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

TOMOYO Linux Cross Reference
Linux/mm/swapfile.c

Version: ~ [ linux-5.8 ] ~ [ linux-5.7.12 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.55 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.136 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.191 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.232 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.232 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ 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-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 // SPDX-License-Identifier: GPL-2.0-only
  2 /*
  3  *  linux/mm/swapfile.c
  4  *
  5  *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
  6  *  Swap reorganised 29.12.95, Stephen Tweedie
  7  */
  8 
  9 #include <linux/mm.h>
 10 #include <linux/sched/mm.h>
 11 #include <linux/sched/task.h>
 12 #include <linux/hugetlb.h>
 13 #include <linux/mman.h>
 14 #include <linux/slab.h>
 15 #include <linux/kernel_stat.h>
 16 #include <linux/swap.h>
 17 #include <linux/vmalloc.h>
 18 #include <linux/pagemap.h>
 19 #include <linux/namei.h>
 20 #include <linux/shmem_fs.h>
 21 #include <linux/blkdev.h>
 22 #include <linux/random.h>
 23 #include <linux/writeback.h>
 24 #include <linux/proc_fs.h>
 25 #include <linux/seq_file.h>
 26 #include <linux/init.h>
 27 #include <linux/ksm.h>
 28 #include <linux/rmap.h>
 29 #include <linux/security.h>
 30 #include <linux/backing-dev.h>
 31 #include <linux/mutex.h>
 32 #include <linux/capability.h>
 33 #include <linux/syscalls.h>
 34 #include <linux/memcontrol.h>
 35 #include <linux/poll.h>
 36 #include <linux/oom.h>
 37 #include <linux/frontswap.h>
 38 #include <linux/swapfile.h>
 39 #include <linux/export.h>
 40 #include <linux/swap_slots.h>
 41 #include <linux/sort.h>
 42 
 43 #include <asm/pgtable.h>
 44 #include <asm/tlbflush.h>
 45 #include <linux/swapops.h>
 46 #include <linux/swap_cgroup.h>
 47 
 48 static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
 49                                  unsigned char);
 50 static void free_swap_count_continuations(struct swap_info_struct *);
 51 static sector_t map_swap_entry(swp_entry_t, struct block_device**);
 52 
 53 DEFINE_SPINLOCK(swap_lock);
 54 static unsigned int nr_swapfiles;
 55 atomic_long_t nr_swap_pages;
 56 /*
 57  * Some modules use swappable objects and may try to swap them out under
 58  * memory pressure (via the shrinker). Before doing so, they may wish to
 59  * check to see if any swap space is available.
 60  */
 61 EXPORT_SYMBOL_GPL(nr_swap_pages);
 62 /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
 63 long total_swap_pages;
 64 static int least_priority = -1;
 65 
 66 static const char Bad_file[] = "Bad swap file entry ";
 67 static const char Unused_file[] = "Unused swap file entry ";
 68 static const char Bad_offset[] = "Bad swap offset entry ";
 69 static const char Unused_offset[] = "Unused swap offset entry ";
 70 
 71 /*
 72  * all active swap_info_structs
 73  * protected with swap_lock, and ordered by priority.
 74  */
 75 PLIST_HEAD(swap_active_head);
 76 
 77 /*
 78  * all available (active, not full) swap_info_structs
 79  * protected with swap_avail_lock, ordered by priority.
 80  * This is used by get_swap_page() instead of swap_active_head
 81  * because swap_active_head includes all swap_info_structs,
 82  * but get_swap_page() doesn't need to look at full ones.
 83  * This uses its own lock instead of swap_lock because when a
 84  * swap_info_struct changes between not-full/full, it needs to
 85  * add/remove itself to/from this list, but the swap_info_struct->lock
 86  * is held and the locking order requires swap_lock to be taken
 87  * before any swap_info_struct->lock.
 88  */
 89 static struct plist_head *swap_avail_heads;
 90 static DEFINE_SPINLOCK(swap_avail_lock);
 91 
 92 struct swap_info_struct *swap_info[MAX_SWAPFILES];
 93 
 94 static DEFINE_MUTEX(swapon_mutex);
 95 
 96 static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
 97 /* Activity counter to indicate that a swapon or swapoff has occurred */
 98 static atomic_t proc_poll_event = ATOMIC_INIT(0);
 99 
100 atomic_t nr_rotate_swap = ATOMIC_INIT(0);
101 
102 static struct swap_info_struct *swap_type_to_swap_info(int type)
103 {
104         if (type >= READ_ONCE(nr_swapfiles))
105                 return NULL;
106 
107         smp_rmb();      /* Pairs with smp_wmb in alloc_swap_info. */
108         return READ_ONCE(swap_info[type]);
109 }
110 
111 static inline unsigned char swap_count(unsigned char ent)
112 {
113         return ent & ~SWAP_HAS_CACHE;   /* may include COUNT_CONTINUED flag */
114 }
115 
116 /* Reclaim the swap entry anyway if possible */
117 #define TTRS_ANYWAY             0x1
118 /*
119  * Reclaim the swap entry if there are no more mappings of the
120  * corresponding page
121  */
122 #define TTRS_UNMAPPED           0x2
123 /* Reclaim the swap entry if swap is getting full*/
124 #define TTRS_FULL               0x4
125 
126 /* returns 1 if swap entry is freed */
127 static int __try_to_reclaim_swap(struct swap_info_struct *si,
128                                  unsigned long offset, unsigned long flags)
129 {
130         swp_entry_t entry = swp_entry(si->type, offset);
131         struct page *page;
132         int ret = 0;
133 
134         page = find_get_page(swap_address_space(entry), offset);
135         if (!page)
136                 return 0;
137         /*
138          * When this function is called from scan_swap_map_slots() and it's
139          * called by vmscan.c at reclaiming pages. So, we hold a lock on a page,
140          * here. We have to use trylock for avoiding deadlock. This is a special
141          * case and you should use try_to_free_swap() with explicit lock_page()
142          * in usual operations.
143          */
144         if (trylock_page(page)) {
145                 if ((flags & TTRS_ANYWAY) ||
146                     ((flags & TTRS_UNMAPPED) && !page_mapped(page)) ||
147                     ((flags & TTRS_FULL) && mem_cgroup_swap_full(page)))
148                         ret = try_to_free_swap(page);
149                 unlock_page(page);
150         }
151         put_page(page);
152         return ret;
153 }
154 
155 static inline struct swap_extent *first_se(struct swap_info_struct *sis)
156 {
157         struct rb_node *rb = rb_first(&sis->swap_extent_root);
158         return rb_entry(rb, struct swap_extent, rb_node);
159 }
160 
161 static inline struct swap_extent *next_se(struct swap_extent *se)
162 {
163         struct rb_node *rb = rb_next(&se->rb_node);
164         return rb ? rb_entry(rb, struct swap_extent, rb_node) : NULL;
165 }
166 
167 /*
168  * swapon tell device that all the old swap contents can be discarded,
169  * to allow the swap device to optimize its wear-levelling.
170  */
171 static int discard_swap(struct swap_info_struct *si)
172 {
173         struct swap_extent *se;
174         sector_t start_block;
175         sector_t nr_blocks;
176         int err = 0;
177 
178         /* Do not discard the swap header page! */
179         se = first_se(si);
180         start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
181         nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
182         if (nr_blocks) {
183                 err = blkdev_issue_discard(si->bdev, start_block,
184                                 nr_blocks, GFP_KERNEL, 0);
185                 if (err)
186                         return err;
187                 cond_resched();
188         }
189 
190         for (se = next_se(se); se; se = next_se(se)) {
191                 start_block = se->start_block << (PAGE_SHIFT - 9);
192                 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
193 
194                 err = blkdev_issue_discard(si->bdev, start_block,
195                                 nr_blocks, GFP_KERNEL, 0);
196                 if (err)
197                         break;
198 
199                 cond_resched();
200         }
201         return err;             /* That will often be -EOPNOTSUPP */
202 }
203 
204 static struct swap_extent *
205 offset_to_swap_extent(struct swap_info_struct *sis, unsigned long offset)
206 {
207         struct swap_extent *se;
208         struct rb_node *rb;
209 
210         rb = sis->swap_extent_root.rb_node;
211         while (rb) {
212                 se = rb_entry(rb, struct swap_extent, rb_node);
213                 if (offset < se->start_page)
214                         rb = rb->rb_left;
215                 else if (offset >= se->start_page + se->nr_pages)
216                         rb = rb->rb_right;
217                 else
218                         return se;
219         }
220         /* It *must* be present */
221         BUG();
222 }
223 
224 /*
225  * swap allocation tell device that a cluster of swap can now be discarded,
226  * to allow the swap device to optimize its wear-levelling.
227  */
228 static void discard_swap_cluster(struct swap_info_struct *si,
229                                  pgoff_t start_page, pgoff_t nr_pages)
230 {
231         struct swap_extent *se = offset_to_swap_extent(si, start_page);
232 
233         while (nr_pages) {
234                 pgoff_t offset = start_page - se->start_page;
235                 sector_t start_block = se->start_block + offset;
236                 sector_t nr_blocks = se->nr_pages - offset;
237 
238                 if (nr_blocks > nr_pages)
239                         nr_blocks = nr_pages;
240                 start_page += nr_blocks;
241                 nr_pages -= nr_blocks;
242 
243                 start_block <<= PAGE_SHIFT - 9;
244                 nr_blocks <<= PAGE_SHIFT - 9;
245                 if (blkdev_issue_discard(si->bdev, start_block,
246                                         nr_blocks, GFP_NOIO, 0))
247                         break;
248 
249                 se = next_se(se);
250         }
251 }
252 
253 #ifdef CONFIG_THP_SWAP
254 #define SWAPFILE_CLUSTER        HPAGE_PMD_NR
255 
256 #define swap_entry_size(size)   (size)
257 #else
258 #define SWAPFILE_CLUSTER        256
259 
260 /*
261  * Define swap_entry_size() as constant to let compiler to optimize
262  * out some code if !CONFIG_THP_SWAP
263  */
264 #define swap_entry_size(size)   1
265 #endif
266 #define LATENCY_LIMIT           256
267 
268 static inline void cluster_set_flag(struct swap_cluster_info *info,
269         unsigned int flag)
270 {
271         info->flags = flag;
272 }
273 
274 static inline unsigned int cluster_count(struct swap_cluster_info *info)
275 {
276         return info->data;
277 }
278 
279 static inline void cluster_set_count(struct swap_cluster_info *info,
280                                      unsigned int c)
281 {
282         info->data = c;
283 }
284 
285 static inline void cluster_set_count_flag(struct swap_cluster_info *info,
286                                          unsigned int c, unsigned int f)
287 {
288         info->flags = f;
289         info->data = c;
290 }
291 
292 static inline unsigned int cluster_next(struct swap_cluster_info *info)
293 {
294         return info->data;
295 }
296 
297 static inline void cluster_set_next(struct swap_cluster_info *info,
298                                     unsigned int n)
299 {
300         info->data = n;
301 }
302 
303 static inline void cluster_set_next_flag(struct swap_cluster_info *info,
304                                          unsigned int n, unsigned int f)
305 {
306         info->flags = f;
307         info->data = n;
308 }
309 
310 static inline bool cluster_is_free(struct swap_cluster_info *info)
311 {
312         return info->flags & CLUSTER_FLAG_FREE;
313 }
314 
315 static inline bool cluster_is_null(struct swap_cluster_info *info)
316 {
317         return info->flags & CLUSTER_FLAG_NEXT_NULL;
318 }
319 
320 static inline void cluster_set_null(struct swap_cluster_info *info)
321 {
322         info->flags = CLUSTER_FLAG_NEXT_NULL;
323         info->data = 0;
324 }
325 
326 static inline bool cluster_is_huge(struct swap_cluster_info *info)
327 {
328         if (IS_ENABLED(CONFIG_THP_SWAP))
329                 return info->flags & CLUSTER_FLAG_HUGE;
330         return false;
331 }
332 
333 static inline void cluster_clear_huge(struct swap_cluster_info *info)
334 {
335         info->flags &= ~CLUSTER_FLAG_HUGE;
336 }
337 
338 static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si,
339                                                      unsigned long offset)
340 {
341         struct swap_cluster_info *ci;
342 
343         ci = si->cluster_info;
344         if (ci) {
345                 ci += offset / SWAPFILE_CLUSTER;
346                 spin_lock(&ci->lock);
347         }
348         return ci;
349 }
350 
351 static inline void unlock_cluster(struct swap_cluster_info *ci)
352 {
353         if (ci)
354                 spin_unlock(&ci->lock);
355 }
356 
357 /*
358  * Determine the locking method in use for this device.  Return
359  * swap_cluster_info if SSD-style cluster-based locking is in place.
360  */
361 static inline struct swap_cluster_info *lock_cluster_or_swap_info(
362                 struct swap_info_struct *si, unsigned long offset)
363 {
364         struct swap_cluster_info *ci;
365 
366         /* Try to use fine-grained SSD-style locking if available: */
367         ci = lock_cluster(si, offset);
368         /* Otherwise, fall back to traditional, coarse locking: */
369         if (!ci)
370                 spin_lock(&si->lock);
371 
372         return ci;
373 }
374 
375 static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si,
376                                                struct swap_cluster_info *ci)
377 {
378         if (ci)
379                 unlock_cluster(ci);
380         else
381                 spin_unlock(&si->lock);
382 }
383 
384 static inline bool cluster_list_empty(struct swap_cluster_list *list)
385 {
386         return cluster_is_null(&list->head);
387 }
388 
389 static inline unsigned int cluster_list_first(struct swap_cluster_list *list)
390 {
391         return cluster_next(&list->head);
392 }
393 
394 static void cluster_list_init(struct swap_cluster_list *list)
395 {
396         cluster_set_null(&list->head);
397         cluster_set_null(&list->tail);
398 }
399 
400 static void cluster_list_add_tail(struct swap_cluster_list *list,
401                                   struct swap_cluster_info *ci,
402                                   unsigned int idx)
403 {
404         if (cluster_list_empty(list)) {
405                 cluster_set_next_flag(&list->head, idx, 0);
406                 cluster_set_next_flag(&list->tail, idx, 0);
407         } else {
408                 struct swap_cluster_info *ci_tail;
409                 unsigned int tail = cluster_next(&list->tail);
410 
411                 /*
412                  * Nested cluster lock, but both cluster locks are
413                  * only acquired when we held swap_info_struct->lock
414                  */
415                 ci_tail = ci + tail;
416                 spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING);
417                 cluster_set_next(ci_tail, idx);
418                 spin_unlock(&ci_tail->lock);
419                 cluster_set_next_flag(&list->tail, idx, 0);
420         }
421 }
422 
423 static unsigned int cluster_list_del_first(struct swap_cluster_list *list,
424                                            struct swap_cluster_info *ci)
425 {
426         unsigned int idx;
427 
428         idx = cluster_next(&list->head);
429         if (cluster_next(&list->tail) == idx) {
430                 cluster_set_null(&list->head);
431                 cluster_set_null(&list->tail);
432         } else
433                 cluster_set_next_flag(&list->head,
434                                       cluster_next(&ci[idx]), 0);
435 
436         return idx;
437 }
438 
439 /* Add a cluster to discard list and schedule it to do discard */
440 static void swap_cluster_schedule_discard(struct swap_info_struct *si,
441                 unsigned int idx)
442 {
443         /*
444          * If scan_swap_map() can't find a free cluster, it will check
445          * si->swap_map directly. To make sure the discarding cluster isn't
446          * taken by scan_swap_map(), mark the swap entries bad (occupied). It
447          * will be cleared after discard
448          */
449         memset(si->swap_map + idx * SWAPFILE_CLUSTER,
450                         SWAP_MAP_BAD, SWAPFILE_CLUSTER);
451 
452         cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx);
453 
454         schedule_work(&si->discard_work);
455 }
456 
457 static void __free_cluster(struct swap_info_struct *si, unsigned long idx)
458 {
459         struct swap_cluster_info *ci = si->cluster_info;
460 
461         cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE);
462         cluster_list_add_tail(&si->free_clusters, ci, idx);
463 }
464 
465 /*
466  * Doing discard actually. After a cluster discard is finished, the cluster
467  * will be added to free cluster list. caller should hold si->lock.
468 */
469 static void swap_do_scheduled_discard(struct swap_info_struct *si)
470 {
471         struct swap_cluster_info *info, *ci;
472         unsigned int idx;
473 
474         info = si->cluster_info;
475 
476         while (!cluster_list_empty(&si->discard_clusters)) {
477                 idx = cluster_list_del_first(&si->discard_clusters, info);
478                 spin_unlock(&si->lock);
479 
480                 discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
481                                 SWAPFILE_CLUSTER);
482 
483                 spin_lock(&si->lock);
484                 ci = lock_cluster(si, idx * SWAPFILE_CLUSTER);
485                 __free_cluster(si, idx);
486                 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
487                                 0, SWAPFILE_CLUSTER);
488                 unlock_cluster(ci);
489         }
490 }
491 
492 static void swap_discard_work(struct work_struct *work)
493 {
494         struct swap_info_struct *si;
495 
496         si = container_of(work, struct swap_info_struct, discard_work);
497 
498         spin_lock(&si->lock);
499         swap_do_scheduled_discard(si);
500         spin_unlock(&si->lock);
501 }
502 
503 static void alloc_cluster(struct swap_info_struct *si, unsigned long idx)
504 {
505         struct swap_cluster_info *ci = si->cluster_info;
506 
507         VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx);
508         cluster_list_del_first(&si->free_clusters, ci);
509         cluster_set_count_flag(ci + idx, 0, 0);
510 }
511 
512 static void free_cluster(struct swap_info_struct *si, unsigned long idx)
513 {
514         struct swap_cluster_info *ci = si->cluster_info + idx;
515 
516         VM_BUG_ON(cluster_count(ci) != 0);
517         /*
518          * If the swap is discardable, prepare discard the cluster
519          * instead of free it immediately. The cluster will be freed
520          * after discard.
521          */
522         if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
523             (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
524                 swap_cluster_schedule_discard(si, idx);
525                 return;
526         }
527 
528         __free_cluster(si, idx);
529 }
530 
531 /*
532  * The cluster corresponding to page_nr will be used. The cluster will be
533  * removed from free cluster list and its usage counter will be increased.
534  */
535 static void inc_cluster_info_page(struct swap_info_struct *p,
536         struct swap_cluster_info *cluster_info, unsigned long page_nr)
537 {
538         unsigned long idx = page_nr / SWAPFILE_CLUSTER;
539 
540         if (!cluster_info)
541                 return;
542         if (cluster_is_free(&cluster_info[idx]))
543                 alloc_cluster(p, idx);
544 
545         VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
546         cluster_set_count(&cluster_info[idx],
547                 cluster_count(&cluster_info[idx]) + 1);
548 }
549 
550 /*
551  * The cluster corresponding to page_nr decreases one usage. If the usage
552  * counter becomes 0, which means no page in the cluster is in using, we can
553  * optionally discard the cluster and add it to free cluster list.
554  */
555 static void dec_cluster_info_page(struct swap_info_struct *p,
556         struct swap_cluster_info *cluster_info, unsigned long page_nr)
557 {
558         unsigned long idx = page_nr / SWAPFILE_CLUSTER;
559 
560         if (!cluster_info)
561                 return;
562 
563         VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
564         cluster_set_count(&cluster_info[idx],
565                 cluster_count(&cluster_info[idx]) - 1);
566 
567         if (cluster_count(&cluster_info[idx]) == 0)
568                 free_cluster(p, idx);
569 }
570 
571 /*
572  * It's possible scan_swap_map() uses a free cluster in the middle of free
573  * cluster list. Avoiding such abuse to avoid list corruption.
574  */
575 static bool
576 scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
577         unsigned long offset)
578 {
579         struct percpu_cluster *percpu_cluster;
580         bool conflict;
581 
582         offset /= SWAPFILE_CLUSTER;
583         conflict = !cluster_list_empty(&si->free_clusters) &&
584                 offset != cluster_list_first(&si->free_clusters) &&
585                 cluster_is_free(&si->cluster_info[offset]);
586 
587         if (!conflict)
588                 return false;
589 
590         percpu_cluster = this_cpu_ptr(si->percpu_cluster);
591         cluster_set_null(&percpu_cluster->index);
592         return true;
593 }
594 
595 /*
596  * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
597  * might involve allocating a new cluster for current CPU too.
598  */
599 static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
600         unsigned long *offset, unsigned long *scan_base)
601 {
602         struct percpu_cluster *cluster;
603         struct swap_cluster_info *ci;
604         bool found_free;
605         unsigned long tmp, max;
606 
607 new_cluster:
608         cluster = this_cpu_ptr(si->percpu_cluster);
609         if (cluster_is_null(&cluster->index)) {
610                 if (!cluster_list_empty(&si->free_clusters)) {
611                         cluster->index = si->free_clusters.head;
612                         cluster->next = cluster_next(&cluster->index) *
613                                         SWAPFILE_CLUSTER;
614                 } else if (!cluster_list_empty(&si->discard_clusters)) {
615                         /*
616                          * we don't have free cluster but have some clusters in
617                          * discarding, do discard now and reclaim them
618                          */
619                         swap_do_scheduled_discard(si);
620                         *scan_base = *offset = si->cluster_next;
621                         goto new_cluster;
622                 } else
623                         return false;
624         }
625 
626         found_free = false;
627 
628         /*
629          * Other CPUs can use our cluster if they can't find a free cluster,
630          * check if there is still free entry in the cluster
631          */
632         tmp = cluster->next;
633         max = min_t(unsigned long, si->max,
634                     (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER);
635         if (tmp >= max) {
636                 cluster_set_null(&cluster->index);
637                 goto new_cluster;
638         }
639         ci = lock_cluster(si, tmp);
640         while (tmp < max) {
641                 if (!si->swap_map[tmp]) {
642                         found_free = true;
643                         break;
644                 }
645                 tmp++;
646         }
647         unlock_cluster(ci);
648         if (!found_free) {
649                 cluster_set_null(&cluster->index);
650                 goto new_cluster;
651         }
652         cluster->next = tmp + 1;
653         *offset = tmp;
654         *scan_base = tmp;
655         return found_free;
656 }
657 
658 static void __del_from_avail_list(struct swap_info_struct *p)
659 {
660         int nid;
661 
662         for_each_node(nid)
663                 plist_del(&p->avail_lists[nid], &swap_avail_heads[nid]);
664 }
665 
666 static void del_from_avail_list(struct swap_info_struct *p)
667 {
668         spin_lock(&swap_avail_lock);
669         __del_from_avail_list(p);
670         spin_unlock(&swap_avail_lock);
671 }
672 
673 static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset,
674                              unsigned int nr_entries)
675 {
676         unsigned int end = offset + nr_entries - 1;
677 
678         if (offset == si->lowest_bit)
679                 si->lowest_bit += nr_entries;
680         if (end == si->highest_bit)
681                 si->highest_bit -= nr_entries;
682         si->inuse_pages += nr_entries;
683         if (si->inuse_pages == si->pages) {
684                 si->lowest_bit = si->max;
685                 si->highest_bit = 0;
686                 del_from_avail_list(si);
687         }
688 }
689 
690 static void add_to_avail_list(struct swap_info_struct *p)
691 {
692         int nid;
693 
694         spin_lock(&swap_avail_lock);
695         for_each_node(nid) {
696                 WARN_ON(!plist_node_empty(&p->avail_lists[nid]));
697                 plist_add(&p->avail_lists[nid], &swap_avail_heads[nid]);
698         }
699         spin_unlock(&swap_avail_lock);
700 }
701 
702 static void swap_range_free(struct swap_info_struct *si, unsigned long offset,
703                             unsigned int nr_entries)
704 {
705         unsigned long end = offset + nr_entries - 1;
706         void (*swap_slot_free_notify)(struct block_device *, unsigned long);
707 
708         if (offset < si->lowest_bit)
709                 si->lowest_bit = offset;
710         if (end > si->highest_bit) {
711                 bool was_full = !si->highest_bit;
712 
713                 si->highest_bit = end;
714                 if (was_full && (si->flags & SWP_WRITEOK))
715                         add_to_avail_list(si);
716         }
717         atomic_long_add(nr_entries, &nr_swap_pages);
718         si->inuse_pages -= nr_entries;
719         if (si->flags & SWP_BLKDEV)
720                 swap_slot_free_notify =
721                         si->bdev->bd_disk->fops->swap_slot_free_notify;
722         else
723                 swap_slot_free_notify = NULL;
724         while (offset <= end) {
725                 frontswap_invalidate_page(si->type, offset);
726                 if (swap_slot_free_notify)
727                         swap_slot_free_notify(si->bdev, offset);
728                 offset++;
729         }
730 }
731 
732 static int scan_swap_map_slots(struct swap_info_struct *si,
733                                unsigned char usage, int nr,
734                                swp_entry_t slots[])
735 {
736         struct swap_cluster_info *ci;
737         unsigned long offset;
738         unsigned long scan_base;
739         unsigned long last_in_cluster = 0;
740         int latency_ration = LATENCY_LIMIT;
741         int n_ret = 0;
742 
743         if (nr > SWAP_BATCH)
744                 nr = SWAP_BATCH;
745 
746         /*
747          * We try to cluster swap pages by allocating them sequentially
748          * in swap.  Once we've allocated SWAPFILE_CLUSTER pages this
749          * way, however, we resort to first-free allocation, starting
750          * a new cluster.  This prevents us from scattering swap pages
751          * all over the entire swap partition, so that we reduce
752          * overall disk seek times between swap pages.  -- sct
753          * But we do now try to find an empty cluster.  -Andrea
754          * And we let swap pages go all over an SSD partition.  Hugh
755          */
756 
757         si->flags += SWP_SCANNING;
758         scan_base = offset = si->cluster_next;
759 
760         /* SSD algorithm */
761         if (si->cluster_info) {
762                 if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
763                         goto checks;
764                 else
765                         goto scan;
766         }
767 
768         if (unlikely(!si->cluster_nr--)) {
769                 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
770                         si->cluster_nr = SWAPFILE_CLUSTER - 1;
771                         goto checks;
772                 }
773 
774                 spin_unlock(&si->lock);
775 
776                 /*
777                  * If seek is expensive, start searching for new cluster from
778                  * start of partition, to minimize the span of allocated swap.
779                  * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
780                  * case, just handled by scan_swap_map_try_ssd_cluster() above.
781                  */
782                 scan_base = offset = si->lowest_bit;
783                 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
784 
785                 /* Locate the first empty (unaligned) cluster */
786                 for (; last_in_cluster <= si->highest_bit; offset++) {
787                         if (si->swap_map[offset])
788                                 last_in_cluster = offset + SWAPFILE_CLUSTER;
789                         else if (offset == last_in_cluster) {
790                                 spin_lock(&si->lock);
791                                 offset -= SWAPFILE_CLUSTER - 1;
792                                 si->cluster_next = offset;
793                                 si->cluster_nr = SWAPFILE_CLUSTER - 1;
794                                 goto checks;
795                         }
796                         if (unlikely(--latency_ration < 0)) {
797                                 cond_resched();
798                                 latency_ration = LATENCY_LIMIT;
799                         }
800                 }
801 
802                 offset = scan_base;
803                 spin_lock(&si->lock);
804                 si->cluster_nr = SWAPFILE_CLUSTER - 1;
805         }
806 
807 checks:
808         if (si->cluster_info) {
809                 while (scan_swap_map_ssd_cluster_conflict(si, offset)) {
810                 /* take a break if we already got some slots */
811                         if (n_ret)
812                                 goto done;
813                         if (!scan_swap_map_try_ssd_cluster(si, &offset,
814                                                         &scan_base))
815                                 goto scan;
816                 }
817         }
818         if (!(si->flags & SWP_WRITEOK))
819                 goto no_page;
820         if (!si->highest_bit)
821                 goto no_page;
822         if (offset > si->highest_bit)
823                 scan_base = offset = si->lowest_bit;
824 
825         ci = lock_cluster(si, offset);
826         /* reuse swap entry of cache-only swap if not busy. */
827         if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
828                 int swap_was_freed;
829                 unlock_cluster(ci);
830                 spin_unlock(&si->lock);
831                 swap_was_freed = __try_to_reclaim_swap(si, offset, TTRS_ANYWAY);
832                 spin_lock(&si->lock);
833                 /* entry was freed successfully, try to use this again */
834                 if (swap_was_freed)
835                         goto checks;
836                 goto scan; /* check next one */
837         }
838 
839         if (si->swap_map[offset]) {
840                 unlock_cluster(ci);
841                 if (!n_ret)
842                         goto scan;
843                 else
844                         goto done;
845         }
846         si->swap_map[offset] = usage;
847         inc_cluster_info_page(si, si->cluster_info, offset);
848         unlock_cluster(ci);
849 
850         swap_range_alloc(si, offset, 1);
851         si->cluster_next = offset + 1;
852         slots[n_ret++] = swp_entry(si->type, offset);
853 
854         /* got enough slots or reach max slots? */
855         if ((n_ret == nr) || (offset >= si->highest_bit))
856                 goto done;
857 
858         /* search for next available slot */
859 
860         /* time to take a break? */
861         if (unlikely(--latency_ration < 0)) {
862                 if (n_ret)
863                         goto done;
864                 spin_unlock(&si->lock);
865                 cond_resched();
866                 spin_lock(&si->lock);
867                 latency_ration = LATENCY_LIMIT;
868         }
869 
870         /* try to get more slots in cluster */
871         if (si->cluster_info) {
872                 if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
873                         goto checks;
874                 else
875                         goto done;
876         }
877         /* non-ssd case */
878         ++offset;
879 
880         /* non-ssd case, still more slots in cluster? */
881         if (si->cluster_nr && !si->swap_map[offset]) {
882                 --si->cluster_nr;
883                 goto checks;
884         }
885 
886 done:
887         si->flags -= SWP_SCANNING;
888         return n_ret;
889 
890 scan:
891         spin_unlock(&si->lock);
892         while (++offset <= si->highest_bit) {
893                 if (!si->swap_map[offset]) {
894                         spin_lock(&si->lock);
895                         goto checks;
896                 }
897                 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
898                         spin_lock(&si->lock);
899                         goto checks;
900                 }
901                 if (unlikely(--latency_ration < 0)) {
902                         cond_resched();
903                         latency_ration = LATENCY_LIMIT;
904                 }
905         }
906         offset = si->lowest_bit;
907         while (offset < scan_base) {
908                 if (!si->swap_map[offset]) {
909                         spin_lock(&si->lock);
910                         goto checks;
911                 }
912                 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
913                         spin_lock(&si->lock);
914                         goto checks;
915                 }
916                 if (unlikely(--latency_ration < 0)) {
917                         cond_resched();
918                         latency_ration = LATENCY_LIMIT;
919                 }
920                 offset++;
921         }
922         spin_lock(&si->lock);
923 
924 no_page:
925         si->flags -= SWP_SCANNING;
926         return n_ret;
927 }
928 
929 static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot)
930 {
931         unsigned long idx;
932         struct swap_cluster_info *ci;
933         unsigned long offset, i;
934         unsigned char *map;
935 
936         /*
937          * Should not even be attempting cluster allocations when huge
938          * page swap is disabled.  Warn and fail the allocation.
939          */
940         if (!IS_ENABLED(CONFIG_THP_SWAP)) {
941                 VM_WARN_ON_ONCE(1);
942                 return 0;
943         }
944 
945         if (cluster_list_empty(&si->free_clusters))
946                 return 0;
947 
948         idx = cluster_list_first(&si->free_clusters);
949         offset = idx * SWAPFILE_CLUSTER;
950         ci = lock_cluster(si, offset);
951         alloc_cluster(si, idx);
952         cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE);
953 
954         map = si->swap_map + offset;
955         for (i = 0; i < SWAPFILE_CLUSTER; i++)
956                 map[i] = SWAP_HAS_CACHE;
957         unlock_cluster(ci);
958         swap_range_alloc(si, offset, SWAPFILE_CLUSTER);
959         *slot = swp_entry(si->type, offset);
960 
961         return 1;
962 }
963 
964 static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx)
965 {
966         unsigned long offset = idx * SWAPFILE_CLUSTER;
967         struct swap_cluster_info *ci;
968 
969         ci = lock_cluster(si, offset);
970         memset(si->swap_map + offset, 0, SWAPFILE_CLUSTER);
971         cluster_set_count_flag(ci, 0, 0);
972         free_cluster(si, idx);
973         unlock_cluster(ci);
974         swap_range_free(si, offset, SWAPFILE_CLUSTER);
975 }
976 
977 static unsigned long scan_swap_map(struct swap_info_struct *si,
978                                    unsigned char usage)
979 {
980         swp_entry_t entry;
981         int n_ret;
982 
983         n_ret = scan_swap_map_slots(si, usage, 1, &entry);
984 
985         if (n_ret)
986                 return swp_offset(entry);
987         else
988                 return 0;
989 
990 }
991 
992 int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size)
993 {
994         unsigned long size = swap_entry_size(entry_size);
995         struct swap_info_struct *si, *next;
996         long avail_pgs;
997         int n_ret = 0;
998         int node;
999 
1000         /* Only single cluster request supported */
1001         WARN_ON_ONCE(n_goal > 1 && size == SWAPFILE_CLUSTER);
1002 
1003         avail_pgs = atomic_long_read(&nr_swap_pages) / size;
1004         if (avail_pgs <= 0)
1005                 goto noswap;
1006 
1007         if (n_goal > SWAP_BATCH)
1008                 n_goal = SWAP_BATCH;
1009 
1010         if (n_goal > avail_pgs)
1011                 n_goal = avail_pgs;
1012 
1013         atomic_long_sub(n_goal * size, &nr_swap_pages);
1014 
1015         spin_lock(&swap_avail_lock);
1016 
1017 start_over:
1018         node = numa_node_id();
1019         plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) {
1020                 /* requeue si to after same-priority siblings */
1021                 plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]);
1022                 spin_unlock(&swap_avail_lock);
1023                 spin_lock(&si->lock);
1024                 if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) {
1025                         spin_lock(&swap_avail_lock);
1026                         if (plist_node_empty(&si->avail_lists[node])) {
1027                                 spin_unlock(&si->lock);
1028                                 goto nextsi;
1029                         }
1030                         WARN(!si->highest_bit,
1031                              "swap_info %d in list but !highest_bit\n",
1032                              si->type);
1033                         WARN(!(si->flags & SWP_WRITEOK),
1034                              "swap_info %d in list but !SWP_WRITEOK\n",
1035                              si->type);
1036                         __del_from_avail_list(si);
1037                         spin_unlock(&si->lock);
1038                         goto nextsi;
1039                 }
1040                 if (size == SWAPFILE_CLUSTER) {
1041                         if (!(si->flags & SWP_FS))
1042                                 n_ret = swap_alloc_cluster(si, swp_entries);
1043                 } else
1044                         n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE,
1045                                                     n_goal, swp_entries);
1046                 spin_unlock(&si->lock);
1047                 if (n_ret || size == SWAPFILE_CLUSTER)
1048                         goto check_out;
1049                 pr_debug("scan_swap_map of si %d failed to find offset\n",
1050                         si->type);
1051 
1052                 spin_lock(&swap_avail_lock);
1053 nextsi:
1054                 /*
1055                  * if we got here, it's likely that si was almost full before,
1056                  * and since scan_swap_map() can drop the si->lock, multiple
1057                  * callers probably all tried to get a page from the same si
1058                  * and it filled up before we could get one; or, the si filled
1059                  * up between us dropping swap_avail_lock and taking si->lock.
1060                  * Since we dropped the swap_avail_lock, the swap_avail_head
1061                  * list may have been modified; so if next is still in the
1062                  * swap_avail_head list then try it, otherwise start over
1063                  * if we have not gotten any slots.
1064                  */
1065                 if (plist_node_empty(&next->avail_lists[node]))
1066                         goto start_over;
1067         }
1068 
1069         spin_unlock(&swap_avail_lock);
1070 
1071 check_out:
1072         if (n_ret < n_goal)
1073                 atomic_long_add((long)(n_goal - n_ret) * size,
1074                                 &nr_swap_pages);
1075 noswap:
1076         return n_ret;
1077 }
1078 
1079 /* The only caller of this function is now suspend routine */
1080 swp_entry_t get_swap_page_of_type(int type)
1081 {
1082         struct swap_info_struct *si = swap_type_to_swap_info(type);
1083         pgoff_t offset;
1084 
1085         if (!si)
1086                 goto fail;
1087 
1088         spin_lock(&si->lock);
1089         if (si->flags & SWP_WRITEOK) {
1090                 atomic_long_dec(&nr_swap_pages);
1091                 /* This is called for allocating swap entry, not cache */
1092                 offset = scan_swap_map(si, 1);
1093                 if (offset) {
1094                         spin_unlock(&si->lock);
1095                         return swp_entry(type, offset);
1096                 }
1097                 atomic_long_inc(&nr_swap_pages);
1098         }
1099         spin_unlock(&si->lock);
1100 fail:
1101         return (swp_entry_t) {0};
1102 }
1103 
1104 static struct swap_info_struct *__swap_info_get(swp_entry_t entry)
1105 {
1106         struct swap_info_struct *p;
1107         unsigned long offset;
1108 
1109         if (!entry.val)
1110                 goto out;
1111         p = swp_swap_info(entry);
1112         if (!p)
1113                 goto bad_nofile;
1114         if (!(p->flags & SWP_USED))
1115                 goto bad_device;
1116         offset = swp_offset(entry);
1117         if (offset >= p->max)
1118                 goto bad_offset;
1119         return p;
1120 
1121 bad_offset:
1122         pr_err("swap_info_get: %s%08lx\n", Bad_offset, entry.val);
1123         goto out;
1124 bad_device:
1125         pr_err("swap_info_get: %s%08lx\n", Unused_file, entry.val);
1126         goto out;
1127 bad_nofile:
1128         pr_err("swap_info_get: %s%08lx\n", Bad_file, entry.val);
1129 out:
1130         return NULL;
1131 }
1132 
1133 static struct swap_info_struct *_swap_info_get(swp_entry_t entry)
1134 {
1135         struct swap_info_struct *p;
1136 
1137         p = __swap_info_get(entry);
1138         if (!p)
1139                 goto out;
1140         if (!p->swap_map[swp_offset(entry)])
1141                 goto bad_free;
1142         return p;
1143 
1144 bad_free:
1145         pr_err("swap_info_get: %s%08lx\n", Unused_offset, entry.val);
1146         goto out;
1147 out:
1148         return NULL;
1149 }
1150 
1151 static struct swap_info_struct *swap_info_get(swp_entry_t entry)
1152 {
1153         struct swap_info_struct *p;
1154 
1155         p = _swap_info_get(entry);
1156         if (p)
1157                 spin_lock(&p->lock);
1158         return p;
1159 }
1160 
1161 static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry,
1162                                         struct swap_info_struct *q)
1163 {
1164         struct swap_info_struct *p;
1165 
1166         p = _swap_info_get(entry);
1167 
1168         if (p != q) {
1169                 if (q != NULL)
1170                         spin_unlock(&q->lock);
1171                 if (p != NULL)
1172                         spin_lock(&p->lock);
1173         }
1174         return p;
1175 }
1176 
1177 static unsigned char __swap_entry_free_locked(struct swap_info_struct *p,
1178                                               unsigned long offset,
1179                                               unsigned char usage)
1180 {
1181         unsigned char count;
1182         unsigned char has_cache;
1183 
1184         count = p->swap_map[offset];
1185 
1186         has_cache = count & SWAP_HAS_CACHE;
1187         count &= ~SWAP_HAS_CACHE;
1188 
1189         if (usage == SWAP_HAS_CACHE) {
1190                 VM_BUG_ON(!has_cache);
1191                 has_cache = 0;
1192         } else if (count == SWAP_MAP_SHMEM) {
1193                 /*
1194                  * Or we could insist on shmem.c using a special
1195                  * swap_shmem_free() and free_shmem_swap_and_cache()...
1196                  */
1197                 count = 0;
1198         } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
1199                 if (count == COUNT_CONTINUED) {
1200                         if (swap_count_continued(p, offset, count))
1201                                 count = SWAP_MAP_MAX | COUNT_CONTINUED;
1202                         else
1203                                 count = SWAP_MAP_MAX;
1204                 } else
1205                         count--;
1206         }
1207 
1208         usage = count | has_cache;
1209         p->swap_map[offset] = usage ? : SWAP_HAS_CACHE;
1210 
1211         return usage;
1212 }
1213 
1214 /*
1215  * Check whether swap entry is valid in the swap device.  If so,
1216  * return pointer to swap_info_struct, and keep the swap entry valid
1217  * via preventing the swap device from being swapoff, until
1218  * put_swap_device() is called.  Otherwise return NULL.
1219  *
1220  * The entirety of the RCU read critical section must come before the
1221  * return from or after the call to synchronize_rcu() in
1222  * enable_swap_info() or swapoff().  So if "si->flags & SWP_VALID" is
1223  * true, the si->map, si->cluster_info, etc. must be valid in the
1224  * critical section.
1225  *
1226  * Notice that swapoff or swapoff+swapon can still happen before the
1227  * rcu_read_lock() in get_swap_device() or after the rcu_read_unlock()
1228  * in put_swap_device() if there isn't any other way to prevent
1229  * swapoff, such as page lock, page table lock, etc.  The caller must
1230  * be prepared for that.  For example, the following situation is
1231  * possible.
1232  *
1233  *   CPU1                               CPU2
1234  *   do_swap_page()
1235  *     ...                              swapoff+swapon
1236  *     __read_swap_cache_async()
1237  *       swapcache_prepare()
1238  *         __swap_duplicate()
1239  *           // check swap_map
1240  *     // verify PTE not changed
1241  *
1242  * In __swap_duplicate(), the swap_map need to be checked before
1243  * changing partly because the specified swap entry may be for another
1244  * swap device which has been swapoff.  And in do_swap_page(), after
1245  * the page is read from the swap device, the PTE is verified not
1246  * changed with the page table locked to check whether the swap device
1247  * has been swapoff or swapoff+swapon.
1248  */
1249 struct swap_info_struct *get_swap_device(swp_entry_t entry)
1250 {
1251         struct swap_info_struct *si;
1252         unsigned long offset;
1253 
1254         if (!entry.val)
1255                 goto out;
1256         si = swp_swap_info(entry);
1257         if (!si)
1258                 goto bad_nofile;
1259 
1260         rcu_read_lock();
1261         if (!(si->flags & SWP_VALID))
1262                 goto unlock_out;
1263         offset = swp_offset(entry);
1264         if (offset >= si->max)
1265                 goto unlock_out;
1266 
1267         return si;
1268 bad_nofile:
1269         pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val);
1270 out:
1271         return NULL;
1272 unlock_out:
1273         rcu_read_unlock();
1274         return NULL;
1275 }
1276 
1277 static unsigned char __swap_entry_free(struct swap_info_struct *p,
1278                                        swp_entry_t entry, unsigned char usage)
1279 {
1280         struct swap_cluster_info *ci;
1281         unsigned long offset = swp_offset(entry);
1282 
1283         ci = lock_cluster_or_swap_info(p, offset);
1284         usage = __swap_entry_free_locked(p, offset, usage);
1285         unlock_cluster_or_swap_info(p, ci);
1286         if (!usage)
1287                 free_swap_slot(entry);
1288 
1289         return usage;
1290 }
1291 
1292 static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry)
1293 {
1294         struct swap_cluster_info *ci;
1295         unsigned long offset = swp_offset(entry);
1296         unsigned char count;
1297 
1298         ci = lock_cluster(p, offset);
1299         count = p->swap_map[offset];
1300         VM_BUG_ON(count != SWAP_HAS_CACHE);
1301         p->swap_map[offset] = 0;
1302         dec_cluster_info_page(p, p->cluster_info, offset);
1303         unlock_cluster(ci);
1304 
1305         mem_cgroup_uncharge_swap(entry, 1);
1306         swap_range_free(p, offset, 1);
1307 }
1308 
1309 /*
1310  * Caller has made sure that the swap device corresponding to entry
1311  * is still around or has not been recycled.
1312  */
1313 void swap_free(swp_entry_t entry)
1314 {
1315         struct swap_info_struct *p;
1316 
1317         p = _swap_info_get(entry);
1318         if (p)
1319                 __swap_entry_free(p, entry, 1);
1320 }
1321 
1322 /*
1323  * Called after dropping swapcache to decrease refcnt to swap entries.
1324  */
1325 void put_swap_page(struct page *page, swp_entry_t entry)
1326 {
1327         unsigned long offset = swp_offset(entry);
1328         unsigned long idx = offset / SWAPFILE_CLUSTER;
1329         struct swap_cluster_info *ci;
1330         struct swap_info_struct *si;
1331         unsigned char *map;
1332         unsigned int i, free_entries = 0;
1333         unsigned char val;
1334         int size = swap_entry_size(hpage_nr_pages(page));
1335 
1336         si = _swap_info_get(entry);
1337         if (!si)
1338                 return;
1339 
1340         ci = lock_cluster_or_swap_info(si, offset);
1341         if (size == SWAPFILE_CLUSTER) {
1342                 VM_BUG_ON(!cluster_is_huge(ci));
1343                 map = si->swap_map + offset;
1344                 for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1345                         val = map[i];
1346                         VM_BUG_ON(!(val & SWAP_HAS_CACHE));
1347                         if (val == SWAP_HAS_CACHE)
1348                                 free_entries++;
1349                 }
1350                 cluster_clear_huge(ci);
1351                 if (free_entries == SWAPFILE_CLUSTER) {
1352                         unlock_cluster_or_swap_info(si, ci);
1353                         spin_lock(&si->lock);
1354                         mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER);
1355                         swap_free_cluster(si, idx);
1356                         spin_unlock(&si->lock);
1357                         return;
1358                 }
1359         }
1360         for (i = 0; i < size; i++, entry.val++) {
1361                 if (!__swap_entry_free_locked(si, offset + i, SWAP_HAS_CACHE)) {
1362                         unlock_cluster_or_swap_info(si, ci);
1363                         free_swap_slot(entry);
1364                         if (i == size - 1)
1365                                 return;
1366                         lock_cluster_or_swap_info(si, offset);
1367                 }
1368         }
1369         unlock_cluster_or_swap_info(si, ci);
1370 }
1371 
1372 #ifdef CONFIG_THP_SWAP
1373 int split_swap_cluster(swp_entry_t entry)
1374 {
1375         struct swap_info_struct *si;
1376         struct swap_cluster_info *ci;
1377         unsigned long offset = swp_offset(entry);
1378 
1379         si = _swap_info_get(entry);
1380         if (!si)
1381                 return -EBUSY;
1382         ci = lock_cluster(si, offset);
1383         cluster_clear_huge(ci);
1384         unlock_cluster(ci);
1385         return 0;
1386 }
1387 #endif
1388 
1389 static int swp_entry_cmp(const void *ent1, const void *ent2)
1390 {
1391         const swp_entry_t *e1 = ent1, *e2 = ent2;
1392 
1393         return (int)swp_type(*e1) - (int)swp_type(*e2);
1394 }
1395 
1396 void swapcache_free_entries(swp_entry_t *entries, int n)
1397 {
1398         struct swap_info_struct *p, *prev;
1399         int i;
1400 
1401         if (n <= 0)
1402                 return;
1403 
1404         prev = NULL;
1405         p = NULL;
1406 
1407         /*
1408          * Sort swap entries by swap device, so each lock is only taken once.
1409          * nr_swapfiles isn't absolutely correct, but the overhead of sort() is
1410          * so low that it isn't necessary to optimize further.
1411          */
1412         if (nr_swapfiles > 1)
1413                 sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL);
1414         for (i = 0; i < n; ++i) {
1415                 p = swap_info_get_cont(entries[i], prev);
1416                 if (p)
1417                         swap_entry_free(p, entries[i]);
1418                 prev = p;
1419         }
1420         if (p)
1421                 spin_unlock(&p->lock);
1422 }
1423 
1424 /*
1425  * How many references to page are currently swapped out?
1426  * This does not give an exact answer when swap count is continued,
1427  * but does include the high COUNT_CONTINUED flag to allow for that.
1428  */
1429 int page_swapcount(struct page *page)
1430 {
1431         int count = 0;
1432         struct swap_info_struct *p;
1433         struct swap_cluster_info *ci;
1434         swp_entry_t entry;
1435         unsigned long offset;
1436 
1437         entry.val = page_private(page);
1438         p = _swap_info_get(entry);
1439         if (p) {
1440                 offset = swp_offset(entry);
1441                 ci = lock_cluster_or_swap_info(p, offset);
1442                 count = swap_count(p->swap_map[offset]);
1443                 unlock_cluster_or_swap_info(p, ci);
1444         }
1445         return count;
1446 }
1447 
1448 int __swap_count(swp_entry_t entry)
1449 {
1450         struct swap_info_struct *si;
1451         pgoff_t offset = swp_offset(entry);
1452         int count = 0;
1453 
1454         si = get_swap_device(entry);
1455         if (si) {
1456                 count = swap_count(si->swap_map[offset]);
1457                 put_swap_device(si);
1458         }
1459         return count;
1460 }
1461 
1462 static int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry)
1463 {
1464         int count = 0;
1465         pgoff_t offset = swp_offset(entry);
1466         struct swap_cluster_info *ci;
1467 
1468         ci = lock_cluster_or_swap_info(si, offset);
1469         count = swap_count(si->swap_map[offset]);
1470         unlock_cluster_or_swap_info(si, ci);
1471         return count;
1472 }
1473 
1474 /*
1475  * How many references to @entry are currently swapped out?
1476  * This does not give an exact answer when swap count is continued,
1477  * but does include the high COUNT_CONTINUED flag to allow for that.
1478  */
1479 int __swp_swapcount(swp_entry_t entry)
1480 {
1481         int count = 0;
1482         struct swap_info_struct *si;
1483 
1484         si = get_swap_device(entry);
1485         if (si) {
1486                 count = swap_swapcount(si, entry);
1487                 put_swap_device(si);
1488         }
1489         return count;
1490 }
1491 
1492 /*
1493  * How many references to @entry are currently swapped out?
1494  * This considers COUNT_CONTINUED so it returns exact answer.
1495  */
1496 int swp_swapcount(swp_entry_t entry)
1497 {
1498         int count, tmp_count, n;
1499         struct swap_info_struct *p;
1500         struct swap_cluster_info *ci;
1501         struct page *page;
1502         pgoff_t offset;
1503         unsigned char *map;
1504 
1505         p = _swap_info_get(entry);
1506         if (!p)
1507                 return 0;
1508 
1509         offset = swp_offset(entry);
1510 
1511         ci = lock_cluster_or_swap_info(p, offset);
1512 
1513         count = swap_count(p->swap_map[offset]);
1514         if (!(count & COUNT_CONTINUED))
1515                 goto out;
1516 
1517         count &= ~COUNT_CONTINUED;
1518         n = SWAP_MAP_MAX + 1;
1519 
1520         page = vmalloc_to_page(p->swap_map + offset);
1521         offset &= ~PAGE_MASK;
1522         VM_BUG_ON(page_private(page) != SWP_CONTINUED);
1523 
1524         do {
1525                 page = list_next_entry(page, lru);
1526                 map = kmap_atomic(page);
1527                 tmp_count = map[offset];
1528                 kunmap_atomic(map);
1529 
1530                 count += (tmp_count & ~COUNT_CONTINUED) * n;
1531                 n *= (SWAP_CONT_MAX + 1);
1532         } while (tmp_count & COUNT_CONTINUED);
1533 out:
1534         unlock_cluster_or_swap_info(p, ci);
1535         return count;
1536 }
1537 
1538 static bool swap_page_trans_huge_swapped(struct swap_info_struct *si,
1539                                          swp_entry_t entry)
1540 {
1541         struct swap_cluster_info *ci;
1542         unsigned char *map = si->swap_map;
1543         unsigned long roffset = swp_offset(entry);
1544         unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER);
1545         int i;
1546         bool ret = false;
1547 
1548         ci = lock_cluster_or_swap_info(si, offset);
1549         if (!ci || !cluster_is_huge(ci)) {
1550                 if (swap_count(map[roffset]))
1551                         ret = true;
1552                 goto unlock_out;
1553         }
1554         for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1555                 if (swap_count(map[offset + i])) {
1556                         ret = true;
1557                         break;
1558                 }
1559         }
1560 unlock_out:
1561         unlock_cluster_or_swap_info(si, ci);
1562         return ret;
1563 }
1564 
1565 static bool page_swapped(struct page *page)
1566 {
1567         swp_entry_t entry;
1568         struct swap_info_struct *si;
1569 
1570         if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page)))
1571                 return page_swapcount(page) != 0;
1572 
1573         page = compound_head(page);
1574         entry.val = page_private(page);
1575         si = _swap_info_get(entry);
1576         if (si)
1577                 return swap_page_trans_huge_swapped(si, entry);
1578         return false;
1579 }
1580 
1581 static int page_trans_huge_map_swapcount(struct page *page, int *total_mapcount,
1582                                          int *total_swapcount)
1583 {
1584         int i, map_swapcount, _total_mapcount, _total_swapcount;
1585         unsigned long offset = 0;
1586         struct swap_info_struct *si;
1587         struct swap_cluster_info *ci = NULL;
1588         unsigned char *map = NULL;
1589         int mapcount, swapcount = 0;
1590 
1591         /* hugetlbfs shouldn't call it */
1592         VM_BUG_ON_PAGE(PageHuge(page), page);
1593 
1594         if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page))) {
1595                 mapcount = page_trans_huge_mapcount(page, total_mapcount);
1596                 if (PageSwapCache(page))
1597                         swapcount = page_swapcount(page);
1598                 if (total_swapcount)
1599                         *total_swapcount = swapcount;
1600                 return mapcount + swapcount;
1601         }
1602 
1603         page = compound_head(page);
1604 
1605         _total_mapcount = _total_swapcount = map_swapcount = 0;
1606         if (PageSwapCache(page)) {
1607                 swp_entry_t entry;
1608 
1609                 entry.val = page_private(page);
1610                 si = _swap_info_get(entry);
1611                 if (si) {
1612                         map = si->swap_map;
1613                         offset = swp_offset(entry);
1614                 }
1615         }
1616         if (map)
1617                 ci = lock_cluster(si, offset);
1618         for (i = 0; i < HPAGE_PMD_NR; i++) {
1619                 mapcount = atomic_read(&page[i]._mapcount) + 1;
1620                 _total_mapcount += mapcount;
1621                 if (map) {
1622                         swapcount = swap_count(map[offset + i]);
1623                         _total_swapcount += swapcount;
1624                 }
1625                 map_swapcount = max(map_swapcount, mapcount + swapcount);
1626         }
1627         unlock_cluster(ci);
1628         if (PageDoubleMap(page)) {
1629                 map_swapcount -= 1;
1630                 _total_mapcount -= HPAGE_PMD_NR;
1631         }
1632         mapcount = compound_mapcount(page);
1633         map_swapcount += mapcount;
1634         _total_mapcount += mapcount;
1635         if (total_mapcount)
1636                 *total_mapcount = _total_mapcount;
1637         if (total_swapcount)
1638                 *total_swapcount = _total_swapcount;
1639 
1640         return map_swapcount;
1641 }
1642 
1643 /*
1644  * We can write to an anon page without COW if there are no other references
1645  * to it.  And as a side-effect, free up its swap: because the old content
1646  * on disk will never be read, and seeking back there to write new content
1647  * later would only waste time away from clustering.
1648  *
1649  * NOTE: total_map_swapcount should not be relied upon by the caller if
1650  * reuse_swap_page() returns false, but it may be always overwritten
1651  * (see the other implementation for CONFIG_SWAP=n).
1652  */
1653 bool reuse_swap_page(struct page *page, int *total_map_swapcount)
1654 {
1655         int count, total_mapcount, total_swapcount;
1656 
1657         VM_BUG_ON_PAGE(!PageLocked(page), page);
1658         if (unlikely(PageKsm(page)))
1659                 return false;
1660         count = page_trans_huge_map_swapcount(page, &total_mapcount,
1661                                               &total_swapcount);
1662         if (total_map_swapcount)
1663                 *total_map_swapcount = total_mapcount + total_swapcount;
1664         if (count == 1 && PageSwapCache(page) &&
1665             (likely(!PageTransCompound(page)) ||
1666              /* The remaining swap count will be freed soon */
1667              total_swapcount == page_swapcount(page))) {
1668                 if (!PageWriteback(page)) {
1669                         page = compound_head(page);
1670                         delete_from_swap_cache(page);
1671                         SetPageDirty(page);
1672                 } else {
1673                         swp_entry_t entry;
1674                         struct swap_info_struct *p;
1675 
1676                         entry.val = page_private(page);
1677                         p = swap_info_get(entry);
1678                         if (p->flags & SWP_STABLE_WRITES) {
1679                                 spin_unlock(&p->lock);
1680                                 return false;
1681                         }
1682                         spin_unlock(&p->lock);
1683                 }
1684         }
1685 
1686         return count <= 1;
1687 }
1688 
1689 /*
1690  * If swap is getting full, or if there are no more mappings of this page,
1691  * then try_to_free_swap is called to free its swap space.
1692  */
1693 int try_to_free_swap(struct page *page)
1694 {
1695         VM_BUG_ON_PAGE(!PageLocked(page), page);
1696 
1697         if (!PageSwapCache(page))
1698                 return 0;
1699         if (PageWriteback(page))
1700                 return 0;
1701         if (page_swapped(page))
1702                 return 0;
1703 
1704         /*
1705          * Once hibernation has begun to create its image of memory,
1706          * there's a danger that one of the calls to try_to_free_swap()
1707          * - most probably a call from __try_to_reclaim_swap() while
1708          * hibernation is allocating its own swap pages for the image,
1709          * but conceivably even a call from memory reclaim - will free
1710          * the swap from a page which has already been recorded in the
1711          * image as a clean swapcache page, and then reuse its swap for
1712          * another page of the image.  On waking from hibernation, the
1713          * original page might be freed under memory pressure, then
1714          * later read back in from swap, now with the wrong data.
1715          *
1716          * Hibernation suspends storage while it is writing the image
1717          * to disk so check that here.
1718          */
1719         if (pm_suspended_storage())
1720                 return 0;
1721 
1722         page = compound_head(page);
1723         delete_from_swap_cache(page);
1724         SetPageDirty(page);
1725         return 1;
1726 }
1727 
1728 /*
1729  * Free the swap entry like above, but also try to
1730  * free the page cache entry if it is the last user.
1731  */
1732 int free_swap_and_cache(swp_entry_t entry)
1733 {
1734         struct swap_info_struct *p;
1735         unsigned char count;
1736 
1737         if (non_swap_entry(entry))
1738                 return 1;
1739 
1740         p = _swap_info_get(entry);
1741         if (p) {
1742                 count = __swap_entry_free(p, entry, 1);
1743                 if (count == SWAP_HAS_CACHE &&
1744                     !swap_page_trans_huge_swapped(p, entry))
1745                         __try_to_reclaim_swap(p, swp_offset(entry),
1746                                               TTRS_UNMAPPED | TTRS_FULL);
1747         }
1748         return p != NULL;
1749 }
1750 
1751 #ifdef CONFIG_HIBERNATION
1752 /*
1753  * Find the swap type that corresponds to given device (if any).
1754  *
1755  * @offset - number of the PAGE_SIZE-sized block of the device, starting
1756  * from 0, in which the swap header is expected to be located.
1757  *
1758  * This is needed for the suspend to disk (aka swsusp).
1759  */
1760 int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
1761 {
1762         struct block_device *bdev = NULL;
1763         int type;
1764 
1765         if (device)
1766                 bdev = bdget(device);
1767 
1768         spin_lock(&swap_lock);
1769         for (type = 0; type < nr_swapfiles; type++) {
1770                 struct swap_info_struct *sis = swap_info[type];
1771 
1772                 if (!(sis->flags & SWP_WRITEOK))
1773                         continue;
1774 
1775                 if (!bdev) {
1776                         if (bdev_p)
1777                                 *bdev_p = bdgrab(sis->bdev);
1778 
1779                         spin_unlock(&swap_lock);
1780                         return type;
1781                 }
1782                 if (bdev == sis->bdev) {
1783                         struct swap_extent *se = first_se(sis);
1784 
1785                         if (se->start_block == offset) {
1786                                 if (bdev_p)
1787                                         *bdev_p = bdgrab(sis->bdev);
1788 
1789                                 spin_unlock(&swap_lock);
1790                                 bdput(bdev);
1791                                 return type;
1792                         }
1793                 }
1794         }
1795         spin_unlock(&swap_lock);
1796         if (bdev)
1797                 bdput(bdev);
1798 
1799         return -ENODEV;
1800 }
1801 
1802 /*
1803  * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1804  * corresponding to given index in swap_info (swap type).
1805  */
1806 sector_t swapdev_block(int type, pgoff_t offset)
1807 {
1808         struct block_device *bdev;
1809         struct swap_info_struct *si = swap_type_to_swap_info(type);
1810 
1811         if (!si || !(si->flags & SWP_WRITEOK))
1812                 return 0;
1813         return map_swap_entry(swp_entry(type, offset), &bdev);
1814 }
1815 
1816 /*
1817  * Return either the total number of swap pages of given type, or the number
1818  * of free pages of that type (depending on @free)
1819  *
1820  * This is needed for software suspend
1821  */
1822 unsigned int count_swap_pages(int type, int free)
1823 {
1824         unsigned int n = 0;
1825 
1826         spin_lock(&swap_lock);
1827         if ((unsigned int)type < nr_swapfiles) {
1828                 struct swap_info_struct *sis = swap_info[type];
1829 
1830                 spin_lock(&sis->lock);
1831                 if (sis->flags & SWP_WRITEOK) {
1832                         n = sis->pages;
1833                         if (free)
1834                                 n -= sis->inuse_pages;
1835                 }
1836                 spin_unlock(&sis->lock);
1837         }
1838         spin_unlock(&swap_lock);
1839         return n;
1840 }
1841 #endif /* CONFIG_HIBERNATION */
1842 
1843 static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte)
1844 {
1845         return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte);
1846 }
1847 
1848 /*
1849  * No need to decide whether this PTE shares the swap entry with others,
1850  * just let do_wp_page work it out if a write is requested later - to
1851  * force COW, vm_page_prot omits write permission from any private vma.
1852  */
1853 static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
1854                 unsigned long addr, swp_entry_t entry, struct page *page)
1855 {
1856         struct page *swapcache;
1857         struct mem_cgroup *memcg;
1858         spinlock_t *ptl;
1859         pte_t *pte;
1860         int ret = 1;
1861 
1862         swapcache = page;
1863         page = ksm_might_need_to_copy(page, vma, addr);
1864         if (unlikely(!page))
1865                 return -ENOMEM;
1866 
1867         if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL,
1868                                 &memcg, false)) {
1869                 ret = -ENOMEM;
1870                 goto out_nolock;
1871         }
1872 
1873         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1874         if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) {
1875                 mem_cgroup_cancel_charge(page, memcg, false);
1876                 ret = 0;
1877                 goto out;
1878         }
1879 
1880         dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1881         inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
1882         get_page(page);
1883         set_pte_at(vma->vm_mm, addr, pte,
1884                    pte_mkold(mk_pte(page, vma->vm_page_prot)));
1885         if (page == swapcache) {
1886                 page_add_anon_rmap(page, vma, addr, false);
1887                 mem_cgroup_commit_charge(page, memcg, true, false);
1888         } else { /* ksm created a completely new copy */
1889                 page_add_new_anon_rmap(page, vma, addr, false);
1890                 mem_cgroup_commit_charge(page, memcg, false, false);
1891                 lru_cache_add_active_or_unevictable(page, vma);
1892         }
1893         swap_free(entry);
1894         /*
1895          * Move the page to the active list so it is not
1896          * immediately swapped out again after swapon.
1897          */
1898         activate_page(page);
1899 out:
1900         pte_unmap_unlock(pte, ptl);
1901 out_nolock:
1902         if (page != swapcache) {
1903                 unlock_page(page);
1904                 put_page(page);
1905         }
1906         return ret;
1907 }
1908 
1909 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1910                         unsigned long addr, unsigned long end,
1911                         unsigned int type, bool frontswap,
1912                         unsigned long *fs_pages_to_unuse)
1913 {
1914         struct page *page;
1915         swp_entry_t entry;
1916         pte_t *pte;
1917         struct swap_info_struct *si;
1918         unsigned long offset;
1919         int ret = 0;
1920         volatile unsigned char *swap_map;
1921 
1922         si = swap_info[type];
1923         pte = pte_offset_map(pmd, addr);
1924         do {
1925                 struct vm_fault vmf;
1926 
1927                 if (!is_swap_pte(*pte))
1928                         continue;
1929 
1930                 entry = pte_to_swp_entry(*pte);
1931                 if (swp_type(entry) != type)
1932                         continue;
1933 
1934                 offset = swp_offset(entry);
1935                 if (frontswap && !frontswap_test(si, offset))
1936                         continue;
1937 
1938                 pte_unmap(pte);
1939                 swap_map = &si->swap_map[offset];
1940                 vmf.vma = vma;
1941                 vmf.address = addr;
1942                 vmf.pmd = pmd;
1943                 page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, &vmf);
1944                 if (!page) {
1945                         if (*swap_map == 0 || *swap_map == SWAP_MAP_BAD)
1946                                 goto try_next;
1947                         return -ENOMEM;
1948                 }
1949 
1950                 lock_page(page);
1951                 wait_on_page_writeback(page);
1952                 ret = unuse_pte(vma, pmd, addr, entry, page);
1953                 if (ret < 0) {
1954                         unlock_page(page);
1955                         put_page(page);
1956                         goto out;
1957                 }
1958 
1959                 try_to_free_swap(page);
1960                 unlock_page(page);
1961                 put_page(page);
1962 
1963                 if (*fs_pages_to_unuse && !--(*fs_pages_to_unuse)) {
1964                         ret = FRONTSWAP_PAGES_UNUSED;
1965                         goto out;
1966                 }
1967 try_next:
1968                 pte = pte_offset_map(pmd, addr);
1969         } while (pte++, addr += PAGE_SIZE, addr != end);
1970         pte_unmap(pte - 1);
1971 
1972         ret = 0;
1973 out:
1974         return ret;
1975 }
1976 
1977 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1978                                 unsigned long addr, unsigned long end,
1979                                 unsigned int type, bool frontswap,
1980                                 unsigned long *fs_pages_to_unuse)
1981 {
1982         pmd_t *pmd;
1983         unsigned long next;
1984         int ret;
1985 
1986         pmd = pmd_offset(pud, addr);
1987         do {
1988                 cond_resched();
1989                 next = pmd_addr_end(addr, end);
1990                 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
1991                         continue;
1992                 ret = unuse_pte_range(vma, pmd, addr, next, type,
1993                                       frontswap, fs_pages_to_unuse);
1994                 if (ret)
1995                         return ret;
1996         } while (pmd++, addr = next, addr != end);
1997         return 0;
1998 }
1999 
2000 static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d,
2001                                 unsigned long addr, unsigned long end,
2002                                 unsigned int type, bool frontswap,
2003                                 unsigned long *fs_pages_to_unuse)
2004 {
2005         pud_t *pud;
2006         unsigned long next;
2007         int ret;
2008 
2009         pud = pud_offset(p4d, addr);
2010         do {
2011                 next = pud_addr_end(addr, end);
2012                 if (pud_none_or_clear_bad(pud))
2013                         continue;
2014                 ret = unuse_pmd_range(vma, pud, addr, next, type,
2015                                       frontswap, fs_pages_to_unuse);
2016                 if (ret)
2017                         return ret;
2018         } while (pud++, addr = next, addr != end);
2019         return 0;
2020 }
2021 
2022 static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd,
2023                                 unsigned long addr, unsigned long end,
2024                                 unsigned int type, bool frontswap,
2025                                 unsigned long *fs_pages_to_unuse)
2026 {
2027         p4d_t *p4d;
2028         unsigned long next;
2029         int ret;
2030 
2031         p4d = p4d_offset(pgd, addr);
2032         do {
2033                 next = p4d_addr_end(addr, end);
2034                 if (p4d_none_or_clear_bad(p4d))
2035                         continue;
2036                 ret = unuse_pud_range(vma, p4d, addr, next, type,
2037                                       frontswap, fs_pages_to_unuse);
2038                 if (ret)
2039                         return ret;
2040         } while (p4d++, addr = next, addr != end);
2041         return 0;
2042 }
2043 
2044 static int unuse_vma(struct vm_area_struct *vma, unsigned int type,
2045                      bool frontswap, unsigned long *fs_pages_to_unuse)
2046 {
2047         pgd_t *pgd;
2048         unsigned long addr, end, next;
2049         int ret;
2050 
2051         addr = vma->vm_start;
2052         end = vma->vm_end;
2053 
2054         pgd = pgd_offset(vma->vm_mm, addr);
2055         do {
2056                 next = pgd_addr_end(addr, end);
2057                 if (pgd_none_or_clear_bad(pgd))
2058                         continue;
2059                 ret = unuse_p4d_range(vma, pgd, addr, next, type,
2060                                       frontswap, fs_pages_to_unuse);
2061                 if (ret)
2062                         return ret;
2063         } while (pgd++, addr = next, addr != end);
2064         return 0;
2065 }
2066 
2067 static int unuse_mm(struct mm_struct *mm, unsigned int type,
2068                     bool frontswap, unsigned long *fs_pages_to_unuse)
2069 {
2070         struct vm_area_struct *vma;
2071         int ret = 0;
2072 
2073         down_read(&mm->mmap_sem);
2074         for (vma = mm->mmap; vma; vma = vma->vm_next) {
2075                 if (vma->anon_vma) {
2076                         ret = unuse_vma(vma, type, frontswap,
2077                                         fs_pages_to_unuse);
2078                         if (ret)
2079                                 break;
2080                 }
2081                 cond_resched();
2082         }
2083         up_read(&mm->mmap_sem);
2084         return ret;
2085 }
2086 
2087 /*
2088  * Scan swap_map (or frontswap_map if frontswap parameter is true)
2089  * from current position to next entry still in use. Return 0
2090  * if there are no inuse entries after prev till end of the map.
2091  */
2092 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
2093                                         unsigned int prev, bool frontswap)
2094 {
2095         unsigned int i;
2096         unsigned char count;
2097 
2098         /*
2099          * No need for swap_lock here: we're just looking
2100          * for whether an entry is in use, not modifying it; false
2101          * hits are okay, and sys_swapoff() has already prevented new
2102          * allocations from this area (while holding swap_lock).
2103          */
2104         for (i = prev + 1; i < si->max; i++) {
2105                 count = READ_ONCE(si->swap_map[i]);
2106                 if (count && swap_count(count) != SWAP_MAP_BAD)
2107                         if (!frontswap || frontswap_test(si, i))
2108                                 break;
2109                 if ((i % LATENCY_LIMIT) == 0)
2110                         cond_resched();
2111         }
2112 
2113         if (i == si->max)
2114                 i = 0;
2115 
2116         return i;
2117 }
2118 
2119 /*
2120  * If the boolean frontswap is true, only unuse pages_to_unuse pages;
2121  * pages_to_unuse==0 means all pages; ignored if frontswap is false
2122  */
2123 int try_to_unuse(unsigned int type, bool frontswap,
2124                  unsigned long pages_to_unuse)
2125 {
2126         struct mm_struct *prev_mm;
2127         struct mm_struct *mm;
2128         struct list_head *p;
2129         int retval = 0;
2130         struct swap_info_struct *si = swap_info[type];
2131         struct page *page;
2132         swp_entry_t entry;
2133         unsigned int i;
2134 
2135         if (!si->inuse_pages)
2136                 return 0;
2137 
2138         if (!frontswap)
2139                 pages_to_unuse = 0;
2140 
2141 retry:
2142         retval = shmem_unuse(type, frontswap, &pages_to_unuse);
2143         if (retval)
2144                 goto out;
2145 
2146         prev_mm = &init_mm;
2147         mmget(prev_mm);
2148 
2149         spin_lock(&mmlist_lock);
2150         p = &init_mm.mmlist;
2151         while (si->inuse_pages &&
2152                !signal_pending(current) &&
2153                (p = p->next) != &init_mm.mmlist) {
2154 
2155                 mm = list_entry(p, struct mm_struct, mmlist);
2156                 if (!mmget_not_zero(mm))
2157                         continue;
2158                 spin_unlock(&mmlist_lock);
2159                 mmput(prev_mm);
2160                 prev_mm = mm;
2161                 retval = unuse_mm(mm, type, frontswap, &pages_to_unuse);
2162 
2163                 if (retval) {
2164                         mmput(prev_mm);
2165                         goto out;
2166                 }
2167 
2168                 /*
2169                  * Make sure that we aren't completely killing
2170                  * interactive performance.
2171                  */
2172                 cond_resched();
2173                 spin_lock(&mmlist_lock);
2174         }
2175         spin_unlock(&mmlist_lock);
2176 
2177         mmput(prev_mm);
2178 
2179         i = 0;
2180         while (si->inuse_pages &&
2181                !signal_pending(current) &&
2182                (i = find_next_to_unuse(si, i, frontswap)) != 0) {
2183 
2184                 entry = swp_entry(type, i);
2185                 page = find_get_page(swap_address_space(entry), i);
2186                 if (!page)
2187                         continue;
2188 
2189                 /*
2190                  * It is conceivable that a racing task removed this page from
2191                  * swap cache just before we acquired the page lock. The page
2192                  * might even be back in swap cache on another swap area. But
2193                  * that is okay, try_to_free_swap() only removes stale pages.
2194                  */
2195                 lock_page(page);
2196                 wait_on_page_writeback(page);
2197                 try_to_free_swap(page);
2198                 unlock_page(page);
2199                 put_page(page);
2200 
2201                 /*
2202                  * For frontswap, we just need to unuse pages_to_unuse, if
2203                  * it was specified. Need not check frontswap again here as
2204                  * we already zeroed out pages_to_unuse if not frontswap.
2205                  */
2206                 if (pages_to_unuse && --pages_to_unuse == 0)
2207                         goto out;
2208         }
2209 
2210         /*
2211          * Lets check again to see if there are still swap entries in the map.
2212          * If yes, we would need to do retry the unuse logic again.
2213          * Under global memory pressure, swap entries can be reinserted back
2214          * into process space after the mmlist loop above passes over them.
2215          *
2216          * Limit the number of retries? No: when mmget_not_zero() above fails,
2217          * that mm is likely to be freeing swap from exit_mmap(), which proceeds
2218          * at its own independent pace; and even shmem_writepage() could have
2219          * been preempted after get_swap_page(), temporarily hiding that swap.
2220          * It's easy and robust (though cpu-intensive) just to keep retrying.
2221          */
2222         if (si->inuse_pages) {
2223                 if (!signal_pending(current))
2224                         goto retry;
2225                 retval = -EINTR;
2226         }
2227 out:
2228         return (retval == FRONTSWAP_PAGES_UNUSED) ? 0 : retval;
2229 }
2230 
2231 /*
2232  * After a successful try_to_unuse, if no swap is now in use, we know
2233  * we can empty the mmlist.  swap_lock must be held on entry and exit.
2234  * Note that mmlist_lock nests inside swap_lock, and an mm must be
2235  * added to the mmlist just after page_duplicate - before would be racy.
2236  */
2237 static void drain_mmlist(void)
2238 {
2239         struct list_head *p, *next;
2240         unsigned int type;
2241 
2242         for (type = 0; type < nr_swapfiles; type++)
2243                 if (swap_info[type]->inuse_pages)
2244                         return;
2245         spin_lock(&mmlist_lock);
2246         list_for_each_safe(p, next, &init_mm.mmlist)
2247                 list_del_init(p);
2248         spin_unlock(&mmlist_lock);
2249 }
2250 
2251 /*
2252  * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
2253  * corresponds to page offset for the specified swap entry.
2254  * Note that the type of this function is sector_t, but it returns page offset
2255  * into the bdev, not sector offset.
2256  */
2257 static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
2258 {
2259         struct swap_info_struct *sis;
2260         struct swap_extent *se;
2261         pgoff_t offset;
2262 
2263         sis = swp_swap_info(entry);
2264         *bdev = sis->bdev;
2265 
2266         offset = swp_offset(entry);
2267         se = offset_to_swap_extent(sis, offset);
2268         return se->start_block + (offset - se->start_page);
2269 }
2270 
2271 /*
2272  * Returns the page offset into bdev for the specified page's swap entry.
2273  */
2274 sector_t map_swap_page(struct page *page, struct block_device **bdev)
2275 {
2276         swp_entry_t entry;
2277         entry.val = page_private(page);
2278         return map_swap_entry(entry, bdev);
2279 }
2280 
2281 /*
2282  * Free all of a swapdev's extent information
2283  */
2284 static void destroy_swap_extents(struct swap_info_struct *sis)
2285 {
2286         while (!RB_EMPTY_ROOT(&sis->swap_extent_root)) {
2287                 struct rb_node *rb = sis->swap_extent_root.rb_node;
2288                 struct swap_extent *se = rb_entry(rb, struct swap_extent, rb_node);
2289 
2290                 rb_erase(rb, &sis->swap_extent_root);
2291                 kfree(se);
2292         }
2293 
2294         if (sis->flags & SWP_ACTIVATED) {
2295                 struct file *swap_file = sis->swap_file;
2296                 struct address_space *mapping = swap_file->f_mapping;
2297 
2298                 sis->flags &= ~SWP_ACTIVATED;
2299                 if (mapping->a_ops->swap_deactivate)
2300                         mapping->a_ops->swap_deactivate(swap_file);
2301         }
2302 }
2303 
2304 /*
2305  * Add a block range (and the corresponding page range) into this swapdev's
2306  * extent tree.
2307  *
2308  * This function rather assumes that it is called in ascending page order.
2309  */
2310 int
2311 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
2312                 unsigned long nr_pages, sector_t start_block)
2313 {
2314         struct rb_node **link = &sis->swap_extent_root.rb_node, *parent = NULL;
2315         struct swap_extent *se;
2316         struct swap_extent *new_se;
2317 
2318         /*
2319          * place the new node at the right most since the
2320          * function is called in ascending page order.
2321          */
2322         while (*link) {
2323                 parent = *link;
2324                 link = &parent->rb_right;
2325         }
2326 
2327         if (parent) {
2328                 se = rb_entry(parent, struct swap_extent, rb_node);
2329                 BUG_ON(se->start_page + se->nr_pages != start_page);
2330                 if (se->start_block + se->nr_pages == start_block) {
2331                         /* Merge it */
2332                         se->nr_pages += nr_pages;
2333                         return 0;
2334                 }
2335         }
2336 
2337         /* No merge, insert a new extent. */
2338         new_se = kmalloc(sizeof(*se), GFP_KERNEL);
2339         if (new_se == NULL)
2340                 return -ENOMEM;
2341         new_se->start_page = start_page;
2342         new_se->nr_pages = nr_pages;
2343         new_se->start_block = start_block;
2344 
2345         rb_link_node(&new_se->rb_node, parent, link);
2346         rb_insert_color(&new_se->rb_node, &sis->swap_extent_root);
2347         return 1;
2348 }
2349 EXPORT_SYMBOL_GPL(add_swap_extent);
2350 
2351 /*
2352  * A `swap extent' is a simple thing which maps a contiguous range of pages
2353  * onto a contiguous range of disk blocks.  An ordered list of swap extents
2354  * is built at swapon time and is then used at swap_writepage/swap_readpage
2355  * time for locating where on disk a page belongs.
2356  *
2357  * If the swapfile is an S_ISBLK block device, a single extent is installed.
2358  * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2359  * swap files identically.
2360  *
2361  * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2362  * extent list operates in PAGE_SIZE disk blocks.  Both S_ISREG and S_ISBLK
2363  * swapfiles are handled *identically* after swapon time.
2364  *
2365  * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2366  * and will parse them into an ordered extent list, in PAGE_SIZE chunks.  If
2367  * some stray blocks are found which do not fall within the PAGE_SIZE alignment
2368  * requirements, they are simply tossed out - we will never use those blocks
2369  * for swapping.
2370  *
2371  * For all swap devices we set S_SWAPFILE across the life of the swapon.  This
2372  * prevents users from writing to the swap device, which will corrupt memory.
2373  *
2374  * The amount of disk space which a single swap extent represents varies.
2375  * Typically it is in the 1-4 megabyte range.  So we can have hundreds of
2376  * extents in the list.  To avoid much list walking, we cache the previous
2377  * search location in `curr_swap_extent', and start new searches from there.
2378  * This is extremely effective.  The average number of iterations in
2379  * map_swap_page() has been measured at about 0.3 per page.  - akpm.
2380  */
2381 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
2382 {
2383         struct file *swap_file = sis->swap_file;
2384         struct address_space *mapping = swap_file->f_mapping;
2385         struct inode *inode = mapping->host;
2386         int ret;
2387 
2388         if (S_ISBLK(inode->i_mode)) {
2389                 ret = add_swap_extent(sis, 0, sis->max, 0);
2390                 *span = sis->pages;
2391                 return ret;
2392         }
2393 
2394         if (mapping->a_ops->swap_activate) {
2395                 ret = mapping->a_ops->swap_activate(sis, swap_file, span);
2396                 if (ret >= 0)
2397                         sis->flags |= SWP_ACTIVATED;
2398                 if (!ret) {
2399                         sis->flags |= SWP_FS;
2400                         ret = add_swap_extent(sis, 0, sis->max, 0);
2401                         *span = sis->pages;
2402                 }
2403                 return ret;
2404         }
2405 
2406         return generic_swapfile_activate(sis, swap_file, span);
2407 }
2408 
2409 static int swap_node(struct swap_info_struct *p)
2410 {
2411         struct block_device *bdev;
2412 
2413         if (p->bdev)
2414                 bdev = p->bdev;
2415         else
2416                 bdev = p->swap_file->f_inode->i_sb->s_bdev;
2417 
2418         return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE;
2419 }
2420 
2421 static void setup_swap_info(struct swap_info_struct *p, int prio,
2422                             unsigned char *swap_map,
2423                             struct swap_cluster_info *cluster_info)
2424 {
2425         int i;
2426 
2427         if (prio >= 0)
2428                 p->prio = prio;
2429         else
2430                 p->prio = --least_priority;
2431         /*
2432          * the plist prio is negated because plist ordering is
2433          * low-to-high, while swap ordering is high-to-low
2434          */
2435         p->list.prio = -p->prio;
2436         for_each_node(i) {
2437                 if (p->prio >= 0)
2438                         p->avail_lists[i].prio = -p->prio;
2439                 else {
2440                         if (swap_node(p) == i)
2441                                 p->avail_lists[i].prio = 1;
2442                         else
2443                                 p->avail_lists[i].prio = -p->prio;
2444                 }
2445         }
2446         p->swap_map = swap_map;
2447         p->cluster_info = cluster_info;
2448 }
2449 
2450 static void _enable_swap_info(struct swap_info_struct *p)
2451 {
2452         p->flags |= SWP_WRITEOK | SWP_VALID;
2453         atomic_long_add(p->pages, &nr_swap_pages);
2454         total_swap_pages += p->pages;
2455 
2456         assert_spin_locked(&swap_lock);
2457         /*
2458          * both lists are plists, and thus priority ordered.
2459          * swap_active_head needs to be priority ordered for swapoff(),
2460          * which on removal of any swap_info_struct with an auto-assigned
2461          * (i.e. negative) priority increments the auto-assigned priority
2462          * of any lower-priority swap_info_structs.
2463          * swap_avail_head needs to be priority ordered for get_swap_page(),
2464          * which allocates swap pages from the highest available priority
2465          * swap_info_struct.
2466          */
2467         plist_add(&p->list, &swap_active_head);
2468         add_to_avail_list(p);
2469 }
2470 
2471 static void enable_swap_info(struct swap_info_struct *p, int prio,
2472                                 unsigned char *swap_map,
2473                                 struct swap_cluster_info *cluster_info,
2474                                 unsigned long *frontswap_map)
2475 {
2476         frontswap_init(p->type, frontswap_map);
2477         spin_lock(&swap_lock);
2478         spin_lock(&p->lock);
2479         setup_swap_info(p, prio, swap_map, cluster_info);
2480         spin_unlock(&p->lock);
2481         spin_unlock(&swap_lock);
2482         /*
2483          * Guarantee swap_map, cluster_info, etc. fields are valid
2484          * between get/put_swap_device() if SWP_VALID bit is set
2485          */
2486         synchronize_rcu();
2487         spin_lock(&swap_lock);
2488         spin_lock(&p->lock);
2489         _enable_swap_info(p);
2490         spin_unlock(&p->lock);
2491         spin_unlock(&swap_lock);
2492 }
2493 
2494 static void reinsert_swap_info(struct swap_info_struct *p)
2495 {
2496         spin_lock(&swap_lock);
2497         spin_lock(&p->lock);
2498         setup_swap_info(p, p->prio, p->swap_map, p->cluster_info);
2499         _enable_swap_info(p);
2500         spin_unlock(&p->lock);
2501         spin_unlock(&swap_lock);
2502 }
2503 
2504 bool has_usable_swap(void)
2505 {
2506         bool ret = true;
2507 
2508         spin_lock(&swap_lock);
2509         if (plist_head_empty(&swap_active_head))
2510                 ret = false;
2511         spin_unlock(&swap_lock);
2512         return ret;
2513 }
2514 
2515 SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
2516 {
2517         struct swap_info_struct *p = NULL;
2518         unsigned char *swap_map;
2519         struct swap_cluster_info *cluster_info;
2520         unsigned long *frontswap_map;
2521         struct file *swap_file, *victim;
2522         struct address_space *mapping;
2523         struct inode *inode;
2524         struct filename *pathname;
2525         int err, found = 0;
2526         unsigned int old_block_size;
2527 
2528         if (!capable(CAP_SYS_ADMIN))
2529                 return -EPERM;
2530 
2531         BUG_ON(!current->mm);
2532 
2533         pathname = getname(specialfile);
2534         if (IS_ERR(pathname))
2535                 return PTR_ERR(pathname);
2536 
2537         victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
2538         err = PTR_ERR(victim);
2539         if (IS_ERR(victim))
2540                 goto out;
2541 
2542         mapping = victim->f_mapping;
2543         spin_lock(&swap_lock);
2544         plist_for_each_entry(p, &swap_active_head, list) {
2545                 if (p->flags & SWP_WRITEOK) {
2546                         if (p->swap_file->f_mapping == mapping) {
2547                                 found = 1;
2548                                 break;
2549                         }
2550                 }
2551         }
2552         if (!found) {
2553                 err = -EINVAL;
2554                 spin_unlock(&swap_lock);
2555                 goto out_dput;
2556         }
2557         if (!security_vm_enough_memory_mm(current->mm, p->pages))
2558                 vm_unacct_memory(p->pages);
2559         else {
2560                 err = -ENOMEM;
2561                 spin_unlock(&swap_lock);
2562                 goto out_dput;
2563         }
2564         del_from_avail_list(p);
2565         spin_lock(&p->lock);
2566         if (p->prio < 0) {
2567                 struct swap_info_struct *si = p;
2568                 int nid;
2569 
2570                 plist_for_each_entry_continue(si, &swap_active_head, list) {
2571                         si->prio++;
2572                         si->list.prio--;
2573                         for_each_node(nid) {
2574                                 if (si->avail_lists[nid].prio != 1)
2575                                         si->avail_lists[nid].prio--;
2576                         }
2577                 }
2578                 least_priority++;
2579         }
2580         plist_del(&p->list, &swap_active_head);
2581         atomic_long_sub(p->pages, &nr_swap_pages);
2582         total_swap_pages -= p->pages;
2583         p->flags &= ~SWP_WRITEOK;
2584         spin_unlock(&p->lock);
2585         spin_unlock(&swap_lock);
2586 
2587         disable_swap_slots_cache_lock();
2588 
2589         set_current_oom_origin();
2590         err = try_to_unuse(p->type, false, 0); /* force unuse all pages */
2591         clear_current_oom_origin();
2592 
2593         if (err) {
2594                 /* re-insert swap space back into swap_list */
2595                 reinsert_swap_info(p);
2596                 reenable_swap_slots_cache_unlock();
2597                 goto out_dput;
2598         }
2599 
2600         reenable_swap_slots_cache_unlock();
2601 
2602         spin_lock(&swap_lock);
2603         spin_lock(&p->lock);
2604         p->flags &= ~SWP_VALID;         /* mark swap device as invalid */
2605         spin_unlock(&p->lock);
2606         spin_unlock(&swap_lock);
2607         /*
2608          * wait for swap operations protected by get/put_swap_device()
2609          * to complete
2610          */
2611         synchronize_rcu();
2612 
2613         flush_work(&p->discard_work);
2614 
2615         destroy_swap_extents(p);
2616         if (p->flags & SWP_CONTINUED)
2617                 free_swap_count_continuations(p);
2618 
2619         if (!p->bdev || !blk_queue_nonrot(bdev_get_queue(p->bdev)))
2620                 atomic_dec(&nr_rotate_swap);
2621 
2622         mutex_lock(&swapon_mutex);
2623         spin_lock(&swap_lock);
2624         spin_lock(&p->lock);
2625         drain_mmlist();
2626 
2627         /* wait for anyone still in scan_swap_map */
2628         p->highest_bit = 0;             /* cuts scans short */
2629         while (p->flags >= SWP_SCANNING) {
2630                 spin_unlock(&p->lock);
2631                 spin_unlock(&swap_lock);
2632                 schedule_timeout_uninterruptible(1);
2633                 spin_lock(&swap_lock);
2634                 spin_lock(&p->lock);
2635         }
2636 
2637         swap_file = p->swap_file;
2638         old_block_size = p->old_block_size;
2639         p->swap_file = NULL;
2640         p->max = 0;
2641         swap_map = p->swap_map;
2642         p->swap_map = NULL;
2643         cluster_info = p->cluster_info;
2644         p->cluster_info = NULL;
2645         frontswap_map = frontswap_map_get(p);
2646         spin_unlock(&p->lock);
2647         spin_unlock(&swap_lock);
2648         frontswap_invalidate_area(p->type);
2649         frontswap_map_set(p, NULL);
2650         mutex_unlock(&swapon_mutex);
2651         free_percpu(p->percpu_cluster);
2652         p->percpu_cluster = NULL;
2653         vfree(swap_map);
2654         kvfree(cluster_info);
2655         kvfree(frontswap_map);
2656         /* Destroy swap account information */
2657         swap_cgroup_swapoff(p->type);
2658         exit_swap_address_space(p->type);
2659 
2660         inode = mapping->host;
2661         if (S_ISBLK(inode->i_mode)) {
2662                 struct block_device *bdev = I_BDEV(inode);
2663 
2664                 set_blocksize(bdev, old_block_size);
2665                 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2666         }
2667 
2668         inode_lock(inode);
2669         inode->i_flags &= ~S_SWAPFILE;
2670         inode_unlock(inode);
2671         filp_close(swap_file, NULL);
2672 
2673         /*
2674          * Clear the SWP_USED flag after all resources are freed so that swapon
2675          * can reuse this swap_info in alloc_swap_info() safely.  It is ok to
2676          * not hold p->lock after we cleared its SWP_WRITEOK.
2677          */
2678         spin_lock(&swap_lock);
2679         p->flags = 0;
2680         spin_unlock(&swap_lock);
2681 
2682         err = 0;
2683         atomic_inc(&proc_poll_event);
2684         wake_up_interruptible(&proc_poll_wait);
2685 
2686 out_dput:
2687         filp_close(victim, NULL);
2688 out:
2689         putname(pathname);
2690         return err;
2691 }
2692 
2693 #ifdef CONFIG_PROC_FS
2694 static __poll_t swaps_poll(struct file *file, poll_table *wait)
2695 {
2696         struct seq_file *seq = file->private_data;
2697 
2698         poll_wait(file, &proc_poll_wait, wait);
2699 
2700         if (seq->poll_event != atomic_read(&proc_poll_event)) {
2701                 seq->poll_event = atomic_read(&proc_poll_event);
2702                 return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
2703         }
2704 
2705         return EPOLLIN | EPOLLRDNORM;
2706 }
2707 
2708 /* iterator */
2709 static void *swap_start(struct seq_file *swap, loff_t *pos)
2710 {
2711         struct swap_info_struct *si;
2712         int type;
2713         loff_t l = *pos;
2714 
2715         mutex_lock(&swapon_mutex);
2716 
2717         if (!l)
2718                 return SEQ_START_TOKEN;
2719 
2720         for (type = 0; (si = swap_type_to_swap_info(type)); type++) {
2721                 if (!(si->flags & SWP_USED) || !si->swap_map)
2722                         continue;
2723                 if (!--l)
2724                         return si;
2725         }
2726 
2727         return NULL;
2728 }
2729 
2730 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2731 {
2732         struct swap_info_struct *si = v;
2733         int type;
2734 
2735         if (v == SEQ_START_TOKEN)
2736                 type = 0;
2737         else
2738                 type = si->type + 1;
2739 
2740         for (; (si = swap_type_to_swap_info(type)); type++) {
2741                 if (!(si->flags & SWP_USED) || !si->swap_map)
2742                         continue;
2743                 ++*pos;
2744                 return si;
2745         }
2746 
2747         return NULL;
2748 }
2749 
2750 static void swap_stop(struct seq_file *swap, void *v)
2751 {
2752         mutex_unlock(&swapon_mutex);
2753 }
2754 
2755 static int swap_show(struct seq_file *swap, void *v)
2756 {
2757         struct swap_info_struct *si = v;
2758         struct file *file;
2759         int len;
2760 
2761         if (si == SEQ_START_TOKEN) {
2762                 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2763                 return 0;
2764         }
2765 
2766         file = si->swap_file;
2767         len = seq_file_path(swap, file, " \t\n\\");
2768         seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
2769                         len < 40 ? 40 - len : 1, " ",
2770                         S_ISBLK(file_inode(file)->i_mode) ?
2771                                 "partition" : "file\t",
2772                         si->pages << (PAGE_SHIFT - 10),
2773                         si->inuse_pages << (PAGE_SHIFT - 10),
2774                         si->prio);
2775         return 0;
2776 }
2777 
2778 static const struct seq_operations swaps_op = {
2779         .start =        swap_start,
2780         .next =         swap_next,
2781         .stop =         swap_stop,
2782         .show =         swap_show
2783 };
2784 
2785 static int swaps_open(struct inode *inode, struct file *file)
2786 {
2787         struct seq_file *seq;
2788         int ret;
2789 
2790         ret = seq_open(file, &swaps_op);
2791         if (ret)
2792                 return ret;
2793 
2794         seq = file->private_data;
2795         seq->poll_event = atomic_read(&proc_poll_event);
2796         return 0;
2797 }
2798 
2799 static const struct file_operations proc_swaps_operations = {
2800         .open           = swaps_open,
2801         .read           = seq_read,
2802         .llseek         = seq_lseek,
2803         .release        = seq_release,
2804         .poll           = swaps_poll,
2805 };
2806 
2807 static int __init procswaps_init(void)
2808 {
2809         proc_create("swaps", 0, NULL, &proc_swaps_operations);
2810         return 0;
2811 }
2812 __initcall(procswaps_init);
2813 #endif /* CONFIG_PROC_FS */
2814 
2815 #ifdef MAX_SWAPFILES_CHECK
2816 static int __init max_swapfiles_check(void)
2817 {
2818         MAX_SWAPFILES_CHECK();
2819         return 0;
2820 }
2821 late_initcall(max_swapfiles_check);
2822 #endif
2823 
2824 static struct swap_info_struct *alloc_swap_info(void)
2825 {
2826         struct swap_info_struct *p;
2827         unsigned int type;
2828         int i;
2829 
2830         p = kvzalloc(struct_size(p, avail_lists, nr_node_ids), GFP_KERNEL);
2831         if (!p)
2832                 return ERR_PTR(-ENOMEM);
2833 
2834         spin_lock(&swap_lock);
2835         for (type = 0; type < nr_swapfiles; type++) {
2836                 if (!(swap_info[type]->flags & SWP_USED))
2837                         break;
2838         }
2839         if (type >= MAX_SWAPFILES) {
2840                 spin_unlock(&swap_lock);
2841                 kvfree(p);
2842                 return ERR_PTR(-EPERM);
2843         }
2844         if (type >= nr_swapfiles) {
2845                 p->type = type;
2846                 WRITE_ONCE(swap_info[type], p);
2847                 /*
2848                  * Write swap_info[type] before nr_swapfiles, in case a
2849                  * racing procfs swap_start() or swap_next() is reading them.
2850                  * (We never shrink nr_swapfiles, we never free this entry.)
2851                  */
2852                 smp_wmb();
2853                 WRITE_ONCE(nr_swapfiles, nr_swapfiles + 1);
2854         } else {
2855                 kvfree(p);
2856                 p = swap_info[type];
2857                 /*
2858                  * Do not memset this entry: a racing procfs swap_next()
2859                  * would be relying on p->type to remain valid.
2860                  */
2861         }
2862         p->swap_extent_root = RB_ROOT;
2863         plist_node_init(&p->list, 0);
2864         for_each_node(i)
2865                 plist_node_init(&p->avail_lists[i], 0);
2866         p->flags = SWP_USED;
2867         spin_unlock(&swap_lock);
2868         spin_lock_init(&p->lock);
2869         spin_lock_init(&p->cont_lock);
2870 
2871         return p;
2872 }
2873 
2874 static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2875 {
2876         int error;
2877 
2878         if (S_ISBLK(inode->i_mode)) {
2879                 p->bdev = bdgrab(I_BDEV(inode));
2880                 error = blkdev_get(p->bdev,
2881                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL, p);
2882                 if (error < 0) {
2883                         p->bdev = NULL;
2884                         return error;
2885                 }
2886                 p->old_block_size = block_size(p->bdev);
2887                 error = set_blocksize(p->bdev, PAGE_SIZE);
2888                 if (error < 0)
2889                         return error;
2890                 /*
2891                  * Zoned block devices contain zones that have a sequential
2892                  * write only restriction.  Hence zoned block devices are not
2893                  * suitable for swapping.  Disallow them here.
2894                  */
2895                 if (blk_queue_is_zoned(p->bdev->bd_queue))
2896                         return -EINVAL;
2897                 p->flags |= SWP_BLKDEV;
2898         } else if (S_ISREG(inode->i_mode)) {
2899                 p->bdev = inode->i_sb->s_bdev;
2900         }
2901 
2902         return 0;
2903 }
2904 
2905 
2906 /*
2907  * Find out how many pages are allowed for a single swap device. There
2908  * are two limiting factors:
2909  * 1) the number of bits for the swap offset in the swp_entry_t type, and
2910  * 2) the number of bits in the swap pte, as defined by the different
2911  * architectures.
2912  *
2913  * In order to find the largest possible bit mask, a swap entry with
2914  * swap type 0 and swap offset ~0UL is created, encoded to a swap pte,
2915  * decoded to a swp_entry_t again, and finally the swap offset is
2916  * extracted.
2917  *
2918  * This will mask all the bits from the initial ~0UL mask that can't
2919  * be encoded in either the swp_entry_t or the architecture definition
2920  * of a swap pte.
2921  */
2922 unsigned long generic_max_swapfile_size(void)
2923 {
2924         return swp_offset(pte_to_swp_entry(
2925                         swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2926 }
2927 
2928 /* Can be overridden by an architecture for additional checks. */
2929 __weak unsigned long max_swapfile_size(void)
2930 {
2931         return generic_max_swapfile_size();
2932 }
2933 
2934 static unsigned long read_swap_header(struct swap_info_struct *p,
2935                                         union swap_header *swap_header,
2936                                         struct inode *inode)
2937 {
2938         int i;
2939         unsigned long maxpages;
2940         unsigned long swapfilepages;
2941         unsigned long last_page;
2942 
2943         if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
2944                 pr_err("Unable to find swap-space signature\n");
2945                 return 0;
2946         }
2947 
2948         /* swap partition endianess hack... */
2949         if (swab32(swap_header->info.version) == 1) {
2950                 swab32s(&swap_header->info.version);
2951                 swab32s(&swap_header->info.last_page);
2952                 swab32s(&swap_header->info.nr_badpages);
2953                 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2954                         return 0;
2955                 for (i = 0; i < swap_header->info.nr_badpages; i++)
2956                         swab32s(&swap_header->info.badpages[i]);
2957         }
2958         /* Check the swap header's sub-version */
2959         if (swap_header->info.version != 1) {
2960                 pr_warn("Unable to handle swap header version %d\n",
2961                         swap_header->info.version);
2962                 return 0;
2963         }
2964 
2965         p->lowest_bit  = 1;
2966         p->cluster_next = 1;
2967         p->cluster_nr = 0;
2968 
2969         maxpages = max_swapfile_size();
2970         last_page = swap_header->info.last_page;
2971         if (!last_page) {
2972                 pr_warn("Empty swap-file\n");
2973                 return 0;
2974         }
2975         if (last_page > maxpages) {
2976                 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
2977                         maxpages << (PAGE_SHIFT - 10),
2978                         last_page << (PAGE_SHIFT - 10));
2979         }
2980         if (maxpages > last_page) {
2981                 maxpages = last_page + 1;
2982                 /* p->max is an unsigned int: don't overflow it */
2983                 if ((unsigned int)maxpages == 0)
2984                         maxpages = UINT_MAX;
2985         }
2986         p->highest_bit = maxpages - 1;
2987 
2988         if (!maxpages)
2989                 return 0;
2990         swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2991         if (swapfilepages && maxpages > swapfilepages) {
2992                 pr_warn("Swap area shorter than signature indicates\n");
2993                 return 0;
2994         }
2995         if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
2996                 return 0;
2997         if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2998                 return 0;
2999 
3000         return maxpages;
3001 }
3002 
3003 #define SWAP_CLUSTER_INFO_COLS                                          \
3004         DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
3005 #define SWAP_CLUSTER_SPACE_COLS                                         \
3006         DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
3007 #define SWAP_CLUSTER_COLS                                               \
3008         max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
3009 
3010 static int setup_swap_map_and_extents(struct swap_info_struct *p,
3011                                         union swap_header *swap_header,
3012                                         unsigned char *swap_map,
3013                                         struct swap_cluster_info *cluster_info,
3014                                         unsigned long maxpages,
3015                                         sector_t *span)
3016 {
3017         unsigned int j, k;
3018         unsigned int nr_good_pages;
3019         int nr_extents;
3020         unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
3021         unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS;
3022         unsigned long i, idx;
3023 
3024         nr_good_pages = maxpages - 1;   /* omit header page */
3025 
3026         cluster_list_init(&p->free_clusters);
3027         cluster_list_init(&p->discard_clusters);
3028 
3029         for (i = 0; i < swap_header->info.nr_badpages; i++) {
3030                 unsigned int page_nr = swap_header->info.badpages[i];
3031                 if (page_nr == 0 || page_nr > swap_header->info.last_page)
3032                         return -EINVAL;
3033                 if (page_nr < maxpages) {
3034                         swap_map[page_nr] = SWAP_MAP_BAD;
3035                         nr_good_pages--;
3036                         /*
3037                          * Haven't marked the cluster free yet, no list
3038                          * operation involved
3039                          */
3040                         inc_cluster_info_page(p, cluster_info, page_nr);
3041                 }
3042         }
3043 
3044         /* Haven't marked the cluster free yet, no list operation involved */
3045         for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
3046                 inc_cluster_info_page(p, cluster_info, i);
3047 
3048         if (nr_good_pages) {
3049                 swap_map[0] = SWAP_MAP_BAD;
3050                 /*
3051                  * Not mark the cluster free yet, no list
3052                  * operation involved
3053                  */
3054                 inc_cluster_info_page(p, cluster_info, 0);
3055                 p->max = maxpages;
3056                 p->pages = nr_good_pages;
3057                 nr_extents = setup_swap_extents(p, span);
3058                 if (nr_extents < 0)
3059                         return nr_extents;
3060                 nr_good_pages = p->pages;
3061         }
3062         if (!nr_good_pages) {
3063                 pr_warn("Empty swap-file\n");
3064                 return -EINVAL;
3065         }
3066 
3067         if (!cluster_info)
3068                 return nr_extents;
3069 
3070 
3071         /*
3072          * Reduce false cache line sharing between cluster_info and
3073          * sharing same address space.
3074          */
3075         for (k = 0; k < SWAP_CLUSTER_COLS; k++) {
3076                 j = (k + col) % SWAP_CLUSTER_COLS;
3077                 for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) {
3078                         idx = i * SWAP_CLUSTER_COLS + j;
3079                         if (idx >= nr_clusters)
3080                                 continue;
3081                         if (cluster_count(&cluster_info[idx]))
3082                                 continue;
3083                         cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
3084                         cluster_list_add_tail(&p->free_clusters, cluster_info,
3085                                               idx);
3086                 }
3087         }
3088         return nr_extents;
3089 }
3090 
3091 /*
3092  * Helper to sys_swapon determining if a given swap
3093  * backing device queue supports DISCARD operations.
3094  */
3095 static bool swap_discardable(struct swap_info_struct *si)
3096 {
3097         struct request_queue *q = bdev_get_queue(si->bdev);
3098 
3099         if (!q || !blk_queue_discard(q))
3100                 return false;
3101 
3102         return true;
3103 }
3104 
3105 SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
3106 {
3107         struct swap_info_struct *p;
3108         struct filename *name;
3109         struct file *swap_file = NULL;
3110         struct address_space *mapping;
3111         int prio;
3112         int error;
3113         union swap_header *swap_header;
3114         int nr_extents;
3115         sector_t span;
3116         unsigned long maxpages;
3117         unsigned char *swap_map = NULL;
3118         struct swap_cluster_info *cluster_info = NULL;
3119         unsigned long *frontswap_map = NULL;
3120         struct page *page = NULL;
3121         struct inode *inode = NULL;
3122         bool inced_nr_rotate_swap = false;
3123 
3124         if (swap_flags & ~SWAP_FLAGS_VALID)
3125                 return -EINVAL;
3126 
3127         if (!capable(CAP_SYS_ADMIN))
3128                 return -EPERM;
3129 
3130         if (!swap_avail_heads)
3131                 return -ENOMEM;
3132 
3133         p = alloc_swap_info();
3134         if (IS_ERR(p))
3135                 return PTR_ERR(p);
3136 
3137         INIT_WORK(&p->discard_work, swap_discard_work);
3138 
3139         name = getname(specialfile);
3140         if (IS_ERR(name)) {
3141                 error = PTR_ERR(name);
3142                 name = NULL;
3143                 goto bad_swap;
3144         }
3145         swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
3146         if (IS_ERR(swap_file)) {
3147                 error = PTR_ERR(swap_file);
3148                 swap_file = NULL;
3149                 goto bad_swap;
3150         }
3151 
3152         p->swap_file = swap_file;
3153         mapping = swap_file->f_mapping;
3154         inode = mapping->host;
3155 
3156         error = claim_swapfile(p, inode);
3157         if (unlikely(error))
3158                 goto bad_swap;
3159 
3160         inode_lock(inode);
3161         if (IS_SWAPFILE(inode)) {
3162                 error = -EBUSY;
3163                 goto bad_swap_unlock_inode;
3164         }
3165 
3166         /*
3167          * Read the swap header.
3168          */
3169         if (!mapping->a_ops->readpage) {
3170                 error = -EINVAL;
3171                 goto bad_swap_unlock_inode;
3172         }
3173         page = read_mapping_page(mapping, 0, swap_file);
3174         if (IS_ERR(page)) {
3175                 error = PTR_ERR(page);
3176                 goto bad_swap;
3177         }
3178         swap_header = kmap(page);
3179 
3180         maxpages = read_swap_header(p, swap_header, inode);
3181         if (unlikely(!maxpages)) {
3182                 error = -EINVAL;
3183                 goto bad_swap_unlock_inode;
3184         }
3185 
3186         /* OK, set up the swap map and apply the bad block list */
3187         swap_map = vzalloc(maxpages);
3188         if (!swap_map) {
3189                 error = -ENOMEM;
3190                 goto bad_swap_unlock_inode;
3191         }
3192 
3193         if (bdi_cap_stable_pages_required(inode_to_bdi(inode)))
3194                 p->flags |= SWP_STABLE_WRITES;
3195 
3196         if (bdi_cap_synchronous_io(inode_to_bdi(inode)))
3197                 p->flags |= SWP_SYNCHRONOUS_IO;
3198 
3199         if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) {
3200                 int cpu;
3201                 unsigned long ci, nr_cluster;
3202 
3203                 p->flags |= SWP_SOLIDSTATE;
3204                 /*
3205                  * select a random position to start with to help wear leveling
3206                  * SSD
3207                  */
3208                 p->cluster_next = 1 + (prandom_u32() % p->highest_bit);
3209                 nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
3210 
3211                 cluster_info = kvcalloc(nr_cluster, sizeof(*cluster_info),
3212                                         GFP_KERNEL);
3213                 if (!cluster_info) {
3214                         error = -ENOMEM;
3215                         goto bad_swap_unlock_inode;
3216                 }
3217 
3218                 for (ci = 0; ci < nr_cluster; ci++)
3219                         spin_lock_init(&((cluster_info + ci)->lock));
3220 
3221                 p->percpu_cluster = alloc_percpu(struct percpu_cluster);
3222                 if (!p->percpu_cluster) {
3223                         error = -ENOMEM;
3224                         goto bad_swap_unlock_inode;
3225                 }
3226                 for_each_possible_cpu(cpu) {
3227                         struct percpu_cluster *cluster;
3228                         cluster = per_cpu_ptr(p->percpu_cluster, cpu);
3229                         cluster_set_null(&cluster->index);
3230                 }
3231         } else {
3232                 atomic_inc(&nr_rotate_swap);
3233                 inced_nr_rotate_swap = true;
3234         }
3235 
3236         error = swap_cgroup_swapon(p->type, maxpages);
3237         if (error)
3238                 goto bad_swap_unlock_inode;
3239 
3240         nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
3241                 cluster_info, maxpages, &span);
3242         if (unlikely(nr_extents < 0)) {
3243                 error = nr_extents;
3244                 goto bad_swap_unlock_inode;
3245         }
3246         /* frontswap enabled? set up bit-per-page map for frontswap */
3247         if (IS_ENABLED(CONFIG_FRONTSWAP))
3248                 frontswap_map = kvcalloc(BITS_TO_LONGS(maxpages),
3249                                          sizeof(long),
3250                                          GFP_KERNEL);
3251 
3252         if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) {
3253                 /*
3254                  * When discard is enabled for swap with no particular
3255                  * policy flagged, we set all swap discard flags here in
3256                  * order to sustain backward compatibility with older
3257                  * swapon(8) releases.
3258                  */
3259                 p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
3260                              SWP_PAGE_DISCARD);
3261 
3262                 /*
3263                  * By flagging sys_swapon, a sysadmin can tell us to
3264                  * either do single-time area discards only, or to just
3265                  * perform discards for released swap page-clusters.
3266                  * Now it's time to adjust the p->flags accordingly.
3267                  */
3268                 if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
3269                         p->flags &= ~SWP_PAGE_DISCARD;
3270                 else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
3271                         p->flags &= ~SWP_AREA_DISCARD;
3272 
3273                 /* issue a swapon-time discard if it's still required */
3274                 if (p->flags & SWP_AREA_DISCARD) {
3275                         int err = discard_swap(p);
3276                         if (unlikely(err))
3277                                 pr_err("swapon: discard_swap(%p): %d\n",
3278                                         p, err);
3279                 }
3280         }
3281 
3282         error = init_swap_address_space(p->type, maxpages);
3283         if (error)
3284                 goto bad_swap_unlock_inode;
3285 
3286         /*
3287          * Flush any pending IO and dirty mappings before we start using this
3288          * swap device.
3289          */
3290         inode->i_flags |= S_SWAPFILE;
3291         error = inode_drain_writes(inode);
3292         if (error) {
3293                 inode->i_flags &= ~S_SWAPFILE;
3294                 goto bad_swap_unlock_inode;
3295         }
3296 
3297         mutex_lock(&swapon_mutex);
3298         prio = -1;
3299         if (swap_flags & SWAP_FLAG_PREFER)
3300                 prio =
3301                   (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
3302         enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
3303 
3304         pr_info("Adding %uk swap on %s.  Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
3305                 p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
3306                 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
3307                 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
3308                 (p->flags & SWP_DISCARDABLE) ? "D" : "",
3309                 (p->flags & SWP_AREA_DISCARD) ? "s" : "",
3310                 (p->flags & SWP_PAGE_DISCARD) ? "c" : "",
3311                 (frontswap_map) ? "FS" : "");
3312 
3313         mutex_unlock(&swapon_mutex);
3314         atomic_inc(&proc_poll_event);
3315         wake_up_interruptible(&proc_poll_wait);
3316 
3317         error = 0;
3318         goto out;
3319 bad_swap_unlock_inode:
3320         inode_unlock(inode);
3321 bad_swap:
3322         free_percpu(p->percpu_cluster);
3323         p->percpu_cluster = NULL;
3324         if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
3325                 set_blocksize(p->bdev, p->old_block_size);
3326                 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
3327         }
3328         inode = NULL;
3329         destroy_swap_extents(p);
3330         swap_cgroup_swapoff(p->type);
3331         spin_lock(&swap_lock);
3332         p->swap_file = NULL;
3333         p->flags = 0;
3334         spin_unlock(&swap_lock);
3335         vfree(swap_map);
3336         kvfree(cluster_info);
3337         kvfree(frontswap_map);
3338         if (inced_nr_rotate_swap)
3339                 atomic_dec(&nr_rotate_swap);
3340         if (swap_file)
3341                 filp_close(swap_file, NULL);
3342 out:
3343         if (page && !IS_ERR(page)) {
3344                 kunmap(page);
3345                 put_page(page);
3346         }
3347         if (name)
3348                 putname(name);
3349         if (inode)
3350                 inode_unlock(inode);
3351         if (!error)
3352                 enable_swap_slots_cache();
3353         return error;
3354 }
3355 
3356 void si_swapinfo(struct sysinfo *val)
3357 {
3358         unsigned int type;
3359         unsigned long nr_to_be_unused = 0;
3360 
3361         spin_lock(&swap_lock);
3362         for (type = 0; type < nr_swapfiles; type++) {
3363                 struct swap_info_struct *si = swap_info[type];
3364 
3365                 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
3366                         nr_to_be_unused += si->inuse_pages;
3367         }
3368         val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
3369         val->totalswap = total_swap_pages + nr_to_be_unused;
3370         spin_unlock(&swap_lock);
3371 }
3372 
3373 /*
3374  * Verify that a swap entry is valid and increment its swap map count.
3375  *
3376  * Returns error code in following case.
3377  * - success -> 0
3378  * - swp_entry is invalid -> EINVAL
3379  * - swp_entry is migration entry -> EINVAL
3380  * - swap-cache reference is requested but there is already one. -> EEXIST
3381  * - swap-cache reference is requested but the entry is not used. -> ENOENT
3382  * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
3383  */
3384 static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
3385 {
3386         struct swap_info_struct *p;
3387         struct swap_cluster_info *ci;
3388         unsigned long offset;
3389         unsigned char count;
3390         unsigned char has_cache;
3391         int err = -EINVAL;
3392 
3393         p = get_swap_device(entry);
3394         if (!p)
3395                 goto out;
3396 
3397         offset = swp_offset(entry);
3398         ci = lock_cluster_or_swap_info(p, offset);
3399 
3400         count = p->swap_map[offset];
3401 
3402         /*
3403          * swapin_readahead() doesn't check if a swap entry is valid, so the
3404          * swap entry could be SWAP_MAP_BAD. Check here with lock held.
3405          */
3406         if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
3407                 err = -ENOENT;
3408                 goto unlock_out;
3409         }
3410 
3411         has_cache = count & SWAP_HAS_CACHE;
3412         count &= ~SWAP_HAS_CACHE;
3413         err = 0;
3414 
3415         if (usage == SWAP_HAS_CACHE) {
3416 
3417                 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
3418                 if (!has_cache && count)
3419                         has_cache = SWAP_HAS_CACHE;
3420                 else if (has_cache)             /* someone else added cache */
3421                         err = -EEXIST;
3422                 else                            /* no users remaining */
3423                         err = -ENOENT;
3424 
3425         } else if (count || has_cache) {
3426 
3427                 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
3428                         count += usage;
3429                 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
3430                         err = -EINVAL;
3431                 else if (swap_count_continued(p, offset, count))
3432                         count = COUNT_CONTINUED;
3433                 else
3434                         err = -ENOMEM;
3435         } else
3436                 err = -ENOENT;                  /* unused swap entry */
3437 
3438         p->swap_map[offset] = count | has_cache;
3439 
3440 unlock_out:
3441         unlock_cluster_or_swap_info(p, ci);
3442 out:
3443         if (p)
3444                 put_swap_device(p);
3445         return err;
3446 }
3447 
3448 /*
3449  * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3450  * (in which case its reference count is never incremented).
3451  */
3452 void swap_shmem_alloc(swp_entry_t entry)
3453 {
3454         __swap_duplicate(entry, SWAP_MAP_SHMEM);
3455 }
3456 
3457 /*
3458  * Increase reference count of swap entry by 1.
3459  * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3460  * but could not be atomically allocated.  Returns 0, just as if it succeeded,
3461  * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3462  * might occur if a page table entry has got corrupted.
3463  */
3464 int swap_duplicate(swp_entry_t entry)
3465 {
3466         int err = 0;
3467 
3468         while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
3469                 err = add_swap_count_continuation(entry, GFP_ATOMIC);
3470         return err;
3471 }
3472 
3473 /*
3474  * @entry: swap entry for which we allocate swap cache.
3475  *
3476  * Called when allocating swap cache for existing swap entry,
3477  * This can return error codes. Returns 0 at success.
3478  * -EBUSY means there is a swap cache.
3479  * Note: return code is different from swap_duplicate().
3480  */
3481 int swapcache_prepare(swp_entry_t entry)
3482 {
3483         return __swap_duplicate(entry, SWAP_HAS_CACHE);
3484 }
3485 
3486 struct swap_info_struct *swp_swap_info(swp_entry_t entry)
3487 {
3488         return swap_type_to_swap_info(swp_type(entry));
3489 }
3490 
3491 struct swap_info_struct *page_swap_info(struct page *page)
3492 {
3493         swp_entry_t entry = { .val = page_private(page) };
3494         return swp_swap_info(entry);
3495 }
3496 
3497 /*
3498  * out-of-line __page_file_ methods to avoid include hell.
3499  */
3500 struct address_space *__page_file_mapping(struct page *page)
3501 {
3502         return page_swap_info(page)->swap_file->f_mapping;
3503 }
3504 EXPORT_SYMBOL_GPL(__page_file_mapping);
3505 
3506 pgoff_t __page_file_index(struct page *page)
3507 {
3508         swp_entry_t swap = { .val = page_private(page) };
3509         return swp_offset(swap);
3510 }
3511 EXPORT_SYMBOL_GPL(__page_file_index);
3512 
3513 /*
3514  * add_swap_count_continuation - called when a swap count is duplicated
3515  * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3516  * page of the original vmalloc'ed swap_map, to hold the continuation count
3517  * (for that entry and for its neighbouring PAGE_SIZE swap entries).  Called
3518  * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3519  *
3520  * These continuation pages are seldom referenced: the common paths all work
3521  * on the original swap_map, only referring to a continuation page when the
3522  * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3523  *
3524  * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3525  * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3526  * can be called after dropping locks.
3527  */
3528 int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
3529 {
3530         struct swap_info_struct *si;
3531         struct swap_cluster_info *ci;
3532         struct page *head;
3533         struct page *page;
3534         struct page *list_page;
3535         pgoff_t offset;
3536         unsigned char count;
3537         int ret = 0;
3538 
3539         /*
3540          * When debugging, it's easier to use __GFP_ZERO here; but it's better
3541          * for latency not to zero a page while GFP_ATOMIC and holding locks.
3542          */
3543         page = alloc_page(gfp_mask | __GFP_HIGHMEM);
3544 
3545         si = get_swap_device(entry);
3546         if (!si) {
3547                 /*
3548                  * An acceptable race has occurred since the failing
3549                  * __swap_duplicate(): the swap device may be swapoff
3550                  */
3551                 goto outer;
3552         }
3553         spin_lock(&si->lock);
3554 
3555         offset = swp_offset(entry);
3556 
3557         ci = lock_cluster(si, offset);
3558 
3559         count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
3560 
3561         if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
3562                 /*
3563                  * The higher the swap count, the more likely it is that tasks
3564                  * will race to add swap count continuation: we need to avoid
3565                  * over-provisioning.
3566                  */
3567                 goto out;
3568         }
3569 
3570         if (!page) {
3571                 ret = -ENOMEM;
3572                 goto out;
3573         }
3574 
3575         /*
3576          * We are fortunate that although vmalloc_to_page uses pte_offset_map,
3577          * no architecture is using highmem pages for kernel page tables: so it
3578          * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
3579          */
3580         head = vmalloc_to_page(si->swap_map + offset);
3581         offset &= ~PAGE_MASK;
3582 
3583         spin_lock(&si->cont_lock);
3584         /*
3585          * Page allocation does not initialize the page's lru field,
3586          * but it does always reset its private field.
3587          */
3588         if (!page_private(head)) {
3589                 BUG_ON(count & COUNT_CONTINUED);
3590                 INIT_LIST_HEAD(&head->lru);
3591                 set_page_private(head, SWP_CONTINUED);
3592                 si->flags |= SWP_CONTINUED;
3593         }
3594 
3595         list_for_each_entry(list_page, &head->lru, lru) {
3596                 unsigned char *map;
3597 
3598                 /*
3599                  * If the previous map said no continuation, but we've found
3600                  * a continuation page, free our allocation and use this one.
3601                  */
3602                 if (!(count & COUNT_CONTINUED))
3603                         goto out_unlock_cont;
3604 
3605                 map = kmap_atomic(list_page) + offset;
3606                 count = *map;
3607                 kunmap_atomic(map);
3608 
3609                 /*
3610                  * If this continuation count now has some space in it,
3611                  * free our allocation and use this one.
3612                  */
3613                 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
3614                         goto out_unlock_cont;
3615         }
3616 
3617         list_add_tail(&page->lru, &head->lru);
3618         page = NULL;                    /* now it's attached, don't free it */
3619 out_unlock_cont:
3620         spin_unlock(&si->cont_lock);
3621 out:
3622         unlock_cluster(ci);
3623         spin_unlock(&si->lock);
3624         put_swap_device(si);
3625 outer:
3626         if (page)
3627                 __free_page(page);
3628         return ret;
3629 }
3630 
3631 /*
3632  * swap_count_continued - when the original swap_map count is incremented
3633  * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3634  * into, carry if so, or else fail until a new continuation page is allocated;
3635  * when the original swap_map count is decremented from 0 with continuation,
3636  * borrow from the continuation and report whether it still holds more.
3637  * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3638  * lock.
3639  */
3640 static bool swap_count_continued(struct swap_info_struct *si,
3641                                  pgoff_t offset, unsigned char count)
3642 {
3643         struct page *head;
3644         struct page *page;
3645         unsigned char *map;
3646         bool ret;
3647 
3648         head = vmalloc_to_page(si->swap_map + offset);
3649         if (page_private(head) != SWP_CONTINUED) {
3650                 BUG_ON(count & COUNT_CONTINUED);
3651                 return false;           /* need to add count continuation */
3652         }
3653 
3654         spin_lock(&si->cont_lock);
3655         offset &= ~PAGE_MASK;
3656         page = list_entry(head->lru.next, struct page, lru);
3657         map = kmap_atomic(page) + offset;
3658 
3659         if (count == SWAP_MAP_MAX)      /* initial increment from swap_map */
3660                 goto init_map;          /* jump over SWAP_CONT_MAX checks */
3661 
3662         if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
3663                 /*
3664                  * Think of how you add 1 to 999
3665                  */
3666                 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
3667                         kunmap_atomic(map);
3668                         page = list_entry(page->lru.next, struct page, lru);
3669                         BUG_ON(page == head);
3670                         map = kmap_atomic(page) + offset;
3671                 }
3672                 if (*map == SWAP_CONT_MAX) {
3673                         kunmap_atomic(map);
3674                         page = list_entry(page->lru.next, struct page, lru);
3675                         if (page == head) {
3676                                 ret = false;    /* add count continuation */
3677                                 goto out;
3678                         }
3679                         map = kmap_atomic(page) + offset;
3680 init_map:               *map = 0;               /* we didn't zero the page */
3681                 }
3682                 *map += 1;
3683                 kunmap_atomic(map);
3684                 page = list_entry(page->lru.prev, struct page, lru);
3685                 while (page != head) {
3686                         map = kmap_atomic(page) + offset;
3687                         *map = COUNT_CONTINUED;
3688                         kunmap_atomic(map);
3689                         page = list_entry(page->lru.prev, struct page, lru);
3690                 }
3691                 ret = true;                     /* incremented */
3692 
3693         } else {                                /* decrementing */
3694                 /*
3695                  * Think of how you subtract 1 from 1000
3696                  */
3697                 BUG_ON(count != COUNT_CONTINUED);
3698                 while (*map == COUNT_CONTINUED) {
3699                         kunmap_atomic(map);
3700                         page = list_entry(page->lru.next, struct page, lru);
3701                         BUG_ON(page == head);
3702                         map = kmap_atomic(page) + offset;
3703                 }
3704                 BUG_ON(*map == 0);
3705                 *map -= 1;
3706                 if (*map == 0)
3707                         count = 0;
3708                 kunmap_atomic(map);
3709                 page = list_entry(page->lru.prev, struct page, lru);
3710                 while (page != head) {
3711                         map = kmap_atomic(page) + offset;
3712                         *map = SWAP_CONT_MAX | count;
3713                         count = COUNT_CONTINUED;
3714                         kunmap_atomic(map);
3715                         page = list_entry(page->lru.prev, struct page, lru);
3716                 }
3717                 ret = count == COUNT_CONTINUED;
3718         }
3719 out:
3720         spin_unlock(&si->cont_lock);
3721         return ret;
3722 }
3723 
3724 /*
3725  * free_swap_count_continuations - swapoff free all the continuation pages
3726  * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3727  */
3728 static void free_swap_count_continuations(struct swap_info_struct *si)
3729 {
3730         pgoff_t offset;
3731 
3732         for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
3733                 struct page *head;
3734                 head = vmalloc_to_page(si->swap_map + offset);
3735                 if (page_private(head)) {
3736                         struct page *page, *next;
3737 
3738                         list_for_each_entry_safe(page, next, &head->lru, lru) {
3739                                 list_del(&page->lru);
3740                                 __free_page(page);
3741                         }
3742                 }
3743         }
3744 }
3745 
3746 #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
3747 void mem_cgroup_throttle_swaprate(struct mem_cgroup *memcg, int node,
3748                                   gfp_t gfp_mask)
3749 {
3750         struct swap_info_struct *si, *next;
3751         if (!(gfp_mask & __GFP_IO) || !memcg)
3752                 return;
3753 
3754         if (!blk_cgroup_congested())
3755                 return;
3756 
3757         /*
3758          * We've already scheduled a throttle, avoid taking the global swap
3759          * lock.
3760          */
3761         if (current->throttle_queue)
3762                 return;
3763 
3764         spin_lock(&swap_avail_lock);
3765         plist_for_each_entry_safe(si, next, &swap_avail_heads[node],
3766                                   avail_lists[node]) {
3767                 if (si->bdev) {
3768                         blkcg_schedule_throttle(bdev_get_queue(si->bdev),
3769                                                 true);
3770                         break;
3771                 }
3772         }
3773         spin_unlock(&swap_avail_lock);
3774 }
3775 #endif
3776 
3777 static int __init swapfile_init(void)
3778 {
3779         int nid;
3780 
3781         swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head),
3782                                          GFP_KERNEL);
3783         if (!swap_avail_heads) {
3784                 pr_emerg("Not enough memory for swap heads, swap is disabled\n");
3785                 return -ENOMEM;
3786         }
3787 
3788         for_each_node(nid)
3789                 plist_head_init(&swap_avail_heads[nid]);
3790 
3791         return 0;
3792 }
3793 subsys_initcall(swapfile_init);
3794 

~ [ 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