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/mm.h>
10 #include <linux/sched/mm.h>
11 #include <linux/sched/task.h>
12 #include <linux/hugetlb.h>
13 #include <linux/mman.h>
14 #include <linux/slab.h>
15 #include <linux/kernel_stat.h>
16 #include <linux/swap.h>
17 #include <linux/vmalloc.h>
18 #include <linux/pagemap.h>
19 #include <linux/namei.h>
20 #include <linux/shmem_fs.h>
21 #include <linux/blkdev.h>
22 #include <linux/random.h>
23 #include <linux/writeback.h>
24 #include <linux/proc_fs.h>
25 #include <linux/seq_file.h>
26 #include <linux/init.h>
27 #include <linux/ksm.h>
28 #include <linux/rmap.h>
29 #include <linux/security.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mutex.h>
32 #include <linux/capability.h>
33 #include <linux/syscalls.h>
34 #include <linux/memcontrol.h>
35 #include <linux/poll.h>
36 #include <linux/oom.h>
37 #include <linux/frontswap.h>
38 #include <linux/swapfile.h>
39 #include <linux/export.h>
40 #include <linux/swap_slots.h>
41 #include <linux/sort.h>
42
43 #include <asm/tlbflush.h>
44 #include <linux/swapops.h>
45 #include <linux/swap_cgroup.h>
46 #include <linux/zswapd.h>
47
48 static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
49 unsigned char);
50 static void free_swap_count_continuations(struct swap_info_struct *);
51 static sector_t map_swap_entry(swp_entry_t, struct block_device**);
52
53 DEFINE_SPINLOCK(swap_lock);
54 static unsigned int nr_swapfiles;
55 atomic_long_t nr_swap_pages;
56 /*
57 * Some modules use swappable objects and may try to swap them out under
58 * memory pressure (via the shrinker). Before doing so, they may wish to
59 * check to see if any swap space is available.
60 */
61 EXPORT_SYMBOL_GPL(nr_swap_pages);
62 /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
63 long total_swap_pages;
64 static int least_priority = -1;
65
66 static const char Bad_file[] = "Bad swap file entry ";
67 static const char Unused_file[] = "Unused swap file entry ";
68 static const char Bad_offset[] = "Bad swap offset entry ";
69 static const char Unused_offset[] = "Unused swap offset entry ";
70
71 /*
72 * all active swap_info_structs
73 * protected with swap_lock, and ordered by priority.
74 */
75 PLIST_HEAD(swap_active_head);
76
77 /*
78 * all available (active, not full) swap_info_structs
79 * protected with swap_avail_lock, ordered by priority.
80 * This is used by get_swap_page() instead of swap_active_head
81 * because swap_active_head includes all swap_info_structs,
82 * but get_swap_page() doesn't need to look at full ones.
83 * This uses its own lock instead of swap_lock because when a
84 * swap_info_struct changes between not-full/full, it needs to
85 * add/remove itself to/from this list, but the swap_info_struct->lock
86 * is held and the locking order requires swap_lock to be taken
87 * before any swap_info_struct->lock.
88 */
89 static struct plist_head *swap_avail_heads;
90 static DEFINE_SPINLOCK(swap_avail_lock);
91
92 struct swap_info_struct *swap_info[MAX_SWAPFILES];
93
94 static DEFINE_MUTEX(swapon_mutex);
95
96 static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
97 /* Activity counter to indicate that a swapon or swapoff has occurred */
98 static atomic_t proc_poll_event = ATOMIC_INIT(0);
99
100 atomic_t nr_rotate_swap = ATOMIC_INIT(0);
101
swap_type_to_swap_info(int type)102 static struct swap_info_struct *swap_type_to_swap_info(int type)
103 {
104 if (type >= READ_ONCE(nr_swapfiles))
105 return NULL;
106
107 smp_rmb(); /* Pairs with smp_wmb in alloc_swap_info. */
108 return READ_ONCE(swap_info[type]);
109 }
110
swap_count(unsigned char ent)111 static inline unsigned char swap_count(unsigned char ent)
112 {
113 return ent & ~SWAP_HAS_CACHE; /* may include COUNT_CONTINUED flag */
114 }
115
116 /* Reclaim the swap entry anyway if possible */
117 #define TTRS_ANYWAY 0x1
118 /*
119 * Reclaim the swap entry if there are no more mappings of the
120 * corresponding page
121 */
122 #define TTRS_UNMAPPED 0x2
123 /* Reclaim the swap entry if swap is getting full*/
124 #define TTRS_FULL 0x4
125
126 /* returns 1 if swap entry is freed */
__try_to_reclaim_swap(struct swap_info_struct * si,unsigned long offset,unsigned long flags)127 static int __try_to_reclaim_swap(struct swap_info_struct *si,
128 unsigned long offset, unsigned long flags)
129 {
130 swp_entry_t entry = swp_entry(si->type, offset);
131 struct page *page;
132 int ret = 0;
133
134 page = find_get_page(swap_address_space(entry), offset);
135 if (!page)
136 return 0;
137 /*
138 * When this function is called from scan_swap_map_slots() and it's
139 * called by vmscan.c at reclaiming pages. So, we hold a lock on a page,
140 * here. We have to use trylock for avoiding deadlock. This is a special
141 * case and you should use try_to_free_swap() with explicit lock_page()
142 * in usual operations.
143 */
144 if (trylock_page(page)) {
145 if ((flags & TTRS_ANYWAY) ||
146 ((flags & TTRS_UNMAPPED) && !page_mapped(page)) ||
147 ((flags & TTRS_FULL) && mem_cgroup_swap_full(page)))
148 ret = try_to_free_swap(page);
149 unlock_page(page);
150 }
151 put_page(page);
152 return ret;
153 }
154
first_se(struct swap_info_struct * sis)155 static inline struct swap_extent *first_se(struct swap_info_struct *sis)
156 {
157 struct rb_node *rb = rb_first(&sis->swap_extent_root);
158 return rb_entry(rb, struct swap_extent, rb_node);
159 }
160
next_se(struct swap_extent * se)161 static inline struct swap_extent *next_se(struct swap_extent *se)
162 {
163 struct rb_node *rb = rb_next(&se->rb_node);
164 return rb ? rb_entry(rb, struct swap_extent, rb_node) : NULL;
165 }
166
167 /*
168 * swapon tell device that all the old swap contents can be discarded,
169 * to allow the swap device to optimize its wear-levelling.
170 */
discard_swap(struct swap_info_struct * si)171 static int discard_swap(struct swap_info_struct *si)
172 {
173 struct swap_extent *se;
174 sector_t start_block;
175 sector_t nr_blocks;
176 int err = 0;
177
178 /* Do not discard the swap header page! */
179 se = first_se(si);
180 start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
181 nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
182 if (nr_blocks) {
183 err = blkdev_issue_discard(si->bdev, start_block,
184 nr_blocks, GFP_KERNEL, 0);
185 if (err)
186 return err;
187 cond_resched();
188 }
189
190 for (se = next_se(se); se; se = next_se(se)) {
191 start_block = se->start_block << (PAGE_SHIFT - 9);
192 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
193
194 err = blkdev_issue_discard(si->bdev, start_block,
195 nr_blocks, GFP_KERNEL, 0);
196 if (err)
197 break;
198
199 cond_resched();
200 }
201 return err; /* That will often be -EOPNOTSUPP */
202 }
203
204 static struct swap_extent *
offset_to_swap_extent(struct swap_info_struct * sis,unsigned long offset)205 offset_to_swap_extent(struct swap_info_struct *sis, unsigned long offset)
206 {
207 struct swap_extent *se;
208 struct rb_node *rb;
209
210 rb = sis->swap_extent_root.rb_node;
211 while (rb) {
212 se = rb_entry(rb, struct swap_extent, rb_node);
213 if (offset < se->start_page)
214 rb = rb->rb_left;
215 else if (offset >= se->start_page + se->nr_pages)
216 rb = rb->rb_right;
217 else
218 return se;
219 }
220 /* It *must* be present */
221 BUG();
222 }
223
swap_page_sector(struct page * page)224 sector_t swap_page_sector(struct page *page)
225 {
226 struct swap_info_struct *sis = page_swap_info(page);
227 struct swap_extent *se;
228 sector_t sector;
229 pgoff_t offset;
230
231 offset = __page_file_index(page);
232 se = offset_to_swap_extent(sis, offset);
233 sector = se->start_block + (offset - se->start_page);
234 return sector << (PAGE_SHIFT - 9);
235 }
236
237 /*
238 * swap allocation tell device that a cluster of swap can now be discarded,
239 * to allow the swap device to optimize its wear-levelling.
240 */
discard_swap_cluster(struct swap_info_struct * si,pgoff_t start_page,pgoff_t nr_pages)241 static void discard_swap_cluster(struct swap_info_struct *si,
242 pgoff_t start_page, pgoff_t nr_pages)
243 {
244 struct swap_extent *se = offset_to_swap_extent(si, start_page);
245
246 while (nr_pages) {
247 pgoff_t offset = start_page - se->start_page;
248 sector_t start_block = se->start_block + offset;
249 sector_t nr_blocks = se->nr_pages - offset;
250
251 if (nr_blocks > nr_pages)
252 nr_blocks = nr_pages;
253 start_page += nr_blocks;
254 nr_pages -= nr_blocks;
255
256 start_block <<= PAGE_SHIFT - 9;
257 nr_blocks <<= PAGE_SHIFT - 9;
258 if (blkdev_issue_discard(si->bdev, start_block,
259 nr_blocks, GFP_NOIO, 0))
260 break;
261
262 se = next_se(se);
263 }
264 }
265
266 #ifdef CONFIG_THP_SWAP
267 #define SWAPFILE_CLUSTER HPAGE_PMD_NR
268
269 #define swap_entry_size(size) (size)
270 #else
271 #define SWAPFILE_CLUSTER 256
272
273 /*
274 * Define swap_entry_size() as constant to let compiler to optimize
275 * out some code if !CONFIG_THP_SWAP
276 */
277 #define swap_entry_size(size) 1
278 #endif
279 #define LATENCY_LIMIT 256
280
cluster_set_flag(struct swap_cluster_info * info,unsigned int flag)281 static inline void cluster_set_flag(struct swap_cluster_info *info,
282 unsigned int flag)
283 {
284 info->flags = flag;
285 }
286
cluster_count(struct swap_cluster_info * info)287 static inline unsigned int cluster_count(struct swap_cluster_info *info)
288 {
289 return info->data;
290 }
291
cluster_set_count(struct swap_cluster_info * info,unsigned int c)292 static inline void cluster_set_count(struct swap_cluster_info *info,
293 unsigned int c)
294 {
295 info->data = c;
296 }
297
cluster_set_count_flag(struct swap_cluster_info * info,unsigned int c,unsigned int f)298 static inline void cluster_set_count_flag(struct swap_cluster_info *info,
299 unsigned int c, unsigned int f)
300 {
301 info->flags = f;
302 info->data = c;
303 }
304
cluster_next(struct swap_cluster_info * info)305 static inline unsigned int cluster_next(struct swap_cluster_info *info)
306 {
307 return info->data;
308 }
309
cluster_set_next(struct swap_cluster_info * info,unsigned int n)310 static inline void cluster_set_next(struct swap_cluster_info *info,
311 unsigned int n)
312 {
313 info->data = n;
314 }
315
cluster_set_next_flag(struct swap_cluster_info * info,unsigned int n,unsigned int f)316 static inline void cluster_set_next_flag(struct swap_cluster_info *info,
317 unsigned int n, unsigned int f)
318 {
319 info->flags = f;
320 info->data = n;
321 }
322
cluster_is_free(struct swap_cluster_info * info)323 static inline bool cluster_is_free(struct swap_cluster_info *info)
324 {
325 return info->flags & CLUSTER_FLAG_FREE;
326 }
327
cluster_is_null(struct swap_cluster_info * info)328 static inline bool cluster_is_null(struct swap_cluster_info *info)
329 {
330 return info->flags & CLUSTER_FLAG_NEXT_NULL;
331 }
332
cluster_set_null(struct swap_cluster_info * info)333 static inline void cluster_set_null(struct swap_cluster_info *info)
334 {
335 info->flags = CLUSTER_FLAG_NEXT_NULL;
336 info->data = 0;
337 }
338
cluster_is_huge(struct swap_cluster_info * info)339 static inline bool cluster_is_huge(struct swap_cluster_info *info)
340 {
341 if (IS_ENABLED(CONFIG_THP_SWAP))
342 return info->flags & CLUSTER_FLAG_HUGE;
343 return false;
344 }
345
cluster_clear_huge(struct swap_cluster_info * info)346 static inline void cluster_clear_huge(struct swap_cluster_info *info)
347 {
348 info->flags &= ~CLUSTER_FLAG_HUGE;
349 }
350
lock_cluster(struct swap_info_struct * si,unsigned long offset)351 static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si,
352 unsigned long offset)
353 {
354 struct swap_cluster_info *ci;
355
356 ci = si->cluster_info;
357 if (ci) {
358 ci += offset / SWAPFILE_CLUSTER;
359 spin_lock(&ci->lock);
360 }
361 return ci;
362 }
363
unlock_cluster(struct swap_cluster_info * ci)364 static inline void unlock_cluster(struct swap_cluster_info *ci)
365 {
366 if (ci)
367 spin_unlock(&ci->lock);
368 }
369
370 /*
371 * Determine the locking method in use for this device. Return
372 * swap_cluster_info if SSD-style cluster-based locking is in place.
373 */
lock_cluster_or_swap_info(struct swap_info_struct * si,unsigned long offset)374 static inline struct swap_cluster_info *lock_cluster_or_swap_info(
375 struct swap_info_struct *si, unsigned long offset)
376 {
377 struct swap_cluster_info *ci;
378
379 /* Try to use fine-grained SSD-style locking if available: */
380 ci = lock_cluster(si, offset);
381 /* Otherwise, fall back to traditional, coarse locking: */
382 if (!ci)
383 spin_lock(&si->lock);
384
385 return ci;
386 }
387
unlock_cluster_or_swap_info(struct swap_info_struct * si,struct swap_cluster_info * ci)388 static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si,
389 struct swap_cluster_info *ci)
390 {
391 if (ci)
392 unlock_cluster(ci);
393 else
394 spin_unlock(&si->lock);
395 }
396
cluster_list_empty(struct swap_cluster_list * list)397 static inline bool cluster_list_empty(struct swap_cluster_list *list)
398 {
399 return cluster_is_null(&list->head);
400 }
401
cluster_list_first(struct swap_cluster_list * list)402 static inline unsigned int cluster_list_first(struct swap_cluster_list *list)
403 {
404 return cluster_next(&list->head);
405 }
406
cluster_list_init(struct swap_cluster_list * list)407 static void cluster_list_init(struct swap_cluster_list *list)
408 {
409 cluster_set_null(&list->head);
410 cluster_set_null(&list->tail);
411 }
412
cluster_list_add_tail(struct swap_cluster_list * list,struct swap_cluster_info * ci,unsigned int idx)413 static void cluster_list_add_tail(struct swap_cluster_list *list,
414 struct swap_cluster_info *ci,
415 unsigned int idx)
416 {
417 if (cluster_list_empty(list)) {
418 cluster_set_next_flag(&list->head, idx, 0);
419 cluster_set_next_flag(&list->tail, idx, 0);
420 } else {
421 struct swap_cluster_info *ci_tail;
422 unsigned int tail = cluster_next(&list->tail);
423
424 /*
425 * Nested cluster lock, but both cluster locks are
426 * only acquired when we held swap_info_struct->lock
427 */
428 ci_tail = ci + tail;
429 spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING);
430 cluster_set_next(ci_tail, idx);
431 spin_unlock(&ci_tail->lock);
432 cluster_set_next_flag(&list->tail, idx, 0);
433 }
434 }
435
cluster_list_del_first(struct swap_cluster_list * list,struct swap_cluster_info * ci)436 static unsigned int cluster_list_del_first(struct swap_cluster_list *list,
437 struct swap_cluster_info *ci)
438 {
439 unsigned int idx;
440
441 idx = cluster_next(&list->head);
442 if (cluster_next(&list->tail) == idx) {
443 cluster_set_null(&list->head);
444 cluster_set_null(&list->tail);
445 } else
446 cluster_set_next_flag(&list->head,
447 cluster_next(&ci[idx]), 0);
448
449 return idx;
450 }
451
452 /* Add a cluster to discard list and schedule it to do discard */
swap_cluster_schedule_discard(struct swap_info_struct * si,unsigned int idx)453 static void swap_cluster_schedule_discard(struct swap_info_struct *si,
454 unsigned int idx)
455 {
456 /*
457 * If scan_swap_map() can't find a free cluster, it will check
458 * si->swap_map directly. To make sure the discarding cluster isn't
459 * taken by scan_swap_map(), mark the swap entries bad (occupied). It
460 * will be cleared after discard
461 */
462 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
463 SWAP_MAP_BAD, SWAPFILE_CLUSTER);
464
465 cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx);
466
467 schedule_work(&si->discard_work);
468 }
469
__free_cluster(struct swap_info_struct * si,unsigned long idx)470 static void __free_cluster(struct swap_info_struct *si, unsigned long idx)
471 {
472 struct swap_cluster_info *ci = si->cluster_info;
473
474 cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE);
475 cluster_list_add_tail(&si->free_clusters, ci, idx);
476 }
477
478 /*
479 * Doing discard actually. After a cluster discard is finished, the cluster
480 * will be added to free cluster list. caller should hold si->lock.
481 */
swap_do_scheduled_discard(struct swap_info_struct * si)482 static void swap_do_scheduled_discard(struct swap_info_struct *si)
483 {
484 struct swap_cluster_info *info, *ci;
485 unsigned int idx;
486
487 info = si->cluster_info;
488
489 while (!cluster_list_empty(&si->discard_clusters)) {
490 idx = cluster_list_del_first(&si->discard_clusters, info);
491 spin_unlock(&si->lock);
492
493 discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
494 SWAPFILE_CLUSTER);
495
496 spin_lock(&si->lock);
497 ci = lock_cluster(si, idx * SWAPFILE_CLUSTER);
498 __free_cluster(si, idx);
499 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
500 0, SWAPFILE_CLUSTER);
501 unlock_cluster(ci);
502 }
503 }
504
swap_discard_work(struct work_struct * work)505 static void swap_discard_work(struct work_struct *work)
506 {
507 struct swap_info_struct *si;
508
509 si = container_of(work, struct swap_info_struct, discard_work);
510
511 spin_lock(&si->lock);
512 swap_do_scheduled_discard(si);
513 spin_unlock(&si->lock);
514 }
515
alloc_cluster(struct swap_info_struct * si,unsigned long idx)516 static void alloc_cluster(struct swap_info_struct *si, unsigned long idx)
517 {
518 struct swap_cluster_info *ci = si->cluster_info;
519
520 VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx);
521 cluster_list_del_first(&si->free_clusters, ci);
522 cluster_set_count_flag(ci + idx, 0, 0);
523 }
524
free_cluster(struct swap_info_struct * si,unsigned long idx)525 static void free_cluster(struct swap_info_struct *si, unsigned long idx)
526 {
527 struct swap_cluster_info *ci = si->cluster_info + idx;
528
529 VM_BUG_ON(cluster_count(ci) != 0);
530 /*
531 * If the swap is discardable, prepare discard the cluster
532 * instead of free it immediately. The cluster will be freed
533 * after discard.
534 */
535 if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
536 (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
537 swap_cluster_schedule_discard(si, idx);
538 return;
539 }
540
541 __free_cluster(si, idx);
542 }
543
544 /*
545 * The cluster corresponding to page_nr will be used. The cluster will be
546 * removed from free cluster list and its usage counter will be increased.
547 */
inc_cluster_info_page(struct swap_info_struct * p,struct swap_cluster_info * cluster_info,unsigned long page_nr)548 static void inc_cluster_info_page(struct swap_info_struct *p,
549 struct swap_cluster_info *cluster_info, unsigned long page_nr)
550 {
551 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
552
553 if (!cluster_info)
554 return;
555 if (cluster_is_free(&cluster_info[idx]))
556 alloc_cluster(p, idx);
557
558 VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
559 cluster_set_count(&cluster_info[idx],
560 cluster_count(&cluster_info[idx]) + 1);
561 }
562
563 /*
564 * The cluster corresponding to page_nr decreases one usage. If the usage
565 * counter becomes 0, which means no page in the cluster is in using, we can
566 * optionally discard the cluster and add it to free cluster list.
567 */
dec_cluster_info_page(struct swap_info_struct * p,struct swap_cluster_info * cluster_info,unsigned long page_nr)568 static void dec_cluster_info_page(struct swap_info_struct *p,
569 struct swap_cluster_info *cluster_info, unsigned long page_nr)
570 {
571 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
572
573 if (!cluster_info)
574 return;
575
576 VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
577 cluster_set_count(&cluster_info[idx],
578 cluster_count(&cluster_info[idx]) - 1);
579
580 if (cluster_count(&cluster_info[idx]) == 0)
581 free_cluster(p, idx);
582 }
583
584 /*
585 * It's possible scan_swap_map() uses a free cluster in the middle of free
586 * cluster list. Avoiding such abuse to avoid list corruption.
587 */
588 static bool
scan_swap_map_ssd_cluster_conflict(struct swap_info_struct * si,unsigned long offset)589 scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
590 unsigned long offset)
591 {
592 struct percpu_cluster *percpu_cluster;
593 bool conflict;
594
595 offset /= SWAPFILE_CLUSTER;
596 conflict = !cluster_list_empty(&si->free_clusters) &&
597 offset != cluster_list_first(&si->free_clusters) &&
598 cluster_is_free(&si->cluster_info[offset]);
599
600 if (!conflict)
601 return false;
602
603 percpu_cluster = this_cpu_ptr(si->percpu_cluster);
604 cluster_set_null(&percpu_cluster->index);
605 return true;
606 }
607
608 /*
609 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
610 * might involve allocating a new cluster for current CPU too.
611 */
scan_swap_map_try_ssd_cluster(struct swap_info_struct * si,unsigned long * offset,unsigned long * scan_base)612 static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
613 unsigned long *offset, unsigned long *scan_base)
614 {
615 struct percpu_cluster *cluster;
616 struct swap_cluster_info *ci;
617 unsigned long tmp, max;
618
619 new_cluster:
620 cluster = this_cpu_ptr(si->percpu_cluster);
621 if (cluster_is_null(&cluster->index)) {
622 if (!cluster_list_empty(&si->free_clusters)) {
623 cluster->index = si->free_clusters.head;
624 cluster->next = cluster_next(&cluster->index) *
625 SWAPFILE_CLUSTER;
626 } else if (!cluster_list_empty(&si->discard_clusters)) {
627 /*
628 * we don't have free cluster but have some clusters in
629 * discarding, do discard now and reclaim them, then
630 * reread cluster_next_cpu since we dropped si->lock
631 */
632 swap_do_scheduled_discard(si);
633 *scan_base = this_cpu_read(*si->cluster_next_cpu);
634 *offset = *scan_base;
635 goto new_cluster;
636 } else
637 return false;
638 }
639
640 /*
641 * Other CPUs can use our cluster if they can't find a free cluster,
642 * check if there is still free entry in the cluster
643 */
644 tmp = cluster->next;
645 max = min_t(unsigned long, si->max,
646 (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER);
647 if (tmp < max) {
648 ci = lock_cluster(si, tmp);
649 while (tmp < max) {
650 if (!si->swap_map[tmp])
651 break;
652 tmp++;
653 }
654 unlock_cluster(ci);
655 }
656 if (tmp >= max) {
657 cluster_set_null(&cluster->index);
658 goto new_cluster;
659 }
660 cluster->next = tmp + 1;
661 *offset = tmp;
662 *scan_base = tmp;
663 return true;
664 }
665
__del_from_avail_list(struct swap_info_struct * p)666 static void __del_from_avail_list(struct swap_info_struct *p)
667 {
668 int nid;
669
670 for_each_node(nid)
671 plist_del(&p->avail_lists[nid], &swap_avail_heads[nid]);
672 }
673
del_from_avail_list(struct swap_info_struct * p)674 static void del_from_avail_list(struct swap_info_struct *p)
675 {
676 spin_lock(&swap_avail_lock);
677 __del_from_avail_list(p);
678 spin_unlock(&swap_avail_lock);
679 }
680
swap_range_alloc(struct swap_info_struct * si,unsigned long offset,unsigned int nr_entries)681 static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset,
682 unsigned int nr_entries)
683 {
684 unsigned int end = offset + nr_entries - 1;
685
686 if (offset == si->lowest_bit)
687 si->lowest_bit += nr_entries;
688 if (end == si->highest_bit)
689 WRITE_ONCE(si->highest_bit, si->highest_bit - nr_entries);
690 si->inuse_pages += nr_entries;
691 if (si->inuse_pages == si->pages) {
692 si->lowest_bit = si->max;
693 si->highest_bit = 0;
694 del_from_avail_list(si);
695 }
696 }
697
add_to_avail_list(struct swap_info_struct * p)698 static void add_to_avail_list(struct swap_info_struct *p)
699 {
700 int nid;
701
702 spin_lock(&swap_avail_lock);
703 for_each_node(nid) {
704 WARN_ON(!plist_node_empty(&p->avail_lists[nid]));
705 plist_add(&p->avail_lists[nid], &swap_avail_heads[nid]);
706 }
707 spin_unlock(&swap_avail_lock);
708 }
709
swap_range_free(struct swap_info_struct * si,unsigned long offset,unsigned int nr_entries)710 static void swap_range_free(struct swap_info_struct *si, unsigned long offset,
711 unsigned int nr_entries)
712 {
713 unsigned long begin = offset;
714 unsigned long end = offset + nr_entries - 1;
715 void (*swap_slot_free_notify)(struct block_device *, unsigned long);
716
717 if (offset < si->lowest_bit)
718 si->lowest_bit = offset;
719 if (end > si->highest_bit) {
720 bool was_full = !si->highest_bit;
721
722 WRITE_ONCE(si->highest_bit, end);
723 if (was_full && (si->flags & SWP_WRITEOK))
724 add_to_avail_list(si);
725 }
726 atomic_long_add(nr_entries, &nr_swap_pages);
727 si->inuse_pages -= nr_entries;
728 if (si->flags & SWP_BLKDEV)
729 swap_slot_free_notify =
730 si->bdev->bd_disk->fops->swap_slot_free_notify;
731 else
732 swap_slot_free_notify = NULL;
733 while (offset <= end) {
734 arch_swap_invalidate_page(si->type, offset);
735 frontswap_invalidate_page(si->type, offset);
736 if (swap_slot_free_notify)
737 swap_slot_free_notify(si->bdev, offset);
738 offset++;
739 }
740 clear_shadow_from_swap_cache(si->type, begin, end);
741 }
742
set_cluster_next(struct swap_info_struct * si,unsigned long next)743 static void set_cluster_next(struct swap_info_struct *si, unsigned long next)
744 {
745 unsigned long prev;
746
747 if (!(si->flags & SWP_SOLIDSTATE)) {
748 si->cluster_next = next;
749 return;
750 }
751
752 prev = this_cpu_read(*si->cluster_next_cpu);
753 /*
754 * Cross the swap address space size aligned trunk, choose
755 * another trunk randomly to avoid lock contention on swap
756 * address space if possible.
757 */
758 if ((prev >> SWAP_ADDRESS_SPACE_SHIFT) !=
759 (next >> SWAP_ADDRESS_SPACE_SHIFT)) {
760 /* No free swap slots available */
761 if (si->highest_bit <= si->lowest_bit)
762 return;
763 next = si->lowest_bit +
764 prandom_u32_max(si->highest_bit - si->lowest_bit + 1);
765 next = ALIGN_DOWN(next, SWAP_ADDRESS_SPACE_PAGES);
766 next = max_t(unsigned int, next, si->lowest_bit);
767 }
768 this_cpu_write(*si->cluster_next_cpu, next);
769 }
770
scan_swap_map_slots(struct swap_info_struct * si,unsigned char usage,int nr,swp_entry_t slots[])771 static int scan_swap_map_slots(struct swap_info_struct *si,
772 unsigned char usage, int nr,
773 swp_entry_t slots[])
774 {
775 struct swap_cluster_info *ci;
776 unsigned long offset;
777 unsigned long scan_base;
778 unsigned long last_in_cluster = 0;
779 int latency_ration = LATENCY_LIMIT;
780 int n_ret = 0;
781 bool scanned_many = false;
782
783 /*
784 * We try to cluster swap pages by allocating them sequentially
785 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
786 * way, however, we resort to first-free allocation, starting
787 * a new cluster. This prevents us from scattering swap pages
788 * all over the entire swap partition, so that we reduce
789 * overall disk seek times between swap pages. -- sct
790 * But we do now try to find an empty cluster. -Andrea
791 * And we let swap pages go all over an SSD partition. Hugh
792 */
793
794 si->flags += SWP_SCANNING;
795 /*
796 * Use percpu scan base for SSD to reduce lock contention on
797 * cluster and swap cache. For HDD, sequential access is more
798 * important.
799 */
800 if (si->flags & SWP_SOLIDSTATE)
801 scan_base = this_cpu_read(*si->cluster_next_cpu);
802 else
803 scan_base = si->cluster_next;
804 offset = scan_base;
805
806 /* SSD algorithm */
807 if (si->cluster_info) {
808 if (!scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
809 goto scan;
810 } else if (unlikely(!si->cluster_nr--)) {
811 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
812 si->cluster_nr = SWAPFILE_CLUSTER - 1;
813 goto checks;
814 }
815
816 spin_unlock(&si->lock);
817
818 /*
819 * If seek is expensive, start searching for new cluster from
820 * start of partition, to minimize the span of allocated swap.
821 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
822 * case, just handled by scan_swap_map_try_ssd_cluster() above.
823 */
824 scan_base = offset = si->lowest_bit;
825 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
826
827 /* Locate the first empty (unaligned) cluster */
828 for (; last_in_cluster <= si->highest_bit; offset++) {
829 if (si->swap_map[offset])
830 last_in_cluster = offset + SWAPFILE_CLUSTER;
831 else if (offset == last_in_cluster) {
832 spin_lock(&si->lock);
833 offset -= SWAPFILE_CLUSTER - 1;
834 si->cluster_next = offset;
835 si->cluster_nr = SWAPFILE_CLUSTER - 1;
836 goto checks;
837 }
838 if (unlikely(--latency_ration < 0)) {
839 cond_resched();
840 latency_ration = LATENCY_LIMIT;
841 }
842 }
843
844 offset = scan_base;
845 spin_lock(&si->lock);
846 si->cluster_nr = SWAPFILE_CLUSTER - 1;
847 }
848
849 checks:
850 if (si->cluster_info) {
851 while (scan_swap_map_ssd_cluster_conflict(si, offset)) {
852 /* take a break if we already got some slots */
853 if (n_ret)
854 goto done;
855 if (!scan_swap_map_try_ssd_cluster(si, &offset,
856 &scan_base))
857 goto scan;
858 }
859 }
860 if (!(si->flags & SWP_WRITEOK))
861 goto no_page;
862 if (!si->highest_bit)
863 goto no_page;
864 if (offset > si->highest_bit)
865 scan_base = offset = si->lowest_bit;
866
867 ci = lock_cluster(si, offset);
868 /* reuse swap entry of cache-only swap if not busy. */
869 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
870 int swap_was_freed;
871 unlock_cluster(ci);
872 spin_unlock(&si->lock);
873 swap_was_freed = __try_to_reclaim_swap(si, offset, TTRS_ANYWAY);
874 spin_lock(&si->lock);
875 /* entry was freed successfully, try to use this again */
876 if (swap_was_freed)
877 goto checks;
878 goto scan; /* check next one */
879 }
880
881 if (si->swap_map[offset]) {
882 unlock_cluster(ci);
883 if (!n_ret)
884 goto scan;
885 else
886 goto done;
887 }
888 WRITE_ONCE(si->swap_map[offset], usage);
889 inc_cluster_info_page(si, si->cluster_info, offset);
890 unlock_cluster(ci);
891
892 swap_range_alloc(si, offset, 1);
893 slots[n_ret++] = swp_entry(si->type, offset);
894
895 /* got enough slots or reach max slots? */
896 if ((n_ret == nr) || (offset >= si->highest_bit))
897 goto done;
898
899 /* search for next available slot */
900
901 /* time to take a break? */
902 if (unlikely(--latency_ration < 0)) {
903 if (n_ret)
904 goto done;
905 spin_unlock(&si->lock);
906 cond_resched();
907 spin_lock(&si->lock);
908 latency_ration = LATENCY_LIMIT;
909 }
910
911 /* try to get more slots in cluster */
912 if (si->cluster_info) {
913 if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
914 goto checks;
915 } else if (si->cluster_nr && !si->swap_map[++offset]) {
916 /* non-ssd case, still more slots in cluster? */
917 --si->cluster_nr;
918 goto checks;
919 }
920
921 /*
922 * Even if there's no free clusters available (fragmented),
923 * try to scan a little more quickly with lock held unless we
924 * have scanned too many slots already.
925 */
926 if (!scanned_many) {
927 unsigned long scan_limit;
928
929 if (offset < scan_base)
930 scan_limit = scan_base;
931 else
932 scan_limit = si->highest_bit;
933 for (; offset <= scan_limit && --latency_ration > 0;
934 offset++) {
935 if (!si->swap_map[offset])
936 goto checks;
937 }
938 }
939
940 done:
941 set_cluster_next(si, offset + 1);
942 si->flags -= SWP_SCANNING;
943 return n_ret;
944
945 scan:
946 spin_unlock(&si->lock);
947 while (++offset <= READ_ONCE(si->highest_bit)) {
948 if (data_race(!si->swap_map[offset])) {
949 spin_lock(&si->lock);
950 goto checks;
951 }
952 if (vm_swap_full() &&
953 READ_ONCE(si->swap_map[offset]) == SWAP_HAS_CACHE) {
954 spin_lock(&si->lock);
955 goto checks;
956 }
957 if (unlikely(--latency_ration < 0)) {
958 cond_resched();
959 latency_ration = LATENCY_LIMIT;
960 scanned_many = true;
961 }
962 }
963 offset = si->lowest_bit;
964 while (offset < scan_base) {
965 if (data_race(!si->swap_map[offset])) {
966 spin_lock(&si->lock);
967 goto checks;
968 }
969 if (vm_swap_full() &&
970 READ_ONCE(si->swap_map[offset]) == SWAP_HAS_CACHE) {
971 spin_lock(&si->lock);
972 goto checks;
973 }
974 if (unlikely(--latency_ration < 0)) {
975 cond_resched();
976 latency_ration = LATENCY_LIMIT;
977 scanned_many = true;
978 }
979 offset++;
980 }
981 spin_lock(&si->lock);
982
983 no_page:
984 si->flags -= SWP_SCANNING;
985 return n_ret;
986 }
987
swap_alloc_cluster(struct swap_info_struct * si,swp_entry_t * slot)988 static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot)
989 {
990 unsigned long idx;
991 struct swap_cluster_info *ci;
992 unsigned long offset, i;
993 unsigned char *map;
994
995 /*
996 * Should not even be attempting cluster allocations when huge
997 * page swap is disabled. Warn and fail the allocation.
998 */
999 if (!IS_ENABLED(CONFIG_THP_SWAP)) {
1000 VM_WARN_ON_ONCE(1);
1001 return 0;
1002 }
1003
1004 if (cluster_list_empty(&si->free_clusters))
1005 return 0;
1006
1007 idx = cluster_list_first(&si->free_clusters);
1008 offset = idx * SWAPFILE_CLUSTER;
1009 ci = lock_cluster(si, offset);
1010 alloc_cluster(si, idx);
1011 cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE);
1012
1013 map = si->swap_map + offset;
1014 for (i = 0; i < SWAPFILE_CLUSTER; i++)
1015 map[i] = SWAP_HAS_CACHE;
1016 unlock_cluster(ci);
1017 swap_range_alloc(si, offset, SWAPFILE_CLUSTER);
1018 *slot = swp_entry(si->type, offset);
1019
1020 return 1;
1021 }
1022
swap_free_cluster(struct swap_info_struct * si,unsigned long idx)1023 static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx)
1024 {
1025 unsigned long offset = idx * SWAPFILE_CLUSTER;
1026 struct swap_cluster_info *ci;
1027
1028 ci = lock_cluster(si, offset);
1029 memset(si->swap_map + offset, 0, SWAPFILE_CLUSTER);
1030 cluster_set_count_flag(ci, 0, 0);
1031 free_cluster(si, idx);
1032 unlock_cluster(ci);
1033 swap_range_free(si, offset, SWAPFILE_CLUSTER);
1034 }
1035
scan_swap_map(struct swap_info_struct * si,unsigned char usage)1036 static unsigned long scan_swap_map(struct swap_info_struct *si,
1037 unsigned char usage)
1038 {
1039 swp_entry_t entry;
1040 int n_ret;
1041
1042 n_ret = scan_swap_map_slots(si, usage, 1, &entry);
1043
1044 if (n_ret)
1045 return swp_offset(entry);
1046 else
1047 return 0;
1048
1049 }
1050
get_swap_pages(int n_goal,swp_entry_t swp_entries[],int entry_size)1051 int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size)
1052 {
1053 unsigned long size = swap_entry_size(entry_size);
1054 struct swap_info_struct *si, *next;
1055 long avail_pgs;
1056 int n_ret = 0;
1057 int node;
1058
1059 /* Only single cluster request supported */
1060 WARN_ON_ONCE(n_goal > 1 && size == SWAPFILE_CLUSTER);
1061
1062 spin_lock(&swap_avail_lock);
1063
1064 avail_pgs = atomic_long_read(&nr_swap_pages) / size;
1065 if (avail_pgs <= 0) {
1066 spin_unlock(&swap_avail_lock);
1067 goto noswap;
1068 }
1069
1070 n_goal = min3((long)n_goal, (long)SWAP_BATCH, avail_pgs);
1071
1072 atomic_long_sub(n_goal * size, &nr_swap_pages);
1073
1074 start_over:
1075 node = numa_node_id();
1076 plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) {
1077 /* requeue si to after same-priority siblings */
1078 plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]);
1079 spin_unlock(&swap_avail_lock);
1080 spin_lock(&si->lock);
1081 if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) {
1082 spin_lock(&swap_avail_lock);
1083 if (plist_node_empty(&si->avail_lists[node])) {
1084 spin_unlock(&si->lock);
1085 goto nextsi;
1086 }
1087 WARN(!si->highest_bit,
1088 "swap_info %d in list but !highest_bit\n",
1089 si->type);
1090 WARN(!(si->flags & SWP_WRITEOK),
1091 "swap_info %d in list but !SWP_WRITEOK\n",
1092 si->type);
1093 __del_from_avail_list(si);
1094 spin_unlock(&si->lock);
1095 goto nextsi;
1096 }
1097 if (size == SWAPFILE_CLUSTER) {
1098 if (si->flags & SWP_BLKDEV)
1099 n_ret = swap_alloc_cluster(si, swp_entries);
1100 } else
1101 n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE,
1102 n_goal, swp_entries);
1103 spin_unlock(&si->lock);
1104 if (n_ret || size == SWAPFILE_CLUSTER)
1105 goto check_out;
1106 pr_debug("scan_swap_map of si %d failed to find offset\n",
1107 si->type);
1108
1109 spin_lock(&swap_avail_lock);
1110 nextsi:
1111 /*
1112 * if we got here, it's likely that si was almost full before,
1113 * and since scan_swap_map() can drop the si->lock, multiple
1114 * callers probably all tried to get a page from the same si
1115 * and it filled up before we could get one; or, the si filled
1116 * up between us dropping swap_avail_lock and taking si->lock.
1117 * Since we dropped the swap_avail_lock, the swap_avail_head
1118 * list may have been modified; so if next is still in the
1119 * swap_avail_head list then try it, otherwise start over
1120 * if we have not gotten any slots.
1121 */
1122 if (plist_node_empty(&next->avail_lists[node]))
1123 goto start_over;
1124 }
1125
1126 spin_unlock(&swap_avail_lock);
1127
1128 check_out:
1129 if (n_ret < n_goal)
1130 atomic_long_add((long)(n_goal - n_ret) * size,
1131 &nr_swap_pages);
1132 noswap:
1133 return n_ret;
1134 }
1135
1136 /* The only caller of this function is now suspend routine */
get_swap_page_of_type(int type)1137 swp_entry_t get_swap_page_of_type(int type)
1138 {
1139 struct swap_info_struct *si = swap_type_to_swap_info(type);
1140 pgoff_t offset;
1141
1142 if (!si)
1143 goto fail;
1144
1145 spin_lock(&si->lock);
1146 if (si->flags & SWP_WRITEOK) {
1147 /* This is called for allocating swap entry, not cache */
1148 offset = scan_swap_map(si, 1);
1149 if (offset) {
1150 atomic_long_dec(&nr_swap_pages);
1151 spin_unlock(&si->lock);
1152 return swp_entry(type, offset);
1153 }
1154 }
1155 spin_unlock(&si->lock);
1156 fail:
1157 return (swp_entry_t) {0};
1158 }
1159
__swap_info_get(swp_entry_t entry)1160 static struct swap_info_struct *__swap_info_get(swp_entry_t entry)
1161 {
1162 struct swap_info_struct *p;
1163 unsigned long offset;
1164
1165 if (!entry.val)
1166 goto out;
1167 p = swp_swap_info(entry);
1168 if (!p)
1169 goto bad_nofile;
1170 if (data_race(!(p->flags & SWP_USED)))
1171 goto bad_device;
1172 offset = swp_offset(entry);
1173 if (offset >= p->max)
1174 goto bad_offset;
1175 return p;
1176
1177 bad_offset:
1178 pr_err("swap_info_get: %s%08lx\n", Bad_offset, entry.val);
1179 goto out;
1180 bad_device:
1181 pr_err("swap_info_get: %s%08lx\n", Unused_file, entry.val);
1182 goto out;
1183 bad_nofile:
1184 pr_err("swap_info_get: %s%08lx\n", Bad_file, entry.val);
1185 out:
1186 return NULL;
1187 }
1188
_swap_info_get(swp_entry_t entry)1189 static struct swap_info_struct *_swap_info_get(swp_entry_t entry)
1190 {
1191 struct swap_info_struct *p;
1192
1193 p = __swap_info_get(entry);
1194 if (!p)
1195 goto out;
1196 if (data_race(!p->swap_map[swp_offset(entry)]))
1197 goto bad_free;
1198 return p;
1199
1200 bad_free:
1201 pr_err("swap_info_get: %s%08lx\n", Unused_offset, entry.val);
1202 out:
1203 return NULL;
1204 }
1205
swap_info_get(swp_entry_t entry)1206 static struct swap_info_struct *swap_info_get(swp_entry_t entry)
1207 {
1208 struct swap_info_struct *p;
1209
1210 p = _swap_info_get(entry);
1211 if (p)
1212 spin_lock(&p->lock);
1213 return p;
1214 }
1215
swap_info_get_cont(swp_entry_t entry,struct swap_info_struct * q)1216 static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry,
1217 struct swap_info_struct *q)
1218 {
1219 struct swap_info_struct *p;
1220
1221 p = _swap_info_get(entry);
1222
1223 if (p != q) {
1224 if (q != NULL)
1225 spin_unlock(&q->lock);
1226 if (p != NULL)
1227 spin_lock(&p->lock);
1228 }
1229 return p;
1230 }
1231
__swap_entry_free_locked(struct swap_info_struct * p,unsigned long offset,unsigned char usage)1232 static unsigned char __swap_entry_free_locked(struct swap_info_struct *p,
1233 unsigned long offset,
1234 unsigned char usage)
1235 {
1236 unsigned char count;
1237 unsigned char has_cache;
1238
1239 count = p->swap_map[offset];
1240
1241 has_cache = count & SWAP_HAS_CACHE;
1242 count &= ~SWAP_HAS_CACHE;
1243
1244 if (usage == SWAP_HAS_CACHE) {
1245 VM_BUG_ON(!has_cache);
1246 has_cache = 0;
1247 } else if (count == SWAP_MAP_SHMEM) {
1248 /*
1249 * Or we could insist on shmem.c using a special
1250 * swap_shmem_free() and free_shmem_swap_and_cache()...
1251 */
1252 count = 0;
1253 } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
1254 if (count == COUNT_CONTINUED) {
1255 if (swap_count_continued(p, offset, count))
1256 count = SWAP_MAP_MAX | COUNT_CONTINUED;
1257 else
1258 count = SWAP_MAP_MAX;
1259 } else
1260 count--;
1261 }
1262
1263 usage = count | has_cache;
1264 if (usage)
1265 WRITE_ONCE(p->swap_map[offset], usage);
1266 else
1267 WRITE_ONCE(p->swap_map[offset], SWAP_HAS_CACHE);
1268
1269 return usage;
1270 }
1271
1272 /*
1273 * Check whether swap entry is valid in the swap device. If so,
1274 * return pointer to swap_info_struct, and keep the swap entry valid
1275 * via preventing the swap device from being swapoff, until
1276 * put_swap_device() is called. Otherwise return NULL.
1277 *
1278 * The entirety of the RCU read critical section must come before the
1279 * return from or after the call to synchronize_rcu() in
1280 * enable_swap_info() or swapoff(). So if "si->flags & SWP_VALID" is
1281 * true, the si->map, si->cluster_info, etc. must be valid in the
1282 * critical section.
1283 *
1284 * Notice that swapoff or swapoff+swapon can still happen before the
1285 * rcu_read_lock() in get_swap_device() or after the rcu_read_unlock()
1286 * in put_swap_device() if there isn't any other way to prevent
1287 * swapoff, such as page lock, page table lock, etc. The caller must
1288 * be prepared for that. For example, the following situation is
1289 * possible.
1290 *
1291 * CPU1 CPU2
1292 * do_swap_page()
1293 * ... swapoff+swapon
1294 * __read_swap_cache_async()
1295 * swapcache_prepare()
1296 * __swap_duplicate()
1297 * // check swap_map
1298 * // verify PTE not changed
1299 *
1300 * In __swap_duplicate(), the swap_map need to be checked before
1301 * changing partly because the specified swap entry may be for another
1302 * swap device which has been swapoff. And in do_swap_page(), after
1303 * the page is read from the swap device, the PTE is verified not
1304 * changed with the page table locked to check whether the swap device
1305 * has been swapoff or swapoff+swapon.
1306 */
get_swap_device(swp_entry_t entry)1307 struct swap_info_struct *get_swap_device(swp_entry_t entry)
1308 {
1309 struct swap_info_struct *si;
1310 unsigned long offset;
1311
1312 if (!entry.val)
1313 goto out;
1314 si = swp_swap_info(entry);
1315 if (!si)
1316 goto bad_nofile;
1317
1318 rcu_read_lock();
1319 if (data_race(!(si->flags & SWP_VALID)))
1320 goto unlock_out;
1321 offset = swp_offset(entry);
1322 if (offset >= si->max)
1323 goto unlock_out;
1324
1325 return si;
1326 bad_nofile:
1327 pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val);
1328 out:
1329 return NULL;
1330 unlock_out:
1331 rcu_read_unlock();
1332 return NULL;
1333 }
1334
__swap_entry_free(struct swap_info_struct * p,swp_entry_t entry)1335 static unsigned char __swap_entry_free(struct swap_info_struct *p,
1336 swp_entry_t entry)
1337 {
1338 struct swap_cluster_info *ci;
1339 unsigned long offset = swp_offset(entry);
1340 unsigned char usage;
1341
1342 ci = lock_cluster_or_swap_info(p, offset);
1343 usage = __swap_entry_free_locked(p, offset, 1);
1344 unlock_cluster_or_swap_info(p, ci);
1345 if (!usage)
1346 free_swap_slot(entry);
1347
1348 return usage;
1349 }
1350
swap_entry_free(struct swap_info_struct * p,swp_entry_t entry)1351 static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry)
1352 {
1353 struct swap_cluster_info *ci;
1354 unsigned long offset = swp_offset(entry);
1355 unsigned char count;
1356
1357 ci = lock_cluster(p, offset);
1358 count = p->swap_map[offset];
1359 VM_BUG_ON(count != SWAP_HAS_CACHE);
1360 p->swap_map[offset] = 0;
1361 dec_cluster_info_page(p, p->cluster_info, offset);
1362 unlock_cluster(ci);
1363
1364 mem_cgroup_uncharge_swap(entry, 1);
1365 swap_range_free(p, offset, 1);
1366 }
1367
1368 /*
1369 * Caller has made sure that the swap device corresponding to entry
1370 * is still around or has not been recycled.
1371 */
swap_free(swp_entry_t entry)1372 void swap_free(swp_entry_t entry)
1373 {
1374 struct swap_info_struct *p;
1375
1376 p = _swap_info_get(entry);
1377 if (p)
1378 __swap_entry_free(p, entry);
1379 }
1380
1381 /*
1382 * Called after dropping swapcache to decrease refcnt to swap entries.
1383 */
put_swap_page(struct page * page,swp_entry_t entry)1384 void put_swap_page(struct page *page, swp_entry_t entry)
1385 {
1386 unsigned long offset = swp_offset(entry);
1387 unsigned long idx = offset / SWAPFILE_CLUSTER;
1388 struct swap_cluster_info *ci;
1389 struct swap_info_struct *si;
1390 unsigned char *map;
1391 unsigned int i, free_entries = 0;
1392 unsigned char val;
1393 int size = swap_entry_size(thp_nr_pages(page));
1394
1395 si = _swap_info_get(entry);
1396 if (!si)
1397 return;
1398
1399 ci = lock_cluster_or_swap_info(si, offset);
1400 if (size == SWAPFILE_CLUSTER) {
1401 VM_BUG_ON(!cluster_is_huge(ci));
1402 map = si->swap_map + offset;
1403 for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1404 val = map[i];
1405 VM_BUG_ON(!(val & SWAP_HAS_CACHE));
1406 if (val == SWAP_HAS_CACHE)
1407 free_entries++;
1408 }
1409 cluster_clear_huge(ci);
1410 if (free_entries == SWAPFILE_CLUSTER) {
1411 unlock_cluster_or_swap_info(si, ci);
1412 spin_lock(&si->lock);
1413 mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER);
1414 swap_free_cluster(si, idx);
1415 spin_unlock(&si->lock);
1416 return;
1417 }
1418 }
1419 for (i = 0; i < size; i++, entry.val++) {
1420 if (!__swap_entry_free_locked(si, offset + i, SWAP_HAS_CACHE)) {
1421 unlock_cluster_or_swap_info(si, ci);
1422 free_swap_slot(entry);
1423 if (i == size - 1)
1424 return;
1425 lock_cluster_or_swap_info(si, offset);
1426 }
1427 }
1428 unlock_cluster_or_swap_info(si, ci);
1429 }
1430
1431 #ifdef CONFIG_THP_SWAP
split_swap_cluster(swp_entry_t entry)1432 int split_swap_cluster(swp_entry_t entry)
1433 {
1434 struct swap_info_struct *si;
1435 struct swap_cluster_info *ci;
1436 unsigned long offset = swp_offset(entry);
1437
1438 si = _swap_info_get(entry);
1439 if (!si)
1440 return -EBUSY;
1441 ci = lock_cluster(si, offset);
1442 cluster_clear_huge(ci);
1443 unlock_cluster(ci);
1444 return 0;
1445 }
1446 #endif
1447
swp_entry_cmp(const void * ent1,const void * ent2)1448 static int swp_entry_cmp(const void *ent1, const void *ent2)
1449 {
1450 const swp_entry_t *e1 = ent1, *e2 = ent2;
1451
1452 return (int)swp_type(*e1) - (int)swp_type(*e2);
1453 }
1454
swapcache_free_entries(swp_entry_t * entries,int n)1455 void swapcache_free_entries(swp_entry_t *entries, int n)
1456 {
1457 struct swap_info_struct *p, *prev;
1458 int i;
1459
1460 if (n <= 0)
1461 return;
1462
1463 prev = NULL;
1464 p = NULL;
1465
1466 /*
1467 * Sort swap entries by swap device, so each lock is only taken once.
1468 * nr_swapfiles isn't absolutely correct, but the overhead of sort() is
1469 * so low that it isn't necessary to optimize further.
1470 */
1471 if (nr_swapfiles > 1)
1472 sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL);
1473 for (i = 0; i < n; ++i) {
1474 p = swap_info_get_cont(entries[i], prev);
1475 if (p)
1476 swap_entry_free(p, entries[i]);
1477 prev = p;
1478 }
1479 if (p)
1480 spin_unlock(&p->lock);
1481 }
1482
1483 /*
1484 * How many references to page are currently swapped out?
1485 * This does not give an exact answer when swap count is continued,
1486 * but does include the high COUNT_CONTINUED flag to allow for that.
1487 */
page_swapcount(struct page * page)1488 int page_swapcount(struct page *page)
1489 {
1490 int count = 0;
1491 struct swap_info_struct *p;
1492 struct swap_cluster_info *ci;
1493 swp_entry_t entry;
1494 unsigned long offset;
1495
1496 entry.val = page_private(page);
1497 p = _swap_info_get(entry);
1498 if (p) {
1499 offset = swp_offset(entry);
1500 ci = lock_cluster_or_swap_info(p, offset);
1501 count = swap_count(p->swap_map[offset]);
1502 unlock_cluster_or_swap_info(p, ci);
1503 }
1504 return count;
1505 }
1506
__swap_count(swp_entry_t entry)1507 int __swap_count(swp_entry_t entry)
1508 {
1509 struct swap_info_struct *si;
1510 pgoff_t offset = swp_offset(entry);
1511 int count = 0;
1512
1513 si = get_swap_device(entry);
1514 if (si) {
1515 count = swap_count(si->swap_map[offset]);
1516 put_swap_device(si);
1517 }
1518 return count;
1519 }
1520
swap_swapcount(struct swap_info_struct * si,swp_entry_t entry)1521 static int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry)
1522 {
1523 int count = 0;
1524 pgoff_t offset = swp_offset(entry);
1525 struct swap_cluster_info *ci;
1526
1527 ci = lock_cluster_or_swap_info(si, offset);
1528 count = swap_count(si->swap_map[offset]);
1529 unlock_cluster_or_swap_info(si, ci);
1530 return count;
1531 }
1532
1533 /*
1534 * How many references to @entry are currently swapped out?
1535 * This does not give an exact answer when swap count is continued,
1536 * but does include the high COUNT_CONTINUED flag to allow for that.
1537 */
__swp_swapcount(swp_entry_t entry)1538 int __swp_swapcount(swp_entry_t entry)
1539 {
1540 int count = 0;
1541 struct swap_info_struct *si;
1542
1543 si = get_swap_device(entry);
1544 if (si) {
1545 count = swap_swapcount(si, entry);
1546 put_swap_device(si);
1547 }
1548 return count;
1549 }
1550
1551 /*
1552 * How many references to @entry are currently swapped out?
1553 * This considers COUNT_CONTINUED so it returns exact answer.
1554 */
swp_swapcount(swp_entry_t entry)1555 int swp_swapcount(swp_entry_t entry)
1556 {
1557 int count, tmp_count, n;
1558 struct swap_info_struct *p;
1559 struct swap_cluster_info *ci;
1560 struct page *page;
1561 pgoff_t offset;
1562 unsigned char *map;
1563
1564 p = _swap_info_get(entry);
1565 if (!p)
1566 return 0;
1567
1568 offset = swp_offset(entry);
1569
1570 ci = lock_cluster_or_swap_info(p, offset);
1571
1572 count = swap_count(p->swap_map[offset]);
1573 if (!(count & COUNT_CONTINUED))
1574 goto out;
1575
1576 count &= ~COUNT_CONTINUED;
1577 n = SWAP_MAP_MAX + 1;
1578
1579 page = vmalloc_to_page(p->swap_map + offset);
1580 offset &= ~PAGE_MASK;
1581 VM_BUG_ON(page_private(page) != SWP_CONTINUED);
1582
1583 do {
1584 page = list_next_entry(page, lru);
1585 map = kmap_atomic(page);
1586 tmp_count = map[offset];
1587 kunmap_atomic(map);
1588
1589 count += (tmp_count & ~COUNT_CONTINUED) * n;
1590 n *= (SWAP_CONT_MAX + 1);
1591 } while (tmp_count & COUNT_CONTINUED);
1592 out:
1593 unlock_cluster_or_swap_info(p, ci);
1594 return count;
1595 }
1596
swap_page_trans_huge_swapped(struct swap_info_struct * si,swp_entry_t entry)1597 static bool swap_page_trans_huge_swapped(struct swap_info_struct *si,
1598 swp_entry_t entry)
1599 {
1600 struct swap_cluster_info *ci;
1601 unsigned char *map = si->swap_map;
1602 unsigned long roffset = swp_offset(entry);
1603 unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER);
1604 int i;
1605 bool ret = false;
1606
1607 ci = lock_cluster_or_swap_info(si, offset);
1608 if (!ci || !cluster_is_huge(ci)) {
1609 if (swap_count(map[roffset]))
1610 ret = true;
1611 goto unlock_out;
1612 }
1613 for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1614 if (swap_count(map[offset + i])) {
1615 ret = true;
1616 break;
1617 }
1618 }
1619 unlock_out:
1620 unlock_cluster_or_swap_info(si, ci);
1621 return ret;
1622 }
1623
page_swapped(struct page * page)1624 static bool page_swapped(struct page *page)
1625 {
1626 swp_entry_t entry;
1627 struct swap_info_struct *si;
1628
1629 if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page)))
1630 return page_swapcount(page) != 0;
1631
1632 page = compound_head(page);
1633 entry.val = page_private(page);
1634 si = _swap_info_get(entry);
1635 if (si)
1636 return swap_page_trans_huge_swapped(si, entry);
1637 return false;
1638 }
1639
page_trans_huge_map_swapcount(struct page * page,int * total_mapcount,int * total_swapcount)1640 static int page_trans_huge_map_swapcount(struct page *page, int *total_mapcount,
1641 int *total_swapcount)
1642 {
1643 int i, map_swapcount, _total_mapcount, _total_swapcount;
1644 unsigned long offset = 0;
1645 struct swap_info_struct *si;
1646 struct swap_cluster_info *ci = NULL;
1647 unsigned char *map = NULL;
1648 int mapcount, swapcount = 0;
1649
1650 /* hugetlbfs shouldn't call it */
1651 VM_BUG_ON_PAGE(PageHuge(page), page);
1652
1653 if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page))) {
1654 mapcount = page_trans_huge_mapcount(page, total_mapcount);
1655 if (PageSwapCache(page))
1656 swapcount = page_swapcount(page);
1657 if (total_swapcount)
1658 *total_swapcount = swapcount;
1659 return mapcount + swapcount;
1660 }
1661
1662 page = compound_head(page);
1663
1664 _total_mapcount = _total_swapcount = map_swapcount = 0;
1665 if (PageSwapCache(page)) {
1666 swp_entry_t entry;
1667
1668 entry.val = page_private(page);
1669 si = _swap_info_get(entry);
1670 if (si) {
1671 map = si->swap_map;
1672 offset = swp_offset(entry);
1673 }
1674 }
1675 if (map)
1676 ci = lock_cluster(si, offset);
1677 for (i = 0; i < HPAGE_PMD_NR; i++) {
1678 mapcount = atomic_read(&page[i]._mapcount) + 1;
1679 _total_mapcount += mapcount;
1680 if (map) {
1681 swapcount = swap_count(map[offset + i]);
1682 _total_swapcount += swapcount;
1683 }
1684 map_swapcount = max(map_swapcount, mapcount + swapcount);
1685 }
1686 unlock_cluster(ci);
1687 if (PageDoubleMap(page)) {
1688 map_swapcount -= 1;
1689 _total_mapcount -= HPAGE_PMD_NR;
1690 }
1691 mapcount = compound_mapcount(page);
1692 map_swapcount += mapcount;
1693 _total_mapcount += mapcount;
1694 if (total_mapcount)
1695 *total_mapcount = _total_mapcount;
1696 if (total_swapcount)
1697 *total_swapcount = _total_swapcount;
1698
1699 return map_swapcount;
1700 }
1701
1702 /*
1703 * We can write to an anon page without COW if there are no other references
1704 * to it. And as a side-effect, free up its swap: because the old content
1705 * on disk will never be read, and seeking back there to write new content
1706 * later would only waste time away from clustering.
1707 *
1708 * NOTE: total_map_swapcount should not be relied upon by the caller if
1709 * reuse_swap_page() returns false, but it may be always overwritten
1710 * (see the other implementation for CONFIG_SWAP=n).
1711 */
reuse_swap_page(struct page * page,int * total_map_swapcount)1712 bool reuse_swap_page(struct page *page, int *total_map_swapcount)
1713 {
1714 int count, total_mapcount, total_swapcount;
1715
1716 VM_BUG_ON_PAGE(!PageLocked(page), page);
1717 if (unlikely(PageKsm(page)))
1718 return false;
1719 count = page_trans_huge_map_swapcount(page, &total_mapcount,
1720 &total_swapcount);
1721 if (total_map_swapcount)
1722 *total_map_swapcount = total_mapcount + total_swapcount;
1723 if (count == 1 && PageSwapCache(page) &&
1724 (likely(!PageTransCompound(page)) ||
1725 /* The remaining swap count will be freed soon */
1726 total_swapcount == page_swapcount(page))) {
1727 if (!PageWriteback(page)) {
1728 page = compound_head(page);
1729 delete_from_swap_cache(page);
1730 SetPageDirty(page);
1731 } else {
1732 swp_entry_t entry;
1733 struct swap_info_struct *p;
1734
1735 entry.val = page_private(page);
1736 p = swap_info_get(entry);
1737 if (p->flags & SWP_STABLE_WRITES) {
1738 spin_unlock(&p->lock);
1739 return false;
1740 }
1741 spin_unlock(&p->lock);
1742 }
1743 }
1744
1745 return count <= 1;
1746 }
1747
1748 /*
1749 * If swap is getting full, or if there are no more mappings of this page,
1750 * then try_to_free_swap is called to free its swap space.
1751 */
try_to_free_swap(struct page * page)1752 int try_to_free_swap(struct page *page)
1753 {
1754 VM_BUG_ON_PAGE(!PageLocked(page), page);
1755
1756 if (!PageSwapCache(page))
1757 return 0;
1758 if (PageWriteback(page))
1759 return 0;
1760 if (page_swapped(page))
1761 return 0;
1762
1763 /*
1764 * Once hibernation has begun to create its image of memory,
1765 * there's a danger that one of the calls to try_to_free_swap()
1766 * - most probably a call from __try_to_reclaim_swap() while
1767 * hibernation is allocating its own swap pages for the image,
1768 * but conceivably even a call from memory reclaim - will free
1769 * the swap from a page which has already been recorded in the
1770 * image as a clean swapcache page, and then reuse its swap for
1771 * another page of the image. On waking from hibernation, the
1772 * original page might be freed under memory pressure, then
1773 * later read back in from swap, now with the wrong data.
1774 *
1775 * Hibernation suspends storage while it is writing the image
1776 * to disk so check that here.
1777 */
1778 if (pm_suspended_storage())
1779 return 0;
1780
1781 page = compound_head(page);
1782 delete_from_swap_cache(page);
1783 SetPageDirty(page);
1784 return 1;
1785 }
1786
1787 /*
1788 * Free the swap entry like above, but also try to
1789 * free the page cache entry if it is the last user.
1790 */
free_swap_and_cache(swp_entry_t entry)1791 int free_swap_and_cache(swp_entry_t entry)
1792 {
1793 struct swap_info_struct *p;
1794 unsigned char count;
1795
1796 if (non_swap_entry(entry))
1797 return 1;
1798
1799 p = _swap_info_get(entry);
1800 if (p) {
1801 count = __swap_entry_free(p, entry);
1802 if (count == SWAP_HAS_CACHE &&
1803 !swap_page_trans_huge_swapped(p, entry))
1804 __try_to_reclaim_swap(p, swp_offset(entry),
1805 TTRS_UNMAPPED | TTRS_FULL);
1806 }
1807 return p != NULL;
1808 }
1809
1810 #ifdef CONFIG_HIBERNATION
1811 /*
1812 * Find the swap type that corresponds to given device (if any).
1813 *
1814 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1815 * from 0, in which the swap header is expected to be located.
1816 *
1817 * This is needed for the suspend to disk (aka swsusp).
1818 */
swap_type_of(dev_t device,sector_t offset)1819 int swap_type_of(dev_t device, sector_t offset)
1820 {
1821 int type;
1822
1823 if (!device)
1824 return -1;
1825
1826 spin_lock(&swap_lock);
1827 for (type = 0; type < nr_swapfiles; type++) {
1828 struct swap_info_struct *sis = swap_info[type];
1829
1830 if (!(sis->flags & SWP_WRITEOK))
1831 continue;
1832
1833 if (device == sis->bdev->bd_dev) {
1834 struct swap_extent *se = first_se(sis);
1835
1836 if (se->start_block == offset) {
1837 spin_unlock(&swap_lock);
1838 return type;
1839 }
1840 }
1841 }
1842 spin_unlock(&swap_lock);
1843 return -ENODEV;
1844 }
1845
find_first_swap(dev_t * device)1846 int find_first_swap(dev_t *device)
1847 {
1848 int type;
1849
1850 spin_lock(&swap_lock);
1851 for (type = 0; type < nr_swapfiles; type++) {
1852 struct swap_info_struct *sis = swap_info[type];
1853
1854 if (!(sis->flags & SWP_WRITEOK))
1855 continue;
1856 *device = sis->bdev->bd_dev;
1857 spin_unlock(&swap_lock);
1858 return type;
1859 }
1860 spin_unlock(&swap_lock);
1861 return -ENODEV;
1862 }
1863
1864 /*
1865 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1866 * corresponding to given index in swap_info (swap type).
1867 */
swapdev_block(int type,pgoff_t offset)1868 sector_t swapdev_block(int type, pgoff_t offset)
1869 {
1870 struct block_device *bdev;
1871 struct swap_info_struct *si = swap_type_to_swap_info(type);
1872
1873 if (!si || !(si->flags & SWP_WRITEOK))
1874 return 0;
1875 return map_swap_entry(swp_entry(type, offset), &bdev);
1876 }
1877
1878 /*
1879 * Return either the total number of swap pages of given type, or the number
1880 * of free pages of that type (depending on @free)
1881 *
1882 * This is needed for software suspend
1883 */
count_swap_pages(int type,int free)1884 unsigned int count_swap_pages(int type, int free)
1885 {
1886 unsigned int n = 0;
1887
1888 spin_lock(&swap_lock);
1889 if ((unsigned int)type < nr_swapfiles) {
1890 struct swap_info_struct *sis = swap_info[type];
1891
1892 spin_lock(&sis->lock);
1893 if (sis->flags & SWP_WRITEOK) {
1894 n = sis->pages;
1895 if (free)
1896 n -= sis->inuse_pages;
1897 }
1898 spin_unlock(&sis->lock);
1899 }
1900 spin_unlock(&swap_lock);
1901 return n;
1902 }
1903 #endif /* CONFIG_HIBERNATION */
1904
pte_same_as_swp(pte_t pte,pte_t swp_pte)1905 static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte)
1906 {
1907 return pte_same(pte_swp_clear_flags(pte), swp_pte);
1908 }
1909
1910 /*
1911 * No need to decide whether this PTE shares the swap entry with others,
1912 * just let do_wp_page work it out if a write is requested later - to
1913 * force COW, vm_page_prot omits write permission from any private vma.
1914 */
unuse_pte(struct vm_area_struct * vma,pmd_t * pmd,unsigned long addr,swp_entry_t entry,struct page * page)1915 static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
1916 unsigned long addr, swp_entry_t entry, struct page *page)
1917 {
1918 struct page *swapcache;
1919 spinlock_t *ptl;
1920 pte_t *pte;
1921 int ret = 1;
1922
1923 swapcache = page;
1924 page = ksm_might_need_to_copy(page, vma, addr);
1925 if (unlikely(!page))
1926 return -ENOMEM;
1927
1928 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1929 if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) {
1930 ret = 0;
1931 goto out;
1932 }
1933
1934 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1935 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
1936 get_page(page);
1937 set_pte_at(vma->vm_mm, addr, pte,
1938 pte_mkold(mk_pte(page, vma->vm_page_prot)));
1939 if (page == swapcache) {
1940 page_add_anon_rmap(page, vma, addr, false);
1941 } else { /* ksm created a completely new copy */
1942 page_add_new_anon_rmap(page, vma, addr, false);
1943 lru_cache_add_inactive_or_unevictable(page, vma);
1944 }
1945 swap_free(entry);
1946 out:
1947 pte_unmap_unlock(pte, ptl);
1948 if (page != swapcache) {
1949 unlock_page(page);
1950 put_page(page);
1951 }
1952 return ret;
1953 }
1954
unuse_pte_range(struct vm_area_struct * vma,pmd_t * pmd,unsigned long addr,unsigned long end,unsigned int type,bool frontswap,unsigned long * fs_pages_to_unuse)1955 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1956 unsigned long addr, unsigned long end,
1957 unsigned int type, bool frontswap,
1958 unsigned long *fs_pages_to_unuse)
1959 {
1960 struct page *page;
1961 swp_entry_t entry;
1962 pte_t *pte;
1963 struct swap_info_struct *si;
1964 unsigned long offset;
1965 int ret = 0;
1966 volatile unsigned char *swap_map;
1967
1968 si = swap_info[type];
1969 pte = pte_offset_map(pmd, addr);
1970 do {
1971 struct vm_fault vmf;
1972
1973 if (!is_swap_pte(*pte))
1974 continue;
1975
1976 entry = pte_to_swp_entry(*pte);
1977 if (swp_type(entry) != type)
1978 continue;
1979
1980 offset = swp_offset(entry);
1981 if (frontswap && !frontswap_test(si, offset))
1982 continue;
1983
1984 pte_unmap(pte);
1985 swap_map = &si->swap_map[offset];
1986 page = lookup_swap_cache(entry, vma, addr);
1987 if (!page) {
1988 vmf.vma = vma;
1989 vmf.address = addr;
1990 vmf.pmd = pmd;
1991 page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE,
1992 &vmf);
1993 }
1994 if (!page) {
1995 if (*swap_map == 0 || *swap_map == SWAP_MAP_BAD)
1996 goto try_next;
1997 return -ENOMEM;
1998 }
1999
2000 lock_page(page);
2001 wait_on_page_writeback(page);
2002 ret = unuse_pte(vma, pmd, addr, entry, page);
2003 if (ret < 0) {
2004 unlock_page(page);
2005 put_page(page);
2006 goto out;
2007 }
2008
2009 try_to_free_swap(page);
2010 unlock_page(page);
2011 put_page(page);
2012
2013 if (*fs_pages_to_unuse && !--(*fs_pages_to_unuse)) {
2014 ret = FRONTSWAP_PAGES_UNUSED;
2015 goto out;
2016 }
2017 try_next:
2018 pte = pte_offset_map(pmd, addr);
2019 } while (pte++, addr += PAGE_SIZE, addr != end);
2020 pte_unmap(pte - 1);
2021
2022 ret = 0;
2023 out:
2024 return ret;
2025 }
2026
unuse_pmd_range(struct vm_area_struct * vma,pud_t * pud,unsigned long addr,unsigned long end,unsigned int type,bool frontswap,unsigned long * fs_pages_to_unuse)2027 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
2028 unsigned long addr, unsigned long end,
2029 unsigned int type, bool frontswap,
2030 unsigned long *fs_pages_to_unuse)
2031 {
2032 pmd_t *pmd;
2033 unsigned long next;
2034 int ret;
2035
2036 pmd = pmd_offset(pud, addr);
2037 do {
2038 cond_resched();
2039 next = pmd_addr_end(addr, end);
2040 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
2041 continue;
2042 ret = unuse_pte_range(vma, pmd, addr, next, type,
2043 frontswap, fs_pages_to_unuse);
2044 if (ret)
2045 return ret;
2046 } while (pmd++, addr = next, addr != end);
2047 return 0;
2048 }
2049
unuse_pud_range(struct vm_area_struct * vma,p4d_t * p4d,unsigned long addr,unsigned long end,unsigned int type,bool frontswap,unsigned long * fs_pages_to_unuse)2050 static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d,
2051 unsigned long addr, unsigned long end,
2052 unsigned int type, bool frontswap,
2053 unsigned long *fs_pages_to_unuse)
2054 {
2055 pud_t *pud;
2056 unsigned long next;
2057 int ret;
2058
2059 pud = pud_offset(p4d, addr);
2060 do {
2061 next = pud_addr_end(addr, end);
2062 if (pud_none_or_clear_bad(pud))
2063 continue;
2064 ret = unuse_pmd_range(vma, pud, addr, next, type,
2065 frontswap, fs_pages_to_unuse);
2066 if (ret)
2067 return ret;
2068 } while (pud++, addr = next, addr != end);
2069 return 0;
2070 }
2071
unuse_p4d_range(struct vm_area_struct * vma,pgd_t * pgd,unsigned long addr,unsigned long end,unsigned int type,bool frontswap,unsigned long * fs_pages_to_unuse)2072 static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd,
2073 unsigned long addr, unsigned long end,
2074 unsigned int type, bool frontswap,
2075 unsigned long *fs_pages_to_unuse)
2076 {
2077 p4d_t *p4d;
2078 unsigned long next;
2079 int ret;
2080
2081 p4d = p4d_offset(pgd, addr);
2082 do {
2083 next = p4d_addr_end(addr, end);
2084 if (p4d_none_or_clear_bad(p4d))
2085 continue;
2086 ret = unuse_pud_range(vma, p4d, addr, next, type,
2087 frontswap, fs_pages_to_unuse);
2088 if (ret)
2089 return ret;
2090 } while (p4d++, addr = next, addr != end);
2091 return 0;
2092 }
2093
unuse_vma(struct vm_area_struct * vma,unsigned int type,bool frontswap,unsigned long * fs_pages_to_unuse)2094 static int unuse_vma(struct vm_area_struct *vma, unsigned int type,
2095 bool frontswap, unsigned long *fs_pages_to_unuse)
2096 {
2097 pgd_t *pgd;
2098 unsigned long addr, end, next;
2099 int ret;
2100
2101 addr = vma->vm_start;
2102 end = vma->vm_end;
2103
2104 pgd = pgd_offset(vma->vm_mm, addr);
2105 do {
2106 next = pgd_addr_end(addr, end);
2107 if (pgd_none_or_clear_bad(pgd))
2108 continue;
2109 ret = unuse_p4d_range(vma, pgd, addr, next, type,
2110 frontswap, fs_pages_to_unuse);
2111 if (ret)
2112 return ret;
2113 } while (pgd++, addr = next, addr != end);
2114 return 0;
2115 }
2116
unuse_mm(struct mm_struct * mm,unsigned int type,bool frontswap,unsigned long * fs_pages_to_unuse)2117 static int unuse_mm(struct mm_struct *mm, unsigned int type,
2118 bool frontswap, unsigned long *fs_pages_to_unuse)
2119 {
2120 struct vm_area_struct *vma;
2121 int ret = 0;
2122
2123 mmap_read_lock(mm);
2124 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2125 if (vma->anon_vma) {
2126 ret = unuse_vma(vma, type, frontswap,
2127 fs_pages_to_unuse);
2128 if (ret)
2129 break;
2130 }
2131 cond_resched();
2132 }
2133 mmap_read_unlock(mm);
2134 return ret;
2135 }
2136
2137 /*
2138 * Scan swap_map (or frontswap_map if frontswap parameter is true)
2139 * from current position to next entry still in use. Return 0
2140 * if there are no inuse entries after prev till end of the map.
2141 */
find_next_to_unuse(struct swap_info_struct * si,unsigned int prev,bool frontswap)2142 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
2143 unsigned int prev, bool frontswap)
2144 {
2145 unsigned int i;
2146 unsigned char count;
2147
2148 /*
2149 * No need for swap_lock here: we're just looking
2150 * for whether an entry is in use, not modifying it; false
2151 * hits are okay, and sys_swapoff() has already prevented new
2152 * allocations from this area (while holding swap_lock).
2153 */
2154 for (i = prev + 1; i < si->max; i++) {
2155 count = READ_ONCE(si->swap_map[i]);
2156 if (count && swap_count(count) != SWAP_MAP_BAD)
2157 if (!frontswap || frontswap_test(si, i))
2158 break;
2159 if ((i % LATENCY_LIMIT) == 0)
2160 cond_resched();
2161 }
2162
2163 if (i == si->max)
2164 i = 0;
2165
2166 return i;
2167 }
2168
2169 /*
2170 * If the boolean frontswap is true, only unuse pages_to_unuse pages;
2171 * pages_to_unuse==0 means all pages; ignored if frontswap is false
2172 */
try_to_unuse(unsigned int type,bool frontswap,unsigned long pages_to_unuse)2173 int try_to_unuse(unsigned int type, bool frontswap,
2174 unsigned long pages_to_unuse)
2175 {
2176 struct mm_struct *prev_mm;
2177 struct mm_struct *mm;
2178 struct list_head *p;
2179 int retval = 0;
2180 struct swap_info_struct *si = swap_info[type];
2181 struct page *page;
2182 swp_entry_t entry;
2183 unsigned int i;
2184
2185 if (!READ_ONCE(si->inuse_pages))
2186 return 0;
2187
2188 if (!frontswap)
2189 pages_to_unuse = 0;
2190
2191 retry:
2192 retval = shmem_unuse(type, frontswap, &pages_to_unuse);
2193 if (retval)
2194 goto out;
2195
2196 prev_mm = &init_mm;
2197 mmget(prev_mm);
2198
2199 spin_lock(&mmlist_lock);
2200 p = &init_mm.mmlist;
2201 while (READ_ONCE(si->inuse_pages) &&
2202 !signal_pending(current) &&
2203 (p = p->next) != &init_mm.mmlist) {
2204
2205 mm = list_entry(p, struct mm_struct, mmlist);
2206 if (!mmget_not_zero(mm))
2207 continue;
2208 spin_unlock(&mmlist_lock);
2209 mmput(prev_mm);
2210 prev_mm = mm;
2211 retval = unuse_mm(mm, type, frontswap, &pages_to_unuse);
2212
2213 if (retval) {
2214 mmput(prev_mm);
2215 goto out;
2216 }
2217
2218 /*
2219 * Make sure that we aren't completely killing
2220 * interactive performance.
2221 */
2222 cond_resched();
2223 spin_lock(&mmlist_lock);
2224 }
2225 spin_unlock(&mmlist_lock);
2226
2227 mmput(prev_mm);
2228
2229 i = 0;
2230 while (READ_ONCE(si->inuse_pages) &&
2231 !signal_pending(current) &&
2232 (i = find_next_to_unuse(si, i, frontswap)) != 0) {
2233
2234 entry = swp_entry(type, i);
2235 page = find_get_page(swap_address_space(entry), i);
2236 if (!page)
2237 continue;
2238
2239 /*
2240 * It is conceivable that a racing task removed this page from
2241 * swap cache just before we acquired the page lock. The page
2242 * might even be back in swap cache on another swap area. But
2243 * that is okay, try_to_free_swap() only removes stale pages.
2244 */
2245 lock_page(page);
2246 wait_on_page_writeback(page);
2247 try_to_free_swap(page);
2248 unlock_page(page);
2249 put_page(page);
2250
2251 /*
2252 * For frontswap, we just need to unuse pages_to_unuse, if
2253 * it was specified. Need not check frontswap again here as
2254 * we already zeroed out pages_to_unuse if not frontswap.
2255 */
2256 if (pages_to_unuse && --pages_to_unuse == 0)
2257 goto out;
2258 }
2259
2260 /*
2261 * Lets check again to see if there are still swap entries in the map.
2262 * If yes, we would need to do retry the unuse logic again.
2263 * Under global memory pressure, swap entries can be reinserted back
2264 * into process space after the mmlist loop above passes over them.
2265 *
2266 * Limit the number of retries? No: when mmget_not_zero() above fails,
2267 * that mm is likely to be freeing swap from exit_mmap(), which proceeds
2268 * at its own independent pace; and even shmem_writepage() could have
2269 * been preempted after get_swap_page(), temporarily hiding that swap.
2270 * It's easy and robust (though cpu-intensive) just to keep retrying.
2271 */
2272 if (READ_ONCE(si->inuse_pages)) {
2273 if (!signal_pending(current))
2274 goto retry;
2275 retval = -EINTR;
2276 }
2277 out:
2278 return (retval == FRONTSWAP_PAGES_UNUSED) ? 0 : retval;
2279 }
2280
2281 /*
2282 * After a successful try_to_unuse, if no swap is now in use, we know
2283 * we can empty the mmlist. swap_lock must be held on entry and exit.
2284 * Note that mmlist_lock nests inside swap_lock, and an mm must be
2285 * added to the mmlist just after page_duplicate - before would be racy.
2286 */
drain_mmlist(void)2287 static void drain_mmlist(void)
2288 {
2289 struct list_head *p, *next;
2290 unsigned int type;
2291
2292 for (type = 0; type < nr_swapfiles; type++)
2293 if (swap_info[type]->inuse_pages)
2294 return;
2295 spin_lock(&mmlist_lock);
2296 list_for_each_safe(p, next, &init_mm.mmlist)
2297 list_del_init(p);
2298 spin_unlock(&mmlist_lock);
2299 }
2300
2301 /*
2302 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
2303 * corresponds to page offset for the specified swap entry.
2304 * Note that the type of this function is sector_t, but it returns page offset
2305 * into the bdev, not sector offset.
2306 */
map_swap_entry(swp_entry_t entry,struct block_device ** bdev)2307 static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
2308 {
2309 struct swap_info_struct *sis;
2310 struct swap_extent *se;
2311 pgoff_t offset;
2312
2313 sis = swp_swap_info(entry);
2314 *bdev = sis->bdev;
2315
2316 offset = swp_offset(entry);
2317 se = offset_to_swap_extent(sis, offset);
2318 return se->start_block + (offset - se->start_page);
2319 }
2320
2321 /*
2322 * Returns the page offset into bdev for the specified page's swap entry.
2323 */
map_swap_page(struct page * page,struct block_device ** bdev)2324 sector_t map_swap_page(struct page *page, struct block_device **bdev)
2325 {
2326 swp_entry_t entry;
2327 entry.val = page_private(page);
2328 return map_swap_entry(entry, bdev);
2329 }
2330
2331 /*
2332 * Free all of a swapdev's extent information
2333 */
destroy_swap_extents(struct swap_info_struct * sis)2334 static void destroy_swap_extents(struct swap_info_struct *sis)
2335 {
2336 while (!RB_EMPTY_ROOT(&sis->swap_extent_root)) {
2337 struct rb_node *rb = sis->swap_extent_root.rb_node;
2338 struct swap_extent *se = rb_entry(rb, struct swap_extent, rb_node);
2339
2340 rb_erase(rb, &sis->swap_extent_root);
2341 kfree(se);
2342 }
2343
2344 if (sis->flags & SWP_ACTIVATED) {
2345 struct file *swap_file = sis->swap_file;
2346 struct address_space *mapping = swap_file->f_mapping;
2347
2348 sis->flags &= ~SWP_ACTIVATED;
2349 if (mapping->a_ops->swap_deactivate)
2350 mapping->a_ops->swap_deactivate(swap_file);
2351 }
2352 }
2353
2354 /*
2355 * Add a block range (and the corresponding page range) into this swapdev's
2356 * extent tree.
2357 *
2358 * This function rather assumes that it is called in ascending page order.
2359 */
2360 int
add_swap_extent(struct swap_info_struct * sis,unsigned long start_page,unsigned long nr_pages,sector_t start_block)2361 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
2362 unsigned long nr_pages, sector_t start_block)
2363 {
2364 struct rb_node **link = &sis->swap_extent_root.rb_node, *parent = NULL;
2365 struct swap_extent *se;
2366 struct swap_extent *new_se;
2367
2368 /*
2369 * place the new node at the right most since the
2370 * function is called in ascending page order.
2371 */
2372 while (*link) {
2373 parent = *link;
2374 link = &parent->rb_right;
2375 }
2376
2377 if (parent) {
2378 se = rb_entry(parent, struct swap_extent, rb_node);
2379 BUG_ON(se->start_page + se->nr_pages != start_page);
2380 if (se->start_block + se->nr_pages == start_block) {
2381 /* Merge it */
2382 se->nr_pages += nr_pages;
2383 return 0;
2384 }
2385 }
2386
2387 /* No merge, insert a new extent. */
2388 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
2389 if (new_se == NULL)
2390 return -ENOMEM;
2391 new_se->start_page = start_page;
2392 new_se->nr_pages = nr_pages;
2393 new_se->start_block = start_block;
2394
2395 rb_link_node(&new_se->rb_node, parent, link);
2396 rb_insert_color(&new_se->rb_node, &sis->swap_extent_root);
2397 return 1;
2398 }
2399 EXPORT_SYMBOL_GPL(add_swap_extent);
2400
2401 /*
2402 * A `swap extent' is a simple thing which maps a contiguous range of pages
2403 * onto a contiguous range of disk blocks. An ordered list of swap extents
2404 * is built at swapon time and is then used at swap_writepage/swap_readpage
2405 * time for locating where on disk a page belongs.
2406 *
2407 * If the swapfile is an S_ISBLK block device, a single extent is installed.
2408 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2409 * swap files identically.
2410 *
2411 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2412 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
2413 * swapfiles are handled *identically* after swapon time.
2414 *
2415 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2416 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
2417 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
2418 * requirements, they are simply tossed out - we will never use those blocks
2419 * for swapping.
2420 *
2421 * For all swap devices we set S_SWAPFILE across the life of the swapon. This
2422 * prevents users from writing to the swap device, which will corrupt memory.
2423 *
2424 * The amount of disk space which a single swap extent represents varies.
2425 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
2426 * extents in the list. To avoid much list walking, we cache the previous
2427 * search location in `curr_swap_extent', and start new searches from there.
2428 * This is extremely effective. The average number of iterations in
2429 * map_swap_page() has been measured at about 0.3 per page. - akpm.
2430 */
setup_swap_extents(struct swap_info_struct * sis,sector_t * span)2431 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
2432 {
2433 struct file *swap_file = sis->swap_file;
2434 struct address_space *mapping = swap_file->f_mapping;
2435 struct inode *inode = mapping->host;
2436 int ret;
2437
2438 if (S_ISBLK(inode->i_mode)) {
2439 ret = add_swap_extent(sis, 0, sis->max, 0);
2440 *span = sis->pages;
2441 return ret;
2442 }
2443
2444 if (mapping->a_ops->swap_activate) {
2445 ret = mapping->a_ops->swap_activate(sis, swap_file, span);
2446 if (ret >= 0)
2447 sis->flags |= SWP_ACTIVATED;
2448 if (!ret) {
2449 sis->flags |= SWP_FS_OPS;
2450 ret = add_swap_extent(sis, 0, sis->max, 0);
2451 *span = sis->pages;
2452 }
2453 return ret;
2454 }
2455
2456 return generic_swapfile_activate(sis, swap_file, span);
2457 }
2458
swap_node(struct swap_info_struct * p)2459 static int swap_node(struct swap_info_struct *p)
2460 {
2461 struct block_device *bdev;
2462
2463 if (p->bdev)
2464 bdev = p->bdev;
2465 else
2466 bdev = p->swap_file->f_inode->i_sb->s_bdev;
2467
2468 return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE;
2469 }
2470
setup_swap_info(struct swap_info_struct * p,int prio,unsigned char * swap_map,struct swap_cluster_info * cluster_info)2471 static void setup_swap_info(struct swap_info_struct *p, int prio,
2472 unsigned char *swap_map,
2473 struct swap_cluster_info *cluster_info)
2474 {
2475 int i;
2476
2477 if (prio >= 0)
2478 p->prio = prio;
2479 else
2480 p->prio = --least_priority;
2481 /*
2482 * the plist prio is negated because plist ordering is
2483 * low-to-high, while swap ordering is high-to-low
2484 */
2485 p->list.prio = -p->prio;
2486 for_each_node(i) {
2487 if (p->prio >= 0)
2488 p->avail_lists[i].prio = -p->prio;
2489 else {
2490 if (swap_node(p) == i)
2491 p->avail_lists[i].prio = 1;
2492 else
2493 p->avail_lists[i].prio = -p->prio;
2494 }
2495 }
2496 p->swap_map = swap_map;
2497 p->cluster_info = cluster_info;
2498 }
2499
_enable_swap_info(struct swap_info_struct * p)2500 static void _enable_swap_info(struct swap_info_struct *p)
2501 {
2502 p->flags |= SWP_WRITEOK | SWP_VALID;
2503 atomic_long_add(p->pages, &nr_swap_pages);
2504 total_swap_pages += p->pages;
2505
2506 assert_spin_locked(&swap_lock);
2507 /*
2508 * both lists are plists, and thus priority ordered.
2509 * swap_active_head needs to be priority ordered for swapoff(),
2510 * which on removal of any swap_info_struct with an auto-assigned
2511 * (i.e. negative) priority increments the auto-assigned priority
2512 * of any lower-priority swap_info_structs.
2513 * swap_avail_head needs to be priority ordered for get_swap_page(),
2514 * which allocates swap pages from the highest available priority
2515 * swap_info_struct.
2516 */
2517 plist_add(&p->list, &swap_active_head);
2518 add_to_avail_list(p);
2519 }
2520
enable_swap_info(struct swap_info_struct * p,int prio,unsigned char * swap_map,struct swap_cluster_info * cluster_info,unsigned long * frontswap_map)2521 static void enable_swap_info(struct swap_info_struct *p, int prio,
2522 unsigned char *swap_map,
2523 struct swap_cluster_info *cluster_info,
2524 unsigned long *frontswap_map)
2525 {
2526 frontswap_init(p->type, frontswap_map);
2527 spin_lock(&swap_lock);
2528 spin_lock(&p->lock);
2529 setup_swap_info(p, prio, swap_map, cluster_info);
2530 spin_unlock(&p->lock);
2531 spin_unlock(&swap_lock);
2532 /*
2533 * Guarantee swap_map, cluster_info, etc. fields are valid
2534 * between get/put_swap_device() if SWP_VALID bit is set
2535 */
2536 synchronize_rcu();
2537 spin_lock(&swap_lock);
2538 spin_lock(&p->lock);
2539 _enable_swap_info(p);
2540 spin_unlock(&p->lock);
2541 spin_unlock(&swap_lock);
2542 }
2543
reinsert_swap_info(struct swap_info_struct * p)2544 static void reinsert_swap_info(struct swap_info_struct *p)
2545 {
2546 spin_lock(&swap_lock);
2547 spin_lock(&p->lock);
2548 setup_swap_info(p, p->prio, p->swap_map, p->cluster_info);
2549 _enable_swap_info(p);
2550 spin_unlock(&p->lock);
2551 spin_unlock(&swap_lock);
2552 }
2553
has_usable_swap(void)2554 bool has_usable_swap(void)
2555 {
2556 bool ret = true;
2557
2558 spin_lock(&swap_lock);
2559 if (plist_head_empty(&swap_active_head))
2560 ret = false;
2561 spin_unlock(&swap_lock);
2562 return ret;
2563 }
2564
SYSCALL_DEFINE1(swapoff,const char __user *,specialfile)2565 SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
2566 {
2567 struct swap_info_struct *p = NULL;
2568 unsigned char *swap_map;
2569 struct swap_cluster_info *cluster_info;
2570 unsigned long *frontswap_map;
2571 struct file *swap_file, *victim;
2572 struct address_space *mapping;
2573 struct inode *inode;
2574 struct filename *pathname;
2575 int err, found = 0;
2576 unsigned int old_block_size;
2577
2578 if (!capable(CAP_SYS_ADMIN))
2579 return -EPERM;
2580
2581 BUG_ON(!current->mm);
2582
2583 pathname = getname(specialfile);
2584 if (IS_ERR(pathname))
2585 return PTR_ERR(pathname);
2586
2587 victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
2588 err = PTR_ERR(victim);
2589 if (IS_ERR(victim))
2590 goto out;
2591
2592 mapping = victim->f_mapping;
2593 spin_lock(&swap_lock);
2594 plist_for_each_entry(p, &swap_active_head, list) {
2595 if (p->flags & SWP_WRITEOK) {
2596 if (p->swap_file->f_mapping == mapping) {
2597 found = 1;
2598 break;
2599 }
2600 }
2601 }
2602 if (!found) {
2603 err = -EINVAL;
2604 spin_unlock(&swap_lock);
2605 goto out_dput;
2606 }
2607 if (!security_vm_enough_memory_mm(current->mm, p->pages))
2608 vm_unacct_memory(p->pages);
2609 else {
2610 err = -ENOMEM;
2611 spin_unlock(&swap_lock);
2612 goto out_dput;
2613 }
2614 del_from_avail_list(p);
2615 spin_lock(&p->lock);
2616 if (p->prio < 0) {
2617 struct swap_info_struct *si = p;
2618 int nid;
2619
2620 plist_for_each_entry_continue(si, &swap_active_head, list) {
2621 si->prio++;
2622 si->list.prio--;
2623 for_each_node(nid) {
2624 if (si->avail_lists[nid].prio != 1)
2625 si->avail_lists[nid].prio--;
2626 }
2627 }
2628 least_priority++;
2629 }
2630 plist_del(&p->list, &swap_active_head);
2631 atomic_long_sub(p->pages, &nr_swap_pages);
2632 total_swap_pages -= p->pages;
2633 p->flags &= ~SWP_WRITEOK;
2634 spin_unlock(&p->lock);
2635 spin_unlock(&swap_lock);
2636
2637 disable_swap_slots_cache_lock();
2638
2639 set_current_oom_origin();
2640 err = try_to_unuse(p->type, false, 0); /* force unuse all pages */
2641 clear_current_oom_origin();
2642
2643 if (err) {
2644 /* re-insert swap space back into swap_list */
2645 reinsert_swap_info(p);
2646 reenable_swap_slots_cache_unlock();
2647 goto out_dput;
2648 }
2649
2650 reenable_swap_slots_cache_unlock();
2651
2652 spin_lock(&swap_lock);
2653 spin_lock(&p->lock);
2654 p->flags &= ~SWP_VALID; /* mark swap device as invalid */
2655 spin_unlock(&p->lock);
2656 spin_unlock(&swap_lock);
2657 /*
2658 * wait for swap operations protected by get/put_swap_device()
2659 * to complete
2660 */
2661 synchronize_rcu();
2662
2663 flush_work(&p->discard_work);
2664
2665 destroy_swap_extents(p);
2666 if (p->flags & SWP_CONTINUED)
2667 free_swap_count_continuations(p);
2668
2669 if (!p->bdev || !blk_queue_nonrot(bdev_get_queue(p->bdev)))
2670 atomic_dec(&nr_rotate_swap);
2671
2672 mutex_lock(&swapon_mutex);
2673 spin_lock(&swap_lock);
2674 spin_lock(&p->lock);
2675 drain_mmlist();
2676
2677 /* wait for anyone still in scan_swap_map */
2678 p->highest_bit = 0; /* cuts scans short */
2679 while (p->flags >= SWP_SCANNING) {
2680 spin_unlock(&p->lock);
2681 spin_unlock(&swap_lock);
2682 schedule_timeout_uninterruptible(1);
2683 spin_lock(&swap_lock);
2684 spin_lock(&p->lock);
2685 }
2686
2687 swap_file = p->swap_file;
2688 old_block_size = p->old_block_size;
2689 p->swap_file = NULL;
2690 p->max = 0;
2691 swap_map = p->swap_map;
2692 p->swap_map = NULL;
2693 cluster_info = p->cluster_info;
2694 p->cluster_info = NULL;
2695 frontswap_map = frontswap_map_get(p);
2696 spin_unlock(&p->lock);
2697 spin_unlock(&swap_lock);
2698 arch_swap_invalidate_area(p->type);
2699 frontswap_invalidate_area(p->type);
2700 frontswap_map_set(p, NULL);
2701 mutex_unlock(&swapon_mutex);
2702 free_percpu(p->percpu_cluster);
2703 p->percpu_cluster = NULL;
2704 free_percpu(p->cluster_next_cpu);
2705 p->cluster_next_cpu = NULL;
2706 vfree(swap_map);
2707 kvfree(cluster_info);
2708 kvfree(frontswap_map);
2709 /* Destroy swap account information */
2710 swap_cgroup_swapoff(p->type);
2711 exit_swap_address_space(p->type);
2712
2713 inode = mapping->host;
2714 if (S_ISBLK(inode->i_mode)) {
2715 struct block_device *bdev = I_BDEV(inode);
2716
2717 set_blocksize(bdev, old_block_size);
2718 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2719 }
2720
2721 inode_lock(inode);
2722 inode->i_flags &= ~S_SWAPFILE;
2723 inode_unlock(inode);
2724 filp_close(swap_file, NULL);
2725
2726 /*
2727 * Clear the SWP_USED flag after all resources are freed so that swapon
2728 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
2729 * not hold p->lock after we cleared its SWP_WRITEOK.
2730 */
2731 spin_lock(&swap_lock);
2732 p->flags = 0;
2733 spin_unlock(&swap_lock);
2734
2735 err = 0;
2736 atomic_inc(&proc_poll_event);
2737 wake_up_interruptible(&proc_poll_wait);
2738
2739 out_dput:
2740 filp_close(victim, NULL);
2741 out:
2742 putname(pathname);
2743 return err;
2744 }
2745
2746 #ifdef CONFIG_PROC_FS
swaps_poll(struct file * file,poll_table * wait)2747 static __poll_t swaps_poll(struct file *file, poll_table *wait)
2748 {
2749 struct seq_file *seq = file->private_data;
2750
2751 poll_wait(file, &proc_poll_wait, wait);
2752
2753 if (seq->poll_event != atomic_read(&proc_poll_event)) {
2754 seq->poll_event = atomic_read(&proc_poll_event);
2755 return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
2756 }
2757
2758 return EPOLLIN | EPOLLRDNORM;
2759 }
2760
2761 /* iterator */
swap_start(struct seq_file * swap,loff_t * pos)2762 static void *swap_start(struct seq_file *swap, loff_t *pos)
2763 {
2764 struct swap_info_struct *si;
2765 int type;
2766 loff_t l = *pos;
2767
2768 mutex_lock(&swapon_mutex);
2769
2770 if (!l)
2771 return SEQ_START_TOKEN;
2772
2773 for (type = 0; (si = swap_type_to_swap_info(type)); type++) {
2774 if (!(si->flags & SWP_USED) || !si->swap_map)
2775 continue;
2776 if (!--l)
2777 return si;
2778 }
2779
2780 return NULL;
2781 }
2782
swap_next(struct seq_file * swap,void * v,loff_t * pos)2783 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2784 {
2785 struct swap_info_struct *si = v;
2786 int type;
2787
2788 if (v == SEQ_START_TOKEN)
2789 type = 0;
2790 else
2791 type = si->type + 1;
2792
2793 ++(*pos);
2794 for (; (si = swap_type_to_swap_info(type)); type++) {
2795 if (!(si->flags & SWP_USED) || !si->swap_map)
2796 continue;
2797 return si;
2798 }
2799
2800 return NULL;
2801 }
2802
swap_stop(struct seq_file * swap,void * v)2803 static void swap_stop(struct seq_file *swap, void *v)
2804 {
2805 mutex_unlock(&swapon_mutex);
2806 }
2807
swap_show(struct seq_file * swap,void * v)2808 static int swap_show(struct seq_file *swap, void *v)
2809 {
2810 struct swap_info_struct *si = v;
2811 struct file *file;
2812 int len;
2813 unsigned int bytes, inuse;
2814
2815 if (si == SEQ_START_TOKEN) {
2816 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\t\tUsed\t\tPriority\n");
2817 return 0;
2818 }
2819
2820 bytes = si->pages << (PAGE_SHIFT - 10);
2821 inuse = si->inuse_pages << (PAGE_SHIFT - 10);
2822
2823 file = si->swap_file;
2824 len = seq_file_path(swap, file, " \t\n\\");
2825 seq_printf(swap, "%*s%s\t%u\t%s%u\t%s%d\n",
2826 len < 40 ? 40 - len : 1, " ",
2827 S_ISBLK(file_inode(file)->i_mode) ?
2828 "partition" : "file\t",
2829 bytes, bytes < 10000000 ? "\t" : "",
2830 inuse, inuse < 10000000 ? "\t" : "",
2831 si->prio);
2832 return 0;
2833 }
2834
2835 static const struct seq_operations swaps_op = {
2836 .start = swap_start,
2837 .next = swap_next,
2838 .stop = swap_stop,
2839 .show = swap_show
2840 };
2841
swaps_open(struct inode * inode,struct file * file)2842 static int swaps_open(struct inode *inode, struct file *file)
2843 {
2844 struct seq_file *seq;
2845 int ret;
2846
2847 ret = seq_open(file, &swaps_op);
2848 if (ret)
2849 return ret;
2850
2851 seq = file->private_data;
2852 seq->poll_event = atomic_read(&proc_poll_event);
2853 return 0;
2854 }
2855
2856 static const struct proc_ops swaps_proc_ops = {
2857 .proc_flags = PROC_ENTRY_PERMANENT,
2858 .proc_open = swaps_open,
2859 .proc_read = seq_read,
2860 .proc_lseek = seq_lseek,
2861 .proc_release = seq_release,
2862 .proc_poll = swaps_poll,
2863 };
2864
procswaps_init(void)2865 static int __init procswaps_init(void)
2866 {
2867 proc_create("swaps", 0, NULL, &swaps_proc_ops);
2868 return 0;
2869 }
2870 __initcall(procswaps_init);
2871 #endif /* CONFIG_PROC_FS */
2872
2873 #ifdef MAX_SWAPFILES_CHECK
max_swapfiles_check(void)2874 static int __init max_swapfiles_check(void)
2875 {
2876 MAX_SWAPFILES_CHECK();
2877 return 0;
2878 }
2879 late_initcall(max_swapfiles_check);
2880 #endif
2881
alloc_swap_info(void)2882 static struct swap_info_struct *alloc_swap_info(void)
2883 {
2884 struct swap_info_struct *p;
2885 struct swap_info_struct *defer = NULL;
2886 unsigned int type;
2887 int i;
2888
2889 p = kvzalloc(struct_size(p, avail_lists, nr_node_ids), GFP_KERNEL);
2890 if (!p)
2891 return ERR_PTR(-ENOMEM);
2892
2893 spin_lock(&swap_lock);
2894 for (type = 0; type < nr_swapfiles; type++) {
2895 if (!(swap_info[type]->flags & SWP_USED))
2896 break;
2897 }
2898 if (type >= MAX_SWAPFILES) {
2899 spin_unlock(&swap_lock);
2900 kvfree(p);
2901 return ERR_PTR(-EPERM);
2902 }
2903 if (type >= nr_swapfiles) {
2904 p->type = type;
2905 WRITE_ONCE(swap_info[type], p);
2906 /*
2907 * Write swap_info[type] before nr_swapfiles, in case a
2908 * racing procfs swap_start() or swap_next() is reading them.
2909 * (We never shrink nr_swapfiles, we never free this entry.)
2910 */
2911 smp_wmb();
2912 WRITE_ONCE(nr_swapfiles, nr_swapfiles + 1);
2913 } else {
2914 defer = p;
2915 p = swap_info[type];
2916 /*
2917 * Do not memset this entry: a racing procfs swap_next()
2918 * would be relying on p->type to remain valid.
2919 */
2920 }
2921 p->swap_extent_root = RB_ROOT;
2922 plist_node_init(&p->list, 0);
2923 for_each_node(i)
2924 plist_node_init(&p->avail_lists[i], 0);
2925 p->flags = SWP_USED;
2926 spin_unlock(&swap_lock);
2927 kvfree(defer);
2928 spin_lock_init(&p->lock);
2929 spin_lock_init(&p->cont_lock);
2930
2931 return p;
2932 }
2933
claim_swapfile(struct swap_info_struct * p,struct inode * inode)2934 static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2935 {
2936 int error;
2937
2938 if (S_ISBLK(inode->i_mode)) {
2939 p->bdev = blkdev_get_by_dev(inode->i_rdev,
2940 FMODE_READ | FMODE_WRITE | FMODE_EXCL, p);
2941 if (IS_ERR(p->bdev)) {
2942 error = PTR_ERR(p->bdev);
2943 p->bdev = NULL;
2944 return error;
2945 }
2946 p->old_block_size = block_size(p->bdev);
2947 error = set_blocksize(p->bdev, PAGE_SIZE);
2948 if (error < 0)
2949 return error;
2950 /*
2951 * Zoned block devices contain zones that have a sequential
2952 * write only restriction. Hence zoned block devices are not
2953 * suitable for swapping. Disallow them here.
2954 */
2955 if (blk_queue_is_zoned(p->bdev->bd_disk->queue))
2956 return -EINVAL;
2957 p->flags |= SWP_BLKDEV;
2958 } else if (S_ISREG(inode->i_mode)) {
2959 p->bdev = inode->i_sb->s_bdev;
2960 }
2961
2962 return 0;
2963 }
2964
2965
2966 /*
2967 * Find out how many pages are allowed for a single swap device. There
2968 * are two limiting factors:
2969 * 1) the number of bits for the swap offset in the swp_entry_t type, and
2970 * 2) the number of bits in the swap pte, as defined by the different
2971 * architectures.
2972 *
2973 * In order to find the largest possible bit mask, a swap entry with
2974 * swap type 0 and swap offset ~0UL is created, encoded to a swap pte,
2975 * decoded to a swp_entry_t again, and finally the swap offset is
2976 * extracted.
2977 *
2978 * This will mask all the bits from the initial ~0UL mask that can't
2979 * be encoded in either the swp_entry_t or the architecture definition
2980 * of a swap pte.
2981 */
generic_max_swapfile_size(void)2982 unsigned long generic_max_swapfile_size(void)
2983 {
2984 return swp_offset(pte_to_swp_entry(
2985 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2986 }
2987
2988 /* Can be overridden by an architecture for additional checks. */
max_swapfile_size(void)2989 __weak unsigned long max_swapfile_size(void)
2990 {
2991 return generic_max_swapfile_size();
2992 }
2993
read_swap_header(struct swap_info_struct * p,union swap_header * swap_header,struct inode * inode)2994 static unsigned long read_swap_header(struct swap_info_struct *p,
2995 union swap_header *swap_header,
2996 struct inode *inode)
2997 {
2998 int i;
2999 unsigned long maxpages;
3000 unsigned long swapfilepages;
3001 unsigned long last_page;
3002
3003 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
3004 pr_err("Unable to find swap-space signature\n");
3005 return 0;
3006 }
3007
3008 /* swap partition endianess hack... */
3009 if (swab32(swap_header->info.version) == 1) {
3010 swab32s(&swap_header->info.version);
3011 swab32s(&swap_header->info.last_page);
3012 swab32s(&swap_header->info.nr_badpages);
3013 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
3014 return 0;
3015 for (i = 0; i < swap_header->info.nr_badpages; i++)
3016 swab32s(&swap_header->info.badpages[i]);
3017 }
3018 /* Check the swap header's sub-version */
3019 if (swap_header->info.version != 1) {
3020 pr_warn("Unable to handle swap header version %d\n",
3021 swap_header->info.version);
3022 return 0;
3023 }
3024
3025 p->lowest_bit = 1;
3026 p->cluster_next = 1;
3027 p->cluster_nr = 0;
3028
3029 maxpages = max_swapfile_size();
3030 last_page = swap_header->info.last_page;
3031 if (!last_page) {
3032 pr_warn("Empty swap-file\n");
3033 return 0;
3034 }
3035 if (last_page > maxpages) {
3036 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
3037 maxpages << (PAGE_SHIFT - 10),
3038 last_page << (PAGE_SHIFT - 10));
3039 }
3040 if (maxpages > last_page) {
3041 maxpages = last_page + 1;
3042 /* p->max is an unsigned int: don't overflow it */
3043 if ((unsigned int)maxpages == 0)
3044 maxpages = UINT_MAX;
3045 }
3046 p->highest_bit = maxpages - 1;
3047
3048 if (!maxpages)
3049 return 0;
3050 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
3051 if (swapfilepages && maxpages > swapfilepages) {
3052 pr_warn("Swap area shorter than signature indicates\n");
3053 return 0;
3054 }
3055 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
3056 return 0;
3057 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
3058 return 0;
3059
3060 return maxpages;
3061 }
3062
3063 #define SWAP_CLUSTER_INFO_COLS \
3064 DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
3065 #define SWAP_CLUSTER_SPACE_COLS \
3066 DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
3067 #define SWAP_CLUSTER_COLS \
3068 max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
3069
setup_swap_map_and_extents(struct swap_info_struct * p,union swap_header * swap_header,unsigned char * swap_map,struct swap_cluster_info * cluster_info,unsigned long maxpages,sector_t * span)3070 static int setup_swap_map_and_extents(struct swap_info_struct *p,
3071 union swap_header *swap_header,
3072 unsigned char *swap_map,
3073 struct swap_cluster_info *cluster_info,
3074 unsigned long maxpages,
3075 sector_t *span)
3076 {
3077 unsigned int j, k;
3078 unsigned int nr_good_pages;
3079 int nr_extents;
3080 unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
3081 unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS;
3082 unsigned long i, idx;
3083
3084 nr_good_pages = maxpages - 1; /* omit header page */
3085
3086 cluster_list_init(&p->free_clusters);
3087 cluster_list_init(&p->discard_clusters);
3088
3089 for (i = 0; i < swap_header->info.nr_badpages; i++) {
3090 unsigned int page_nr = swap_header->info.badpages[i];
3091 if (page_nr == 0 || page_nr > swap_header->info.last_page)
3092 return -EINVAL;
3093 if (page_nr < maxpages) {
3094 swap_map[page_nr] = SWAP_MAP_BAD;
3095 nr_good_pages--;
3096 /*
3097 * Haven't marked the cluster free yet, no list
3098 * operation involved
3099 */
3100 inc_cluster_info_page(p, cluster_info, page_nr);
3101 }
3102 }
3103
3104 /* Haven't marked the cluster free yet, no list operation involved */
3105 for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
3106 inc_cluster_info_page(p, cluster_info, i);
3107
3108 if (nr_good_pages) {
3109 swap_map[0] = SWAP_MAP_BAD;
3110 /*
3111 * Not mark the cluster free yet, no list
3112 * operation involved
3113 */
3114 inc_cluster_info_page(p, cluster_info, 0);
3115 p->max = maxpages;
3116 p->pages = nr_good_pages;
3117 nr_extents = setup_swap_extents(p, span);
3118 if (nr_extents < 0)
3119 return nr_extents;
3120 nr_good_pages = p->pages;
3121 }
3122 if (!nr_good_pages) {
3123 pr_warn("Empty swap-file\n");
3124 return -EINVAL;
3125 }
3126
3127 if (!cluster_info)
3128 return nr_extents;
3129
3130
3131 /*
3132 * Reduce false cache line sharing between cluster_info and
3133 * sharing same address space.
3134 */
3135 for (k = 0; k < SWAP_CLUSTER_COLS; k++) {
3136 j = (k + col) % SWAP_CLUSTER_COLS;
3137 for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) {
3138 idx = i * SWAP_CLUSTER_COLS + j;
3139 if (idx >= nr_clusters)
3140 continue;
3141 if (cluster_count(&cluster_info[idx]))
3142 continue;
3143 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
3144 cluster_list_add_tail(&p->free_clusters, cluster_info,
3145 idx);
3146 }
3147 }
3148 return nr_extents;
3149 }
3150
3151 /*
3152 * Helper to sys_swapon determining if a given swap
3153 * backing device queue supports DISCARD operations.
3154 */
swap_discardable(struct swap_info_struct * si)3155 static bool swap_discardable(struct swap_info_struct *si)
3156 {
3157 struct request_queue *q = bdev_get_queue(si->bdev);
3158
3159 if (!q || !blk_queue_discard(q))
3160 return false;
3161
3162 return true;
3163 }
3164
SYSCALL_DEFINE2(swapon,const char __user *,specialfile,int,swap_flags)3165 SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
3166 {
3167 struct swap_info_struct *p;
3168 struct filename *name;
3169 struct file *swap_file = NULL;
3170 struct address_space *mapping;
3171 int prio;
3172 int error;
3173 union swap_header *swap_header;
3174 int nr_extents;
3175 sector_t span;
3176 unsigned long maxpages;
3177 unsigned char *swap_map = NULL;
3178 struct swap_cluster_info *cluster_info = NULL;
3179 unsigned long *frontswap_map = NULL;
3180 struct page *page = NULL;
3181 struct inode *inode = NULL;
3182 bool inced_nr_rotate_swap = false;
3183
3184 if (swap_flags & ~SWAP_FLAGS_VALID)
3185 return -EINVAL;
3186
3187 if (!capable(CAP_SYS_ADMIN))
3188 return -EPERM;
3189
3190 if (!swap_avail_heads)
3191 return -ENOMEM;
3192
3193 p = alloc_swap_info();
3194 if (IS_ERR(p))
3195 return PTR_ERR(p);
3196
3197 INIT_WORK(&p->discard_work, swap_discard_work);
3198
3199 name = getname(specialfile);
3200 if (IS_ERR(name)) {
3201 error = PTR_ERR(name);
3202 name = NULL;
3203 goto bad_swap;
3204 }
3205 swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
3206 if (IS_ERR(swap_file)) {
3207 error = PTR_ERR(swap_file);
3208 swap_file = NULL;
3209 goto bad_swap;
3210 }
3211
3212 p->swap_file = swap_file;
3213 mapping = swap_file->f_mapping;
3214 inode = mapping->host;
3215
3216 error = claim_swapfile(p, inode);
3217 if (unlikely(error))
3218 goto bad_swap;
3219
3220 inode_lock(inode);
3221 if (IS_SWAPFILE(inode)) {
3222 error = -EBUSY;
3223 goto bad_swap_unlock_inode;
3224 }
3225
3226 /*
3227 * Read the swap header.
3228 */
3229 if (!mapping->a_ops->readpage) {
3230 error = -EINVAL;
3231 goto bad_swap_unlock_inode;
3232 }
3233 page = read_mapping_page(mapping, 0, swap_file);
3234 if (IS_ERR(page)) {
3235 error = PTR_ERR(page);
3236 goto bad_swap_unlock_inode;
3237 }
3238 swap_header = kmap(page);
3239
3240 maxpages = read_swap_header(p, swap_header, inode);
3241 if (unlikely(!maxpages)) {
3242 error = -EINVAL;
3243 goto bad_swap_unlock_inode;
3244 }
3245
3246 /* OK, set up the swap map and apply the bad block list */
3247 swap_map = vzalloc(maxpages);
3248 if (!swap_map) {
3249 error = -ENOMEM;
3250 goto bad_swap_unlock_inode;
3251 }
3252
3253 if (p->bdev && blk_queue_stable_writes(p->bdev->bd_disk->queue))
3254 p->flags |= SWP_STABLE_WRITES;
3255
3256 if (p->bdev && p->bdev->bd_disk->fops->rw_page)
3257 p->flags |= SWP_SYNCHRONOUS_IO;
3258
3259 if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) {
3260 int cpu;
3261 unsigned long ci, nr_cluster;
3262
3263 p->flags |= SWP_SOLIDSTATE;
3264 p->cluster_next_cpu = alloc_percpu(unsigned int);
3265 if (!p->cluster_next_cpu) {
3266 error = -ENOMEM;
3267 goto bad_swap_unlock_inode;
3268 }
3269 /*
3270 * select a random position to start with to help wear leveling
3271 * SSD
3272 */
3273 for_each_possible_cpu(cpu) {
3274 per_cpu(*p->cluster_next_cpu, cpu) =
3275 1 + prandom_u32_max(p->highest_bit);
3276 }
3277 nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
3278
3279 cluster_info = kvcalloc(nr_cluster, sizeof(*cluster_info),
3280 GFP_KERNEL);
3281 if (!cluster_info) {
3282 error = -ENOMEM;
3283 goto bad_swap_unlock_inode;
3284 }
3285
3286 for (ci = 0; ci < nr_cluster; ci++)
3287 spin_lock_init(&((cluster_info + ci)->lock));
3288
3289 p->percpu_cluster = alloc_percpu(struct percpu_cluster);
3290 if (!p->percpu_cluster) {
3291 error = -ENOMEM;
3292 goto bad_swap_unlock_inode;
3293 }
3294 for_each_possible_cpu(cpu) {
3295 struct percpu_cluster *cluster;
3296 cluster = per_cpu_ptr(p->percpu_cluster, cpu);
3297 cluster_set_null(&cluster->index);
3298 }
3299 } else {
3300 atomic_inc(&nr_rotate_swap);
3301 inced_nr_rotate_swap = true;
3302 }
3303
3304 error = swap_cgroup_swapon(p->type, maxpages);
3305 if (error)
3306 goto bad_swap_unlock_inode;
3307
3308 nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
3309 cluster_info, maxpages, &span);
3310 if (unlikely(nr_extents < 0)) {
3311 error = nr_extents;
3312 goto bad_swap_unlock_inode;
3313 }
3314 /* frontswap enabled? set up bit-per-page map for frontswap */
3315 if (IS_ENABLED(CONFIG_FRONTSWAP))
3316 frontswap_map = kvcalloc(BITS_TO_LONGS(maxpages),
3317 sizeof(long),
3318 GFP_KERNEL);
3319
3320 if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) {
3321 /*
3322 * When discard is enabled for swap with no particular
3323 * policy flagged, we set all swap discard flags here in
3324 * order to sustain backward compatibility with older
3325 * swapon(8) releases.
3326 */
3327 p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
3328 SWP_PAGE_DISCARD);
3329
3330 /*
3331 * By flagging sys_swapon, a sysadmin can tell us to
3332 * either do single-time area discards only, or to just
3333 * perform discards for released swap page-clusters.
3334 * Now it's time to adjust the p->flags accordingly.
3335 */
3336 if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
3337 p->flags &= ~SWP_PAGE_DISCARD;
3338 else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
3339 p->flags &= ~SWP_AREA_DISCARD;
3340
3341 /* issue a swapon-time discard if it's still required */
3342 if (p->flags & SWP_AREA_DISCARD) {
3343 int err = discard_swap(p);
3344 if (unlikely(err))
3345 pr_err("swapon: discard_swap(%p): %d\n",
3346 p, err);
3347 }
3348 }
3349
3350 error = init_swap_address_space(p->type, maxpages);
3351 if (error)
3352 goto bad_swap_unlock_inode;
3353
3354 /*
3355 * Flush any pending IO and dirty mappings before we start using this
3356 * swap device.
3357 */
3358 inode->i_flags |= S_SWAPFILE;
3359 error = inode_drain_writes(inode);
3360 if (error) {
3361 inode->i_flags &= ~S_SWAPFILE;
3362 goto free_swap_address_space;
3363 }
3364
3365 mutex_lock(&swapon_mutex);
3366 prio = -1;
3367 if (swap_flags & SWAP_FLAG_PREFER)
3368 prio =
3369 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
3370 enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
3371
3372 pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
3373 p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
3374 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
3375 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
3376 (p->flags & SWP_DISCARDABLE) ? "D" : "",
3377 (p->flags & SWP_AREA_DISCARD) ? "s" : "",
3378 (p->flags & SWP_PAGE_DISCARD) ? "c" : "",
3379 (frontswap_map) ? "FS" : "");
3380
3381 mutex_unlock(&swapon_mutex);
3382 atomic_inc(&proc_poll_event);
3383 wake_up_interruptible(&proc_poll_wait);
3384
3385 error = 0;
3386 goto out;
3387 free_swap_address_space:
3388 exit_swap_address_space(p->type);
3389 bad_swap_unlock_inode:
3390 inode_unlock(inode);
3391 bad_swap:
3392 free_percpu(p->percpu_cluster);
3393 p->percpu_cluster = NULL;
3394 free_percpu(p->cluster_next_cpu);
3395 p->cluster_next_cpu = NULL;
3396 if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
3397 set_blocksize(p->bdev, p->old_block_size);
3398 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
3399 }
3400 inode = NULL;
3401 destroy_swap_extents(p);
3402 swap_cgroup_swapoff(p->type);
3403 spin_lock(&swap_lock);
3404 p->swap_file = NULL;
3405 p->flags = 0;
3406 spin_unlock(&swap_lock);
3407 vfree(swap_map);
3408 kvfree(cluster_info);
3409 kvfree(frontswap_map);
3410 if (inced_nr_rotate_swap)
3411 atomic_dec(&nr_rotate_swap);
3412 if (swap_file)
3413 filp_close(swap_file, NULL);
3414 out:
3415 if (page && !IS_ERR(page)) {
3416 kunmap(page);
3417 put_page(page);
3418 }
3419 if (name)
3420 putname(name);
3421 if (inode)
3422 inode_unlock(inode);
3423 if (!error)
3424 enable_swap_slots_cache();
3425 return error;
3426 }
3427
si_swapinfo(struct sysinfo * val)3428 void si_swapinfo(struct sysinfo *val)
3429 {
3430 unsigned int type;
3431 unsigned long nr_to_be_unused = 0;
3432
3433 spin_lock(&swap_lock);
3434 for (type = 0; type < nr_swapfiles; type++) {
3435 struct swap_info_struct *si = swap_info[type];
3436
3437 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
3438 nr_to_be_unused += si->inuse_pages;
3439 }
3440 val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
3441 val->totalswap = total_swap_pages + nr_to_be_unused;
3442 spin_unlock(&swap_lock);
3443 }
3444
3445 #ifdef CONFIG_HYPERHOLD_ZSWAPD
free_swap_is_low(void)3446 bool free_swap_is_low(void)
3447 {
3448 unsigned int type;
3449 unsigned long long freeswap = 0;
3450 unsigned long nr_to_be_unused = 0;
3451
3452 spin_lock(&swap_lock);
3453 for (type = 0; type < nr_swapfiles; type++) {
3454 struct swap_info_struct *si = swap_info[type];
3455
3456 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
3457 nr_to_be_unused += si->inuse_pages;
3458 }
3459 freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
3460 spin_unlock(&swap_lock);
3461
3462 return (freeswap < get_free_swap_threshold());
3463 }
3464 EXPORT_SYMBOL(free_swap_is_low);
3465 #endif
3466
3467 /*
3468 * Verify that a swap entry is valid and increment its swap map count.
3469 *
3470 * Returns error code in following case.
3471 * - success -> 0
3472 * - swp_entry is invalid -> EINVAL
3473 * - swp_entry is migration entry -> EINVAL
3474 * - swap-cache reference is requested but there is already one. -> EEXIST
3475 * - swap-cache reference is requested but the entry is not used. -> ENOENT
3476 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
3477 */
__swap_duplicate(swp_entry_t entry,unsigned char usage)3478 static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
3479 {
3480 struct swap_info_struct *p;
3481 struct swap_cluster_info *ci;
3482 unsigned long offset;
3483 unsigned char count;
3484 unsigned char has_cache;
3485 int err = -EINVAL;
3486
3487 p = get_swap_device(entry);
3488 if (!p)
3489 goto out;
3490
3491 offset = swp_offset(entry);
3492 ci = lock_cluster_or_swap_info(p, offset);
3493
3494 count = p->swap_map[offset];
3495
3496 /*
3497 * swapin_readahead() doesn't check if a swap entry is valid, so the
3498 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
3499 */
3500 if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
3501 err = -ENOENT;
3502 goto unlock_out;
3503 }
3504
3505 has_cache = count & SWAP_HAS_CACHE;
3506 count &= ~SWAP_HAS_CACHE;
3507 err = 0;
3508
3509 if (usage == SWAP_HAS_CACHE) {
3510
3511 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
3512 if (!has_cache && count)
3513 has_cache = SWAP_HAS_CACHE;
3514 else if (has_cache) /* someone else added cache */
3515 err = -EEXIST;
3516 else /* no users remaining */
3517 err = -ENOENT;
3518
3519 } else if (count || has_cache) {
3520
3521 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
3522 count += usage;
3523 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
3524 err = -EINVAL;
3525 else if (swap_count_continued(p, offset, count))
3526 count = COUNT_CONTINUED;
3527 else
3528 err = -ENOMEM;
3529 } else
3530 err = -ENOENT; /* unused swap entry */
3531
3532 WRITE_ONCE(p->swap_map[offset], count | has_cache);
3533
3534 unlock_out:
3535 unlock_cluster_or_swap_info(p, ci);
3536 out:
3537 if (p)
3538 put_swap_device(p);
3539 return err;
3540 }
3541
3542 /*
3543 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3544 * (in which case its reference count is never incremented).
3545 */
swap_shmem_alloc(swp_entry_t entry)3546 void swap_shmem_alloc(swp_entry_t entry)
3547 {
3548 __swap_duplicate(entry, SWAP_MAP_SHMEM);
3549 }
3550
3551 /*
3552 * Increase reference count of swap entry by 1.
3553 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3554 * but could not be atomically allocated. Returns 0, just as if it succeeded,
3555 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3556 * might occur if a page table entry has got corrupted.
3557 */
swap_duplicate(swp_entry_t entry)3558 int swap_duplicate(swp_entry_t entry)
3559 {
3560 int err = 0;
3561
3562 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
3563 err = add_swap_count_continuation(entry, GFP_ATOMIC);
3564 return err;
3565 }
3566
3567 /*
3568 * @entry: swap entry for which we allocate swap cache.
3569 *
3570 * Called when allocating swap cache for existing swap entry,
3571 * This can return error codes. Returns 0 at success.
3572 * -EEXIST means there is a swap cache.
3573 * Note: return code is different from swap_duplicate().
3574 */
swapcache_prepare(swp_entry_t entry)3575 int swapcache_prepare(swp_entry_t entry)
3576 {
3577 return __swap_duplicate(entry, SWAP_HAS_CACHE);
3578 }
3579
swp_swap_info(swp_entry_t entry)3580 struct swap_info_struct *swp_swap_info(swp_entry_t entry)
3581 {
3582 return swap_type_to_swap_info(swp_type(entry));
3583 }
3584
page_swap_info(struct page * page)3585 struct swap_info_struct *page_swap_info(struct page *page)
3586 {
3587 swp_entry_t entry = { .val = page_private(page) };
3588 return swp_swap_info(entry);
3589 }
3590
3591 /*
3592 * out-of-line __page_file_ methods to avoid include hell.
3593 */
__page_file_mapping(struct page * page)3594 struct address_space *__page_file_mapping(struct page *page)
3595 {
3596 return page_swap_info(page)->swap_file->f_mapping;
3597 }
3598 EXPORT_SYMBOL_GPL(__page_file_mapping);
3599
__page_file_index(struct page * page)3600 pgoff_t __page_file_index(struct page *page)
3601 {
3602 swp_entry_t swap = { .val = page_private(page) };
3603 return swp_offset(swap);
3604 }
3605 EXPORT_SYMBOL_GPL(__page_file_index);
3606
3607 /*
3608 * add_swap_count_continuation - called when a swap count is duplicated
3609 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3610 * page of the original vmalloc'ed swap_map, to hold the continuation count
3611 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
3612 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3613 *
3614 * These continuation pages are seldom referenced: the common paths all work
3615 * on the original swap_map, only referring to a continuation page when the
3616 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3617 *
3618 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3619 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3620 * can be called after dropping locks.
3621 */
add_swap_count_continuation(swp_entry_t entry,gfp_t gfp_mask)3622 int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
3623 {
3624 struct swap_info_struct *si;
3625 struct swap_cluster_info *ci;
3626 struct page *head;
3627 struct page *page;
3628 struct page *list_page;
3629 pgoff_t offset;
3630 unsigned char count;
3631 int ret = 0;
3632
3633 /*
3634 * When debugging, it's easier to use __GFP_ZERO here; but it's better
3635 * for latency not to zero a page while GFP_ATOMIC and holding locks.
3636 */
3637 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
3638
3639 si = get_swap_device(entry);
3640 if (!si) {
3641 /*
3642 * An acceptable race has occurred since the failing
3643 * __swap_duplicate(): the swap device may be swapoff
3644 */
3645 goto outer;
3646 }
3647 spin_lock(&si->lock);
3648
3649 offset = swp_offset(entry);
3650
3651 ci = lock_cluster(si, offset);
3652
3653 count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
3654
3655 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
3656 /*
3657 * The higher the swap count, the more likely it is that tasks
3658 * will race to add swap count continuation: we need to avoid
3659 * over-provisioning.
3660 */
3661 goto out;
3662 }
3663
3664 if (!page) {
3665 ret = -ENOMEM;
3666 goto out;
3667 }
3668
3669 /*
3670 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
3671 * no architecture is using highmem pages for kernel page tables: so it
3672 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
3673 */
3674 head = vmalloc_to_page(si->swap_map + offset);
3675 offset &= ~PAGE_MASK;
3676
3677 spin_lock(&si->cont_lock);
3678 /*
3679 * Page allocation does not initialize the page's lru field,
3680 * but it does always reset its private field.
3681 */
3682 if (!page_private(head)) {
3683 BUG_ON(count & COUNT_CONTINUED);
3684 INIT_LIST_HEAD(&head->lru);
3685 set_page_private(head, SWP_CONTINUED);
3686 si->flags |= SWP_CONTINUED;
3687 }
3688
3689 list_for_each_entry(list_page, &head->lru, lru) {
3690 unsigned char *map;
3691
3692 /*
3693 * If the previous map said no continuation, but we've found
3694 * a continuation page, free our allocation and use this one.
3695 */
3696 if (!(count & COUNT_CONTINUED))
3697 goto out_unlock_cont;
3698
3699 map = kmap_atomic(list_page) + offset;
3700 count = *map;
3701 kunmap_atomic(map);
3702
3703 /*
3704 * If this continuation count now has some space in it,
3705 * free our allocation and use this one.
3706 */
3707 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
3708 goto out_unlock_cont;
3709 }
3710
3711 list_add_tail(&page->lru, &head->lru);
3712 page = NULL; /* now it's attached, don't free it */
3713 out_unlock_cont:
3714 spin_unlock(&si->cont_lock);
3715 out:
3716 unlock_cluster(ci);
3717 spin_unlock(&si->lock);
3718 put_swap_device(si);
3719 outer:
3720 if (page)
3721 __free_page(page);
3722 return ret;
3723 }
3724
3725 /*
3726 * swap_count_continued - when the original swap_map count is incremented
3727 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3728 * into, carry if so, or else fail until a new continuation page is allocated;
3729 * when the original swap_map count is decremented from 0 with continuation,
3730 * borrow from the continuation and report whether it still holds more.
3731 * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3732 * lock.
3733 */
swap_count_continued(struct swap_info_struct * si,pgoff_t offset,unsigned char count)3734 static bool swap_count_continued(struct swap_info_struct *si,
3735 pgoff_t offset, unsigned char count)
3736 {
3737 struct page *head;
3738 struct page *page;
3739 unsigned char *map;
3740 bool ret;
3741
3742 head = vmalloc_to_page(si->swap_map + offset);
3743 if (page_private(head) != SWP_CONTINUED) {
3744 BUG_ON(count & COUNT_CONTINUED);
3745 return false; /* need to add count continuation */
3746 }
3747
3748 spin_lock(&si->cont_lock);
3749 offset &= ~PAGE_MASK;
3750 page = list_next_entry(head, lru);
3751 map = kmap_atomic(page) + offset;
3752
3753 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
3754 goto init_map; /* jump over SWAP_CONT_MAX checks */
3755
3756 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
3757 /*
3758 * Think of how you add 1 to 999
3759 */
3760 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
3761 kunmap_atomic(map);
3762 page = list_next_entry(page, lru);
3763 BUG_ON(page == head);
3764 map = kmap_atomic(page) + offset;
3765 }
3766 if (*map == SWAP_CONT_MAX) {
3767 kunmap_atomic(map);
3768 page = list_next_entry(page, lru);
3769 if (page == head) {
3770 ret = false; /* add count continuation */
3771 goto out;
3772 }
3773 map = kmap_atomic(page) + offset;
3774 init_map: *map = 0; /* we didn't zero the page */
3775 }
3776 *map += 1;
3777 kunmap_atomic(map);
3778 while ((page = list_prev_entry(page, lru)) != head) {
3779 map = kmap_atomic(page) + offset;
3780 *map = COUNT_CONTINUED;
3781 kunmap_atomic(map);
3782 }
3783 ret = true; /* incremented */
3784
3785 } else { /* decrementing */
3786 /*
3787 * Think of how you subtract 1 from 1000
3788 */
3789 BUG_ON(count != COUNT_CONTINUED);
3790 while (*map == COUNT_CONTINUED) {
3791 kunmap_atomic(map);
3792 page = list_next_entry(page, lru);
3793 BUG_ON(page == head);
3794 map = kmap_atomic(page) + offset;
3795 }
3796 BUG_ON(*map == 0);
3797 *map -= 1;
3798 if (*map == 0)
3799 count = 0;
3800 kunmap_atomic(map);
3801 while ((page = list_prev_entry(page, lru)) != head) {
3802 map = kmap_atomic(page) + offset;
3803 *map = SWAP_CONT_MAX | count;
3804 count = COUNT_CONTINUED;
3805 kunmap_atomic(map);
3806 }
3807 ret = count == COUNT_CONTINUED;
3808 }
3809 out:
3810 spin_unlock(&si->cont_lock);
3811 return ret;
3812 }
3813
3814 /*
3815 * free_swap_count_continuations - swapoff free all the continuation pages
3816 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3817 */
free_swap_count_continuations(struct swap_info_struct * si)3818 static void free_swap_count_continuations(struct swap_info_struct *si)
3819 {
3820 pgoff_t offset;
3821
3822 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
3823 struct page *head;
3824 head = vmalloc_to_page(si->swap_map + offset);
3825 if (page_private(head)) {
3826 struct page *page, *next;
3827
3828 list_for_each_entry_safe(page, next, &head->lru, lru) {
3829 list_del(&page->lru);
3830 __free_page(page);
3831 }
3832 }
3833 }
3834 }
3835
3836 #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
cgroup_throttle_swaprate(struct page * page,gfp_t gfp_mask)3837 void cgroup_throttle_swaprate(struct page *page, gfp_t gfp_mask)
3838 {
3839 struct swap_info_struct *si, *next;
3840 int nid = page_to_nid(page);
3841
3842 if (!(gfp_mask & __GFP_IO))
3843 return;
3844
3845 if (!blk_cgroup_congested())
3846 return;
3847
3848 /*
3849 * We've already scheduled a throttle, avoid taking the global swap
3850 * lock.
3851 */
3852 if (current->throttle_queue)
3853 return;
3854
3855 spin_lock(&swap_avail_lock);
3856 plist_for_each_entry_safe(si, next, &swap_avail_heads[nid],
3857 avail_lists[nid]) {
3858 if (si->bdev) {
3859 blkcg_schedule_throttle(bdev_get_queue(si->bdev), true);
3860 break;
3861 }
3862 }
3863 spin_unlock(&swap_avail_lock);
3864 }
3865 #endif
3866
swapfile_init(void)3867 static int __init swapfile_init(void)
3868 {
3869 int nid;
3870
3871 swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head),
3872 GFP_KERNEL);
3873 if (!swap_avail_heads) {
3874 pr_emerg("Not enough memory for swap heads, swap is disabled\n");
3875 return -ENOMEM;
3876 }
3877
3878 for_each_node(nid)
3879 plist_head_init(&swap_avail_heads[nid]);
3880
3881 return 0;
3882 }
3883 subsys_initcall(swapfile_init);
3884