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

TOMOYO Linux Cross Reference
Linux/mm/swapfile.c

Version: ~ [ linux-5.1-rc1 ] ~ [ linux-5.0.3 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.30 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.107 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.164 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.176 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.136 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.63 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ 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 static 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 struct swap_info_struct *swap_type_to_swap_info(int type)
102 {
103         if (type >= READ_ONCE(nr_swapfiles))
104                 return NULL;
105 
106         smp_rmb();      /* Pairs with smp_wmb in alloc_swap_info. */
107         return READ_ONCE(swap_info[type]);
108 }
109 
110 static inline unsigned char swap_count(unsigned char ent)
111 {
112         return ent & ~SWAP_HAS_CACHE;   /* may include COUNT_CONTINUED flag */
113 }
114 
115 /* Reclaim the swap entry anyway if possible */
116 #define TTRS_ANYWAY             0x1
117 /*
118  * Reclaim the swap entry if there are no more mappings of the
119  * corresponding page
120  */
121 #define TTRS_UNMAPPED           0x2
122 /* Reclaim the swap entry if swap is getting full*/
123 #define TTRS_FULL               0x4
124 
125 /* returns 1 if swap entry is freed */
126 static int __try_to_reclaim_swap(struct swap_info_struct *si,
127                                  unsigned long offset, unsigned long flags)
128 {
129         swp_entry_t entry = swp_entry(si->type, offset);
130         struct page *page;
131         int ret = 0;
132 
133         page = find_get_page(swap_address_space(entry), offset);
134         if (!page)
135                 return 0;
136         /*
137          * When this function is called from scan_swap_map_slots() and it's
138          * called by vmscan.c at reclaiming pages. So, we hold a lock on a page,
139          * here. We have to use trylock for avoiding deadlock. This is a special
140          * case and you should use try_to_free_swap() with explicit lock_page()
141          * in usual operations.
142          */
143         if (trylock_page(page)) {
144                 if ((flags & TTRS_ANYWAY) ||
145                     ((flags & TTRS_UNMAPPED) && !page_mapped(page)) ||
146                     ((flags & TTRS_FULL) && mem_cgroup_swap_full(page)))
147                         ret = try_to_free_swap(page);
148                 unlock_page(page);
149         }
150         put_page(page);
151         return ret;
152 }
153 
154 /*
155  * swapon tell device that all the old swap contents can be discarded,
156  * to allow the swap device to optimize its wear-levelling.
157  */
158 static int discard_swap(struct swap_info_struct *si)
159 {
160         struct swap_extent *se;
161         sector_t start_block;
162         sector_t nr_blocks;
163         int err = 0;
164 
165         /* Do not discard the swap header page! */
166         se = &si->first_swap_extent;
167         start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
168         nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
169         if (nr_blocks) {
170                 err = blkdev_issue_discard(si->bdev, start_block,
171                                 nr_blocks, GFP_KERNEL, 0);
172                 if (err)
173                         return err;
174                 cond_resched();
175         }
176 
177         list_for_each_entry(se, &si->first_swap_extent.list, list) {
178                 start_block = se->start_block << (PAGE_SHIFT - 9);
179                 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
180 
181                 err = blkdev_issue_discard(si->bdev, start_block,
182                                 nr_blocks, GFP_KERNEL, 0);
183                 if (err)
184                         break;
185 
186                 cond_resched();
187         }
188         return err;             /* That will often be -EOPNOTSUPP */
189 }
190 
191 /*
192  * swap allocation tell device that a cluster of swap can now be discarded,
193  * to allow the swap device to optimize its wear-levelling.
194  */
195 static void discard_swap_cluster(struct swap_info_struct *si,
196                                  pgoff_t start_page, pgoff_t nr_pages)
197 {
198         struct swap_extent *se = si->curr_swap_extent;
199         int found_extent = 0;
200 
201         while (nr_pages) {
202                 if (se->start_page <= start_page &&
203                     start_page < se->start_page + se->nr_pages) {
204                         pgoff_t offset = start_page - se->start_page;
205                         sector_t start_block = se->start_block + offset;
206                         sector_t nr_blocks = se->nr_pages - offset;
207 
208                         if (nr_blocks > nr_pages)
209                                 nr_blocks = nr_pages;
210                         start_page += nr_blocks;
211                         nr_pages -= nr_blocks;
212 
213                         if (!found_extent++)
214                                 si->curr_swap_extent = se;
215 
216                         start_block <<= PAGE_SHIFT - 9;
217                         nr_blocks <<= PAGE_SHIFT - 9;
218                         if (blkdev_issue_discard(si->bdev, start_block,
219                                     nr_blocks, GFP_NOIO, 0))
220                                 break;
221                 }
222 
223                 se = list_next_entry(se, list);
224         }
225 }
226 
227 #ifdef CONFIG_THP_SWAP
228 #define SWAPFILE_CLUSTER        HPAGE_PMD_NR
229 
230 #define swap_entry_size(size)   (size)
231 #else
232 #define SWAPFILE_CLUSTER        256
233 
234 /*
235  * Define swap_entry_size() as constant to let compiler to optimize
236  * out some code if !CONFIG_THP_SWAP
237  */
238 #define swap_entry_size(size)   1
239 #endif
240 #define LATENCY_LIMIT           256
241 
242 static inline void cluster_set_flag(struct swap_cluster_info *info,
243         unsigned int flag)
244 {
245         info->flags = flag;
246 }
247 
248 static inline unsigned int cluster_count(struct swap_cluster_info *info)
249 {
250         return info->data;
251 }
252 
253 static inline void cluster_set_count(struct swap_cluster_info *info,
254                                      unsigned int c)
255 {
256         info->data = c;
257 }
258 
259 static inline void cluster_set_count_flag(struct swap_cluster_info *info,
260                                          unsigned int c, unsigned int f)
261 {
262         info->flags = f;
263         info->data = c;
264 }
265 
266 static inline unsigned int cluster_next(struct swap_cluster_info *info)
267 {
268         return info->data;
269 }
270 
271 static inline void cluster_set_next(struct swap_cluster_info *info,
272                                     unsigned int n)
273 {
274         info->data = n;
275 }
276 
277 static inline void cluster_set_next_flag(struct swap_cluster_info *info,
278                                          unsigned int n, unsigned int f)
279 {
280         info->flags = f;
281         info->data = n;
282 }
283 
284 static inline bool cluster_is_free(struct swap_cluster_info *info)
285 {
286         return info->flags & CLUSTER_FLAG_FREE;
287 }
288 
289 static inline bool cluster_is_null(struct swap_cluster_info *info)
290 {
291         return info->flags & CLUSTER_FLAG_NEXT_NULL;
292 }
293 
294 static inline void cluster_set_null(struct swap_cluster_info *info)
295 {
296         info->flags = CLUSTER_FLAG_NEXT_NULL;
297         info->data = 0;
298 }
299 
300 static inline bool cluster_is_huge(struct swap_cluster_info *info)
301 {
302         if (IS_ENABLED(CONFIG_THP_SWAP))
303                 return info->flags & CLUSTER_FLAG_HUGE;
304         return false;
305 }
306 
307 static inline void cluster_clear_huge(struct swap_cluster_info *info)
308 {
309         info->flags &= ~CLUSTER_FLAG_HUGE;
310 }
311 
312 static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si,
313                                                      unsigned long offset)
314 {
315         struct swap_cluster_info *ci;
316 
317         ci = si->cluster_info;
318         if (ci) {
319                 ci += offset / SWAPFILE_CLUSTER;
320                 spin_lock(&ci->lock);
321         }
322         return ci;
323 }
324 
325 static inline void unlock_cluster(struct swap_cluster_info *ci)
326 {
327         if (ci)
328                 spin_unlock(&ci->lock);
329 }
330 
331 /*
332  * Determine the locking method in use for this device.  Return
333  * swap_cluster_info if SSD-style cluster-based locking is in place.
334  */
335 static inline struct swap_cluster_info *lock_cluster_or_swap_info(
336                 struct swap_info_struct *si, unsigned long offset)
337 {
338         struct swap_cluster_info *ci;
339 
340         /* Try to use fine-grained SSD-style locking if available: */
341         ci = lock_cluster(si, offset);
342         /* Otherwise, fall back to traditional, coarse locking: */
343         if (!ci)
344                 spin_lock(&si->lock);
345 
346         return ci;
347 }
348 
349 static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si,
350                                                struct swap_cluster_info *ci)
351 {
352         if (ci)
353                 unlock_cluster(ci);
354         else
355                 spin_unlock(&si->lock);
356 }
357 
358 static inline bool cluster_list_empty(struct swap_cluster_list *list)
359 {
360         return cluster_is_null(&list->head);
361 }
362 
363 static inline unsigned int cluster_list_first(struct swap_cluster_list *list)
364 {
365         return cluster_next(&list->head);
366 }
367 
368 static void cluster_list_init(struct swap_cluster_list *list)
369 {
370         cluster_set_null(&list->head);
371         cluster_set_null(&list->tail);
372 }
373 
374 static void cluster_list_add_tail(struct swap_cluster_list *list,
375                                   struct swap_cluster_info *ci,
376                                   unsigned int idx)
377 {
378         if (cluster_list_empty(list)) {
379                 cluster_set_next_flag(&list->head, idx, 0);
380                 cluster_set_next_flag(&list->tail, idx, 0);
381         } else {
382                 struct swap_cluster_info *ci_tail;
383                 unsigned int tail = cluster_next(&list->tail);
384 
385                 /*
386                  * Nested cluster lock, but both cluster locks are
387                  * only acquired when we held swap_info_struct->lock
388                  */
389                 ci_tail = ci + tail;
390                 spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING);
391                 cluster_set_next(ci_tail, idx);
392                 spin_unlock(&ci_tail->lock);
393                 cluster_set_next_flag(&list->tail, idx, 0);
394         }
395 }
396 
397 static unsigned int cluster_list_del_first(struct swap_cluster_list *list,
398                                            struct swap_cluster_info *ci)
399 {
400         unsigned int idx;
401 
402         idx = cluster_next(&list->head);
403         if (cluster_next(&list->tail) == idx) {
404                 cluster_set_null(&list->head);
405                 cluster_set_null(&list->tail);
406         } else
407                 cluster_set_next_flag(&list->head,
408                                       cluster_next(&ci[idx]), 0);
409 
410         return idx;
411 }
412 
413 /* Add a cluster to discard list and schedule it to do discard */
414 static void swap_cluster_schedule_discard(struct swap_info_struct *si,
415                 unsigned int idx)
416 {
417         /*
418          * If scan_swap_map() can't find a free cluster, it will check
419          * si->swap_map directly. To make sure the discarding cluster isn't
420          * taken by scan_swap_map(), mark the swap entries bad (occupied). It
421          * will be cleared after discard
422          */
423         memset(si->swap_map + idx * SWAPFILE_CLUSTER,
424                         SWAP_MAP_BAD, SWAPFILE_CLUSTER);
425 
426         cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx);
427 
428         schedule_work(&si->discard_work);
429 }
430 
431 static void __free_cluster(struct swap_info_struct *si, unsigned long idx)
432 {
433         struct swap_cluster_info *ci = si->cluster_info;
434 
435         cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE);
436         cluster_list_add_tail(&si->free_clusters, ci, idx);
437 }
438 
439 /*
440  * Doing discard actually. After a cluster discard is finished, the cluster
441  * will be added to free cluster list. caller should hold si->lock.
442 */
443 static void swap_do_scheduled_discard(struct swap_info_struct *si)
444 {
445         struct swap_cluster_info *info, *ci;
446         unsigned int idx;
447 
448         info = si->cluster_info;
449 
450         while (!cluster_list_empty(&si->discard_clusters)) {
451                 idx = cluster_list_del_first(&si->discard_clusters, info);
452                 spin_unlock(&si->lock);
453 
454                 discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
455                                 SWAPFILE_CLUSTER);
456 
457                 spin_lock(&si->lock);
458                 ci = lock_cluster(si, idx * SWAPFILE_CLUSTER);
459                 __free_cluster(si, idx);
460                 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
461                                 0, SWAPFILE_CLUSTER);
462                 unlock_cluster(ci);
463         }
464 }
465 
466 static void swap_discard_work(struct work_struct *work)
467 {
468         struct swap_info_struct *si;
469 
470         si = container_of(work, struct swap_info_struct, discard_work);
471 
472         spin_lock(&si->lock);
473         swap_do_scheduled_discard(si);
474         spin_unlock(&si->lock);
475 }
476 
477 static void alloc_cluster(struct swap_info_struct *si, unsigned long idx)
478 {
479         struct swap_cluster_info *ci = si->cluster_info;
480 
481         VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx);
482         cluster_list_del_first(&si->free_clusters, ci);
483         cluster_set_count_flag(ci + idx, 0, 0);
484 }
485 
486 static void free_cluster(struct swap_info_struct *si, unsigned long idx)
487 {
488         struct swap_cluster_info *ci = si->cluster_info + idx;
489 
490         VM_BUG_ON(cluster_count(ci) != 0);
491         /*
492          * If the swap is discardable, prepare discard the cluster
493          * instead of free it immediately. The cluster will be freed
494          * after discard.
495          */
496         if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
497             (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
498                 swap_cluster_schedule_discard(si, idx);
499                 return;
500         }
501 
502         __free_cluster(si, idx);
503 }
504 
505 /*
506  * The cluster corresponding to page_nr will be used. The cluster will be
507  * removed from free cluster list and its usage counter will be increased.
508  */
509 static void inc_cluster_info_page(struct swap_info_struct *p,
510         struct swap_cluster_info *cluster_info, unsigned long page_nr)
511 {
512         unsigned long idx = page_nr / SWAPFILE_CLUSTER;
513 
514         if (!cluster_info)
515                 return;
516         if (cluster_is_free(&cluster_info[idx]))
517                 alloc_cluster(p, idx);
518 
519         VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
520         cluster_set_count(&cluster_info[idx],
521                 cluster_count(&cluster_info[idx]) + 1);
522 }
523 
524 /*
525  * The cluster corresponding to page_nr decreases one usage. If the usage
526  * counter becomes 0, which means no page in the cluster is in using, we can
527  * optionally discard the cluster and add it to free cluster list.
528  */
529 static void dec_cluster_info_page(struct swap_info_struct *p,
530         struct swap_cluster_info *cluster_info, unsigned long page_nr)
531 {
532         unsigned long idx = page_nr / SWAPFILE_CLUSTER;
533 
534         if (!cluster_info)
535                 return;
536 
537         VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
538         cluster_set_count(&cluster_info[idx],
539                 cluster_count(&cluster_info[idx]) - 1);
540 
541         if (cluster_count(&cluster_info[idx]) == 0)
542                 free_cluster(p, idx);
543 }
544 
545 /*
546  * It's possible scan_swap_map() uses a free cluster in the middle of free
547  * cluster list. Avoiding such abuse to avoid list corruption.
548  */
549 static bool
550 scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
551         unsigned long offset)
552 {
553         struct percpu_cluster *percpu_cluster;
554         bool conflict;
555 
556         offset /= SWAPFILE_CLUSTER;
557         conflict = !cluster_list_empty(&si->free_clusters) &&
558                 offset != cluster_list_first(&si->free_clusters) &&
559                 cluster_is_free(&si->cluster_info[offset]);
560 
561         if (!conflict)
562                 return false;
563 
564         percpu_cluster = this_cpu_ptr(si->percpu_cluster);
565         cluster_set_null(&percpu_cluster->index);
566         return true;
567 }
568 
569 /*
570  * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
571  * might involve allocating a new cluster for current CPU too.
572  */
573 static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
574         unsigned long *offset, unsigned long *scan_base)
575 {
576         struct percpu_cluster *cluster;
577         struct swap_cluster_info *ci;
578         bool found_free;
579         unsigned long tmp, max;
580 
581 new_cluster:
582         cluster = this_cpu_ptr(si->percpu_cluster);
583         if (cluster_is_null(&cluster->index)) {
584                 if (!cluster_list_empty(&si->free_clusters)) {
585                         cluster->index = si->free_clusters.head;
586                         cluster->next = cluster_next(&cluster->index) *
587                                         SWAPFILE_CLUSTER;
588                 } else if (!cluster_list_empty(&si->discard_clusters)) {
589                         /*
590                          * we don't have free cluster but have some clusters in
591                          * discarding, do discard now and reclaim them
592                          */
593                         swap_do_scheduled_discard(si);
594                         *scan_base = *offset = si->cluster_next;
595                         goto new_cluster;
596                 } else
597                         return false;
598         }
599 
600         found_free = false;
601 
602         /*
603          * Other CPUs can use our cluster if they can't find a free cluster,
604          * check if there is still free entry in the cluster
605          */
606         tmp = cluster->next;
607         max = min_t(unsigned long, si->max,
608                     (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER);
609         if (tmp >= max) {
610                 cluster_set_null(&cluster->index);
611                 goto new_cluster;
612         }
613         ci = lock_cluster(si, tmp);
614         while (tmp < max) {
615                 if (!si->swap_map[tmp]) {
616                         found_free = true;
617                         break;
618                 }
619                 tmp++;
620         }
621         unlock_cluster(ci);
622         if (!found_free) {
623                 cluster_set_null(&cluster->index);
624                 goto new_cluster;
625         }
626         cluster->next = tmp + 1;
627         *offset = tmp;
628         *scan_base = tmp;
629         return found_free;
630 }
631 
632 static void __del_from_avail_list(struct swap_info_struct *p)
633 {
634         int nid;
635 
636         for_each_node(nid)
637                 plist_del(&p->avail_lists[nid], &swap_avail_heads[nid]);
638 }
639 
640 static void del_from_avail_list(struct swap_info_struct *p)
641 {
642         spin_lock(&swap_avail_lock);
643         __del_from_avail_list(p);
644         spin_unlock(&swap_avail_lock);
645 }
646 
647 static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset,
648                              unsigned int nr_entries)
649 {
650         unsigned int end = offset + nr_entries - 1;
651 
652         if (offset == si->lowest_bit)
653                 si->lowest_bit += nr_entries;
654         if (end == si->highest_bit)
655                 si->highest_bit -= nr_entries;
656         si->inuse_pages += nr_entries;
657         if (si->inuse_pages == si->pages) {
658                 si->lowest_bit = si->max;
659                 si->highest_bit = 0;
660                 del_from_avail_list(si);
661         }
662 }
663 
664 static void add_to_avail_list(struct swap_info_struct *p)
665 {
666         int nid;
667 
668         spin_lock(&swap_avail_lock);
669         for_each_node(nid) {
670                 WARN_ON(!plist_node_empty(&p->avail_lists[nid]));
671                 plist_add(&p->avail_lists[nid], &swap_avail_heads[nid]);
672         }
673         spin_unlock(&swap_avail_lock);
674 }
675 
676 static void swap_range_free(struct swap_info_struct *si, unsigned long offset,
677                             unsigned int nr_entries)
678 {
679         unsigned long end = offset + nr_entries - 1;
680         void (*swap_slot_free_notify)(struct block_device *, unsigned long);
681 
682         if (offset < si->lowest_bit)
683                 si->lowest_bit = offset;
684         if (end > si->highest_bit) {
685                 bool was_full = !si->highest_bit;
686 
687                 si->highest_bit = end;
688                 if (was_full && (si->flags & SWP_WRITEOK))
689                         add_to_avail_list(si);
690         }
691         atomic_long_add(nr_entries, &nr_swap_pages);
692         si->inuse_pages -= nr_entries;
693         if (si->flags & SWP_BLKDEV)
694                 swap_slot_free_notify =
695                         si->bdev->bd_disk->fops->swap_slot_free_notify;
696         else
697                 swap_slot_free_notify = NULL;
698         while (offset <= end) {
699                 frontswap_invalidate_page(si->type, offset);
700                 if (swap_slot_free_notify)
701                         swap_slot_free_notify(si->bdev, offset);
702                 offset++;
703         }
704 }
705 
706 static int scan_swap_map_slots(struct swap_info_struct *si,
707                                unsigned char usage, int nr,
708                                swp_entry_t slots[])
709 {
710         struct swap_cluster_info *ci;
711         unsigned long offset;
712         unsigned long scan_base;
713         unsigned long last_in_cluster = 0;
714         int latency_ration = LATENCY_LIMIT;
715         int n_ret = 0;
716 
717         if (nr > SWAP_BATCH)
718                 nr = SWAP_BATCH;
719 
720         /*
721          * We try to cluster swap pages by allocating them sequentially
722          * in swap.  Once we've allocated SWAPFILE_CLUSTER pages this
723          * way, however, we resort to first-free allocation, starting
724          * a new cluster.  This prevents us from scattering swap pages
725          * all over the entire swap partition, so that we reduce
726          * overall disk seek times between swap pages.  -- sct
727          * But we do now try to find an empty cluster.  -Andrea
728          * And we let swap pages go all over an SSD partition.  Hugh
729          */
730 
731         si->flags += SWP_SCANNING;
732         scan_base = offset = si->cluster_next;
733 
734         /* SSD algorithm */
735         if (si->cluster_info) {
736                 if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
737                         goto checks;
738                 else
739                         goto scan;
740         }
741 
742         if (unlikely(!si->cluster_nr--)) {
743                 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
744                         si->cluster_nr = SWAPFILE_CLUSTER - 1;
745                         goto checks;
746                 }
747 
748                 spin_unlock(&si->lock);
749 
750                 /*
751                  * If seek is expensive, start searching for new cluster from
752                  * start of partition, to minimize the span of allocated swap.
753                  * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
754                  * case, just handled by scan_swap_map_try_ssd_cluster() above.
755                  */
756                 scan_base = offset = si->lowest_bit;
757                 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
758 
759                 /* Locate the first empty (unaligned) cluster */
760                 for (; last_in_cluster <= si->highest_bit; offset++) {
761                         if (si->swap_map[offset])
762                                 last_in_cluster = offset + SWAPFILE_CLUSTER;
763                         else if (offset == last_in_cluster) {
764                                 spin_lock(&si->lock);
765                                 offset -= SWAPFILE_CLUSTER - 1;
766                                 si->cluster_next = offset;
767                                 si->cluster_nr = SWAPFILE_CLUSTER - 1;
768                                 goto checks;
769                         }
770                         if (unlikely(--latency_ration < 0)) {
771                                 cond_resched();
772                                 latency_ration = LATENCY_LIMIT;
773                         }
774                 }
775 
776                 offset = scan_base;
777                 spin_lock(&si->lock);
778                 si->cluster_nr = SWAPFILE_CLUSTER - 1;
779         }
780 
781 checks:
782         if (si->cluster_info) {
783                 while (scan_swap_map_ssd_cluster_conflict(si, offset)) {
784                 /* take a break if we already got some slots */
785                         if (n_ret)
786                                 goto done;
787                         if (!scan_swap_map_try_ssd_cluster(si, &offset,
788                                                         &scan_base))
789                                 goto scan;
790                 }
791         }
792         if (!(si->flags & SWP_WRITEOK))
793                 goto no_page;
794         if (!si->highest_bit)
795                 goto no_page;
796         if (offset > si->highest_bit)
797                 scan_base = offset = si->lowest_bit;
798 
799         ci = lock_cluster(si, offset);
800         /* reuse swap entry of cache-only swap if not busy. */
801         if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
802                 int swap_was_freed;
803                 unlock_cluster(ci);
804                 spin_unlock(&si->lock);
805                 swap_was_freed = __try_to_reclaim_swap(si, offset, TTRS_ANYWAY);
806                 spin_lock(&si->lock);
807                 /* entry was freed successfully, try to use this again */
808                 if (swap_was_freed)
809                         goto checks;
810                 goto scan; /* check next one */
811         }
812 
813         if (si->swap_map[offset]) {
814                 unlock_cluster(ci);
815                 if (!n_ret)
816                         goto scan;
817                 else
818                         goto done;
819         }
820         si->swap_map[offset] = usage;
821         inc_cluster_info_page(si, si->cluster_info, offset);
822         unlock_cluster(ci);
823 
824         swap_range_alloc(si, offset, 1);
825         si->cluster_next = offset + 1;
826         slots[n_ret++] = swp_entry(si->type, offset);
827 
828         /* got enough slots or reach max slots? */
829         if ((n_ret == nr) || (offset >= si->highest_bit))
830                 goto done;
831 
832         /* search for next available slot */
833 
834         /* time to take a break? */
835         if (unlikely(--latency_ration < 0)) {
836                 if (n_ret)
837                         goto done;
838                 spin_unlock(&si->lock);
839                 cond_resched();
840                 spin_lock(&si->lock);
841                 latency_ration = LATENCY_LIMIT;
842         }
843 
844         /* try to get more slots in cluster */
845         if (si->cluster_info) {
846                 if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
847                         goto checks;
848                 else
849                         goto done;
850         }
851         /* non-ssd case */
852         ++offset;
853 
854         /* non-ssd case, still more slots in cluster? */
855         if (si->cluster_nr && !si->swap_map[offset]) {
856                 --si->cluster_nr;
857                 goto checks;
858         }
859 
860 done:
861         si->flags -= SWP_SCANNING;
862         return n_ret;
863 
864 scan:
865         spin_unlock(&si->lock);
866         while (++offset <= si->highest_bit) {
867                 if (!si->swap_map[offset]) {
868                         spin_lock(&si->lock);
869                         goto checks;
870                 }
871                 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
872                         spin_lock(&si->lock);
873                         goto checks;
874                 }
875                 if (unlikely(--latency_ration < 0)) {
876                         cond_resched();
877                         latency_ration = LATENCY_LIMIT;
878                 }
879         }
880         offset = si->lowest_bit;
881         while (offset < scan_base) {
882                 if (!si->swap_map[offset]) {
883                         spin_lock(&si->lock);
884                         goto checks;
885                 }
886                 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
887                         spin_lock(&si->lock);
888                         goto checks;
889                 }
890                 if (unlikely(--latency_ration < 0)) {
891                         cond_resched();
892                         latency_ration = LATENCY_LIMIT;
893                 }
894                 offset++;
895         }
896         spin_lock(&si->lock);
897 
898 no_page:
899         si->flags -= SWP_SCANNING;
900         return n_ret;
901 }
902 
903 static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot)
904 {
905         unsigned long idx;
906         struct swap_cluster_info *ci;
907         unsigned long offset, i;
908         unsigned char *map;
909 
910         /*
911          * Should not even be attempting cluster allocations when huge
912          * page swap is disabled.  Warn and fail the allocation.
913          */
914         if (!IS_ENABLED(CONFIG_THP_SWAP)) {
915                 VM_WARN_ON_ONCE(1);
916                 return 0;
917         }
918 
919         if (cluster_list_empty(&si->free_clusters))
920                 return 0;
921 
922         idx = cluster_list_first(&si->free_clusters);
923         offset = idx * SWAPFILE_CLUSTER;
924         ci = lock_cluster(si, offset);
925         alloc_cluster(si, idx);
926         cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE);
927 
928         map = si->swap_map + offset;
929         for (i = 0; i < SWAPFILE_CLUSTER; i++)
930                 map[i] = SWAP_HAS_CACHE;
931         unlock_cluster(ci);
932         swap_range_alloc(si, offset, SWAPFILE_CLUSTER);
933         *slot = swp_entry(si->type, offset);
934 
935         return 1;
936 }
937 
938 static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx)
939 {
940         unsigned long offset = idx * SWAPFILE_CLUSTER;
941         struct swap_cluster_info *ci;
942 
943         ci = lock_cluster(si, offset);
944         memset(si->swap_map + offset, 0, SWAPFILE_CLUSTER);
945         cluster_set_count_flag(ci, 0, 0);
946         free_cluster(si, idx);
947         unlock_cluster(ci);
948         swap_range_free(si, offset, SWAPFILE_CLUSTER);
949 }
950 
951 static unsigned long scan_swap_map(struct swap_info_struct *si,
952                                    unsigned char usage)
953 {
954         swp_entry_t entry;
955         int n_ret;
956 
957         n_ret = scan_swap_map_slots(si, usage, 1, &entry);
958 
959         if (n_ret)
960                 return swp_offset(entry);
961         else
962                 return 0;
963 
964 }
965 
966 int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size)
967 {
968         unsigned long size = swap_entry_size(entry_size);
969         struct swap_info_struct *si, *next;
970         long avail_pgs;
971         int n_ret = 0;
972         int node;
973 
974         /* Only single cluster request supported */
975         WARN_ON_ONCE(n_goal > 1 && size == SWAPFILE_CLUSTER);
976 
977         avail_pgs = atomic_long_read(&nr_swap_pages) / size;
978         if (avail_pgs <= 0)
979                 goto noswap;
980 
981         if (n_goal > SWAP_BATCH)
982                 n_goal = SWAP_BATCH;
983 
984         if (n_goal > avail_pgs)
985                 n_goal = avail_pgs;
986 
987         atomic_long_sub(n_goal * size, &nr_swap_pages);
988 
989         spin_lock(&swap_avail_lock);
990 
991 start_over:
992         node = numa_node_id();
993         plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) {
994                 /* requeue si to after same-priority siblings */
995                 plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]);
996                 spin_unlock(&swap_avail_lock);
997                 spin_lock(&si->lock);
998                 if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) {
999                         spin_lock(&swap_avail_lock);
1000                         if (plist_node_empty(&si->avail_lists[node])) {
1001                                 spin_unlock(&si->lock);
1002                                 goto nextsi;
1003                         }
1004                         WARN(!si->highest_bit,
1005                              "swap_info %d in list but !highest_bit\n",
1006                              si->type);
1007                         WARN(!(si->flags & SWP_WRITEOK),
1008                              "swap_info %d in list but !SWP_WRITEOK\n",
1009                              si->type);
1010                         __del_from_avail_list(si);
1011                         spin_unlock(&si->lock);
1012                         goto nextsi;
1013                 }
1014                 if (size == SWAPFILE_CLUSTER) {
1015                         if (!(si->flags & SWP_FS))
1016                                 n_ret = swap_alloc_cluster(si, swp_entries);
1017                 } else
1018                         n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE,
1019                                                     n_goal, swp_entries);
1020                 spin_unlock(&si->lock);
1021                 if (n_ret || size == SWAPFILE_CLUSTER)
1022                         goto check_out;
1023                 pr_debug("scan_swap_map of si %d failed to find offset\n",
1024                         si->type);
1025 
1026                 spin_lock(&swap_avail_lock);
1027 nextsi:
1028                 /*
1029                  * if we got here, it's likely that si was almost full before,
1030                  * and since scan_swap_map() can drop the si->lock, multiple
1031                  * callers probably all tried to get a page from the same si
1032                  * and it filled up before we could get one; or, the si filled
1033                  * up between us dropping swap_avail_lock and taking si->lock.
1034                  * Since we dropped the swap_avail_lock, the swap_avail_head
1035                  * list may have been modified; so if next is still in the
1036                  * swap_avail_head list then try it, otherwise start over
1037                  * if we have not gotten any slots.
1038                  */
1039                 if (plist_node_empty(&next->avail_lists[node]))
1040                         goto start_over;
1041         }
1042 
1043         spin_unlock(&swap_avail_lock);
1044 
1045 check_out:
1046         if (n_ret < n_goal)
1047                 atomic_long_add((long)(n_goal - n_ret) * size,
1048                                 &nr_swap_pages);
1049 noswap:
1050         return n_ret;
1051 }
1052 
1053 /* The only caller of this function is now suspend routine */
1054 swp_entry_t get_swap_page_of_type(int type)
1055 {
1056         struct swap_info_struct *si = swap_type_to_swap_info(type);
1057         pgoff_t offset;
1058 
1059         if (!si)
1060                 goto fail;
1061 
1062         spin_lock(&si->lock);
1063         if (si->flags & SWP_WRITEOK) {
1064                 atomic_long_dec(&nr_swap_pages);
1065                 /* This is called for allocating swap entry, not cache */
1066                 offset = scan_swap_map(si, 1);
1067                 if (offset) {
1068                         spin_unlock(&si->lock);
1069                         return swp_entry(type, offset);
1070                 }
1071                 atomic_long_inc(&nr_swap_pages);
1072         }
1073         spin_unlock(&si->lock);
1074 fail:
1075         return (swp_entry_t) {0};
1076 }
1077 
1078 static struct swap_info_struct *__swap_info_get(swp_entry_t entry)
1079 {
1080         struct swap_info_struct *p;
1081         unsigned long offset, type;
1082 
1083         if (!entry.val)
1084                 goto out;
1085         type = swp_type(entry);
1086         p = swap_type_to_swap_info(type);
1087         if (!p)
1088                 goto bad_nofile;
1089         if (!(p->flags & SWP_USED))
1090                 goto bad_device;
1091         offset = swp_offset(entry);
1092         if (offset >= p->max)
1093                 goto bad_offset;
1094         return p;
1095 
1096 bad_offset:
1097         pr_err("swap_info_get: %s%08lx\n", Bad_offset, entry.val);
1098         goto out;
1099 bad_device:
1100         pr_err("swap_info_get: %s%08lx\n", Unused_file, entry.val);
1101         goto out;
1102 bad_nofile:
1103         pr_err("swap_info_get: %s%08lx\n", Bad_file, entry.val);
1104 out:
1105         return NULL;
1106 }
1107 
1108 static struct swap_info_struct *_swap_info_get(swp_entry_t entry)
1109 {
1110         struct swap_info_struct *p;
1111 
1112         p = __swap_info_get(entry);
1113         if (!p)
1114                 goto out;
1115         if (!p->swap_map[swp_offset(entry)])
1116                 goto bad_free;
1117         return p;
1118 
1119 bad_free:
1120         pr_err("swap_info_get: %s%08lx\n", Unused_offset, entry.val);
1121         goto out;
1122 out:
1123         return NULL;
1124 }
1125 
1126 static struct swap_info_struct *swap_info_get(swp_entry_t entry)
1127 {
1128         struct swap_info_struct *p;
1129 
1130         p = _swap_info_get(entry);
1131         if (p)
1132                 spin_lock(&p->lock);
1133         return p;
1134 }
1135 
1136 static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry,
1137                                         struct swap_info_struct *q)
1138 {
1139         struct swap_info_struct *p;
1140 
1141         p = _swap_info_get(entry);
1142 
1143         if (p != q) {
1144                 if (q != NULL)
1145                         spin_unlock(&q->lock);
1146                 if (p != NULL)
1147                         spin_lock(&p->lock);
1148         }
1149         return p;
1150 }
1151 
1152 static unsigned char __swap_entry_free_locked(struct swap_info_struct *p,
1153                                               unsigned long offset,
1154                                               unsigned char usage)
1155 {
1156         unsigned char count;
1157         unsigned char has_cache;
1158 
1159         count = p->swap_map[offset];
1160 
1161         has_cache = count & SWAP_HAS_CACHE;
1162         count &= ~SWAP_HAS_CACHE;
1163 
1164         if (usage == SWAP_HAS_CACHE) {
1165                 VM_BUG_ON(!has_cache);
1166                 has_cache = 0;
1167         } else if (count == SWAP_MAP_SHMEM) {
1168                 /*
1169                  * Or we could insist on shmem.c using a special
1170                  * swap_shmem_free() and free_shmem_swap_and_cache()...
1171                  */
1172                 count = 0;
1173         } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
1174                 if (count == COUNT_CONTINUED) {
1175                         if (swap_count_continued(p, offset, count))
1176                                 count = SWAP_MAP_MAX | COUNT_CONTINUED;
1177                         else
1178                                 count = SWAP_MAP_MAX;
1179                 } else
1180                         count--;
1181         }
1182 
1183         usage = count | has_cache;
1184         p->swap_map[offset] = usage ? : SWAP_HAS_CACHE;
1185 
1186         return usage;
1187 }
1188 
1189 static unsigned char __swap_entry_free(struct swap_info_struct *p,
1190                                        swp_entry_t entry, unsigned char usage)
1191 {
1192         struct swap_cluster_info *ci;
1193         unsigned long offset = swp_offset(entry);
1194 
1195         ci = lock_cluster_or_swap_info(p, offset);
1196         usage = __swap_entry_free_locked(p, offset, usage);
1197         unlock_cluster_or_swap_info(p, ci);
1198         if (!usage)
1199                 free_swap_slot(entry);
1200 
1201         return usage;
1202 }
1203 
1204 static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry)
1205 {
1206         struct swap_cluster_info *ci;
1207         unsigned long offset = swp_offset(entry);
1208         unsigned char count;
1209 
1210         ci = lock_cluster(p, offset);
1211         count = p->swap_map[offset];
1212         VM_BUG_ON(count != SWAP_HAS_CACHE);
1213         p->swap_map[offset] = 0;
1214         dec_cluster_info_page(p, p->cluster_info, offset);
1215         unlock_cluster(ci);
1216 
1217         mem_cgroup_uncharge_swap(entry, 1);
1218         swap_range_free(p, offset, 1);
1219 }
1220 
1221 /*
1222  * Caller has made sure that the swap device corresponding to entry
1223  * is still around or has not been recycled.
1224  */
1225 void swap_free(swp_entry_t entry)
1226 {
1227         struct swap_info_struct *p;
1228 
1229         p = _swap_info_get(entry);
1230         if (p)
1231                 __swap_entry_free(p, entry, 1);
1232 }
1233 
1234 /*
1235  * Called after dropping swapcache to decrease refcnt to swap entries.
1236  */
1237 void put_swap_page(struct page *page, swp_entry_t entry)
1238 {
1239         unsigned long offset = swp_offset(entry);
1240         unsigned long idx = offset / SWAPFILE_CLUSTER;
1241         struct swap_cluster_info *ci;
1242         struct swap_info_struct *si;
1243         unsigned char *map;
1244         unsigned int i, free_entries = 0;
1245         unsigned char val;
1246         int size = swap_entry_size(hpage_nr_pages(page));
1247 
1248         si = _swap_info_get(entry);
1249         if (!si)
1250                 return;
1251 
1252         ci = lock_cluster_or_swap_info(si, offset);
1253         if (size == SWAPFILE_CLUSTER) {
1254                 VM_BUG_ON(!cluster_is_huge(ci));
1255                 map = si->swap_map + offset;
1256                 for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1257                         val = map[i];
1258                         VM_BUG_ON(!(val & SWAP_HAS_CACHE));
1259                         if (val == SWAP_HAS_CACHE)
1260                                 free_entries++;
1261                 }
1262                 cluster_clear_huge(ci);
1263                 if (free_entries == SWAPFILE_CLUSTER) {
1264                         unlock_cluster_or_swap_info(si, ci);
1265                         spin_lock(&si->lock);
1266                         mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER);
1267                         swap_free_cluster(si, idx);
1268                         spin_unlock(&si->lock);
1269                         return;
1270                 }
1271         }
1272         for (i = 0; i < size; i++, entry.val++) {
1273                 if (!__swap_entry_free_locked(si, offset + i, SWAP_HAS_CACHE)) {
1274                         unlock_cluster_or_swap_info(si, ci);
1275                         free_swap_slot(entry);
1276                         if (i == size - 1)
1277                                 return;
1278                         lock_cluster_or_swap_info(si, offset);
1279                 }
1280         }
1281         unlock_cluster_or_swap_info(si, ci);
1282 }
1283 
1284 #ifdef CONFIG_THP_SWAP
1285 int split_swap_cluster(swp_entry_t entry)
1286 {
1287         struct swap_info_struct *si;
1288         struct swap_cluster_info *ci;
1289         unsigned long offset = swp_offset(entry);
1290 
1291         si = _swap_info_get(entry);
1292         if (!si)
1293                 return -EBUSY;
1294         ci = lock_cluster(si, offset);
1295         cluster_clear_huge(ci);
1296         unlock_cluster(ci);
1297         return 0;
1298 }
1299 #endif
1300 
1301 static int swp_entry_cmp(const void *ent1, const void *ent2)
1302 {
1303         const swp_entry_t *e1 = ent1, *e2 = ent2;
1304 
1305         return (int)swp_type(*e1) - (int)swp_type(*e2);
1306 }
1307 
1308 void swapcache_free_entries(swp_entry_t *entries, int n)
1309 {
1310         struct swap_info_struct *p, *prev;
1311         int i;
1312 
1313         if (n <= 0)
1314                 return;
1315 
1316         prev = NULL;
1317         p = NULL;
1318 
1319         /*
1320          * Sort swap entries by swap device, so each lock is only taken once.
1321          * nr_swapfiles isn't absolutely correct, but the overhead of sort() is
1322          * so low that it isn't necessary to optimize further.
1323          */
1324         if (nr_swapfiles > 1)
1325                 sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL);
1326         for (i = 0; i < n; ++i) {
1327                 p = swap_info_get_cont(entries[i], prev);
1328                 if (p)
1329                         swap_entry_free(p, entries[i]);
1330                 prev = p;
1331         }
1332         if (p)
1333                 spin_unlock(&p->lock);
1334 }
1335 
1336 /*
1337  * How many references to page are currently swapped out?
1338  * This does not give an exact answer when swap count is continued,
1339  * but does include the high COUNT_CONTINUED flag to allow for that.
1340  */
1341 int page_swapcount(struct page *page)
1342 {
1343         int count = 0;
1344         struct swap_info_struct *p;
1345         struct swap_cluster_info *ci;
1346         swp_entry_t entry;
1347         unsigned long offset;
1348 
1349         entry.val = page_private(page);
1350         p = _swap_info_get(entry);
1351         if (p) {
1352                 offset = swp_offset(entry);
1353                 ci = lock_cluster_or_swap_info(p, offset);
1354                 count = swap_count(p->swap_map[offset]);
1355                 unlock_cluster_or_swap_info(p, ci);
1356         }
1357         return count;
1358 }
1359 
1360 int __swap_count(struct swap_info_struct *si, swp_entry_t entry)
1361 {
1362         pgoff_t offset = swp_offset(entry);
1363 
1364         return swap_count(si->swap_map[offset]);
1365 }
1366 
1367 static int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry)
1368 {
1369         int count = 0;
1370         pgoff_t offset = swp_offset(entry);
1371         struct swap_cluster_info *ci;
1372 
1373         ci = lock_cluster_or_swap_info(si, offset);
1374         count = swap_count(si->swap_map[offset]);
1375         unlock_cluster_or_swap_info(si, ci);
1376         return count;
1377 }
1378 
1379 /*
1380  * How many references to @entry are currently swapped out?
1381  * This does not give an exact answer when swap count is continued,
1382  * but does include the high COUNT_CONTINUED flag to allow for that.
1383  */
1384 int __swp_swapcount(swp_entry_t entry)
1385 {
1386         int count = 0;
1387         struct swap_info_struct *si;
1388 
1389         si = __swap_info_get(entry);
1390         if (si)
1391                 count = swap_swapcount(si, entry);
1392         return count;
1393 }
1394 
1395 /*
1396  * How many references to @entry are currently swapped out?
1397  * This considers COUNT_CONTINUED so it returns exact answer.
1398  */
1399 int swp_swapcount(swp_entry_t entry)
1400 {
1401         int count, tmp_count, n;
1402         struct swap_info_struct *p;
1403         struct swap_cluster_info *ci;
1404         struct page *page;
1405         pgoff_t offset;
1406         unsigned char *map;
1407 
1408         p = _swap_info_get(entry);
1409         if (!p)
1410                 return 0;
1411 
1412         offset = swp_offset(entry);
1413 
1414         ci = lock_cluster_or_swap_info(p, offset);
1415 
1416         count = swap_count(p->swap_map[offset]);
1417         if (!(count & COUNT_CONTINUED))
1418                 goto out;
1419 
1420         count &= ~COUNT_CONTINUED;
1421         n = SWAP_MAP_MAX + 1;
1422 
1423         page = vmalloc_to_page(p->swap_map + offset);
1424         offset &= ~PAGE_MASK;
1425         VM_BUG_ON(page_private(page) != SWP_CONTINUED);
1426 
1427         do {
1428                 page = list_next_entry(page, lru);
1429                 map = kmap_atomic(page);
1430                 tmp_count = map[offset];
1431                 kunmap_atomic(map);
1432 
1433                 count += (tmp_count & ~COUNT_CONTINUED) * n;
1434                 n *= (SWAP_CONT_MAX + 1);
1435         } while (tmp_count & COUNT_CONTINUED);
1436 out:
1437         unlock_cluster_or_swap_info(p, ci);
1438         return count;
1439 }
1440 
1441 static bool swap_page_trans_huge_swapped(struct swap_info_struct *si,
1442                                          swp_entry_t entry)
1443 {
1444         struct swap_cluster_info *ci;
1445         unsigned char *map = si->swap_map;
1446         unsigned long roffset = swp_offset(entry);
1447         unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER);
1448         int i;
1449         bool ret = false;
1450 
1451         ci = lock_cluster_or_swap_info(si, offset);
1452         if (!ci || !cluster_is_huge(ci)) {
1453                 if (swap_count(map[roffset]))
1454                         ret = true;
1455                 goto unlock_out;
1456         }
1457         for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1458                 if (swap_count(map[offset + i])) {
1459                         ret = true;
1460                         break;
1461                 }
1462         }
1463 unlock_out:
1464         unlock_cluster_or_swap_info(si, ci);
1465         return ret;
1466 }
1467 
1468 static bool page_swapped(struct page *page)
1469 {
1470         swp_entry_t entry;
1471         struct swap_info_struct *si;
1472 
1473         if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page)))
1474                 return page_swapcount(page) != 0;
1475 
1476         page = compound_head(page);
1477         entry.val = page_private(page);
1478         si = _swap_info_get(entry);
1479         if (si)
1480                 return swap_page_trans_huge_swapped(si, entry);
1481         return false;
1482 }
1483 
1484 static int page_trans_huge_map_swapcount(struct page *page, int *total_mapcount,
1485                                          int *total_swapcount)
1486 {
1487         int i, map_swapcount, _total_mapcount, _total_swapcount;
1488         unsigned long offset = 0;
1489         struct swap_info_struct *si;
1490         struct swap_cluster_info *ci = NULL;
1491         unsigned char *map = NULL;
1492         int mapcount, swapcount = 0;
1493 
1494         /* hugetlbfs shouldn't call it */
1495         VM_BUG_ON_PAGE(PageHuge(page), page);
1496 
1497         if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page))) {
1498                 mapcount = page_trans_huge_mapcount(page, total_mapcount);
1499                 if (PageSwapCache(page))
1500                         swapcount = page_swapcount(page);
1501                 if (total_swapcount)
1502                         *total_swapcount = swapcount;
1503                 return mapcount + swapcount;
1504         }
1505 
1506         page = compound_head(page);
1507 
1508         _total_mapcount = _total_swapcount = map_swapcount = 0;
1509         if (PageSwapCache(page)) {
1510                 swp_entry_t entry;
1511 
1512                 entry.val = page_private(page);
1513                 si = _swap_info_get(entry);
1514                 if (si) {
1515                         map = si->swap_map;
1516                         offset = swp_offset(entry);
1517                 }
1518         }
1519         if (map)
1520                 ci = lock_cluster(si, offset);
1521         for (i = 0; i < HPAGE_PMD_NR; i++) {
1522                 mapcount = atomic_read(&page[i]._mapcount) + 1;
1523                 _total_mapcount += mapcount;
1524                 if (map) {
1525                         swapcount = swap_count(map[offset + i]);
1526                         _total_swapcount += swapcount;
1527                 }
1528                 map_swapcount = max(map_swapcount, mapcount + swapcount);
1529         }
1530         unlock_cluster(ci);
1531         if (PageDoubleMap(page)) {
1532                 map_swapcount -= 1;
1533                 _total_mapcount -= HPAGE_PMD_NR;
1534         }
1535         mapcount = compound_mapcount(page);
1536         map_swapcount += mapcount;
1537         _total_mapcount += mapcount;
1538         if (total_mapcount)
1539                 *total_mapcount = _total_mapcount;
1540         if (total_swapcount)
1541                 *total_swapcount = _total_swapcount;
1542 
1543         return map_swapcount;
1544 }
1545 
1546 /*
1547  * We can write to an anon page without COW if there are no other references
1548  * to it.  And as a side-effect, free up its swap: because the old content
1549  * on disk will never be read, and seeking back there to write new content
1550  * later would only waste time away from clustering.
1551  *
1552  * NOTE: total_map_swapcount should not be relied upon by the caller if
1553  * reuse_swap_page() returns false, but it may be always overwritten
1554  * (see the other implementation for CONFIG_SWAP=n).
1555  */
1556 bool reuse_swap_page(struct page *page, int *total_map_swapcount)
1557 {
1558         int count, total_mapcount, total_swapcount;
1559 
1560         VM_BUG_ON_PAGE(!PageLocked(page), page);
1561         if (unlikely(PageKsm(page)))
1562                 return false;
1563         count = page_trans_huge_map_swapcount(page, &total_mapcount,
1564                                               &total_swapcount);
1565         if (total_map_swapcount)
1566                 *total_map_swapcount = total_mapcount + total_swapcount;
1567         if (count == 1 && PageSwapCache(page) &&
1568             (likely(!PageTransCompound(page)) ||
1569              /* The remaining swap count will be freed soon */
1570              total_swapcount == page_swapcount(page))) {
1571                 if (!PageWriteback(page)) {
1572                         page = compound_head(page);
1573                         delete_from_swap_cache(page);
1574                         SetPageDirty(page);
1575                 } else {
1576                         swp_entry_t entry;
1577                         struct swap_info_struct *p;
1578 
1579                         entry.val = page_private(page);
1580                         p = swap_info_get(entry);
1581                         if (p->flags & SWP_STABLE_WRITES) {
1582                                 spin_unlock(&p->lock);
1583                                 return false;
1584                         }
1585                         spin_unlock(&p->lock);
1586                 }
1587         }
1588 
1589         return count <= 1;
1590 }
1591 
1592 /*
1593  * If swap is getting full, or if there are no more mappings of this page,
1594  * then try_to_free_swap is called to free its swap space.
1595  */
1596 int try_to_free_swap(struct page *page)
1597 {
1598         VM_BUG_ON_PAGE(!PageLocked(page), page);
1599 
1600         if (!PageSwapCache(page))
1601                 return 0;
1602         if (PageWriteback(page))
1603                 return 0;
1604         if (page_swapped(page))
1605                 return 0;
1606 
1607         /*
1608          * Once hibernation has begun to create its image of memory,
1609          * there's a danger that one of the calls to try_to_free_swap()
1610          * - most probably a call from __try_to_reclaim_swap() while
1611          * hibernation is allocating its own swap pages for the image,
1612          * but conceivably even a call from memory reclaim - will free
1613          * the swap from a page which has already been recorded in the
1614          * image as a clean swapcache page, and then reuse its swap for
1615          * another page of the image.  On waking from hibernation, the
1616          * original page might be freed under memory pressure, then
1617          * later read back in from swap, now with the wrong data.
1618          *
1619          * Hibernation suspends storage while it is writing the image
1620          * to disk so check that here.
1621          */
1622         if (pm_suspended_storage())
1623                 return 0;
1624 
1625         page = compound_head(page);
1626         delete_from_swap_cache(page);
1627         SetPageDirty(page);
1628         return 1;
1629 }
1630 
1631 /*
1632  * Free the swap entry like above, but also try to
1633  * free the page cache entry if it is the last user.
1634  */
1635 int free_swap_and_cache(swp_entry_t entry)
1636 {
1637         struct swap_info_struct *p;
1638         unsigned char count;
1639 
1640         if (non_swap_entry(entry))
1641                 return 1;
1642 
1643         p = _swap_info_get(entry);
1644         if (p) {
1645                 count = __swap_entry_free(p, entry, 1);
1646                 if (count == SWAP_HAS_CACHE &&
1647                     !swap_page_trans_huge_swapped(p, entry))
1648                         __try_to_reclaim_swap(p, swp_offset(entry),
1649                                               TTRS_UNMAPPED | TTRS_FULL);
1650         }
1651         return p != NULL;
1652 }
1653 
1654 #ifdef CONFIG_HIBERNATION
1655 /*
1656  * Find the swap type that corresponds to given device (if any).
1657  *
1658  * @offset - number of the PAGE_SIZE-sized block of the device, starting
1659  * from 0, in which the swap header is expected to be located.
1660  *
1661  * This is needed for the suspend to disk (aka swsusp).
1662  */
1663 int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
1664 {
1665         struct block_device *bdev = NULL;
1666         int type;
1667 
1668         if (device)
1669                 bdev = bdget(device);
1670 
1671         spin_lock(&swap_lock);
1672         for (type = 0; type < nr_swapfiles; type++) {
1673                 struct swap_info_struct *sis = swap_info[type];
1674 
1675                 if (!(sis->flags & SWP_WRITEOK))
1676                         continue;
1677 
1678                 if (!bdev) {
1679                         if (bdev_p)
1680                                 *bdev_p = bdgrab(sis->bdev);
1681 
1682                         spin_unlock(&swap_lock);
1683                         return type;
1684                 }
1685                 if (bdev == sis->bdev) {
1686                         struct swap_extent *se = &sis->first_swap_extent;
1687 
1688                         if (se->start_block == offset) {
1689                                 if (bdev_p)
1690                                         *bdev_p = bdgrab(sis->bdev);
1691 
1692                                 spin_unlock(&swap_lock);
1693                                 bdput(bdev);
1694                                 return type;
1695                         }
1696                 }
1697         }
1698         spin_unlock(&swap_lock);
1699         if (bdev)
1700                 bdput(bdev);
1701 
1702         return -ENODEV;
1703 }
1704 
1705 /*
1706  * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1707  * corresponding to given index in swap_info (swap type).
1708  */
1709 sector_t swapdev_block(int type, pgoff_t offset)
1710 {
1711         struct block_device *bdev;
1712         struct swap_info_struct *si = swap_type_to_swap_info(type);
1713 
1714         if (!si || !(si->flags & SWP_WRITEOK))
1715                 return 0;
1716         return map_swap_entry(swp_entry(type, offset), &bdev);
1717 }
1718 
1719 /*
1720  * Return either the total number of swap pages of given type, or the number
1721  * of free pages of that type (depending on @free)
1722  *
1723  * This is needed for software suspend
1724  */
1725 unsigned int count_swap_pages(int type, int free)
1726 {
1727         unsigned int n = 0;
1728 
1729         spin_lock(&swap_lock);
1730         if ((unsigned int)type < nr_swapfiles) {
1731                 struct swap_info_struct *sis = swap_info[type];
1732 
1733                 spin_lock(&sis->lock);
1734                 if (sis->flags & SWP_WRITEOK) {
1735                         n = sis->pages;
1736                         if (free)
1737                                 n -= sis->inuse_pages;
1738                 }
1739                 spin_unlock(&sis->lock);
1740         }
1741         spin_unlock(&swap_lock);
1742         return n;
1743 }
1744 #endif /* CONFIG_HIBERNATION */
1745 
1746 static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte)
1747 {
1748         return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte);
1749 }
1750 
1751 /*
1752  * No need to decide whether this PTE shares the swap entry with others,
1753  * just let do_wp_page work it out if a write is requested later - to
1754  * force COW, vm_page_prot omits write permission from any private vma.
1755  */
1756 static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
1757                 unsigned long addr, swp_entry_t entry, struct page *page)
1758 {
1759         struct page *swapcache;
1760         struct mem_cgroup *memcg;
1761         spinlock_t *ptl;
1762         pte_t *pte;
1763         int ret = 1;
1764 
1765         swapcache = page;
1766         page = ksm_might_need_to_copy(page, vma, addr);
1767         if (unlikely(!page))
1768                 return -ENOMEM;
1769 
1770         if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL,
1771                                 &memcg, false)) {
1772                 ret = -ENOMEM;
1773                 goto out_nolock;
1774         }
1775 
1776         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1777         if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) {
1778                 mem_cgroup_cancel_charge(page, memcg, false);
1779                 ret = 0;
1780                 goto out;
1781         }
1782 
1783         dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1784         inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
1785         get_page(page);
1786         set_pte_at(vma->vm_mm, addr, pte,
1787                    pte_mkold(mk_pte(page, vma->vm_page_prot)));
1788         if (page == swapcache) {
1789                 page_add_anon_rmap(page, vma, addr, false);
1790                 mem_cgroup_commit_charge(page, memcg, true, false);
1791         } else { /* ksm created a completely new copy */
1792                 page_add_new_anon_rmap(page, vma, addr, false);
1793                 mem_cgroup_commit_charge(page, memcg, false, false);
1794                 lru_cache_add_active_or_unevictable(page, vma);
1795         }
1796         swap_free(entry);
1797         /*
1798          * Move the page to the active list so it is not
1799          * immediately swapped out again after swapon.
1800          */
1801         activate_page(page);
1802 out:
1803         pte_unmap_unlock(pte, ptl);
1804 out_nolock:
1805         if (page != swapcache) {
1806                 unlock_page(page);
1807                 put_page(page);
1808         }
1809         return ret;
1810 }
1811 
1812 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1813                         unsigned long addr, unsigned long end,
1814                         unsigned int type, bool frontswap,
1815                         unsigned long *fs_pages_to_unuse)
1816 {
1817         struct page *page;
1818         swp_entry_t entry;
1819         pte_t *pte;
1820         struct swap_info_struct *si;
1821         unsigned long offset;
1822         int ret = 0;
1823         volatile unsigned char *swap_map;
1824 
1825         si = swap_info[type];
1826         pte = pte_offset_map(pmd, addr);
1827         do {
1828                 struct vm_fault vmf;
1829 
1830                 if (!is_swap_pte(*pte))
1831                         continue;
1832 
1833                 entry = pte_to_swp_entry(*pte);
1834                 if (swp_type(entry) != type)
1835                         continue;
1836 
1837                 offset = swp_offset(entry);
1838                 if (frontswap && !frontswap_test(si, offset))
1839                         continue;
1840 
1841                 pte_unmap(pte);
1842                 swap_map = &si->swap_map[offset];
1843                 vmf.vma = vma;
1844                 vmf.address = addr;
1845                 vmf.pmd = pmd;
1846                 page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, &vmf);
1847                 if (!page) {
1848                         if (*swap_map == 0 || *swap_map == SWAP_MAP_BAD)
1849                                 goto try_next;
1850                         return -ENOMEM;
1851                 }
1852 
1853                 lock_page(page);
1854                 wait_on_page_writeback(page);
1855                 ret = unuse_pte(vma, pmd, addr, entry, page);
1856                 if (ret < 0) {
1857                         unlock_page(page);
1858                         put_page(page);
1859                         goto out;
1860                 }
1861 
1862                 try_to_free_swap(page);
1863                 unlock_page(page);
1864                 put_page(page);
1865 
1866                 if (*fs_pages_to_unuse && !--(*fs_pages_to_unuse)) {
1867                         ret = FRONTSWAP_PAGES_UNUSED;
1868                         goto out;
1869                 }
1870 try_next:
1871                 pte = pte_offset_map(pmd, addr);
1872         } while (pte++, addr += PAGE_SIZE, addr != end);
1873         pte_unmap(pte - 1);
1874 
1875         ret = 0;
1876 out:
1877         return ret;
1878 }
1879 
1880 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1881                                 unsigned long addr, unsigned long end,
1882                                 unsigned int type, bool frontswap,
1883                                 unsigned long *fs_pages_to_unuse)
1884 {
1885         pmd_t *pmd;
1886         unsigned long next;
1887         int ret;
1888 
1889         pmd = pmd_offset(pud, addr);
1890         do {
1891                 cond_resched();
1892                 next = pmd_addr_end(addr, end);
1893                 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
1894                         continue;
1895                 ret = unuse_pte_range(vma, pmd, addr, next, type,
1896                                       frontswap, fs_pages_to_unuse);
1897                 if (ret)
1898                         return ret;
1899         } while (pmd++, addr = next, addr != end);
1900         return 0;
1901 }
1902 
1903 static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d,
1904                                 unsigned long addr, unsigned long end,
1905                                 unsigned int type, bool frontswap,
1906                                 unsigned long *fs_pages_to_unuse)
1907 {
1908         pud_t *pud;
1909         unsigned long next;
1910         int ret;
1911 
1912         pud = pud_offset(p4d, addr);
1913         do {
1914                 next = pud_addr_end(addr, end);
1915                 if (pud_none_or_clear_bad(pud))
1916                         continue;
1917                 ret = unuse_pmd_range(vma, pud, addr, next, type,
1918                                       frontswap, fs_pages_to_unuse);
1919                 if (ret)
1920                         return ret;
1921         } while (pud++, addr = next, addr != end);
1922         return 0;
1923 }
1924 
1925 static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd,
1926                                 unsigned long addr, unsigned long end,
1927                                 unsigned int type, bool frontswap,
1928                                 unsigned long *fs_pages_to_unuse)
1929 {
1930         p4d_t *p4d;
1931         unsigned long next;
1932         int ret;
1933 
1934         p4d = p4d_offset(pgd, addr);
1935         do {
1936                 next = p4d_addr_end(addr, end);
1937                 if (p4d_none_or_clear_bad(p4d))
1938                         continue;
1939                 ret = unuse_pud_range(vma, p4d, addr, next, type,
1940                                       frontswap, fs_pages_to_unuse);
1941                 if (ret)
1942                         return ret;
1943         } while (p4d++, addr = next, addr != end);
1944         return 0;
1945 }
1946 
1947 static int unuse_vma(struct vm_area_struct *vma, unsigned int type,
1948                      bool frontswap, unsigned long *fs_pages_to_unuse)
1949 {
1950         pgd_t *pgd;
1951         unsigned long addr, end, next;
1952         int ret;
1953 
1954         addr = vma->vm_start;
1955         end = vma->vm_end;
1956 
1957         pgd = pgd_offset(vma->vm_mm, addr);
1958         do {
1959                 next = pgd_addr_end(addr, end);
1960                 if (pgd_none_or_clear_bad(pgd))
1961                         continue;
1962                 ret = unuse_p4d_range(vma, pgd, addr, next, type,
1963                                       frontswap, fs_pages_to_unuse);
1964                 if (ret)
1965                         return ret;
1966         } while (pgd++, addr = next, addr != end);
1967         return 0;
1968 }
1969 
1970 static int unuse_mm(struct mm_struct *mm, unsigned int type,
1971                     bool frontswap, unsigned long *fs_pages_to_unuse)
1972 {
1973         struct vm_area_struct *vma;
1974         int ret = 0;
1975 
1976         down_read(&mm->mmap_sem);
1977         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1978                 if (vma->anon_vma) {
1979                         ret = unuse_vma(vma, type, frontswap,
1980                                         fs_pages_to_unuse);
1981                         if (ret)
1982                                 break;
1983                 }
1984                 cond_resched();
1985         }
1986         up_read(&mm->mmap_sem);
1987         return ret;
1988 }
1989 
1990 /*
1991  * Scan swap_map (or frontswap_map if frontswap parameter is true)
1992  * from current position to next entry still in use. Return 0
1993  * if there are no inuse entries after prev till end of the map.
1994  */
1995 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
1996                                         unsigned int prev, bool frontswap)
1997 {
1998         unsigned int i;
1999         unsigned char count;
2000 
2001         /*
2002          * No need for swap_lock here: we're just looking
2003          * for whether an entry is in use, not modifying it; false
2004          * hits are okay, and sys_swapoff() has already prevented new
2005          * allocations from this area (while holding swap_lock).
2006          */
2007         for (i = prev + 1; i < si->max; i++) {
2008                 count = READ_ONCE(si->swap_map[i]);
2009                 if (count && swap_count(count) != SWAP_MAP_BAD)
2010                         if (!frontswap || frontswap_test(si, i))
2011                                 break;
2012                 if ((i % LATENCY_LIMIT) == 0)
2013                         cond_resched();
2014         }
2015 
2016         if (i == si->max)
2017                 i = 0;
2018 
2019         return i;
2020 }
2021 
2022 /*
2023  * If the boolean frontswap is true, only unuse pages_to_unuse pages;
2024  * pages_to_unuse==0 means all pages; ignored if frontswap is false
2025  */
2026 #define SWAP_UNUSE_MAX_TRIES 3
2027 int try_to_unuse(unsigned int type, bool frontswap,
2028                  unsigned long pages_to_unuse)
2029 {
2030         struct mm_struct *prev_mm;
2031         struct mm_struct *mm;
2032         struct list_head *p;
2033         int retval = 0;
2034         struct swap_info_struct *si = swap_info[type];
2035         struct page *page;
2036         swp_entry_t entry;
2037         unsigned int i;
2038         int retries = 0;
2039 
2040         if (!si->inuse_pages)
2041                 return 0;
2042 
2043         if (!frontswap)
2044                 pages_to_unuse = 0;
2045 
2046 retry:
2047         retval = shmem_unuse(type, frontswap, &pages_to_unuse);
2048         if (retval)
2049                 goto out;
2050 
2051         prev_mm = &init_mm;
2052         mmget(prev_mm);
2053 
2054         spin_lock(&mmlist_lock);
2055         p = &init_mm.mmlist;
2056         while ((p = p->next) != &init_mm.mmlist) {
2057                 if (signal_pending(current)) {
2058                         retval = -EINTR;
2059                         break;
2060                 }
2061 
2062                 mm = list_entry(p, struct mm_struct, mmlist);
2063                 if (!mmget_not_zero(mm))
2064                         continue;
2065                 spin_unlock(&mmlist_lock);
2066                 mmput(prev_mm);
2067                 prev_mm = mm;
2068                 retval = unuse_mm(mm, type, frontswap, &pages_to_unuse);
2069 
2070                 if (retval) {
2071                         mmput(prev_mm);
2072                         goto out;
2073                 }
2074 
2075                 /*
2076                  * Make sure that we aren't completely killing
2077                  * interactive performance.
2078                  */
2079                 cond_resched();
2080                 spin_lock(&mmlist_lock);
2081         }
2082         spin_unlock(&mmlist_lock);
2083 
2084         mmput(prev_mm);
2085 
2086         i = 0;
2087         while ((i = find_next_to_unuse(si, i, frontswap)) != 0) {
2088 
2089                 entry = swp_entry(type, i);
2090                 page = find_get_page(swap_address_space(entry), i);
2091                 if (!page)
2092                         continue;
2093 
2094                 /*
2095                  * It is conceivable that a racing task removed this page from
2096                  * swap cache just before we acquired the page lock. The page
2097                  * might even be back in swap cache on another swap area. But
2098                  * that is okay, try_to_free_swap() only removes stale pages.
2099                  */
2100                 lock_page(page);
2101                 wait_on_page_writeback(page);
2102                 try_to_free_swap(page);
2103                 unlock_page(page);
2104                 put_page(page);
2105 
2106                 /*
2107                  * For frontswap, we just need to unuse pages_to_unuse, if
2108                  * it was specified. Need not check frontswap again here as
2109                  * we already zeroed out pages_to_unuse if not frontswap.
2110                  */
2111                 if (pages_to_unuse && --pages_to_unuse == 0)
2112                         goto out;
2113         }
2114 
2115         /*
2116          * Lets check again to see if there are still swap entries in the map.
2117          * If yes, we would need to do retry the unuse logic again.
2118          * Under global memory pressure, swap entries can be reinserted back
2119          * into process space after the mmlist loop above passes over them.
2120          * Its not worth continuosuly retrying to unuse the swap in this case.
2121          * So we try SWAP_UNUSE_MAX_TRIES times.
2122          */
2123         if (++retries >= SWAP_UNUSE_MAX_TRIES)
2124                 retval = -EBUSY;
2125         else if (si->inuse_pages)
2126                 goto retry;
2127 
2128 out:
2129         return (retval == FRONTSWAP_PAGES_UNUSED) ? 0 : retval;
2130 }
2131 
2132 /*
2133  * After a successful try_to_unuse, if no swap is now in use, we know
2134  * we can empty the mmlist.  swap_lock must be held on entry and exit.
2135  * Note that mmlist_lock nests inside swap_lock, and an mm must be
2136  * added to the mmlist just after page_duplicate - before would be racy.
2137  */
2138 static void drain_mmlist(void)
2139 {
2140         struct list_head *p, *next;
2141         unsigned int type;
2142 
2143         for (type = 0; type < nr_swapfiles; type++)
2144                 if (swap_info[type]->inuse_pages)
2145                         return;
2146         spin_lock(&mmlist_lock);
2147         list_for_each_safe(p, next, &init_mm.mmlist)
2148                 list_del_init(p);
2149         spin_unlock(&mmlist_lock);
2150 }
2151 
2152 /*
2153  * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
2154  * corresponds to page offset for the specified swap entry.
2155  * Note that the type of this function is sector_t, but it returns page offset
2156  * into the bdev, not sector offset.
2157  */
2158 static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
2159 {
2160         struct swap_info_struct *sis;
2161         struct swap_extent *start_se;
2162         struct swap_extent *se;
2163         pgoff_t offset;
2164 
2165         sis = swp_swap_info(entry);
2166         *bdev = sis->bdev;
2167 
2168         offset = swp_offset(entry);
2169         start_se = sis->curr_swap_extent;
2170         se = start_se;
2171 
2172         for ( ; ; ) {
2173                 if (se->start_page <= offset &&
2174                                 offset < (se->start_page + se->nr_pages)) {
2175                         return se->start_block + (offset - se->start_page);
2176                 }
2177                 se = list_next_entry(se, list);
2178                 sis->curr_swap_extent = se;
2179                 BUG_ON(se == start_se);         /* It *must* be present */
2180         }
2181 }
2182 
2183 /*
2184  * Returns the page offset into bdev for the specified page's swap entry.
2185  */
2186 sector_t map_swap_page(struct page *page, struct block_device **bdev)
2187 {
2188         swp_entry_t entry;
2189         entry.val = page_private(page);
2190         return map_swap_entry(entry, bdev);
2191 }
2192 
2193 /*
2194  * Free all of a swapdev's extent information
2195  */
2196 static void destroy_swap_extents(struct swap_info_struct *sis)
2197 {
2198         while (!list_empty(&sis->first_swap_extent.list)) {
2199                 struct swap_extent *se;
2200 
2201                 se = list_first_entry(&sis->first_swap_extent.list,
2202                                 struct swap_extent, list);
2203                 list_del(&se->list);
2204                 kfree(se);
2205         }
2206 
2207         if (sis->flags & SWP_ACTIVATED) {
2208                 struct file *swap_file = sis->swap_file;
2209                 struct address_space *mapping = swap_file->f_mapping;
2210 
2211                 sis->flags &= ~SWP_ACTIVATED;
2212                 if (mapping->a_ops->swap_deactivate)
2213                         mapping->a_ops->swap_deactivate(swap_file);
2214         }
2215 }
2216 
2217 /*
2218  * Add a block range (and the corresponding page range) into this swapdev's
2219  * extent list.  The extent list is kept sorted in page order.
2220  *
2221  * This function rather assumes that it is called in ascending page order.
2222  */
2223 int
2224 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
2225                 unsigned long nr_pages, sector_t start_block)
2226 {
2227         struct swap_extent *se;
2228         struct swap_extent *new_se;
2229         struct list_head *lh;
2230 
2231         if (start_page == 0) {
2232                 se = &sis->first_swap_extent;
2233                 sis->curr_swap_extent = se;
2234                 se->start_page = 0;
2235                 se->nr_pages = nr_pages;
2236                 se->start_block = start_block;
2237                 return 1;
2238         } else {
2239                 lh = sis->first_swap_extent.list.prev;  /* Highest extent */
2240                 se = list_entry(lh, struct swap_extent, list);
2241                 BUG_ON(se->start_page + se->nr_pages != start_page);
2242                 if (se->start_block + se->nr_pages == start_block) {
2243                         /* Merge it */
2244                         se->nr_pages += nr_pages;
2245                         return 0;
2246                 }
2247         }
2248 
2249         /*
2250          * No merge.  Insert a new extent, preserving ordering.
2251          */
2252         new_se = kmalloc(sizeof(*se), GFP_KERNEL);
2253         if (new_se == NULL)
2254                 return -ENOMEM;
2255         new_se->start_page = start_page;
2256         new_se->nr_pages = nr_pages;
2257         new_se->start_block = start_block;
2258 
2259         list_add_tail(&new_se->list, &sis->first_swap_extent.list);
2260         return 1;
2261 }
2262 EXPORT_SYMBOL_GPL(add_swap_extent);
2263 
2264 /*
2265  * A `swap extent' is a simple thing which maps a contiguous range of pages
2266  * onto a contiguous range of disk blocks.  An ordered list of swap extents
2267  * is built at swapon time and is then used at swap_writepage/swap_readpage
2268  * time for locating where on disk a page belongs.
2269  *
2270  * If the swapfile is an S_ISBLK block device, a single extent is installed.
2271  * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2272  * swap files identically.
2273  *
2274  * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2275  * extent list operates in PAGE_SIZE disk blocks.  Both S_ISREG and S_ISBLK
2276  * swapfiles are handled *identically* after swapon time.
2277  *
2278  * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2279  * and will parse them into an ordered extent list, in PAGE_SIZE chunks.  If
2280  * some stray blocks are found which do not fall within the PAGE_SIZE alignment
2281  * requirements, they are simply tossed out - we will never use those blocks
2282  * for swapping.
2283  *
2284  * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon.  This
2285  * prevents root from shooting her foot off by ftruncating an in-use swapfile,
2286  * which will scribble on the fs.
2287  *
2288  * The amount of disk space which a single swap extent represents varies.
2289  * Typically it is in the 1-4 megabyte range.  So we can have hundreds of
2290  * extents in the list.  To avoid much list walking, we cache the previous
2291  * search location in `curr_swap_extent', and start new searches from there.
2292  * This is extremely effective.  The average number of iterations in
2293  * map_swap_page() has been measured at about 0.3 per page.  - akpm.
2294  */
2295 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
2296 {
2297         struct file *swap_file = sis->swap_file;
2298         struct address_space *mapping = swap_file->f_mapping;
2299         struct inode *inode = mapping->host;
2300         int ret;
2301 
2302         if (S_ISBLK(inode->i_mode)) {
2303                 ret = add_swap_extent(sis, 0, sis->max, 0);
2304                 *span = sis->pages;
2305                 return ret;
2306         }
2307 
2308         if (mapping->a_ops->swap_activate) {
2309                 ret = mapping->a_ops->swap_activate(sis, swap_file, span);
2310                 if (ret >= 0)
2311                         sis->flags |= SWP_ACTIVATED;
2312                 if (!ret) {
2313                         sis->flags |= SWP_FS;
2314                         ret = add_swap_extent(sis, 0, sis->max, 0);
2315                         *span = sis->pages;
2316                 }
2317                 return ret;
2318         }
2319 
2320         return generic_swapfile_activate(sis, swap_file, span);
2321 }
2322 
2323 static int swap_node(struct swap_info_struct *p)
2324 {
2325         struct block_device *bdev;
2326 
2327         if (p->bdev)
2328                 bdev = p->bdev;
2329         else
2330                 bdev = p->swap_file->f_inode->i_sb->s_bdev;
2331 
2332         return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE;
2333 }
2334 
2335 static void _enable_swap_info(struct swap_info_struct *p, int prio,
2336                                 unsigned char *swap_map,
2337                                 struct swap_cluster_info *cluster_info)
2338 {
2339         int i;
2340 
2341         if (prio >= 0)
2342                 p->prio = prio;
2343         else
2344                 p->prio = --least_priority;
2345         /*
2346          * the plist prio is negated because plist ordering is
2347          * low-to-high, while swap ordering is high-to-low
2348          */
2349         p->list.prio = -p->prio;
2350         for_each_node(i) {
2351                 if (p->prio >= 0)
2352                         p->avail_lists[i].prio = -p->prio;
2353                 else {
2354                         if (swap_node(p) == i)
2355                                 p->avail_lists[i].prio = 1;
2356                         else
2357                                 p->avail_lists[i].prio = -p->prio;
2358                 }
2359         }
2360         p->swap_map = swap_map;
2361         p->cluster_info = cluster_info;
2362         p->flags |= SWP_WRITEOK;
2363         atomic_long_add(p->pages, &nr_swap_pages);
2364         total_swap_pages += p->pages;
2365 
2366         assert_spin_locked(&swap_lock);
2367         /*
2368          * both lists are plists, and thus priority ordered.
2369          * swap_active_head needs to be priority ordered for swapoff(),
2370          * which on removal of any swap_info_struct with an auto-assigned
2371          * (i.e. negative) priority increments the auto-assigned priority
2372          * of any lower-priority swap_info_structs.
2373          * swap_avail_head needs to be priority ordered for get_swap_page(),
2374          * which allocates swap pages from the highest available priority
2375          * swap_info_struct.
2376          */
2377         plist_add(&p->list, &swap_active_head);
2378         add_to_avail_list(p);
2379 }
2380 
2381 static void enable_swap_info(struct swap_info_struct *p, int prio,
2382                                 unsigned char *swap_map,
2383                                 struct swap_cluster_info *cluster_info,
2384                                 unsigned long *frontswap_map)
2385 {
2386         frontswap_init(p->type, frontswap_map);
2387         spin_lock(&swap_lock);
2388         spin_lock(&p->lock);
2389          _enable_swap_info(p, prio, swap_map, cluster_info);
2390         spin_unlock(&p->lock);
2391         spin_unlock(&swap_lock);
2392 }
2393 
2394 static void reinsert_swap_info(struct swap_info_struct *p)
2395 {
2396         spin_lock(&swap_lock);
2397         spin_lock(&p->lock);
2398         _enable_swap_info(p, p->prio, p->swap_map, p->cluster_info);
2399         spin_unlock(&p->lock);
2400         spin_unlock(&swap_lock);
2401 }
2402 
2403 bool has_usable_swap(void)
2404 {
2405         bool ret = true;
2406 
2407         spin_lock(&swap_lock);
2408         if (plist_head_empty(&swap_active_head))
2409                 ret = false;
2410         spin_unlock(&swap_lock);
2411         return ret;
2412 }
2413 
2414 SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
2415 {
2416         struct swap_info_struct *p = NULL;
2417         unsigned char *swap_map;
2418         struct swap_cluster_info *cluster_info;
2419         unsigned long *frontswap_map;
2420         struct file *swap_file, *victim;
2421         struct address_space *mapping;
2422         struct inode *inode;
2423         struct filename *pathname;
2424         int err, found = 0;
2425         unsigned int old_block_size;
2426 
2427         if (!capable(CAP_SYS_ADMIN))
2428                 return -EPERM;
2429 
2430         BUG_ON(!current->mm);
2431 
2432         pathname = getname(specialfile);
2433         if (IS_ERR(pathname))
2434                 return PTR_ERR(pathname);
2435 
2436         victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
2437         err = PTR_ERR(victim);
2438         if (IS_ERR(victim))
2439                 goto out;
2440 
2441         mapping = victim->f_mapping;
2442         spin_lock(&swap_lock);
2443         plist_for_each_entry(p, &swap_active_head, list) {
2444                 if (p->flags & SWP_WRITEOK) {
2445                         if (p->swap_file->f_mapping == mapping) {
2446                                 found = 1;
2447                                 break;
2448                         }
2449                 }
2450         }
2451         if (!found) {
2452                 err = -EINVAL;
2453                 spin_unlock(&swap_lock);
2454                 goto out_dput;
2455         }
2456         if (!security_vm_enough_memory_mm(current->mm, p->pages))
2457                 vm_unacct_memory(p->pages);
2458         else {
2459                 err = -ENOMEM;
2460                 spin_unlock(&swap_lock);
2461                 goto out_dput;
2462         }
2463         del_from_avail_list(p);
2464         spin_lock(&p->lock);
2465         if (p->prio < 0) {
2466                 struct swap_info_struct *si = p;
2467                 int nid;
2468 
2469                 plist_for_each_entry_continue(si, &swap_active_head, list) {
2470                         si->prio++;
2471                         si->list.prio--;
2472                         for_each_node(nid) {
2473                                 if (si->avail_lists[nid].prio != 1)
2474                                         si->avail_lists[nid].prio--;
2475                         }
2476                 }
2477                 least_priority++;
2478         }
2479         plist_del(&p->list, &swap_active_head);
2480         atomic_long_sub(p->pages, &nr_swap_pages);
2481         total_swap_pages -= p->pages;
2482         p->flags &= ~SWP_WRITEOK;
2483         spin_unlock(&p->lock);
2484         spin_unlock(&swap_lock);
2485 
2486         disable_swap_slots_cache_lock();
2487 
2488         set_current_oom_origin();
2489         err = try_to_unuse(p->type, false, 0); /* force unuse all pages */
2490         clear_current_oom_origin();
2491 
2492         if (err) {
2493                 /* re-insert swap space back into swap_list */
2494                 reinsert_swap_info(p);
2495                 reenable_swap_slots_cache_unlock();
2496                 goto out_dput;
2497         }
2498 
2499         reenable_swap_slots_cache_unlock();
2500 
2501         flush_work(&p->discard_work);
2502 
2503         destroy_swap_extents(p);
2504         if (p->flags & SWP_CONTINUED)
2505                 free_swap_count_continuations(p);
2506 
2507         if (!p->bdev || !blk_queue_nonrot(bdev_get_queue(p->bdev)))
2508                 atomic_dec(&nr_rotate_swap);
2509 
2510         mutex_lock(&swapon_mutex);
2511         spin_lock(&swap_lock);
2512         spin_lock(&p->lock);
2513         drain_mmlist();
2514 
2515         /* wait for anyone still in scan_swap_map */
2516         p->highest_bit = 0;             /* cuts scans short */
2517         while (p->flags >= SWP_SCANNING) {
2518                 spin_unlock(&p->lock);
2519                 spin_unlock(&swap_lock);
2520                 schedule_timeout_uninterruptible(1);
2521                 spin_lock(&swap_lock);
2522                 spin_lock(&p->lock);
2523         }
2524 
2525         swap_file = p->swap_file;
2526         old_block_size = p->old_block_size;
2527         p->swap_file = NULL;
2528         p->max = 0;
2529         swap_map = p->swap_map;
2530         p->swap_map = NULL;
2531         cluster_info = p->cluster_info;
2532         p->cluster_info = NULL;
2533         frontswap_map = frontswap_map_get(p);
2534         spin_unlock(&p->lock);
2535         spin_unlock(&swap_lock);
2536         frontswap_invalidate_area(p->type);
2537         frontswap_map_set(p, NULL);
2538         mutex_unlock(&swapon_mutex);
2539         free_percpu(p->percpu_cluster);
2540         p->percpu_cluster = NULL;
2541         vfree(swap_map);
2542         kvfree(cluster_info);
2543         kvfree(frontswap_map);
2544         /* Destroy swap account information */
2545         swap_cgroup_swapoff(p->type);
2546         exit_swap_address_space(p->type);
2547 
2548         inode = mapping->host;
2549         if (S_ISBLK(inode->i_mode)) {
2550                 struct block_device *bdev = I_BDEV(inode);
2551                 set_blocksize(bdev, old_block_size);
2552                 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2553         } else {
2554                 inode_lock(inode);
2555                 inode->i_flags &= ~S_SWAPFILE;
2556                 inode_unlock(inode);
2557         }
2558         filp_close(swap_file, NULL);
2559 
2560         /*
2561          * Clear the SWP_USED flag after all resources are freed so that swapon
2562          * can reuse this swap_info in alloc_swap_info() safely.  It is ok to
2563          * not hold p->lock after we cleared its SWP_WRITEOK.
2564          */
2565         spin_lock(&swap_lock);
2566         p->flags = 0;
2567         spin_unlock(&swap_lock);
2568 
2569         err = 0;
2570         atomic_inc(&proc_poll_event);
2571         wake_up_interruptible(&proc_poll_wait);
2572 
2573 out_dput:
2574         filp_close(victim, NULL);
2575 out:
2576         putname(pathname);
2577         return err;
2578 }
2579 
2580 #ifdef CONFIG_PROC_FS
2581 static __poll_t swaps_poll(struct file *file, poll_table *wait)
2582 {
2583         struct seq_file *seq = file->private_data;
2584 
2585         poll_wait(file, &proc_poll_wait, wait);
2586 
2587         if (seq->poll_event != atomic_read(&proc_poll_event)) {
2588                 seq->poll_event = atomic_read(&proc_poll_event);
2589                 return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
2590         }
2591 
2592         return EPOLLIN | EPOLLRDNORM;
2593 }
2594 
2595 /* iterator */
2596 static void *swap_start(struct seq_file *swap, loff_t *pos)
2597 {
2598         struct swap_info_struct *si;
2599         int type;
2600         loff_t l = *pos;
2601 
2602         mutex_lock(&swapon_mutex);
2603 
2604         if (!l)
2605                 return SEQ_START_TOKEN;
2606 
2607         for (type = 0; (si = swap_type_to_swap_info(type)); type++) {
2608                 if (!(si->flags & SWP_USED) || !si->swap_map)
2609                         continue;
2610                 if (!--l)
2611                         return si;
2612         }
2613 
2614         return NULL;
2615 }
2616 
2617 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2618 {
2619         struct swap_info_struct *si = v;
2620         int type;
2621 
2622         if (v == SEQ_START_TOKEN)
2623                 type = 0;
2624         else
2625                 type = si->type + 1;
2626 
2627         for (; (si = swap_type_to_swap_info(type)); type++) {
2628                 if (!(si->flags & SWP_USED) || !si->swap_map)
2629                         continue;
2630                 ++*pos;
2631                 return si;
2632         }
2633 
2634         return NULL;
2635 }
2636 
2637 static void swap_stop(struct seq_file *swap, void *v)
2638 {
2639         mutex_unlock(&swapon_mutex);
2640 }
2641 
2642 static int swap_show(struct seq_file *swap, void *v)
2643 {
2644         struct swap_info_struct *si = v;
2645         struct file *file;
2646         int len;
2647 
2648         if (si == SEQ_START_TOKEN) {
2649                 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2650                 return 0;
2651         }
2652 
2653         file = si->swap_file;
2654         len = seq_file_path(swap, file, " \t\n\\");
2655         seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
2656                         len < 40 ? 40 - len : 1, " ",
2657                         S_ISBLK(file_inode(file)->i_mode) ?
2658                                 "partition" : "file\t",
2659                         si->pages << (PAGE_SHIFT - 10),
2660                         si->inuse_pages << (PAGE_SHIFT - 10),
2661                         si->prio);
2662         return 0;
2663 }
2664 
2665 static const struct seq_operations swaps_op = {
2666         .start =        swap_start,
2667         .next =         swap_next,
2668         .stop =         swap_stop,
2669         .show =         swap_show
2670 };
2671 
2672 static int swaps_open(struct inode *inode, struct file *file)
2673 {
2674         struct seq_file *seq;
2675         int ret;
2676 
2677         ret = seq_open(file, &swaps_op);
2678         if (ret)
2679                 return ret;
2680 
2681         seq = file->private_data;
2682         seq->poll_event = atomic_read(&proc_poll_event);
2683         return 0;
2684 }
2685 
2686 static const struct file_operations proc_swaps_operations = {
2687         .open           = swaps_open,
2688         .read           = seq_read,
2689         .llseek         = seq_lseek,
2690         .release        = seq_release,
2691         .poll           = swaps_poll,
2692 };
2693 
2694 static int __init procswaps_init(void)
2695 {
2696         proc_create("swaps", 0, NULL, &proc_swaps_operations);
2697         return 0;
2698 }
2699 __initcall(procswaps_init);
2700 #endif /* CONFIG_PROC_FS */
2701 
2702 #ifdef MAX_SWAPFILES_CHECK
2703 static int __init max_swapfiles_check(void)
2704 {
2705         MAX_SWAPFILES_CHECK();
2706         return 0;
2707 }
2708 late_initcall(max_swapfiles_check);
2709 #endif
2710 
2711 static struct swap_info_struct *alloc_swap_info(void)
2712 {
2713         struct swap_info_struct *p;
2714         unsigned int type;
2715         int i;
2716 
2717         p = kvzalloc(struct_size(p, avail_lists, nr_node_ids), GFP_KERNEL);
2718         if (!p)
2719                 return ERR_PTR(-ENOMEM);
2720 
2721         spin_lock(&swap_lock);
2722         for (type = 0; type < nr_swapfiles; type++) {
2723                 if (!(swap_info[type]->flags & SWP_USED))
2724                         break;
2725         }
2726         if (type >= MAX_SWAPFILES) {
2727                 spin_unlock(&swap_lock);
2728                 kvfree(p);
2729                 return ERR_PTR(-EPERM);
2730         }
2731         if (type >= nr_swapfiles) {
2732                 p->type = type;
2733                 WRITE_ONCE(swap_info[type], p);
2734                 /*
2735                  * Write swap_info[type] before nr_swapfiles, in case a
2736                  * racing procfs swap_start() or swap_next() is reading them.
2737                  * (We never shrink nr_swapfiles, we never free this entry.)
2738                  */
2739                 smp_wmb();
2740                 WRITE_ONCE(nr_swapfiles, nr_swapfiles + 1);
2741         } else {
2742                 kvfree(p);
2743                 p = swap_info[type];
2744                 /*
2745                  * Do not memset this entry: a racing procfs swap_next()
2746                  * would be relying on p->type to remain valid.
2747                  */
2748         }
2749         INIT_LIST_HEAD(&p->first_swap_extent.list);
2750         plist_node_init(&p->list, 0);
2751         for_each_node(i)
2752                 plist_node_init(&p->avail_lists[i], 0);
2753         p->flags = SWP_USED;
2754         spin_unlock(&swap_lock);
2755         spin_lock_init(&p->lock);
2756         spin_lock_init(&p->cont_lock);
2757 
2758         return p;
2759 }
2760 
2761 static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2762 {
2763         int error;
2764 
2765         if (S_ISBLK(inode->i_mode)) {
2766                 p->bdev = bdgrab(I_BDEV(inode));
2767                 error = blkdev_get(p->bdev,
2768                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL, p);
2769                 if (error < 0) {
2770                         p->bdev = NULL;
2771                         return error;
2772                 }
2773                 p->old_block_size = block_size(p->bdev);
2774                 error = set_blocksize(p->bdev, PAGE_SIZE);
2775                 if (error < 0)
2776                         return error;
2777                 p->flags |= SWP_BLKDEV;
2778         } else if (S_ISREG(inode->i_mode)) {
2779                 p->bdev = inode->i_sb->s_bdev;
2780                 inode_lock(inode);
2781                 if (IS_SWAPFILE(inode))
2782                         return -EBUSY;
2783         } else
2784                 return -EINVAL;
2785 
2786         return 0;
2787 }
2788 
2789 
2790 /*
2791  * Find out how many pages are allowed for a single swap device. There
2792  * are two limiting factors:
2793  * 1) the number of bits for the swap offset in the swp_entry_t type, and
2794  * 2) the number of bits in the swap pte, as defined by the different
2795  * architectures.
2796  *
2797  * In order to find the largest possible bit mask, a swap entry with
2798  * swap type 0 and swap offset ~0UL is created, encoded to a swap pte,
2799  * decoded to a swp_entry_t again, and finally the swap offset is
2800  * extracted.
2801  *
2802  * This will mask all the bits from the initial ~0UL mask that can't
2803  * be encoded in either the swp_entry_t or the architecture definition
2804  * of a swap pte.
2805  */
2806 unsigned long generic_max_swapfile_size(void)
2807 {
2808         return swp_offset(pte_to_swp_entry(
2809                         swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2810 }
2811 
2812 /* Can be overridden by an architecture for additional checks. */
2813 __weak unsigned long max_swapfile_size(void)
2814 {
2815         return generic_max_swapfile_size();
2816 }
2817 
2818 static unsigned long read_swap_header(struct swap_info_struct *p,
2819                                         union swap_header *swap_header,
2820                                         struct inode *inode)
2821 {
2822         int i;
2823         unsigned long maxpages;
2824         unsigned long swapfilepages;
2825         unsigned long last_page;
2826 
2827         if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
2828                 pr_err("Unable to find swap-space signature\n");
2829                 return 0;
2830         }
2831 
2832         /* swap partition endianess hack... */
2833         if (swab32(swap_header->info.version) == 1) {
2834                 swab32s(&swap_header->info.version);
2835                 swab32s(&swap_header->info.last_page);
2836                 swab32s(&swap_header->info.nr_badpages);
2837                 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2838                         return 0;
2839                 for (i = 0; i < swap_header->info.nr_badpages; i++)
2840                         swab32s(&swap_header->info.badpages[i]);
2841         }
2842         /* Check the swap header's sub-version */
2843         if (swap_header->info.version != 1) {
2844                 pr_warn("Unable to handle swap header version %d\n",
2845                         swap_header->info.version);
2846                 return 0;
2847         }
2848 
2849         p->lowest_bit  = 1;
2850         p->cluster_next = 1;
2851         p->cluster_nr = 0;
2852 
2853         maxpages = max_swapfile_size();
2854         last_page = swap_header->info.last_page;
2855         if (!last_page) {
2856                 pr_warn("Empty swap-file\n");
2857                 return 0;
2858         }
2859         if (last_page > maxpages) {
2860                 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
2861                         maxpages << (PAGE_SHIFT - 10),
2862                         last_page << (PAGE_SHIFT - 10));
2863         }
2864         if (maxpages > last_page) {
2865                 maxpages = last_page + 1;
2866                 /* p->max is an unsigned int: don't overflow it */
2867                 if ((unsigned int)maxpages == 0)
2868                         maxpages = UINT_MAX;
2869         }
2870         p->highest_bit = maxpages - 1;
2871 
2872         if (!maxpages)
2873                 return 0;
2874         swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2875         if (swapfilepages && maxpages > swapfilepages) {
2876                 pr_warn("Swap area shorter than signature indicates\n");
2877                 return 0;
2878         }
2879         if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
2880                 return 0;
2881         if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2882                 return 0;
2883 
2884         return maxpages;
2885 }
2886 
2887 #define SWAP_CLUSTER_INFO_COLS                                          \
2888         DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
2889 #define SWAP_CLUSTER_SPACE_COLS                                         \
2890         DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
2891 #define SWAP_CLUSTER_COLS                                               \
2892         max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
2893 
2894 static int setup_swap_map_and_extents(struct swap_info_struct *p,
2895                                         union swap_header *swap_header,
2896                                         unsigned char *swap_map,
2897                                         struct swap_cluster_info *cluster_info,
2898                                         unsigned long maxpages,
2899                                         sector_t *span)
2900 {
2901         unsigned int j, k;
2902         unsigned int nr_good_pages;
2903         int nr_extents;
2904         unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
2905         unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS;
2906         unsigned long i, idx;
2907 
2908         nr_good_pages = maxpages - 1;   /* omit header page */
2909 
2910         cluster_list_init(&p->free_clusters);
2911         cluster_list_init(&p->discard_clusters);
2912 
2913         for (i = 0; i < swap_header->info.nr_badpages; i++) {
2914                 unsigned int page_nr = swap_header->info.badpages[i];
2915                 if (page_nr == 0 || page_nr > swap_header->info.last_page)
2916                         return -EINVAL;
2917                 if (page_nr < maxpages) {
2918                         swap_map[page_nr] = SWAP_MAP_BAD;
2919                         nr_good_pages--;
2920                         /*
2921                          * Haven't marked the cluster free yet, no list
2922                          * operation involved
2923                          */
2924                         inc_cluster_info_page(p, cluster_info, page_nr);
2925                 }
2926         }
2927 
2928         /* Haven't marked the cluster free yet, no list operation involved */
2929         for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
2930                 inc_cluster_info_page(p, cluster_info, i);
2931 
2932         if (nr_good_pages) {
2933                 swap_map[0] = SWAP_MAP_BAD;
2934                 /*
2935                  * Not mark the cluster free yet, no list
2936                  * operation involved
2937                  */
2938                 inc_cluster_info_page(p, cluster_info, 0);
2939                 p->max = maxpages;
2940                 p->pages = nr_good_pages;
2941                 nr_extents = setup_swap_extents(p, span);
2942                 if (nr_extents < 0)
2943                         return nr_extents;
2944                 nr_good_pages = p->pages;
2945         }
2946         if (!nr_good_pages) {
2947                 pr_warn("Empty swap-file\n");
2948                 return -EINVAL;
2949         }
2950 
2951         if (!cluster_info)
2952                 return nr_extents;
2953 
2954 
2955         /*
2956          * Reduce false cache line sharing between cluster_info and
2957          * sharing same address space.
2958          */
2959         for (k = 0; k < SWAP_CLUSTER_COLS; k++) {
2960                 j = (k + col) % SWAP_CLUSTER_COLS;
2961                 for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) {
2962                         idx = i * SWAP_CLUSTER_COLS + j;
2963                         if (idx >= nr_clusters)
2964                                 continue;
2965                         if (cluster_count(&cluster_info[idx]))
2966                                 continue;
2967                         cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
2968                         cluster_list_add_tail(&p->free_clusters, cluster_info,
2969                                               idx);
2970                 }
2971         }
2972         return nr_extents;
2973 }
2974 
2975 /*
2976  * Helper to sys_swapon determining if a given swap
2977  * backing device queue supports DISCARD operations.
2978  */
2979 static bool swap_discardable(struct swap_info_struct *si)
2980 {
2981         struct request_queue *q = bdev_get_queue(si->bdev);
2982 
2983         if (!q || !blk_queue_discard(q))
2984                 return false;
2985 
2986         return true;
2987 }
2988 
2989 SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2990 {
2991         struct swap_info_struct *p;
2992         struct filename *name;
2993         struct file *swap_file = NULL;
2994         struct address_space *mapping;
2995         int prio;
2996         int error;
2997         union swap_header *swap_header;
2998         int nr_extents;
2999         sector_t span;
3000         unsigned long maxpages;
3001         unsigned char *swap_map = NULL;
3002         struct swap_cluster_info *cluster_info = NULL;
3003         unsigned long *frontswap_map = NULL;
3004         struct page *page = NULL;
3005         struct inode *inode = NULL;
3006         bool inced_nr_rotate_swap = false;
3007 
3008         if (swap_flags & ~SWAP_FLAGS_VALID)
3009                 return -EINVAL;
3010 
3011         if (!capable(CAP_SYS_ADMIN))
3012                 return -EPERM;
3013 
3014         if (!swap_avail_heads)
3015                 return -ENOMEM;
3016 
3017         p = alloc_swap_info();
3018         if (IS_ERR(p))
3019                 return PTR_ERR(p);
3020 
3021         INIT_WORK(&p->discard_work, swap_discard_work);
3022 
3023         name = getname(specialfile);
3024         if (IS_ERR(name)) {
3025                 error = PTR_ERR(name);
3026                 name = NULL;
3027                 goto bad_swap;
3028         }
3029         swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
3030         if (IS_ERR(swap_file)) {
3031                 error = PTR_ERR(swap_file);
3032                 swap_file = NULL;
3033                 goto bad_swap;
3034         }
3035 
3036         p->swap_file = swap_file;
3037         mapping = swap_file->f_mapping;
3038         inode = mapping->host;
3039 
3040         /* If S_ISREG(inode->i_mode) will do inode_lock(inode); */
3041         error = claim_swapfile(p, inode);
3042         if (unlikely(error))
3043                 goto bad_swap;
3044 
3045         /*
3046          * Read the swap header.
3047          */
3048         if (!mapping->a_ops->readpage) {
3049                 error = -EINVAL;
3050                 goto bad_swap;
3051         }
3052         page = read_mapping_page(mapping, 0, swap_file);
3053         if (IS_ERR(page)) {
3054                 error = PTR_ERR(page);
3055                 goto bad_swap;
3056         }
3057         swap_header = kmap(page);
3058 
3059         maxpages = read_swap_header(p, swap_header, inode);
3060         if (unlikely(!maxpages)) {
3061                 error = -EINVAL;
3062                 goto bad_swap;
3063         }
3064 
3065         /* OK, set up the swap map and apply the bad block list */
3066         swap_map = vzalloc(maxpages);
3067         if (!swap_map) {
3068                 error = -ENOMEM;
3069                 goto bad_swap;
3070         }
3071 
3072         if (bdi_cap_stable_pages_required(inode_to_bdi(inode)))
3073                 p->flags |= SWP_STABLE_WRITES;
3074 
3075         if (bdi_cap_synchronous_io(inode_to_bdi(inode)))
3076                 p->flags |= SWP_SYNCHRONOUS_IO;
3077 
3078         if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) {
3079                 int cpu;
3080                 unsigned long ci, nr_cluster;
3081 
3082                 p->flags |= SWP_SOLIDSTATE;
3083                 /*
3084                  * select a random position to start with to help wear leveling
3085                  * SSD
3086                  */
3087                 p->cluster_next = 1 + (prandom_u32() % p->highest_bit);
3088                 nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
3089 
3090                 cluster_info = kvcalloc(nr_cluster, sizeof(*cluster_info),
3091                                         GFP_KERNEL);
3092                 if (!cluster_info) {
3093                         error = -ENOMEM;
3094                         goto bad_swap;
3095                 }
3096 
3097                 for (ci = 0; ci < nr_cluster; ci++)
3098                         spin_lock_init(&((cluster_info + ci)->lock));
3099 
3100                 p->percpu_cluster = alloc_percpu(struct percpu_cluster);
3101                 if (!p->percpu_cluster) {
3102                         error = -ENOMEM;
3103                         goto bad_swap;
3104                 }
3105                 for_each_possible_cpu(cpu) {
3106                         struct percpu_cluster *cluster;
3107                         cluster = per_cpu_ptr(p->percpu_cluster, cpu);
3108                         cluster_set_null(&cluster->index);
3109                 }
3110         } else {
3111                 atomic_inc(&nr_rotate_swap);
3112                 inced_nr_rotate_swap = true;
3113         }
3114 
3115         error = swap_cgroup_swapon(p->type, maxpages);
3116         if (error)
3117                 goto bad_swap;
3118 
3119         nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
3120                 cluster_info, maxpages, &span);
3121         if (unlikely(nr_extents < 0)) {
3122                 error = nr_extents;
3123                 goto bad_swap;
3124         }
3125         /* frontswap enabled? set up bit-per-page map for frontswap */
3126         if (IS_ENABLED(CONFIG_FRONTSWAP))
3127                 frontswap_map = kvcalloc(BITS_TO_LONGS(maxpages),
3128                                          sizeof(long),
3129                                          GFP_KERNEL);
3130 
3131         if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) {
3132                 /*
3133                  * When discard is enabled for swap with no particular
3134                  * policy flagged, we set all swap discard flags here in
3135                  * order to sustain backward compatibility with older
3136                  * swapon(8) releases.
3137                  */
3138                 p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
3139                              SWP_PAGE_DISCARD);
3140 
3141                 /*
3142                  * By flagging sys_swapon, a sysadmin can tell us to
3143                  * either do single-time area discards only, or to just
3144                  * perform discards for released swap page-clusters.
3145                  * Now it's time to adjust the p->flags accordingly.
3146                  */
3147                 if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
3148                         p->flags &= ~SWP_PAGE_DISCARD;
3149                 else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
3150                         p->flags &= ~SWP_AREA_DISCARD;
3151 
3152                 /* issue a swapon-time discard if it's still required */
3153                 if (p->flags & SWP_AREA_DISCARD) {
3154                         int err = discard_swap(p);
3155                         if (unlikely(err))
3156                                 pr_err("swapon: discard_swap(%p): %d\n",
3157                                         p, err);
3158                 }
3159         }
3160 
3161         error = init_swap_address_space(p->type, maxpages);
3162         if (error)
3163                 goto bad_swap;
3164 
3165         mutex_lock(&swapon_mutex);
3166         prio = -1;
3167         if (swap_flags & SWAP_FLAG_PREFER)
3168                 prio =
3169                   (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
3170         enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
3171 
3172         pr_info("Adding %uk swap on %s.  Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
3173                 p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
3174                 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
3175                 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
3176                 (p->flags & SWP_DISCARDABLE) ? "D" : "",
3177                 (p->flags & SWP_AREA_DISCARD) ? "s" : "",
3178                 (p->flags & SWP_PAGE_DISCARD) ? "c" : "",
3179                 (frontswap_map) ? "FS" : "");
3180 
3181         mutex_unlock(&swapon_mutex);
3182         atomic_inc(&proc_poll_event);
3183         wake_up_interruptible(&proc_poll_wait);
3184 
3185         if (S_ISREG(inode->i_mode))
3186                 inode->i_flags |= S_SWAPFILE;
3187         error = 0;
3188         goto out;
3189 bad_swap:
3190         free_percpu(p->percpu_cluster);
3191         p->percpu_cluster = NULL;
3192         if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
3193                 set_blocksize(p->bdev, p->old_block_size);
3194                 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
3195         }
3196         destroy_swap_extents(p);
3197         swap_cgroup_swapoff(p->type);
3198         spin_lock(&swap_lock);
3199         p->swap_file = NULL;
3200         p->flags = 0;
3201         spin_unlock(&swap_lock);
3202         vfree(swap_map);
3203         kvfree(cluster_info);
3204         kvfree(frontswap_map);
3205         if (inced_nr_rotate_swap)
3206                 atomic_dec(&nr_rotate_swap);
3207         if (swap_file) {
3208                 if (inode && S_ISREG(inode->i_mode)) {
3209                         inode_unlock(inode);
3210                         inode = NULL;
3211                 }
3212                 filp_close(swap_file, NULL);
3213         }
3214 out:
3215         if (page && !IS_ERR(page)) {
3216                 kunmap(page);
3217                 put_page(page);
3218         }
3219         if (name)
3220                 putname(name);
3221         if (inode && S_ISREG(inode->i_mode))
3222                 inode_unlock(inode);
3223         if (!error)
3224                 enable_swap_slots_cache();
3225         return error;
3226 }
3227 
3228 void si_swapinfo(struct sysinfo *val)
3229 {
3230         unsigned int type;
3231         unsigned long nr_to_be_unused = 0;
3232 
3233         spin_lock(&swap_lock);
3234         for (type = 0; type < nr_swapfiles; type++) {
3235                 struct swap_info_struct *si = swap_info[type];
3236 
3237                 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
3238                         nr_to_be_unused += si->inuse_pages;
3239         }
3240         val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
3241         val->totalswap = total_swap_pages + nr_to_be_unused;
3242         spin_unlock(&swap_lock);
3243 }
3244 
3245 /*
3246  * Verify that a swap entry is valid and increment its swap map count.
3247  *
3248  * Returns error code in following case.
3249  * - success -> 0
3250  * - swp_entry is invalid -> EINVAL
3251  * - swp_entry is migration entry -> EINVAL
3252  * - swap-cache reference is requested but there is already one. -> EEXIST
3253  * - swap-cache reference is requested but the entry is not used. -> ENOENT
3254  * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
3255  */
3256 static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
3257 {
3258         struct swap_info_struct *p;
3259         struct swap_cluster_info *ci;
3260         unsigned long offset;
3261         unsigned char count;
3262         unsigned char has_cache;
3263         int err = -EINVAL;
3264 
3265         if (non_swap_entry(entry))
3266                 goto out;
3267 
3268         p = swp_swap_info(entry);
3269         if (!p)
3270                 goto bad_file;
3271 
3272         offset = swp_offset(entry);
3273         if (unlikely(offset >= p->max))
3274                 goto out;
3275 
3276         ci = lock_cluster_or_swap_info(p, offset);
3277 
3278         count = p->swap_map[offset];
3279 
3280         /*
3281          * swapin_readahead() doesn't check if a swap entry is valid, so the
3282          * swap entry could be SWAP_MAP_BAD. Check here with lock held.
3283          */
3284         if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
3285                 err = -ENOENT;
3286                 goto unlock_out;
3287         }
3288 
3289         has_cache = count & SWAP_HAS_CACHE;
3290         count &= ~SWAP_HAS_CACHE;
3291         err = 0;
3292 
3293         if (usage == SWAP_HAS_CACHE) {
3294 
3295                 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
3296                 if (!has_cache && count)
3297                         has_cache = SWAP_HAS_CACHE;
3298                 else if (has_cache)             /* someone else added cache */
3299                         err = -EEXIST;
3300                 else                            /* no users remaining */
3301                         err = -ENOENT;
3302 
3303         } else if (count || has_cache) {
3304 
3305                 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
3306                         count += usage;
3307                 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
3308                         err = -EINVAL;
3309                 else if (swap_count_continued(p, offset, count))
3310                         count = COUNT_CONTINUED;
3311                 else
3312                         err = -ENOMEM;
3313         } else
3314                 err = -ENOENT;                  /* unused swap entry */
3315 
3316         p->swap_map[offset] = count | has_cache;
3317 
3318 unlock_out:
3319         unlock_cluster_or_swap_info(p, ci);
3320 out:
3321         return err;
3322 
3323 bad_file:
3324         pr_err("swap_dup: %s%08lx\n", Bad_file, entry.val);
3325         goto out;
3326 }
3327 
3328 /*
3329  * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3330  * (in which case its reference count is never incremented).
3331  */
3332 void swap_shmem_alloc(swp_entry_t entry)
3333 {
3334         __swap_duplicate(entry, SWAP_MAP_SHMEM);
3335 }
3336 
3337 /*
3338  * Increase reference count of swap entry by 1.
3339  * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3340  * but could not be atomically allocated.  Returns 0, just as if it succeeded,
3341  * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3342  * might occur if a page table entry has got corrupted.
3343  */
3344 int swap_duplicate(swp_entry_t entry)
3345 {
3346         int err = 0;
3347 
3348         while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
3349                 err = add_swap_count_continuation(entry, GFP_ATOMIC);
3350         return err;
3351 }
3352 
3353 /*
3354  * @entry: swap entry for which we allocate swap cache.
3355  *
3356  * Called when allocating swap cache for existing swap entry,
3357  * This can return error codes. Returns 0 at success.
3358  * -EBUSY means there is a swap cache.
3359  * Note: return code is different from swap_duplicate().
3360  */
3361 int swapcache_prepare(swp_entry_t entry)
3362 {
3363         return __swap_duplicate(entry, SWAP_HAS_CACHE);
3364 }
3365 
3366 struct swap_info_struct *swp_swap_info(swp_entry_t entry)
3367 {
3368         return swap_type_to_swap_info(swp_type(entry));
3369 }
3370 
3371 struct swap_info_struct *page_swap_info(struct page *page)
3372 {
3373         swp_entry_t entry = { .val = page_private(page) };
3374         return swp_swap_info(entry);
3375 }
3376 
3377 /*
3378  * out-of-line __page_file_ methods to avoid include hell.
3379  */
3380 struct address_space *__page_file_mapping(struct page *page)
3381 {
3382         return page_swap_info(page)->swap_file->f_mapping;
3383 }
3384 EXPORT_SYMBOL_GPL(__page_file_mapping);
3385 
3386 pgoff_t __page_file_index(struct page *page)
3387 {
3388         swp_entry_t swap = { .val = page_private(page) };
3389         return swp_offset(swap);
3390 }
3391 EXPORT_SYMBOL_GPL(__page_file_index);
3392 
3393 /*
3394  * add_swap_count_continuation - called when a swap count is duplicated
3395  * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3396  * page of the original vmalloc'ed swap_map, to hold the continuation count
3397  * (for that entry and for its neighbouring PAGE_SIZE swap entries).  Called
3398  * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3399  *
3400  * These continuation pages are seldom referenced: the common paths all work
3401  * on the original swap_map, only referring to a continuation page when the
3402  * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3403  *
3404  * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3405  * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3406  * can be called after dropping locks.
3407  */
3408 int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
3409 {
3410         struct swap_info_struct *si;
3411         struct swap_cluster_info *ci;
3412         struct page *head;
3413         struct page *page;
3414         struct page *list_page;
3415         pgoff_t offset;
3416         unsigned char count;
3417 
3418         /*
3419          * When debugging, it's easier to use __GFP_ZERO here; but it's better
3420          * for latency not to zero a page while GFP_ATOMIC and holding locks.
3421          */
3422         page = alloc_page(gfp_mask | __GFP_HIGHMEM);
3423 
3424         si = swap_info_get(entry);
3425         if (!si) {
3426                 /*
3427                  * An acceptable race has occurred since the failing
3428                  * __swap_duplicate(): the swap entry has been freed,
3429                  * perhaps even the whole swap_map cleared for swapoff.
3430                  */
3431                 goto outer;
3432         }
3433 
3434         offset = swp_offset(entry);
3435 
3436         ci = lock_cluster(si, offset);
3437 
3438         count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
3439 
3440         if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
3441                 /*
3442                  * The higher the swap count, the more likely it is that tasks
3443                  * will race to add swap count continuation: we need to avoid
3444                  * over-provisioning.
3445                  */
3446                 goto out;
3447         }
3448 
3449         if (!page) {
3450                 unlock_cluster(ci);
3451                 spin_unlock(&si->lock);
3452                 return -ENOMEM;
3453         }
3454 
3455         /*
3456          * We are fortunate that although vmalloc_to_page uses pte_offset_map,
3457          * no architecture is using highmem pages for kernel page tables: so it
3458          * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
3459          */
3460         head = vmalloc_to_page(si->swap_map + offset);
3461         offset &= ~PAGE_MASK;
3462 
3463         spin_lock(&si->cont_lock);
3464         /*
3465          * Page allocation does not initialize the page's lru field,
3466          * but it does always reset its private field.
3467          */
3468         if (!page_private(head)) {
3469                 BUG_ON(count & COUNT_CONTINUED);
3470                 INIT_LIST_HEAD(&head->lru);
3471                 set_page_private(head, SWP_CONTINUED);
3472                 si->flags |= SWP_CONTINUED;
3473         }
3474 
3475         list_for_each_entry(list_page, &head->lru, lru) {
3476                 unsigned char *map;
3477 
3478                 /*
3479                  * If the previous map said no continuation, but we've found
3480                  * a continuation page, free our allocation and use this one.
3481                  */
3482                 if (!(count & COUNT_CONTINUED))
3483                         goto out_unlock_cont;
3484 
3485                 map = kmap_atomic(list_page) + offset;
3486                 count = *map;
3487                 kunmap_atomic(map);
3488 
3489                 /*
3490                  * If this continuation count now has some space in it,
3491                  * free our allocation and use this one.
3492                  */
3493                 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
3494                         goto out_unlock_cont;
3495         }
3496 
3497         list_add_tail(&page->lru, &head->lru);
3498         page = NULL;                    /* now it's attached, don't free it */
3499 out_unlock_cont:
3500         spin_unlock(&si->cont_lock);
3501 out:
3502         unlock_cluster(ci);
3503         spin_unlock(&si->lock);
3504 outer:
3505         if (page)
3506                 __free_page(page);
3507         return 0;
3508 }
3509 
3510 /*
3511  * swap_count_continued - when the original swap_map count is incremented
3512  * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3513  * into, carry if so, or else fail until a new continuation page is allocated;
3514  * when the original swap_map count is decremented from 0 with continuation,
3515  * borrow from the continuation and report whether it still holds more.
3516  * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3517  * lock.
3518  */
3519 static bool swap_count_continued(struct swap_info_struct *si,
3520                                  pgoff_t offset, unsigned char count)
3521 {
3522         struct page *head;
3523         struct page *page;
3524         unsigned char *map;
3525         bool ret;
3526 
3527         head = vmalloc_to_page(si->swap_map + offset);
3528         if (page_private(head) != SWP_CONTINUED) {
3529                 BUG_ON(count & COUNT_CONTINUED);
3530                 return false;           /* need to add count continuation */
3531         }
3532 
3533         spin_lock(&si->cont_lock);
3534         offset &= ~PAGE_MASK;
3535         page = list_entry(head->lru.next, struct page, lru);
3536         map = kmap_atomic(page) + offset;
3537 
3538         if (count == SWAP_MAP_MAX)      /* initial increment from swap_map */
3539                 goto init_map;          /* jump over SWAP_CONT_MAX checks */
3540 
3541         if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
3542                 /*
3543                  * Think of how you add 1 to 999
3544                  */
3545                 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
3546                         kunmap_atomic(map);
3547                         page = list_entry(page->lru.next, struct page, lru);
3548                         BUG_ON(page == head);
3549                         map = kmap_atomic(page) + offset;
3550                 }
3551                 if (*map == SWAP_CONT_MAX) {
3552                         kunmap_atomic(map);
3553                         page = list_entry(page->lru.next, struct page, lru);
3554                         if (page == head) {
3555                                 ret = false;    /* add count continuation */
3556                                 goto out;
3557                         }
3558                         map = kmap_atomic(page) + offset;
3559 init_map:               *map = 0;               /* we didn't zero the page */
3560                 }
3561                 *map += 1;
3562                 kunmap_atomic(map);
3563                 page = list_entry(page->lru.prev, struct page, lru);
3564                 while (page != head) {
3565                         map = kmap_atomic(page) + offset;
3566                         *map = COUNT_CONTINUED;
3567                         kunmap_atomic(map);
3568                         page = list_entry(page->lru.prev, struct page, lru);
3569                 }
3570                 ret = true;                     /* incremented */
3571 
3572         } else {                                /* decrementing */
3573                 /*
3574                  * Think of how you subtract 1 from 1000
3575                  */
3576                 BUG_ON(count != COUNT_CONTINUED);
3577                 while (*map == COUNT_CONTINUED) {
3578                         kunmap_atomic(map);
3579                         page = list_entry(page->lru.next, struct page, lru);
3580                         BUG_ON(page == head);
3581                         map = kmap_atomic(page) + offset;
3582                 }
3583                 BUG_ON(*map == 0);
3584                 *map -= 1;
3585                 if (*map == 0)
3586                         count = 0;
3587                 kunmap_atomic(map);
3588                 page = list_entry(page->lru.prev, struct page, lru);
3589                 while (page != head) {
3590                         map = kmap_atomic(page) + offset;
3591                         *map = SWAP_CONT_MAX | count;
3592                         count = COUNT_CONTINUED;
3593                         kunmap_atomic(map);
3594                         page = list_entry(page->lru.prev, struct page, lru);
3595                 }
3596                 ret = count == COUNT_CONTINUED;
3597         }
3598 out:
3599         spin_unlock(&si->cont_lock);
3600         return ret;
3601 }
3602 
3603 /*
3604  * free_swap_count_continuations - swapoff free all the continuation pages
3605  * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3606  */
3607 static void free_swap_count_continuations(struct swap_info_struct *si)
3608 {
3609         pgoff_t offset;
3610 
3611         for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
3612                 struct page *head;
3613                 head = vmalloc_to_page(si->swap_map + offset);
3614                 if (page_private(head)) {
3615                         struct page *page, *next;
3616 
3617                         list_for_each_entry_safe(page, next, &head->lru, lru) {
3618                                 list_del(&page->lru);
3619                                 __free_page(page);
3620                         }
3621                 }
3622         }
3623 }
3624 
3625 #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
3626 void mem_cgroup_throttle_swaprate(struct mem_cgroup *memcg, int node,
3627                                   gfp_t gfp_mask)
3628 {
3629         struct swap_info_struct *si, *next;
3630         if (!(gfp_mask & __GFP_IO) || !memcg)
3631                 return;
3632 
3633         if (!blk_cgroup_congested())
3634                 return;
3635 
3636         /*
3637          * We've already scheduled a throttle, avoid taking the global swap
3638          * lock.
3639          */
3640         if (current->throttle_queue)
3641                 return;
3642 
3643         spin_lock(&swap_avail_lock);
3644         plist_for_each_entry_safe(si, next, &swap_avail_heads[node],
3645                                   avail_lists[node]) {
3646                 if (si->bdev) {
3647                         blkcg_schedule_throttle(bdev_get_queue(si->bdev),
3648                                                 true);
3649                         break;
3650                 }
3651         }
3652         spin_unlock(&swap_avail_lock);
3653 }
3654 #endif
3655 
3656 static int __init swapfile_init(void)
3657 {
3658         int nid;
3659 
3660         swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head),
3661                                          GFP_KERNEL);
3662         if (!swap_avail_heads) {
3663                 pr_emerg("Not enough memory for swap heads, swap is disabled\n");
3664                 return -ENOMEM;
3665         }
3666 
3667         for_each_node(nid)
3668                 plist_head_init(&swap_avail_heads[nid]);
3669 
3670         return 0;
3671 }
3672 subsys_initcall(swapfile_init);
3673 

~ [ 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