]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - mm/swapfile.c
Merge tag 'linux-kselftest-4.12-rc1' of git://git.kernel.org/pub/scm/linux/kernel...
[mirror_ubuntu-bionic-kernel.git] / mm / swapfile.c
1 /*
2 * linux/mm/swapfile.c
3 *
4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 * Swap reorganised 29.12.95, Stephen Tweedie
6 */
7
8 #include <linux/mm.h>
9 #include <linux/sched/mm.h>
10 #include <linux/sched/task.h>
11 #include <linux/hugetlb.h>
12 #include <linux/mman.h>
13 #include <linux/slab.h>
14 #include <linux/kernel_stat.h>
15 #include <linux/swap.h>
16 #include <linux/vmalloc.h>
17 #include <linux/pagemap.h>
18 #include <linux/namei.h>
19 #include <linux/shmem_fs.h>
20 #include <linux/blkdev.h>
21 #include <linux/random.h>
22 #include <linux/writeback.h>
23 #include <linux/proc_fs.h>
24 #include <linux/seq_file.h>
25 #include <linux/init.h>
26 #include <linux/ksm.h>
27 #include <linux/rmap.h>
28 #include <linux/security.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mutex.h>
31 #include <linux/capability.h>
32 #include <linux/syscalls.h>
33 #include <linux/memcontrol.h>
34 #include <linux/poll.h>
35 #include <linux/oom.h>
36 #include <linux/frontswap.h>
37 #include <linux/swapfile.h>
38 #include <linux/export.h>
39 #include <linux/swap_slots.h>
40
41 #include <asm/pgtable.h>
42 #include <asm/tlbflush.h>
43 #include <linux/swapops.h>
44 #include <linux/swap_cgroup.h>
45
46 static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
47 unsigned char);
48 static void free_swap_count_continuations(struct swap_info_struct *);
49 static sector_t map_swap_entry(swp_entry_t, struct block_device**);
50
51 DEFINE_SPINLOCK(swap_lock);
52 static unsigned int nr_swapfiles;
53 atomic_long_t nr_swap_pages;
54 /*
55 * Some modules use swappable objects and may try to swap them out under
56 * memory pressure (via the shrinker). Before doing so, they may wish to
57 * check to see if any swap space is available.
58 */
59 EXPORT_SYMBOL_GPL(nr_swap_pages);
60 /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
61 long total_swap_pages;
62 static int least_priority;
63
64 static const char Bad_file[] = "Bad swap file entry ";
65 static const char Unused_file[] = "Unused swap file entry ";
66 static const char Bad_offset[] = "Bad swap offset entry ";
67 static const char Unused_offset[] = "Unused swap offset entry ";
68
69 /*
70 * all active swap_info_structs
71 * protected with swap_lock, and ordered by priority.
72 */
73 PLIST_HEAD(swap_active_head);
74
75 /*
76 * all available (active, not full) swap_info_structs
77 * protected with swap_avail_lock, ordered by priority.
78 * This is used by get_swap_page() instead of swap_active_head
79 * because swap_active_head includes all swap_info_structs,
80 * but get_swap_page() doesn't need to look at full ones.
81 * This uses its own lock instead of swap_lock because when a
82 * swap_info_struct changes between not-full/full, it needs to
83 * add/remove itself to/from this list, but the swap_info_struct->lock
84 * is held and the locking order requires swap_lock to be taken
85 * before any swap_info_struct->lock.
86 */
87 static PLIST_HEAD(swap_avail_head);
88 static DEFINE_SPINLOCK(swap_avail_lock);
89
90 struct swap_info_struct *swap_info[MAX_SWAPFILES];
91
92 static DEFINE_MUTEX(swapon_mutex);
93
94 static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
95 /* Activity counter to indicate that a swapon or swapoff has occurred */
96 static atomic_t proc_poll_event = ATOMIC_INIT(0);
97
98 static inline unsigned char swap_count(unsigned char ent)
99 {
100 return ent & ~SWAP_HAS_CACHE; /* may include SWAP_HAS_CONT flag */
101 }
102
103 /* returns 1 if swap entry is freed */
104 static int
105 __try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset)
106 {
107 swp_entry_t entry = swp_entry(si->type, offset);
108 struct page *page;
109 int ret = 0;
110
111 page = find_get_page(swap_address_space(entry), swp_offset(entry));
112 if (!page)
113 return 0;
114 /*
115 * This function is called from scan_swap_map() and it's called
116 * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
117 * We have to use trylock for avoiding deadlock. This is a special
118 * case and you should use try_to_free_swap() with explicit lock_page()
119 * in usual operations.
120 */
121 if (trylock_page(page)) {
122 ret = try_to_free_swap(page);
123 unlock_page(page);
124 }
125 put_page(page);
126 return ret;
127 }
128
129 /*
130 * swapon tell device that all the old swap contents can be discarded,
131 * to allow the swap device to optimize its wear-levelling.
132 */
133 static int discard_swap(struct swap_info_struct *si)
134 {
135 struct swap_extent *se;
136 sector_t start_block;
137 sector_t nr_blocks;
138 int err = 0;
139
140 /* Do not discard the swap header page! */
141 se = &si->first_swap_extent;
142 start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
143 nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
144 if (nr_blocks) {
145 err = blkdev_issue_discard(si->bdev, start_block,
146 nr_blocks, GFP_KERNEL, 0);
147 if (err)
148 return err;
149 cond_resched();
150 }
151
152 list_for_each_entry(se, &si->first_swap_extent.list, list) {
153 start_block = se->start_block << (PAGE_SHIFT - 9);
154 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
155
156 err = blkdev_issue_discard(si->bdev, start_block,
157 nr_blocks, GFP_KERNEL, 0);
158 if (err)
159 break;
160
161 cond_resched();
162 }
163 return err; /* That will often be -EOPNOTSUPP */
164 }
165
166 /*
167 * swap allocation tell device that a cluster of swap can now be discarded,
168 * to allow the swap device to optimize its wear-levelling.
169 */
170 static void discard_swap_cluster(struct swap_info_struct *si,
171 pgoff_t start_page, pgoff_t nr_pages)
172 {
173 struct swap_extent *se = si->curr_swap_extent;
174 int found_extent = 0;
175
176 while (nr_pages) {
177 if (se->start_page <= start_page &&
178 start_page < se->start_page + se->nr_pages) {
179 pgoff_t offset = start_page - se->start_page;
180 sector_t start_block = se->start_block + offset;
181 sector_t nr_blocks = se->nr_pages - offset;
182
183 if (nr_blocks > nr_pages)
184 nr_blocks = nr_pages;
185 start_page += nr_blocks;
186 nr_pages -= nr_blocks;
187
188 if (!found_extent++)
189 si->curr_swap_extent = se;
190
191 start_block <<= PAGE_SHIFT - 9;
192 nr_blocks <<= PAGE_SHIFT - 9;
193 if (blkdev_issue_discard(si->bdev, start_block,
194 nr_blocks, GFP_NOIO, 0))
195 break;
196 }
197
198 se = list_next_entry(se, list);
199 }
200 }
201
202 #define SWAPFILE_CLUSTER 256
203 #define LATENCY_LIMIT 256
204
205 static inline void cluster_set_flag(struct swap_cluster_info *info,
206 unsigned int flag)
207 {
208 info->flags = flag;
209 }
210
211 static inline unsigned int cluster_count(struct swap_cluster_info *info)
212 {
213 return info->data;
214 }
215
216 static inline void cluster_set_count(struct swap_cluster_info *info,
217 unsigned int c)
218 {
219 info->data = c;
220 }
221
222 static inline void cluster_set_count_flag(struct swap_cluster_info *info,
223 unsigned int c, unsigned int f)
224 {
225 info->flags = f;
226 info->data = c;
227 }
228
229 static inline unsigned int cluster_next(struct swap_cluster_info *info)
230 {
231 return info->data;
232 }
233
234 static inline void cluster_set_next(struct swap_cluster_info *info,
235 unsigned int n)
236 {
237 info->data = n;
238 }
239
240 static inline void cluster_set_next_flag(struct swap_cluster_info *info,
241 unsigned int n, unsigned int f)
242 {
243 info->flags = f;
244 info->data = n;
245 }
246
247 static inline bool cluster_is_free(struct swap_cluster_info *info)
248 {
249 return info->flags & CLUSTER_FLAG_FREE;
250 }
251
252 static inline bool cluster_is_null(struct swap_cluster_info *info)
253 {
254 return info->flags & CLUSTER_FLAG_NEXT_NULL;
255 }
256
257 static inline void cluster_set_null(struct swap_cluster_info *info)
258 {
259 info->flags = CLUSTER_FLAG_NEXT_NULL;
260 info->data = 0;
261 }
262
263 static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si,
264 unsigned long offset)
265 {
266 struct swap_cluster_info *ci;
267
268 ci = si->cluster_info;
269 if (ci) {
270 ci += offset / SWAPFILE_CLUSTER;
271 spin_lock(&ci->lock);
272 }
273 return ci;
274 }
275
276 static inline void unlock_cluster(struct swap_cluster_info *ci)
277 {
278 if (ci)
279 spin_unlock(&ci->lock);
280 }
281
282 static inline struct swap_cluster_info *lock_cluster_or_swap_info(
283 struct swap_info_struct *si,
284 unsigned long offset)
285 {
286 struct swap_cluster_info *ci;
287
288 ci = lock_cluster(si, offset);
289 if (!ci)
290 spin_lock(&si->lock);
291
292 return ci;
293 }
294
295 static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si,
296 struct swap_cluster_info *ci)
297 {
298 if (ci)
299 unlock_cluster(ci);
300 else
301 spin_unlock(&si->lock);
302 }
303
304 static inline bool cluster_list_empty(struct swap_cluster_list *list)
305 {
306 return cluster_is_null(&list->head);
307 }
308
309 static inline unsigned int cluster_list_first(struct swap_cluster_list *list)
310 {
311 return cluster_next(&list->head);
312 }
313
314 static void cluster_list_init(struct swap_cluster_list *list)
315 {
316 cluster_set_null(&list->head);
317 cluster_set_null(&list->tail);
318 }
319
320 static void cluster_list_add_tail(struct swap_cluster_list *list,
321 struct swap_cluster_info *ci,
322 unsigned int idx)
323 {
324 if (cluster_list_empty(list)) {
325 cluster_set_next_flag(&list->head, idx, 0);
326 cluster_set_next_flag(&list->tail, idx, 0);
327 } else {
328 struct swap_cluster_info *ci_tail;
329 unsigned int tail = cluster_next(&list->tail);
330
331 /*
332 * Nested cluster lock, but both cluster locks are
333 * only acquired when we held swap_info_struct->lock
334 */
335 ci_tail = ci + tail;
336 spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING);
337 cluster_set_next(ci_tail, idx);
338 spin_unlock(&ci_tail->lock);
339 cluster_set_next_flag(&list->tail, idx, 0);
340 }
341 }
342
343 static unsigned int cluster_list_del_first(struct swap_cluster_list *list,
344 struct swap_cluster_info *ci)
345 {
346 unsigned int idx;
347
348 idx = cluster_next(&list->head);
349 if (cluster_next(&list->tail) == idx) {
350 cluster_set_null(&list->head);
351 cluster_set_null(&list->tail);
352 } else
353 cluster_set_next_flag(&list->head,
354 cluster_next(&ci[idx]), 0);
355
356 return idx;
357 }
358
359 /* Add a cluster to discard list and schedule it to do discard */
360 static void swap_cluster_schedule_discard(struct swap_info_struct *si,
361 unsigned int idx)
362 {
363 /*
364 * If scan_swap_map() can't find a free cluster, it will check
365 * si->swap_map directly. To make sure the discarding cluster isn't
366 * taken by scan_swap_map(), mark the swap entries bad (occupied). It
367 * will be cleared after discard
368 */
369 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
370 SWAP_MAP_BAD, SWAPFILE_CLUSTER);
371
372 cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx);
373
374 schedule_work(&si->discard_work);
375 }
376
377 /*
378 * Doing discard actually. After a cluster discard is finished, the cluster
379 * will be added to free cluster list. caller should hold si->lock.
380 */
381 static void swap_do_scheduled_discard(struct swap_info_struct *si)
382 {
383 struct swap_cluster_info *info, *ci;
384 unsigned int idx;
385
386 info = si->cluster_info;
387
388 while (!cluster_list_empty(&si->discard_clusters)) {
389 idx = cluster_list_del_first(&si->discard_clusters, info);
390 spin_unlock(&si->lock);
391
392 discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
393 SWAPFILE_CLUSTER);
394
395 spin_lock(&si->lock);
396 ci = lock_cluster(si, idx * SWAPFILE_CLUSTER);
397 cluster_set_flag(ci, CLUSTER_FLAG_FREE);
398 unlock_cluster(ci);
399 cluster_list_add_tail(&si->free_clusters, info, idx);
400 ci = lock_cluster(si, idx * SWAPFILE_CLUSTER);
401 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
402 0, SWAPFILE_CLUSTER);
403 unlock_cluster(ci);
404 }
405 }
406
407 static void swap_discard_work(struct work_struct *work)
408 {
409 struct swap_info_struct *si;
410
411 si = container_of(work, struct swap_info_struct, discard_work);
412
413 spin_lock(&si->lock);
414 swap_do_scheduled_discard(si);
415 spin_unlock(&si->lock);
416 }
417
418 /*
419 * The cluster corresponding to page_nr will be used. The cluster will be
420 * removed from free cluster list and its usage counter will be increased.
421 */
422 static void inc_cluster_info_page(struct swap_info_struct *p,
423 struct swap_cluster_info *cluster_info, unsigned long page_nr)
424 {
425 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
426
427 if (!cluster_info)
428 return;
429 if (cluster_is_free(&cluster_info[idx])) {
430 VM_BUG_ON(cluster_list_first(&p->free_clusters) != idx);
431 cluster_list_del_first(&p->free_clusters, cluster_info);
432 cluster_set_count_flag(&cluster_info[idx], 0, 0);
433 }
434
435 VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
436 cluster_set_count(&cluster_info[idx],
437 cluster_count(&cluster_info[idx]) + 1);
438 }
439
440 /*
441 * The cluster corresponding to page_nr decreases one usage. If the usage
442 * counter becomes 0, which means no page in the cluster is in using, we can
443 * optionally discard the cluster and add it to free cluster list.
444 */
445 static void dec_cluster_info_page(struct swap_info_struct *p,
446 struct swap_cluster_info *cluster_info, unsigned long page_nr)
447 {
448 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
449
450 if (!cluster_info)
451 return;
452
453 VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
454 cluster_set_count(&cluster_info[idx],
455 cluster_count(&cluster_info[idx]) - 1);
456
457 if (cluster_count(&cluster_info[idx]) == 0) {
458 /*
459 * If the swap is discardable, prepare discard the cluster
460 * instead of free it immediately. The cluster will be freed
461 * after discard.
462 */
463 if ((p->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
464 (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
465 swap_cluster_schedule_discard(p, idx);
466 return;
467 }
468
469 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
470 cluster_list_add_tail(&p->free_clusters, cluster_info, idx);
471 }
472 }
473
474 /*
475 * It's possible scan_swap_map() uses a free cluster in the middle of free
476 * cluster list. Avoiding such abuse to avoid list corruption.
477 */
478 static bool
479 scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
480 unsigned long offset)
481 {
482 struct percpu_cluster *percpu_cluster;
483 bool conflict;
484
485 offset /= SWAPFILE_CLUSTER;
486 conflict = !cluster_list_empty(&si->free_clusters) &&
487 offset != cluster_list_first(&si->free_clusters) &&
488 cluster_is_free(&si->cluster_info[offset]);
489
490 if (!conflict)
491 return false;
492
493 percpu_cluster = this_cpu_ptr(si->percpu_cluster);
494 cluster_set_null(&percpu_cluster->index);
495 return true;
496 }
497
498 /*
499 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
500 * might involve allocating a new cluster for current CPU too.
501 */
502 static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
503 unsigned long *offset, unsigned long *scan_base)
504 {
505 struct percpu_cluster *cluster;
506 struct swap_cluster_info *ci;
507 bool found_free;
508 unsigned long tmp, max;
509
510 new_cluster:
511 cluster = this_cpu_ptr(si->percpu_cluster);
512 if (cluster_is_null(&cluster->index)) {
513 if (!cluster_list_empty(&si->free_clusters)) {
514 cluster->index = si->free_clusters.head;
515 cluster->next = cluster_next(&cluster->index) *
516 SWAPFILE_CLUSTER;
517 } else if (!cluster_list_empty(&si->discard_clusters)) {
518 /*
519 * we don't have free cluster but have some clusters in
520 * discarding, do discard now and reclaim them
521 */
522 swap_do_scheduled_discard(si);
523 *scan_base = *offset = si->cluster_next;
524 goto new_cluster;
525 } else
526 return false;
527 }
528
529 found_free = false;
530
531 /*
532 * Other CPUs can use our cluster if they can't find a free cluster,
533 * check if there is still free entry in the cluster
534 */
535 tmp = cluster->next;
536 max = min_t(unsigned long, si->max,
537 (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER);
538 if (tmp >= max) {
539 cluster_set_null(&cluster->index);
540 goto new_cluster;
541 }
542 ci = lock_cluster(si, tmp);
543 while (tmp < max) {
544 if (!si->swap_map[tmp]) {
545 found_free = true;
546 break;
547 }
548 tmp++;
549 }
550 unlock_cluster(ci);
551 if (!found_free) {
552 cluster_set_null(&cluster->index);
553 goto new_cluster;
554 }
555 cluster->next = tmp + 1;
556 *offset = tmp;
557 *scan_base = tmp;
558 return found_free;
559 }
560
561 static int scan_swap_map_slots(struct swap_info_struct *si,
562 unsigned char usage, int nr,
563 swp_entry_t slots[])
564 {
565 struct swap_cluster_info *ci;
566 unsigned long offset;
567 unsigned long scan_base;
568 unsigned long last_in_cluster = 0;
569 int latency_ration = LATENCY_LIMIT;
570 int n_ret = 0;
571
572 if (nr > SWAP_BATCH)
573 nr = SWAP_BATCH;
574
575 /*
576 * We try to cluster swap pages by allocating them sequentially
577 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
578 * way, however, we resort to first-free allocation, starting
579 * a new cluster. This prevents us from scattering swap pages
580 * all over the entire swap partition, so that we reduce
581 * overall disk seek times between swap pages. -- sct
582 * But we do now try to find an empty cluster. -Andrea
583 * And we let swap pages go all over an SSD partition. Hugh
584 */
585
586 si->flags += SWP_SCANNING;
587 scan_base = offset = si->cluster_next;
588
589 /* SSD algorithm */
590 if (si->cluster_info) {
591 if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
592 goto checks;
593 else
594 goto scan;
595 }
596
597 if (unlikely(!si->cluster_nr--)) {
598 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
599 si->cluster_nr = SWAPFILE_CLUSTER - 1;
600 goto checks;
601 }
602
603 spin_unlock(&si->lock);
604
605 /*
606 * If seek is expensive, start searching for new cluster from
607 * start of partition, to minimize the span of allocated swap.
608 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
609 * case, just handled by scan_swap_map_try_ssd_cluster() above.
610 */
611 scan_base = offset = si->lowest_bit;
612 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
613
614 /* Locate the first empty (unaligned) cluster */
615 for (; last_in_cluster <= si->highest_bit; offset++) {
616 if (si->swap_map[offset])
617 last_in_cluster = offset + SWAPFILE_CLUSTER;
618 else if (offset == last_in_cluster) {
619 spin_lock(&si->lock);
620 offset -= SWAPFILE_CLUSTER - 1;
621 si->cluster_next = offset;
622 si->cluster_nr = SWAPFILE_CLUSTER - 1;
623 goto checks;
624 }
625 if (unlikely(--latency_ration < 0)) {
626 cond_resched();
627 latency_ration = LATENCY_LIMIT;
628 }
629 }
630
631 offset = scan_base;
632 spin_lock(&si->lock);
633 si->cluster_nr = SWAPFILE_CLUSTER - 1;
634 }
635
636 checks:
637 if (si->cluster_info) {
638 while (scan_swap_map_ssd_cluster_conflict(si, offset)) {
639 /* take a break if we already got some slots */
640 if (n_ret)
641 goto done;
642 if (!scan_swap_map_try_ssd_cluster(si, &offset,
643 &scan_base))
644 goto scan;
645 }
646 }
647 if (!(si->flags & SWP_WRITEOK))
648 goto no_page;
649 if (!si->highest_bit)
650 goto no_page;
651 if (offset > si->highest_bit)
652 scan_base = offset = si->lowest_bit;
653
654 ci = lock_cluster(si, offset);
655 /* reuse swap entry of cache-only swap if not busy. */
656 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
657 int swap_was_freed;
658 unlock_cluster(ci);
659 spin_unlock(&si->lock);
660 swap_was_freed = __try_to_reclaim_swap(si, offset);
661 spin_lock(&si->lock);
662 /* entry was freed successfully, try to use this again */
663 if (swap_was_freed)
664 goto checks;
665 goto scan; /* check next one */
666 }
667
668 if (si->swap_map[offset]) {
669 unlock_cluster(ci);
670 if (!n_ret)
671 goto scan;
672 else
673 goto done;
674 }
675 si->swap_map[offset] = usage;
676 inc_cluster_info_page(si, si->cluster_info, offset);
677 unlock_cluster(ci);
678
679 if (offset == si->lowest_bit)
680 si->lowest_bit++;
681 if (offset == si->highest_bit)
682 si->highest_bit--;
683 si->inuse_pages++;
684 if (si->inuse_pages == si->pages) {
685 si->lowest_bit = si->max;
686 si->highest_bit = 0;
687 spin_lock(&swap_avail_lock);
688 plist_del(&si->avail_list, &swap_avail_head);
689 spin_unlock(&swap_avail_lock);
690 }
691 si->cluster_next = offset + 1;
692 slots[n_ret++] = swp_entry(si->type, offset);
693
694 /* got enough slots or reach max slots? */
695 if ((n_ret == nr) || (offset >= si->highest_bit))
696 goto done;
697
698 /* search for next available slot */
699
700 /* time to take a break? */
701 if (unlikely(--latency_ration < 0)) {
702 if (n_ret)
703 goto done;
704 spin_unlock(&si->lock);
705 cond_resched();
706 spin_lock(&si->lock);
707 latency_ration = LATENCY_LIMIT;
708 }
709
710 /* try to get more slots in cluster */
711 if (si->cluster_info) {
712 if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
713 goto checks;
714 else
715 goto done;
716 }
717 /* non-ssd case */
718 ++offset;
719
720 /* non-ssd case, still more slots in cluster? */
721 if (si->cluster_nr && !si->swap_map[offset]) {
722 --si->cluster_nr;
723 goto checks;
724 }
725
726 done:
727 si->flags -= SWP_SCANNING;
728 return n_ret;
729
730 scan:
731 spin_unlock(&si->lock);
732 while (++offset <= si->highest_bit) {
733 if (!si->swap_map[offset]) {
734 spin_lock(&si->lock);
735 goto checks;
736 }
737 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
738 spin_lock(&si->lock);
739 goto checks;
740 }
741 if (unlikely(--latency_ration < 0)) {
742 cond_resched();
743 latency_ration = LATENCY_LIMIT;
744 }
745 }
746 offset = si->lowest_bit;
747 while (offset < scan_base) {
748 if (!si->swap_map[offset]) {
749 spin_lock(&si->lock);
750 goto checks;
751 }
752 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
753 spin_lock(&si->lock);
754 goto checks;
755 }
756 if (unlikely(--latency_ration < 0)) {
757 cond_resched();
758 latency_ration = LATENCY_LIMIT;
759 }
760 offset++;
761 }
762 spin_lock(&si->lock);
763
764 no_page:
765 si->flags -= SWP_SCANNING;
766 return n_ret;
767 }
768
769 static unsigned long scan_swap_map(struct swap_info_struct *si,
770 unsigned char usage)
771 {
772 swp_entry_t entry;
773 int n_ret;
774
775 n_ret = scan_swap_map_slots(si, usage, 1, &entry);
776
777 if (n_ret)
778 return swp_offset(entry);
779 else
780 return 0;
781
782 }
783
784 int get_swap_pages(int n_goal, swp_entry_t swp_entries[])
785 {
786 struct swap_info_struct *si, *next;
787 long avail_pgs;
788 int n_ret = 0;
789
790 avail_pgs = atomic_long_read(&nr_swap_pages);
791 if (avail_pgs <= 0)
792 goto noswap;
793
794 if (n_goal > SWAP_BATCH)
795 n_goal = SWAP_BATCH;
796
797 if (n_goal > avail_pgs)
798 n_goal = avail_pgs;
799
800 atomic_long_sub(n_goal, &nr_swap_pages);
801
802 spin_lock(&swap_avail_lock);
803
804 start_over:
805 plist_for_each_entry_safe(si, next, &swap_avail_head, avail_list) {
806 /* requeue si to after same-priority siblings */
807 plist_requeue(&si->avail_list, &swap_avail_head);
808 spin_unlock(&swap_avail_lock);
809 spin_lock(&si->lock);
810 if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) {
811 spin_lock(&swap_avail_lock);
812 if (plist_node_empty(&si->avail_list)) {
813 spin_unlock(&si->lock);
814 goto nextsi;
815 }
816 WARN(!si->highest_bit,
817 "swap_info %d in list but !highest_bit\n",
818 si->type);
819 WARN(!(si->flags & SWP_WRITEOK),
820 "swap_info %d in list but !SWP_WRITEOK\n",
821 si->type);
822 plist_del(&si->avail_list, &swap_avail_head);
823 spin_unlock(&si->lock);
824 goto nextsi;
825 }
826 n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE,
827 n_goal, swp_entries);
828 spin_unlock(&si->lock);
829 if (n_ret)
830 goto check_out;
831 pr_debug("scan_swap_map of si %d failed to find offset\n",
832 si->type);
833
834 spin_lock(&swap_avail_lock);
835 nextsi:
836 /*
837 * if we got here, it's likely that si was almost full before,
838 * and since scan_swap_map() can drop the si->lock, multiple
839 * callers probably all tried to get a page from the same si
840 * and it filled up before we could get one; or, the si filled
841 * up between us dropping swap_avail_lock and taking si->lock.
842 * Since we dropped the swap_avail_lock, the swap_avail_head
843 * list may have been modified; so if next is still in the
844 * swap_avail_head list then try it, otherwise start over
845 * if we have not gotten any slots.
846 */
847 if (plist_node_empty(&next->avail_list))
848 goto start_over;
849 }
850
851 spin_unlock(&swap_avail_lock);
852
853 check_out:
854 if (n_ret < n_goal)
855 atomic_long_add((long) (n_goal-n_ret), &nr_swap_pages);
856 noswap:
857 return n_ret;
858 }
859
860 /* The only caller of this function is now suspend routine */
861 swp_entry_t get_swap_page_of_type(int type)
862 {
863 struct swap_info_struct *si;
864 pgoff_t offset;
865
866 si = swap_info[type];
867 spin_lock(&si->lock);
868 if (si && (si->flags & SWP_WRITEOK)) {
869 atomic_long_dec(&nr_swap_pages);
870 /* This is called for allocating swap entry, not cache */
871 offset = scan_swap_map(si, 1);
872 if (offset) {
873 spin_unlock(&si->lock);
874 return swp_entry(type, offset);
875 }
876 atomic_long_inc(&nr_swap_pages);
877 }
878 spin_unlock(&si->lock);
879 return (swp_entry_t) {0};
880 }
881
882 static struct swap_info_struct *__swap_info_get(swp_entry_t entry)
883 {
884 struct swap_info_struct *p;
885 unsigned long offset, type;
886
887 if (!entry.val)
888 goto out;
889 type = swp_type(entry);
890 if (type >= nr_swapfiles)
891 goto bad_nofile;
892 p = swap_info[type];
893 if (!(p->flags & SWP_USED))
894 goto bad_device;
895 offset = swp_offset(entry);
896 if (offset >= p->max)
897 goto bad_offset;
898 return p;
899
900 bad_offset:
901 pr_err("swap_info_get: %s%08lx\n", Bad_offset, entry.val);
902 goto out;
903 bad_device:
904 pr_err("swap_info_get: %s%08lx\n", Unused_file, entry.val);
905 goto out;
906 bad_nofile:
907 pr_err("swap_info_get: %s%08lx\n", Bad_file, entry.val);
908 out:
909 return NULL;
910 }
911
912 static struct swap_info_struct *_swap_info_get(swp_entry_t entry)
913 {
914 struct swap_info_struct *p;
915
916 p = __swap_info_get(entry);
917 if (!p)
918 goto out;
919 if (!p->swap_map[swp_offset(entry)])
920 goto bad_free;
921 return p;
922
923 bad_free:
924 pr_err("swap_info_get: %s%08lx\n", Unused_offset, entry.val);
925 goto out;
926 out:
927 return NULL;
928 }
929
930 static struct swap_info_struct *swap_info_get(swp_entry_t entry)
931 {
932 struct swap_info_struct *p;
933
934 p = _swap_info_get(entry);
935 if (p)
936 spin_lock(&p->lock);
937 return p;
938 }
939
940 static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry,
941 struct swap_info_struct *q)
942 {
943 struct swap_info_struct *p;
944
945 p = _swap_info_get(entry);
946
947 if (p != q) {
948 if (q != NULL)
949 spin_unlock(&q->lock);
950 if (p != NULL)
951 spin_lock(&p->lock);
952 }
953 return p;
954 }
955
956 static unsigned char __swap_entry_free(struct swap_info_struct *p,
957 swp_entry_t entry, unsigned char usage)
958 {
959 struct swap_cluster_info *ci;
960 unsigned long offset = swp_offset(entry);
961 unsigned char count;
962 unsigned char has_cache;
963
964 ci = lock_cluster_or_swap_info(p, offset);
965
966 count = p->swap_map[offset];
967
968 has_cache = count & SWAP_HAS_CACHE;
969 count &= ~SWAP_HAS_CACHE;
970
971 if (usage == SWAP_HAS_CACHE) {
972 VM_BUG_ON(!has_cache);
973 has_cache = 0;
974 } else if (count == SWAP_MAP_SHMEM) {
975 /*
976 * Or we could insist on shmem.c using a special
977 * swap_shmem_free() and free_shmem_swap_and_cache()...
978 */
979 count = 0;
980 } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
981 if (count == COUNT_CONTINUED) {
982 if (swap_count_continued(p, offset, count))
983 count = SWAP_MAP_MAX | COUNT_CONTINUED;
984 else
985 count = SWAP_MAP_MAX;
986 } else
987 count--;
988 }
989
990 usage = count | has_cache;
991 p->swap_map[offset] = usage ? : SWAP_HAS_CACHE;
992
993 unlock_cluster_or_swap_info(p, ci);
994
995 return usage;
996 }
997
998 static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry)
999 {
1000 struct swap_cluster_info *ci;
1001 unsigned long offset = swp_offset(entry);
1002 unsigned char count;
1003
1004 ci = lock_cluster(p, offset);
1005 count = p->swap_map[offset];
1006 VM_BUG_ON(count != SWAP_HAS_CACHE);
1007 p->swap_map[offset] = 0;
1008 dec_cluster_info_page(p, p->cluster_info, offset);
1009 unlock_cluster(ci);
1010
1011 mem_cgroup_uncharge_swap(entry);
1012 if (offset < p->lowest_bit)
1013 p->lowest_bit = offset;
1014 if (offset > p->highest_bit) {
1015 bool was_full = !p->highest_bit;
1016
1017 p->highest_bit = offset;
1018 if (was_full && (p->flags & SWP_WRITEOK)) {
1019 spin_lock(&swap_avail_lock);
1020 WARN_ON(!plist_node_empty(&p->avail_list));
1021 if (plist_node_empty(&p->avail_list))
1022 plist_add(&p->avail_list,
1023 &swap_avail_head);
1024 spin_unlock(&swap_avail_lock);
1025 }
1026 }
1027 atomic_long_inc(&nr_swap_pages);
1028 p->inuse_pages--;
1029 frontswap_invalidate_page(p->type, offset);
1030 if (p->flags & SWP_BLKDEV) {
1031 struct gendisk *disk = p->bdev->bd_disk;
1032
1033 if (disk->fops->swap_slot_free_notify)
1034 disk->fops->swap_slot_free_notify(p->bdev,
1035 offset);
1036 }
1037 }
1038
1039 /*
1040 * Caller has made sure that the swap device corresponding to entry
1041 * is still around or has not been recycled.
1042 */
1043 void swap_free(swp_entry_t entry)
1044 {
1045 struct swap_info_struct *p;
1046
1047 p = _swap_info_get(entry);
1048 if (p) {
1049 if (!__swap_entry_free(p, entry, 1))
1050 free_swap_slot(entry);
1051 }
1052 }
1053
1054 /*
1055 * Called after dropping swapcache to decrease refcnt to swap entries.
1056 */
1057 void swapcache_free(swp_entry_t entry)
1058 {
1059 struct swap_info_struct *p;
1060
1061 p = _swap_info_get(entry);
1062 if (p) {
1063 if (!__swap_entry_free(p, entry, SWAP_HAS_CACHE))
1064 free_swap_slot(entry);
1065 }
1066 }
1067
1068 void swapcache_free_entries(swp_entry_t *entries, int n)
1069 {
1070 struct swap_info_struct *p, *prev;
1071 int i;
1072
1073 if (n <= 0)
1074 return;
1075
1076 prev = NULL;
1077 p = NULL;
1078 for (i = 0; i < n; ++i) {
1079 p = swap_info_get_cont(entries[i], prev);
1080 if (p)
1081 swap_entry_free(p, entries[i]);
1082 prev = p;
1083 }
1084 if (p)
1085 spin_unlock(&p->lock);
1086 }
1087
1088 /*
1089 * How many references to page are currently swapped out?
1090 * This does not give an exact answer when swap count is continued,
1091 * but does include the high COUNT_CONTINUED flag to allow for that.
1092 */
1093 int page_swapcount(struct page *page)
1094 {
1095 int count = 0;
1096 struct swap_info_struct *p;
1097 struct swap_cluster_info *ci;
1098 swp_entry_t entry;
1099 unsigned long offset;
1100
1101 entry.val = page_private(page);
1102 p = _swap_info_get(entry);
1103 if (p) {
1104 offset = swp_offset(entry);
1105 ci = lock_cluster_or_swap_info(p, offset);
1106 count = swap_count(p->swap_map[offset]);
1107 unlock_cluster_or_swap_info(p, ci);
1108 }
1109 return count;
1110 }
1111
1112 static int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry)
1113 {
1114 int count = 0;
1115 pgoff_t offset = swp_offset(entry);
1116 struct swap_cluster_info *ci;
1117
1118 ci = lock_cluster_or_swap_info(si, offset);
1119 count = swap_count(si->swap_map[offset]);
1120 unlock_cluster_or_swap_info(si, ci);
1121 return count;
1122 }
1123
1124 /*
1125 * How many references to @entry are currently swapped out?
1126 * This does not give an exact answer when swap count is continued,
1127 * but does include the high COUNT_CONTINUED flag to allow for that.
1128 */
1129 int __swp_swapcount(swp_entry_t entry)
1130 {
1131 int count = 0;
1132 struct swap_info_struct *si;
1133
1134 si = __swap_info_get(entry);
1135 if (si)
1136 count = swap_swapcount(si, entry);
1137 return count;
1138 }
1139
1140 /*
1141 * How many references to @entry are currently swapped out?
1142 * This considers COUNT_CONTINUED so it returns exact answer.
1143 */
1144 int swp_swapcount(swp_entry_t entry)
1145 {
1146 int count, tmp_count, n;
1147 struct swap_info_struct *p;
1148 struct swap_cluster_info *ci;
1149 struct page *page;
1150 pgoff_t offset;
1151 unsigned char *map;
1152
1153 p = _swap_info_get(entry);
1154 if (!p)
1155 return 0;
1156
1157 offset = swp_offset(entry);
1158
1159 ci = lock_cluster_or_swap_info(p, offset);
1160
1161 count = swap_count(p->swap_map[offset]);
1162 if (!(count & COUNT_CONTINUED))
1163 goto out;
1164
1165 count &= ~COUNT_CONTINUED;
1166 n = SWAP_MAP_MAX + 1;
1167
1168 page = vmalloc_to_page(p->swap_map + offset);
1169 offset &= ~PAGE_MASK;
1170 VM_BUG_ON(page_private(page) != SWP_CONTINUED);
1171
1172 do {
1173 page = list_next_entry(page, lru);
1174 map = kmap_atomic(page);
1175 tmp_count = map[offset];
1176 kunmap_atomic(map);
1177
1178 count += (tmp_count & ~COUNT_CONTINUED) * n;
1179 n *= (SWAP_CONT_MAX + 1);
1180 } while (tmp_count & COUNT_CONTINUED);
1181 out:
1182 unlock_cluster_or_swap_info(p, ci);
1183 return count;
1184 }
1185
1186 /*
1187 * We can write to an anon page without COW if there are no other references
1188 * to it. And as a side-effect, free up its swap: because the old content
1189 * on disk will never be read, and seeking back there to write new content
1190 * later would only waste time away from clustering.
1191 *
1192 * NOTE: total_mapcount should not be relied upon by the caller if
1193 * reuse_swap_page() returns false, but it may be always overwritten
1194 * (see the other implementation for CONFIG_SWAP=n).
1195 */
1196 bool reuse_swap_page(struct page *page, int *total_mapcount)
1197 {
1198 int count;
1199
1200 VM_BUG_ON_PAGE(!PageLocked(page), page);
1201 if (unlikely(PageKsm(page)))
1202 return false;
1203 count = page_trans_huge_mapcount(page, total_mapcount);
1204 if (count <= 1 && PageSwapCache(page)) {
1205 count += page_swapcount(page);
1206 if (count != 1)
1207 goto out;
1208 if (!PageWriteback(page)) {
1209 delete_from_swap_cache(page);
1210 SetPageDirty(page);
1211 } else {
1212 swp_entry_t entry;
1213 struct swap_info_struct *p;
1214
1215 entry.val = page_private(page);
1216 p = swap_info_get(entry);
1217 if (p->flags & SWP_STABLE_WRITES) {
1218 spin_unlock(&p->lock);
1219 return false;
1220 }
1221 spin_unlock(&p->lock);
1222 }
1223 }
1224 out:
1225 return count <= 1;
1226 }
1227
1228 /*
1229 * If swap is getting full, or if there are no more mappings of this page,
1230 * then try_to_free_swap is called to free its swap space.
1231 */
1232 int try_to_free_swap(struct page *page)
1233 {
1234 VM_BUG_ON_PAGE(!PageLocked(page), page);
1235
1236 if (!PageSwapCache(page))
1237 return 0;
1238 if (PageWriteback(page))
1239 return 0;
1240 if (page_swapcount(page))
1241 return 0;
1242
1243 /*
1244 * Once hibernation has begun to create its image of memory,
1245 * there's a danger that one of the calls to try_to_free_swap()
1246 * - most probably a call from __try_to_reclaim_swap() while
1247 * hibernation is allocating its own swap pages for the image,
1248 * but conceivably even a call from memory reclaim - will free
1249 * the swap from a page which has already been recorded in the
1250 * image as a clean swapcache page, and then reuse its swap for
1251 * another page of the image. On waking from hibernation, the
1252 * original page might be freed under memory pressure, then
1253 * later read back in from swap, now with the wrong data.
1254 *
1255 * Hibernation suspends storage while it is writing the image
1256 * to disk so check that here.
1257 */
1258 if (pm_suspended_storage())
1259 return 0;
1260
1261 delete_from_swap_cache(page);
1262 SetPageDirty(page);
1263 return 1;
1264 }
1265
1266 /*
1267 * Free the swap entry like above, but also try to
1268 * free the page cache entry if it is the last user.
1269 */
1270 int free_swap_and_cache(swp_entry_t entry)
1271 {
1272 struct swap_info_struct *p;
1273 struct page *page = NULL;
1274 unsigned char count;
1275
1276 if (non_swap_entry(entry))
1277 return 1;
1278
1279 p = _swap_info_get(entry);
1280 if (p) {
1281 count = __swap_entry_free(p, entry, 1);
1282 if (count == SWAP_HAS_CACHE) {
1283 page = find_get_page(swap_address_space(entry),
1284 swp_offset(entry));
1285 if (page && !trylock_page(page)) {
1286 put_page(page);
1287 page = NULL;
1288 }
1289 } else if (!count)
1290 free_swap_slot(entry);
1291 }
1292 if (page) {
1293 /*
1294 * Not mapped elsewhere, or swap space full? Free it!
1295 * Also recheck PageSwapCache now page is locked (above).
1296 */
1297 if (PageSwapCache(page) && !PageWriteback(page) &&
1298 (!page_mapped(page) || mem_cgroup_swap_full(page)) &&
1299 !swap_swapcount(p, entry)) {
1300 delete_from_swap_cache(page);
1301 SetPageDirty(page);
1302 }
1303 unlock_page(page);
1304 put_page(page);
1305 }
1306 return p != NULL;
1307 }
1308
1309 #ifdef CONFIG_HIBERNATION
1310 /*
1311 * Find the swap type that corresponds to given device (if any).
1312 *
1313 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1314 * from 0, in which the swap header is expected to be located.
1315 *
1316 * This is needed for the suspend to disk (aka swsusp).
1317 */
1318 int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
1319 {
1320 struct block_device *bdev = NULL;
1321 int type;
1322
1323 if (device)
1324 bdev = bdget(device);
1325
1326 spin_lock(&swap_lock);
1327 for (type = 0; type < nr_swapfiles; type++) {
1328 struct swap_info_struct *sis = swap_info[type];
1329
1330 if (!(sis->flags & SWP_WRITEOK))
1331 continue;
1332
1333 if (!bdev) {
1334 if (bdev_p)
1335 *bdev_p = bdgrab(sis->bdev);
1336
1337 spin_unlock(&swap_lock);
1338 return type;
1339 }
1340 if (bdev == sis->bdev) {
1341 struct swap_extent *se = &sis->first_swap_extent;
1342
1343 if (se->start_block == offset) {
1344 if (bdev_p)
1345 *bdev_p = bdgrab(sis->bdev);
1346
1347 spin_unlock(&swap_lock);
1348 bdput(bdev);
1349 return type;
1350 }
1351 }
1352 }
1353 spin_unlock(&swap_lock);
1354 if (bdev)
1355 bdput(bdev);
1356
1357 return -ENODEV;
1358 }
1359
1360 /*
1361 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1362 * corresponding to given index in swap_info (swap type).
1363 */
1364 sector_t swapdev_block(int type, pgoff_t offset)
1365 {
1366 struct block_device *bdev;
1367
1368 if ((unsigned int)type >= nr_swapfiles)
1369 return 0;
1370 if (!(swap_info[type]->flags & SWP_WRITEOK))
1371 return 0;
1372 return map_swap_entry(swp_entry(type, offset), &bdev);
1373 }
1374
1375 /*
1376 * Return either the total number of swap pages of given type, or the number
1377 * of free pages of that type (depending on @free)
1378 *
1379 * This is needed for software suspend
1380 */
1381 unsigned int count_swap_pages(int type, int free)
1382 {
1383 unsigned int n = 0;
1384
1385 spin_lock(&swap_lock);
1386 if ((unsigned int)type < nr_swapfiles) {
1387 struct swap_info_struct *sis = swap_info[type];
1388
1389 spin_lock(&sis->lock);
1390 if (sis->flags & SWP_WRITEOK) {
1391 n = sis->pages;
1392 if (free)
1393 n -= sis->inuse_pages;
1394 }
1395 spin_unlock(&sis->lock);
1396 }
1397 spin_unlock(&swap_lock);
1398 return n;
1399 }
1400 #endif /* CONFIG_HIBERNATION */
1401
1402 static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte)
1403 {
1404 return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte);
1405 }
1406
1407 /*
1408 * No need to decide whether this PTE shares the swap entry with others,
1409 * just let do_wp_page work it out if a write is requested later - to
1410 * force COW, vm_page_prot omits write permission from any private vma.
1411 */
1412 static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
1413 unsigned long addr, swp_entry_t entry, struct page *page)
1414 {
1415 struct page *swapcache;
1416 struct mem_cgroup *memcg;
1417 spinlock_t *ptl;
1418 pte_t *pte;
1419 int ret = 1;
1420
1421 swapcache = page;
1422 page = ksm_might_need_to_copy(page, vma, addr);
1423 if (unlikely(!page))
1424 return -ENOMEM;
1425
1426 if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL,
1427 &memcg, false)) {
1428 ret = -ENOMEM;
1429 goto out_nolock;
1430 }
1431
1432 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1433 if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) {
1434 mem_cgroup_cancel_charge(page, memcg, false);
1435 ret = 0;
1436 goto out;
1437 }
1438
1439 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1440 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
1441 get_page(page);
1442 set_pte_at(vma->vm_mm, addr, pte,
1443 pte_mkold(mk_pte(page, vma->vm_page_prot)));
1444 if (page == swapcache) {
1445 page_add_anon_rmap(page, vma, addr, false);
1446 mem_cgroup_commit_charge(page, memcg, true, false);
1447 } else { /* ksm created a completely new copy */
1448 page_add_new_anon_rmap(page, vma, addr, false);
1449 mem_cgroup_commit_charge(page, memcg, false, false);
1450 lru_cache_add_active_or_unevictable(page, vma);
1451 }
1452 swap_free(entry);
1453 /*
1454 * Move the page to the active list so it is not
1455 * immediately swapped out again after swapon.
1456 */
1457 activate_page(page);
1458 out:
1459 pte_unmap_unlock(pte, ptl);
1460 out_nolock:
1461 if (page != swapcache) {
1462 unlock_page(page);
1463 put_page(page);
1464 }
1465 return ret;
1466 }
1467
1468 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1469 unsigned long addr, unsigned long end,
1470 swp_entry_t entry, struct page *page)
1471 {
1472 pte_t swp_pte = swp_entry_to_pte(entry);
1473 pte_t *pte;
1474 int ret = 0;
1475
1476 /*
1477 * We don't actually need pte lock while scanning for swp_pte: since
1478 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
1479 * page table while we're scanning; though it could get zapped, and on
1480 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
1481 * of unmatched parts which look like swp_pte, so unuse_pte must
1482 * recheck under pte lock. Scanning without pte lock lets it be
1483 * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
1484 */
1485 pte = pte_offset_map(pmd, addr);
1486 do {
1487 /*
1488 * swapoff spends a _lot_ of time in this loop!
1489 * Test inline before going to call unuse_pte.
1490 */
1491 if (unlikely(pte_same_as_swp(*pte, swp_pte))) {
1492 pte_unmap(pte);
1493 ret = unuse_pte(vma, pmd, addr, entry, page);
1494 if (ret)
1495 goto out;
1496 pte = pte_offset_map(pmd, addr);
1497 }
1498 } while (pte++, addr += PAGE_SIZE, addr != end);
1499 pte_unmap(pte - 1);
1500 out:
1501 return ret;
1502 }
1503
1504 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1505 unsigned long addr, unsigned long end,
1506 swp_entry_t entry, struct page *page)
1507 {
1508 pmd_t *pmd;
1509 unsigned long next;
1510 int ret;
1511
1512 pmd = pmd_offset(pud, addr);
1513 do {
1514 cond_resched();
1515 next = pmd_addr_end(addr, end);
1516 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
1517 continue;
1518 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
1519 if (ret)
1520 return ret;
1521 } while (pmd++, addr = next, addr != end);
1522 return 0;
1523 }
1524
1525 static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d,
1526 unsigned long addr, unsigned long end,
1527 swp_entry_t entry, struct page *page)
1528 {
1529 pud_t *pud;
1530 unsigned long next;
1531 int ret;
1532
1533 pud = pud_offset(p4d, addr);
1534 do {
1535 next = pud_addr_end(addr, end);
1536 if (pud_none_or_clear_bad(pud))
1537 continue;
1538 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
1539 if (ret)
1540 return ret;
1541 } while (pud++, addr = next, addr != end);
1542 return 0;
1543 }
1544
1545 static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd,
1546 unsigned long addr, unsigned long end,
1547 swp_entry_t entry, struct page *page)
1548 {
1549 p4d_t *p4d;
1550 unsigned long next;
1551 int ret;
1552
1553 p4d = p4d_offset(pgd, addr);
1554 do {
1555 next = p4d_addr_end(addr, end);
1556 if (p4d_none_or_clear_bad(p4d))
1557 continue;
1558 ret = unuse_pud_range(vma, p4d, addr, next, entry, page);
1559 if (ret)
1560 return ret;
1561 } while (p4d++, addr = next, addr != end);
1562 return 0;
1563 }
1564
1565 static int unuse_vma(struct vm_area_struct *vma,
1566 swp_entry_t entry, struct page *page)
1567 {
1568 pgd_t *pgd;
1569 unsigned long addr, end, next;
1570 int ret;
1571
1572 if (page_anon_vma(page)) {
1573 addr = page_address_in_vma(page, vma);
1574 if (addr == -EFAULT)
1575 return 0;
1576 else
1577 end = addr + PAGE_SIZE;
1578 } else {
1579 addr = vma->vm_start;
1580 end = vma->vm_end;
1581 }
1582
1583 pgd = pgd_offset(vma->vm_mm, addr);
1584 do {
1585 next = pgd_addr_end(addr, end);
1586 if (pgd_none_or_clear_bad(pgd))
1587 continue;
1588 ret = unuse_p4d_range(vma, pgd, addr, next, entry, page);
1589 if (ret)
1590 return ret;
1591 } while (pgd++, addr = next, addr != end);
1592 return 0;
1593 }
1594
1595 static int unuse_mm(struct mm_struct *mm,
1596 swp_entry_t entry, struct page *page)
1597 {
1598 struct vm_area_struct *vma;
1599 int ret = 0;
1600
1601 if (!down_read_trylock(&mm->mmap_sem)) {
1602 /*
1603 * Activate page so shrink_inactive_list is unlikely to unmap
1604 * its ptes while lock is dropped, so swapoff can make progress.
1605 */
1606 activate_page(page);
1607 unlock_page(page);
1608 down_read(&mm->mmap_sem);
1609 lock_page(page);
1610 }
1611 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1612 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
1613 break;
1614 cond_resched();
1615 }
1616 up_read(&mm->mmap_sem);
1617 return (ret < 0)? ret: 0;
1618 }
1619
1620 /*
1621 * Scan swap_map (or frontswap_map if frontswap parameter is true)
1622 * from current position to next entry still in use.
1623 * Recycle to start on reaching the end, returning 0 when empty.
1624 */
1625 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
1626 unsigned int prev, bool frontswap)
1627 {
1628 unsigned int max = si->max;
1629 unsigned int i = prev;
1630 unsigned char count;
1631
1632 /*
1633 * No need for swap_lock here: we're just looking
1634 * for whether an entry is in use, not modifying it; false
1635 * hits are okay, and sys_swapoff() has already prevented new
1636 * allocations from this area (while holding swap_lock).
1637 */
1638 for (;;) {
1639 if (++i >= max) {
1640 if (!prev) {
1641 i = 0;
1642 break;
1643 }
1644 /*
1645 * No entries in use at top of swap_map,
1646 * loop back to start and recheck there.
1647 */
1648 max = prev + 1;
1649 prev = 0;
1650 i = 1;
1651 }
1652 count = READ_ONCE(si->swap_map[i]);
1653 if (count && swap_count(count) != SWAP_MAP_BAD)
1654 if (!frontswap || frontswap_test(si, i))
1655 break;
1656 if ((i % LATENCY_LIMIT) == 0)
1657 cond_resched();
1658 }
1659 return i;
1660 }
1661
1662 /*
1663 * We completely avoid races by reading each swap page in advance,
1664 * and then search for the process using it. All the necessary
1665 * page table adjustments can then be made atomically.
1666 *
1667 * if the boolean frontswap is true, only unuse pages_to_unuse pages;
1668 * pages_to_unuse==0 means all pages; ignored if frontswap is false
1669 */
1670 int try_to_unuse(unsigned int type, bool frontswap,
1671 unsigned long pages_to_unuse)
1672 {
1673 struct swap_info_struct *si = swap_info[type];
1674 struct mm_struct *start_mm;
1675 volatile unsigned char *swap_map; /* swap_map is accessed without
1676 * locking. Mark it as volatile
1677 * to prevent compiler doing
1678 * something odd.
1679 */
1680 unsigned char swcount;
1681 struct page *page;
1682 swp_entry_t entry;
1683 unsigned int i = 0;
1684 int retval = 0;
1685
1686 /*
1687 * When searching mms for an entry, a good strategy is to
1688 * start at the first mm we freed the previous entry from
1689 * (though actually we don't notice whether we or coincidence
1690 * freed the entry). Initialize this start_mm with a hold.
1691 *
1692 * A simpler strategy would be to start at the last mm we
1693 * freed the previous entry from; but that would take less
1694 * advantage of mmlist ordering, which clusters forked mms
1695 * together, child after parent. If we race with dup_mmap(), we
1696 * prefer to resolve parent before child, lest we miss entries
1697 * duplicated after we scanned child: using last mm would invert
1698 * that.
1699 */
1700 start_mm = &init_mm;
1701 mmget(&init_mm);
1702
1703 /*
1704 * Keep on scanning until all entries have gone. Usually,
1705 * one pass through swap_map is enough, but not necessarily:
1706 * there are races when an instance of an entry might be missed.
1707 */
1708 while ((i = find_next_to_unuse(si, i, frontswap)) != 0) {
1709 if (signal_pending(current)) {
1710 retval = -EINTR;
1711 break;
1712 }
1713
1714 /*
1715 * Get a page for the entry, using the existing swap
1716 * cache page if there is one. Otherwise, get a clean
1717 * page and read the swap into it.
1718 */
1719 swap_map = &si->swap_map[i];
1720 entry = swp_entry(type, i);
1721 page = read_swap_cache_async(entry,
1722 GFP_HIGHUSER_MOVABLE, NULL, 0);
1723 if (!page) {
1724 /*
1725 * Either swap_duplicate() failed because entry
1726 * has been freed independently, and will not be
1727 * reused since sys_swapoff() already disabled
1728 * allocation from here, or alloc_page() failed.
1729 */
1730 swcount = *swap_map;
1731 /*
1732 * We don't hold lock here, so the swap entry could be
1733 * SWAP_MAP_BAD (when the cluster is discarding).
1734 * Instead of fail out, We can just skip the swap
1735 * entry because swapoff will wait for discarding
1736 * finish anyway.
1737 */
1738 if (!swcount || swcount == SWAP_MAP_BAD)
1739 continue;
1740 retval = -ENOMEM;
1741 break;
1742 }
1743
1744 /*
1745 * Don't hold on to start_mm if it looks like exiting.
1746 */
1747 if (atomic_read(&start_mm->mm_users) == 1) {
1748 mmput(start_mm);
1749 start_mm = &init_mm;
1750 mmget(&init_mm);
1751 }
1752
1753 /*
1754 * Wait for and lock page. When do_swap_page races with
1755 * try_to_unuse, do_swap_page can handle the fault much
1756 * faster than try_to_unuse can locate the entry. This
1757 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1758 * defer to do_swap_page in such a case - in some tests,
1759 * do_swap_page and try_to_unuse repeatedly compete.
1760 */
1761 wait_on_page_locked(page);
1762 wait_on_page_writeback(page);
1763 lock_page(page);
1764 wait_on_page_writeback(page);
1765
1766 /*
1767 * Remove all references to entry.
1768 */
1769 swcount = *swap_map;
1770 if (swap_count(swcount) == SWAP_MAP_SHMEM) {
1771 retval = shmem_unuse(entry, page);
1772 /* page has already been unlocked and released */
1773 if (retval < 0)
1774 break;
1775 continue;
1776 }
1777 if (swap_count(swcount) && start_mm != &init_mm)
1778 retval = unuse_mm(start_mm, entry, page);
1779
1780 if (swap_count(*swap_map)) {
1781 int set_start_mm = (*swap_map >= swcount);
1782 struct list_head *p = &start_mm->mmlist;
1783 struct mm_struct *new_start_mm = start_mm;
1784 struct mm_struct *prev_mm = start_mm;
1785 struct mm_struct *mm;
1786
1787 mmget(new_start_mm);
1788 mmget(prev_mm);
1789 spin_lock(&mmlist_lock);
1790 while (swap_count(*swap_map) && !retval &&
1791 (p = p->next) != &start_mm->mmlist) {
1792 mm = list_entry(p, struct mm_struct, mmlist);
1793 if (!mmget_not_zero(mm))
1794 continue;
1795 spin_unlock(&mmlist_lock);
1796 mmput(prev_mm);
1797 prev_mm = mm;
1798
1799 cond_resched();
1800
1801 swcount = *swap_map;
1802 if (!swap_count(swcount)) /* any usage ? */
1803 ;
1804 else if (mm == &init_mm)
1805 set_start_mm = 1;
1806 else
1807 retval = unuse_mm(mm, entry, page);
1808
1809 if (set_start_mm && *swap_map < swcount) {
1810 mmput(new_start_mm);
1811 mmget(mm);
1812 new_start_mm = mm;
1813 set_start_mm = 0;
1814 }
1815 spin_lock(&mmlist_lock);
1816 }
1817 spin_unlock(&mmlist_lock);
1818 mmput(prev_mm);
1819 mmput(start_mm);
1820 start_mm = new_start_mm;
1821 }
1822 if (retval) {
1823 unlock_page(page);
1824 put_page(page);
1825 break;
1826 }
1827
1828 /*
1829 * If a reference remains (rare), we would like to leave
1830 * the page in the swap cache; but try_to_unmap could
1831 * then re-duplicate the entry once we drop page lock,
1832 * so we might loop indefinitely; also, that page could
1833 * not be swapped out to other storage meanwhile. So:
1834 * delete from cache even if there's another reference,
1835 * after ensuring that the data has been saved to disk -
1836 * since if the reference remains (rarer), it will be
1837 * read from disk into another page. Splitting into two
1838 * pages would be incorrect if swap supported "shared
1839 * private" pages, but they are handled by tmpfs files.
1840 *
1841 * Given how unuse_vma() targets one particular offset
1842 * in an anon_vma, once the anon_vma has been determined,
1843 * this splitting happens to be just what is needed to
1844 * handle where KSM pages have been swapped out: re-reading
1845 * is unnecessarily slow, but we can fix that later on.
1846 */
1847 if (swap_count(*swap_map) &&
1848 PageDirty(page) && PageSwapCache(page)) {
1849 struct writeback_control wbc = {
1850 .sync_mode = WB_SYNC_NONE,
1851 };
1852
1853 swap_writepage(page, &wbc);
1854 lock_page(page);
1855 wait_on_page_writeback(page);
1856 }
1857
1858 /*
1859 * It is conceivable that a racing task removed this page from
1860 * swap cache just before we acquired the page lock at the top,
1861 * or while we dropped it in unuse_mm(). The page might even
1862 * be back in swap cache on another swap area: that we must not
1863 * delete, since it may not have been written out to swap yet.
1864 */
1865 if (PageSwapCache(page) &&
1866 likely(page_private(page) == entry.val))
1867 delete_from_swap_cache(page);
1868
1869 /*
1870 * So we could skip searching mms once swap count went
1871 * to 1, we did not mark any present ptes as dirty: must
1872 * mark page dirty so shrink_page_list will preserve it.
1873 */
1874 SetPageDirty(page);
1875 unlock_page(page);
1876 put_page(page);
1877
1878 /*
1879 * Make sure that we aren't completely killing
1880 * interactive performance.
1881 */
1882 cond_resched();
1883 if (frontswap && pages_to_unuse > 0) {
1884 if (!--pages_to_unuse)
1885 break;
1886 }
1887 }
1888
1889 mmput(start_mm);
1890 return retval;
1891 }
1892
1893 /*
1894 * After a successful try_to_unuse, if no swap is now in use, we know
1895 * we can empty the mmlist. swap_lock must be held on entry and exit.
1896 * Note that mmlist_lock nests inside swap_lock, and an mm must be
1897 * added to the mmlist just after page_duplicate - before would be racy.
1898 */
1899 static void drain_mmlist(void)
1900 {
1901 struct list_head *p, *next;
1902 unsigned int type;
1903
1904 for (type = 0; type < nr_swapfiles; type++)
1905 if (swap_info[type]->inuse_pages)
1906 return;
1907 spin_lock(&mmlist_lock);
1908 list_for_each_safe(p, next, &init_mm.mmlist)
1909 list_del_init(p);
1910 spin_unlock(&mmlist_lock);
1911 }
1912
1913 /*
1914 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
1915 * corresponds to page offset for the specified swap entry.
1916 * Note that the type of this function is sector_t, but it returns page offset
1917 * into the bdev, not sector offset.
1918 */
1919 static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
1920 {
1921 struct swap_info_struct *sis;
1922 struct swap_extent *start_se;
1923 struct swap_extent *se;
1924 pgoff_t offset;
1925
1926 sis = swap_info[swp_type(entry)];
1927 *bdev = sis->bdev;
1928
1929 offset = swp_offset(entry);
1930 start_se = sis->curr_swap_extent;
1931 se = start_se;
1932
1933 for ( ; ; ) {
1934 if (se->start_page <= offset &&
1935 offset < (se->start_page + se->nr_pages)) {
1936 return se->start_block + (offset - se->start_page);
1937 }
1938 se = list_next_entry(se, list);
1939 sis->curr_swap_extent = se;
1940 BUG_ON(se == start_se); /* It *must* be present */
1941 }
1942 }
1943
1944 /*
1945 * Returns the page offset into bdev for the specified page's swap entry.
1946 */
1947 sector_t map_swap_page(struct page *page, struct block_device **bdev)
1948 {
1949 swp_entry_t entry;
1950 entry.val = page_private(page);
1951 return map_swap_entry(entry, bdev);
1952 }
1953
1954 /*
1955 * Free all of a swapdev's extent information
1956 */
1957 static void destroy_swap_extents(struct swap_info_struct *sis)
1958 {
1959 while (!list_empty(&sis->first_swap_extent.list)) {
1960 struct swap_extent *se;
1961
1962 se = list_first_entry(&sis->first_swap_extent.list,
1963 struct swap_extent, list);
1964 list_del(&se->list);
1965 kfree(se);
1966 }
1967
1968 if (sis->flags & SWP_FILE) {
1969 struct file *swap_file = sis->swap_file;
1970 struct address_space *mapping = swap_file->f_mapping;
1971
1972 sis->flags &= ~SWP_FILE;
1973 mapping->a_ops->swap_deactivate(swap_file);
1974 }
1975 }
1976
1977 /*
1978 * Add a block range (and the corresponding page range) into this swapdev's
1979 * extent list. The extent list is kept sorted in page order.
1980 *
1981 * This function rather assumes that it is called in ascending page order.
1982 */
1983 int
1984 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1985 unsigned long nr_pages, sector_t start_block)
1986 {
1987 struct swap_extent *se;
1988 struct swap_extent *new_se;
1989 struct list_head *lh;
1990
1991 if (start_page == 0) {
1992 se = &sis->first_swap_extent;
1993 sis->curr_swap_extent = se;
1994 se->start_page = 0;
1995 se->nr_pages = nr_pages;
1996 se->start_block = start_block;
1997 return 1;
1998 } else {
1999 lh = sis->first_swap_extent.list.prev; /* Highest extent */
2000 se = list_entry(lh, struct swap_extent, list);
2001 BUG_ON(se->start_page + se->nr_pages != start_page);
2002 if (se->start_block + se->nr_pages == start_block) {
2003 /* Merge it */
2004 se->nr_pages += nr_pages;
2005 return 0;
2006 }
2007 }
2008
2009 /*
2010 * No merge. Insert a new extent, preserving ordering.
2011 */
2012 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
2013 if (new_se == NULL)
2014 return -ENOMEM;
2015 new_se->start_page = start_page;
2016 new_se->nr_pages = nr_pages;
2017 new_se->start_block = start_block;
2018
2019 list_add_tail(&new_se->list, &sis->first_swap_extent.list);
2020 return 1;
2021 }
2022
2023 /*
2024 * A `swap extent' is a simple thing which maps a contiguous range of pages
2025 * onto a contiguous range of disk blocks. An ordered list of swap extents
2026 * is built at swapon time and is then used at swap_writepage/swap_readpage
2027 * time for locating where on disk a page belongs.
2028 *
2029 * If the swapfile is an S_ISBLK block device, a single extent is installed.
2030 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2031 * swap files identically.
2032 *
2033 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2034 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
2035 * swapfiles are handled *identically* after swapon time.
2036 *
2037 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2038 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
2039 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
2040 * requirements, they are simply tossed out - we will never use those blocks
2041 * for swapping.
2042 *
2043 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
2044 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
2045 * which will scribble on the fs.
2046 *
2047 * The amount of disk space which a single swap extent represents varies.
2048 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
2049 * extents in the list. To avoid much list walking, we cache the previous
2050 * search location in `curr_swap_extent', and start new searches from there.
2051 * This is extremely effective. The average number of iterations in
2052 * map_swap_page() has been measured at about 0.3 per page. - akpm.
2053 */
2054 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
2055 {
2056 struct file *swap_file = sis->swap_file;
2057 struct address_space *mapping = swap_file->f_mapping;
2058 struct inode *inode = mapping->host;
2059 int ret;
2060
2061 if (S_ISBLK(inode->i_mode)) {
2062 ret = add_swap_extent(sis, 0, sis->max, 0);
2063 *span = sis->pages;
2064 return ret;
2065 }
2066
2067 if (mapping->a_ops->swap_activate) {
2068 ret = mapping->a_ops->swap_activate(sis, swap_file, span);
2069 if (!ret) {
2070 sis->flags |= SWP_FILE;
2071 ret = add_swap_extent(sis, 0, sis->max, 0);
2072 *span = sis->pages;
2073 }
2074 return ret;
2075 }
2076
2077 return generic_swapfile_activate(sis, swap_file, span);
2078 }
2079
2080 static void _enable_swap_info(struct swap_info_struct *p, int prio,
2081 unsigned char *swap_map,
2082 struct swap_cluster_info *cluster_info)
2083 {
2084 if (prio >= 0)
2085 p->prio = prio;
2086 else
2087 p->prio = --least_priority;
2088 /*
2089 * the plist prio is negated because plist ordering is
2090 * low-to-high, while swap ordering is high-to-low
2091 */
2092 p->list.prio = -p->prio;
2093 p->avail_list.prio = -p->prio;
2094 p->swap_map = swap_map;
2095 p->cluster_info = cluster_info;
2096 p->flags |= SWP_WRITEOK;
2097 atomic_long_add(p->pages, &nr_swap_pages);
2098 total_swap_pages += p->pages;
2099
2100 assert_spin_locked(&swap_lock);
2101 /*
2102 * both lists are plists, and thus priority ordered.
2103 * swap_active_head needs to be priority ordered for swapoff(),
2104 * which on removal of any swap_info_struct with an auto-assigned
2105 * (i.e. negative) priority increments the auto-assigned priority
2106 * of any lower-priority swap_info_structs.
2107 * swap_avail_head needs to be priority ordered for get_swap_page(),
2108 * which allocates swap pages from the highest available priority
2109 * swap_info_struct.
2110 */
2111 plist_add(&p->list, &swap_active_head);
2112 spin_lock(&swap_avail_lock);
2113 plist_add(&p->avail_list, &swap_avail_head);
2114 spin_unlock(&swap_avail_lock);
2115 }
2116
2117 static void enable_swap_info(struct swap_info_struct *p, int prio,
2118 unsigned char *swap_map,
2119 struct swap_cluster_info *cluster_info,
2120 unsigned long *frontswap_map)
2121 {
2122 frontswap_init(p->type, frontswap_map);
2123 spin_lock(&swap_lock);
2124 spin_lock(&p->lock);
2125 _enable_swap_info(p, prio, swap_map, cluster_info);
2126 spin_unlock(&p->lock);
2127 spin_unlock(&swap_lock);
2128 }
2129
2130 static void reinsert_swap_info(struct swap_info_struct *p)
2131 {
2132 spin_lock(&swap_lock);
2133 spin_lock(&p->lock);
2134 _enable_swap_info(p, p->prio, p->swap_map, p->cluster_info);
2135 spin_unlock(&p->lock);
2136 spin_unlock(&swap_lock);
2137 }
2138
2139 bool has_usable_swap(void)
2140 {
2141 bool ret = true;
2142
2143 spin_lock(&swap_lock);
2144 if (plist_head_empty(&swap_active_head))
2145 ret = false;
2146 spin_unlock(&swap_lock);
2147 return ret;
2148 }
2149
2150 SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
2151 {
2152 struct swap_info_struct *p = NULL;
2153 unsigned char *swap_map;
2154 struct swap_cluster_info *cluster_info;
2155 unsigned long *frontswap_map;
2156 struct file *swap_file, *victim;
2157 struct address_space *mapping;
2158 struct inode *inode;
2159 struct filename *pathname;
2160 int err, found = 0;
2161 unsigned int old_block_size;
2162
2163 if (!capable(CAP_SYS_ADMIN))
2164 return -EPERM;
2165
2166 BUG_ON(!current->mm);
2167
2168 pathname = getname(specialfile);
2169 if (IS_ERR(pathname))
2170 return PTR_ERR(pathname);
2171
2172 victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
2173 err = PTR_ERR(victim);
2174 if (IS_ERR(victim))
2175 goto out;
2176
2177 mapping = victim->f_mapping;
2178 spin_lock(&swap_lock);
2179 plist_for_each_entry(p, &swap_active_head, list) {
2180 if (p->flags & SWP_WRITEOK) {
2181 if (p->swap_file->f_mapping == mapping) {
2182 found = 1;
2183 break;
2184 }
2185 }
2186 }
2187 if (!found) {
2188 err = -EINVAL;
2189 spin_unlock(&swap_lock);
2190 goto out_dput;
2191 }
2192 if (!security_vm_enough_memory_mm(current->mm, p->pages))
2193 vm_unacct_memory(p->pages);
2194 else {
2195 err = -ENOMEM;
2196 spin_unlock(&swap_lock);
2197 goto out_dput;
2198 }
2199 spin_lock(&swap_avail_lock);
2200 plist_del(&p->avail_list, &swap_avail_head);
2201 spin_unlock(&swap_avail_lock);
2202 spin_lock(&p->lock);
2203 if (p->prio < 0) {
2204 struct swap_info_struct *si = p;
2205
2206 plist_for_each_entry_continue(si, &swap_active_head, list) {
2207 si->prio++;
2208 si->list.prio--;
2209 si->avail_list.prio--;
2210 }
2211 least_priority++;
2212 }
2213 plist_del(&p->list, &swap_active_head);
2214 atomic_long_sub(p->pages, &nr_swap_pages);
2215 total_swap_pages -= p->pages;
2216 p->flags &= ~SWP_WRITEOK;
2217 spin_unlock(&p->lock);
2218 spin_unlock(&swap_lock);
2219
2220 disable_swap_slots_cache_lock();
2221
2222 set_current_oom_origin();
2223 err = try_to_unuse(p->type, false, 0); /* force unuse all pages */
2224 clear_current_oom_origin();
2225
2226 if (err) {
2227 /* re-insert swap space back into swap_list */
2228 reinsert_swap_info(p);
2229 reenable_swap_slots_cache_unlock();
2230 goto out_dput;
2231 }
2232
2233 reenable_swap_slots_cache_unlock();
2234
2235 flush_work(&p->discard_work);
2236
2237 destroy_swap_extents(p);
2238 if (p->flags & SWP_CONTINUED)
2239 free_swap_count_continuations(p);
2240
2241 mutex_lock(&swapon_mutex);
2242 spin_lock(&swap_lock);
2243 spin_lock(&p->lock);
2244 drain_mmlist();
2245
2246 /* wait for anyone still in scan_swap_map */
2247 p->highest_bit = 0; /* cuts scans short */
2248 while (p->flags >= SWP_SCANNING) {
2249 spin_unlock(&p->lock);
2250 spin_unlock(&swap_lock);
2251 schedule_timeout_uninterruptible(1);
2252 spin_lock(&swap_lock);
2253 spin_lock(&p->lock);
2254 }
2255
2256 swap_file = p->swap_file;
2257 old_block_size = p->old_block_size;
2258 p->swap_file = NULL;
2259 p->max = 0;
2260 swap_map = p->swap_map;
2261 p->swap_map = NULL;
2262 cluster_info = p->cluster_info;
2263 p->cluster_info = NULL;
2264 frontswap_map = frontswap_map_get(p);
2265 spin_unlock(&p->lock);
2266 spin_unlock(&swap_lock);
2267 frontswap_invalidate_area(p->type);
2268 frontswap_map_set(p, NULL);
2269 mutex_unlock(&swapon_mutex);
2270 free_percpu(p->percpu_cluster);
2271 p->percpu_cluster = NULL;
2272 vfree(swap_map);
2273 kvfree(cluster_info);
2274 kvfree(frontswap_map);
2275 /* Destroy swap account information */
2276 swap_cgroup_swapoff(p->type);
2277 exit_swap_address_space(p->type);
2278
2279 inode = mapping->host;
2280 if (S_ISBLK(inode->i_mode)) {
2281 struct block_device *bdev = I_BDEV(inode);
2282 set_blocksize(bdev, old_block_size);
2283 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2284 } else {
2285 inode_lock(inode);
2286 inode->i_flags &= ~S_SWAPFILE;
2287 inode_unlock(inode);
2288 }
2289 filp_close(swap_file, NULL);
2290
2291 /*
2292 * Clear the SWP_USED flag after all resources are freed so that swapon
2293 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
2294 * not hold p->lock after we cleared its SWP_WRITEOK.
2295 */
2296 spin_lock(&swap_lock);
2297 p->flags = 0;
2298 spin_unlock(&swap_lock);
2299
2300 err = 0;
2301 atomic_inc(&proc_poll_event);
2302 wake_up_interruptible(&proc_poll_wait);
2303
2304 out_dput:
2305 filp_close(victim, NULL);
2306 out:
2307 putname(pathname);
2308 return err;
2309 }
2310
2311 #ifdef CONFIG_PROC_FS
2312 static unsigned swaps_poll(struct file *file, poll_table *wait)
2313 {
2314 struct seq_file *seq = file->private_data;
2315
2316 poll_wait(file, &proc_poll_wait, wait);
2317
2318 if (seq->poll_event != atomic_read(&proc_poll_event)) {
2319 seq->poll_event = atomic_read(&proc_poll_event);
2320 return POLLIN | POLLRDNORM | POLLERR | POLLPRI;
2321 }
2322
2323 return POLLIN | POLLRDNORM;
2324 }
2325
2326 /* iterator */
2327 static void *swap_start(struct seq_file *swap, loff_t *pos)
2328 {
2329 struct swap_info_struct *si;
2330 int type;
2331 loff_t l = *pos;
2332
2333 mutex_lock(&swapon_mutex);
2334
2335 if (!l)
2336 return SEQ_START_TOKEN;
2337
2338 for (type = 0; type < nr_swapfiles; type++) {
2339 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2340 si = swap_info[type];
2341 if (!(si->flags & SWP_USED) || !si->swap_map)
2342 continue;
2343 if (!--l)
2344 return si;
2345 }
2346
2347 return NULL;
2348 }
2349
2350 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2351 {
2352 struct swap_info_struct *si = v;
2353 int type;
2354
2355 if (v == SEQ_START_TOKEN)
2356 type = 0;
2357 else
2358 type = si->type + 1;
2359
2360 for (; type < nr_swapfiles; type++) {
2361 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2362 si = swap_info[type];
2363 if (!(si->flags & SWP_USED) || !si->swap_map)
2364 continue;
2365 ++*pos;
2366 return si;
2367 }
2368
2369 return NULL;
2370 }
2371
2372 static void swap_stop(struct seq_file *swap, void *v)
2373 {
2374 mutex_unlock(&swapon_mutex);
2375 }
2376
2377 static int swap_show(struct seq_file *swap, void *v)
2378 {
2379 struct swap_info_struct *si = v;
2380 struct file *file;
2381 int len;
2382
2383 if (si == SEQ_START_TOKEN) {
2384 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2385 return 0;
2386 }
2387
2388 file = si->swap_file;
2389 len = seq_file_path(swap, file, " \t\n\\");
2390 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
2391 len < 40 ? 40 - len : 1, " ",
2392 S_ISBLK(file_inode(file)->i_mode) ?
2393 "partition" : "file\t",
2394 si->pages << (PAGE_SHIFT - 10),
2395 si->inuse_pages << (PAGE_SHIFT - 10),
2396 si->prio);
2397 return 0;
2398 }
2399
2400 static const struct seq_operations swaps_op = {
2401 .start = swap_start,
2402 .next = swap_next,
2403 .stop = swap_stop,
2404 .show = swap_show
2405 };
2406
2407 static int swaps_open(struct inode *inode, struct file *file)
2408 {
2409 struct seq_file *seq;
2410 int ret;
2411
2412 ret = seq_open(file, &swaps_op);
2413 if (ret)
2414 return ret;
2415
2416 seq = file->private_data;
2417 seq->poll_event = atomic_read(&proc_poll_event);
2418 return 0;
2419 }
2420
2421 static const struct file_operations proc_swaps_operations = {
2422 .open = swaps_open,
2423 .read = seq_read,
2424 .llseek = seq_lseek,
2425 .release = seq_release,
2426 .poll = swaps_poll,
2427 };
2428
2429 static int __init procswaps_init(void)
2430 {
2431 proc_create("swaps", 0, NULL, &proc_swaps_operations);
2432 return 0;
2433 }
2434 __initcall(procswaps_init);
2435 #endif /* CONFIG_PROC_FS */
2436
2437 #ifdef MAX_SWAPFILES_CHECK
2438 static int __init max_swapfiles_check(void)
2439 {
2440 MAX_SWAPFILES_CHECK();
2441 return 0;
2442 }
2443 late_initcall(max_swapfiles_check);
2444 #endif
2445
2446 static struct swap_info_struct *alloc_swap_info(void)
2447 {
2448 struct swap_info_struct *p;
2449 unsigned int type;
2450
2451 p = kzalloc(sizeof(*p), GFP_KERNEL);
2452 if (!p)
2453 return ERR_PTR(-ENOMEM);
2454
2455 spin_lock(&swap_lock);
2456 for (type = 0; type < nr_swapfiles; type++) {
2457 if (!(swap_info[type]->flags & SWP_USED))
2458 break;
2459 }
2460 if (type >= MAX_SWAPFILES) {
2461 spin_unlock(&swap_lock);
2462 kfree(p);
2463 return ERR_PTR(-EPERM);
2464 }
2465 if (type >= nr_swapfiles) {
2466 p->type = type;
2467 swap_info[type] = p;
2468 /*
2469 * Write swap_info[type] before nr_swapfiles, in case a
2470 * racing procfs swap_start() or swap_next() is reading them.
2471 * (We never shrink nr_swapfiles, we never free this entry.)
2472 */
2473 smp_wmb();
2474 nr_swapfiles++;
2475 } else {
2476 kfree(p);
2477 p = swap_info[type];
2478 /*
2479 * Do not memset this entry: a racing procfs swap_next()
2480 * would be relying on p->type to remain valid.
2481 */
2482 }
2483 INIT_LIST_HEAD(&p->first_swap_extent.list);
2484 plist_node_init(&p->list, 0);
2485 plist_node_init(&p->avail_list, 0);
2486 p->flags = SWP_USED;
2487 spin_unlock(&swap_lock);
2488 spin_lock_init(&p->lock);
2489
2490 return p;
2491 }
2492
2493 static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2494 {
2495 int error;
2496
2497 if (S_ISBLK(inode->i_mode)) {
2498 p->bdev = bdgrab(I_BDEV(inode));
2499 error = blkdev_get(p->bdev,
2500 FMODE_READ | FMODE_WRITE | FMODE_EXCL, p);
2501 if (error < 0) {
2502 p->bdev = NULL;
2503 return error;
2504 }
2505 p->old_block_size = block_size(p->bdev);
2506 error = set_blocksize(p->bdev, PAGE_SIZE);
2507 if (error < 0)
2508 return error;
2509 p->flags |= SWP_BLKDEV;
2510 } else if (S_ISREG(inode->i_mode)) {
2511 p->bdev = inode->i_sb->s_bdev;
2512 inode_lock(inode);
2513 if (IS_SWAPFILE(inode))
2514 return -EBUSY;
2515 } else
2516 return -EINVAL;
2517
2518 return 0;
2519 }
2520
2521 static unsigned long read_swap_header(struct swap_info_struct *p,
2522 union swap_header *swap_header,
2523 struct inode *inode)
2524 {
2525 int i;
2526 unsigned long maxpages;
2527 unsigned long swapfilepages;
2528 unsigned long last_page;
2529
2530 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
2531 pr_err("Unable to find swap-space signature\n");
2532 return 0;
2533 }
2534
2535 /* swap partition endianess hack... */
2536 if (swab32(swap_header->info.version) == 1) {
2537 swab32s(&swap_header->info.version);
2538 swab32s(&swap_header->info.last_page);
2539 swab32s(&swap_header->info.nr_badpages);
2540 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2541 return 0;
2542 for (i = 0; i < swap_header->info.nr_badpages; i++)
2543 swab32s(&swap_header->info.badpages[i]);
2544 }
2545 /* Check the swap header's sub-version */
2546 if (swap_header->info.version != 1) {
2547 pr_warn("Unable to handle swap header version %d\n",
2548 swap_header->info.version);
2549 return 0;
2550 }
2551
2552 p->lowest_bit = 1;
2553 p->cluster_next = 1;
2554 p->cluster_nr = 0;
2555
2556 /*
2557 * Find out how many pages are allowed for a single swap
2558 * device. There are two limiting factors: 1) the number
2559 * of bits for the swap offset in the swp_entry_t type, and
2560 * 2) the number of bits in the swap pte as defined by the
2561 * different architectures. In order to find the
2562 * largest possible bit mask, a swap entry with swap type 0
2563 * and swap offset ~0UL is created, encoded to a swap pte,
2564 * decoded to a swp_entry_t again, and finally the swap
2565 * offset is extracted. This will mask all the bits from
2566 * the initial ~0UL mask that can't be encoded in either
2567 * the swp_entry_t or the architecture definition of a
2568 * swap pte.
2569 */
2570 maxpages = swp_offset(pte_to_swp_entry(
2571 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2572 last_page = swap_header->info.last_page;
2573 if (last_page > maxpages) {
2574 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
2575 maxpages << (PAGE_SHIFT - 10),
2576 last_page << (PAGE_SHIFT - 10));
2577 }
2578 if (maxpages > last_page) {
2579 maxpages = last_page + 1;
2580 /* p->max is an unsigned int: don't overflow it */
2581 if ((unsigned int)maxpages == 0)
2582 maxpages = UINT_MAX;
2583 }
2584 p->highest_bit = maxpages - 1;
2585
2586 if (!maxpages)
2587 return 0;
2588 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2589 if (swapfilepages && maxpages > swapfilepages) {
2590 pr_warn("Swap area shorter than signature indicates\n");
2591 return 0;
2592 }
2593 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
2594 return 0;
2595 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2596 return 0;
2597
2598 return maxpages;
2599 }
2600
2601 #define SWAP_CLUSTER_INFO_COLS \
2602 DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
2603 #define SWAP_CLUSTER_SPACE_COLS \
2604 DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
2605 #define SWAP_CLUSTER_COLS \
2606 max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
2607
2608 static int setup_swap_map_and_extents(struct swap_info_struct *p,
2609 union swap_header *swap_header,
2610 unsigned char *swap_map,
2611 struct swap_cluster_info *cluster_info,
2612 unsigned long maxpages,
2613 sector_t *span)
2614 {
2615 unsigned int j, k;
2616 unsigned int nr_good_pages;
2617 int nr_extents;
2618 unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
2619 unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS;
2620 unsigned long i, idx;
2621
2622 nr_good_pages = maxpages - 1; /* omit header page */
2623
2624 cluster_list_init(&p->free_clusters);
2625 cluster_list_init(&p->discard_clusters);
2626
2627 for (i = 0; i < swap_header->info.nr_badpages; i++) {
2628 unsigned int page_nr = swap_header->info.badpages[i];
2629 if (page_nr == 0 || page_nr > swap_header->info.last_page)
2630 return -EINVAL;
2631 if (page_nr < maxpages) {
2632 swap_map[page_nr] = SWAP_MAP_BAD;
2633 nr_good_pages--;
2634 /*
2635 * Haven't marked the cluster free yet, no list
2636 * operation involved
2637 */
2638 inc_cluster_info_page(p, cluster_info, page_nr);
2639 }
2640 }
2641
2642 /* Haven't marked the cluster free yet, no list operation involved */
2643 for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
2644 inc_cluster_info_page(p, cluster_info, i);
2645
2646 if (nr_good_pages) {
2647 swap_map[0] = SWAP_MAP_BAD;
2648 /*
2649 * Not mark the cluster free yet, no list
2650 * operation involved
2651 */
2652 inc_cluster_info_page(p, cluster_info, 0);
2653 p->max = maxpages;
2654 p->pages = nr_good_pages;
2655 nr_extents = setup_swap_extents(p, span);
2656 if (nr_extents < 0)
2657 return nr_extents;
2658 nr_good_pages = p->pages;
2659 }
2660 if (!nr_good_pages) {
2661 pr_warn("Empty swap-file\n");
2662 return -EINVAL;
2663 }
2664
2665 if (!cluster_info)
2666 return nr_extents;
2667
2668
2669 /*
2670 * Reduce false cache line sharing between cluster_info and
2671 * sharing same address space.
2672 */
2673 for (k = 0; k < SWAP_CLUSTER_COLS; k++) {
2674 j = (k + col) % SWAP_CLUSTER_COLS;
2675 for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) {
2676 idx = i * SWAP_CLUSTER_COLS + j;
2677 if (idx >= nr_clusters)
2678 continue;
2679 if (cluster_count(&cluster_info[idx]))
2680 continue;
2681 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
2682 cluster_list_add_tail(&p->free_clusters, cluster_info,
2683 idx);
2684 }
2685 }
2686 return nr_extents;
2687 }
2688
2689 /*
2690 * Helper to sys_swapon determining if a given swap
2691 * backing device queue supports DISCARD operations.
2692 */
2693 static bool swap_discardable(struct swap_info_struct *si)
2694 {
2695 struct request_queue *q = bdev_get_queue(si->bdev);
2696
2697 if (!q || !blk_queue_discard(q))
2698 return false;
2699
2700 return true;
2701 }
2702
2703 SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2704 {
2705 struct swap_info_struct *p;
2706 struct filename *name;
2707 struct file *swap_file = NULL;
2708 struct address_space *mapping;
2709 int prio;
2710 int error;
2711 union swap_header *swap_header;
2712 int nr_extents;
2713 sector_t span;
2714 unsigned long maxpages;
2715 unsigned char *swap_map = NULL;
2716 struct swap_cluster_info *cluster_info = NULL;
2717 unsigned long *frontswap_map = NULL;
2718 struct page *page = NULL;
2719 struct inode *inode = NULL;
2720
2721 if (swap_flags & ~SWAP_FLAGS_VALID)
2722 return -EINVAL;
2723
2724 if (!capable(CAP_SYS_ADMIN))
2725 return -EPERM;
2726
2727 p = alloc_swap_info();
2728 if (IS_ERR(p))
2729 return PTR_ERR(p);
2730
2731 INIT_WORK(&p->discard_work, swap_discard_work);
2732
2733 name = getname(specialfile);
2734 if (IS_ERR(name)) {
2735 error = PTR_ERR(name);
2736 name = NULL;
2737 goto bad_swap;
2738 }
2739 swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
2740 if (IS_ERR(swap_file)) {
2741 error = PTR_ERR(swap_file);
2742 swap_file = NULL;
2743 goto bad_swap;
2744 }
2745
2746 p->swap_file = swap_file;
2747 mapping = swap_file->f_mapping;
2748 inode = mapping->host;
2749
2750 /* If S_ISREG(inode->i_mode) will do inode_lock(inode); */
2751 error = claim_swapfile(p, inode);
2752 if (unlikely(error))
2753 goto bad_swap;
2754
2755 /*
2756 * Read the swap header.
2757 */
2758 if (!mapping->a_ops->readpage) {
2759 error = -EINVAL;
2760 goto bad_swap;
2761 }
2762 page = read_mapping_page(mapping, 0, swap_file);
2763 if (IS_ERR(page)) {
2764 error = PTR_ERR(page);
2765 goto bad_swap;
2766 }
2767 swap_header = kmap(page);
2768
2769 maxpages = read_swap_header(p, swap_header, inode);
2770 if (unlikely(!maxpages)) {
2771 error = -EINVAL;
2772 goto bad_swap;
2773 }
2774
2775 /* OK, set up the swap map and apply the bad block list */
2776 swap_map = vzalloc(maxpages);
2777 if (!swap_map) {
2778 error = -ENOMEM;
2779 goto bad_swap;
2780 }
2781
2782 if (bdi_cap_stable_pages_required(inode_to_bdi(inode)))
2783 p->flags |= SWP_STABLE_WRITES;
2784
2785 if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) {
2786 int cpu;
2787 unsigned long ci, nr_cluster;
2788
2789 p->flags |= SWP_SOLIDSTATE;
2790 /*
2791 * select a random position to start with to help wear leveling
2792 * SSD
2793 */
2794 p->cluster_next = 1 + (prandom_u32() % p->highest_bit);
2795 nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
2796
2797 cluster_info = kvzalloc(nr_cluster * sizeof(*cluster_info),
2798 GFP_KERNEL);
2799 if (!cluster_info) {
2800 error = -ENOMEM;
2801 goto bad_swap;
2802 }
2803
2804 for (ci = 0; ci < nr_cluster; ci++)
2805 spin_lock_init(&((cluster_info + ci)->lock));
2806
2807 p->percpu_cluster = alloc_percpu(struct percpu_cluster);
2808 if (!p->percpu_cluster) {
2809 error = -ENOMEM;
2810 goto bad_swap;
2811 }
2812 for_each_possible_cpu(cpu) {
2813 struct percpu_cluster *cluster;
2814 cluster = per_cpu_ptr(p->percpu_cluster, cpu);
2815 cluster_set_null(&cluster->index);
2816 }
2817 }
2818
2819 error = swap_cgroup_swapon(p->type, maxpages);
2820 if (error)
2821 goto bad_swap;
2822
2823 nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
2824 cluster_info, maxpages, &span);
2825 if (unlikely(nr_extents < 0)) {
2826 error = nr_extents;
2827 goto bad_swap;
2828 }
2829 /* frontswap enabled? set up bit-per-page map for frontswap */
2830 if (IS_ENABLED(CONFIG_FRONTSWAP))
2831 frontswap_map = kvzalloc(BITS_TO_LONGS(maxpages) * sizeof(long),
2832 GFP_KERNEL);
2833
2834 if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) {
2835 /*
2836 * When discard is enabled for swap with no particular
2837 * policy flagged, we set all swap discard flags here in
2838 * order to sustain backward compatibility with older
2839 * swapon(8) releases.
2840 */
2841 p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
2842 SWP_PAGE_DISCARD);
2843
2844 /*
2845 * By flagging sys_swapon, a sysadmin can tell us to
2846 * either do single-time area discards only, or to just
2847 * perform discards for released swap page-clusters.
2848 * Now it's time to adjust the p->flags accordingly.
2849 */
2850 if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
2851 p->flags &= ~SWP_PAGE_DISCARD;
2852 else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
2853 p->flags &= ~SWP_AREA_DISCARD;
2854
2855 /* issue a swapon-time discard if it's still required */
2856 if (p->flags & SWP_AREA_DISCARD) {
2857 int err = discard_swap(p);
2858 if (unlikely(err))
2859 pr_err("swapon: discard_swap(%p): %d\n",
2860 p, err);
2861 }
2862 }
2863
2864 error = init_swap_address_space(p->type, maxpages);
2865 if (error)
2866 goto bad_swap;
2867
2868 mutex_lock(&swapon_mutex);
2869 prio = -1;
2870 if (swap_flags & SWAP_FLAG_PREFER)
2871 prio =
2872 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
2873 enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
2874
2875 pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
2876 p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
2877 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
2878 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
2879 (p->flags & SWP_DISCARDABLE) ? "D" : "",
2880 (p->flags & SWP_AREA_DISCARD) ? "s" : "",
2881 (p->flags & SWP_PAGE_DISCARD) ? "c" : "",
2882 (frontswap_map) ? "FS" : "");
2883
2884 mutex_unlock(&swapon_mutex);
2885 atomic_inc(&proc_poll_event);
2886 wake_up_interruptible(&proc_poll_wait);
2887
2888 if (S_ISREG(inode->i_mode))
2889 inode->i_flags |= S_SWAPFILE;
2890 error = 0;
2891 goto out;
2892 bad_swap:
2893 free_percpu(p->percpu_cluster);
2894 p->percpu_cluster = NULL;
2895 if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
2896 set_blocksize(p->bdev, p->old_block_size);
2897 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2898 }
2899 destroy_swap_extents(p);
2900 swap_cgroup_swapoff(p->type);
2901 spin_lock(&swap_lock);
2902 p->swap_file = NULL;
2903 p->flags = 0;
2904 spin_unlock(&swap_lock);
2905 vfree(swap_map);
2906 vfree(cluster_info);
2907 if (swap_file) {
2908 if (inode && S_ISREG(inode->i_mode)) {
2909 inode_unlock(inode);
2910 inode = NULL;
2911 }
2912 filp_close(swap_file, NULL);
2913 }
2914 out:
2915 if (page && !IS_ERR(page)) {
2916 kunmap(page);
2917 put_page(page);
2918 }
2919 if (name)
2920 putname(name);
2921 if (inode && S_ISREG(inode->i_mode))
2922 inode_unlock(inode);
2923 if (!error)
2924 enable_swap_slots_cache();
2925 return error;
2926 }
2927
2928 void si_swapinfo(struct sysinfo *val)
2929 {
2930 unsigned int type;
2931 unsigned long nr_to_be_unused = 0;
2932
2933 spin_lock(&swap_lock);
2934 for (type = 0; type < nr_swapfiles; type++) {
2935 struct swap_info_struct *si = swap_info[type];
2936
2937 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
2938 nr_to_be_unused += si->inuse_pages;
2939 }
2940 val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
2941 val->totalswap = total_swap_pages + nr_to_be_unused;
2942 spin_unlock(&swap_lock);
2943 }
2944
2945 /*
2946 * Verify that a swap entry is valid and increment its swap map count.
2947 *
2948 * Returns error code in following case.
2949 * - success -> 0
2950 * - swp_entry is invalid -> EINVAL
2951 * - swp_entry is migration entry -> EINVAL
2952 * - swap-cache reference is requested but there is already one. -> EEXIST
2953 * - swap-cache reference is requested but the entry is not used. -> ENOENT
2954 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
2955 */
2956 static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
2957 {
2958 struct swap_info_struct *p;
2959 struct swap_cluster_info *ci;
2960 unsigned long offset, type;
2961 unsigned char count;
2962 unsigned char has_cache;
2963 int err = -EINVAL;
2964
2965 if (non_swap_entry(entry))
2966 goto out;
2967
2968 type = swp_type(entry);
2969 if (type >= nr_swapfiles)
2970 goto bad_file;
2971 p = swap_info[type];
2972 offset = swp_offset(entry);
2973 if (unlikely(offset >= p->max))
2974 goto out;
2975
2976 ci = lock_cluster_or_swap_info(p, offset);
2977
2978 count = p->swap_map[offset];
2979
2980 /*
2981 * swapin_readahead() doesn't check if a swap entry is valid, so the
2982 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
2983 */
2984 if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
2985 err = -ENOENT;
2986 goto unlock_out;
2987 }
2988
2989 has_cache = count & SWAP_HAS_CACHE;
2990 count &= ~SWAP_HAS_CACHE;
2991 err = 0;
2992
2993 if (usage == SWAP_HAS_CACHE) {
2994
2995 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
2996 if (!has_cache && count)
2997 has_cache = SWAP_HAS_CACHE;
2998 else if (has_cache) /* someone else added cache */
2999 err = -EEXIST;
3000 else /* no users remaining */
3001 err = -ENOENT;
3002
3003 } else if (count || has_cache) {
3004
3005 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
3006 count += usage;
3007 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
3008 err = -EINVAL;
3009 else if (swap_count_continued(p, offset, count))
3010 count = COUNT_CONTINUED;
3011 else
3012 err = -ENOMEM;
3013 } else
3014 err = -ENOENT; /* unused swap entry */
3015
3016 p->swap_map[offset] = count | has_cache;
3017
3018 unlock_out:
3019 unlock_cluster_or_swap_info(p, ci);
3020 out:
3021 return err;
3022
3023 bad_file:
3024 pr_err("swap_dup: %s%08lx\n", Bad_file, entry.val);
3025 goto out;
3026 }
3027
3028 /*
3029 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3030 * (in which case its reference count is never incremented).
3031 */
3032 void swap_shmem_alloc(swp_entry_t entry)
3033 {
3034 __swap_duplicate(entry, SWAP_MAP_SHMEM);
3035 }
3036
3037 /*
3038 * Increase reference count of swap entry by 1.
3039 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3040 * but could not be atomically allocated. Returns 0, just as if it succeeded,
3041 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3042 * might occur if a page table entry has got corrupted.
3043 */
3044 int swap_duplicate(swp_entry_t entry)
3045 {
3046 int err = 0;
3047
3048 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
3049 err = add_swap_count_continuation(entry, GFP_ATOMIC);
3050 return err;
3051 }
3052
3053 /*
3054 * @entry: swap entry for which we allocate swap cache.
3055 *
3056 * Called when allocating swap cache for existing swap entry,
3057 * This can return error codes. Returns 0 at success.
3058 * -EBUSY means there is a swap cache.
3059 * Note: return code is different from swap_duplicate().
3060 */
3061 int swapcache_prepare(swp_entry_t entry)
3062 {
3063 return __swap_duplicate(entry, SWAP_HAS_CACHE);
3064 }
3065
3066 struct swap_info_struct *page_swap_info(struct page *page)
3067 {
3068 swp_entry_t swap = { .val = page_private(page) };
3069 return swap_info[swp_type(swap)];
3070 }
3071
3072 /*
3073 * out-of-line __page_file_ methods to avoid include hell.
3074 */
3075 struct address_space *__page_file_mapping(struct page *page)
3076 {
3077 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
3078 return page_swap_info(page)->swap_file->f_mapping;
3079 }
3080 EXPORT_SYMBOL_GPL(__page_file_mapping);
3081
3082 pgoff_t __page_file_index(struct page *page)
3083 {
3084 swp_entry_t swap = { .val = page_private(page) };
3085 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
3086 return swp_offset(swap);
3087 }
3088 EXPORT_SYMBOL_GPL(__page_file_index);
3089
3090 /*
3091 * add_swap_count_continuation - called when a swap count is duplicated
3092 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3093 * page of the original vmalloc'ed swap_map, to hold the continuation count
3094 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
3095 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3096 *
3097 * These continuation pages are seldom referenced: the common paths all work
3098 * on the original swap_map, only referring to a continuation page when the
3099 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3100 *
3101 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3102 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3103 * can be called after dropping locks.
3104 */
3105 int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
3106 {
3107 struct swap_info_struct *si;
3108 struct swap_cluster_info *ci;
3109 struct page *head;
3110 struct page *page;
3111 struct page *list_page;
3112 pgoff_t offset;
3113 unsigned char count;
3114
3115 /*
3116 * When debugging, it's easier to use __GFP_ZERO here; but it's better
3117 * for latency not to zero a page while GFP_ATOMIC and holding locks.
3118 */
3119 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
3120
3121 si = swap_info_get(entry);
3122 if (!si) {
3123 /*
3124 * An acceptable race has occurred since the failing
3125 * __swap_duplicate(): the swap entry has been freed,
3126 * perhaps even the whole swap_map cleared for swapoff.
3127 */
3128 goto outer;
3129 }
3130
3131 offset = swp_offset(entry);
3132
3133 ci = lock_cluster(si, offset);
3134
3135 count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
3136
3137 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
3138 /*
3139 * The higher the swap count, the more likely it is that tasks
3140 * will race to add swap count continuation: we need to avoid
3141 * over-provisioning.
3142 */
3143 goto out;
3144 }
3145
3146 if (!page) {
3147 unlock_cluster(ci);
3148 spin_unlock(&si->lock);
3149 return -ENOMEM;
3150 }
3151
3152 /*
3153 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
3154 * no architecture is using highmem pages for kernel page tables: so it
3155 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
3156 */
3157 head = vmalloc_to_page(si->swap_map + offset);
3158 offset &= ~PAGE_MASK;
3159
3160 /*
3161 * Page allocation does not initialize the page's lru field,
3162 * but it does always reset its private field.
3163 */
3164 if (!page_private(head)) {
3165 BUG_ON(count & COUNT_CONTINUED);
3166 INIT_LIST_HEAD(&head->lru);
3167 set_page_private(head, SWP_CONTINUED);
3168 si->flags |= SWP_CONTINUED;
3169 }
3170
3171 list_for_each_entry(list_page, &head->lru, lru) {
3172 unsigned char *map;
3173
3174 /*
3175 * If the previous map said no continuation, but we've found
3176 * a continuation page, free our allocation and use this one.
3177 */
3178 if (!(count & COUNT_CONTINUED))
3179 goto out;
3180
3181 map = kmap_atomic(list_page) + offset;
3182 count = *map;
3183 kunmap_atomic(map);
3184
3185 /*
3186 * If this continuation count now has some space in it,
3187 * free our allocation and use this one.
3188 */
3189 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
3190 goto out;
3191 }
3192
3193 list_add_tail(&page->lru, &head->lru);
3194 page = NULL; /* now it's attached, don't free it */
3195 out:
3196 unlock_cluster(ci);
3197 spin_unlock(&si->lock);
3198 outer:
3199 if (page)
3200 __free_page(page);
3201 return 0;
3202 }
3203
3204 /*
3205 * swap_count_continued - when the original swap_map count is incremented
3206 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3207 * into, carry if so, or else fail until a new continuation page is allocated;
3208 * when the original swap_map count is decremented from 0 with continuation,
3209 * borrow from the continuation and report whether it still holds more.
3210 * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3211 * lock.
3212 */
3213 static bool swap_count_continued(struct swap_info_struct *si,
3214 pgoff_t offset, unsigned char count)
3215 {
3216 struct page *head;
3217 struct page *page;
3218 unsigned char *map;
3219
3220 head = vmalloc_to_page(si->swap_map + offset);
3221 if (page_private(head) != SWP_CONTINUED) {
3222 BUG_ON(count & COUNT_CONTINUED);
3223 return false; /* need to add count continuation */
3224 }
3225
3226 offset &= ~PAGE_MASK;
3227 page = list_entry(head->lru.next, struct page, lru);
3228 map = kmap_atomic(page) + offset;
3229
3230 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
3231 goto init_map; /* jump over SWAP_CONT_MAX checks */
3232
3233 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
3234 /*
3235 * Think of how you add 1 to 999
3236 */
3237 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
3238 kunmap_atomic(map);
3239 page = list_entry(page->lru.next, struct page, lru);
3240 BUG_ON(page == head);
3241 map = kmap_atomic(page) + offset;
3242 }
3243 if (*map == SWAP_CONT_MAX) {
3244 kunmap_atomic(map);
3245 page = list_entry(page->lru.next, struct page, lru);
3246 if (page == head)
3247 return false; /* add count continuation */
3248 map = kmap_atomic(page) + offset;
3249 init_map: *map = 0; /* we didn't zero the page */
3250 }
3251 *map += 1;
3252 kunmap_atomic(map);
3253 page = list_entry(page->lru.prev, struct page, lru);
3254 while (page != head) {
3255 map = kmap_atomic(page) + offset;
3256 *map = COUNT_CONTINUED;
3257 kunmap_atomic(map);
3258 page = list_entry(page->lru.prev, struct page, lru);
3259 }
3260 return true; /* incremented */
3261
3262 } else { /* decrementing */
3263 /*
3264 * Think of how you subtract 1 from 1000
3265 */
3266 BUG_ON(count != COUNT_CONTINUED);
3267 while (*map == COUNT_CONTINUED) {
3268 kunmap_atomic(map);
3269 page = list_entry(page->lru.next, struct page, lru);
3270 BUG_ON(page == head);
3271 map = kmap_atomic(page) + offset;
3272 }
3273 BUG_ON(*map == 0);
3274 *map -= 1;
3275 if (*map == 0)
3276 count = 0;
3277 kunmap_atomic(map);
3278 page = list_entry(page->lru.prev, struct page, lru);
3279 while (page != head) {
3280 map = kmap_atomic(page) + offset;
3281 *map = SWAP_CONT_MAX | count;
3282 count = COUNT_CONTINUED;
3283 kunmap_atomic(map);
3284 page = list_entry(page->lru.prev, struct page, lru);
3285 }
3286 return count == COUNT_CONTINUED;
3287 }
3288 }
3289
3290 /*
3291 * free_swap_count_continuations - swapoff free all the continuation pages
3292 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3293 */
3294 static void free_swap_count_continuations(struct swap_info_struct *si)
3295 {
3296 pgoff_t offset;
3297
3298 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
3299 struct page *head;
3300 head = vmalloc_to_page(si->swap_map + offset);
3301 if (page_private(head)) {
3302 struct page *page, *next;
3303
3304 list_for_each_entry_safe(page, next, &head->lru, lru) {
3305 list_del(&page->lru);
3306 __free_page(page);
3307 }
3308 }
3309 }
3310 }