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

TOMOYO Linux Cross Reference
Linux/mm/swapfile.c

Version: ~ [ linux-5.1-rc5 ] ~ [ linux-5.0.7 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.34 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.111 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.168 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.178 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.138 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.65 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  *  linux/mm/swapfile.c
  3  *
  4  *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
  5  *  Swap reorganised 29.12.95, Stephen Tweedie
  6  */
  7 
  8 #include <linux/mm.h>
  9 #include <linux/hugetlb.h>
 10 #include <linux/mman.h>
 11 #include <linux/slab.h>
 12 #include <linux/kernel_stat.h>
 13 #include <linux/swap.h>
 14 #include <linux/vmalloc.h>
 15 #include <linux/pagemap.h>
 16 #include <linux/namei.h>
 17 #include <linux/shmem_fs.h>
 18 #include <linux/blkdev.h>
 19 #include <linux/random.h>
 20 #include <linux/writeback.h>
 21 #include <linux/proc_fs.h>
 22 #include <linux/seq_file.h>
 23 #include <linux/init.h>
 24 #include <linux/module.h>
 25 #include <linux/ksm.h>
 26 #include <linux/rmap.h>
 27 #include <linux/security.h>
 28 #include <linux/backing-dev.h>
 29 #include <linux/mutex.h>
 30 #include <linux/capability.h>
 31 #include <linux/syscalls.h>
 32 #include <linux/memcontrol.h>
 33 #include <linux/poll.h>
 34 #include <linux/oom.h>
 35 
 36 #include <asm/pgtable.h>
 37 #include <asm/tlbflush.h>
 38 #include <linux/swapops.h>
 39 #include <linux/page_cgroup.h>
 40 
 41 static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
 42                                  unsigned char);
 43 static void free_swap_count_continuations(struct swap_info_struct *);
 44 static sector_t map_swap_entry(swp_entry_t, struct block_device**);
 45 
 46 static DEFINE_SPINLOCK(swap_lock);
 47 static unsigned int nr_swapfiles;
 48 long nr_swap_pages;
 49 long total_swap_pages;
 50 static int least_priority;
 51 
 52 static const char Bad_file[] = "Bad swap file entry ";
 53 static const char Unused_file[] = "Unused swap file entry ";
 54 static const char Bad_offset[] = "Bad swap offset entry ";
 55 static const char Unused_offset[] = "Unused swap offset entry ";
 56 
 57 static struct swap_list_t swap_list = {-1, -1};
 58 
 59 static struct swap_info_struct *swap_info[MAX_SWAPFILES];
 60 
 61 static DEFINE_MUTEX(swapon_mutex);
 62 
 63 static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
 64 /* Activity counter to indicate that a swapon or swapoff has occurred */
 65 static atomic_t proc_poll_event = ATOMIC_INIT(0);
 66 
 67 static inline unsigned char swap_count(unsigned char ent)
 68 {
 69         return ent & ~SWAP_HAS_CACHE;   /* may include SWAP_HAS_CONT flag */
 70 }
 71 
 72 /* returns 1 if swap entry is freed */
 73 static int
 74 __try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset)
 75 {
 76         swp_entry_t entry = swp_entry(si->type, offset);
 77         struct page *page;
 78         int ret = 0;
 79 
 80         page = find_get_page(&swapper_space, entry.val);
 81         if (!page)
 82                 return 0;
 83         /*
 84          * This function is called from scan_swap_map() and it's called
 85          * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
 86          * We have to use trylock for avoiding deadlock. This is a special
 87          * case and you should use try_to_free_swap() with explicit lock_page()
 88          * in usual operations.
 89          */
 90         if (trylock_page(page)) {
 91                 ret = try_to_free_swap(page);
 92                 unlock_page(page);
 93         }
 94         page_cache_release(page);
 95         return ret;
 96 }
 97 
 98 /*
 99  * swapon tell device that all the old swap contents can be discarded,
100  * to allow the swap device to optimize its wear-levelling.
101  */
102 static int discard_swap(struct swap_info_struct *si)
103 {
104         struct swap_extent *se;
105         sector_t start_block;
106         sector_t nr_blocks;
107         int err = 0;
108 
109         /* Do not discard the swap header page! */
110         se = &si->first_swap_extent;
111         start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
112         nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
113         if (nr_blocks) {
114                 err = blkdev_issue_discard(si->bdev, start_block,
115                                 nr_blocks, GFP_KERNEL, 0);
116                 if (err)
117                         return err;
118                 cond_resched();
119         }
120 
121         list_for_each_entry(se, &si->first_swap_extent.list, list) {
122                 start_block = se->start_block << (PAGE_SHIFT - 9);
123                 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
124 
125                 err = blkdev_issue_discard(si->bdev, start_block,
126                                 nr_blocks, GFP_KERNEL, 0);
127                 if (err)
128                         break;
129 
130                 cond_resched();
131         }
132         return err;             /* That will often be -EOPNOTSUPP */
133 }
134 
135 /*
136  * swap allocation tell device that a cluster of swap can now be discarded,
137  * to allow the swap device to optimize its wear-levelling.
138  */
139 static void discard_swap_cluster(struct swap_info_struct *si,
140                                  pgoff_t start_page, pgoff_t nr_pages)
141 {
142         struct swap_extent *se = si->curr_swap_extent;
143         int found_extent = 0;
144 
145         while (nr_pages) {
146                 struct list_head *lh;
147 
148                 if (se->start_page <= start_page &&
149                     start_page < se->start_page + se->nr_pages) {
150                         pgoff_t offset = start_page - se->start_page;
151                         sector_t start_block = se->start_block + offset;
152                         sector_t nr_blocks = se->nr_pages - offset;
153 
154                         if (nr_blocks > nr_pages)
155                                 nr_blocks = nr_pages;
156                         start_page += nr_blocks;
157                         nr_pages -= nr_blocks;
158 
159                         if (!found_extent++)
160                                 si->curr_swap_extent = se;
161 
162                         start_block <<= PAGE_SHIFT - 9;
163                         nr_blocks <<= PAGE_SHIFT - 9;
164                         if (blkdev_issue_discard(si->bdev, start_block,
165                                     nr_blocks, GFP_NOIO, 0))
166                                 break;
167                 }
168 
169                 lh = se->list.next;
170                 se = list_entry(lh, struct swap_extent, list);
171         }
172 }
173 
174 static int wait_for_discard(void *word)
175 {
176         schedule();
177         return 0;
178 }
179 
180 #define SWAPFILE_CLUSTER        256
181 #define LATENCY_LIMIT           256
182 
183 static unsigned long scan_swap_map(struct swap_info_struct *si,
184                                    unsigned char usage)
185 {
186         unsigned long offset;
187         unsigned long scan_base;
188         unsigned long last_in_cluster = 0;
189         int latency_ration = LATENCY_LIMIT;
190         int found_free_cluster = 0;
191 
192         /*
193          * We try to cluster swap pages by allocating them sequentially
194          * in swap.  Once we've allocated SWAPFILE_CLUSTER pages this
195          * way, however, we resort to first-free allocation, starting
196          * a new cluster.  This prevents us from scattering swap pages
197          * all over the entire swap partition, so that we reduce
198          * overall disk seek times between swap pages.  -- sct
199          * But we do now try to find an empty cluster.  -Andrea
200          * And we let swap pages go all over an SSD partition.  Hugh
201          */
202 
203         si->flags += SWP_SCANNING;
204         scan_base = offset = si->cluster_next;
205 
206         if (unlikely(!si->cluster_nr--)) {
207                 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
208                         si->cluster_nr = SWAPFILE_CLUSTER - 1;
209                         goto checks;
210                 }
211                 if (si->flags & SWP_DISCARDABLE) {
212                         /*
213                          * Start range check on racing allocations, in case
214                          * they overlap the cluster we eventually decide on
215                          * (we scan without swap_lock to allow preemption).
216                          * It's hardly conceivable that cluster_nr could be
217                          * wrapped during our scan, but don't depend on it.
218                          */
219                         if (si->lowest_alloc)
220                                 goto checks;
221                         si->lowest_alloc = si->max;
222                         si->highest_alloc = 0;
223                 }
224                 spin_unlock(&swap_lock);
225 
226                 /*
227                  * If seek is expensive, start searching for new cluster from
228                  * start of partition, to minimize the span of allocated swap.
229                  * But if seek is cheap, search from our current position, so
230                  * that swap is allocated from all over the partition: if the
231                  * Flash Translation Layer only remaps within limited zones,
232                  * we don't want to wear out the first zone too quickly.
233                  */
234                 if (!(si->flags & SWP_SOLIDSTATE))
235                         scan_base = offset = si->lowest_bit;
236                 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
237 
238                 /* Locate the first empty (unaligned) cluster */
239                 for (; last_in_cluster <= si->highest_bit; offset++) {
240                         if (si->swap_map[offset])
241                                 last_in_cluster = offset + SWAPFILE_CLUSTER;
242                         else if (offset == last_in_cluster) {
243                                 spin_lock(&swap_lock);
244                                 offset -= SWAPFILE_CLUSTER - 1;
245                                 si->cluster_next = offset;
246                                 si->cluster_nr = SWAPFILE_CLUSTER - 1;
247                                 found_free_cluster = 1;
248                                 goto checks;
249                         }
250                         if (unlikely(--latency_ration < 0)) {
251                                 cond_resched();
252                                 latency_ration = LATENCY_LIMIT;
253                         }
254                 }
255 
256                 offset = si->lowest_bit;
257                 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
258 
259                 /* Locate the first empty (unaligned) cluster */
260                 for (; last_in_cluster < scan_base; offset++) {
261                         if (si->swap_map[offset])
262                                 last_in_cluster = offset + SWAPFILE_CLUSTER;
263                         else if (offset == last_in_cluster) {
264                                 spin_lock(&swap_lock);
265                                 offset -= SWAPFILE_CLUSTER - 1;
266                                 si->cluster_next = offset;
267                                 si->cluster_nr = SWAPFILE_CLUSTER - 1;
268                                 found_free_cluster = 1;
269                                 goto checks;
270                         }
271                         if (unlikely(--latency_ration < 0)) {
272                                 cond_resched();
273                                 latency_ration = LATENCY_LIMIT;
274                         }
275                 }
276 
277                 offset = scan_base;
278                 spin_lock(&swap_lock);
279                 si->cluster_nr = SWAPFILE_CLUSTER - 1;
280                 si->lowest_alloc = 0;
281         }
282 
283 checks:
284         if (!(si->flags & SWP_WRITEOK))
285                 goto no_page;
286         if (!si->highest_bit)
287                 goto no_page;
288         if (offset > si->highest_bit)
289                 scan_base = offset = si->lowest_bit;
290 
291         /* reuse swap entry of cache-only swap if not busy. */
292         if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
293                 int swap_was_freed;
294                 spin_unlock(&swap_lock);
295                 swap_was_freed = __try_to_reclaim_swap(si, offset);
296                 spin_lock(&swap_lock);
297                 /* entry was freed successfully, try to use this again */
298                 if (swap_was_freed)
299                         goto checks;
300                 goto scan; /* check next one */
301         }
302 
303         if (si->swap_map[offset])
304                 goto scan;
305 
306         if (offset == si->lowest_bit)
307                 si->lowest_bit++;
308         if (offset == si->highest_bit)
309                 si->highest_bit--;
310         si->inuse_pages++;
311         if (si->inuse_pages == si->pages) {
312                 si->lowest_bit = si->max;
313                 si->highest_bit = 0;
314         }
315         si->swap_map[offset] = usage;
316         si->cluster_next = offset + 1;
317         si->flags -= SWP_SCANNING;
318 
319         if (si->lowest_alloc) {
320                 /*
321                  * Only set when SWP_DISCARDABLE, and there's a scan
322                  * for a free cluster in progress or just completed.
323                  */
324                 if (found_free_cluster) {
325                         /*
326                          * To optimize wear-levelling, discard the
327                          * old data of the cluster, taking care not to
328                          * discard any of its pages that have already
329                          * been allocated by racing tasks (offset has
330                          * already stepped over any at the beginning).
331                          */
332                         if (offset < si->highest_alloc &&
333                             si->lowest_alloc <= last_in_cluster)
334                                 last_in_cluster = si->lowest_alloc - 1;
335                         si->flags |= SWP_DISCARDING;
336                         spin_unlock(&swap_lock);
337 
338                         if (offset < last_in_cluster)
339                                 discard_swap_cluster(si, offset,
340                                         last_in_cluster - offset + 1);
341 
342                         spin_lock(&swap_lock);
343                         si->lowest_alloc = 0;
344                         si->flags &= ~SWP_DISCARDING;
345 
346                         smp_mb();       /* wake_up_bit advises this */
347                         wake_up_bit(&si->flags, ilog2(SWP_DISCARDING));
348 
349                 } else if (si->flags & SWP_DISCARDING) {
350                         /*
351                          * Delay using pages allocated by racing tasks
352                          * until the whole discard has been issued. We
353                          * could defer that delay until swap_writepage,
354                          * but it's easier to keep this self-contained.
355                          */
356                         spin_unlock(&swap_lock);
357                         wait_on_bit(&si->flags, ilog2(SWP_DISCARDING),
358                                 wait_for_discard, TASK_UNINTERRUPTIBLE);
359                         spin_lock(&swap_lock);
360                 } else {
361                         /*
362                          * Note pages allocated by racing tasks while
363                          * scan for a free cluster is in progress, so
364                          * that its final discard can exclude them.
365                          */
366                         if (offset < si->lowest_alloc)
367                                 si->lowest_alloc = offset;
368                         if (offset > si->highest_alloc)
369                                 si->highest_alloc = offset;
370                 }
371         }
372         return offset;
373 
374 scan:
375         spin_unlock(&swap_lock);
376         while (++offset <= si->highest_bit) {
377                 if (!si->swap_map[offset]) {
378                         spin_lock(&swap_lock);
379                         goto checks;
380                 }
381                 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
382                         spin_lock(&swap_lock);
383                         goto checks;
384                 }
385                 if (unlikely(--latency_ration < 0)) {
386                         cond_resched();
387                         latency_ration = LATENCY_LIMIT;
388                 }
389         }
390         offset = si->lowest_bit;
391         while (++offset < scan_base) {
392                 if (!si->swap_map[offset]) {
393                         spin_lock(&swap_lock);
394                         goto checks;
395                 }
396                 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
397                         spin_lock(&swap_lock);
398                         goto checks;
399                 }
400                 if (unlikely(--latency_ration < 0)) {
401                         cond_resched();
402                         latency_ration = LATENCY_LIMIT;
403                 }
404         }
405         spin_lock(&swap_lock);
406 
407 no_page:
408         si->flags -= SWP_SCANNING;
409         return 0;
410 }
411 
412 swp_entry_t get_swap_page(void)
413 {
414         struct swap_info_struct *si;
415         pgoff_t offset;
416         int type, next;
417         int wrapped = 0;
418 
419         spin_lock(&swap_lock);
420         if (nr_swap_pages <= 0)
421                 goto noswap;
422         nr_swap_pages--;
423 
424         for (type = swap_list.next; type >= 0 && wrapped < 2; type = next) {
425                 si = swap_info[type];
426                 next = si->next;
427                 if (next < 0 ||
428                     (!wrapped && si->prio != swap_info[next]->prio)) {
429                         next = swap_list.head;
430                         wrapped++;
431                 }
432 
433                 if (!si->highest_bit)
434                         continue;
435                 if (!(si->flags & SWP_WRITEOK))
436                         continue;
437 
438                 swap_list.next = next;
439                 /* This is called for allocating swap entry for cache */
440                 offset = scan_swap_map(si, SWAP_HAS_CACHE);
441                 if (offset) {
442                         spin_unlock(&swap_lock);
443                         return swp_entry(type, offset);
444                 }
445                 next = swap_list.next;
446         }
447 
448         nr_swap_pages++;
449 noswap:
450         spin_unlock(&swap_lock);
451         return (swp_entry_t) {0};
452 }
453 
454 /* The only caller of this function is now susupend routine */
455 swp_entry_t get_swap_page_of_type(int type)
456 {
457         struct swap_info_struct *si;
458         pgoff_t offset;
459 
460         spin_lock(&swap_lock);
461         si = swap_info[type];
462         if (si && (si->flags & SWP_WRITEOK)) {
463                 nr_swap_pages--;
464                 /* This is called for allocating swap entry, not cache */
465                 offset = scan_swap_map(si, 1);
466                 if (offset) {
467                         spin_unlock(&swap_lock);
468                         return swp_entry(type, offset);
469                 }
470                 nr_swap_pages++;
471         }
472         spin_unlock(&swap_lock);
473         return (swp_entry_t) {0};
474 }
475 
476 static struct swap_info_struct *swap_info_get(swp_entry_t entry)
477 {
478         struct swap_info_struct *p;
479         unsigned long offset, type;
480 
481         if (!entry.val)
482                 goto out;
483         type = swp_type(entry);
484         if (type >= nr_swapfiles)
485                 goto bad_nofile;
486         p = swap_info[type];
487         if (!(p->flags & SWP_USED))
488                 goto bad_device;
489         offset = swp_offset(entry);
490         if (offset >= p->max)
491                 goto bad_offset;
492         if (!p->swap_map[offset])
493                 goto bad_free;
494         spin_lock(&swap_lock);
495         return p;
496 
497 bad_free:
498         printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val);
499         goto out;
500 bad_offset:
501         printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val);
502         goto out;
503 bad_device:
504         printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val);
505         goto out;
506 bad_nofile:
507         printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val);
508 out:
509         return NULL;
510 }
511 
512 static unsigned char swap_entry_free(struct swap_info_struct *p,
513                                      swp_entry_t entry, unsigned char usage)
514 {
515         unsigned long offset = swp_offset(entry);
516         unsigned char count;
517         unsigned char has_cache;
518 
519         count = p->swap_map[offset];
520         has_cache = count & SWAP_HAS_CACHE;
521         count &= ~SWAP_HAS_CACHE;
522 
523         if (usage == SWAP_HAS_CACHE) {
524                 VM_BUG_ON(!has_cache);
525                 has_cache = 0;
526         } else if (count == SWAP_MAP_SHMEM) {
527                 /*
528                  * Or we could insist on shmem.c using a special
529                  * swap_shmem_free() and free_shmem_swap_and_cache()...
530                  */
531                 count = 0;
532         } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
533                 if (count == COUNT_CONTINUED) {
534                         if (swap_count_continued(p, offset, count))
535                                 count = SWAP_MAP_MAX | COUNT_CONTINUED;
536                         else
537                                 count = SWAP_MAP_MAX;
538                 } else
539                         count--;
540         }
541 
542         if (!count)
543                 mem_cgroup_uncharge_swap(entry);
544 
545         usage = count | has_cache;
546         p->swap_map[offset] = usage;
547 
548         /* free if no reference */
549         if (!usage) {
550                 struct gendisk *disk = p->bdev->bd_disk;
551                 if (offset < p->lowest_bit)
552                         p->lowest_bit = offset;
553                 if (offset > p->highest_bit)
554                         p->highest_bit = offset;
555                 if (swap_list.next >= 0 &&
556                     p->prio > swap_info[swap_list.next]->prio)
557                         swap_list.next = p->type;
558                 nr_swap_pages++;
559                 p->inuse_pages--;
560                 if ((p->flags & SWP_BLKDEV) &&
561                                 disk->fops->swap_slot_free_notify)
562                         disk->fops->swap_slot_free_notify(p->bdev, offset);
563         }
564 
565         return usage;
566 }
567 
568 /*
569  * Caller has made sure that the swapdevice corresponding to entry
570  * is still around or has not been recycled.
571  */
572 void swap_free(swp_entry_t entry)
573 {
574         struct swap_info_struct *p;
575 
576         p = swap_info_get(entry);
577         if (p) {
578                 swap_entry_free(p, entry, 1);
579                 spin_unlock(&swap_lock);
580         }
581 }
582 
583 /*
584  * Called after dropping swapcache to decrease refcnt to swap entries.
585  */
586 void swapcache_free(swp_entry_t entry, struct page *page)
587 {
588         struct swap_info_struct *p;
589         unsigned char count;
590 
591         p = swap_info_get(entry);
592         if (p) {
593                 count = swap_entry_free(p, entry, SWAP_HAS_CACHE);
594                 if (page)
595                         mem_cgroup_uncharge_swapcache(page, entry, count != 0);
596                 spin_unlock(&swap_lock);
597         }
598 }
599 
600 /*
601  * How many references to page are currently swapped out?
602  * This does not give an exact answer when swap count is continued,
603  * but does include the high COUNT_CONTINUED flag to allow for that.
604  */
605 static inline int page_swapcount(struct page *page)
606 {
607         int count = 0;
608         struct swap_info_struct *p;
609         swp_entry_t entry;
610 
611         entry.val = page_private(page);
612         p = swap_info_get(entry);
613         if (p) {
614                 count = swap_count(p->swap_map[swp_offset(entry)]);
615                 spin_unlock(&swap_lock);
616         }
617         return count;
618 }
619 
620 /*
621  * We can write to an anon page without COW if there are no other references
622  * to it.  And as a side-effect, free up its swap: because the old content
623  * on disk will never be read, and seeking back there to write new content
624  * later would only waste time away from clustering.
625  */
626 int reuse_swap_page(struct page *page)
627 {
628         int count;
629 
630         VM_BUG_ON(!PageLocked(page));
631         if (unlikely(PageKsm(page)))
632                 return 0;
633         count = page_mapcount(page);
634         if (count <= 1 && PageSwapCache(page)) {
635                 count += page_swapcount(page);
636                 if (count == 1 && !PageWriteback(page)) {
637                         delete_from_swap_cache(page);
638                         SetPageDirty(page);
639                 }
640         }
641         return count <= 1;
642 }
643 
644 /*
645  * If swap is getting full, or if there are no more mappings of this page,
646  * then try_to_free_swap is called to free its swap space.
647  */
648 int try_to_free_swap(struct page *page)
649 {
650         VM_BUG_ON(!PageLocked(page));
651 
652         if (!PageSwapCache(page))
653                 return 0;
654         if (PageWriteback(page))
655                 return 0;
656         if (page_swapcount(page))
657                 return 0;
658 
659         /*
660          * Once hibernation has begun to create its image of memory,
661          * there's a danger that one of the calls to try_to_free_swap()
662          * - most probably a call from __try_to_reclaim_swap() while
663          * hibernation is allocating its own swap pages for the image,
664          * but conceivably even a call from memory reclaim - will free
665          * the swap from a page which has already been recorded in the
666          * image as a clean swapcache page, and then reuse its swap for
667          * another page of the image.  On waking from hibernation, the
668          * original page might be freed under memory pressure, then
669          * later read back in from swap, now with the wrong data.
670          *
671          * Hibernation clears bits from gfp_allowed_mask to prevent
672          * memory reclaim from writing to disk, so check that here.
673          */
674         if (!(gfp_allowed_mask & __GFP_IO))
675                 return 0;
676 
677         delete_from_swap_cache(page);
678         SetPageDirty(page);
679         return 1;
680 }
681 
682 /*
683  * Free the swap entry like above, but also try to
684  * free the page cache entry if it is the last user.
685  */
686 int free_swap_and_cache(swp_entry_t entry)
687 {
688         struct swap_info_struct *p;
689         struct page *page = NULL;
690 
691         if (non_swap_entry(entry))
692                 return 1;
693 
694         p = swap_info_get(entry);
695         if (p) {
696                 if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) {
697                         page = find_get_page(&swapper_space, entry.val);
698                         if (page && !trylock_page(page)) {
699                                 page_cache_release(page);
700                                 page = NULL;
701                         }
702                 }
703                 spin_unlock(&swap_lock);
704         }
705         if (page) {
706                 /*
707                  * Not mapped elsewhere, or swap space full? Free it!
708                  * Also recheck PageSwapCache now page is locked (above).
709                  */
710                 if (PageSwapCache(page) && !PageWriteback(page) &&
711                                 (!page_mapped(page) || vm_swap_full())) {
712                         delete_from_swap_cache(page);
713                         SetPageDirty(page);
714                 }
715                 unlock_page(page);
716                 page_cache_release(page);
717         }
718         return p != NULL;
719 }
720 
721 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
722 /**
723  * mem_cgroup_count_swap_user - count the user of a swap entry
724  * @ent: the swap entry to be checked
725  * @pagep: the pointer for the swap cache page of the entry to be stored
726  *
727  * Returns the number of the user of the swap entry. The number is valid only
728  * for swaps of anonymous pages.
729  * If the entry is found on swap cache, the page is stored to pagep with
730  * refcount of it being incremented.
731  */
732 int mem_cgroup_count_swap_user(swp_entry_t ent, struct page **pagep)
733 {
734         struct page *page;
735         struct swap_info_struct *p;
736         int count = 0;
737 
738         page = find_get_page(&swapper_space, ent.val);
739         if (page)
740                 count += page_mapcount(page);
741         p = swap_info_get(ent);
742         if (p) {
743                 count += swap_count(p->swap_map[swp_offset(ent)]);
744                 spin_unlock(&swap_lock);
745         }
746 
747         *pagep = page;
748         return count;
749 }
750 #endif
751 
752 #ifdef CONFIG_HIBERNATION
753 /*
754  * Find the swap type that corresponds to given device (if any).
755  *
756  * @offset - number of the PAGE_SIZE-sized block of the device, starting
757  * from 0, in which the swap header is expected to be located.
758  *
759  * This is needed for the suspend to disk (aka swsusp).
760  */
761 int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
762 {
763         struct block_device *bdev = NULL;
764         int type;
765 
766         if (device)
767                 bdev = bdget(device);
768 
769         spin_lock(&swap_lock);
770         for (type = 0; type < nr_swapfiles; type++) {
771                 struct swap_info_struct *sis = swap_info[type];
772 
773                 if (!(sis->flags & SWP_WRITEOK))
774                         continue;
775 
776                 if (!bdev) {
777                         if (bdev_p)
778                                 *bdev_p = bdgrab(sis->bdev);
779 
780                         spin_unlock(&swap_lock);
781                         return type;
782                 }
783                 if (bdev == sis->bdev) {
784                         struct swap_extent *se = &sis->first_swap_extent;
785 
786                         if (se->start_block == offset) {
787                                 if (bdev_p)
788                                         *bdev_p = bdgrab(sis->bdev);
789 
790                                 spin_unlock(&swap_lock);
791                                 bdput(bdev);
792                                 return type;
793                         }
794                 }
795         }
796         spin_unlock(&swap_lock);
797         if (bdev)
798                 bdput(bdev);
799 
800         return -ENODEV;
801 }
802 
803 /*
804  * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
805  * corresponding to given index in swap_info (swap type).
806  */
807 sector_t swapdev_block(int type, pgoff_t offset)
808 {
809         struct block_device *bdev;
810 
811         if ((unsigned int)type >= nr_swapfiles)
812                 return 0;
813         if (!(swap_info[type]->flags & SWP_WRITEOK))
814                 return 0;
815         return map_swap_entry(swp_entry(type, offset), &bdev);
816 }
817 
818 /*
819  * Return either the total number of swap pages of given type, or the number
820  * of free pages of that type (depending on @free)
821  *
822  * This is needed for software suspend
823  */
824 unsigned int count_swap_pages(int type, int free)
825 {
826         unsigned int n = 0;
827 
828         spin_lock(&swap_lock);
829         if ((unsigned int)type < nr_swapfiles) {
830                 struct swap_info_struct *sis = swap_info[type];
831 
832                 if (sis->flags & SWP_WRITEOK) {
833                         n = sis->pages;
834                         if (free)
835                                 n -= sis->inuse_pages;
836                 }
837         }
838         spin_unlock(&swap_lock);
839         return n;
840 }
841 #endif /* CONFIG_HIBERNATION */
842 
843 /*
844  * No need to decide whether this PTE shares the swap entry with others,
845  * just let do_wp_page work it out if a write is requested later - to
846  * force COW, vm_page_prot omits write permission from any private vma.
847  */
848 static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
849                 unsigned long addr, swp_entry_t entry, struct page *page)
850 {
851         struct mem_cgroup *ptr;
852         spinlock_t *ptl;
853         pte_t *pte;
854         int ret = 1;
855 
856         if (mem_cgroup_try_charge_swapin(vma->vm_mm, page, GFP_KERNEL, &ptr)) {
857                 ret = -ENOMEM;
858                 goto out_nolock;
859         }
860 
861         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
862         if (unlikely(!pte_same(*pte, swp_entry_to_pte(entry)))) {
863                 if (ret > 0)
864                         mem_cgroup_cancel_charge_swapin(ptr);
865                 ret = 0;
866                 goto out;
867         }
868 
869         dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
870         inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
871         get_page(page);
872         set_pte_at(vma->vm_mm, addr, pte,
873                    pte_mkold(mk_pte(page, vma->vm_page_prot)));
874         page_add_anon_rmap(page, vma, addr);
875         mem_cgroup_commit_charge_swapin(page, ptr);
876         swap_free(entry);
877         /*
878          * Move the page to the active list so it is not
879          * immediately swapped out again after swapon.
880          */
881         activate_page(page);
882 out:
883         pte_unmap_unlock(pte, ptl);
884 out_nolock:
885         return ret;
886 }
887 
888 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
889                                 unsigned long addr, unsigned long end,
890                                 swp_entry_t entry, struct page *page)
891 {
892         pte_t swp_pte = swp_entry_to_pte(entry);
893         pte_t *pte;
894         int ret = 0;
895 
896         /*
897          * We don't actually need pte lock while scanning for swp_pte: since
898          * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
899          * page table while we're scanning; though it could get zapped, and on
900          * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
901          * of unmatched parts which look like swp_pte, so unuse_pte must
902          * recheck under pte lock.  Scanning without pte lock lets it be
903          * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
904          */
905         pte = pte_offset_map(pmd, addr);
906         do {
907                 /*
908                  * swapoff spends a _lot_ of time in this loop!
909                  * Test inline before going to call unuse_pte.
910                  */
911                 if (unlikely(pte_same(*pte, swp_pte))) {
912                         pte_unmap(pte);
913                         ret = unuse_pte(vma, pmd, addr, entry, page);
914                         if (ret)
915                                 goto out;
916                         pte = pte_offset_map(pmd, addr);
917                 }
918         } while (pte++, addr += PAGE_SIZE, addr != end);
919         pte_unmap(pte - 1);
920 out:
921         return ret;
922 }
923 
924 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
925                                 unsigned long addr, unsigned long end,
926                                 swp_entry_t entry, struct page *page)
927 {
928         pmd_t *pmd;
929         unsigned long next;
930         int ret;
931 
932         pmd = pmd_offset(pud, addr);
933         do {
934                 next = pmd_addr_end(addr, end);
935                 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
936                         continue;
937                 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
938                 if (ret)
939                         return ret;
940         } while (pmd++, addr = next, addr != end);
941         return 0;
942 }
943 
944 static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
945                                 unsigned long addr, unsigned long end,
946                                 swp_entry_t entry, struct page *page)
947 {
948         pud_t *pud;
949         unsigned long next;
950         int ret;
951 
952         pud = pud_offset(pgd, addr);
953         do {
954                 next = pud_addr_end(addr, end);
955                 if (pud_none_or_clear_bad(pud))
956                         continue;
957                 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
958                 if (ret)
959                         return ret;
960         } while (pud++, addr = next, addr != end);
961         return 0;
962 }
963 
964 static int unuse_vma(struct vm_area_struct *vma,
965                                 swp_entry_t entry, struct page *page)
966 {
967         pgd_t *pgd;
968         unsigned long addr, end, next;
969         int ret;
970 
971         if (page_anon_vma(page)) {
972                 addr = page_address_in_vma(page, vma);
973                 if (addr == -EFAULT)
974                         return 0;
975                 else
976                         end = addr + PAGE_SIZE;
977         } else {
978                 addr = vma->vm_start;
979                 end = vma->vm_end;
980         }
981 
982         pgd = pgd_offset(vma->vm_mm, addr);
983         do {
984                 next = pgd_addr_end(addr, end);
985                 if (pgd_none_or_clear_bad(pgd))
986                         continue;
987                 ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
988                 if (ret)
989                         return ret;
990         } while (pgd++, addr = next, addr != end);
991         return 0;
992 }
993 
994 static int unuse_mm(struct mm_struct *mm,
995                                 swp_entry_t entry, struct page *page)
996 {
997         struct vm_area_struct *vma;
998         int ret = 0;
999 
1000         if (!down_read_trylock(&mm->mmap_sem)) {
1001                 /*
1002                  * Activate page so shrink_inactive_list is unlikely to unmap
1003                  * its ptes while lock is dropped, so swapoff can make progress.
1004                  */
1005                 activate_page(page);
1006                 unlock_page(page);
1007                 down_read(&mm->mmap_sem);
1008                 lock_page(page);
1009         }
1010         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1011                 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
1012                         break;
1013         }
1014         up_read(&mm->mmap_sem);
1015         return (ret < 0)? ret: 0;
1016 }
1017 
1018 /*
1019  * Scan swap_map from current position to next entry still in use.
1020  * Recycle to start on reaching the end, returning 0 when empty.
1021  */
1022 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
1023                                         unsigned int prev)
1024 {
1025         unsigned int max = si->max;
1026         unsigned int i = prev;
1027         unsigned char count;
1028 
1029         /*
1030          * No need for swap_lock here: we're just looking
1031          * for whether an entry is in use, not modifying it; false
1032          * hits are okay, and sys_swapoff() has already prevented new
1033          * allocations from this area (while holding swap_lock).
1034          */
1035         for (;;) {
1036                 if (++i >= max) {
1037                         if (!prev) {
1038                                 i = 0;
1039                                 break;
1040                         }
1041                         /*
1042                          * No entries in use at top of swap_map,
1043                          * loop back to start and recheck there.
1044                          */
1045                         max = prev + 1;
1046                         prev = 0;
1047                         i = 1;
1048                 }
1049                 count = si->swap_map[i];
1050                 if (count && swap_count(count) != SWAP_MAP_BAD)
1051                         break;
1052         }
1053         return i;
1054 }
1055 
1056 /*
1057  * We completely avoid races by reading each swap page in advance,
1058  * and then search for the process using it.  All the necessary
1059  * page table adjustments can then be made atomically.
1060  */
1061 static int try_to_unuse(unsigned int type)
1062 {
1063         struct swap_info_struct *si = swap_info[type];
1064         struct mm_struct *start_mm;
1065         unsigned char *swap_map;
1066         unsigned char swcount;
1067         struct page *page;
1068         swp_entry_t entry;
1069         unsigned int i = 0;
1070         int retval = 0;
1071 
1072         /*
1073          * When searching mms for an entry, a good strategy is to
1074          * start at the first mm we freed the previous entry from
1075          * (though actually we don't notice whether we or coincidence
1076          * freed the entry).  Initialize this start_mm with a hold.
1077          *
1078          * A simpler strategy would be to start at the last mm we
1079          * freed the previous entry from; but that would take less
1080          * advantage of mmlist ordering, which clusters forked mms
1081          * together, child after parent.  If we race with dup_mmap(), we
1082          * prefer to resolve parent before child, lest we miss entries
1083          * duplicated after we scanned child: using last mm would invert
1084          * that.
1085          */
1086         start_mm = &init_mm;
1087         atomic_inc(&init_mm.mm_users);
1088 
1089         /*
1090          * Keep on scanning until all entries have gone.  Usually,
1091          * one pass through swap_map is enough, but not necessarily:
1092          * there are races when an instance of an entry might be missed.
1093          */
1094         while ((i = find_next_to_unuse(si, i)) != 0) {
1095                 if (signal_pending(current)) {
1096                         retval = -EINTR;
1097                         break;
1098                 }
1099 
1100                 /*
1101                  * Get a page for the entry, using the existing swap
1102                  * cache page if there is one.  Otherwise, get a clean
1103                  * page and read the swap into it.
1104                  */
1105                 swap_map = &si->swap_map[i];
1106                 entry = swp_entry(type, i);
1107                 page = read_swap_cache_async(entry,
1108                                         GFP_HIGHUSER_MOVABLE, NULL, 0);
1109                 if (!page) {
1110                         /*
1111                          * Either swap_duplicate() failed because entry
1112                          * has been freed independently, and will not be
1113                          * reused since sys_swapoff() already disabled
1114                          * allocation from here, or alloc_page() failed.
1115                          */
1116                         if (!*swap_map)
1117                                 continue;
1118                         retval = -ENOMEM;
1119                         break;
1120                 }
1121 
1122                 /*
1123                  * Don't hold on to start_mm if it looks like exiting.
1124                  */
1125                 if (atomic_read(&start_mm->mm_users) == 1) {
1126                         mmput(start_mm);
1127                         start_mm = &init_mm;
1128                         atomic_inc(&init_mm.mm_users);
1129                 }
1130 
1131                 /*
1132                  * Wait for and lock page.  When do_swap_page races with
1133                  * try_to_unuse, do_swap_page can handle the fault much
1134                  * faster than try_to_unuse can locate the entry.  This
1135                  * apparently redundant "wait_on_page_locked" lets try_to_unuse
1136                  * defer to do_swap_page in such a case - in some tests,
1137                  * do_swap_page and try_to_unuse repeatedly compete.
1138                  */
1139                 wait_on_page_locked(page);
1140                 wait_on_page_writeback(page);
1141                 lock_page(page);
1142                 wait_on_page_writeback(page);
1143 
1144                 /*
1145                  * Remove all references to entry.
1146                  */
1147                 swcount = *swap_map;
1148                 if (swap_count(swcount) == SWAP_MAP_SHMEM) {
1149                         retval = shmem_unuse(entry, page);
1150                         /* page has already been unlocked and released */
1151                         if (retval < 0)
1152                                 break;
1153                         continue;
1154                 }
1155                 if (swap_count(swcount) && start_mm != &init_mm)
1156                         retval = unuse_mm(start_mm, entry, page);
1157 
1158                 if (swap_count(*swap_map)) {
1159                         int set_start_mm = (*swap_map >= swcount);
1160                         struct list_head *p = &start_mm->mmlist;
1161                         struct mm_struct *new_start_mm = start_mm;
1162                         struct mm_struct *prev_mm = start_mm;
1163                         struct mm_struct *mm;
1164 
1165                         atomic_inc(&new_start_mm->mm_users);
1166                         atomic_inc(&prev_mm->mm_users);
1167                         spin_lock(&mmlist_lock);
1168                         while (swap_count(*swap_map) && !retval &&
1169                                         (p = p->next) != &start_mm->mmlist) {
1170                                 mm = list_entry(p, struct mm_struct, mmlist);
1171                                 if (!atomic_inc_not_zero(&mm->mm_users))
1172                                         continue;
1173                                 spin_unlock(&mmlist_lock);
1174                                 mmput(prev_mm);
1175                                 prev_mm = mm;
1176 
1177                                 cond_resched();
1178 
1179                                 swcount = *swap_map;
1180                                 if (!swap_count(swcount)) /* any usage ? */
1181                                         ;
1182                                 else if (mm == &init_mm)
1183                                         set_start_mm = 1;
1184                                 else
1185                                         retval = unuse_mm(mm, entry, page);
1186 
1187                                 if (set_start_mm && *swap_map < swcount) {
1188                                         mmput(new_start_mm);
1189                                         atomic_inc(&mm->mm_users);
1190                                         new_start_mm = mm;
1191                                         set_start_mm = 0;
1192                                 }
1193                                 spin_lock(&mmlist_lock);
1194                         }
1195                         spin_unlock(&mmlist_lock);
1196                         mmput(prev_mm);
1197                         mmput(start_mm);
1198                         start_mm = new_start_mm;
1199                 }
1200                 if (retval) {
1201                         unlock_page(page);
1202                         page_cache_release(page);
1203                         break;
1204                 }
1205 
1206                 /*
1207                  * If a reference remains (rare), we would like to leave
1208                  * the page in the swap cache; but try_to_unmap could
1209                  * then re-duplicate the entry once we drop page lock,
1210                  * so we might loop indefinitely; also, that page could
1211                  * not be swapped out to other storage meanwhile.  So:
1212                  * delete from cache even if there's another reference,
1213                  * after ensuring that the data has been saved to disk -
1214                  * since if the reference remains (rarer), it will be
1215                  * read from disk into another page.  Splitting into two
1216                  * pages would be incorrect if swap supported "shared
1217                  * private" pages, but they are handled by tmpfs files.
1218                  *
1219                  * Given how unuse_vma() targets one particular offset
1220                  * in an anon_vma, once the anon_vma has been determined,
1221                  * this splitting happens to be just what is needed to
1222                  * handle where KSM pages have been swapped out: re-reading
1223                  * is unnecessarily slow, but we can fix that later on.
1224                  */
1225                 if (swap_count(*swap_map) &&
1226                      PageDirty(page) && PageSwapCache(page)) {
1227                         struct writeback_control wbc = {
1228                                 .sync_mode = WB_SYNC_NONE,
1229                         };
1230 
1231                         swap_writepage(page, &wbc);
1232                         lock_page(page);
1233                         wait_on_page_writeback(page);
1234                 }
1235 
1236                 /*
1237                  * It is conceivable that a racing task removed this page from
1238                  * swap cache just before we acquired the page lock at the top,
1239                  * or while we dropped it in unuse_mm().  The page might even
1240                  * be back in swap cache on another swap area: that we must not
1241                  * delete, since it may not have been written out to swap yet.
1242                  */
1243                 if (PageSwapCache(page) &&
1244                     likely(page_private(page) == entry.val))
1245                         delete_from_swap_cache(page);
1246 
1247                 /*
1248                  * So we could skip searching mms once swap count went
1249                  * to 1, we did not mark any present ptes as dirty: must
1250                  * mark page dirty so shrink_page_list will preserve it.
1251                  */
1252                 SetPageDirty(page);
1253                 unlock_page(page);
1254                 page_cache_release(page);
1255 
1256                 /*
1257                  * Make sure that we aren't completely killing
1258                  * interactive performance.
1259                  */
1260                 cond_resched();
1261         }
1262 
1263         mmput(start_mm);
1264         return retval;
1265 }
1266 
1267 /*
1268  * After a successful try_to_unuse, if no swap is now in use, we know
1269  * we can empty the mmlist.  swap_lock must be held on entry and exit.
1270  * Note that mmlist_lock nests inside swap_lock, and an mm must be
1271  * added to the mmlist just after page_duplicate - before would be racy.
1272  */
1273 static void drain_mmlist(void)
1274 {
1275         struct list_head *p, *next;
1276         unsigned int type;
1277 
1278         for (type = 0; type < nr_swapfiles; type++)
1279                 if (swap_info[type]->inuse_pages)
1280                         return;
1281         spin_lock(&mmlist_lock);
1282         list_for_each_safe(p, next, &init_mm.mmlist)
1283                 list_del_init(p);
1284         spin_unlock(&mmlist_lock);
1285 }
1286 
1287 /*
1288  * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
1289  * corresponds to page offset for the specified swap entry.
1290  * Note that the type of this function is sector_t, but it returns page offset
1291  * into the bdev, not sector offset.
1292  */
1293 static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
1294 {
1295         struct swap_info_struct *sis;
1296         struct swap_extent *start_se;
1297         struct swap_extent *se;
1298         pgoff_t offset;
1299 
1300         sis = swap_info[swp_type(entry)];
1301         *bdev = sis->bdev;
1302 
1303         offset = swp_offset(entry);
1304         start_se = sis->curr_swap_extent;
1305         se = start_se;
1306 
1307         for ( ; ; ) {
1308                 struct list_head *lh;
1309 
1310                 if (se->start_page <= offset &&
1311                                 offset < (se->start_page + se->nr_pages)) {
1312                         return se->start_block + (offset - se->start_page);
1313                 }
1314                 lh = se->list.next;
1315                 se = list_entry(lh, struct swap_extent, list);
1316                 sis->curr_swap_extent = se;
1317                 BUG_ON(se == start_se);         /* It *must* be present */
1318         }
1319 }
1320 
1321 /*
1322  * Returns the page offset into bdev for the specified page's swap entry.
1323  */
1324 sector_t map_swap_page(struct page *page, struct block_device **bdev)
1325 {
1326         swp_entry_t entry;
1327         entry.val = page_private(page);
1328         return map_swap_entry(entry, bdev);
1329 }
1330 
1331 /*
1332  * Free all of a swapdev's extent information
1333  */
1334 static void destroy_swap_extents(struct swap_info_struct *sis)
1335 {
1336         while (!list_empty(&sis->first_swap_extent.list)) {
1337                 struct swap_extent *se;
1338 
1339                 se = list_entry(sis->first_swap_extent.list.next,
1340                                 struct swap_extent, list);
1341                 list_del(&se->list);
1342                 kfree(se);
1343         }
1344 }
1345 
1346 /*
1347  * Add a block range (and the corresponding page range) into this swapdev's
1348  * extent list.  The extent list is kept sorted in page order.
1349  *
1350  * This function rather assumes that it is called in ascending page order.
1351  */
1352 static int
1353 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1354                 unsigned long nr_pages, sector_t start_block)
1355 {
1356         struct swap_extent *se;
1357         struct swap_extent *new_se;
1358         struct list_head *lh;
1359 
1360         if (start_page == 0) {
1361                 se = &sis->first_swap_extent;
1362                 sis->curr_swap_extent = se;
1363                 se->start_page = 0;
1364                 se->nr_pages = nr_pages;
1365                 se->start_block = start_block;
1366                 return 1;
1367         } else {
1368                 lh = sis->first_swap_extent.list.prev;  /* Highest extent */
1369                 se = list_entry(lh, struct swap_extent, list);
1370                 BUG_ON(se->start_page + se->nr_pages != start_page);
1371                 if (se->start_block + se->nr_pages == start_block) {
1372                         /* Merge it */
1373                         se->nr_pages += nr_pages;
1374                         return 0;
1375                 }
1376         }
1377 
1378         /*
1379          * No merge.  Insert a new extent, preserving ordering.
1380          */
1381         new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1382         if (new_se == NULL)
1383                 return -ENOMEM;
1384         new_se->start_page = start_page;
1385         new_se->nr_pages = nr_pages;
1386         new_se->start_block = start_block;
1387 
1388         list_add_tail(&new_se->list, &sis->first_swap_extent.list);
1389         return 1;
1390 }
1391 
1392 /*
1393  * A `swap extent' is a simple thing which maps a contiguous range of pages
1394  * onto a contiguous range of disk blocks.  An ordered list of swap extents
1395  * is built at swapon time and is then used at swap_writepage/swap_readpage
1396  * time for locating where on disk a page belongs.
1397  *
1398  * If the swapfile is an S_ISBLK block device, a single extent is installed.
1399  * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1400  * swap files identically.
1401  *
1402  * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1403  * extent list operates in PAGE_SIZE disk blocks.  Both S_ISREG and S_ISBLK
1404  * swapfiles are handled *identically* after swapon time.
1405  *
1406  * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1407  * and will parse them into an ordered extent list, in PAGE_SIZE chunks.  If
1408  * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1409  * requirements, they are simply tossed out - we will never use those blocks
1410  * for swapping.
1411  *
1412  * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon.  This
1413  * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1414  * which will scribble on the fs.
1415  *
1416  * The amount of disk space which a single swap extent represents varies.
1417  * Typically it is in the 1-4 megabyte range.  So we can have hundreds of
1418  * extents in the list.  To avoid much list walking, we cache the previous
1419  * search location in `curr_swap_extent', and start new searches from there.
1420  * This is extremely effective.  The average number of iterations in
1421  * map_swap_page() has been measured at about 0.3 per page.  - akpm.
1422  */
1423 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
1424 {
1425         struct inode *inode;
1426         unsigned blocks_per_page;
1427         unsigned long page_no;
1428         unsigned blkbits;
1429         sector_t probe_block;
1430         sector_t last_block;
1431         sector_t lowest_block = -1;
1432         sector_t highest_block = 0;
1433         int nr_extents = 0;
1434         int ret;
1435 
1436         inode = sis->swap_file->f_mapping->host;
1437         if (S_ISBLK(inode->i_mode)) {
1438                 ret = add_swap_extent(sis, 0, sis->max, 0);
1439                 *span = sis->pages;
1440                 goto out;
1441         }
1442 
1443         blkbits = inode->i_blkbits;
1444         blocks_per_page = PAGE_SIZE >> blkbits;
1445 
1446         /*
1447          * Map all the blocks into the extent list.  This code doesn't try
1448          * to be very smart.
1449          */
1450         probe_block = 0;
1451         page_no = 0;
1452         last_block = i_size_read(inode) >> blkbits;
1453         while ((probe_block + blocks_per_page) <= last_block &&
1454                         page_no < sis->max) {
1455                 unsigned block_in_page;
1456                 sector_t first_block;
1457 
1458                 first_block = bmap(inode, probe_block);
1459                 if (first_block == 0)
1460                         goto bad_bmap;
1461 
1462                 /*
1463                  * It must be PAGE_SIZE aligned on-disk
1464                  */
1465                 if (first_block & (blocks_per_page - 1)) {
1466                         probe_block++;
1467                         goto reprobe;
1468                 }
1469 
1470                 for (block_in_page = 1; block_in_page < blocks_per_page;
1471                                         block_in_page++) {
1472                         sector_t block;
1473 
1474                         block = bmap(inode, probe_block + block_in_page);
1475                         if (block == 0)
1476                                 goto bad_bmap;
1477                         if (block != first_block + block_in_page) {
1478                                 /* Discontiguity */
1479                                 probe_block++;
1480                                 goto reprobe;
1481                         }
1482                 }
1483 
1484                 first_block >>= (PAGE_SHIFT - blkbits);
1485                 if (page_no) {  /* exclude the header page */
1486                         if (first_block < lowest_block)
1487                                 lowest_block = first_block;
1488                         if (first_block > highest_block)
1489                                 highest_block = first_block;
1490                 }
1491 
1492                 /*
1493                  * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1494                  */
1495                 ret = add_swap_extent(sis, page_no, 1, first_block);
1496                 if (ret < 0)
1497                         goto out;
1498                 nr_extents += ret;
1499                 page_no++;
1500                 probe_block += blocks_per_page;
1501 reprobe:
1502                 continue;
1503         }
1504         ret = nr_extents;
1505         *span = 1 + highest_block - lowest_block;
1506         if (page_no == 0)
1507                 page_no = 1;    /* force Empty message */
1508         sis->max = page_no;
1509         sis->pages = page_no - 1;
1510         sis->highest_bit = page_no - 1;
1511 out:
1512         return ret;
1513 bad_bmap:
1514         printk(KERN_ERR "swapon: swapfile has holes\n");
1515         ret = -EINVAL;
1516         goto out;
1517 }
1518 
1519 static void enable_swap_info(struct swap_info_struct *p, int prio,
1520                                 unsigned char *swap_map)
1521 {
1522         int i, prev;
1523 
1524         spin_lock(&swap_lock);
1525         if (prio >= 0)
1526                 p->prio = prio;
1527         else
1528                 p->prio = --least_priority;
1529         p->swap_map = swap_map;
1530         p->flags |= SWP_WRITEOK;
1531         nr_swap_pages += p->pages;
1532         total_swap_pages += p->pages;
1533 
1534         /* insert swap space into swap_list: */
1535         prev = -1;
1536         for (i = swap_list.head; i >= 0; i = swap_info[i]->next) {
1537                 if (p->prio >= swap_info[i]->prio)
1538                         break;
1539                 prev = i;
1540         }
1541         p->next = i;
1542         if (prev < 0)
1543                 swap_list.head = swap_list.next = p->type;
1544         else
1545                 swap_info[prev]->next = p->type;
1546         spin_unlock(&swap_lock);
1547 }
1548 
1549 SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
1550 {
1551         struct swap_info_struct *p = NULL;
1552         unsigned char *swap_map;
1553         struct file *swap_file, *victim;
1554         struct address_space *mapping;
1555         struct inode *inode;
1556         char *pathname;
1557         int oom_score_adj;
1558         int i, type, prev;
1559         int err;
1560 
1561         if (!capable(CAP_SYS_ADMIN))
1562                 return -EPERM;
1563 
1564         pathname = getname(specialfile);
1565         err = PTR_ERR(pathname);
1566         if (IS_ERR(pathname))
1567                 goto out;
1568 
1569         victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0);
1570         putname(pathname);
1571         err = PTR_ERR(victim);
1572         if (IS_ERR(victim))
1573                 goto out;
1574 
1575         mapping = victim->f_mapping;
1576         prev = -1;
1577         spin_lock(&swap_lock);
1578         for (type = swap_list.head; type >= 0; type = swap_info[type]->next) {
1579                 p = swap_info[type];
1580                 if (p->flags & SWP_WRITEOK) {
1581                         if (p->swap_file->f_mapping == mapping)
1582                                 break;
1583                 }
1584                 prev = type;
1585         }
1586         if (type < 0) {
1587                 err = -EINVAL;
1588                 spin_unlock(&swap_lock);
1589                 goto out_dput;
1590         }
1591         if (!security_vm_enough_memory(p->pages))
1592                 vm_unacct_memory(p->pages);
1593         else {
1594                 err = -ENOMEM;
1595                 spin_unlock(&swap_lock);
1596                 goto out_dput;
1597         }
1598         if (prev < 0)
1599                 swap_list.head = p->next;
1600         else
1601                 swap_info[prev]->next = p->next;
1602         if (type == swap_list.next) {
1603                 /* just pick something that's safe... */
1604                 swap_list.next = swap_list.head;
1605         }
1606         if (p->prio < 0) {
1607                 for (i = p->next; i >= 0; i = swap_info[i]->next)
1608                         swap_info[i]->prio = p->prio--;
1609                 least_priority++;
1610         }
1611         nr_swap_pages -= p->pages;
1612         total_swap_pages -= p->pages;
1613         p->flags &= ~SWP_WRITEOK;
1614         spin_unlock(&swap_lock);
1615 
1616         oom_score_adj = test_set_oom_score_adj(OOM_SCORE_ADJ_MAX);
1617         err = try_to_unuse(type);
1618         test_set_oom_score_adj(oom_score_adj);
1619 
1620         if (err) {
1621                 /*
1622                  * reading p->prio and p->swap_map outside the lock is
1623                  * safe here because only sys_swapon and sys_swapoff
1624                  * change them, and there can be no other sys_swapon or
1625                  * sys_swapoff for this swap_info_struct at this point.
1626                  */
1627                 /* re-insert swap space back into swap_list */
1628                 enable_swap_info(p, p->prio, p->swap_map);
1629                 goto out_dput;
1630         }
1631 
1632         destroy_swap_extents(p);
1633         if (p->flags & SWP_CONTINUED)
1634                 free_swap_count_continuations(p);
1635 
1636         mutex_lock(&swapon_mutex);
1637         spin_lock(&swap_lock);
1638         drain_mmlist();
1639 
1640         /* wait for anyone still in scan_swap_map */
1641         p->highest_bit = 0;             /* cuts scans short */
1642         while (p->flags >= SWP_SCANNING) {
1643                 spin_unlock(&swap_lock);
1644                 schedule_timeout_uninterruptible(1);
1645                 spin_lock(&swap_lock);
1646         }
1647 
1648         swap_file = p->swap_file;
1649         p->swap_file = NULL;
1650         p->max = 0;
1651         swap_map = p->swap_map;
1652         p->swap_map = NULL;
1653         p->flags = 0;
1654         spin_unlock(&swap_lock);
1655         mutex_unlock(&swapon_mutex);
1656         vfree(swap_map);
1657         /* Destroy swap account informatin */
1658         swap_cgroup_swapoff(type);
1659 
1660         inode = mapping->host;
1661         if (S_ISBLK(inode->i_mode)) {
1662                 struct block_device *bdev = I_BDEV(inode);
1663                 set_blocksize(bdev, p->old_block_size);
1664                 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1665         } else {
1666                 mutex_lock(&inode->i_mutex);
1667                 inode->i_flags &= ~S_SWAPFILE;
1668                 mutex_unlock(&inode->i_mutex);
1669         }
1670         filp_close(swap_file, NULL);
1671         err = 0;
1672         atomic_inc(&proc_poll_event);
1673         wake_up_interruptible(&proc_poll_wait);
1674 
1675 out_dput:
1676         filp_close(victim, NULL);
1677 out:
1678         return err;
1679 }
1680 
1681 #ifdef CONFIG_PROC_FS
1682 struct proc_swaps {
1683         struct seq_file seq;
1684         int event;
1685 };
1686 
1687 static unsigned swaps_poll(struct file *file, poll_table *wait)
1688 {
1689         struct proc_swaps *s = file->private_data;
1690 
1691         poll_wait(file, &proc_poll_wait, wait);
1692 
1693         if (s->event != atomic_read(&proc_poll_event)) {
1694                 s->event = atomic_read(&proc_poll_event);
1695                 return POLLIN | POLLRDNORM | POLLERR | POLLPRI;
1696         }
1697 
1698         return POLLIN | POLLRDNORM;
1699 }
1700 
1701 /* iterator */
1702 static void *swap_start(struct seq_file *swap, loff_t *pos)
1703 {
1704         struct swap_info_struct *si;
1705         int type;
1706         loff_t l = *pos;
1707 
1708         mutex_lock(&swapon_mutex);
1709 
1710         if (!l)
1711                 return SEQ_START_TOKEN;
1712 
1713         for (type = 0; type < nr_swapfiles; type++) {
1714                 smp_rmb();      /* read nr_swapfiles before swap_info[type] */
1715                 si = swap_info[type];
1716                 if (!(si->flags & SWP_USED) || !si->swap_map)
1717                         continue;
1718                 if (!--l)
1719                         return si;
1720         }
1721 
1722         return NULL;
1723 }
1724 
1725 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1726 {
1727         struct swap_info_struct *si = v;
1728         int type;
1729 
1730         if (v == SEQ_START_TOKEN)
1731                 type = 0;
1732         else
1733                 type = si->type + 1;
1734 
1735         for (; type < nr_swapfiles; type++) {
1736                 smp_rmb();      /* read nr_swapfiles before swap_info[type] */
1737                 si = swap_info[type];
1738                 if (!(si->flags & SWP_USED) || !si->swap_map)
1739                         continue;
1740                 ++*pos;
1741                 return si;
1742         }
1743 
1744         return NULL;
1745 }
1746 
1747 static void swap_stop(struct seq_file *swap, void *v)
1748 {
1749         mutex_unlock(&swapon_mutex);
1750 }
1751 
1752 static int swap_show(struct seq_file *swap, void *v)
1753 {
1754         struct swap_info_struct *si = v;
1755         struct file *file;
1756         int len;
1757 
1758         if (si == SEQ_START_TOKEN) {
1759                 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1760                 return 0;
1761         }
1762 
1763         file = si->swap_file;
1764         len = seq_path(swap, &file->f_path, " \t\n\\");
1765         seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
1766                         len < 40 ? 40 - len : 1, " ",
1767                         S_ISBLK(file->f_path.dentry->d_inode->i_mode) ?
1768                                 "partition" : "file\t",
1769                         si->pages << (PAGE_SHIFT - 10),
1770                         si->inuse_pages << (PAGE_SHIFT - 10),
1771                         si->prio);
1772         return 0;
1773 }
1774 
1775 static const struct seq_operations swaps_op = {
1776         .start =        swap_start,
1777         .next =         swap_next,
1778         .stop =         swap_stop,
1779         .show =         swap_show
1780 };
1781 
1782 static int swaps_open(struct inode *inode, struct file *file)
1783 {
1784         struct proc_swaps *s;
1785         int ret;
1786 
1787         s = kmalloc(sizeof(struct proc_swaps), GFP_KERNEL);
1788         if (!s)
1789                 return -ENOMEM;
1790 
1791         file->private_data = s;
1792 
1793         ret = seq_open(file, &swaps_op);
1794         if (ret) {
1795                 kfree(s);
1796                 return ret;
1797         }
1798 
1799         s->seq.private = s;
1800         s->event = atomic_read(&proc_poll_event);
1801         return ret;
1802 }
1803 
1804 static const struct file_operations proc_swaps_operations = {
1805         .open           = swaps_open,
1806         .read           = seq_read,
1807         .llseek         = seq_lseek,
1808         .release        = seq_release,
1809         .poll           = swaps_poll,
1810 };
1811 
1812 static int __init procswaps_init(void)
1813 {
1814         proc_create("swaps", 0, NULL, &proc_swaps_operations);
1815         return 0;
1816 }
1817 __initcall(procswaps_init);
1818 #endif /* CONFIG_PROC_FS */
1819 
1820 #ifdef MAX_SWAPFILES_CHECK
1821 static int __init max_swapfiles_check(void)
1822 {
1823         MAX_SWAPFILES_CHECK();
1824         return 0;
1825 }
1826 late_initcall(max_swapfiles_check);
1827 #endif
1828 
1829 static struct swap_info_struct *alloc_swap_info(void)
1830 {
1831         struct swap_info_struct *p;
1832         unsigned int type;
1833 
1834         p = kzalloc(sizeof(*p), GFP_KERNEL);
1835         if (!p)
1836                 return ERR_PTR(-ENOMEM);
1837 
1838         spin_lock(&swap_lock);
1839         for (type = 0; type < nr_swapfiles; type++) {
1840                 if (!(swap_info[type]->flags & SWP_USED))
1841                         break;
1842         }
1843         if (type >= MAX_SWAPFILES) {
1844                 spin_unlock(&swap_lock);
1845                 kfree(p);
1846                 return ERR_PTR(-EPERM);
1847         }
1848         if (type >= nr_swapfiles) {
1849                 p->type = type;
1850                 swap_info[type] = p;
1851                 /*
1852                  * Write swap_info[type] before nr_swapfiles, in case a
1853                  * racing procfs swap_start() or swap_next() is reading them.
1854                  * (We never shrink nr_swapfiles, we never free this entry.)
1855                  */
1856                 smp_wmb();
1857                 nr_swapfiles++;
1858         } else {
1859                 kfree(p);
1860                 p = swap_info[type];
1861                 /*
1862                  * Do not memset this entry: a racing procfs swap_next()
1863                  * would be relying on p->type to remain valid.
1864                  */
1865         }
1866         INIT_LIST_HEAD(&p->first_swap_extent.list);
1867         p->flags = SWP_USED;
1868         p->next = -1;
1869         spin_unlock(&swap_lock);
1870 
1871         return p;
1872 }
1873 
1874 static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
1875 {
1876         int error;
1877 
1878         if (S_ISBLK(inode->i_mode)) {
1879                 p->bdev = bdgrab(I_BDEV(inode));
1880                 error = blkdev_get(p->bdev,
1881                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1882                                    sys_swapon);
1883                 if (error < 0) {
1884                         p->bdev = NULL;
1885                         return -EINVAL;
1886                 }
1887                 p->old_block_size = block_size(p->bdev);
1888                 error = set_blocksize(p->bdev, PAGE_SIZE);
1889                 if (error < 0)
1890                         return error;
1891                 p->flags |= SWP_BLKDEV;
1892         } else if (S_ISREG(inode->i_mode)) {
1893                 p->bdev = inode->i_sb->s_bdev;
1894                 mutex_lock(&inode->i_mutex);
1895                 if (IS_SWAPFILE(inode))
1896                         return -EBUSY;
1897         } else
1898                 return -EINVAL;
1899 
1900         return 0;
1901 }
1902 
1903 static unsigned long read_swap_header(struct swap_info_struct *p,
1904                                         union swap_header *swap_header,
1905                                         struct inode *inode)
1906 {
1907         int i;
1908         unsigned long maxpages;
1909         unsigned long swapfilepages;
1910 
1911         if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
1912                 printk(KERN_ERR "Unable to find swap-space signature\n");
1913                 return 0;
1914         }
1915 
1916         /* swap partition endianess hack... */
1917         if (swab32(swap_header->info.version) == 1) {
1918                 swab32s(&swap_header->info.version);
1919                 swab32s(&swap_header->info.last_page);
1920                 swab32s(&swap_header->info.nr_badpages);
1921                 for (i = 0; i < swap_header->info.nr_badpages; i++)
1922                         swab32s(&swap_header->info.badpages[i]);
1923         }
1924         /* Check the swap header's sub-version */
1925         if (swap_header->info.version != 1) {
1926                 printk(KERN_WARNING
1927                        "Unable to handle swap header version %d\n",
1928                        swap_header->info.version);
1929                 return 0;
1930         }
1931 
1932         p->lowest_bit  = 1;
1933         p->cluster_next = 1;
1934         p->cluster_nr = 0;
1935 
1936         /*
1937          * Find out how many pages are allowed for a single swap
1938          * device. There are two limiting factors: 1) the number of
1939          * bits for the swap offset in the swp_entry_t type and
1940          * 2) the number of bits in the a swap pte as defined by
1941          * the different architectures. In order to find the
1942          * largest possible bit mask a swap entry with swap type 0
1943          * and swap offset ~0UL is created, encoded to a swap pte,
1944          * decoded to a swp_entry_t again and finally the swap
1945          * offset is extracted. This will mask all the bits from
1946          * the initial ~0UL mask that can't be encoded in either
1947          * the swp_entry_t or the architecture definition of a
1948          * swap pte.
1949          */
1950         maxpages = swp_offset(pte_to_swp_entry(
1951                         swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
1952         if (maxpages > swap_header->info.last_page) {
1953                 maxpages = swap_header->info.last_page + 1;
1954                 /* p->max is an unsigned int: don't overflow it */
1955                 if ((unsigned int)maxpages == 0)
1956                         maxpages = UINT_MAX;
1957         }
1958         p->highest_bit = maxpages - 1;
1959 
1960         if (!maxpages)
1961                 return 0;
1962         swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
1963         if (swapfilepages && maxpages > swapfilepages) {
1964                 printk(KERN_WARNING
1965                        "Swap area shorter than signature indicates\n");
1966                 return 0;
1967         }
1968         if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
1969                 return 0;
1970         if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
1971                 return 0;
1972 
1973         return maxpages;
1974 }
1975 
1976 static int setup_swap_map_and_extents(struct swap_info_struct *p,
1977                                         union swap_header *swap_header,
1978                                         unsigned char *swap_map,
1979                                         unsigned long maxpages,
1980                                         sector_t *span)
1981 {
1982         int i;
1983         unsigned int nr_good_pages;
1984         int nr_extents;
1985 
1986         nr_good_pages = maxpages - 1;   /* omit header page */
1987 
1988         for (i = 0; i < swap_header->info.nr_badpages; i++) {
1989                 unsigned int page_nr = swap_header->info.badpages[i];
1990                 if (page_nr == 0 || page_nr > swap_header->info.last_page)
1991                         return -EINVAL;
1992                 if (page_nr < maxpages) {
1993                         swap_map[page_nr] = SWAP_MAP_BAD;
1994                         nr_good_pages--;
1995                 }
1996         }
1997 
1998         if (nr_good_pages) {
1999                 swap_map[0] = SWAP_MAP_BAD;
2000                 p->max = maxpages;
2001                 p->pages = nr_good_pages;
2002                 nr_extents = setup_swap_extents(p, span);
2003                 if (nr_extents < 0)
2004                         return nr_extents;
2005                 nr_good_pages = p->pages;
2006         }
2007         if (!nr_good_pages) {
2008                 printk(KERN_WARNING "Empty swap-file\n");
2009                 return -EINVAL;
2010         }
2011 
2012         return nr_extents;
2013 }
2014 
2015 SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2016 {
2017         struct swap_info_struct *p;
2018         char *name;
2019         struct file *swap_file = NULL;
2020         struct address_space *mapping;
2021         int i;
2022         int prio;
2023         int error;
2024         union swap_header *swap_header;
2025         int nr_extents;
2026         sector_t span;
2027         unsigned long maxpages;
2028         unsigned char *swap_map = NULL;
2029         struct page *page = NULL;
2030         struct inode *inode = NULL;
2031 
2032         if (!capable(CAP_SYS_ADMIN))
2033                 return -EPERM;
2034 
2035         p = alloc_swap_info();
2036         if (IS_ERR(p))
2037                 return PTR_ERR(p);
2038 
2039         name = getname(specialfile);
2040         if (IS_ERR(name)) {
2041                 error = PTR_ERR(name);
2042                 name = NULL;
2043                 goto bad_swap;
2044         }
2045         swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0);
2046         if (IS_ERR(swap_file)) {
2047                 error = PTR_ERR(swap_file);
2048                 swap_file = NULL;
2049                 goto bad_swap;
2050         }
2051 
2052         p->swap_file = swap_file;
2053         mapping = swap_file->f_mapping;
2054 
2055         for (i = 0; i < nr_swapfiles; i++) {
2056                 struct swap_info_struct *q = swap_info[i];
2057 
2058                 if (q == p || !q->swap_file)
2059                         continue;
2060                 if (mapping == q->swap_file->f_mapping) {
2061                         error = -EBUSY;
2062                         goto bad_swap;
2063                 }
2064         }
2065 
2066         inode = mapping->host;
2067         /* If S_ISREG(inode->i_mode) will do mutex_lock(&inode->i_mutex); */
2068         error = claim_swapfile(p, inode);
2069         if (unlikely(error))
2070                 goto bad_swap;
2071 
2072         /*
2073          * Read the swap header.
2074          */
2075         if (!mapping->a_ops->readpage) {
2076                 error = -EINVAL;
2077                 goto bad_swap;
2078         }
2079         page = read_mapping_page(mapping, 0, swap_file);
2080         if (IS_ERR(page)) {
2081                 error = PTR_ERR(page);
2082                 goto bad_swap;
2083         }
2084         swap_header = kmap(page);
2085 
2086         maxpages = read_swap_header(p, swap_header, inode);
2087         if (unlikely(!maxpages)) {
2088                 error = -EINVAL;
2089                 goto bad_swap;
2090         }
2091 
2092         /* OK, set up the swap map and apply the bad block list */
2093         swap_map = vzalloc(maxpages);
2094         if (!swap_map) {
2095                 error = -ENOMEM;
2096                 goto bad_swap;
2097         }
2098 
2099         error = swap_cgroup_swapon(p->type, maxpages);
2100         if (error)
2101                 goto bad_swap;
2102 
2103         nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
2104                 maxpages, &span);
2105         if (unlikely(nr_extents < 0)) {
2106                 error = nr_extents;
2107                 goto bad_swap;
2108         }
2109 
2110         if (p->bdev) {
2111                 if (blk_queue_nonrot(bdev_get_queue(p->bdev))) {
2112                         p->flags |= SWP_SOLIDSTATE;
2113                         p->cluster_next = 1 + (random32() % p->highest_bit);
2114                 }
2115                 if (discard_swap(p) == 0 && (swap_flags & SWAP_FLAG_DISCARD))
2116                         p->flags |= SWP_DISCARDABLE;
2117         }
2118 
2119         mutex_lock(&swapon_mutex);
2120         prio = -1;
2121         if (swap_flags & SWAP_FLAG_PREFER)
2122                 prio =
2123                   (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
2124         enable_swap_info(p, prio, swap_map);
2125 
2126         printk(KERN_INFO "Adding %uk swap on %s.  "
2127                         "Priority:%d extents:%d across:%lluk %s%s\n",
2128                 p->pages<<(PAGE_SHIFT-10), name, p->prio,
2129                 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
2130                 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
2131                 (p->flags & SWP_DISCARDABLE) ? "D" : "");
2132 
2133         mutex_unlock(&swapon_mutex);
2134         atomic_inc(&proc_poll_event);
2135         wake_up_interruptible(&proc_poll_wait);
2136 
2137         if (S_ISREG(inode->i_mode))
2138                 inode->i_flags |= S_SWAPFILE;
2139         error = 0;
2140         goto out;
2141 bad_swap:
2142         if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
2143                 set_blocksize(p->bdev, p->old_block_size);
2144                 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2145         }
2146         destroy_swap_extents(p);
2147         swap_cgroup_swapoff(p->type);
2148         spin_lock(&swap_lock);
2149         p->swap_file = NULL;
2150         p->flags = 0;
2151         spin_unlock(&swap_lock);
2152         vfree(swap_map);
2153         if (swap_file) {
2154                 if (inode && S_ISREG(inode->i_mode)) {
2155                         mutex_unlock(&inode->i_mutex);
2156                         inode = NULL;
2157                 }
2158                 filp_close(swap_file, NULL);
2159         }
2160 out:
2161         if (page && !IS_ERR(page)) {
2162                 kunmap(page);
2163                 page_cache_release(page);
2164         }
2165         if (name)
2166                 putname(name);
2167         if (inode && S_ISREG(inode->i_mode))
2168                 mutex_unlock(&inode->i_mutex);
2169         return error;
2170 }
2171 
2172 void si_swapinfo(struct sysinfo *val)
2173 {
2174         unsigned int type;
2175         unsigned long nr_to_be_unused = 0;
2176 
2177         spin_lock(&swap_lock);
2178         for (type = 0; type < nr_swapfiles; type++) {
2179                 struct swap_info_struct *si = swap_info[type];
2180 
2181                 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
2182                         nr_to_be_unused += si->inuse_pages;
2183         }
2184         val->freeswap = nr_swap_pages + nr_to_be_unused;
2185         val->totalswap = total_swap_pages + nr_to_be_unused;
2186         spin_unlock(&swap_lock);
2187 }
2188 
2189 /*
2190  * Verify that a swap entry is valid and increment its swap map count.
2191  *
2192  * Returns error code in following case.
2193  * - success -> 0
2194  * - swp_entry is invalid -> EINVAL
2195  * - swp_entry is migration entry -> EINVAL
2196  * - swap-cache reference is requested but there is already one. -> EEXIST
2197  * - swap-cache reference is requested but the entry is not used. -> ENOENT
2198  * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
2199  */
2200 static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
2201 {
2202         struct swap_info_struct *p;
2203         unsigned long offset, type;
2204         unsigned char count;
2205         unsigned char has_cache;
2206         int err = -EINVAL;
2207 
2208         if (non_swap_entry(entry))
2209                 goto out;
2210 
2211         type = swp_type(entry);
2212         if (type >= nr_swapfiles)
2213                 goto bad_file;
2214         p = swap_info[type];
2215         offset = swp_offset(entry);
2216 
2217         spin_lock(&swap_lock);
2218         if (unlikely(offset >= p->max))
2219                 goto unlock_out;
2220 
2221         count = p->swap_map[offset];
2222         has_cache = count & SWAP_HAS_CACHE;
2223         count &= ~SWAP_HAS_CACHE;
2224         err = 0;
2225 
2226         if (usage == SWAP_HAS_CACHE) {
2227 
2228                 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
2229                 if (!has_cache && count)
2230                         has_cache = SWAP_HAS_CACHE;
2231                 else if (has_cache)             /* someone else added cache */
2232                         err = -EEXIST;
2233                 else                            /* no users remaining */
2234                         err = -ENOENT;
2235 
2236         } else if (count || has_cache) {
2237 
2238                 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
2239                         count += usage;
2240                 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
2241                         err = -EINVAL;
2242                 else if (swap_count_continued(p, offset, count))
2243                         count = COUNT_CONTINUED;
2244                 else
2245                         err = -ENOMEM;
2246         } else
2247                 err = -ENOENT;                  /* unused swap entry */
2248 
2249         p->swap_map[offset] = count | has_cache;
2250 
2251 unlock_out:
2252         spin_unlock(&swap_lock);
2253 out:
2254         return err;
2255 
2256 bad_file:
2257         printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
2258         goto out;
2259 }
2260 
2261 /*
2262  * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2263  * (in which case its reference count is never incremented).
2264  */
2265 void swap_shmem_alloc(swp_entry_t entry)
2266 {
2267         __swap_duplicate(entry, SWAP_MAP_SHMEM);
2268 }
2269 
2270 /*
2271  * Increase reference count of swap entry by 1.
2272  * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2273  * but could not be atomically allocated.  Returns 0, just as if it succeeded,
2274  * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2275  * might occur if a page table entry has got corrupted.
2276  */
2277 int swap_duplicate(swp_entry_t entry)
2278 {
2279         int err = 0;
2280 
2281         while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
2282                 err = add_swap_count_continuation(entry, GFP_ATOMIC);
2283         return err;
2284 }
2285 
2286 /*
2287  * @entry: swap entry for which we allocate swap cache.
2288  *
2289  * Called when allocating swap cache for existing swap entry,
2290  * This can return error codes. Returns 0 at success.
2291  * -EBUSY means there is a swap cache.
2292  * Note: return code is different from swap_duplicate().
2293  */
2294 int swapcache_prepare(swp_entry_t entry)
2295 {
2296         return __swap_duplicate(entry, SWAP_HAS_CACHE);
2297 }
2298 
2299 /*
2300  * swap_lock prevents swap_map being freed. Don't grab an extra
2301  * reference on the swaphandle, it doesn't matter if it becomes unused.
2302  */
2303 int valid_swaphandles(swp_entry_t entry, unsigned long *offset)
2304 {
2305         struct swap_info_struct *si;
2306         int our_page_cluster = page_cluster;
2307         pgoff_t target, toff;
2308         pgoff_t base, end;
2309         int nr_pages = 0;
2310 
2311         if (!our_page_cluster)  /* no readahead */
2312                 return 0;
2313 
2314         si = swap_info[swp_type(entry)];
2315         target = swp_offset(entry);
2316         base = (target >> our_page_cluster) << our_page_cluster;
2317         end = base + (1 << our_page_cluster);
2318         if (!base)              /* first page is swap header */
2319                 base++;
2320 
2321         spin_lock(&swap_lock);
2322         if (end > si->max)      /* don't go beyond end of map */
2323                 end = si->max;
2324 
2325         /* Count contiguous allocated slots above our target */
2326         for (toff = target; ++toff < end; nr_pages++) {
2327                 /* Don't read in free or bad pages */
2328                 if (!si->swap_map[toff])
2329                         break;
2330                 if (swap_count(si->swap_map[toff]) == SWAP_MAP_BAD)
2331                         break;
2332         }
2333         /* Count contiguous allocated slots below our target */
2334         for (toff = target; --toff >= base; nr_pages++) {
2335                 /* Don't read in free or bad pages */
2336                 if (!si->swap_map[toff])
2337                         break;
2338                 if (swap_count(si->swap_map[toff]) == SWAP_MAP_BAD)
2339                         break;
2340         }
2341         spin_unlock(&swap_lock);
2342 
2343         /*
2344          * Indicate starting offset, and return number of pages to get:
2345          * if only 1, say 0, since there's then no readahead to be done.
2346          */
2347         *offset = ++toff;
2348         return nr_pages? ++nr_pages: 0;
2349 }
2350 
2351 /*
2352  * add_swap_count_continuation - called when a swap count is duplicated
2353  * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2354  * page of the original vmalloc'ed swap_map, to hold the continuation count
2355  * (for that entry and for its neighbouring PAGE_SIZE swap entries).  Called
2356  * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2357  *
2358  * These continuation pages are seldom referenced: the common paths all work
2359  * on the original swap_map, only referring to a continuation page when the
2360  * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2361  *
2362  * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2363  * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2364  * can be called after dropping locks.
2365  */
2366 int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
2367 {
2368         struct swap_info_struct *si;
2369         struct page *head;
2370         struct page *page;
2371         struct page *list_page;
2372         pgoff_t offset;
2373         unsigned char count;
2374 
2375         /*
2376          * When debugging, it's easier to use __GFP_ZERO here; but it's better
2377          * for latency not to zero a page while GFP_ATOMIC and holding locks.
2378          */
2379         page = alloc_page(gfp_mask | __GFP_HIGHMEM);
2380 
2381         si = swap_info_get(entry);
2382         if (!si) {
2383                 /*
2384                  * An acceptable race has occurred since the failing
2385                  * __swap_duplicate(): the swap entry has been freed,
2386                  * perhaps even the whole swap_map cleared for swapoff.
2387                  */
2388                 goto outer;
2389         }
2390 
2391         offset = swp_offset(entry);
2392         count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
2393 
2394         if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
2395                 /*
2396                  * The higher the swap count, the more likely it is that tasks
2397                  * will race to add swap count continuation: we need to avoid
2398                  * over-provisioning.
2399                  */
2400                 goto out;
2401         }
2402 
2403         if (!page) {
2404                 spin_unlock(&swap_lock);
2405                 return -ENOMEM;
2406         }
2407 
2408         /*
2409          * We are fortunate that although vmalloc_to_page uses pte_offset_map,
2410          * no architecture is using highmem pages for kernel pagetables: so it
2411          * will not corrupt the GFP_ATOMIC caller's atomic pagetable kmaps.
2412          */
2413         head = vmalloc_to_page(si->swap_map + offset);
2414         offset &= ~PAGE_MASK;
2415 
2416         /*
2417          * Page allocation does not initialize the page's lru field,
2418          * but it does always reset its private field.
2419          */
2420         if (!page_private(head)) {
2421                 BUG_ON(count & COUNT_CONTINUED);
2422                 INIT_LIST_HEAD(&head->lru);
2423                 set_page_private(head, SWP_CONTINUED);
2424                 si->flags |= SWP_CONTINUED;
2425         }
2426 
2427         list_for_each_entry(list_page, &head->lru, lru) {
2428                 unsigned char *map;
2429 
2430                 /*
2431                  * If the previous map said no continuation, but we've found
2432                  * a continuation page, free our allocation and use this one.
2433                  */
2434                 if (!(count & COUNT_CONTINUED))
2435                         goto out;
2436 
2437                 map = kmap_atomic(list_page, KM_USER0) + offset;
2438                 count = *map;
2439                 kunmap_atomic(map, KM_USER0);
2440 
2441                 /*
2442                  * If this continuation count now has some space in it,
2443                  * free our allocation and use this one.
2444                  */
2445                 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
2446                         goto out;
2447         }
2448 
2449         list_add_tail(&page->lru, &head->lru);
2450         page = NULL;                    /* now it's attached, don't free it */
2451 out:
2452         spin_unlock(&swap_lock);
2453 outer:
2454         if (page)
2455                 __free_page(page);
2456         return 0;
2457 }
2458 
2459 /*
2460  * swap_count_continued - when the original swap_map count is incremented
2461  * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2462  * into, carry if so, or else fail until a new continuation page is allocated;
2463  * when the original swap_map count is decremented from 0 with continuation,
2464  * borrow from the continuation and report whether it still holds more.
2465  * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2466  */
2467 static bool swap_count_continued(struct swap_info_struct *si,
2468                                  pgoff_t offset, unsigned char count)
2469 {
2470         struct page *head;
2471         struct page *page;
2472         unsigned char *map;
2473 
2474         head = vmalloc_to_page(si->swap_map + offset);
2475         if (page_private(head) != SWP_CONTINUED) {
2476                 BUG_ON(count & COUNT_CONTINUED);
2477                 return false;           /* need to add count continuation */
2478         }
2479 
2480         offset &= ~PAGE_MASK;
2481         page = list_entry(head->lru.next, struct page, lru);
2482         map = kmap_atomic(page, KM_USER0) + offset;
2483 
2484         if (count == SWAP_MAP_MAX)      /* initial increment from swap_map */
2485                 goto init_map;          /* jump over SWAP_CONT_MAX checks */
2486 
2487         if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
2488                 /*
2489                  * Think of how you add 1 to 999
2490                  */
2491                 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
2492                         kunmap_atomic(map, KM_USER0);
2493                         page = list_entry(page->lru.next, struct page, lru);
2494                         BUG_ON(page == head);
2495                         map = kmap_atomic(page, KM_USER0) + offset;
2496                 }
2497                 if (*map == SWAP_CONT_MAX) {
2498                         kunmap_atomic(map, KM_USER0);
2499                         page = list_entry(page->lru.next, struct page, lru);
2500                         if (page == head)
2501                                 return false;   /* add count continuation */
2502                         map = kmap_atomic(page, KM_USER0) + offset;
2503 init_map:               *map = 0;               /* we didn't zero the page */
2504                 }
2505                 *map += 1;
2506                 kunmap_atomic(map, KM_USER0);
2507                 page = list_entry(page->lru.prev, struct page, lru);
2508                 while (page != head) {
2509                         map = kmap_atomic(page, KM_USER0) + offset;
2510                         *map = COUNT_CONTINUED;
2511                         kunmap_atomic(map, KM_USER0);
2512                         page = list_entry(page->lru.prev, struct page, lru);
2513                 }
2514                 return true;                    /* incremented */
2515 
2516         } else {                                /* decrementing */
2517                 /*
2518                  * Think of how you subtract 1 from 1000
2519                  */
2520                 BUG_ON(count != COUNT_CONTINUED);
2521                 while (*map == COUNT_CONTINUED) {
2522                         kunmap_atomic(map, KM_USER0);
2523                         page = list_entry(page->lru.next, struct page, lru);
2524                         BUG_ON(page == head);
2525                         map = kmap_atomic(page, KM_USER0) + offset;
2526                 }
2527                 BUG_ON(*map == 0);
2528                 *map -= 1;
2529                 if (*map == 0)
2530                         count = 0;
2531                 kunmap_atomic(map, KM_USER0);
2532                 page = list_entry(page->lru.prev, struct page, lru);
2533                 while (page != head) {
2534                         map = kmap_atomic(page, KM_USER0) + offset;
2535                         *map = SWAP_CONT_MAX | count;
2536                         count = COUNT_CONTINUED;
2537                         kunmap_atomic(map, KM_USER0);
2538                         page = list_entry(page->lru.prev, struct page, lru);
2539                 }
2540                 return count == COUNT_CONTINUED;
2541         }
2542 }
2543 
2544 /*
2545  * free_swap_count_continuations - swapoff free all the continuation pages
2546  * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
2547  */
2548 static void free_swap_count_continuations(struct swap_info_struct *si)
2549 {
2550         pgoff_t offset;
2551 
2552         for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
2553                 struct page *head;
2554                 head = vmalloc_to_page(si->swap_map + offset);
2555                 if (page_private(head)) {
2556                         struct list_head *this, *next;
2557                         list_for_each_safe(this, next, &head->lru) {
2558                                 struct page *page;
2559                                 page = list_entry(this, struct page, lru);
2560                                 list_del(this);
2561                                 __free_page(page);
2562                         }
2563                 }
2564         }
2565 }
2566 

~ [ 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