~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/mm/swapfile.c

Version: ~ [ linux-5.13-rc5 ] ~ [ linux-5.12.9 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.42 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.124 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.193 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.235 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.271 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.271 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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 #include <linux/sort.h>
 41 
 42 #include <asm/pgtable.h>
 43 #include <asm/tlbflush.h>
 44 #include <linux/swapops.h>
 45 #include <linux/swap_cgroup.h>
 46 
 47 static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
 48                                  unsigned char);
 49 static void free_swap_count_continuations(struct swap_info_struct *);
 50 static sector_t map_swap_entry(swp_entry_t, struct block_device**);
 51 
 52 DEFINE_SPINLOCK(swap_lock);
 53 static unsigned int nr_swapfiles;
 54 atomic_long_t nr_swap_pages;
 55 /*
 56  * Some modules use swappable objects and may try to swap them out under
 57  * memory pressure (via the shrinker). Before doing so, they may wish to
 58  * check to see if any swap space is available.
 59  */
 60 EXPORT_SYMBOL_GPL(nr_swap_pages);
 61 /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
 62 long total_swap_pages;
 63 static int least_priority = -1;
 64 
 65 static const char Bad_file[] = "Bad swap file entry ";
 66 static const char Unused_file[] = "Unused swap file entry ";
 67 static const char Bad_offset[] = "Bad swap offset entry ";
 68 static const char Unused_offset[] = "Unused swap offset entry ";
 69 
 70 /*
 71  * all active swap_info_structs
 72  * protected with swap_lock, and ordered by priority.
 73  */
 74 PLIST_HEAD(swap_active_head);
 75 
 76 /*
 77  * all available (active, not full) swap_info_structs
 78  * protected with swap_avail_lock, ordered by priority.
 79  * This is used by get_swap_page() instead of swap_active_head
 80  * because swap_active_head includes all swap_info_structs,
 81  * but get_swap_page() doesn't need to look at full ones.
 82  * This uses its own lock instead of swap_lock because when a
 83  * swap_info_struct changes between not-full/full, it needs to
 84  * add/remove itself to/from this list, but the swap_info_struct->lock
 85  * is held and the locking order requires swap_lock to be taken
 86  * before any swap_info_struct->lock.
 87  */
 88 struct plist_head *swap_avail_heads;
 89 static DEFINE_SPINLOCK(swap_avail_lock);
 90 
 91 struct swap_info_struct *swap_info[MAX_SWAPFILES];
 92 
 93 static DEFINE_MUTEX(swapon_mutex);
 94 
 95 static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
 96 /* Activity counter to indicate that a swapon or swapoff has occurred */
 97 static atomic_t proc_poll_event = ATOMIC_INIT(0);
 98 
 99 atomic_t nr_rotate_swap = ATOMIC_INIT(0);
100 
101 static inline unsigned char swap_count(unsigned char ent)
102 {
103         return ent & ~SWAP_HAS_CACHE;   /* may include SWAP_HAS_CONT flag */
104 }
105 
106 /* returns 1 if swap entry is freed */
107 static int
108 __try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset)
109 {
110         swp_entry_t entry = swp_entry(si->type, offset);
111         struct page *page;
112         int ret = 0;
113 
114         page = find_get_page(swap_address_space(entry), swp_offset(entry));
115         if (!page)
116                 return 0;
117         /*
118          * This function is called from scan_swap_map() and it's called
119          * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
120          * We have to use trylock for avoiding deadlock. This is a special
121          * case and you should use try_to_free_swap() with explicit lock_page()
122          * in usual operations.
123          */
124         if (trylock_page(page)) {
125                 ret = try_to_free_swap(page);
126                 unlock_page(page);
127         }
128         put_page(page);
129         return ret;
130 }
131 
132 /*
133  * swapon tell device that all the old swap contents can be discarded,
134  * to allow the swap device to optimize its wear-levelling.
135  */
136 static int discard_swap(struct swap_info_struct *si)
137 {
138         struct swap_extent *se;
139         sector_t start_block;
140         sector_t nr_blocks;
141         int err = 0;
142 
143         /* Do not discard the swap header page! */
144         se = &si->first_swap_extent;
145         start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
146         nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
147         if (nr_blocks) {
148                 err = blkdev_issue_discard(si->bdev, start_block,
149                                 nr_blocks, GFP_KERNEL, 0);
150                 if (err)
151                         return err;
152                 cond_resched();
153         }
154 
155         list_for_each_entry(se, &si->first_swap_extent.list, list) {
156                 start_block = se->start_block << (PAGE_SHIFT - 9);
157                 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
158 
159                 err = blkdev_issue_discard(si->bdev, start_block,
160                                 nr_blocks, GFP_KERNEL, 0);
161                 if (err)
162                         break;
163 
164                 cond_resched();
165         }
166         return err;             /* That will often be -EOPNOTSUPP */
167 }
168 
169 /*
170  * swap allocation tell device that a cluster of swap can now be discarded,
171  * to allow the swap device to optimize its wear-levelling.
172  */
173 static void discard_swap_cluster(struct swap_info_struct *si,
174                                  pgoff_t start_page, pgoff_t nr_pages)
175 {
176         struct swap_extent *se = si->curr_swap_extent;
177         int found_extent = 0;
178 
179         while (nr_pages) {
180                 if (se->start_page <= start_page &&
181                     start_page < se->start_page + se->nr_pages) {
182                         pgoff_t offset = start_page - se->start_page;
183                         sector_t start_block = se->start_block + offset;
184                         sector_t nr_blocks = se->nr_pages - offset;
185 
186                         if (nr_blocks > nr_pages)
187                                 nr_blocks = nr_pages;
188                         start_page += nr_blocks;
189                         nr_pages -= nr_blocks;
190 
191                         if (!found_extent++)
192                                 si->curr_swap_extent = se;
193 
194                         start_block <<= PAGE_SHIFT - 9;
195                         nr_blocks <<= PAGE_SHIFT - 9;
196                         if (blkdev_issue_discard(si->bdev, start_block,
197                                     nr_blocks, GFP_NOIO, 0))
198                                 break;
199                 }
200 
201                 se = list_next_entry(se, list);
202         }
203 }
204 
205 #ifdef CONFIG_THP_SWAP
206 #define SWAPFILE_CLUSTER        HPAGE_PMD_NR
207 #else
208 #define SWAPFILE_CLUSTER        256
209 #endif
210 #define LATENCY_LIMIT           256
211 
212 static inline void cluster_set_flag(struct swap_cluster_info *info,
213         unsigned int flag)
214 {
215         info->flags = flag;
216 }
217 
218 static inline unsigned int cluster_count(struct swap_cluster_info *info)
219 {
220         return info->data;
221 }
222 
223 static inline void cluster_set_count(struct swap_cluster_info *info,
224                                      unsigned int c)
225 {
226         info->data = c;
227 }
228 
229 static inline void cluster_set_count_flag(struct swap_cluster_info *info,
230                                          unsigned int c, unsigned int f)
231 {
232         info->flags = f;
233         info->data = c;
234 }
235 
236 static inline unsigned int cluster_next(struct swap_cluster_info *info)
237 {
238         return info->data;
239 }
240 
241 static inline void cluster_set_next(struct swap_cluster_info *info,
242                                     unsigned int n)
243 {
244         info->data = n;
245 }
246 
247 static inline void cluster_set_next_flag(struct swap_cluster_info *info,
248                                          unsigned int n, unsigned int f)
249 {
250         info->flags = f;
251         info->data = n;
252 }
253 
254 static inline bool cluster_is_free(struct swap_cluster_info *info)
255 {
256         return info->flags & CLUSTER_FLAG_FREE;
257 }
258 
259 static inline bool cluster_is_null(struct swap_cluster_info *info)
260 {
261         return info->flags & CLUSTER_FLAG_NEXT_NULL;
262 }
263 
264 static inline void cluster_set_null(struct swap_cluster_info *info)
265 {
266         info->flags = CLUSTER_FLAG_NEXT_NULL;
267         info->data = 0;
268 }
269 
270 static inline bool cluster_is_huge(struct swap_cluster_info *info)
271 {
272         return info->flags & CLUSTER_FLAG_HUGE;
273 }
274 
275 static inline void cluster_clear_huge(struct swap_cluster_info *info)
276 {
277         info->flags &= ~CLUSTER_FLAG_HUGE;
278 }
279 
280 static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si,
281                                                      unsigned long offset)
282 {
283         struct swap_cluster_info *ci;
284 
285         ci = si->cluster_info;
286         if (ci) {
287                 ci += offset / SWAPFILE_CLUSTER;
288                 spin_lock(&ci->lock);
289         }
290         return ci;
291 }
292 
293 static inline void unlock_cluster(struct swap_cluster_info *ci)
294 {
295         if (ci)
296                 spin_unlock(&ci->lock);
297 }
298 
299 static inline struct swap_cluster_info *lock_cluster_or_swap_info(
300         struct swap_info_struct *si,
301         unsigned long offset)
302 {
303         struct swap_cluster_info *ci;
304 
305         ci = lock_cluster(si, offset);
306         if (!ci)
307                 spin_lock(&si->lock);
308 
309         return ci;
310 }
311 
312 static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si,
313                                                struct swap_cluster_info *ci)
314 {
315         if (ci)
316                 unlock_cluster(ci);
317         else
318                 spin_unlock(&si->lock);
319 }
320 
321 static inline bool cluster_list_empty(struct swap_cluster_list *list)
322 {
323         return cluster_is_null(&list->head);
324 }
325 
326 static inline unsigned int cluster_list_first(struct swap_cluster_list *list)
327 {
328         return cluster_next(&list->head);
329 }
330 
331 static void cluster_list_init(struct swap_cluster_list *list)
332 {
333         cluster_set_null(&list->head);
334         cluster_set_null(&list->tail);
335 }
336 
337 static void cluster_list_add_tail(struct swap_cluster_list *list,
338                                   struct swap_cluster_info *ci,
339                                   unsigned int idx)
340 {
341         if (cluster_list_empty(list)) {
342                 cluster_set_next_flag(&list->head, idx, 0);
343                 cluster_set_next_flag(&list->tail, idx, 0);
344         } else {
345                 struct swap_cluster_info *ci_tail;
346                 unsigned int tail = cluster_next(&list->tail);
347 
348                 /*
349                  * Nested cluster lock, but both cluster locks are
350                  * only acquired when we held swap_info_struct->lock
351                  */
352                 ci_tail = ci + tail;
353                 spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING);
354                 cluster_set_next(ci_tail, idx);
355                 spin_unlock(&ci_tail->lock);
356                 cluster_set_next_flag(&list->tail, idx, 0);
357         }
358 }
359 
360 static unsigned int cluster_list_del_first(struct swap_cluster_list *list,
361                                            struct swap_cluster_info *ci)
362 {
363         unsigned int idx;
364 
365         idx = cluster_next(&list->head);
366         if (cluster_next(&list->tail) == idx) {
367                 cluster_set_null(&list->head);
368                 cluster_set_null(&list->tail);
369         } else
370                 cluster_set_next_flag(&list->head,
371                                       cluster_next(&ci[idx]), 0);
372 
373         return idx;
374 }
375 
376 /* Add a cluster to discard list and schedule it to do discard */
377 static void swap_cluster_schedule_discard(struct swap_info_struct *si,
378                 unsigned int idx)
379 {
380         /*
381          * If scan_swap_map() can't find a free cluster, it will check
382          * si->swap_map directly. To make sure the discarding cluster isn't
383          * taken by scan_swap_map(), mark the swap entries bad (occupied). It
384          * will be cleared after discard
385          */
386         memset(si->swap_map + idx * SWAPFILE_CLUSTER,
387                         SWAP_MAP_BAD, SWAPFILE_CLUSTER);
388 
389         cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx);
390 
391         schedule_work(&si->discard_work);
392 }
393 
394 static void __free_cluster(struct swap_info_struct *si, unsigned long idx)
395 {
396         struct swap_cluster_info *ci = si->cluster_info;
397 
398         cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE);
399         cluster_list_add_tail(&si->free_clusters, ci, idx);
400 }
401 
402 /*
403  * Doing discard actually. After a cluster discard is finished, the cluster
404  * will be added to free cluster list. caller should hold si->lock.
405 */
406 static void swap_do_scheduled_discard(struct swap_info_struct *si)
407 {
408         struct swap_cluster_info *info, *ci;
409         unsigned int idx;
410 
411         info = si->cluster_info;
412 
413         while (!cluster_list_empty(&si->discard_clusters)) {
414                 idx = cluster_list_del_first(&si->discard_clusters, info);
415                 spin_unlock(&si->lock);
416 
417                 discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
418                                 SWAPFILE_CLUSTER);
419 
420                 spin_lock(&si->lock);
421                 ci = lock_cluster(si, idx * SWAPFILE_CLUSTER);
422                 __free_cluster(si, idx);
423                 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
424                                 0, SWAPFILE_CLUSTER);
425                 unlock_cluster(ci);
426         }
427 }
428 
429 static void swap_discard_work(struct work_struct *work)
430 {
431         struct swap_info_struct *si;
432 
433         si = container_of(work, struct swap_info_struct, discard_work);
434 
435         spin_lock(&si->lock);
436         swap_do_scheduled_discard(si);
437         spin_unlock(&si->lock);
438 }
439 
440 static void alloc_cluster(struct swap_info_struct *si, unsigned long idx)
441 {
442         struct swap_cluster_info *ci = si->cluster_info;
443 
444         VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx);
445         cluster_list_del_first(&si->free_clusters, ci);
446         cluster_set_count_flag(ci + idx, 0, 0);
447 }
448 
449 static void free_cluster(struct swap_info_struct *si, unsigned long idx)
450 {
451         struct swap_cluster_info *ci = si->cluster_info + idx;
452 
453         VM_BUG_ON(cluster_count(ci) != 0);
454         /*
455          * If the swap is discardable, prepare discard the cluster
456          * instead of free it immediately. The cluster will be freed
457          * after discard.
458          */
459         if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
460             (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
461                 swap_cluster_schedule_discard(si, idx);
462                 return;
463         }
464 
465         __free_cluster(si, idx);
466 }
467 
468 /*
469  * The cluster corresponding to page_nr will be used. The cluster will be
470  * removed from free cluster list and its usage counter will be increased.
471  */
472 static void inc_cluster_info_page(struct swap_info_struct *p,
473         struct swap_cluster_info *cluster_info, unsigned long page_nr)
474 {
475         unsigned long idx = page_nr / SWAPFILE_CLUSTER;
476 
477         if (!cluster_info)
478                 return;
479         if (cluster_is_free(&cluster_info[idx]))
480                 alloc_cluster(p, idx);
481 
482         VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
483         cluster_set_count(&cluster_info[idx],
484                 cluster_count(&cluster_info[idx]) + 1);
485 }
486 
487 /*
488  * The cluster corresponding to page_nr decreases one usage. If the usage
489  * counter becomes 0, which means no page in the cluster is in using, we can
490  * optionally discard the cluster and add it to free cluster list.
491  */
492 static void dec_cluster_info_page(struct swap_info_struct *p,
493         struct swap_cluster_info *cluster_info, unsigned long page_nr)
494 {
495         unsigned long idx = page_nr / SWAPFILE_CLUSTER;
496 
497         if (!cluster_info)
498                 return;
499 
500         VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
501         cluster_set_count(&cluster_info[idx],
502                 cluster_count(&cluster_info[idx]) - 1);
503 
504         if (cluster_count(&cluster_info[idx]) == 0)
505                 free_cluster(p, idx);
506 }
507 
508 /*
509  * It's possible scan_swap_map() uses a free cluster in the middle of free
510  * cluster list. Avoiding such abuse to avoid list corruption.
511  */
512 static bool
513 scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
514         unsigned long offset)
515 {
516         struct percpu_cluster *percpu_cluster;
517         bool conflict;
518 
519         offset /= SWAPFILE_CLUSTER;
520         conflict = !cluster_list_empty(&si->free_clusters) &&
521                 offset != cluster_list_first(&si->free_clusters) &&
522                 cluster_is_free(&si->cluster_info[offset]);
523 
524         if (!conflict)
525                 return false;
526 
527         percpu_cluster = this_cpu_ptr(si->percpu_cluster);
528         cluster_set_null(&percpu_cluster->index);
529         return true;
530 }
531 
532 /*
533  * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
534  * might involve allocating a new cluster for current CPU too.
535  */
536 static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
537         unsigned long *offset, unsigned long *scan_base)
538 {
539         struct percpu_cluster *cluster;
540         struct swap_cluster_info *ci;
541         bool found_free;
542         unsigned long tmp, max;
543 
544 new_cluster:
545         cluster = this_cpu_ptr(si->percpu_cluster);
546         if (cluster_is_null(&cluster->index)) {
547                 if (!cluster_list_empty(&si->free_clusters)) {
548                         cluster->index = si->free_clusters.head;
549                         cluster->next = cluster_next(&cluster->index) *
550                                         SWAPFILE_CLUSTER;
551                 } else if (!cluster_list_empty(&si->discard_clusters)) {
552                         /*
553                          * we don't have free cluster but have some clusters in
554                          * discarding, do discard now and reclaim them
555                          */
556                         swap_do_scheduled_discard(si);
557                         *scan_base = *offset = si->cluster_next;
558                         goto new_cluster;
559                 } else
560                         return false;
561         }
562 
563         found_free = false;
564 
565         /*
566          * Other CPUs can use our cluster if they can't find a free cluster,
567          * check if there is still free entry in the cluster
568          */
569         tmp = cluster->next;
570         max = min_t(unsigned long, si->max,
571                     (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER);
572         if (tmp >= max) {
573                 cluster_set_null(&cluster->index);
574                 goto new_cluster;
575         }
576         ci = lock_cluster(si, tmp);
577         while (tmp < max) {
578                 if (!si->swap_map[tmp]) {
579                         found_free = true;
580                         break;
581                 }
582                 tmp++;
583         }
584         unlock_cluster(ci);
585         if (!found_free) {
586                 cluster_set_null(&cluster->index);
587                 goto new_cluster;
588         }
589         cluster->next = tmp + 1;
590         *offset = tmp;
591         *scan_base = tmp;
592         return found_free;
593 }
594 
595 static void __del_from_avail_list(struct swap_info_struct *p)
596 {
597         int nid;
598 
599         for_each_node(nid)
600                 plist_del(&p->avail_lists[nid], &swap_avail_heads[nid]);
601 }
602 
603 static void del_from_avail_list(struct swap_info_struct *p)
604 {
605         spin_lock(&swap_avail_lock);
606         __del_from_avail_list(p);
607         spin_unlock(&swap_avail_lock);
608 }
609 
610 static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset,
611                              unsigned int nr_entries)
612 {
613         unsigned int end = offset + nr_entries - 1;
614 
615         if (offset == si->lowest_bit)
616                 si->lowest_bit += nr_entries;
617         if (end == si->highest_bit)
618                 si->highest_bit -= nr_entries;
619         si->inuse_pages += nr_entries;
620         if (si->inuse_pages == si->pages) {
621                 si->lowest_bit = si->max;
622                 si->highest_bit = 0;
623                 del_from_avail_list(si);
624         }
625 }
626 
627 static void add_to_avail_list(struct swap_info_struct *p)
628 {
629         int nid;
630 
631         spin_lock(&swap_avail_lock);
632         for_each_node(nid) {
633                 WARN_ON(!plist_node_empty(&p->avail_lists[nid]));
634                 plist_add(&p->avail_lists[nid], &swap_avail_heads[nid]);
635         }
636         spin_unlock(&swap_avail_lock);
637 }
638 
639 static void swap_range_free(struct swap_info_struct *si, unsigned long offset,
640                             unsigned int nr_entries)
641 {
642         unsigned long end = offset + nr_entries - 1;
643         void (*swap_slot_free_notify)(struct block_device *, unsigned long);
644 
645         if (offset < si->lowest_bit)
646                 si->lowest_bit = offset;
647         if (end > si->highest_bit) {
648                 bool was_full = !si->highest_bit;
649 
650                 si->highest_bit = end;
651                 if (was_full && (si->flags & SWP_WRITEOK))
652                         add_to_avail_list(si);
653         }
654         atomic_long_add(nr_entries, &nr_swap_pages);
655         si->inuse_pages -= nr_entries;
656         if (si->flags & SWP_BLKDEV)
657                 swap_slot_free_notify =
658                         si->bdev->bd_disk->fops->swap_slot_free_notify;
659         else
660                 swap_slot_free_notify = NULL;
661         while (offset <= end) {
662                 frontswap_invalidate_page(si->type, offset);
663                 if (swap_slot_free_notify)
664                         swap_slot_free_notify(si->bdev, offset);
665                 offset++;
666         }
667 }
668 
669 static int scan_swap_map_slots(struct swap_info_struct *si,
670                                unsigned char usage, int nr,
671                                swp_entry_t slots[])
672 {
673         struct swap_cluster_info *ci;
674         unsigned long offset;
675         unsigned long scan_base;
676         unsigned long last_in_cluster = 0;
677         int latency_ration = LATENCY_LIMIT;
678         int n_ret = 0;
679 
680         if (nr > SWAP_BATCH)
681                 nr = SWAP_BATCH;
682 
683         /*
684          * We try to cluster swap pages by allocating them sequentially
685          * in swap.  Once we've allocated SWAPFILE_CLUSTER pages this
686          * way, however, we resort to first-free allocation, starting
687          * a new cluster.  This prevents us from scattering swap pages
688          * all over the entire swap partition, so that we reduce
689          * overall disk seek times between swap pages.  -- sct
690          * But we do now try to find an empty cluster.  -Andrea
691          * And we let swap pages go all over an SSD partition.  Hugh
692          */
693 
694         si->flags += SWP_SCANNING;
695         scan_base = offset = si->cluster_next;
696 
697         /* SSD algorithm */
698         if (si->cluster_info) {
699                 if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
700                         goto checks;
701                 else
702                         goto scan;
703         }
704 
705         if (unlikely(!si->cluster_nr--)) {
706                 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
707                         si->cluster_nr = SWAPFILE_CLUSTER - 1;
708                         goto checks;
709                 }
710 
711                 spin_unlock(&si->lock);
712 
713                 /*
714                  * If seek is expensive, start searching for new cluster from
715                  * start of partition, to minimize the span of allocated swap.
716                  * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
717                  * case, just handled by scan_swap_map_try_ssd_cluster() above.
718                  */
719                 scan_base = offset = si->lowest_bit;
720                 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
721 
722                 /* Locate the first empty (unaligned) cluster */
723                 for (; last_in_cluster <= si->highest_bit; offset++) {
724                         if (si->swap_map[offset])
725                                 last_in_cluster = offset + SWAPFILE_CLUSTER;
726                         else if (offset == last_in_cluster) {
727                                 spin_lock(&si->lock);
728                                 offset -= SWAPFILE_CLUSTER - 1;
729                                 si->cluster_next = offset;
730                                 si->cluster_nr = SWAPFILE_CLUSTER - 1;
731                                 goto checks;
732                         }
733                         if (unlikely(--latency_ration < 0)) {
734                                 cond_resched();
735                                 latency_ration = LATENCY_LIMIT;
736                         }
737                 }
738 
739                 offset = scan_base;
740                 spin_lock(&si->lock);
741                 si->cluster_nr = SWAPFILE_CLUSTER - 1;
742         }
743 
744 checks:
745         if (si->cluster_info) {
746                 while (scan_swap_map_ssd_cluster_conflict(si, offset)) {
747                 /* take a break if we already got some slots */
748                         if (n_ret)
749                                 goto done;
750                         if (!scan_swap_map_try_ssd_cluster(si, &offset,
751                                                         &scan_base))
752                                 goto scan;
753                 }
754         }
755         if (!(si->flags & SWP_WRITEOK))
756                 goto no_page;
757         if (!si->highest_bit)
758                 goto no_page;
759         if (offset > si->highest_bit)
760                 scan_base = offset = si->lowest_bit;
761 
762         ci = lock_cluster(si, offset);
763         /* reuse swap entry of cache-only swap if not busy. */
764         if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
765                 int swap_was_freed;
766                 unlock_cluster(ci);
767                 spin_unlock(&si->lock);
768                 swap_was_freed = __try_to_reclaim_swap(si, offset);
769                 spin_lock(&si->lock);
770                 /* entry was freed successfully, try to use this again */
771                 if (swap_was_freed)
772                         goto checks;
773                 goto scan; /* check next one */
774         }
775 
776         if (si->swap_map[offset]) {
777                 unlock_cluster(ci);
778                 if (!n_ret)
779                         goto scan;
780                 else
781                         goto done;
782         }
783         si->swap_map[offset] = usage;
784         inc_cluster_info_page(si, si->cluster_info, offset);
785         unlock_cluster(ci);
786 
787         swap_range_alloc(si, offset, 1);
788         si->cluster_next = offset + 1;
789         slots[n_ret++] = swp_entry(si->type, offset);
790 
791         /* got enough slots or reach max slots? */
792         if ((n_ret == nr) || (offset >= si->highest_bit))
793                 goto done;
794 
795         /* search for next available slot */
796 
797         /* time to take a break? */
798         if (unlikely(--latency_ration < 0)) {
799                 if (n_ret)
800                         goto done;
801                 spin_unlock(&si->lock);
802                 cond_resched();
803                 spin_lock(&si->lock);
804                 latency_ration = LATENCY_LIMIT;
805         }
806 
807         /* try to get more slots in cluster */
808         if (si->cluster_info) {
809                 if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
810                         goto checks;
811                 else
812                         goto done;
813         }
814         /* non-ssd case */
815         ++offset;
816 
817         /* non-ssd case, still more slots in cluster? */
818         if (si->cluster_nr && !si->swap_map[offset]) {
819                 --si->cluster_nr;
820                 goto checks;
821         }
822 
823 done:
824         si->flags -= SWP_SCANNING;
825         return n_ret;
826 
827 scan:
828         spin_unlock(&si->lock);
829         while (++offset <= si->highest_bit) {
830                 if (!si->swap_map[offset]) {
831                         spin_lock(&si->lock);
832                         goto checks;
833                 }
834                 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
835                         spin_lock(&si->lock);
836                         goto checks;
837                 }
838                 if (unlikely(--latency_ration < 0)) {
839                         cond_resched();
840                         latency_ration = LATENCY_LIMIT;
841                 }
842         }
843         offset = si->lowest_bit;
844         while (offset < scan_base) {
845                 if (!si->swap_map[offset]) {
846                         spin_lock(&si->lock);
847                         goto checks;
848                 }
849                 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
850                         spin_lock(&si->lock);
851                         goto checks;
852                 }
853                 if (unlikely(--latency_ration < 0)) {
854                         cond_resched();
855                         latency_ration = LATENCY_LIMIT;
856                 }
857                 offset++;
858         }
859         spin_lock(&si->lock);
860 
861 no_page:
862         si->flags -= SWP_SCANNING;
863         return n_ret;
864 }
865 
866 #ifdef CONFIG_THP_SWAP
867 static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot)
868 {
869         unsigned long idx;
870         struct swap_cluster_info *ci;
871         unsigned long offset, i;
872         unsigned char *map;
873 
874         if (cluster_list_empty(&si->free_clusters))
875                 return 0;
876 
877         idx = cluster_list_first(&si->free_clusters);
878         offset = idx * SWAPFILE_CLUSTER;
879         ci = lock_cluster(si, offset);
880         alloc_cluster(si, idx);
881         cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE);
882 
883         map = si->swap_map + offset;
884         for (i = 0; i < SWAPFILE_CLUSTER; i++)
885                 map[i] = SWAP_HAS_CACHE;
886         unlock_cluster(ci);
887         swap_range_alloc(si, offset, SWAPFILE_CLUSTER);
888         *slot = swp_entry(si->type, offset);
889 
890         return 1;
891 }
892 
893 static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx)
894 {
895         unsigned long offset = idx * SWAPFILE_CLUSTER;
896         struct swap_cluster_info *ci;
897 
898         ci = lock_cluster(si, offset);
899         cluster_set_count_flag(ci, 0, 0);
900         free_cluster(si, idx);
901         unlock_cluster(ci);
902         swap_range_free(si, offset, SWAPFILE_CLUSTER);
903 }
904 #else
905 static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot)
906 {
907         VM_WARN_ON_ONCE(1);
908         return 0;
909 }
910 #endif /* CONFIG_THP_SWAP */
911 
912 static unsigned long scan_swap_map(struct swap_info_struct *si,
913                                    unsigned char usage)
914 {
915         swp_entry_t entry;
916         int n_ret;
917 
918         n_ret = scan_swap_map_slots(si, usage, 1, &entry);
919 
920         if (n_ret)
921                 return swp_offset(entry);
922         else
923                 return 0;
924 
925 }
926 
927 int get_swap_pages(int n_goal, bool cluster, swp_entry_t swp_entries[])
928 {
929         unsigned long nr_pages = cluster ? SWAPFILE_CLUSTER : 1;
930         struct swap_info_struct *si, *next;
931         long avail_pgs;
932         int n_ret = 0;
933         int node;
934 
935         /* Only single cluster request supported */
936         WARN_ON_ONCE(n_goal > 1 && cluster);
937 
938         avail_pgs = atomic_long_read(&nr_swap_pages) / nr_pages;
939         if (avail_pgs <= 0)
940                 goto noswap;
941 
942         if (n_goal > SWAP_BATCH)
943                 n_goal = SWAP_BATCH;
944 
945         if (n_goal > avail_pgs)
946                 n_goal = avail_pgs;
947 
948         atomic_long_sub(n_goal * nr_pages, &nr_swap_pages);
949 
950         spin_lock(&swap_avail_lock);
951 
952 start_over:
953         node = numa_node_id();
954         plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) {
955                 /* requeue si to after same-priority siblings */
956                 plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]);
957                 spin_unlock(&swap_avail_lock);
958                 spin_lock(&si->lock);
959                 if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) {
960                         spin_lock(&swap_avail_lock);
961                         if (plist_node_empty(&si->avail_lists[node])) {
962                                 spin_unlock(&si->lock);
963                                 goto nextsi;
964                         }
965                         WARN(!si->highest_bit,
966                              "swap_info %d in list but !highest_bit\n",
967                              si->type);
968                         WARN(!(si->flags & SWP_WRITEOK),
969                              "swap_info %d in list but !SWP_WRITEOK\n",
970                              si->type);
971                         __del_from_avail_list(si);
972                         spin_unlock(&si->lock);
973                         goto nextsi;
974                 }
975                 if (cluster) {
976                         if (!(si->flags & SWP_FILE))
977                                 n_ret = swap_alloc_cluster(si, swp_entries);
978                 } else
979                         n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE,
980                                                     n_goal, swp_entries);
981                 spin_unlock(&si->lock);
982                 if (n_ret || cluster)
983                         goto check_out;
984                 pr_debug("scan_swap_map of si %d failed to find offset\n",
985                         si->type);
986 
987                 spin_lock(&swap_avail_lock);
988 nextsi:
989                 /*
990                  * if we got here, it's likely that si was almost full before,
991                  * and since scan_swap_map() can drop the si->lock, multiple
992                  * callers probably all tried to get a page from the same si
993                  * and it filled up before we could get one; or, the si filled
994                  * up between us dropping swap_avail_lock and taking si->lock.
995                  * Since we dropped the swap_avail_lock, the swap_avail_head
996                  * list may have been modified; so if next is still in the
997                  * swap_avail_head list then try it, otherwise start over
998                  * if we have not gotten any slots.
999                  */
1000                 if (plist_node_empty(&next->avail_lists[node]))
1001                         goto start_over;
1002         }
1003 
1004         spin_unlock(&swap_avail_lock);
1005 
1006 check_out:
1007         if (n_ret < n_goal)
1008                 atomic_long_add((long)(n_goal - n_ret) * nr_pages,
1009                                 &nr_swap_pages);
1010 noswap:
1011         return n_ret;
1012 }
1013 
1014 /* The only caller of this function is now suspend routine */
1015 swp_entry_t get_swap_page_of_type(int type)
1016 {
1017         struct swap_info_struct *si;
1018         pgoff_t offset;
1019 
1020         si = swap_info[type];
1021         spin_lock(&si->lock);
1022         if (si && (si->flags & SWP_WRITEOK)) {
1023                 atomic_long_dec(&nr_swap_pages);
1024                 /* This is called for allocating swap entry, not cache */
1025                 offset = scan_swap_map(si, 1);
1026                 if (offset) {
1027                         spin_unlock(&si->lock);
1028                         return swp_entry(type, offset);
1029                 }
1030                 atomic_long_inc(&nr_swap_pages);
1031         }
1032         spin_unlock(&si->lock);
1033         return (swp_entry_t) {0};
1034 }
1035 
1036 static struct swap_info_struct *__swap_info_get(swp_entry_t entry)
1037 {
1038         struct swap_info_struct *p;
1039         unsigned long offset, type;
1040 
1041         if (!entry.val)
1042                 goto out;
1043         type = swp_type(entry);
1044         if (type >= nr_swapfiles)
1045                 goto bad_nofile;
1046         p = swap_info[type];
1047         if (!(p->flags & SWP_USED))
1048                 goto bad_device;
1049         offset = swp_offset(entry);
1050         if (offset >= p->max)
1051                 goto bad_offset;
1052         return p;
1053 
1054 bad_offset:
1055         pr_err("swap_info_get: %s%08lx\n", Bad_offset, entry.val);
1056         goto out;
1057 bad_device:
1058         pr_err("swap_info_get: %s%08lx\n", Unused_file, entry.val);
1059         goto out;
1060 bad_nofile:
1061         pr_err("swap_info_get: %s%08lx\n", Bad_file, entry.val);
1062 out:
1063         return NULL;
1064 }
1065 
1066 static struct swap_info_struct *_swap_info_get(swp_entry_t entry)
1067 {
1068         struct swap_info_struct *p;
1069 
1070         p = __swap_info_get(entry);
1071         if (!p)
1072                 goto out;
1073         if (!p->swap_map[swp_offset(entry)])
1074                 goto bad_free;
1075         return p;
1076 
1077 bad_free:
1078         pr_err("swap_info_get: %s%08lx\n", Unused_offset, entry.val);
1079         goto out;
1080 out:
1081         return NULL;
1082 }
1083 
1084 static struct swap_info_struct *swap_info_get(swp_entry_t entry)
1085 {
1086         struct swap_info_struct *p;
1087 
1088         p = _swap_info_get(entry);
1089         if (p)
1090                 spin_lock(&p->lock);
1091         return p;
1092 }
1093 
1094 static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry,
1095                                         struct swap_info_struct *q)
1096 {
1097         struct swap_info_struct *p;
1098 
1099         p = _swap_info_get(entry);
1100 
1101         if (p != q) {
1102                 if (q != NULL)
1103                         spin_unlock(&q->lock);
1104                 if (p != NULL)
1105                         spin_lock(&p->lock);
1106         }
1107         return p;
1108 }
1109 
1110 static unsigned char __swap_entry_free(struct swap_info_struct *p,
1111                                        swp_entry_t entry, unsigned char usage)
1112 {
1113         struct swap_cluster_info *ci;
1114         unsigned long offset = swp_offset(entry);
1115         unsigned char count;
1116         unsigned char has_cache;
1117 
1118         ci = lock_cluster_or_swap_info(p, offset);
1119 
1120         count = p->swap_map[offset];
1121 
1122         has_cache = count & SWAP_HAS_CACHE;
1123         count &= ~SWAP_HAS_CACHE;
1124 
1125         if (usage == SWAP_HAS_CACHE) {
1126                 VM_BUG_ON(!has_cache);
1127                 has_cache = 0;
1128         } else if (count == SWAP_MAP_SHMEM) {
1129                 /*
1130                  * Or we could insist on shmem.c using a special
1131                  * swap_shmem_free() and free_shmem_swap_and_cache()...
1132                  */
1133                 count = 0;
1134         } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
1135                 if (count == COUNT_CONTINUED) {
1136                         if (swap_count_continued(p, offset, count))
1137                                 count = SWAP_MAP_MAX | COUNT_CONTINUED;
1138                         else
1139                                 count = SWAP_MAP_MAX;
1140                 } else
1141                         count--;
1142         }
1143 
1144         usage = count | has_cache;
1145         p->swap_map[offset] = usage ? : SWAP_HAS_CACHE;
1146 
1147         unlock_cluster_or_swap_info(p, ci);
1148 
1149         return usage;
1150 }
1151 
1152 static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry)
1153 {
1154         struct swap_cluster_info *ci;
1155         unsigned long offset = swp_offset(entry);
1156         unsigned char count;
1157 
1158         ci = lock_cluster(p, offset);
1159         count = p->swap_map[offset];
1160         VM_BUG_ON(count != SWAP_HAS_CACHE);
1161         p->swap_map[offset] = 0;
1162         dec_cluster_info_page(p, p->cluster_info, offset);
1163         unlock_cluster(ci);
1164 
1165         mem_cgroup_uncharge_swap(entry, 1);
1166         swap_range_free(p, offset, 1);
1167 }
1168 
1169 /*
1170  * Caller has made sure that the swap device corresponding to entry
1171  * is still around or has not been recycled.
1172  */
1173 void swap_free(swp_entry_t entry)
1174 {
1175         struct swap_info_struct *p;
1176 
1177         p = _swap_info_get(entry);
1178         if (p) {
1179                 if (!__swap_entry_free(p, entry, 1))
1180                         free_swap_slot(entry);
1181         }
1182 }
1183 
1184 /*
1185  * Called after dropping swapcache to decrease refcnt to swap entries.
1186  */
1187 static void swapcache_free(swp_entry_t entry)
1188 {
1189         struct swap_info_struct *p;
1190 
1191         p = _swap_info_get(entry);
1192         if (p) {
1193                 if (!__swap_entry_free(p, entry, SWAP_HAS_CACHE))
1194                         free_swap_slot(entry);
1195         }
1196 }
1197 
1198 #ifdef CONFIG_THP_SWAP
1199 static void swapcache_free_cluster(swp_entry_t entry)
1200 {
1201         unsigned long offset = swp_offset(entry);
1202         unsigned long idx = offset / SWAPFILE_CLUSTER;
1203         struct swap_cluster_info *ci;
1204         struct swap_info_struct *si;
1205         unsigned char *map;
1206         unsigned int i, free_entries = 0;
1207         unsigned char val;
1208 
1209         si = _swap_info_get(entry);
1210         if (!si)
1211                 return;
1212 
1213         ci = lock_cluster(si, offset);
1214         VM_BUG_ON(!cluster_is_huge(ci));
1215         map = si->swap_map + offset;
1216         for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1217                 val = map[i];
1218                 VM_BUG_ON(!(val & SWAP_HAS_CACHE));
1219                 if (val == SWAP_HAS_CACHE)
1220                         free_entries++;
1221         }
1222         if (!free_entries) {
1223                 for (i = 0; i < SWAPFILE_CLUSTER; i++)
1224                         map[i] &= ~SWAP_HAS_CACHE;
1225         }
1226         cluster_clear_huge(ci);
1227         unlock_cluster(ci);
1228         if (free_entries == SWAPFILE_CLUSTER) {
1229                 spin_lock(&si->lock);
1230                 ci = lock_cluster(si, offset);
1231                 memset(map, 0, SWAPFILE_CLUSTER);
1232                 unlock_cluster(ci);
1233                 mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER);
1234                 swap_free_cluster(si, idx);
1235                 spin_unlock(&si->lock);
1236         } else if (free_entries) {
1237                 for (i = 0; i < SWAPFILE_CLUSTER; i++, entry.val++) {
1238                         if (!__swap_entry_free(si, entry, SWAP_HAS_CACHE))
1239                                 free_swap_slot(entry);
1240                 }
1241         }
1242 }
1243 
1244 int split_swap_cluster(swp_entry_t entry)
1245 {
1246         struct swap_info_struct *si;
1247         struct swap_cluster_info *ci;
1248         unsigned long offset = swp_offset(entry);
1249 
1250         si = _swap_info_get(entry);
1251         if (!si)
1252                 return -EBUSY;
1253         ci = lock_cluster(si, offset);
1254         cluster_clear_huge(ci);
1255         unlock_cluster(ci);
1256         return 0;
1257 }
1258 #else
1259 static inline void swapcache_free_cluster(swp_entry_t entry)
1260 {
1261 }
1262 #endif /* CONFIG_THP_SWAP */
1263 
1264 void put_swap_page(struct page *page, swp_entry_t entry)
1265 {
1266         if (!PageTransHuge(page))
1267                 swapcache_free(entry);
1268         else
1269                 swapcache_free_cluster(entry);
1270 }
1271 
1272 static int swp_entry_cmp(const void *ent1, const void *ent2)
1273 {
1274         const swp_entry_t *e1 = ent1, *e2 = ent2;
1275 
1276         return (int)swp_type(*e1) - (int)swp_type(*e2);
1277 }
1278 
1279 void swapcache_free_entries(swp_entry_t *entries, int n)
1280 {
1281         struct swap_info_struct *p, *prev;
1282         int i;
1283 
1284         if (n <= 0)
1285                 return;
1286 
1287         prev = NULL;
1288         p = NULL;
1289 
1290         /*
1291          * Sort swap entries by swap device, so each lock is only taken once.
1292          * nr_swapfiles isn't absolutely correct, but the overhead of sort() is
1293          * so low that it isn't necessary to optimize further.
1294          */
1295         if (nr_swapfiles > 1)
1296                 sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL);
1297         for (i = 0; i < n; ++i) {
1298                 p = swap_info_get_cont(entries[i], prev);
1299                 if (p)
1300                         swap_entry_free(p, entries[i]);
1301                 prev = p;
1302         }
1303         if (p)
1304                 spin_unlock(&p->lock);
1305 }
1306 
1307 /*
1308  * How many references to page are currently swapped out?
1309  * This does not give an exact answer when swap count is continued,
1310  * but does include the high COUNT_CONTINUED flag to allow for that.
1311  */
1312 int page_swapcount(struct page *page)
1313 {
1314         int count = 0;
1315         struct swap_info_struct *p;
1316         struct swap_cluster_info *ci;
1317         swp_entry_t entry;
1318         unsigned long offset;
1319 
1320         entry.val = page_private(page);
1321         p = _swap_info_get(entry);
1322         if (p) {
1323                 offset = swp_offset(entry);
1324                 ci = lock_cluster_or_swap_info(p, offset);
1325                 count = swap_count(p->swap_map[offset]);
1326                 unlock_cluster_or_swap_info(p, ci);
1327         }
1328         return count;
1329 }
1330 
1331 int __swap_count(struct swap_info_struct *si, swp_entry_t entry)
1332 {
1333         pgoff_t offset = swp_offset(entry);
1334 
1335         return swap_count(si->swap_map[offset]);
1336 }
1337 
1338 static int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry)
1339 {
1340         int count = 0;
1341         pgoff_t offset = swp_offset(entry);
1342         struct swap_cluster_info *ci;
1343 
1344         ci = lock_cluster_or_swap_info(si, offset);
1345         count = swap_count(si->swap_map[offset]);
1346         unlock_cluster_or_swap_info(si, ci);
1347         return count;
1348 }
1349 
1350 /*
1351  * How many references to @entry are currently swapped out?
1352  * This does not give an exact answer when swap count is continued,
1353  * but does include the high COUNT_CONTINUED flag to allow for that.
1354  */
1355 int __swp_swapcount(swp_entry_t entry)
1356 {
1357         int count = 0;
1358         struct swap_info_struct *si;
1359 
1360         si = __swap_info_get(entry);
1361         if (si)
1362                 count = swap_swapcount(si, entry);
1363         return count;
1364 }
1365 
1366 /*
1367  * How many references to @entry are currently swapped out?
1368  * This considers COUNT_CONTINUED so it returns exact answer.
1369  */
1370 int swp_swapcount(swp_entry_t entry)
1371 {
1372         int count, tmp_count, n;
1373         struct swap_info_struct *p;
1374         struct swap_cluster_info *ci;
1375         struct page *page;
1376         pgoff_t offset;
1377         unsigned char *map;
1378 
1379         p = _swap_info_get(entry);
1380         if (!p)
1381                 return 0;
1382 
1383         offset = swp_offset(entry);
1384 
1385         ci = lock_cluster_or_swap_info(p, offset);
1386 
1387         count = swap_count(p->swap_map[offset]);
1388         if (!(count & COUNT_CONTINUED))
1389                 goto out;
1390 
1391         count &= ~COUNT_CONTINUED;
1392         n = SWAP_MAP_MAX + 1;
1393 
1394         page = vmalloc_to_page(p->swap_map + offset);
1395         offset &= ~PAGE_MASK;
1396         VM_BUG_ON(page_private(page) != SWP_CONTINUED);
1397 
1398         do {
1399                 page = list_next_entry(page, lru);
1400                 map = kmap_atomic(page);
1401                 tmp_count = map[offset];
1402                 kunmap_atomic(map);
1403 
1404                 count += (tmp_count & ~COUNT_CONTINUED) * n;
1405                 n *= (SWAP_CONT_MAX + 1);
1406         } while (tmp_count & COUNT_CONTINUED);
1407 out:
1408         unlock_cluster_or_swap_info(p, ci);
1409         return count;
1410 }
1411 
1412 #ifdef CONFIG_THP_SWAP
1413 static bool swap_page_trans_huge_swapped(struct swap_info_struct *si,
1414                                          swp_entry_t entry)
1415 {
1416         struct swap_cluster_info *ci;
1417         unsigned char *map = si->swap_map;
1418         unsigned long roffset = swp_offset(entry);
1419         unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER);
1420         int i;
1421         bool ret = false;
1422 
1423         ci = lock_cluster_or_swap_info(si, offset);
1424         if (!ci || !cluster_is_huge(ci)) {
1425                 if (map[roffset] != SWAP_HAS_CACHE)
1426                         ret = true;
1427                 goto unlock_out;
1428         }
1429         for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1430                 if (map[offset + i] != SWAP_HAS_CACHE) {
1431                         ret = true;
1432                         break;
1433                 }
1434         }
1435 unlock_out:
1436         unlock_cluster_or_swap_info(si, ci);
1437         return ret;
1438 }
1439 
1440 static bool page_swapped(struct page *page)
1441 {
1442         swp_entry_t entry;
1443         struct swap_info_struct *si;
1444 
1445         if (likely(!PageTransCompound(page)))
1446                 return page_swapcount(page) != 0;
1447 
1448         page = compound_head(page);
1449         entry.val = page_private(page);
1450         si = _swap_info_get(entry);
1451         if (si)
1452                 return swap_page_trans_huge_swapped(si, entry);
1453         return false;
1454 }
1455 
1456 static int page_trans_huge_map_swapcount(struct page *page, int *total_mapcount,
1457                                          int *total_swapcount)
1458 {
1459         int i, map_swapcount, _total_mapcount, _total_swapcount;
1460         unsigned long offset = 0;
1461         struct swap_info_struct *si;
1462         struct swap_cluster_info *ci = NULL;
1463         unsigned char *map = NULL;
1464         int mapcount, swapcount = 0;
1465 
1466         /* hugetlbfs shouldn't call it */
1467         VM_BUG_ON_PAGE(PageHuge(page), page);
1468 
1469         if (likely(!PageTransCompound(page))) {
1470                 mapcount = atomic_read(&page->_mapcount) + 1;
1471                 if (total_mapcount)
1472                         *total_mapcount = mapcount;
1473                 if (PageSwapCache(page))
1474                         swapcount = page_swapcount(page);
1475                 if (total_swapcount)
1476                         *total_swapcount = swapcount;
1477                 return mapcount + swapcount;
1478         }
1479 
1480         page = compound_head(page);
1481 
1482         _total_mapcount = _total_swapcount = map_swapcount = 0;
1483         if (PageSwapCache(page)) {
1484                 swp_entry_t entry;
1485 
1486                 entry.val = page_private(page);
1487                 si = _swap_info_get(entry);
1488                 if (si) {
1489                         map = si->swap_map;
1490                         offset = swp_offset(entry);
1491                 }
1492         }
1493         if (map)
1494                 ci = lock_cluster(si, offset);
1495         for (i = 0; i < HPAGE_PMD_NR; i++) {
1496                 mapcount = atomic_read(&page[i]._mapcount) + 1;
1497                 _total_mapcount += mapcount;
1498                 if (map) {
1499                         swapcount = swap_count(map[offset + i]);
1500                         _total_swapcount += swapcount;
1501                 }
1502                 map_swapcount = max(map_swapcount, mapcount + swapcount);
1503         }
1504         unlock_cluster(ci);
1505         if (PageDoubleMap(page)) {
1506                 map_swapcount -= 1;
1507                 _total_mapcount -= HPAGE_PMD_NR;
1508         }
1509         mapcount = compound_mapcount(page);
1510         map_swapcount += mapcount;
1511         _total_mapcount += mapcount;
1512         if (total_mapcount)
1513                 *total_mapcount = _total_mapcount;
1514         if (total_swapcount)
1515                 *total_swapcount = _total_swapcount;
1516 
1517         return map_swapcount;
1518 }
1519 #else
1520 #define swap_page_trans_huge_swapped(si, entry) swap_swapcount(si, entry)
1521 #define page_swapped(page)                      (page_swapcount(page) != 0)
1522 
1523 static int page_trans_huge_map_swapcount(struct page *page, int *total_mapcount,
1524                                          int *total_swapcount)
1525 {
1526         int mapcount, swapcount = 0;
1527 
1528         /* hugetlbfs shouldn't call it */
1529         VM_BUG_ON_PAGE(PageHuge(page), page);
1530 
1531         mapcount = page_trans_huge_mapcount(page, total_mapcount);
1532         if (PageSwapCache(page))
1533                 swapcount = page_swapcount(page);
1534         if (total_swapcount)
1535                 *total_swapcount = swapcount;
1536         return mapcount + swapcount;
1537 }
1538 #endif
1539 
1540 /*
1541  * We can write to an anon page without COW if there are no other references
1542  * to it.  And as a side-effect, free up its swap: because the old content
1543  * on disk will never be read, and seeking back there to write new content
1544  * later would only waste time away from clustering.
1545  *
1546  * NOTE: total_map_swapcount should not be relied upon by the caller if
1547  * reuse_swap_page() returns false, but it may be always overwritten
1548  * (see the other implementation for CONFIG_SWAP=n).
1549  */
1550 bool reuse_swap_page(struct page *page, int *total_map_swapcount)
1551 {
1552         int count, total_mapcount, total_swapcount;
1553 
1554         VM_BUG_ON_PAGE(!PageLocked(page), page);
1555         if (unlikely(PageKsm(page)))
1556                 return false;
1557         count = page_trans_huge_map_swapcount(page, &total_mapcount,
1558                                               &total_swapcount);
1559         if (total_map_swapcount)
1560                 *total_map_swapcount = total_mapcount + total_swapcount;
1561         if (count == 1 && PageSwapCache(page) &&
1562             (likely(!PageTransCompound(page)) ||
1563              /* The remaining swap count will be freed soon */
1564              total_swapcount == page_swapcount(page))) {
1565                 if (!PageWriteback(page)) {
1566                         page = compound_head(page);
1567                         delete_from_swap_cache(page);
1568                         SetPageDirty(page);
1569                 } else {
1570                         swp_entry_t entry;
1571                         struct swap_info_struct *p;
1572 
1573                         entry.val = page_private(page);
1574                         p = swap_info_get(entry);
1575                         if (p->flags & SWP_STABLE_WRITES) {
1576                                 spin_unlock(&p->lock);
1577                                 return false;
1578                         }
1579                         spin_unlock(&p->lock);
1580                 }
1581         }
1582 
1583         return count <= 1;
1584 }
1585 
1586 /*
1587  * If swap is getting full, or if there are no more mappings of this page,
1588  * then try_to_free_swap is called to free its swap space.
1589  */
1590 int try_to_free_swap(struct page *page)
1591 {
1592         VM_BUG_ON_PAGE(!PageLocked(page), page);
1593 
1594         if (!PageSwapCache(page))
1595                 return 0;
1596         if (PageWriteback(page))
1597                 return 0;
1598         if (page_swapped(page))
1599                 return 0;
1600 
1601         /*
1602          * Once hibernation has begun to create its image of memory,
1603          * there's a danger that one of the calls to try_to_free_swap()
1604          * - most probably a call from __try_to_reclaim_swap() while
1605          * hibernation is allocating its own swap pages for the image,
1606          * but conceivably even a call from memory reclaim - will free
1607          * the swap from a page which has already been recorded in the
1608          * image as a clean swapcache page, and then reuse its swap for
1609          * another page of the image.  On waking from hibernation, the
1610          * original page might be freed under memory pressure, then
1611          * later read back in from swap, now with the wrong data.
1612          *
1613          * Hibernation suspends storage while it is writing the image
1614          * to disk so check that here.
1615          */
1616         if (pm_suspended_storage())
1617                 return 0;
1618 
1619         page = compound_head(page);
1620         delete_from_swap_cache(page);
1621         SetPageDirty(page);
1622         return 1;
1623 }
1624 
1625 /*
1626  * Free the swap entry like above, but also try to
1627  * free the page cache entry if it is the last user.
1628  */
1629 int free_swap_and_cache(swp_entry_t entry)
1630 {
1631         struct swap_info_struct *p;
1632         struct page *page = NULL;
1633         unsigned char count;
1634 
1635         if (non_swap_entry(entry))
1636                 return 1;
1637 
1638         p = _swap_info_get(entry);
1639         if (p) {
1640                 count = __swap_entry_free(p, entry, 1);
1641                 if (count == SWAP_HAS_CACHE &&
1642                     !swap_page_trans_huge_swapped(p, entry)) {
1643                         page = find_get_page(swap_address_space(entry),
1644                                              swp_offset(entry));
1645                         if (page && !trylock_page(page)) {
1646                                 put_page(page);
1647                                 page = NULL;
1648                         }
1649                 } else if (!count)
1650                         free_swap_slot(entry);
1651         }
1652         if (page) {
1653                 /*
1654                  * Not mapped elsewhere, or swap space full? Free it!
1655                  * Also recheck PageSwapCache now page is locked (above).
1656                  */
1657                 if (PageSwapCache(page) && !PageWriteback(page) &&
1658                     (!page_mapped(page) || mem_cgroup_swap_full(page)) &&
1659                     !swap_page_trans_huge_swapped(p, entry)) {
1660                         page = compound_head(page);
1661                         delete_from_swap_cache(page);
1662                         SetPageDirty(page);
1663                 }
1664                 unlock_page(page);
1665                 put_page(page);
1666         }
1667         return p != NULL;
1668 }
1669 
1670 #ifdef CONFIG_HIBERNATION
1671 /*
1672  * Find the swap type that corresponds to given device (if any).
1673  *
1674  * @offset - number of the PAGE_SIZE-sized block of the device, starting
1675  * from 0, in which the swap header is expected to be located.
1676  *
1677  * This is needed for the suspend to disk (aka swsusp).
1678  */
1679 int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
1680 {
1681         struct block_device *bdev = NULL;
1682         int type;
1683 
1684         if (device)
1685                 bdev = bdget(device);
1686 
1687         spin_lock(&swap_lock);
1688         for (type = 0; type < nr_swapfiles; type++) {
1689                 struct swap_info_struct *sis = swap_info[type];
1690 
1691                 if (!(sis->flags & SWP_WRITEOK))
1692                         continue;
1693 
1694                 if (!bdev) {
1695                         if (bdev_p)
1696                                 *bdev_p = bdgrab(sis->bdev);
1697 
1698                         spin_unlock(&swap_lock);
1699                         return type;
1700                 }
1701                 if (bdev == sis->bdev) {
1702                         struct swap_extent *se = &sis->first_swap_extent;
1703 
1704                         if (se->start_block == offset) {
1705                                 if (bdev_p)
1706                                         *bdev_p = bdgrab(sis->bdev);
1707 
1708                                 spin_unlock(&swap_lock);
1709                                 bdput(bdev);
1710                                 return type;
1711                         }
1712                 }
1713         }
1714         spin_unlock(&swap_lock);
1715         if (bdev)
1716                 bdput(bdev);
1717 
1718         return -ENODEV;
1719 }
1720 
1721 /*
1722  * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1723  * corresponding to given index in swap_info (swap type).
1724  */
1725 sector_t swapdev_block(int type, pgoff_t offset)
1726 {
1727         struct block_device *bdev;
1728 
1729         if ((unsigned int)type >= nr_swapfiles)
1730                 return 0;
1731         if (!(swap_info[type]->flags & SWP_WRITEOK))
1732                 return 0;
1733         return map_swap_entry(swp_entry(type, offset), &bdev);
1734 }
1735 
1736 /*
1737  * Return either the total number of swap pages of given type, or the number
1738  * of free pages of that type (depending on @free)
1739  *
1740  * This is needed for software suspend
1741  */
1742 unsigned int count_swap_pages(int type, int free)
1743 {
1744         unsigned int n = 0;
1745 
1746         spin_lock(&swap_lock);
1747         if ((unsigned int)type < nr_swapfiles) {
1748                 struct swap_info_struct *sis = swap_info[type];
1749 
1750                 spin_lock(&sis->lock);
1751                 if (sis->flags & SWP_WRITEOK) {
1752                         n = sis->pages;
1753                         if (free)
1754                                 n -= sis->inuse_pages;
1755                 }
1756                 spin_unlock(&sis->lock);
1757         }
1758         spin_unlock(&swap_lock);
1759         return n;
1760 }
1761 #endif /* CONFIG_HIBERNATION */
1762 
1763 static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte)
1764 {
1765         return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte);
1766 }
1767 
1768 /*
1769  * No need to decide whether this PTE shares the swap entry with others,
1770  * just let do_wp_page work it out if a write is requested later - to
1771  * force COW, vm_page_prot omits write permission from any private vma.
1772  */
1773 static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
1774                 unsigned long addr, swp_entry_t entry, struct page *page)
1775 {
1776         struct page *swapcache;
1777         struct mem_cgroup *memcg;
1778         spinlock_t *ptl;
1779         pte_t *pte;
1780         int ret = 1;
1781 
1782         swapcache = page;
1783         page = ksm_might_need_to_copy(page, vma, addr);
1784         if (unlikely(!page))
1785                 return -ENOMEM;
1786 
1787         if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL,
1788                                 &memcg, false)) {
1789                 ret = -ENOMEM;
1790                 goto out_nolock;
1791         }
1792 
1793         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1794         if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) {
1795                 mem_cgroup_cancel_charge(page, memcg, false);
1796                 ret = 0;
1797                 goto out;
1798         }
1799 
1800         dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1801         inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
1802         get_page(page);
1803         set_pte_at(vma->vm_mm, addr, pte,
1804                    pte_mkold(mk_pte(page, vma->vm_page_prot)));
1805         if (page == swapcache) {
1806                 page_add_anon_rmap(page, vma, addr, false);
1807                 mem_cgroup_commit_charge(page, memcg, true, false);
1808         } else { /* ksm created a completely new copy */
1809                 page_add_new_anon_rmap(page, vma, addr, false);
1810                 mem_cgroup_commit_charge(page, memcg, false, false);
1811                 lru_cache_add_active_or_unevictable(page, vma);
1812         }
1813         swap_free(entry);
1814         /*
1815          * Move the page to the active list so it is not
1816          * immediately swapped out again after swapon.
1817          */
1818         activate_page(page);
1819 out:
1820         pte_unmap_unlock(pte, ptl);
1821 out_nolock:
1822         if (page != swapcache) {
1823                 unlock_page(page);
1824                 put_page(page);
1825         }
1826         return ret;
1827 }
1828 
1829 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1830                                 unsigned long addr, unsigned long end,
1831                                 swp_entry_t entry, struct page *page)
1832 {
1833         pte_t swp_pte = swp_entry_to_pte(entry);
1834         pte_t *pte;
1835         int ret = 0;
1836 
1837         /*
1838          * We don't actually need pte lock while scanning for swp_pte: since
1839          * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
1840          * page table while we're scanning; though it could get zapped, and on
1841          * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
1842          * of unmatched parts which look like swp_pte, so unuse_pte must
1843          * recheck under pte lock.  Scanning without pte lock lets it be
1844          * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
1845          */
1846         pte = pte_offset_map(pmd, addr);
1847         do {
1848                 /*
1849                  * swapoff spends a _lot_ of time in this loop!
1850                  * Test inline before going to call unuse_pte.
1851                  */
1852                 if (unlikely(pte_same_as_swp(*pte, swp_pte))) {
1853                         pte_unmap(pte);
1854                         ret = unuse_pte(vma, pmd, addr, entry, page);
1855                         if (ret)
1856                                 goto out;
1857                         pte = pte_offset_map(pmd, addr);
1858                 }
1859         } while (pte++, addr += PAGE_SIZE, addr != end);
1860         pte_unmap(pte - 1);
1861 out:
1862         return ret;
1863 }
1864 
1865 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1866                                 unsigned long addr, unsigned long end,
1867                                 swp_entry_t entry, struct page *page)
1868 {
1869         pmd_t *pmd;
1870         unsigned long next;
1871         int ret;
1872 
1873         pmd = pmd_offset(pud, addr);
1874         do {
1875                 cond_resched();
1876                 next = pmd_addr_end(addr, end);
1877                 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
1878                         continue;
1879                 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
1880                 if (ret)
1881                         return ret;
1882         } while (pmd++, addr = next, addr != end);
1883         return 0;
1884 }
1885 
1886 static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d,
1887                                 unsigned long addr, unsigned long end,
1888                                 swp_entry_t entry, struct page *page)
1889 {
1890         pud_t *pud;
1891         unsigned long next;
1892         int ret;
1893 
1894         pud = pud_offset(p4d, addr);
1895         do {
1896                 next = pud_addr_end(addr, end);
1897                 if (pud_none_or_clear_bad(pud))
1898                         continue;
1899                 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
1900                 if (ret)
1901                         return ret;
1902         } while (pud++, addr = next, addr != end);
1903         return 0;
1904 }
1905 
1906 static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd,
1907                                 unsigned long addr, unsigned long end,
1908                                 swp_entry_t entry, struct page *page)
1909 {
1910         p4d_t *p4d;
1911         unsigned long next;
1912         int ret;
1913 
1914         p4d = p4d_offset(pgd, addr);
1915         do {
1916                 next = p4d_addr_end(addr, end);
1917                 if (p4d_none_or_clear_bad(p4d))
1918                         continue;
1919                 ret = unuse_pud_range(vma, p4d, addr, next, entry, page);
1920                 if (ret)
1921                         return ret;
1922         } while (p4d++, addr = next, addr != end);
1923         return 0;
1924 }
1925 
1926 static int unuse_vma(struct vm_area_struct *vma,
1927                                 swp_entry_t entry, struct page *page)
1928 {
1929         pgd_t *pgd;
1930         unsigned long addr, end, next;
1931         int ret;
1932 
1933         if (page_anon_vma(page)) {
1934                 addr = page_address_in_vma(page, vma);
1935                 if (addr == -EFAULT)
1936                         return 0;
1937                 else
1938                         end = addr + PAGE_SIZE;
1939         } else {
1940                 addr = vma->vm_start;
1941                 end = vma->vm_end;
1942         }
1943 
1944         pgd = pgd_offset(vma->vm_mm, addr);
1945         do {
1946                 next = pgd_addr_end(addr, end);
1947                 if (pgd_none_or_clear_bad(pgd))
1948                         continue;
1949                 ret = unuse_p4d_range(vma, pgd, addr, next, entry, page);
1950                 if (ret)
1951                         return ret;
1952         } while (pgd++, addr = next, addr != end);
1953         return 0;
1954 }
1955 
1956 static int unuse_mm(struct mm_struct *mm,
1957                                 swp_entry_t entry, struct page *page)
1958 {
1959         struct vm_area_struct *vma;
1960         int ret = 0;
1961 
1962         if (!down_read_trylock(&mm->mmap_sem)) {
1963                 /*
1964                  * Activate page so shrink_inactive_list is unlikely to unmap
1965                  * its ptes while lock is dropped, so swapoff can make progress.
1966                  */
1967                 activate_page(page);
1968                 unlock_page(page);
1969                 down_read(&mm->mmap_sem);
1970                 lock_page(page);
1971         }
1972         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1973                 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
1974                         break;
1975                 cond_resched();
1976         }
1977         up_read(&mm->mmap_sem);
1978         return (ret < 0)? ret: 0;
1979 }
1980 
1981 /*
1982  * Scan swap_map (or frontswap_map if frontswap parameter is true)
1983  * from current position to next entry still in use.
1984  * Recycle to start on reaching the end, returning 0 when empty.
1985  */
1986 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
1987                                         unsigned int prev, bool frontswap)
1988 {
1989         unsigned int max = si->max;
1990         unsigned int i = prev;
1991         unsigned char count;
1992 
1993         /*
1994          * No need for swap_lock here: we're just looking
1995          * for whether an entry is in use, not modifying it; false
1996          * hits are okay, and sys_swapoff() has already prevented new
1997          * allocations from this area (while holding swap_lock).
1998          */
1999         for (;;) {
2000                 if (++i >= max) {
2001                         if (!prev) {
2002                                 i = 0;
2003                                 break;
2004                         }
2005                         /*
2006                          * No entries in use at top of swap_map,
2007                          * loop back to start and recheck there.
2008                          */
2009                         max = prev + 1;
2010                         prev = 0;
2011                         i = 1;
2012                 }
2013                 count = READ_ONCE(si->swap_map[i]);
2014                 if (count && swap_count(count) != SWAP_MAP_BAD)
2015                         if (!frontswap || frontswap_test(si, i))
2016                                 break;
2017                 if ((i % LATENCY_LIMIT) == 0)
2018                         cond_resched();
2019         }
2020         return i;
2021 }
2022 
2023 /*
2024  * We completely avoid races by reading each swap page in advance,
2025  * and then search for the process using it.  All the necessary
2026  * page table adjustments can then be made atomically.
2027  *
2028  * if the boolean frontswap is true, only unuse pages_to_unuse pages;
2029  * pages_to_unuse==0 means all pages; ignored if frontswap is false
2030  */
2031 int try_to_unuse(unsigned int type, bool frontswap,
2032                  unsigned long pages_to_unuse)
2033 {
2034         struct swap_info_struct *si = swap_info[type];
2035         struct mm_struct *start_mm;
2036         volatile unsigned char *swap_map; /* swap_map is accessed without
2037                                            * locking. Mark it as volatile
2038                                            * to prevent compiler doing
2039                                            * something odd.
2040                                            */
2041         unsigned char swcount;
2042         struct page *page;
2043         swp_entry_t entry;
2044         unsigned int i = 0;
2045         int retval = 0;
2046 
2047         /*
2048          * When searching mms for an entry, a good strategy is to
2049          * start at the first mm we freed the previous entry from
2050          * (though actually we don't notice whether we or coincidence
2051          * freed the entry).  Initialize this start_mm with a hold.
2052          *
2053          * A simpler strategy would be to start at the last mm we
2054          * freed the previous entry from; but that would take less
2055          * advantage of mmlist ordering, which clusters forked mms
2056          * together, child after parent.  If we race with dup_mmap(), we
2057          * prefer to resolve parent before child, lest we miss entries
2058          * duplicated after we scanned child: using last mm would invert
2059          * that.
2060          */
2061         start_mm = &init_mm;
2062         mmget(&init_mm);
2063 
2064         /*
2065          * Keep on scanning until all entries have gone.  Usually,
2066          * one pass through swap_map is enough, but not necessarily:
2067          * there are races when an instance of an entry might be missed.
2068          */
2069         while ((i = find_next_to_unuse(si, i, frontswap)) != 0) {
2070                 if (signal_pending(current)) {
2071                         retval = -EINTR;
2072                         break;
2073                 }
2074 
2075                 /*
2076                  * Get a page for the entry, using the existing swap
2077                  * cache page if there is one.  Otherwise, get a clean
2078                  * page and read the swap into it.
2079                  */
2080                 swap_map = &si->swap_map[i];
2081                 entry = swp_entry(type, i);
2082                 page = read_swap_cache_async(entry,
2083                                         GFP_HIGHUSER_MOVABLE, NULL, 0, false);
2084                 if (!page) {
2085                         /*
2086                          * Either swap_duplicate() failed because entry
2087                          * has been freed independently, and will not be
2088                          * reused since sys_swapoff() already disabled
2089                          * allocation from here, or alloc_page() failed.
2090                          */
2091                         swcount = *swap_map;
2092                         /*
2093                          * We don't hold lock here, so the swap entry could be
2094                          * SWAP_MAP_BAD (when the cluster is discarding).
2095                          * Instead of fail out, We can just skip the swap
2096                          * entry because swapoff will wait for discarding
2097                          * finish anyway.
2098                          */
2099                         if (!swcount || swcount == SWAP_MAP_BAD)
2100                                 continue;
2101                         retval = -ENOMEM;
2102                         break;
2103                 }
2104 
2105                 /*
2106                  * Don't hold on to start_mm if it looks like exiting.
2107                  */
2108                 if (atomic_read(&start_mm->mm_users) == 1) {
2109                         mmput(start_mm);
2110                         start_mm = &init_mm;
2111                         mmget(&init_mm);
2112                 }
2113 
2114                 /*
2115                  * Wait for and lock page.  When do_swap_page races with
2116                  * try_to_unuse, do_swap_page can handle the fault much
2117                  * faster than try_to_unuse can locate the entry.  This
2118                  * apparently redundant "wait_on_page_locked" lets try_to_unuse
2119                  * defer to do_swap_page in such a case - in some tests,
2120                  * do_swap_page and try_to_unuse repeatedly compete.
2121                  */
2122                 wait_on_page_locked(page);
2123                 wait_on_page_writeback(page);
2124                 lock_page(page);
2125                 wait_on_page_writeback(page);
2126 
2127                 /*
2128                  * Remove all references to entry.
2129                  */
2130                 swcount = *swap_map;
2131                 if (swap_count(swcount) == SWAP_MAP_SHMEM) {
2132                         retval = shmem_unuse(entry, page);
2133                         /* page has already been unlocked and released */
2134                         if (retval < 0)
2135                                 break;
2136                         continue;
2137                 }
2138                 if (swap_count(swcount) && start_mm != &init_mm)
2139                         retval = unuse_mm(start_mm, entry, page);
2140 
2141                 if (swap_count(*swap_map)) {
2142                         int set_start_mm = (*swap_map >= swcount);
2143                         struct list_head *p = &start_mm->mmlist;
2144                         struct mm_struct *new_start_mm = start_mm;
2145                         struct mm_struct *prev_mm = start_mm;
2146                         struct mm_struct *mm;
2147 
2148                         mmget(new_start_mm);
2149                         mmget(prev_mm);
2150                         spin_lock(&mmlist_lock);
2151                         while (swap_count(*swap_map) && !retval &&
2152                                         (p = p->next) != &start_mm->mmlist) {
2153                                 mm = list_entry(p, struct mm_struct, mmlist);
2154                                 if (!mmget_not_zero(mm))
2155                                         continue;
2156                                 spin_unlock(&mmlist_lock);
2157                                 mmput(prev_mm);
2158                                 prev_mm = mm;
2159 
2160                                 cond_resched();
2161 
2162                                 swcount = *swap_map;
2163                                 if (!swap_count(swcount)) /* any usage ? */
2164                                         ;
2165                                 else if (mm == &init_mm)
2166                                         set_start_mm = 1;
2167                                 else
2168                                         retval = unuse_mm(mm, entry, page);
2169 
2170                                 if (set_start_mm && *swap_map < swcount) {
2171                                         mmput(new_start_mm);
2172                                         mmget(mm);
2173                                         new_start_mm = mm;
2174                                         set_start_mm = 0;
2175                                 }
2176                                 spin_lock(&mmlist_lock);
2177                         }
2178                         spin_unlock(&mmlist_lock);
2179                         mmput(prev_mm);
2180                         mmput(start_mm);
2181                         start_mm = new_start_mm;
2182                 }
2183                 if (retval) {
2184                         unlock_page(page);
2185                         put_page(page);
2186                         break;
2187                 }
2188 
2189                 /*
2190                  * If a reference remains (rare), we would like to leave
2191                  * the page in the swap cache; but try_to_unmap could
2192                  * then re-duplicate the entry once we drop page lock,
2193                  * so we might loop indefinitely; also, that page could
2194                  * not be swapped out to other storage meanwhile.  So:
2195                  * delete from cache even if there's another reference,
2196                  * after ensuring that the data has been saved to disk -
2197                  * since if the reference remains (rarer), it will be
2198                  * read from disk into another page.  Splitting into two
2199                  * pages would be incorrect if swap supported "shared
2200                  * private" pages, but they are handled by tmpfs files.
2201                  *
2202                  * Given how unuse_vma() targets one particular offset
2203                  * in an anon_vma, once the anon_vma has been determined,
2204                  * this splitting happens to be just what is needed to
2205                  * handle where KSM pages have been swapped out: re-reading
2206                  * is unnecessarily slow, but we can fix that later on.
2207                  */
2208                 if (swap_count(*swap_map) &&
2209                      PageDirty(page) && PageSwapCache(page)) {
2210                         struct writeback_control wbc = {
2211                                 .sync_mode = WB_SYNC_NONE,
2212                         };
2213 
2214                         swap_writepage(compound_head(page), &wbc);
2215                         lock_page(page);
2216                         wait_on_page_writeback(page);
2217                 }
2218 
2219                 /*
2220                  * It is conceivable that a racing task removed this page from
2221                  * swap cache just before we acquired the page lock at the top,
2222                  * or while we dropped it in unuse_mm().  The page might even
2223                  * be back in swap cache on another swap area: that we must not
2224                  * delete, since it may not have been written out to swap yet.
2225                  */
2226                 if (PageSwapCache(page) &&
2227                     likely(page_private(page) == entry.val) &&
2228                     !page_swapped(page))
2229                         delete_from_swap_cache(compound_head(page));
2230 
2231                 /*
2232                  * So we could skip searching mms once swap count went
2233                  * to 1, we did not mark any present ptes as dirty: must
2234                  * mark page dirty so shrink_page_list will preserve it.
2235                  */
2236                 SetPageDirty(page);
2237                 unlock_page(page);
2238                 put_page(page);
2239 
2240                 /*
2241                  * Make sure that we aren't completely killing
2242                  * interactive performance.
2243                  */
2244                 cond_resched();
2245                 if (frontswap && pages_to_unuse > 0) {
2246                         if (!--pages_to_unuse)
2247                                 break;
2248                 }
2249         }
2250 
2251         mmput(start_mm);
2252         return retval;
2253 }
2254 
2255 /*
2256  * After a successful try_to_unuse, if no swap is now in use, we know
2257  * we can empty the mmlist.  swap_lock must be held on entry and exit.
2258  * Note that mmlist_lock nests inside swap_lock, and an mm must be
2259  * added to the mmlist just after page_duplicate - before would be racy.
2260  */
2261 static void drain_mmlist(void)
2262 {
2263         struct list_head *p, *next;
2264         unsigned int type;
2265 
2266         for (type = 0; type < nr_swapfiles; type++)
2267                 if (swap_info[type]->inuse_pages)
2268                         return;
2269         spin_lock(&mmlist_lock);
2270         list_for_each_safe(p, next, &init_mm.mmlist)
2271                 list_del_init(p);
2272         spin_unlock(&mmlist_lock);
2273 }
2274 
2275 /*
2276  * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
2277  * corresponds to page offset for the specified swap entry.
2278  * Note that the type of this function is sector_t, but it returns page offset
2279  * into the bdev, not sector offset.
2280  */
2281 static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
2282 {
2283         struct swap_info_struct *sis;
2284         struct swap_extent *start_se;
2285         struct swap_extent *se;
2286         pgoff_t offset;
2287 
2288         sis = swap_info[swp_type(entry)];
2289         *bdev = sis->bdev;
2290 
2291         offset = swp_offset(entry);
2292         start_se = sis->curr_swap_extent;
2293         se = start_se;
2294 
2295         for ( ; ; ) {
2296                 if (se->start_page <= offset &&
2297                                 offset < (se->start_page + se->nr_pages)) {
2298                         return se->start_block + (offset - se->start_page);
2299                 }
2300                 se = list_next_entry(se, list);
2301                 sis->curr_swap_extent = se;
2302                 BUG_ON(se == start_se);         /* It *must* be present */
2303         }
2304 }
2305 
2306 /*
2307  * Returns the page offset into bdev for the specified page's swap entry.
2308  */
2309 sector_t map_swap_page(struct page *page, struct block_device **bdev)
2310 {
2311         swp_entry_t entry;
2312         entry.val = page_private(page);
2313         return map_swap_entry(entry, bdev);
2314 }
2315 
2316 /*
2317  * Free all of a swapdev's extent information
2318  */
2319 static void destroy_swap_extents(struct swap_info_struct *sis)
2320 {
2321         while (!list_empty(&sis->first_swap_extent.list)) {
2322                 struct swap_extent *se;
2323 
2324                 se = list_first_entry(&sis->first_swap_extent.list,
2325                                 struct swap_extent, list);
2326                 list_del(&se->list);
2327                 kfree(se);
2328         }
2329 
2330         if (sis->flags & SWP_FILE) {
2331                 struct file *swap_file = sis->swap_file;
2332                 struct address_space *mapping = swap_file->f_mapping;
2333 
2334                 sis->flags &= ~SWP_FILE;
2335                 mapping->a_ops->swap_deactivate(swap_file);
2336         }
2337 }
2338 
2339 /*
2340  * Add a block range (and the corresponding page range) into this swapdev's
2341  * extent list.  The extent list is kept sorted in page order.
2342  *
2343  * This function rather assumes that it is called in ascending page order.
2344  */
2345 int
2346 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
2347                 unsigned long nr_pages, sector_t start_block)
2348 {
2349         struct swap_extent *se;
2350         struct swap_extent *new_se;
2351         struct list_head *lh;
2352 
2353         if (start_page == 0) {
2354                 se = &sis->first_swap_extent;
2355                 sis->curr_swap_extent = se;
2356                 se->start_page = 0;
2357                 se->nr_pages = nr_pages;
2358                 se->start_block = start_block;
2359                 return 1;
2360         } else {
2361                 lh = sis->first_swap_extent.list.prev;  /* Highest extent */
2362                 se = list_entry(lh, struct swap_extent, list);
2363                 BUG_ON(se->start_page + se->nr_pages != start_page);
2364                 if (se->start_block + se->nr_pages == start_block) {
2365                         /* Merge it */
2366                         se->nr_pages += nr_pages;
2367                         return 0;
2368                 }
2369         }
2370 
2371         /*
2372          * No merge.  Insert a new extent, preserving ordering.
2373          */
2374         new_se = kmalloc(sizeof(*se), GFP_KERNEL);
2375         if (new_se == NULL)
2376                 return -ENOMEM;
2377         new_se->start_page = start_page;
2378         new_se->nr_pages = nr_pages;
2379         new_se->start_block = start_block;
2380 
2381         list_add_tail(&new_se->list, &sis->first_swap_extent.list);
2382         return 1;
2383 }
2384 
2385 /*
2386  * A `swap extent' is a simple thing which maps a contiguous range of pages
2387  * onto a contiguous range of disk blocks.  An ordered list of swap extents
2388  * is built at swapon time and is then used at swap_writepage/swap_readpage
2389  * time for locating where on disk a page belongs.
2390  *
2391  * If the swapfile is an S_ISBLK block device, a single extent is installed.
2392  * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2393  * swap files identically.
2394  *
2395  * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2396  * extent list operates in PAGE_SIZE disk blocks.  Both S_ISREG and S_ISBLK
2397  * swapfiles are handled *identically* after swapon time.
2398  *
2399  * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2400  * and will parse them into an ordered extent list, in PAGE_SIZE chunks.  If
2401  * some stray blocks are found which do not fall within the PAGE_SIZE alignment
2402  * requirements, they are simply tossed out - we will never use those blocks
2403  * for swapping.
2404  *
2405  * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon.  This
2406  * prevents root from shooting her foot off by ftruncating an in-use swapfile,
2407  * which will scribble on the fs.
2408  *
2409  * The amount of disk space which a single swap extent represents varies.
2410  * Typically it is in the 1-4 megabyte range.  So we can have hundreds of
2411  * extents in the list.  To avoid much list walking, we cache the previous
2412  * search location in `curr_swap_extent', and start new searches from there.
2413  * This is extremely effective.  The average number of iterations in
2414  * map_swap_page() has been measured at about 0.3 per page.  - akpm.
2415  */
2416 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
2417 {
2418         struct file *swap_file = sis->swap_file;
2419         struct address_space *mapping = swap_file->f_mapping;
2420         struct inode *inode = mapping->host;
2421         int ret;
2422 
2423         if (S_ISBLK(inode->i_mode)) {
2424                 ret = add_swap_extent(sis, 0, sis->max, 0);
2425                 *span = sis->pages;
2426                 return ret;
2427         }
2428 
2429         if (mapping->a_ops->swap_activate) {
2430                 ret = mapping->a_ops->swap_activate(sis, swap_file, span);
2431                 if (!ret) {
2432                         sis->flags |= SWP_FILE;
2433                         ret = add_swap_extent(sis, 0, sis->max, 0);
2434                         *span = sis->pages;
2435                 }
2436                 return ret;
2437         }
2438 
2439         return generic_swapfile_activate(sis, swap_file, span);
2440 }
2441 
2442 static int swap_node(struct swap_info_struct *p)
2443 {
2444         struct block_device *bdev;
2445 
2446         if (p->bdev)
2447                 bdev = p->bdev;
2448         else
2449                 bdev = p->swap_file->f_inode->i_sb->s_bdev;
2450 
2451         return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE;
2452 }
2453 
2454 static void _enable_swap_info(struct swap_info_struct *p, int prio,
2455                                 unsigned char *swap_map,
2456                                 struct swap_cluster_info *cluster_info)
2457 {
2458         int i;
2459 
2460         if (prio >= 0)
2461                 p->prio = prio;
2462         else
2463                 p->prio = --least_priority;
2464         /*
2465          * the plist prio is negated because plist ordering is
2466          * low-to-high, while swap ordering is high-to-low
2467          */
2468         p->list.prio = -p->prio;
2469         for_each_node(i) {
2470                 if (p->prio >= 0)
2471                         p->avail_lists[i].prio = -p->prio;
2472                 else {
2473                         if (swap_node(p) == i)
2474                                 p->avail_lists[i].prio = 1;
2475                         else
2476                                 p->avail_lists[i].prio = -p->prio;
2477                 }
2478         }
2479         p->swap_map = swap_map;
2480         p->cluster_info = cluster_info;
2481         p->flags |= SWP_WRITEOK;
2482         atomic_long_add(p->pages, &nr_swap_pages);
2483         total_swap_pages += p->pages;
2484 
2485         assert_spin_locked(&swap_lock);
2486         /*
2487          * both lists are plists, and thus priority ordered.
2488          * swap_active_head needs to be priority ordered for swapoff(),
2489          * which on removal of any swap_info_struct with an auto-assigned
2490          * (i.e. negative) priority increments the auto-assigned priority
2491          * of any lower-priority swap_info_structs.
2492          * swap_avail_head needs to be priority ordered for get_swap_page(),
2493          * which allocates swap pages from the highest available priority
2494          * swap_info_struct.
2495          */
2496         plist_add(&p->list, &swap_active_head);
2497         add_to_avail_list(p);
2498 }
2499 
2500 static void enable_swap_info(struct swap_info_struct *p, int prio,
2501                                 unsigned char *swap_map,
2502                                 struct swap_cluster_info *cluster_info,
2503                                 unsigned long *frontswap_map)
2504 {
2505         frontswap_init(p->type, frontswap_map);
2506         spin_lock(&swap_lock);
2507         spin_lock(&p->lock);
2508          _enable_swap_info(p, prio, swap_map, cluster_info);
2509         spin_unlock(&p->lock);
2510         spin_unlock(&swap_lock);
2511 }
2512 
2513 static void reinsert_swap_info(struct swap_info_struct *p)
2514 {
2515         spin_lock(&swap_lock);
2516         spin_lock(&p->lock);
2517         _enable_swap_info(p, p->prio, p->swap_map, p->cluster_info);
2518         spin_unlock(&p->lock);
2519         spin_unlock(&swap_lock);
2520 }
2521 
2522 bool has_usable_swap(void)
2523 {
2524         bool ret = true;
2525 
2526         spin_lock(&swap_lock);
2527         if (plist_head_empty(&swap_active_head))
2528                 ret = false;
2529         spin_unlock(&swap_lock);
2530         return ret;
2531 }
2532 
2533 SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
2534 {
2535         struct swap_info_struct *p = NULL;
2536         unsigned char *swap_map;
2537         struct swap_cluster_info *cluster_info;
2538         unsigned long *frontswap_map;
2539         struct file *swap_file, *victim;
2540         struct address_space *mapping;
2541         struct inode *inode;
2542         struct filename *pathname;
2543         int err, found = 0;
2544         unsigned int old_block_size;
2545 
2546         if (!capable(CAP_SYS_ADMIN))
2547                 return -EPERM;
2548 
2549         BUG_ON(!current->mm);
2550 
2551         pathname = getname(specialfile);
2552         if (IS_ERR(pathname))
2553                 return PTR_ERR(pathname);
2554 
2555         victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
2556         err = PTR_ERR(victim);
2557         if (IS_ERR(victim))
2558                 goto out;
2559 
2560         mapping = victim->f_mapping;
2561         spin_lock(&swap_lock);
2562         plist_for_each_entry(p, &swap_active_head, list) {
2563                 if (p->flags & SWP_WRITEOK) {
2564                         if (p->swap_file->f_mapping == mapping) {
2565                                 found = 1;
2566                                 break;
2567                         }
2568                 }
2569         }
2570         if (!found) {
2571                 err = -EINVAL;
2572                 spin_unlock(&swap_lock);
2573                 goto out_dput;
2574         }
2575         if (!security_vm_enough_memory_mm(current->mm, p->pages))
2576                 vm_unacct_memory(p->pages);
2577         else {
2578                 err = -ENOMEM;
2579                 spin_unlock(&swap_lock);
2580                 goto out_dput;
2581         }
2582         del_from_avail_list(p);
2583         spin_lock(&p->lock);
2584         if (p->prio < 0) {
2585                 struct swap_info_struct *si = p;
2586                 int nid;
2587 
2588                 plist_for_each_entry_continue(si, &swap_active_head, list) {
2589                         si->prio++;
2590                         si->list.prio--;
2591                         for_each_node(nid) {
2592                                 if (si->avail_lists[nid].prio != 1)
2593                                         si->avail_lists[nid].prio--;
2594                         }
2595                 }
2596                 least_priority++;
2597         }
2598         plist_del(&p->list, &swap_active_head);
2599         atomic_long_sub(p->pages, &nr_swap_pages);
2600         total_swap_pages -= p->pages;
2601         p->flags &= ~SWP_WRITEOK;
2602         spin_unlock(&p->lock);
2603         spin_unlock(&swap_lock);
2604 
2605         disable_swap_slots_cache_lock();
2606 
2607         set_current_oom_origin();
2608         err = try_to_unuse(p->type, false, 0); /* force unuse all pages */
2609         clear_current_oom_origin();
2610 
2611         if (err) {
2612                 /* re-insert swap space back into swap_list */
2613                 reinsert_swap_info(p);
2614                 reenable_swap_slots_cache_unlock();
2615                 goto out_dput;
2616         }
2617 
2618         reenable_swap_slots_cache_unlock();
2619 
2620         flush_work(&p->discard_work);
2621 
2622         destroy_swap_extents(p);
2623         if (p->flags & SWP_CONTINUED)
2624                 free_swap_count_continuations(p);
2625 
2626         if (!p->bdev || !blk_queue_nonrot(bdev_get_queue(p->bdev)))
2627                 atomic_dec(&nr_rotate_swap);
2628 
2629         mutex_lock(&swapon_mutex);
2630         spin_lock(&swap_lock);
2631         spin_lock(&p->lock);
2632         drain_mmlist();
2633 
2634         /* wait for anyone still in scan_swap_map */
2635         p->highest_bit = 0;             /* cuts scans short */
2636         while (p->flags >= SWP_SCANNING) {
2637                 spin_unlock(&p->lock);
2638                 spin_unlock(&swap_lock);
2639                 schedule_timeout_uninterruptible(1);
2640                 spin_lock(&swap_lock);
2641                 spin_lock(&p->lock);
2642         }
2643 
2644         swap_file = p->swap_file;
2645         old_block_size = p->old_block_size;
2646         p->swap_file = NULL;
2647         p->max = 0;
2648         swap_map = p->swap_map;
2649         p->swap_map = NULL;
2650         cluster_info = p->cluster_info;
2651         p->cluster_info = NULL;
2652         frontswap_map = frontswap_map_get(p);
2653         spin_unlock(&p->lock);
2654         spin_unlock(&swap_lock);
2655         frontswap_invalidate_area(p->type);
2656         frontswap_map_set(p, NULL);
2657         mutex_unlock(&swapon_mutex);
2658         free_percpu(p->percpu_cluster);
2659         p->percpu_cluster = NULL;
2660         vfree(swap_map);
2661         kvfree(cluster_info);
2662         kvfree(frontswap_map);
2663         /* Destroy swap account information */
2664         swap_cgroup_swapoff(p->type);
2665         exit_swap_address_space(p->type);
2666 
2667         inode = mapping->host;
2668         if (S_ISBLK(inode->i_mode)) {
2669                 struct block_device *bdev = I_BDEV(inode);
2670                 set_blocksize(bdev, old_block_size);
2671                 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2672         } else {
2673                 inode_lock(inode);
2674                 inode->i_flags &= ~S_SWAPFILE;
2675                 inode_unlock(inode);
2676         }
2677         filp_close(swap_file, NULL);
2678 
2679         /*
2680          * Clear the SWP_USED flag after all resources are freed so that swapon
2681          * can reuse this swap_info in alloc_swap_info() safely.  It is ok to
2682          * not hold p->lock after we cleared its SWP_WRITEOK.
2683          */
2684         spin_lock(&swap_lock);
2685         p->flags = 0;
2686         spin_unlock(&swap_lock);
2687 
2688         err = 0;
2689         atomic_inc(&proc_poll_event);
2690         wake_up_interruptible(&proc_poll_wait);
2691 
2692 out_dput:
2693         filp_close(victim, NULL);
2694 out:
2695         putname(pathname);
2696         return err;
2697 }
2698 
2699 #ifdef CONFIG_PROC_FS
2700 static __poll_t swaps_poll(struct file *file, poll_table *wait)
2701 {
2702         struct seq_file *seq = file->private_data;
2703 
2704         poll_wait(file, &proc_poll_wait, wait);
2705 
2706         if (seq->poll_event != atomic_read(&proc_poll_event)) {
2707                 seq->poll_event = atomic_read(&proc_poll_event);
2708                 return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
2709         }
2710 
2711         return EPOLLIN | EPOLLRDNORM;
2712 }
2713 
2714 /* iterator */
2715 static void *swap_start(struct seq_file *swap, loff_t *pos)
2716 {
2717         struct swap_info_struct *si;
2718         int type;
2719         loff_t l = *pos;
2720 
2721         mutex_lock(&swapon_mutex);
2722 
2723         if (!l)
2724                 return SEQ_START_TOKEN;
2725 
2726         for (type = 0; type < nr_swapfiles; type++) {
2727                 smp_rmb();      /* read nr_swapfiles before swap_info[type] */
2728                 si = swap_info[type];
2729                 if (!(si->flags & SWP_USED) || !si->swap_map)
2730                         continue;
2731                 if (!--l)
2732                         return si;
2733         }
2734 
2735         return NULL;
2736 }
2737 
2738 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2739 {
2740         struct swap_info_struct *si = v;
2741         int type;
2742 
2743         if (v == SEQ_START_TOKEN)
2744                 type = 0;
2745         else
2746                 type = si->type + 1;
2747 
2748         for (; type < nr_swapfiles; type++) {
2749                 smp_rmb();      /* read nr_swapfiles before swap_info[type] */
2750                 si = swap_info[type];
2751                 if (!(si->flags & SWP_USED) || !si->swap_map)
2752                         continue;
2753                 ++*pos;
2754                 return si;
2755         }
2756 
2757         return NULL;
2758 }
2759 
2760 static void swap_stop(struct seq_file *swap, void *v)
2761 {
2762         mutex_unlock(&swapon_mutex);
2763 }
2764 
2765 static int swap_show(struct seq_file *swap, void *v)
2766 {
2767         struct swap_info_struct *si = v;
2768         struct file *file;
2769         int len;
2770 
2771         if (si == SEQ_START_TOKEN) {
2772                 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2773                 return 0;
2774         }
2775 
2776         file = si->swap_file;
2777         len = seq_file_path(swap, file, " \t\n\\");
2778         seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
2779                         len < 40 ? 40 - len : 1, " ",
2780                         S_ISBLK(file_inode(file)->i_mode) ?
2781                                 "partition" : "file\t",
2782                         si->pages << (PAGE_SHIFT - 10),
2783                         si->inuse_pages << (PAGE_SHIFT - 10),
2784                         si->prio);
2785         return 0;
2786 }
2787 
2788 static const struct seq_operations swaps_op = {
2789         .start =        swap_start,
2790         .next =         swap_next,
2791         .stop =         swap_stop,
2792         .show =         swap_show
2793 };
2794 
2795 static int swaps_open(struct inode *inode, struct file *file)
2796 {
2797         struct seq_file *seq;
2798         int ret;
2799 
2800         ret = seq_open(file, &swaps_op);
2801         if (ret)
2802                 return ret;
2803 
2804         seq = file->private_data;
2805         seq->poll_event = atomic_read(&proc_poll_event);
2806         return 0;
2807 }
2808 
2809 static const struct file_operations proc_swaps_operations = {
2810         .open           = swaps_open,
2811         .read           = seq_read,
2812         .llseek         = seq_lseek,
2813         .release        = seq_release,
2814         .poll           = swaps_poll,
2815 };
2816 
2817 static int __init procswaps_init(void)
2818 {
2819         proc_create("swaps", 0, NULL, &proc_swaps_operations);
2820         return 0;
2821 }
2822 __initcall(procswaps_init);
2823 #endif /* CONFIG_PROC_FS */
2824 
2825 #ifdef MAX_SWAPFILES_CHECK
2826 static int __init max_swapfiles_check(void)
2827 {
2828         MAX_SWAPFILES_CHECK();
2829         return 0;
2830 }
2831 late_initcall(max_swapfiles_check);
2832 #endif
2833 
2834 static struct swap_info_struct *alloc_swap_info(void)
2835 {
2836         struct swap_info_struct *p;
2837         unsigned int type;
2838         int i;
2839 
2840         p = kzalloc(sizeof(*p), GFP_KERNEL);
2841         if (!p)
2842                 return ERR_PTR(-ENOMEM);
2843 
2844         spin_lock(&swap_lock);
2845         for (type = 0; type < nr_swapfiles; type++) {
2846                 if (!(swap_info[type]->flags & SWP_USED))
2847                         break;
2848         }
2849         if (type >= MAX_SWAPFILES) {
2850                 spin_unlock(&swap_lock);
2851                 kfree(p);
2852                 return ERR_PTR(-EPERM);
2853         }
2854         if (type >= nr_swapfiles) {
2855                 p->type = type;
2856                 swap_info[type] = p;
2857                 /*
2858                  * Write swap_info[type] before nr_swapfiles, in case a
2859                  * racing procfs swap_start() or swap_next() is reading them.
2860                  * (We never shrink nr_swapfiles, we never free this entry.)
2861                  */
2862                 smp_wmb();
2863                 nr_swapfiles++;
2864         } else {
2865                 kfree(p);
2866                 p = swap_info[type];
2867                 /*
2868                  * Do not memset this entry: a racing procfs swap_next()
2869                  * would be relying on p->type to remain valid.
2870                  */
2871         }
2872         INIT_LIST_HEAD(&p->first_swap_extent.list);
2873         plist_node_init(&p->list, 0);
2874         for_each_node(i)
2875                 plist_node_init(&p->avail_lists[i], 0);
2876         p->flags = SWP_USED;
2877         spin_unlock(&swap_lock);
2878         spin_lock_init(&p->lock);
2879         spin_lock_init(&p->cont_lock);
2880 
2881         return p;
2882 }
2883 
2884 static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2885 {
2886         int error;
2887 
2888         if (S_ISBLK(inode->i_mode)) {
2889                 p->bdev = bdgrab(I_BDEV(inode));
2890                 error = blkdev_get(p->bdev,
2891                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL, p);
2892                 if (error < 0) {
2893                         p->bdev = NULL;
2894                         return error;
2895                 }
2896                 p->old_block_size = block_size(p->bdev);
2897                 error = set_blocksize(p->bdev, PAGE_SIZE);
2898                 if (error < 0)
2899                         return error;
2900                 p->flags |= SWP_BLKDEV;
2901         } else if (S_ISREG(inode->i_mode)) {
2902                 p->bdev = inode->i_sb->s_bdev;
2903                 inode_lock(inode);
2904                 if (IS_SWAPFILE(inode))
2905                         return -EBUSY;
2906         } else
2907                 return -EINVAL;
2908 
2909         return 0;
2910 }
2911 
2912 static unsigned long read_swap_header(struct swap_info_struct *p,
2913                                         union swap_header *swap_header,
2914                                         struct inode *inode)
2915 {
2916         int i;
2917         unsigned long maxpages;
2918         unsigned long swapfilepages;
2919         unsigned long last_page;
2920 
2921         if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
2922                 pr_err("Unable to find swap-space signature\n");
2923                 return 0;
2924         }
2925 
2926         /* swap partition endianess hack... */
2927         if (swab32(swap_header->info.version) == 1) {
2928                 swab32s(&swap_header->info.version);
2929                 swab32s(&swap_header->info.last_page);
2930                 swab32s(&swap_header->info.nr_badpages);
2931                 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2932                         return 0;
2933                 for (i = 0; i < swap_header->info.nr_badpages; i++)
2934                         swab32s(&swap_header->info.badpages[i]);
2935         }
2936         /* Check the swap header's sub-version */
2937         if (swap_header->info.version != 1) {
2938                 pr_warn("Unable to handle swap header version %d\n",
2939                         swap_header->info.version);
2940                 return 0;
2941         }
2942 
2943         p->lowest_bit  = 1;
2944         p->cluster_next = 1;
2945         p->cluster_nr = 0;
2946 
2947         /*
2948          * Find out how many pages are allowed for a single swap
2949          * device. There are two limiting factors: 1) the number
2950          * of bits for the swap offset in the swp_entry_t type, and
2951          * 2) the number of bits in the swap pte as defined by the
2952          * different architectures. In order to find the
2953          * largest possible bit mask, a swap entry with swap type 0
2954          * and swap offset ~0UL is created, encoded to a swap pte,
2955          * decoded to a swp_entry_t again, and finally the swap
2956          * offset is extracted. This will mask all the bits from
2957          * the initial ~0UL mask that can't be encoded in either
2958          * the swp_entry_t or the architecture definition of a
2959          * swap pte.
2960          */
2961         maxpages = swp_offset(pte_to_swp_entry(
2962                         swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2963         last_page = swap_header->info.last_page;
2964         if (!last_page) {
2965                 pr_warn("Empty swap-file\n");
2966                 return 0;
2967         }
2968         if (last_page > maxpages) {
2969                 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
2970                         maxpages << (PAGE_SHIFT - 10),
2971                         last_page << (PAGE_SHIFT - 10));
2972         }
2973         if (maxpages > last_page) {
2974                 maxpages = last_page + 1;
2975                 /* p->max is an unsigned int: don't overflow it */
2976                 if ((unsigned int)maxpages == 0)
2977                         maxpages = UINT_MAX;
2978         }
2979         p->highest_bit = maxpages - 1;
2980 
2981         if (!maxpages)
2982                 return 0;
2983         swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2984         if (swapfilepages && maxpages > swapfilepages) {
2985                 pr_warn("Swap area shorter than signature indicates\n");
2986                 return 0;
2987         }
2988         if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
2989                 return 0;
2990         if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2991                 return 0;
2992 
2993         return maxpages;
2994 }
2995 
2996 #define SWAP_CLUSTER_INFO_COLS                                          \
2997         DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
2998 #define SWAP_CLUSTER_SPACE_COLS                                         \
2999         DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
3000 #define SWAP_CLUSTER_COLS                                               \
3001         max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
3002 
3003 static int setup_swap_map_and_extents(struct swap_info_struct *p,
3004                                         union swap_header *swap_header,
3005                                         unsigned char *swap_map,
3006                                         struct swap_cluster_info *cluster_info,
3007                                         unsigned long maxpages,
3008                                         sector_t *span)
3009 {
3010         unsigned int j, k;
3011         unsigned int nr_good_pages;
3012         int nr_extents;
3013         unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
3014         unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS;
3015         unsigned long i, idx;
3016 
3017         nr_good_pages = maxpages - 1;   /* omit header page */
3018 
3019         cluster_list_init(&p->free_clusters);
3020         cluster_list_init(&p->discard_clusters);
3021 
3022         for (i = 0; i < swap_header->info.nr_badpages; i++) {
3023                 unsigned int page_nr = swap_header->info.badpages[i];
3024                 if (page_nr == 0 || page_nr > swap_header->info.last_page)
3025                         return -EINVAL;
3026                 if (page_nr < maxpages) {
3027                         swap_map[page_nr] = SWAP_MAP_BAD;
3028                         nr_good_pages--;
3029                         /*
3030                          * Haven't marked the cluster free yet, no list
3031                          * operation involved
3032                          */
3033                         inc_cluster_info_page(p, cluster_info, page_nr);
3034                 }
3035         }
3036 
3037         /* Haven't marked the cluster free yet, no list operation involved */
3038         for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
3039                 inc_cluster_info_page(p, cluster_info, i);
3040 
3041         if (nr_good_pages) {
3042                 swap_map[0] = SWAP_MAP_BAD;
3043                 /*
3044                  * Not mark the cluster free yet, no list
3045                  * operation involved
3046                  */
3047                 inc_cluster_info_page(p, cluster_info, 0);
3048                 p->max = maxpages;
3049                 p->pages = nr_good_pages;
3050                 nr_extents = setup_swap_extents(p, span);
3051                 if (nr_extents < 0)
3052                         return nr_extents;
3053                 nr_good_pages = p->pages;
3054         }
3055         if (!nr_good_pages) {
3056                 pr_warn("Empty swap-file\n");
3057                 return -EINVAL;
3058         }
3059 
3060         if (!cluster_info)
3061                 return nr_extents;
3062 
3063 
3064         /*
3065          * Reduce false cache line sharing between cluster_info and
3066          * sharing same address space.
3067          */
3068         for (k = 0; k < SWAP_CLUSTER_COLS; k++) {
3069                 j = (k + col) % SWAP_CLUSTER_COLS;
3070                 for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) {
3071                         idx = i * SWAP_CLUSTER_COLS + j;
3072                         if (idx >= nr_clusters)
3073                                 continue;
3074                         if (cluster_count(&cluster_info[idx]))
3075                                 continue;
3076                         cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
3077                         cluster_list_add_tail(&p->free_clusters, cluster_info,
3078                                               idx);
3079                 }
3080         }
3081         return nr_extents;
3082 }
3083 
3084 /*
3085  * Helper to sys_swapon determining if a given swap
3086  * backing device queue supports DISCARD operations.
3087  */
3088 static bool swap_discardable(struct swap_info_struct *si)
3089 {
3090         struct request_queue *q = bdev_get_queue(si->bdev);
3091 
3092         if (!q || !blk_queue_discard(q))
3093                 return false;
3094 
3095         return true;
3096 }
3097 
3098 SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
3099 {
3100         struct swap_info_struct *p;
3101         struct filename *name;
3102         struct file *swap_file = NULL;
3103         struct address_space *mapping;
3104         int prio;
3105         int error;
3106         union swap_header *swap_header;
3107         int nr_extents;
3108         sector_t span;
3109         unsigned long maxpages;
3110         unsigned char *swap_map = NULL;
3111         struct swap_cluster_info *cluster_info = NULL;
3112         unsigned long *frontswap_map = NULL;
3113         struct page *page = NULL;
3114         struct inode *inode = NULL;
3115 
3116         if (swap_flags & ~SWAP_FLAGS_VALID)
3117                 return -EINVAL;
3118 
3119         if (!capable(CAP_SYS_ADMIN))
3120                 return -EPERM;
3121 
3122         if (!swap_avail_heads)
3123                 return -ENOMEM;
3124 
3125         p = alloc_swap_info();
3126         if (IS_ERR(p))
3127                 return PTR_ERR(p);
3128 
3129         INIT_WORK(&p->discard_work, swap_discard_work);
3130 
3131         name = getname(specialfile);
3132         if (IS_ERR(name)) {
3133                 error = PTR_ERR(name);
3134                 name = NULL;
3135                 goto bad_swap;
3136         }
3137         swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
3138         if (IS_ERR(swap_file)) {
3139                 error = PTR_ERR(swap_file);
3140                 swap_file = NULL;
3141                 goto bad_swap;
3142         }
3143 
3144         p->swap_file = swap_file;
3145         mapping = swap_file->f_mapping;
3146         inode = mapping->host;
3147 
3148         /* If S_ISREG(inode->i_mode) will do inode_lock(inode); */
3149         error = claim_swapfile(p, inode);
3150         if (unlikely(error))
3151                 goto bad_swap;
3152 
3153         /*
3154          * Read the swap header.
3155          */
3156         if (!mapping->a_ops->readpage) {
3157                 error = -EINVAL;
3158                 goto bad_swap;
3159         }
3160         page = read_mapping_page(mapping, 0, swap_file);
3161         if (IS_ERR(page)) {
3162                 error = PTR_ERR(page);
3163                 goto bad_swap;
3164         }
3165         swap_header = kmap(page);
3166 
3167         maxpages = read_swap_header(p, swap_header, inode);
3168         if (unlikely(!maxpages)) {
3169                 error = -EINVAL;
3170                 goto bad_swap;
3171         }
3172 
3173         /* OK, set up the swap map and apply the bad block list */
3174         swap_map = vzalloc(maxpages);
3175         if (!swap_map) {
3176                 error = -ENOMEM;
3177                 goto bad_swap;
3178         }
3179 
3180         if (bdi_cap_stable_pages_required(inode_to_bdi(inode)))
3181                 p->flags |= SWP_STABLE_WRITES;
3182 
3183         if (bdi_cap_synchronous_io(inode_to_bdi(inode)))
3184                 p->flags |= SWP_SYNCHRONOUS_IO;
3185 
3186         if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) {
3187                 int cpu;
3188                 unsigned long ci, nr_cluster;
3189 
3190                 p->flags |= SWP_SOLIDSTATE;
3191                 /*
3192                  * select a random position to start with to help wear leveling
3193                  * SSD
3194                  */
3195                 p->cluster_next = 1 + (prandom_u32() % p->highest_bit);
3196                 nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
3197 
3198                 cluster_info = kvzalloc(nr_cluster * sizeof(*cluster_info),
3199                                         GFP_KERNEL);
3200                 if (!cluster_info) {
3201                         error = -ENOMEM;
3202                         goto bad_swap;
3203                 }
3204 
3205                 for (ci = 0; ci < nr_cluster; ci++)
3206                         spin_lock_init(&((cluster_info + ci)->lock));
3207 
3208                 p->percpu_cluster = alloc_percpu(struct percpu_cluster);
3209                 if (!p->percpu_cluster) {
3210                         error = -ENOMEM;
3211                         goto bad_swap;
3212                 }
3213                 for_each_possible_cpu(cpu) {
3214                         struct percpu_cluster *cluster;
3215                         cluster = per_cpu_ptr(p->percpu_cluster, cpu);
3216                         cluster_set_null(&cluster->index);
3217                 }
3218         } else
3219                 atomic_inc(&nr_rotate_swap);
3220 
3221         error = swap_cgroup_swapon(p->type, maxpages);
3222         if (error)
3223                 goto bad_swap;
3224 
3225         nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
3226                 cluster_info, maxpages, &span);
3227         if (unlikely(nr_extents < 0)) {
3228                 error = nr_extents;
3229                 goto bad_swap;
3230         }
3231         /* frontswap enabled? set up bit-per-page map for frontswap */
3232         if (IS_ENABLED(CONFIG_FRONTSWAP))
3233                 frontswap_map = kvzalloc(BITS_TO_LONGS(maxpages) * sizeof(long),
3234                                          GFP_KERNEL);
3235 
3236         if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) {
3237                 /*
3238                  * When discard is enabled for swap with no particular
3239                  * policy flagged, we set all swap discard flags here in
3240                  * order to sustain backward compatibility with older
3241                  * swapon(8) releases.
3242                  */
3243                 p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
3244                              SWP_PAGE_DISCARD);
3245 
3246                 /*
3247                  * By flagging sys_swapon, a sysadmin can tell us to
3248                  * either do single-time area discards only, or to just
3249                  * perform discards for released swap page-clusters.
3250                  * Now it's time to adjust the p->flags accordingly.
3251                  */
3252                 if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
3253                         p->flags &= ~SWP_PAGE_DISCARD;
3254                 else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
3255                         p->flags &= ~SWP_AREA_DISCARD;
3256 
3257                 /* issue a swapon-time discard if it's still required */
3258                 if (p->flags & SWP_AREA_DISCARD) {
3259                         int err = discard_swap(p);
3260                         if (unlikely(err))
3261                                 pr_err("swapon: discard_swap(%p): %d\n",
3262                                         p, err);
3263                 }
3264         }
3265 
3266         error = init_swap_address_space(p->type, maxpages);
3267         if (error)
3268                 goto bad_swap;
3269 
3270         mutex_lock(&swapon_mutex);
3271         prio = -1;
3272         if (swap_flags & SWAP_FLAG_PREFER)
3273                 prio =
3274                   (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
3275         enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
3276 
3277         pr_info("Adding %uk swap on %s.  Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
3278                 p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
3279                 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
3280                 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
3281                 (p->flags & SWP_DISCARDABLE) ? "D" : "",
3282                 (p->flags & SWP_AREA_DISCARD) ? "s" : "",
3283                 (p->flags & SWP_PAGE_DISCARD) ? "c" : "",
3284                 (frontswap_map) ? "FS" : "");
3285 
3286         mutex_unlock(&swapon_mutex);
3287         atomic_inc(&proc_poll_event);
3288         wake_up_interruptible(&proc_poll_wait);
3289 
3290         if (S_ISREG(inode->i_mode))
3291                 inode->i_flags |= S_SWAPFILE;
3292         error = 0;
3293         goto out;
3294 bad_swap:
3295         free_percpu(p->percpu_cluster);
3296         p->percpu_cluster = NULL;
3297         if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
3298                 set_blocksize(p->bdev, p->old_block_size);
3299                 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
3300         }
3301         destroy_swap_extents(p);
3302         swap_cgroup_swapoff(p->type);
3303         spin_lock(&swap_lock);
3304         p->swap_file = NULL;
3305         p->flags = 0;
3306         spin_unlock(&swap_lock);
3307         vfree(swap_map);
3308         kvfree(cluster_info);
3309         kvfree(frontswap_map);
3310         if (swap_file) {
3311                 if (inode && S_ISREG(inode->i_mode)) {
3312                         inode_unlock(inode);
3313                         inode = NULL;
3314                 }
3315                 filp_close(swap_file, NULL);
3316         }
3317 out:
3318         if (page && !IS_ERR(page)) {
3319                 kunmap(page);
3320                 put_page(page);
3321         }
3322         if (name)
3323                 putname(name);
3324         if (inode && S_ISREG(inode->i_mode))
3325                 inode_unlock(inode);
3326         if (!error)
3327                 enable_swap_slots_cache();
3328         return error;
3329 }
3330 
3331 void si_swapinfo(struct sysinfo *val)
3332 {
3333         unsigned int type;
3334         unsigned long nr_to_be_unused = 0;
3335 
3336         spin_lock(&swap_lock);
3337         for (type = 0; type < nr_swapfiles; type++) {
3338                 struct swap_info_struct *si = swap_info[type];
3339 
3340                 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
3341                         nr_to_be_unused += si->inuse_pages;
3342         }
3343         val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
3344         val->totalswap = total_swap_pages + nr_to_be_unused;
3345         spin_unlock(&swap_lock);
3346 }
3347 
3348 /*
3349  * Verify that a swap entry is valid and increment its swap map count.
3350  *
3351  * Returns error code in following case.
3352  * - success -> 0
3353  * - swp_entry is invalid -> EINVAL
3354  * - swp_entry is migration entry -> EINVAL
3355  * - swap-cache reference is requested but there is already one. -> EEXIST
3356  * - swap-cache reference is requested but the entry is not used. -> ENOENT
3357  * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
3358  */
3359 static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
3360 {
3361         struct swap_info_struct *p;
3362         struct swap_cluster_info *ci;
3363         unsigned long offset, type;
3364         unsigned char count;
3365         unsigned char has_cache;
3366         int err = -EINVAL;
3367 
3368         if (non_swap_entry(entry))
3369                 goto out;
3370 
3371         type = swp_type(entry);
3372         if (type >= nr_swapfiles)
3373                 goto bad_file;
3374         p = swap_info[type];
3375         offset = swp_offset(entry);
3376         if (unlikely(offset >= p->max))
3377                 goto out;
3378 
3379         ci = lock_cluster_or_swap_info(p, offset);
3380 
3381         count = p->swap_map[offset];
3382 
3383         /*
3384          * swapin_readahead() doesn't check if a swap entry is valid, so the
3385          * swap entry could be SWAP_MAP_BAD. Check here with lock held.
3386          */
3387         if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
3388                 err = -ENOENT;
3389                 goto unlock_out;
3390         }
3391 
3392         has_cache = count & SWAP_HAS_CACHE;
3393         count &= ~SWAP_HAS_CACHE;
3394         err = 0;
3395 
3396         if (usage == SWAP_HAS_CACHE) {
3397 
3398                 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
3399                 if (!has_cache && count)
3400                         has_cache = SWAP_HAS_CACHE;
3401                 else if (has_cache)             /* someone else added cache */
3402                         err = -EEXIST;
3403                 else                            /* no users remaining */
3404                         err = -ENOENT;
3405 
3406         } else if (count || has_cache) {
3407 
3408                 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
3409                         count += usage;
3410                 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
3411                         err = -EINVAL;
3412                 else if (swap_count_continued(p, offset, count))
3413                         count = COUNT_CONTINUED;
3414                 else
3415                         err = -ENOMEM;
3416         } else
3417                 err = -ENOENT;                  /* unused swap entry */
3418 
3419         p->swap_map[offset] = count | has_cache;
3420 
3421 unlock_out:
3422         unlock_cluster_or_swap_info(p, ci);
3423 out:
3424         return err;
3425 
3426 bad_file:
3427         pr_err("swap_dup: %s%08lx\n", Bad_file, entry.val);
3428         goto out;
3429 }
3430 
3431 /*
3432  * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3433  * (in which case its reference count is never incremented).
3434  */
3435 void swap_shmem_alloc(swp_entry_t entry)
3436 {
3437         __swap_duplicate(entry, SWAP_MAP_SHMEM);
3438 }
3439 
3440 /*
3441  * Increase reference count of swap entry by 1.
3442  * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3443  * but could not be atomically allocated.  Returns 0, just as if it succeeded,
3444  * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3445  * might occur if a page table entry has got corrupted.
3446  */
3447 int swap_duplicate(swp_entry_t entry)
3448 {
3449         int err = 0;
3450 
3451         while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
3452                 err = add_swap_count_continuation(entry, GFP_ATOMIC);
3453         return err;
3454 }
3455 
3456 /*
3457  * @entry: swap entry for which we allocate swap cache.
3458  *
3459  * Called when allocating swap cache for existing swap entry,
3460  * This can return error codes. Returns 0 at success.
3461  * -EBUSY means there is a swap cache.
3462  * Note: return code is different from swap_duplicate().
3463  */
3464 int swapcache_prepare(swp_entry_t entry)
3465 {
3466         return __swap_duplicate(entry, SWAP_HAS_CACHE);
3467 }
3468 
3469 struct swap_info_struct *swp_swap_info(swp_entry_t entry)
3470 {
3471         return swap_info[swp_type(entry)];
3472 }
3473 
3474 struct swap_info_struct *page_swap_info(struct page *page)
3475 {
3476         swp_entry_t entry = { .val = page_private(page) };
3477         return swp_swap_info(entry);
3478 }
3479 
3480 /*
3481  * out-of-line __page_file_ methods to avoid include hell.
3482  */
3483 struct address_space *__page_file_mapping(struct page *page)
3484 {
3485         return page_swap_info(page)->swap_file->f_mapping;
3486 }
3487 EXPORT_SYMBOL_GPL(__page_file_mapping);
3488 
3489 pgoff_t __page_file_index(struct page *page)
3490 {
3491         swp_entry_t swap = { .val = page_private(page) };
3492         return swp_offset(swap);
3493 }
3494 EXPORT_SYMBOL_GPL(__page_file_index);
3495 
3496 /*
3497  * add_swap_count_continuation - called when a swap count is duplicated
3498  * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3499  * page of the original vmalloc'ed swap_map, to hold the continuation count
3500  * (for that entry and for its neighbouring PAGE_SIZE swap entries).  Called
3501  * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3502  *
3503  * These continuation pages are seldom referenced: the common paths all work
3504  * on the original swap_map, only referring to a continuation page when the
3505  * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3506  *
3507  * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3508  * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3509  * can be called after dropping locks.
3510  */
3511 int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
3512 {
3513         struct swap_info_struct *si;
3514         struct swap_cluster_info *ci;
3515         struct page *head;
3516         struct page *page;
3517         struct page *list_page;
3518         pgoff_t offset;
3519         unsigned char count;
3520 
3521         /*
3522          * When debugging, it's easier to use __GFP_ZERO here; but it's better
3523          * for latency not to zero a page while GFP_ATOMIC and holding locks.
3524          */
3525         page = alloc_page(gfp_mask | __GFP_HIGHMEM);
3526 
3527         si = swap_info_get(entry);
3528         if (!si) {
3529                 /*
3530                  * An acceptable race has occurred since the failing
3531                  * __swap_duplicate(): the swap entry has been freed,
3532                  * perhaps even the whole swap_map cleared for swapoff.
3533                  */
3534                 goto outer;
3535         }
3536 
3537         offset = swp_offset(entry);
3538 
3539         ci = lock_cluster(si, offset);
3540 
3541         count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
3542 
3543         if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
3544                 /*
3545                  * The higher the swap count, the more likely it is that tasks
3546                  * will race to add swap count continuation: we need to avoid
3547                  * over-provisioning.
3548                  */
3549                 goto out;
3550         }
3551 
3552         if (!page) {
3553                 unlock_cluster(ci);
3554                 spin_unlock(&si->lock);
3555                 return -ENOMEM;
3556         }
3557 
3558         /*
3559          * We are fortunate that although vmalloc_to_page uses pte_offset_map,
3560          * no architecture is using highmem pages for kernel page tables: so it
3561          * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
3562          */
3563         head = vmalloc_to_page(si->swap_map + offset);
3564         offset &= ~PAGE_MASK;
3565 
3566         spin_lock(&si->cont_lock);
3567         /*
3568          * Page allocation does not initialize the page's lru field,
3569          * but it does always reset its private field.
3570          */
3571         if (!page_private(head)) {
3572                 BUG_ON(count & COUNT_CONTINUED);
3573                 INIT_LIST_HEAD(&head->lru);
3574                 set_page_private(head, SWP_CONTINUED);
3575                 si->flags |= SWP_CONTINUED;
3576         }
3577 
3578         list_for_each_entry(list_page, &head->lru, lru) {
3579                 unsigned char *map;
3580 
3581                 /*
3582                  * If the previous map said no continuation, but we've found
3583                  * a continuation page, free our allocation and use this one.
3584                  */
3585                 if (!(count & COUNT_CONTINUED))
3586                         goto out_unlock_cont;
3587 
3588                 map = kmap_atomic(list_page) + offset;
3589                 count = *map;
3590                 kunmap_atomic(map);
3591 
3592                 /*
3593                  * If this continuation count now has some space in it,
3594                  * free our allocation and use this one.
3595                  */
3596                 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
3597                         goto out_unlock_cont;
3598         }
3599 
3600         list_add_tail(&page->lru, &head->lru);
3601         page = NULL;                    /* now it's attached, don't free it */
3602 out_unlock_cont:
3603         spin_unlock(&si->cont_lock);
3604 out:
3605         unlock_cluster(ci);
3606         spin_unlock(&si->lock);
3607 outer:
3608         if (page)
3609                 __free_page(page);
3610         return 0;
3611 }
3612 
3613 /*
3614  * swap_count_continued - when the original swap_map count is incremented
3615  * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3616  * into, carry if so, or else fail until a new continuation page is allocated;
3617  * when the original swap_map count is decremented from 0 with continuation,
3618  * borrow from the continuation and report whether it still holds more.
3619  * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3620  * lock.
3621  */
3622 static bool swap_count_continued(struct swap_info_struct *si,
3623                                  pgoff_t offset, unsigned char count)
3624 {
3625         struct page *head;
3626         struct page *page;
3627         unsigned char *map;
3628         bool ret;
3629 
3630         head = vmalloc_to_page(si->swap_map + offset);
3631         if (page_private(head) != SWP_CONTINUED) {
3632                 BUG_ON(count & COUNT_CONTINUED);
3633                 return false;           /* need to add count continuation */
3634         }
3635 
3636         spin_lock(&si->cont_lock);
3637         offset &= ~PAGE_MASK;
3638         page = list_entry(head->lru.next, struct page, lru);
3639         map = kmap_atomic(page) + offset;
3640 
3641         if (count == SWAP_MAP_MAX)      /* initial increment from swap_map */
3642                 goto init_map;          /* jump over SWAP_CONT_MAX checks */
3643 
3644         if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
3645                 /*
3646                  * Think of how you add 1 to 999
3647                  */
3648                 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
3649                         kunmap_atomic(map);
3650                         page = list_entry(page->lru.next, struct page, lru);
3651                         BUG_ON(page == head);
3652                         map = kmap_atomic(page) + offset;
3653                 }
3654                 if (*map == SWAP_CONT_MAX) {
3655                         kunmap_atomic(map);
3656                         page = list_entry(page->lru.next, struct page, lru);
3657                         if (page == head) {
3658                                 ret = false;    /* add count continuation */
3659                                 goto out;
3660                         }
3661                         map = kmap_atomic(page) + offset;
3662 init_map:               *map = 0;               /* we didn't zero the page */
3663                 }
3664                 *map += 1;
3665                 kunmap_atomic(map);
3666                 page = list_entry(page->lru.prev, struct page, lru);
3667                 while (page != head) {
3668                         map = kmap_atomic(page) + offset;
3669                         *map = COUNT_CONTINUED;
3670                         kunmap_atomic(map);
3671                         page = list_entry(page->lru.prev, struct page, lru);
3672                 }
3673                 ret = true;                     /* incremented */
3674 
3675         } else {                                /* decrementing */
3676                 /*
3677                  * Think of how you subtract 1 from 1000
3678                  */
3679                 BUG_ON(count != COUNT_CONTINUED);
3680                 while (*map == COUNT_CONTINUED) {
3681                         kunmap_atomic(map);
3682                         page = list_entry(page->lru.next, struct page, lru);
3683                         BUG_ON(page == head);
3684                         map = kmap_atomic(page) + offset;
3685                 }
3686                 BUG_ON(*map == 0);
3687                 *map -= 1;
3688                 if (*map == 0)
3689                         count = 0;
3690                 kunmap_atomic(map);
3691                 page = list_entry(page->lru.prev, struct page, lru);
3692                 while (page != head) {
3693                         map = kmap_atomic(page) + offset;
3694                         *map = SWAP_CONT_MAX | count;
3695                         count = COUNT_CONTINUED;
3696                         kunmap_atomic(map);
3697                         page = list_entry(page->lru.prev, struct page, lru);
3698                 }
3699                 ret = count == COUNT_CONTINUED;
3700         }
3701 out:
3702         spin_unlock(&si->cont_lock);
3703         return ret;
3704 }
3705 
3706 /*
3707  * free_swap_count_continuations - swapoff free all the continuation pages
3708  * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3709  */
3710 static void free_swap_count_continuations(struct swap_info_struct *si)
3711 {
3712         pgoff_t offset;
3713 
3714         for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
3715                 struct page *head;
3716                 head = vmalloc_to_page(si->swap_map + offset);
3717                 if (page_private(head)) {
3718                         struct page *page, *next;
3719 
3720                         list_for_each_entry_safe(page, next, &head->lru, lru) {
3721                                 list_del(&page->lru);
3722                                 __free_page(page);
3723                         }
3724                 }
3725         }
3726 }
3727 
3728 static int __init swapfile_init(void)
3729 {
3730         int nid;
3731 
3732         swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head),
3733                                          GFP_KERNEL);
3734         if (!swap_avail_heads) {
3735                 pr_emerg("Not enough memory for swap heads, swap is disabled\n");
3736                 return -ENOMEM;
3737         }
3738 
3739         for_each_node(nid)
3740                 plist_head_init(&swap_avail_heads[nid]);
3741 
3742         return 0;
3743 }
3744 subsys_initcall(swapfile_init);
3745 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp