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