NULL; /* now it's attached, don't free it */
3515 : out_unlock_cont:
3516 0 : spin_unlock(&si->cont_lock);
3517 : out:
3518 0 : unlock_cluster(ci);
3519 0 : spin_unlock(&si->lock);
3520 : put_swap_device(si);
3521 : outer:
3522 0 : if (page)
3523 0 : __free_page(page);
3524 0 : return ret;
3525 : }
3526 :
3527 : /*
3528 : * swap_count_continued - when the original swap_map count is incremented
3529 : * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3530 : * into, carry if so, or else fail until a new continuation page is allocated;
3531 : * when the original swap_map count is decremented from 0 with continuation,
3532 : * borrow from the continuation and report whether it still holds more.
3533 : * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3534 : * lock.
3535 : */
3536 0 : static bool swap_count_continued(struct swap_info_struct *si,
3537 : pgoff_t offset, unsigned char count)
3538 : {
3539 : struct page *head;
3540 : struct page *page;
3541 : unsigned char *map;
3542 : bool ret;
3543 :
3544 0 : head = vmalloc_to_page(si->swap_map + offset);
3545 0 : if (page_private(head) != SWP_CONTINUED) {
3546 0 : BUG_ON(count & COUNT_CONTINUED);
3547 : return false; /* need to add count continuation */
3548 : }
3549 :
3550 0 : spin_lock(&si->cont_lock);
3551 0 : offset &= ~PAGE_MASK;
3552 0 : page = list_next_entry(head, lru);
3553 0 : map = kmap_atomic(page) + offset;
3554 :
3555 0 : if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
3556 : goto init_map; /* jump over SWAP_CONT_MAX checks */
3557 :
3558 0 : if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
3559 : /*
3560 : * Think of how you add 1 to 999
3561 : */
3562 0 : while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
3563 0 : kunmap_atomic(map);
3564 0 : page = list_next_entry(page, lru);
3565 0 : BUG_ON(page == head);
3566 0 : map = kmap_atomic(page) + offset;
3567 : }
3568 0 : if (*map == SWAP_CONT_MAX) {
3569 0 : kunmap_atomic(map);
3570 0 : page = list_next_entry(page, lru);
3571 0 : if (page == head) {
3572 : ret = false; /* add count continuation */
3573 : goto out;
3574 : }
3575 0 : map = kmap_atomic(page) + offset;
3576 0 : init_map: *map = 0; /* we didn't zero the page */
3577 : }
3578 0 : *map += 1;
3579 0 : kunmap_atomic(map);
3580 0 : while ((page = list_prev_entry(page, lru)) != head) {
3581 0 : map = kmap_atomic(page) + offset;
3582 0 : *map = COUNT_CONTINUED;
3583 0 : kunmap_atomic(map);
3584 : }
3585 : ret = true; /* incremented */
3586 :
3587 : } else { /* decrementing */
3588 : /*
3589 : * Think of how you subtract 1 from 1000
3590 : */
3591 0 : BUG_ON(count != COUNT_CONTINUED);
3592 0 : while (*map == COUNT_CONTINUED) {
3593 0 : kunmap_atomic(map);
3594 0 : page = list_next_entry(page, lru);
3595 0 : BUG_ON(page == head);
3596 0 : map = kmap_atomic(page) + offset;
3597 : }
3598 0 : BUG_ON(*map == 0);
3599 0 : *map -= 1;
3600 0 : if (*map == 0)
3601 0 : count = 0;
3602 0 : kunmap_atomic(map);
3603 0 : while ((page = list_prev_entry(page, lru)) != head) {
3604 0 : map = kmap_atomic(page) + offset;
3605 0 : *map = SWAP_CONT_MAX | count;
3606 0 : count = COUNT_CONTINUED;
3607 0 : kunmap_atomic(map);
3608 : }
3609 0 : ret = count == COUNT_CONTINUED;
3610 : }
3611 : out:
3612 0 : spin_unlock(&si->cont_lock);
3613 0 : return ret;
3614 : }
3615 :
3616 : /*
3617 : * free_swap_count_continuations - swapoff free all the continuation pages
3618 : * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3619 : */
3620 0 : static void free_swap_count_continuations(struct swap_info_struct *si)
3621 : {
3622 : pgoff_t offset;
3623 :
3624 0 : for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
3625 : struct page *head;
3626 0 : head = vmalloc_to_page(si->swap_map + offset);
3627 0 : if (page_private(head)) {
3628 : struct page *page, *next;
3629 :
3630 0 : list_for_each_entry_safe(page, next, &head->lru, lru) {
3631 0 : list_del(&page->lru);
3632 0 : __free_page(page);
3633 : }
3634 : }
3635 : }
3636 0 : }
3637 :
3638 : #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
3639 : void __folio_throttle_swaprate(struct folio *folio, gfp_t gfp)
3640 : {
3641 : struct swap_info_struct *si, *next;
3642 : int nid = folio_nid(folio);
3643 :
3644 : if (!(gfp & __GFP_IO))
3645 : return;
3646 :
3647 : if (!blk_cgroup_congested())
3648 : return;
3649 :
3650 : /*
3651 : * We've already scheduled a throttle, avoid taking the global swap
3652 : * lock.
3653 : */
3654 : if (current->throttle_disk)
3655 : return;
3656 :
3657 : spin_lock(&swap_avail_lock);
3658 : plist_for_each_entry_safe(si, next, &swap_avail_heads[nid],
3659 : avail_lists[nid]) {
3660 : if (si->bdev) {
3661 : blkcg_schedule_throttle(si->bdev->bd_disk, true);
3662 : break;
3663 : }
3664 : }
3665 : spin_unlock(&swap_avail_lock);
3666 : }
3667 : #endif
3668 :
3669 1 : static int __init swapfile_init(void)
3670 : {
3671 : int nid;
3672 :
3673 1 : swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head),
3674 : GFP_KERNEL);
3675 1 : if (!swap_avail_heads) {
3676 0 : pr_emerg("Not enough memory for swap heads, swap is disabled\n");
3677 0 : return -ENOMEM;
3678 : }
3679 :
3680 1 : for_each_node(nid)
3681 2 : plist_head_init(&swap_avail_heads[nid]);
3682 :
3683 1 : swapfile_maximum_size = arch_max_swapfile_size();
3684 :
3685 : #ifdef CONFIG_MIGRATION
3686 1 : if (swapfile_maximum_size >= (1UL << SWP_MIG_TOTAL_BITS))
3687 0 : swap_migration_ad_supported = true;
3688 : #endif /* CONFIG_MIGRATION */
3689 :
3690 : return 0;
3691 : }
3692 : subsys_initcall(swapfile_init);
|
