source: src/linux/universal/linux-3.18/mm/swapfile.c @ 31869

Last change on this file since 31869 was 31869, checked in by brainslayer, 2 months ago

update

File size: 77.0 KB
Line 
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/ksm.h>
25#include <linux/rmap.h>
26#include <linux/security.h>
27#include <linux/backing-dev.h>
28#include <linux/mutex.h>
29#include <linux/capability.h>
30#include <linux/syscalls.h>
31#include <linux/memcontrol.h>
32#include <linux/poll.h>
33#include <linux/oom.h>
34#include <linux/frontswap.h>
35#include <linux/swapfile.h>
36#include <linux/export.h>
37
38#include <asm/pgtable.h>
39#include <asm/tlbflush.h>
40#include <linux/swapops.h>
41#include <linux/page_cgroup.h>
42
43static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
44                                 unsigned char);
45static void free_swap_count_continuations(struct swap_info_struct *);
46static sector_t map_swap_entry(swp_entry_t, struct block_device**);
47
48DEFINE_SPINLOCK(swap_lock);
49static unsigned int nr_swapfiles;
50atomic_long_t nr_swap_pages;
51/* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
52long total_swap_pages;
53static int least_priority;
54
55static const char Bad_file[] = "Bad swap file entry ";
56static const char Unused_file[] = "Unused swap file entry ";
57static const char Bad_offset[] = "Bad swap offset entry ";
58static const char Unused_offset[] = "Unused swap offset entry ";
59
60/*
61 * all active swap_info_structs
62 * protected with swap_lock, and ordered by priority.
63 */
64PLIST_HEAD(swap_active_head);
65
66/*
67 * all available (active, not full) swap_info_structs
68 * protected with swap_avail_lock, ordered by priority.
69 * This is used by get_swap_page() instead of swap_active_head
70 * because swap_active_head includes all swap_info_structs,
71 * but get_swap_page() doesn't need to look at full ones.
72 * This uses its own lock instead of swap_lock because when a
73 * swap_info_struct changes between not-full/full, it needs to
74 * add/remove itself to/from this list, but the swap_info_struct->lock
75 * is held and the locking order requires swap_lock to be taken
76 * before any swap_info_struct->lock.
77 */
78static PLIST_HEAD(swap_avail_head);
79static DEFINE_SPINLOCK(swap_avail_lock);
80
81struct swap_info_struct *swap_info[MAX_SWAPFILES];
82
83static DEFINE_MUTEX(swapon_mutex);
84
85static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
86/* Activity counter to indicate that a swapon or swapoff has occurred */
87static atomic_t proc_poll_event = ATOMIC_INIT(0);
88
89static inline unsigned char swap_count(unsigned char ent)
90{
91        return ent & ~SWAP_HAS_CACHE;   /* may include SWAP_HAS_CONT flag */
92}
93
94/* returns 1 if swap entry is freed */
95static int
96__try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset)
97{
98        swp_entry_t entry = swp_entry(si->type, offset);
99        struct page *page;
100        int ret = 0;
101
102        page = find_get_page(swap_address_space(entry), entry.val);
103        if (!page)
104                return 0;
105        /*
106         * This function is called from scan_swap_map() and it's called
107         * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
108         * We have to use trylock for avoiding deadlock. This is a special
109         * case and you should use try_to_free_swap() with explicit lock_page()
110         * in usual operations.
111         */
112        if (trylock_page(page)) {
113                ret = try_to_free_swap(page);
114                unlock_page(page);
115        }
116        page_cache_release(page);
117        return ret;
118}
119
120/*
121 * swapon tell device that all the old swap contents can be discarded,
122 * to allow the swap device to optimize its wear-levelling.
123 */
124static int discard_swap(struct swap_info_struct *si)
125{
126        struct swap_extent *se;
127        sector_t start_block;
128        sector_t nr_blocks;
129        int err = 0;
130
131        /* Do not discard the swap header page! */
132        se = &si->first_swap_extent;
133        start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
134        nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
135        if (nr_blocks) {
136                err = blkdev_issue_discard(si->bdev, start_block,
137                                nr_blocks, GFP_KERNEL, 0);
138                if (err)
139                        return err;
140                cond_resched();
141        }
142
143        list_for_each_entry(se, &si->first_swap_extent.list, list) {
144                start_block = se->start_block << (PAGE_SHIFT - 9);
145                nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
146
147                err = blkdev_issue_discard(si->bdev, start_block,
148                                nr_blocks, GFP_KERNEL, 0);
149                if (err)
150                        break;
151
152                cond_resched();
153        }
154        return err;             /* That will often be -EOPNOTSUPP */
155}
156
157/*
158 * swap allocation tell device that a cluster of swap can now be discarded,
159 * to allow the swap device to optimize its wear-levelling.
160 */
161static void discard_swap_cluster(struct swap_info_struct *si,
162                                 pgoff_t start_page, pgoff_t nr_pages)
163{
164        struct swap_extent *se = si->curr_swap_extent;
165        int found_extent = 0;
166
167        while (nr_pages) {
168                struct list_head *lh;
169
170                if (se->start_page <= start_page &&
171                    start_page < se->start_page + se->nr_pages) {
172                        pgoff_t offset = start_page - se->start_page;
173                        sector_t start_block = se->start_block + offset;
174                        sector_t nr_blocks = se->nr_pages - offset;
175
176                        if (nr_blocks > nr_pages)
177                                nr_blocks = nr_pages;
178                        start_page += nr_blocks;
179                        nr_pages -= nr_blocks;
180
181                        if (!found_extent++)
182                                si->curr_swap_extent = se;
183
184                        start_block <<= PAGE_SHIFT - 9;
185                        nr_blocks <<= PAGE_SHIFT - 9;
186                        if (blkdev_issue_discard(si->bdev, start_block,
187                                    nr_blocks, GFP_NOIO, 0))
188                                break;
189                }
190
191                lh = se->list.next;
192                se = list_entry(lh, struct swap_extent, list);
193        }
194}
195
196#define SWAPFILE_CLUSTER        256
197#define LATENCY_LIMIT           256
198
199static inline void cluster_set_flag(struct swap_cluster_info *info,
200        unsigned int flag)
201{
202        info->flags = flag;
203}
204
205static inline unsigned int cluster_count(struct swap_cluster_info *info)
206{
207        return info->data;
208}
209
210static inline void cluster_set_count(struct swap_cluster_info *info,
211                                     unsigned int c)
212{
213        info->data = c;
214}
215
216static inline void cluster_set_count_flag(struct swap_cluster_info *info,
217                                         unsigned int c, unsigned int f)
218{
219        info->flags = f;
220        info->data = c;
221}
222
223static inline unsigned int cluster_next(struct swap_cluster_info *info)
224{
225        return info->data;
226}
227
228static inline void cluster_set_next(struct swap_cluster_info *info,
229                                    unsigned int n)
230{
231        info->data = n;
232}
233
234static inline void cluster_set_next_flag(struct swap_cluster_info *info,
235                                         unsigned int n, unsigned int f)
236{
237        info->flags = f;
238        info->data = n;
239}
240
241static inline bool cluster_is_free(struct swap_cluster_info *info)
242{
243        return info->flags & CLUSTER_FLAG_FREE;
244}
245
246static inline bool cluster_is_null(struct swap_cluster_info *info)
247{
248        return info->flags & CLUSTER_FLAG_NEXT_NULL;
249}
250
251static inline void cluster_set_null(struct swap_cluster_info *info)
252{
253        info->flags = CLUSTER_FLAG_NEXT_NULL;
254        info->data = 0;
255}
256
257/* Add a cluster to discard list and schedule it to do discard */
258static void swap_cluster_schedule_discard(struct swap_info_struct *si,
259                unsigned int idx)
260{
261        /*
262         * If scan_swap_map() can't find a free cluster, it will check
263         * si->swap_map directly. To make sure the discarding cluster isn't
264         * taken by scan_swap_map(), mark the swap entries bad (occupied). It
265         * will be cleared after discard
266         */
267        memset(si->swap_map + idx * SWAPFILE_CLUSTER,
268                        SWAP_MAP_BAD, SWAPFILE_CLUSTER);
269
270        if (cluster_is_null(&si->discard_cluster_head)) {
271                cluster_set_next_flag(&si->discard_cluster_head,
272                                                idx, 0);
273                cluster_set_next_flag(&si->discard_cluster_tail,
274                                                idx, 0);
275        } else {
276                unsigned int tail = cluster_next(&si->discard_cluster_tail);
277                cluster_set_next(&si->cluster_info[tail], idx);
278                cluster_set_next_flag(&si->discard_cluster_tail,
279                                                idx, 0);
280        }
281
282        schedule_work(&si->discard_work);
283}
284
285/*
286 * Doing discard actually. After a cluster discard is finished, the cluster
287 * will be added to free cluster list. caller should hold si->lock.
288*/
289static void swap_do_scheduled_discard(struct swap_info_struct *si)
290{
291        struct swap_cluster_info *info;
292        unsigned int idx;
293
294        info = si->cluster_info;
295
296        while (!cluster_is_null(&si->discard_cluster_head)) {
297                idx = cluster_next(&si->discard_cluster_head);
298
299                cluster_set_next_flag(&si->discard_cluster_head,
300                                                cluster_next(&info[idx]), 0);
301                if (cluster_next(&si->discard_cluster_tail) == idx) {
302                        cluster_set_null(&si->discard_cluster_head);
303                        cluster_set_null(&si->discard_cluster_tail);
304                }
305                spin_unlock(&si->lock);
306
307                discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
308                                SWAPFILE_CLUSTER);
309
310                spin_lock(&si->lock);
311                cluster_set_flag(&info[idx], CLUSTER_FLAG_FREE);
312                if (cluster_is_null(&si->free_cluster_head)) {
313                        cluster_set_next_flag(&si->free_cluster_head,
314                                                idx, 0);
315                        cluster_set_next_flag(&si->free_cluster_tail,
316                                                idx, 0);
317                } else {
318                        unsigned int tail;
319
320                        tail = cluster_next(&si->free_cluster_tail);
321                        cluster_set_next(&info[tail], idx);
322                        cluster_set_next_flag(&si->free_cluster_tail,
323                                                idx, 0);
324                }
325                memset(si->swap_map + idx * SWAPFILE_CLUSTER,
326                                0, SWAPFILE_CLUSTER);
327        }
328}
329
330static void swap_discard_work(struct work_struct *work)
331{
332        struct swap_info_struct *si;
333
334        si = container_of(work, struct swap_info_struct, discard_work);
335
336        spin_lock(&si->lock);
337        swap_do_scheduled_discard(si);
338        spin_unlock(&si->lock);
339}
340
341/*
342 * The cluster corresponding to page_nr will be used. The cluster will be
343 * removed from free cluster list and its usage counter will be increased.
344 */
345static void inc_cluster_info_page(struct swap_info_struct *p,
346        struct swap_cluster_info *cluster_info, unsigned long page_nr)
347{
348        unsigned long idx = page_nr / SWAPFILE_CLUSTER;
349
350        if (!cluster_info)
351                return;
352        if (cluster_is_free(&cluster_info[idx])) {
353                VM_BUG_ON(cluster_next(&p->free_cluster_head) != idx);
354                cluster_set_next_flag(&p->free_cluster_head,
355                        cluster_next(&cluster_info[idx]), 0);
356                if (cluster_next(&p->free_cluster_tail) == idx) {
357                        cluster_set_null(&p->free_cluster_tail);
358                        cluster_set_null(&p->free_cluster_head);
359                }
360                cluster_set_count_flag(&cluster_info[idx], 0, 0);
361        }
362
363        VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
364        cluster_set_count(&cluster_info[idx],
365                cluster_count(&cluster_info[idx]) + 1);
366}
367
368/*
369 * The cluster corresponding to page_nr decreases one usage. If the usage
370 * counter becomes 0, which means no page in the cluster is in using, we can
371 * optionally discard the cluster and add it to free cluster list.
372 */
373static void dec_cluster_info_page(struct swap_info_struct *p,
374        struct swap_cluster_info *cluster_info, unsigned long page_nr)
375{
376        unsigned long idx = page_nr / SWAPFILE_CLUSTER;
377
378        if (!cluster_info)
379                return;
380
381        VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
382        cluster_set_count(&cluster_info[idx],
383                cluster_count(&cluster_info[idx]) - 1);
384
385        if (cluster_count(&cluster_info[idx]) == 0) {
386                /*
387                 * If the swap is discardable, prepare discard the cluster
388                 * instead of free it immediately. The cluster will be freed
389                 * after discard.
390                 */
391                if ((p->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
392                                 (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
393                        swap_cluster_schedule_discard(p, idx);
394                        return;
395                }
396
397                cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
398                if (cluster_is_null(&p->free_cluster_head)) {
399                        cluster_set_next_flag(&p->free_cluster_head, idx, 0);
400                        cluster_set_next_flag(&p->free_cluster_tail, idx, 0);
401                } else {
402                        unsigned int tail = cluster_next(&p->free_cluster_tail);
403                        cluster_set_next(&cluster_info[tail], idx);
404                        cluster_set_next_flag(&p->free_cluster_tail, idx, 0);
405                }
406        }
407}
408
409/*
410 * It's possible scan_swap_map() uses a free cluster in the middle of free
411 * cluster list. Avoiding such abuse to avoid list corruption.
412 */
413static bool
414scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
415        unsigned long offset)
416{
417        struct percpu_cluster *percpu_cluster;
418        bool conflict;
419
420        offset /= SWAPFILE_CLUSTER;
421        conflict = !cluster_is_null(&si->free_cluster_head) &&
422                offset != cluster_next(&si->free_cluster_head) &&
423                cluster_is_free(&si->cluster_info[offset]);
424
425        if (!conflict)
426                return false;
427
428        percpu_cluster = this_cpu_ptr(si->percpu_cluster);
429        cluster_set_null(&percpu_cluster->index);
430        return true;
431}
432
433/*
434 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
435 * might involve allocating a new cluster for current CPU too.
436 */
437static void scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
438        unsigned long *offset, unsigned long *scan_base)
439{
440        struct percpu_cluster *cluster;
441        bool found_free;
442        unsigned long tmp;
443
444new_cluster:
445        cluster = this_cpu_ptr(si->percpu_cluster);
446        if (cluster_is_null(&cluster->index)) {
447                if (!cluster_is_null(&si->free_cluster_head)) {
448                        cluster->index = si->free_cluster_head;
449                        cluster->next = cluster_next(&cluster->index) *
450                                        SWAPFILE_CLUSTER;
451                } else if (!cluster_is_null(&si->discard_cluster_head)) {
452                        /*
453                         * we don't have free cluster but have some clusters in
454                         * discarding, do discard now and reclaim them
455                         */
456                        swap_do_scheduled_discard(si);
457                        *scan_base = *offset = si->cluster_next;
458                        goto new_cluster;
459                } else
460                        return;
461        }
462
463        found_free = false;
464
465        /*
466         * Other CPUs can use our cluster if they can't find a free cluster,
467         * check if there is still free entry in the cluster
468         */
469        tmp = cluster->next;
470        while (tmp < si->max && tmp < (cluster_next(&cluster->index) + 1) *
471               SWAPFILE_CLUSTER) {
472                if (!si->swap_map[tmp]) {
473                        found_free = true;
474                        break;
475                }
476                tmp++;
477        }
478        if (!found_free) {
479                cluster_set_null(&cluster->index);
480                goto new_cluster;
481        }
482        cluster->next = tmp + 1;
483        *offset = tmp;
484        *scan_base = tmp;
485}
486
487static unsigned long scan_swap_map(struct swap_info_struct *si,
488                                   unsigned char usage)
489{
490        unsigned long offset;
491        unsigned long scan_base;
492        unsigned long last_in_cluster = 0;
493        int latency_ration = LATENCY_LIMIT;
494
495        /*
496         * We try to cluster swap pages by allocating them sequentially
497         * in swap.  Once we've allocated SWAPFILE_CLUSTER pages this
498         * way, however, we resort to first-free allocation, starting
499         * a new cluster.  This prevents us from scattering swap pages
500         * all over the entire swap partition, so that we reduce
501         * overall disk seek times between swap pages.  -- sct
502         * But we do now try to find an empty cluster.  -Andrea
503         * And we let swap pages go all over an SSD partition.  Hugh
504         */
505
506        si->flags += SWP_SCANNING;
507        scan_base = offset = si->cluster_next;
508
509        /* SSD algorithm */
510        if (si->cluster_info) {
511                scan_swap_map_try_ssd_cluster(si, &offset, &scan_base);
512                goto checks;
513        }
514
515        if (unlikely(!si->cluster_nr--)) {
516                if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
517                        si->cluster_nr = SWAPFILE_CLUSTER - 1;
518                        goto checks;
519                }
520
521                spin_unlock(&si->lock);
522
523                /*
524                 * If seek is expensive, start searching for new cluster from
525                 * start of partition, to minimize the span of allocated swap.
526                 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
527                 * case, just handled by scan_swap_map_try_ssd_cluster() above.
528                 */
529                scan_base = offset = si->lowest_bit;
530                last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
531
532                /* Locate the first empty (unaligned) cluster */
533                for (; last_in_cluster <= si->highest_bit; offset++) {
534                        if (si->swap_map[offset])
535                                last_in_cluster = offset + SWAPFILE_CLUSTER;
536                        else if (offset == last_in_cluster) {
537                                spin_lock(&si->lock);
538                                offset -= SWAPFILE_CLUSTER - 1;
539                                si->cluster_next = offset;
540                                si->cluster_nr = SWAPFILE_CLUSTER - 1;
541                                goto checks;
542                        }
543                        if (unlikely(--latency_ration < 0)) {
544                                cond_resched();
545                                latency_ration = LATENCY_LIMIT;
546                        }
547                }
548
549                offset = scan_base;
550                spin_lock(&si->lock);
551                si->cluster_nr = SWAPFILE_CLUSTER - 1;
552        }
553
554checks:
555        if (si->cluster_info) {
556                while (scan_swap_map_ssd_cluster_conflict(si, offset))
557                        scan_swap_map_try_ssd_cluster(si, &offset, &scan_base);
558        }
559        if (!(si->flags & SWP_WRITEOK))
560                goto no_page;
561        if (!si->highest_bit)
562                goto no_page;
563        if (offset > si->highest_bit)
564                scan_base = offset = si->lowest_bit;
565
566        /* reuse swap entry of cache-only swap if not busy. */
567        if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
568                int swap_was_freed;
569                spin_unlock(&si->lock);
570                swap_was_freed = __try_to_reclaim_swap(si, offset);
571                spin_lock(&si->lock);
572                /* entry was freed successfully, try to use this again */
573                if (swap_was_freed)
574                        goto checks;
575                goto scan; /* check next one */
576        }
577
578        if (si->swap_map[offset])
579                goto scan;
580
581        if (offset == si->lowest_bit)
582                si->lowest_bit++;
583        if (offset == si->highest_bit)
584                si->highest_bit--;
585        si->inuse_pages++;
586        if (si->inuse_pages == si->pages) {
587                si->lowest_bit = si->max;
588                si->highest_bit = 0;
589                spin_lock(&swap_avail_lock);
590                plist_del(&si->avail_list, &swap_avail_head);
591                spin_unlock(&swap_avail_lock);
592        }
593        si->swap_map[offset] = usage;
594        inc_cluster_info_page(si, si->cluster_info, offset);
595        si->cluster_next = offset + 1;
596        si->flags -= SWP_SCANNING;
597
598        return offset;
599
600scan:
601        spin_unlock(&si->lock);
602        while (++offset <= si->highest_bit) {
603                if (!si->swap_map[offset]) {
604                        spin_lock(&si->lock);
605                        goto checks;
606                }
607                if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
608                        spin_lock(&si->lock);
609                        goto checks;
610                }
611                if (unlikely(--latency_ration < 0)) {
612                        cond_resched();
613                        latency_ration = LATENCY_LIMIT;
614                }
615        }
616        offset = si->lowest_bit;
617        while (offset < scan_base) {
618                if (!si->swap_map[offset]) {
619                        spin_lock(&si->lock);
620                        goto checks;
621                }
622                if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
623                        spin_lock(&si->lock);
624                        goto checks;
625                }
626                if (unlikely(--latency_ration < 0)) {
627                        cond_resched();
628                        latency_ration = LATENCY_LIMIT;
629                }
630                offset++;
631        }
632        spin_lock(&si->lock);
633
634no_page:
635        si->flags -= SWP_SCANNING;
636        return 0;
637}
638
639swp_entry_t get_swap_page(void)
640{
641        struct swap_info_struct *si, *next;
642        pgoff_t offset;
643
644        if (atomic_long_read(&nr_swap_pages) <= 0)
645                goto noswap;
646        atomic_long_dec(&nr_swap_pages);
647
648        spin_lock(&swap_avail_lock);
649
650start_over:
651        plist_for_each_entry_safe(si, next, &swap_avail_head, avail_list) {
652                /* requeue si to after same-priority siblings */
653                plist_requeue(&si->avail_list, &swap_avail_head);
654                spin_unlock(&swap_avail_lock);
655                spin_lock(&si->lock);
656                if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) {
657                        spin_lock(&swap_avail_lock);
658                        if (plist_node_empty(&si->avail_list)) {
659                                spin_unlock(&si->lock);
660                                goto nextsi;
661                        }
662                        WARN(!si->highest_bit,
663                             "swap_info %d in list but !highest_bit\n",
664                             si->type);
665                        WARN(!(si->flags & SWP_WRITEOK),
666                             "swap_info %d in list but !SWP_WRITEOK\n",
667                             si->type);
668                        plist_del(&si->avail_list, &swap_avail_head);
669                        spin_unlock(&si->lock);
670                        goto nextsi;
671                }
672
673                /* This is called for allocating swap entry for cache */
674                offset = scan_swap_map(si, SWAP_HAS_CACHE);
675                spin_unlock(&si->lock);
676                if (offset)
677                        return swp_entry(si->type, offset);
678                pr_debug("scan_swap_map of si %d failed to find offset\n",
679                       si->type);
680                spin_lock(&swap_avail_lock);
681nextsi:
682                /*
683                 * if we got here, it's likely that si was almost full before,
684                 * and since scan_swap_map() can drop the si->lock, multiple
685                 * callers probably all tried to get a page from the same si
686                 * and it filled up before we could get one; or, the si filled
687                 * up between us dropping swap_avail_lock and taking si->lock.
688                 * Since we dropped the swap_avail_lock, the swap_avail_head
689                 * list may have been modified; so if next is still in the
690                 * swap_avail_head list then try it, otherwise start over.
691                 */
692                if (plist_node_empty(&next->avail_list))
693                        goto start_over;
694        }
695
696        spin_unlock(&swap_avail_lock);
697
698        atomic_long_inc(&nr_swap_pages);
699noswap:
700        return (swp_entry_t) {0};
701}
702
703/* The only caller of this function is now suspend routine */
704swp_entry_t get_swap_page_of_type(int type)
705{
706        struct swap_info_struct *si;
707        pgoff_t offset;
708
709        si = swap_info[type];
710        spin_lock(&si->lock);
711        if (si && (si->flags & SWP_WRITEOK)) {
712                atomic_long_dec(&nr_swap_pages);
713                /* This is called for allocating swap entry, not cache */
714                offset = scan_swap_map(si, 1);
715                if (offset) {
716                        spin_unlock(&si->lock);
717                        return swp_entry(type, offset);
718                }
719                atomic_long_inc(&nr_swap_pages);
720        }
721        spin_unlock(&si->lock);
722        return (swp_entry_t) {0};
723}
724
725static struct swap_info_struct *swap_info_get(swp_entry_t entry)
726{
727        struct swap_info_struct *p;
728        unsigned long offset, type;
729
730        if (!entry.val)
731                goto out;
732        type = swp_type(entry);
733        if (type >= nr_swapfiles)
734                goto bad_nofile;
735        p = swap_info[type];
736        if (!(p->flags & SWP_USED))
737                goto bad_device;
738        offset = swp_offset(entry);
739        if (offset >= p->max)
740                goto bad_offset;
741        if (!p->swap_map[offset])
742                goto bad_free;
743        spin_lock(&p->lock);
744        return p;
745
746bad_free:
747        pr_err("swap_free: %s%08lx\n", Unused_offset, entry.val);
748        goto out;
749bad_offset:
750        pr_err("swap_free: %s%08lx\n", Bad_offset, entry.val);
751        goto out;
752bad_device:
753        pr_err("swap_free: %s%08lx\n", Unused_file, entry.val);
754        goto out;
755bad_nofile:
756        pr_err("swap_free: %s%08lx\n", Bad_file, entry.val);
757out:
758        return NULL;
759}
760
761static unsigned char swap_entry_free(struct swap_info_struct *p,
762                                     swp_entry_t entry, unsigned char usage)
763{
764        unsigned long offset = swp_offset(entry);
765        unsigned char count;
766        unsigned char has_cache;
767
768        count = p->swap_map[offset];
769        has_cache = count & SWAP_HAS_CACHE;
770        count &= ~SWAP_HAS_CACHE;
771
772        if (usage == SWAP_HAS_CACHE) {
773                VM_BUG_ON(!has_cache);
774                has_cache = 0;
775        } else if (count == SWAP_MAP_SHMEM) {
776                /*
777                 * Or we could insist on shmem.c using a special
778                 * swap_shmem_free() and free_shmem_swap_and_cache()...
779                 */
780                count = 0;
781        } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
782                if (count == COUNT_CONTINUED) {
783                        if (swap_count_continued(p, offset, count))
784                                count = SWAP_MAP_MAX | COUNT_CONTINUED;
785                        else
786                                count = SWAP_MAP_MAX;
787                } else
788                        count--;
789        }
790
791        if (!count)
792                mem_cgroup_uncharge_swap(entry);
793
794        usage = count | has_cache;
795        p->swap_map[offset] = usage;
796
797        /* free if no reference */
798        if (!usage) {
799                dec_cluster_info_page(p, p->cluster_info, offset);
800                if (offset < p->lowest_bit)
801                        p->lowest_bit = offset;
802                if (offset > p->highest_bit) {
803                        bool was_full = !p->highest_bit;
804                        p->highest_bit = offset;
805                        if (was_full && (p->flags & SWP_WRITEOK)) {
806                                spin_lock(&swap_avail_lock);
807                                WARN_ON(!plist_node_empty(&p->avail_list));
808                                if (plist_node_empty(&p->avail_list))
809                                        plist_add(&p->avail_list,
810                                                  &swap_avail_head);
811                                spin_unlock(&swap_avail_lock);
812                        }
813                }
814                atomic_long_inc(&nr_swap_pages);
815                p->inuse_pages--;
816                frontswap_invalidate_page(p->type, offset);
817                if (p->flags & SWP_BLKDEV) {
818                        struct gendisk *disk = p->bdev->bd_disk;
819                        if (disk->fops->swap_slot_free_notify)
820                                disk->fops->swap_slot_free_notify(p->bdev,
821                                                                  offset);
822                }
823        }
824
825        return usage;
826}
827
828/*
829 * Caller has made sure that the swap device corresponding to entry
830 * is still around or has not been recycled.
831 */
832void swap_free(swp_entry_t entry)
833{
834        struct swap_info_struct *p;
835
836        p = swap_info_get(entry);
837        if (p) {
838                swap_entry_free(p, entry, 1);
839                spin_unlock(&p->lock);
840        }
841}
842
843/*
844 * Called after dropping swapcache to decrease refcnt to swap entries.
845 */
846void swapcache_free(swp_entry_t entry)
847{
848        struct swap_info_struct *p;
849
850        p = swap_info_get(entry);
851        if (p) {
852                swap_entry_free(p, entry, SWAP_HAS_CACHE);
853                spin_unlock(&p->lock);
854        }
855}
856
857/*
858 * How many references to page are currently swapped out?
859 * This does not give an exact answer when swap count is continued,
860 * but does include the high COUNT_CONTINUED flag to allow for that.
861 */
862int page_swapcount(struct page *page)
863{
864        int count = 0;
865        struct swap_info_struct *p;
866        swp_entry_t entry;
867
868        entry.val = page_private(page);
869        p = swap_info_get(entry);
870        if (p) {
871                count = swap_count(p->swap_map[swp_offset(entry)]);
872                spin_unlock(&p->lock);
873        }
874        return count;
875}
876
877/*
878 * We can write to an anon page without COW if there are no other references
879 * to it.  And as a side-effect, free up its swap: because the old content
880 * on disk will never be read, and seeking back there to write new content
881 * later would only waste time away from clustering.
882 */
883int reuse_swap_page(struct page *page)
884{
885        int count;
886
887        VM_BUG_ON_PAGE(!PageLocked(page), page);
888        if (unlikely(PageKsm(page)))
889                return 0;
890        count = page_mapcount(page);
891        if (count <= 1 && PageSwapCache(page)) {
892                count += page_swapcount(page);
893                if (count == 1 && !PageWriteback(page)) {
894                        delete_from_swap_cache(page);
895                        SetPageDirty(page);
896                }
897        }
898        return count <= 1;
899}
900
901/*
902 * If swap is getting full, or if there are no more mappings of this page,
903 * then try_to_free_swap is called to free its swap space.
904 */
905int try_to_free_swap(struct page *page)
906{
907        VM_BUG_ON_PAGE(!PageLocked(page), page);
908
909        if (!PageSwapCache(page))
910                return 0;
911        if (PageWriteback(page))
912                return 0;
913        if (page_swapcount(page))
914                return 0;
915
916        /*
917         * Once hibernation has begun to create its image of memory,
918         * there's a danger that one of the calls to try_to_free_swap()
919         * - most probably a call from __try_to_reclaim_swap() while
920         * hibernation is allocating its own swap pages for the image,
921         * but conceivably even a call from memory reclaim - will free
922         * the swap from a page which has already been recorded in the
923         * image as a clean swapcache page, and then reuse its swap for
924         * another page of the image.  On waking from hibernation, the
925         * original page might be freed under memory pressure, then
926         * later read back in from swap, now with the wrong data.
927         *
928         * Hibernation suspends storage while it is writing the image
929         * to disk so check that here.
930         */
931        if (pm_suspended_storage())
932                return 0;
933
934        delete_from_swap_cache(page);
935        SetPageDirty(page);
936        return 1;
937}
938
939/*
940 * Free the swap entry like above, but also try to
941 * free the page cache entry if it is the last user.
942 */
943int free_swap_and_cache(swp_entry_t entry)
944{
945        struct swap_info_struct *p;
946        struct page *page = NULL;
947
948        if (non_swap_entry(entry))
949                return 1;
950
951        p = swap_info_get(entry);
952        if (p) {
953                if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) {
954                        page = find_get_page(swap_address_space(entry),
955                                                entry.val);
956                        if (page && !trylock_page(page)) {
957                                page_cache_release(page);
958                                page = NULL;
959                        }
960                }
961                spin_unlock(&p->lock);
962        }
963        if (page) {
964                /*
965                 * Not mapped elsewhere, or swap space full? Free it!
966                 * Also recheck PageSwapCache now page is locked (above).
967                 */
968                if (PageSwapCache(page) && !PageWriteback(page) &&
969                                (!page_mapped(page) || vm_swap_full())) {
970                        delete_from_swap_cache(page);
971                        SetPageDirty(page);
972                }
973                unlock_page(page);
974                page_cache_release(page);
975        }
976        return p != NULL;
977}
978
979#ifdef CONFIG_HIBERNATION
980/*
981 * Find the swap type that corresponds to given device (if any).
982 *
983 * @offset - number of the PAGE_SIZE-sized block of the device, starting
984 * from 0, in which the swap header is expected to be located.
985 *
986 * This is needed for the suspend to disk (aka swsusp).
987 */
988int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
989{
990        struct block_device *bdev = NULL;
991        int type;
992
993        if (device)
994                bdev = bdget(device);
995
996        spin_lock(&swap_lock);
997        for (type = 0; type < nr_swapfiles; type++) {
998                struct swap_info_struct *sis = swap_info[type];
999
1000                if (!(sis->flags & SWP_WRITEOK))
1001                        continue;
1002
1003                if (!bdev) {
1004                        if (bdev_p)
1005                                *bdev_p = bdgrab(sis->bdev);
1006
1007                        spin_unlock(&swap_lock);
1008                        return type;
1009                }
1010                if (bdev == sis->bdev) {
1011                        struct swap_extent *se = &sis->first_swap_extent;
1012
1013                        if (se->start_block == offset) {
1014                                if (bdev_p)
1015                                        *bdev_p = bdgrab(sis->bdev);
1016
1017                                spin_unlock(&swap_lock);
1018                                bdput(bdev);
1019                                return type;
1020                        }
1021                }
1022        }
1023        spin_unlock(&swap_lock);
1024        if (bdev)
1025                bdput(bdev);
1026
1027        return -ENODEV;
1028}
1029
1030/*
1031 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1032 * corresponding to given index in swap_info (swap type).
1033 */
1034sector_t swapdev_block(int type, pgoff_t offset)
1035{
1036        struct block_device *bdev;
1037
1038        if ((unsigned int)type >= nr_swapfiles)
1039                return 0;
1040        if (!(swap_info[type]->flags & SWP_WRITEOK))
1041                return 0;
1042        return map_swap_entry(swp_entry(type, offset), &bdev);
1043}
1044
1045/*
1046 * Return either the total number of swap pages of given type, or the number
1047 * of free pages of that type (depending on @free)
1048 *
1049 * This is needed for software suspend
1050 */
1051unsigned int count_swap_pages(int type, int free)
1052{
1053        unsigned int n = 0;
1054
1055        spin_lock(&swap_lock);
1056        if ((unsigned int)type < nr_swapfiles) {
1057                struct swap_info_struct *sis = swap_info[type];
1058
1059                spin_lock(&sis->lock);
1060                if (sis->flags & SWP_WRITEOK) {
1061                        n = sis->pages;
1062                        if (free)
1063                                n -= sis->inuse_pages;
1064                }
1065                spin_unlock(&sis->lock);
1066        }
1067        spin_unlock(&swap_lock);
1068        return n;
1069}
1070#endif /* CONFIG_HIBERNATION */
1071
1072static inline int maybe_same_pte(pte_t pte, pte_t swp_pte)
1073{
1074#ifdef CONFIG_MEM_SOFT_DIRTY
1075        /*
1076         * When pte keeps soft dirty bit the pte generated
1077         * from swap entry does not has it, still it's same
1078         * pte from logical point of view.
1079         */
1080        pte_t swp_pte_dirty = pte_swp_mksoft_dirty(swp_pte);
1081        return pte_same(pte, swp_pte) || pte_same(pte, swp_pte_dirty);
1082#else
1083        return pte_same(pte, swp_pte);
1084#endif
1085}
1086
1087/*
1088 * No need to decide whether this PTE shares the swap entry with others,
1089 * just let do_wp_page work it out if a write is requested later - to
1090 * force COW, vm_page_prot omits write permission from any private vma.
1091 */
1092static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
1093                unsigned long addr, swp_entry_t entry, struct page *page)
1094{
1095        struct page *swapcache;
1096        struct mem_cgroup *memcg;
1097        spinlock_t *ptl;
1098        pte_t *pte;
1099        int ret = 1;
1100
1101        swapcache = page;
1102        page = ksm_might_need_to_copy(page, vma, addr);
1103        if (unlikely(!page))
1104                return -ENOMEM;
1105
1106        if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL, &memcg)) {
1107                ret = -ENOMEM;
1108                goto out_nolock;
1109        }
1110
1111        pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1112        if (unlikely(!maybe_same_pte(*pte, swp_entry_to_pte(entry)))) {
1113                mem_cgroup_cancel_charge(page, memcg);
1114                ret = 0;
1115                goto out;
1116        }
1117
1118        dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1119        inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
1120        get_page(page);
1121        set_pte_at(vma->vm_mm, addr, pte,
1122                   pte_mkold(mk_pte(page, vma->vm_page_prot)));
1123        if (page == swapcache) {
1124                page_add_anon_rmap(page, vma, addr);
1125                mem_cgroup_commit_charge(page, memcg, true);
1126        } else { /* ksm created a completely new copy */
1127                page_add_new_anon_rmap(page, vma, addr);
1128                mem_cgroup_commit_charge(page, memcg, false);
1129                lru_cache_add_active_or_unevictable(page, vma);
1130        }
1131        swap_free(entry);
1132        /*
1133         * Move the page to the active list so it is not
1134         * immediately swapped out again after swapon.
1135         */
1136        activate_page(page);
1137out:
1138        pte_unmap_unlock(pte, ptl);
1139out_nolock:
1140        if (page != swapcache) {
1141                unlock_page(page);
1142                put_page(page);
1143        }
1144        return ret;
1145}
1146
1147static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1148                                unsigned long addr, unsigned long end,
1149                                swp_entry_t entry, struct page *page)
1150{
1151        pte_t swp_pte = swp_entry_to_pte(entry);
1152        pte_t *pte;
1153        int ret = 0;
1154
1155        /*
1156         * We don't actually need pte lock while scanning for swp_pte: since
1157         * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
1158         * page table while we're scanning; though it could get zapped, and on
1159         * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
1160         * of unmatched parts which look like swp_pte, so unuse_pte must
1161         * recheck under pte lock.  Scanning without pte lock lets it be
1162         * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
1163         */
1164        pte = pte_offset_map(pmd, addr);
1165        do {
1166                /*
1167                 * swapoff spends a _lot_ of time in this loop!
1168                 * Test inline before going to call unuse_pte.
1169                 */
1170                if (unlikely(maybe_same_pte(*pte, swp_pte))) {
1171                        pte_unmap(pte);
1172                        ret = unuse_pte(vma, pmd, addr, entry, page);
1173                        if (ret)
1174                                goto out;
1175                        pte = pte_offset_map(pmd, addr);
1176                }
1177        } while (pte++, addr += PAGE_SIZE, addr != end);
1178        pte_unmap(pte - 1);
1179out:
1180        return ret;
1181}
1182
1183static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1184                                unsigned long addr, unsigned long end,
1185                                swp_entry_t entry, struct page *page)
1186{
1187        pmd_t *pmd;
1188        unsigned long next;
1189        int ret;
1190
1191        pmd = pmd_offset(pud, addr);
1192        do {
1193                next = pmd_addr_end(addr, end);
1194                if (pmd_none_or_trans_huge_or_clear_bad(pmd))
1195                        continue;
1196                ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
1197                if (ret)
1198                        return ret;
1199        } while (pmd++, addr = next, addr != end);
1200        return 0;
1201}
1202
1203static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
1204                                unsigned long addr, unsigned long end,
1205                                swp_entry_t entry, struct page *page)
1206{
1207        pud_t *pud;
1208        unsigned long next;
1209        int ret;
1210
1211        pud = pud_offset(pgd, addr);
1212        do {
1213                next = pud_addr_end(addr, end);
1214                if (pud_none_or_clear_bad(pud))
1215                        continue;
1216                ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
1217                if (ret)
1218                        return ret;
1219        } while (pud++, addr = next, addr != end);
1220        return 0;
1221}
1222
1223static int unuse_vma(struct vm_area_struct *vma,
1224                                swp_entry_t entry, struct page *page)
1225{
1226        pgd_t *pgd;
1227        unsigned long addr, end, next;
1228        int ret;
1229
1230        if (page_anon_vma(page)) {
1231                addr = page_address_in_vma(page, vma);
1232                if (addr == -EFAULT)
1233                        return 0;
1234                else
1235                        end = addr + PAGE_SIZE;
1236        } else {
1237                addr = vma->vm_start;
1238                end = vma->vm_end;
1239        }
1240
1241        pgd = pgd_offset(vma->vm_mm, addr);
1242        do {
1243                next = pgd_addr_end(addr, end);
1244                if (pgd_none_or_clear_bad(pgd))
1245                        continue;
1246                ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
1247                if (ret)
1248                        return ret;
1249        } while (pgd++, addr = next, addr != end);
1250        return 0;
1251}
1252
1253static int unuse_mm(struct mm_struct *mm,
1254                                swp_entry_t entry, struct page *page)
1255{
1256        struct vm_area_struct *vma;
1257        int ret = 0;
1258
1259        if (!down_read_trylock(&mm->mmap_sem)) {
1260                /*
1261                 * Activate page so shrink_inactive_list is unlikely to unmap
1262                 * its ptes while lock is dropped, so swapoff can make progress.
1263                 */
1264                activate_page(page);
1265                unlock_page(page);
1266                down_read(&mm->mmap_sem);
1267                lock_page(page);
1268        }
1269        for (vma = mm->mmap; vma; vma = vma->vm_next) {
1270                if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
1271                        break;
1272        }
1273        up_read(&mm->mmap_sem);
1274        return (ret < 0)? ret: 0;
1275}
1276
1277/*
1278 * Scan swap_map (or frontswap_map if frontswap parameter is true)
1279 * from current position to next entry still in use.
1280 * Recycle to start on reaching the end, returning 0 when empty.
1281 */
1282static unsigned int find_next_to_unuse(struct swap_info_struct *si,
1283                                        unsigned int prev, bool frontswap)
1284{
1285        unsigned int max = si->max;
1286        unsigned int i = prev;
1287        unsigned char count;
1288
1289        /*
1290         * No need for swap_lock here: we're just looking
1291         * for whether an entry is in use, not modifying it; false
1292         * hits are okay, and sys_swapoff() has already prevented new
1293         * allocations from this area (while holding swap_lock).
1294         */
1295        for (;;) {
1296                if (++i >= max) {
1297                        if (!prev) {
1298                                i = 0;
1299                                break;
1300                        }
1301                        /*
1302                         * No entries in use at top of swap_map,
1303                         * loop back to start and recheck there.
1304                         */
1305                        max = prev + 1;
1306                        prev = 0;
1307                        i = 1;
1308                }
1309                if (frontswap) {
1310                        if (frontswap_test(si, i))
1311                                break;
1312                        else
1313                                continue;
1314                }
1315                count = ACCESS_ONCE(si->swap_map[i]);
1316                if (count && swap_count(count) != SWAP_MAP_BAD)
1317                        break;
1318        }
1319        return i;
1320}
1321
1322/*
1323 * We completely avoid races by reading each swap page in advance,
1324 * and then search for the process using it.  All the necessary
1325 * page table adjustments can then be made atomically.
1326 *
1327 * if the boolean frontswap is true, only unuse pages_to_unuse pages;
1328 * pages_to_unuse==0 means all pages; ignored if frontswap is false
1329 */
1330int try_to_unuse(unsigned int type, bool frontswap,
1331                 unsigned long pages_to_unuse)
1332{
1333        struct swap_info_struct *si = swap_info[type];
1334        struct mm_struct *start_mm;
1335        volatile unsigned char *swap_map; /* swap_map is accessed without
1336                                           * locking. Mark it as volatile
1337                                           * to prevent compiler doing
1338                                           * something odd.
1339                                           */
1340        unsigned char swcount;
1341        struct page *page;
1342        swp_entry_t entry;
1343        unsigned int i = 0;
1344        int retval = 0;
1345
1346        /*
1347         * When searching mms for an entry, a good strategy is to
1348         * start at the first mm we freed the previous entry from
1349         * (though actually we don't notice whether we or coincidence
1350         * freed the entry).  Initialize this start_mm with a hold.
1351         *
1352         * A simpler strategy would be to start at the last mm we
1353         * freed the previous entry from; but that would take less
1354         * advantage of mmlist ordering, which clusters forked mms
1355         * together, child after parent.  If we race with dup_mmap(), we
1356         * prefer to resolve parent before child, lest we miss entries
1357         * duplicated after we scanned child: using last mm would invert
1358         * that.
1359         */
1360        start_mm = &init_mm;
1361        atomic_inc(&init_mm.mm_users);
1362
1363        /*
1364         * Keep on scanning until all entries have gone.  Usually,
1365         * one pass through swap_map is enough, but not necessarily:
1366         * there are races when an instance of an entry might be missed.
1367         */
1368        while ((i = find_next_to_unuse(si, i, frontswap)) != 0) {
1369                if (signal_pending(current)) {
1370                        retval = -EINTR;
1371                        break;
1372                }
1373
1374                /*
1375                 * Get a page for the entry, using the existing swap
1376                 * cache page if there is one.  Otherwise, get a clean
1377                 * page and read the swap into it.
1378                 */
1379                swap_map = &si->swap_map[i];
1380                entry = swp_entry(type, i);
1381                page = read_swap_cache_async(entry,
1382                                        GFP_HIGHUSER_MOVABLE, NULL, 0);
1383                if (!page) {
1384                        /*
1385                         * Either swap_duplicate() failed because entry
1386                         * has been freed independently, and will not be
1387                         * reused since sys_swapoff() already disabled
1388                         * allocation from here, or alloc_page() failed.
1389                         */
1390                        swcount = *swap_map;
1391                        /*
1392                         * We don't hold lock here, so the swap entry could be
1393                         * SWAP_MAP_BAD (when the cluster is discarding).
1394                         * Instead of fail out, We can just skip the swap
1395                         * entry because swapoff will wait for discarding
1396                         * finish anyway.
1397                         */
1398                        if (!swcount || swcount == SWAP_MAP_BAD)
1399                                continue;
1400                        retval = -ENOMEM;
1401                        break;
1402                }
1403
1404                /*
1405                 * Don't hold on to start_mm if it looks like exiting.
1406                 */
1407                if (atomic_read(&start_mm->mm_users) == 1) {
1408                        mmput(start_mm);
1409                        start_mm = &init_mm;
1410                        atomic_inc(&init_mm.mm_users);
1411                }
1412
1413                /*
1414                 * Wait for and lock page.  When do_swap_page races with
1415                 * try_to_unuse, do_swap_page can handle the fault much
1416                 * faster than try_to_unuse can locate the entry.  This
1417                 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1418                 * defer to do_swap_page in such a case - in some tests,
1419                 * do_swap_page and try_to_unuse repeatedly compete.
1420                 */
1421                wait_on_page_locked(page);
1422                wait_on_page_writeback(page);
1423                lock_page(page);
1424                wait_on_page_writeback(page);
1425
1426                /*
1427                 * Remove all references to entry.
1428                 */
1429                swcount = *swap_map;
1430                if (swap_count(swcount) == SWAP_MAP_SHMEM) {
1431                        retval = shmem_unuse(entry, page);
1432                        /* page has already been unlocked and released */
1433                        if (retval < 0)
1434                                break;
1435                        continue;
1436                }
1437                if (swap_count(swcount) && start_mm != &init_mm)
1438                        retval = unuse_mm(start_mm, entry, page);
1439
1440                if (swap_count(*swap_map)) {
1441                        int set_start_mm = (*swap_map >= swcount);
1442                        struct list_head *p = &start_mm->mmlist;
1443                        struct mm_struct *new_start_mm = start_mm;
1444                        struct mm_struct *prev_mm = start_mm;
1445                        struct mm_struct *mm;
1446
1447                        atomic_inc(&new_start_mm->mm_users);
1448                        atomic_inc(&prev_mm->mm_users);
1449                        spin_lock(&mmlist_lock);
1450                        while (swap_count(*swap_map) && !retval &&
1451                                        (p = p->next) != &start_mm->mmlist) {
1452                                mm = list_entry(p, struct mm_struct, mmlist);
1453                                if (!atomic_inc_not_zero(&mm->mm_users))
1454                                        continue;
1455                                spin_unlock(&mmlist_lock);
1456                                mmput(prev_mm);
1457                                prev_mm = mm;
1458
1459                                cond_resched();
1460
1461                                swcount = *swap_map;
1462                                if (!swap_count(swcount)) /* any usage ? */
1463                                        ;
1464                                else if (mm == &init_mm)
1465                                        set_start_mm = 1;
1466                                else
1467                                        retval = unuse_mm(mm, entry, page);
1468
1469                                if (set_start_mm && *swap_map < swcount) {
1470                                        mmput(new_start_mm);
1471                                        atomic_inc(&mm->mm_users);
1472                                        new_start_mm = mm;
1473                                        set_start_mm = 0;
1474                                }
1475                                spin_lock(&mmlist_lock);
1476                        }
1477                        spin_unlock(&mmlist_lock);
1478                        mmput(prev_mm);
1479                        mmput(start_mm);
1480                        start_mm = new_start_mm;
1481                }
1482                if (retval) {
1483                        unlock_page(page);
1484                        page_cache_release(page);
1485                        break;
1486                }
1487
1488                /*
1489                 * If a reference remains (rare), we would like to leave
1490                 * the page in the swap cache; but try_to_unmap could
1491                 * then re-duplicate the entry once we drop page lock,
1492                 * so we might loop indefinitely; also, that page could
1493                 * not be swapped out to other storage meanwhile.  So:
1494                 * delete from cache even if there's another reference,
1495                 * after ensuring that the data has been saved to disk -
1496                 * since if the reference remains (rarer), it will be
1497                 * read from disk into another page.  Splitting into two
1498                 * pages would be incorrect if swap supported "shared
1499                 * private" pages, but they are handled by tmpfs files.
1500                 *
1501                 * Given how unuse_vma() targets one particular offset
1502                 * in an anon_vma, once the anon_vma has been determined,
1503                 * this splitting happens to be just what is needed to
1504                 * handle where KSM pages have been swapped out: re-reading
1505                 * is unnecessarily slow, but we can fix that later on.
1506                 */
1507                if (swap_count(*swap_map) &&
1508                     PageDirty(page) && PageSwapCache(page)) {
1509                        struct writeback_control wbc = {
1510                                .sync_mode = WB_SYNC_NONE,
1511                        };
1512
1513                        swap_writepage(page, &wbc);
1514                        lock_page(page);
1515                        wait_on_page_writeback(page);
1516                }
1517
1518                /*
1519                 * It is conceivable that a racing task removed this page from
1520                 * swap cache just before we acquired the page lock at the top,
1521                 * or while we dropped it in unuse_mm().  The page might even
1522                 * be back in swap cache on another swap area: that we must not
1523                 * delete, since it may not have been written out to swap yet.
1524                 */
1525                if (PageSwapCache(page) &&
1526                    likely(page_private(page) == entry.val))
1527                        delete_from_swap_cache(page);
1528
1529                /*
1530                 * So we could skip searching mms once swap count went
1531                 * to 1, we did not mark any present ptes as dirty: must
1532                 * mark page dirty so shrink_page_list will preserve it.
1533                 */
1534                SetPageDirty(page);
1535                unlock_page(page);
1536                page_cache_release(page);
1537
1538                /*
1539                 * Make sure that we aren't completely killing
1540                 * interactive performance.
1541                 */
1542                cond_resched();
1543                if (frontswap && pages_to_unuse > 0) {
1544                        if (!--pages_to_unuse)
1545                                break;
1546                }
1547        }
1548
1549        mmput(start_mm);
1550        return retval;
1551}
1552
1553/*
1554 * After a successful try_to_unuse, if no swap is now in use, we know
1555 * we can empty the mmlist.  swap_lock must be held on entry and exit.
1556 * Note that mmlist_lock nests inside swap_lock, and an mm must be
1557 * added to the mmlist just after page_duplicate - before would be racy.
1558 */
1559static void drain_mmlist(void)
1560{
1561        struct list_head *p, *next;
1562        unsigned int type;
1563
1564        for (type = 0; type < nr_swapfiles; type++)
1565                if (swap_info[type]->inuse_pages)
1566                        return;
1567        spin_lock(&mmlist_lock);
1568        list_for_each_safe(p, next, &init_mm.mmlist)
1569                list_del_init(p);
1570        spin_unlock(&mmlist_lock);
1571}
1572
1573/*
1574 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
1575 * corresponds to page offset for the specified swap entry.
1576 * Note that the type of this function is sector_t, but it returns page offset
1577 * into the bdev, not sector offset.
1578 */
1579static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
1580{
1581        struct swap_info_struct *sis;
1582        struct swap_extent *start_se;
1583        struct swap_extent *se;
1584        pgoff_t offset;
1585
1586        sis = swap_info[swp_type(entry)];
1587        *bdev = sis->bdev;
1588
1589        offset = swp_offset(entry);
1590        start_se = sis->curr_swap_extent;
1591        se = start_se;
1592
1593        for ( ; ; ) {
1594                struct list_head *lh;
1595
1596                if (se->start_page <= offset &&
1597                                offset < (se->start_page + se->nr_pages)) {
1598                        return se->start_block + (offset - se->start_page);
1599                }
1600                lh = se->list.next;
1601                se = list_entry(lh, struct swap_extent, list);
1602                sis->curr_swap_extent = se;
1603                BUG_ON(se == start_se);         /* It *must* be present */
1604        }
1605}
1606
1607/*
1608 * Returns the page offset into bdev for the specified page's swap entry.
1609 */
1610sector_t map_swap_page(struct page *page, struct block_device **bdev)
1611{
1612        swp_entry_t entry;
1613        entry.val = page_private(page);
1614        return map_swap_entry(entry, bdev);
1615}
1616
1617/*
1618 * Free all of a swapdev's extent information
1619 */
1620static void destroy_swap_extents(struct swap_info_struct *sis)
1621{
1622        while (!list_empty(&sis->first_swap_extent.list)) {
1623                struct swap_extent *se;
1624
1625                se = list_entry(sis->first_swap_extent.list.next,
1626                                struct swap_extent, list);
1627                list_del(&se->list);
1628                kfree(se);
1629        }
1630
1631        if (sis->flags & SWP_FILE) {
1632                struct file *swap_file = sis->swap_file;
1633                struct address_space *mapping = swap_file->f_mapping;
1634
1635                sis->flags &= ~SWP_FILE;
1636                mapping->a_ops->swap_deactivate(swap_file);
1637        }
1638}
1639
1640/*
1641 * Add a block range (and the corresponding page range) into this swapdev's
1642 * extent list.  The extent list is kept sorted in page order.
1643 *
1644 * This function rather assumes that it is called in ascending page order.
1645 */
1646int
1647add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1648                unsigned long nr_pages, sector_t start_block)
1649{
1650        struct swap_extent *se;
1651        struct swap_extent *new_se;
1652        struct list_head *lh;
1653
1654        if (start_page == 0) {
1655                se = &sis->first_swap_extent;
1656                sis->curr_swap_extent = se;
1657                se->start_page = 0;
1658                se->nr_pages = nr_pages;
1659                se->start_block = start_block;
1660                return 1;
1661        } else {
1662                lh = sis->first_swap_extent.list.prev;  /* Highest extent */
1663                se = list_entry(lh, struct swap_extent, list);
1664                BUG_ON(se->start_page + se->nr_pages != start_page);
1665                if (se->start_block + se->nr_pages == start_block) {
1666                        /* Merge it */
1667                        se->nr_pages += nr_pages;
1668                        return 0;
1669                }
1670        }
1671
1672        /*
1673         * No merge.  Insert a new extent, preserving ordering.
1674         */
1675        new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1676        if (new_se == NULL)
1677                return -ENOMEM;
1678        new_se->start_page = start_page;
1679        new_se->nr_pages = nr_pages;
1680        new_se->start_block = start_block;
1681
1682        list_add_tail(&new_se->list, &sis->first_swap_extent.list);
1683        return 1;
1684}
1685
1686/*
1687 * A `swap extent' is a simple thing which maps a contiguous range of pages
1688 * onto a contiguous range of disk blocks.  An ordered list of swap extents
1689 * is built at swapon time and is then used at swap_writepage/swap_readpage
1690 * time for locating where on disk a page belongs.
1691 *
1692 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1693 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1694 * swap files identically.
1695 *
1696 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1697 * extent list operates in PAGE_SIZE disk blocks.  Both S_ISREG and S_ISBLK
1698 * swapfiles are handled *identically* after swapon time.
1699 *
1700 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1701 * and will parse them into an ordered extent list, in PAGE_SIZE chunks.  If
1702 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1703 * requirements, they are simply tossed out - we will never use those blocks
1704 * for swapping.
1705 *
1706 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon.  This
1707 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1708 * which will scribble on the fs.
1709 *
1710 * The amount of disk space which a single swap extent represents varies.
1711 * Typically it is in the 1-4 megabyte range.  So we can have hundreds of
1712 * extents in the list.  To avoid much list walking, we cache the previous
1713 * search location in `curr_swap_extent', and start new searches from there.
1714 * This is extremely effective.  The average number of iterations in
1715 * map_swap_page() has been measured at about 0.3 per page.  - akpm.
1716 */
1717static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
1718{
1719        struct file *swap_file = sis->swap_file;
1720        struct address_space *mapping = swap_file->f_mapping;
1721        struct inode *inode = mapping->host;
1722        int ret;
1723
1724        if (S_ISBLK(inode->i_mode)) {
1725                ret = add_swap_extent(sis, 0, sis->max, 0);
1726                *span = sis->pages;
1727                return ret;
1728        }
1729
1730        if (mapping->a_ops->swap_activate) {
1731                ret = mapping->a_ops->swap_activate(sis, swap_file, span);
1732                if (!ret) {
1733                        sis->flags |= SWP_FILE;
1734                        ret = add_swap_extent(sis, 0, sis->max, 0);
1735                        *span = sis->pages;
1736                }
1737                return ret;
1738        }
1739
1740        return generic_swapfile_activate(sis, swap_file, span);
1741}
1742
1743static void _enable_swap_info(struct swap_info_struct *p, int prio,
1744                                unsigned char *swap_map,
1745                                struct swap_cluster_info *cluster_info)
1746{
1747        if (prio >= 0)
1748                p->prio = prio;
1749        else
1750                p->prio = --least_priority;
1751        /*
1752         * the plist prio is negated because plist ordering is
1753         * low-to-high, while swap ordering is high-to-low
1754         */
1755        p->list.prio = -p->prio;
1756        p->avail_list.prio = -p->prio;
1757        p->swap_map = swap_map;
1758        p->cluster_info = cluster_info;
1759        p->flags |= SWP_WRITEOK;
1760        atomic_long_add(p->pages, &nr_swap_pages);
1761        total_swap_pages += p->pages;
1762
1763        assert_spin_locked(&swap_lock);
1764        /*
1765         * both lists are plists, and thus priority ordered.
1766         * swap_active_head needs to be priority ordered for swapoff(),
1767         * which on removal of any swap_info_struct with an auto-assigned
1768         * (i.e. negative) priority increments the auto-assigned priority
1769         * of any lower-priority swap_info_structs.
1770         * swap_avail_head needs to be priority ordered for get_swap_page(),
1771         * which allocates swap pages from the highest available priority
1772         * swap_info_struct.
1773         */
1774        plist_add(&p->list, &swap_active_head);
1775        spin_lock(&swap_avail_lock);
1776        plist_add(&p->avail_list, &swap_avail_head);
1777        spin_unlock(&swap_avail_lock);
1778}
1779
1780static void enable_swap_info(struct swap_info_struct *p, int prio,
1781                                unsigned char *swap_map,
1782                                struct swap_cluster_info *cluster_info,
1783                                unsigned long *frontswap_map)
1784{
1785        frontswap_init(p->type, frontswap_map);
1786        spin_lock(&swap_lock);
1787        spin_lock(&p->lock);
1788         _enable_swap_info(p, prio, swap_map, cluster_info);
1789        spin_unlock(&p->lock);
1790        spin_unlock(&swap_lock);
1791}
1792
1793static void reinsert_swap_info(struct swap_info_struct *p)
1794{
1795        spin_lock(&swap_lock);
1796        spin_lock(&p->lock);
1797        _enable_swap_info(p, p->prio, p->swap_map, p->cluster_info);
1798        spin_unlock(&p->lock);
1799        spin_unlock(&swap_lock);
1800}
1801
1802SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
1803{
1804        struct swap_info_struct *p = NULL;
1805        unsigned char *swap_map;
1806        struct swap_cluster_info *cluster_info;
1807        unsigned long *frontswap_map;
1808        struct file *swap_file, *victim;
1809        struct address_space *mapping;
1810        struct inode *inode;
1811        struct filename *pathname;
1812        int err, found = 0;
1813        unsigned int old_block_size;
1814
1815        if (!capable(CAP_SYS_ADMIN))
1816                return -EPERM;
1817
1818        BUG_ON(!current->mm);
1819
1820        pathname = getname(specialfile);
1821        if (IS_ERR(pathname))
1822                return PTR_ERR(pathname);
1823
1824        victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
1825        err = PTR_ERR(victim);
1826        if (IS_ERR(victim))
1827                goto out;
1828
1829        mapping = victim->f_mapping;
1830        spin_lock(&swap_lock);
1831        plist_for_each_entry(p, &swap_active_head, list) {
1832                if (p->flags & SWP_WRITEOK) {
1833                        if (p->swap_file->f_mapping == mapping) {
1834                                found = 1;
1835                                break;
1836                        }
1837                }
1838        }
1839        if (!found) {
1840                err = -EINVAL;
1841                spin_unlock(&swap_lock);
1842                goto out_dput;
1843        }
1844        if (!security_vm_enough_memory_mm(current->mm, p->pages))
1845                vm_unacct_memory(p->pages);
1846        else {
1847                err = -ENOMEM;
1848                spin_unlock(&swap_lock);
1849                goto out_dput;
1850        }
1851        spin_lock(&swap_avail_lock);
1852        plist_del(&p->avail_list, &swap_avail_head);
1853        spin_unlock(&swap_avail_lock);
1854        spin_lock(&p->lock);
1855        if (p->prio < 0) {
1856                struct swap_info_struct *si = p;
1857
1858                plist_for_each_entry_continue(si, &swap_active_head, list) {
1859                        si->prio++;
1860                        si->list.prio--;
1861                        si->avail_list.prio--;
1862                }
1863                least_priority++;
1864        }
1865        plist_del(&p->list, &swap_active_head);
1866        atomic_long_sub(p->pages, &nr_swap_pages);
1867        total_swap_pages -= p->pages;
1868        p->flags &= ~SWP_WRITEOK;
1869        spin_unlock(&p->lock);
1870        spin_unlock(&swap_lock);
1871
1872        set_current_oom_origin();
1873        err = try_to_unuse(p->type, false, 0); /* force unuse all pages */
1874        clear_current_oom_origin();
1875
1876        if (err) {
1877                /* re-insert swap space back into swap_list */
1878                reinsert_swap_info(p);
1879                goto out_dput;
1880        }
1881
1882        flush_work(&p->discard_work);
1883
1884        destroy_swap_extents(p);
1885        if (p->flags & SWP_CONTINUED)
1886                free_swap_count_continuations(p);
1887
1888        mutex_lock(&swapon_mutex);
1889        spin_lock(&swap_lock);
1890        spin_lock(&p->lock);
1891        drain_mmlist();
1892
1893        /* wait for anyone still in scan_swap_map */
1894        p->highest_bit = 0;             /* cuts scans short */
1895        while (p->flags >= SWP_SCANNING) {
1896                spin_unlock(&p->lock);
1897                spin_unlock(&swap_lock);
1898                schedule_timeout_uninterruptible(1);
1899                spin_lock(&swap_lock);
1900                spin_lock(&p->lock);
1901        }
1902
1903        swap_file = p->swap_file;
1904        old_block_size = p->old_block_size;
1905        p->swap_file = NULL;
1906        p->max = 0;
1907        swap_map = p->swap_map;
1908        p->swap_map = NULL;
1909        cluster_info = p->cluster_info;
1910        p->cluster_info = NULL;
1911        frontswap_map = frontswap_map_get(p);
1912        spin_unlock(&p->lock);
1913        spin_unlock(&swap_lock);
1914        frontswap_invalidate_area(p->type);
1915        frontswap_map_set(p, NULL);
1916        mutex_unlock(&swapon_mutex);
1917        free_percpu(p->percpu_cluster);
1918        p->percpu_cluster = NULL;
1919        vfree(swap_map);
1920        vfree(cluster_info);
1921        vfree(frontswap_map);
1922        /* Destroy swap account information */
1923        swap_cgroup_swapoff(p->type);
1924
1925        inode = mapping->host;
1926        if (S_ISBLK(inode->i_mode)) {
1927                struct block_device *bdev = I_BDEV(inode);
1928                set_blocksize(bdev, old_block_size);
1929                blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1930        } else {
1931                mutex_lock(&inode->i_mutex);
1932                inode->i_flags &= ~S_SWAPFILE;
1933                mutex_unlock(&inode->i_mutex);
1934        }
1935        filp_close(swap_file, NULL);
1936
1937        /*
1938         * Clear the SWP_USED flag after all resources are freed so that swapon
1939         * can reuse this swap_info in alloc_swap_info() safely.  It is ok to
1940         * not hold p->lock after we cleared its SWP_WRITEOK.
1941         */
1942        spin_lock(&swap_lock);
1943        p->flags = 0;
1944        spin_unlock(&swap_lock);
1945
1946        err = 0;
1947        atomic_inc(&proc_poll_event);
1948        wake_up_interruptible(&proc_poll_wait);
1949
1950out_dput:
1951        filp_close(victim, NULL);
1952out:
1953        putname(pathname);
1954        return err;
1955}
1956
1957#ifdef CONFIG_PROC_FS
1958static unsigned swaps_poll(struct file *file, poll_table *wait)
1959{
1960        struct seq_file *seq = file->private_data;
1961
1962        poll_wait(file, &proc_poll_wait, wait);
1963
1964        if (seq->poll_event != atomic_read(&proc_poll_event)) {
1965                seq->poll_event = atomic_read(&proc_poll_event);
1966                return POLLIN | POLLRDNORM | POLLERR | POLLPRI;
1967        }
1968
1969        return POLLIN | POLLRDNORM;
1970}
1971
1972/* iterator */
1973static void *swap_start(struct seq_file *swap, loff_t *pos)
1974{
1975        struct swap_info_struct *si;
1976        int type;
1977        loff_t l = *pos;
1978
1979        mutex_lock(&swapon_mutex);
1980
1981        if (!l)
1982                return SEQ_START_TOKEN;
1983
1984        for (type = 0; type < nr_swapfiles; type++) {
1985                smp_rmb();      /* read nr_swapfiles before swap_info[type] */
1986                si = swap_info[type];
1987                if (!(si->flags & SWP_USED) || !si->swap_map)
1988                        continue;
1989                if (!--l)
1990                        return si;
1991        }
1992
1993        return NULL;
1994}
1995
1996static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1997{
1998        struct swap_info_struct *si = v;
1999        int type;
2000
2001        if (v == SEQ_START_TOKEN)
2002                type = 0;
2003        else
2004                type = si->type + 1;
2005
2006        for (; type < nr_swapfiles; type++) {
2007                smp_rmb();      /* read nr_swapfiles before swap_info[type] */
2008                si = swap_info[type];
2009                if (!(si->flags & SWP_USED) || !si->swap_map)
2010                        continue;
2011                ++*pos;
2012                return si;
2013        }
2014
2015        return NULL;
2016}
2017
2018static void swap_stop(struct seq_file *swap, void *v)
2019{
2020        mutex_unlock(&swapon_mutex);
2021}
2022
2023static int swap_show(struct seq_file *swap, void *v)
2024{
2025        struct swap_info_struct *si = v;
2026        struct file *file;
2027        int len;
2028
2029        if (si == SEQ_START_TOKEN) {
2030                seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2031                return 0;
2032        }
2033
2034        file = si->swap_file;
2035        len = seq_path(swap, &file->f_path, " \t\n\\");
2036        seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
2037                        len < 40 ? 40 - len : 1, " ",
2038                        S_ISBLK(file_inode(file)->i_mode) ?
2039                                "partition" : "file\t",
2040                        si->pages << (PAGE_SHIFT - 10),
2041                        si->inuse_pages << (PAGE_SHIFT - 10),
2042                        si->prio);
2043        return 0;
2044}
2045
2046static const struct seq_operations swaps_op = {
2047        .start =        swap_start,
2048        .next =         swap_next,
2049        .stop =         swap_stop,
2050        .show =         swap_show
2051};
2052
2053static int swaps_open(struct inode *inode, struct file *file)
2054{
2055        struct seq_file *seq;
2056        int ret;
2057
2058        ret = seq_open(file, &swaps_op);
2059        if (ret)
2060                return ret;
2061
2062        seq = file->private_data;
2063        seq->poll_event = atomic_read(&proc_poll_event);
2064        return 0;
2065}
2066
2067static const struct file_operations proc_swaps_operations = {
2068        .open           = swaps_open,
2069        .read           = seq_read,
2070        .llseek         = seq_lseek,
2071        .release        = seq_release,
2072        .poll           = swaps_poll,
2073};
2074
2075static int __init procswaps_init(void)
2076{
2077        proc_create("swaps", 0, NULL, &proc_swaps_operations);
2078        return 0;
2079}
2080__initcall(procswaps_init);
2081#endif /* CONFIG_PROC_FS */
2082
2083#ifdef MAX_SWAPFILES_CHECK
2084static int __init max_swapfiles_check(void)
2085{
2086        MAX_SWAPFILES_CHECK();
2087        return 0;
2088}
2089late_initcall(max_swapfiles_check);
2090#endif
2091
2092static struct swap_info_struct *alloc_swap_info(void)
2093{
2094        struct swap_info_struct *p;
2095        unsigned int type;
2096
2097        p = kzalloc(sizeof(*p), GFP_KERNEL);
2098        if (!p)
2099                return ERR_PTR(-ENOMEM);
2100
2101        spin_lock(&swap_lock);
2102        for (type = 0; type < nr_swapfiles; type++) {
2103                if (!(swap_info[type]->flags & SWP_USED))
2104                        break;
2105        }
2106        if (type >= MAX_SWAPFILES) {
2107                spin_unlock(&swap_lock);
2108                kfree(p);
2109                return ERR_PTR(-EPERM);
2110        }
2111        if (type >= nr_swapfiles) {
2112                p->type = type;
2113                swap_info[type] = p;
2114                /*
2115                 * Write swap_info[type] before nr_swapfiles, in case a
2116                 * racing procfs swap_start() or swap_next() is reading them.
2117                 * (We never shrink nr_swapfiles, we never free this entry.)
2118                 */
2119                smp_wmb();
2120                nr_swapfiles++;
2121        } else {
2122                kfree(p);
2123                p = swap_info[type];
2124                /*
2125                 * Do not memset this entry: a racing procfs swap_next()
2126                 * would be relying on p->type to remain valid.
2127                 */
2128        }
2129        INIT_LIST_HEAD(&p->first_swap_extent.list);
2130        plist_node_init(&p->list, 0);
2131        plist_node_init(&p->avail_list, 0);
2132        p->flags = SWP_USED;
2133        spin_unlock(&swap_lock);
2134        spin_lock_init(&p->lock);
2135
2136        return p;
2137}
2138
2139static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2140{
2141        int error;
2142
2143        if (S_ISBLK(inode->i_mode)) {
2144                p->bdev = bdgrab(I_BDEV(inode));
2145                error = blkdev_get(p->bdev,
2146                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL,
2147                                   sys_swapon);
2148                if (error < 0) {
2149                        p->bdev = NULL;
2150                        return -EINVAL;
2151                }
2152                p->old_block_size = block_size(p->bdev);
2153                error = set_blocksize(p->bdev, PAGE_SIZE);
2154                if (error < 0)
2155                        return error;
2156                p->flags |= SWP_BLKDEV;
2157        } else if (S_ISREG(inode->i_mode)) {
2158                p->bdev = inode->i_sb->s_bdev;
2159                mutex_lock(&inode->i_mutex);
2160                if (IS_SWAPFILE(inode))
2161                        return -EBUSY;
2162        } else
2163                return -EINVAL;
2164
2165        return 0;
2166}
2167
2168static unsigned long read_swap_header(struct swap_info_struct *p,
2169                                        union swap_header *swap_header,
2170                                        struct inode *inode)
2171{
2172        int i;
2173        unsigned long maxpages;
2174        unsigned long swapfilepages;
2175        unsigned long last_page;
2176
2177        if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
2178                pr_err("Unable to find swap-space signature\n");
2179                return 0;
2180        }
2181
2182        /* swap partition endianess hack... */
2183        if (swab32(swap_header->info.version) == 1) {
2184                swab32s(&swap_header->info.version);
2185                swab32s(&swap_header->info.last_page);
2186                swab32s(&swap_header->info.nr_badpages);
2187                if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2188                        return 0;
2189                for (i = 0; i < swap_header->info.nr_badpages; i++)
2190                        swab32s(&swap_header->info.badpages[i]);
2191        }
2192        /* Check the swap header's sub-version */
2193        if (swap_header->info.version != 1) {
2194                pr_warn("Unable to handle swap header version %d\n",
2195                        swap_header->info.version);
2196                return 0;
2197        }
2198
2199        p->lowest_bit  = 1;
2200        p->cluster_next = 1;
2201        p->cluster_nr = 0;
2202
2203        /*
2204         * Find out how many pages are allowed for a single swap
2205         * device. There are two limiting factors: 1) the number
2206         * of bits for the swap offset in the swp_entry_t type, and
2207         * 2) the number of bits in the swap pte as defined by the
2208         * different architectures. In order to find the
2209         * largest possible bit mask, a swap entry with swap type 0
2210         * and swap offset ~0UL is created, encoded to a swap pte,
2211         * decoded to a swp_entry_t again, and finally the swap
2212         * offset is extracted. This will mask all the bits from
2213         * the initial ~0UL mask that can't be encoded in either
2214         * the swp_entry_t or the architecture definition of a
2215         * swap pte.
2216         */
2217        maxpages = swp_offset(pte_to_swp_entry(
2218                        swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2219        last_page = swap_header->info.last_page;
2220        if (last_page > maxpages) {
2221                pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
2222                        maxpages << (PAGE_SHIFT - 10),
2223                        last_page << (PAGE_SHIFT - 10));
2224        }
2225        if (maxpages > last_page) {
2226                maxpages = last_page + 1;
2227                /* p->max is an unsigned int: don't overflow it */
2228                if ((unsigned int)maxpages == 0)
2229                        maxpages = UINT_MAX;
2230        }
2231        p->highest_bit = maxpages - 1;
2232
2233        if (!maxpages)
2234                return 0;
2235        swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2236        if (swapfilepages && maxpages > swapfilepages) {
2237                pr_warn("Swap area shorter than signature indicates\n");
2238                return 0;
2239        }
2240        if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
2241                return 0;
2242        if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2243                return 0;
2244
2245        return maxpages;
2246}
2247
2248static int setup_swap_map_and_extents(struct swap_info_struct *p,
2249                                        union swap_header *swap_header,
2250                                        unsigned char *swap_map,
2251                                        struct swap_cluster_info *cluster_info,
2252                                        unsigned long maxpages,
2253                                        sector_t *span)
2254{
2255        int i;
2256        unsigned int nr_good_pages;
2257        int nr_extents;
2258        unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
2259        unsigned long idx = p->cluster_next / SWAPFILE_CLUSTER;
2260
2261        nr_good_pages = maxpages - 1;   /* omit header page */
2262
2263        cluster_set_null(&p->free_cluster_head);
2264        cluster_set_null(&p->free_cluster_tail);
2265        cluster_set_null(&p->discard_cluster_head);
2266        cluster_set_null(&p->discard_cluster_tail);
2267
2268        for (i = 0; i < swap_header->info.nr_badpages; i++) {
2269                unsigned int page_nr = swap_header->info.badpages[i];
2270                if (page_nr == 0 || page_nr > swap_header->info.last_page)
2271                        return -EINVAL;
2272                if (page_nr < maxpages) {
2273                        swap_map[page_nr] = SWAP_MAP_BAD;
2274                        nr_good_pages--;
2275                        /*
2276                         * Haven't marked the cluster free yet, no list
2277                         * operation involved
2278                         */
2279                        inc_cluster_info_page(p, cluster_info, page_nr);
2280                }
2281        }
2282
2283        /* Haven't marked the cluster free yet, no list operation involved */
2284        for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
2285                inc_cluster_info_page(p, cluster_info, i);
2286
2287        if (nr_good_pages) {
2288                swap_map[0] = SWAP_MAP_BAD;
2289                /*
2290                 * Not mark the cluster free yet, no list
2291                 * operation involved
2292                 */
2293                inc_cluster_info_page(p, cluster_info, 0);
2294                p->max = maxpages;
2295                p->pages = nr_good_pages;
2296                nr_extents = setup_swap_extents(p, span);
2297                if (nr_extents < 0)
2298                        return nr_extents;
2299                nr_good_pages = p->pages;
2300        }
2301        if (!nr_good_pages) {
2302                pr_warn("Empty swap-file\n");
2303                return -EINVAL;
2304        }
2305
2306        if (!cluster_info)
2307                return nr_extents;
2308
2309        for (i = 0; i < nr_clusters; i++) {
2310                if (!cluster_count(&cluster_info[idx])) {
2311                        cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
2312                        if (cluster_is_null(&p->free_cluster_head)) {
2313                                cluster_set_next_flag(&p->free_cluster_head,
2314                                                                idx, 0);
2315                                cluster_set_next_flag(&p->free_cluster_tail,
2316                                                                idx, 0);
2317                        } else {
2318                                unsigned int tail;
2319
2320                                tail = cluster_next(&p->free_cluster_tail);
2321                                cluster_set_next(&cluster_info[tail], idx);
2322                                cluster_set_next_flag(&p->free_cluster_tail,
2323                                                                idx, 0);
2324                        }
2325                }
2326                idx++;
2327                if (idx == nr_clusters)
2328                        idx = 0;
2329        }
2330        return nr_extents;
2331}
2332
2333/*
2334 * Helper to sys_swapon determining if a given swap
2335 * backing device queue supports DISCARD operations.
2336 */
2337static bool swap_discardable(struct swap_info_struct *si)
2338{
2339        struct request_queue *q = bdev_get_queue(si->bdev);
2340
2341        if (!q || !blk_queue_discard(q))
2342                return false;
2343
2344        return true;
2345}
2346
2347SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2348{
2349        struct swap_info_struct *p;
2350        struct filename *name;
2351        struct file *swap_file = NULL;
2352        struct address_space *mapping;
2353        int i;
2354        int prio;
2355        int error;
2356        union swap_header *swap_header;
2357        int nr_extents;
2358        sector_t span;
2359        unsigned long maxpages;
2360        unsigned char *swap_map = NULL;
2361        struct swap_cluster_info *cluster_info = NULL;
2362        unsigned long *frontswap_map = NULL;
2363        struct page *page = NULL;
2364        struct inode *inode = NULL;
2365
2366        if (swap_flags & ~SWAP_FLAGS_VALID)
2367                return -EINVAL;
2368
2369        if (!capable(CAP_SYS_ADMIN))
2370                return -EPERM;
2371
2372        p = alloc_swap_info();
2373        if (IS_ERR(p))
2374                return PTR_ERR(p);
2375
2376        INIT_WORK(&p->discard_work, swap_discard_work);
2377
2378        name = getname(specialfile);
2379        if (IS_ERR(name)) {
2380                error = PTR_ERR(name);
2381                name = NULL;
2382                goto bad_swap;
2383        }
2384        swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
2385        if (IS_ERR(swap_file)) {
2386                error = PTR_ERR(swap_file);
2387                swap_file = NULL;
2388                goto bad_swap;
2389        }
2390
2391        p->swap_file = swap_file;
2392        mapping = swap_file->f_mapping;
2393
2394        for (i = 0; i < nr_swapfiles; i++) {
2395                struct swap_info_struct *q = swap_info[i];
2396
2397                if (q == p || !q->swap_file)
2398                        continue;
2399                if (mapping == q->swap_file->f_mapping) {
2400                        error = -EBUSY;
2401                        goto bad_swap;
2402                }
2403        }
2404
2405        inode = mapping->host;
2406        /* If S_ISREG(inode->i_mode) will do mutex_lock(&inode->i_mutex); */
2407        error = claim_swapfile(p, inode);
2408        if (unlikely(error))
2409                goto bad_swap;
2410
2411        /*
2412         * Read the swap header.
2413         */
2414        if (!mapping->a_ops->readpage) {
2415                error = -EINVAL;
2416                goto bad_swap;
2417        }
2418        page = read_mapping_page(mapping, 0, swap_file);
2419        if (IS_ERR(page)) {
2420                error = PTR_ERR(page);
2421                goto bad_swap;
2422        }
2423        swap_header = kmap(page);
2424
2425        maxpages = read_swap_header(p, swap_header, inode);
2426        if (unlikely(!maxpages)) {
2427                error = -EINVAL;
2428                goto bad_swap;
2429        }
2430
2431        /* OK, set up the swap map and apply the bad block list */
2432        swap_map = vzalloc(maxpages);
2433        if (!swap_map) {
2434                error = -ENOMEM;
2435                goto bad_swap;
2436        }
2437        if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) {
2438                p->flags |= SWP_SOLIDSTATE;
2439                /*
2440                 * select a random position to start with to help wear leveling
2441                 * SSD
2442                 */
2443                p->cluster_next = 1 + (prandom_u32() % p->highest_bit);
2444
2445                cluster_info = vzalloc(DIV_ROUND_UP(maxpages,
2446                        SWAPFILE_CLUSTER) * sizeof(*cluster_info));
2447                if (!cluster_info) {
2448                        error = -ENOMEM;
2449                        goto bad_swap;
2450                }
2451                p->percpu_cluster = alloc_percpu(struct percpu_cluster);
2452                if (!p->percpu_cluster) {
2453                        error = -ENOMEM;
2454                        goto bad_swap;
2455                }
2456                for_each_possible_cpu(i) {
2457                        struct percpu_cluster *cluster;
2458                        cluster = per_cpu_ptr(p->percpu_cluster, i);
2459                        cluster_set_null(&cluster->index);
2460                }
2461        }
2462
2463        error = swap_cgroup_swapon(p->type, maxpages);
2464        if (error)
2465                goto bad_swap;
2466
2467        nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
2468                cluster_info, maxpages, &span);
2469        if (unlikely(nr_extents < 0)) {
2470                error = nr_extents;
2471                goto bad_swap;
2472        }
2473        /* frontswap enabled? set up bit-per-page map for frontswap */
2474        if (frontswap_enabled)
2475                frontswap_map = vzalloc(BITS_TO_LONGS(maxpages) * sizeof(long));
2476
2477        if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) {
2478                /*
2479                 * When discard is enabled for swap with no particular
2480                 * policy flagged, we set all swap discard flags here in
2481                 * order to sustain backward compatibility with older
2482                 * swapon(8) releases.
2483                 */
2484                p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
2485                             SWP_PAGE_DISCARD);
2486
2487                /*
2488                 * By flagging sys_swapon, a sysadmin can tell us to
2489                 * either do single-time area discards only, or to just
2490                 * perform discards for released swap page-clusters.
2491                 * Now it's time to adjust the p->flags accordingly.
2492                 */
2493                if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
2494                        p->flags &= ~SWP_PAGE_DISCARD;
2495                else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
2496                        p->flags &= ~SWP_AREA_DISCARD;
2497
2498                /* issue a swapon-time discard if it's still required */
2499                if (p->flags & SWP_AREA_DISCARD) {
2500                        int err = discard_swap(p);
2501                        if (unlikely(err))
2502                                pr_err("swapon: discard_swap(%p): %d\n",
2503                                        p, err);
2504                }
2505        }
2506
2507        mutex_lock(&swapon_mutex);
2508        prio = -1;
2509        if (swap_flags & SWAP_FLAG_PREFER)
2510                prio =
2511                  (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
2512        enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
2513
2514        pr_info("Adding %uk swap on %s.  "
2515                        "Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
2516                p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
2517                nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
2518                (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
2519                (p->flags & SWP_DISCARDABLE) ? "D" : "",
2520                (p->flags & SWP_AREA_DISCARD) ? "s" : "",
2521                (p->flags & SWP_PAGE_DISCARD) ? "c" : "",
2522                (frontswap_map) ? "FS" : "");
2523
2524        mutex_unlock(&swapon_mutex);
2525        atomic_inc(&proc_poll_event);
2526        wake_up_interruptible(&proc_poll_wait);
2527
2528        if (S_ISREG(inode->i_mode))
2529                inode->i_flags |= S_SWAPFILE;
2530        error = 0;
2531        goto out;
2532bad_swap:
2533        free_percpu(p->percpu_cluster);
2534        p->percpu_cluster = NULL;
2535        if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
2536                set_blocksize(p->bdev, p->old_block_size);
2537                blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2538        }
2539        destroy_swap_extents(p);
2540        swap_cgroup_swapoff(p->type);
2541        spin_lock(&swap_lock);
2542        p->swap_file = NULL;
2543        p->flags = 0;
2544        spin_unlock(&swap_lock);
2545        vfree(swap_map);
2546        vfree(cluster_info);
2547        if (swap_file) {
2548                if (inode && S_ISREG(inode->i_mode)) {
2549                        mutex_unlock(&inode->i_mutex);
2550                        inode = NULL;
2551                }
2552                filp_close(swap_file, NULL);
2553        }
2554out:
2555        if (page && !IS_ERR(page)) {
2556                kunmap(page);
2557                page_cache_release(page);
2558        }
2559        if (name)
2560                putname(name);
2561        if (inode && S_ISREG(inode->i_mode))
2562                mutex_unlock(&inode->i_mutex);
2563        return error;
2564}
2565
2566void si_swapinfo(struct sysinfo *val)
2567{
2568        unsigned int type;
2569        unsigned long nr_to_be_unused = 0;
2570
2571        spin_lock(&swap_lock);
2572        for (type = 0; type < nr_swapfiles; type++) {
2573                struct swap_info_struct *si = swap_info[type];
2574
2575                if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
2576                        nr_to_be_unused += si->inuse_pages;
2577        }
2578        val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
2579        val->totalswap = total_swap_pages + nr_to_be_unused;
2580        spin_unlock(&swap_lock);
2581}
2582
2583/*
2584 * Verify that a swap entry is valid and increment its swap map count.
2585 *
2586 * Returns error code in following case.
2587 * - success -> 0
2588 * - swp_entry is invalid -> EINVAL
2589 * - swp_entry is migration entry -> EINVAL
2590 * - swap-cache reference is requested but there is already one. -> EEXIST
2591 * - swap-cache reference is requested but the entry is not used. -> ENOENT
2592 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
2593 */
2594static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
2595{
2596        struct swap_info_struct *p;
2597        unsigned long offset, type;
2598        unsigned char count;
2599        unsigned char has_cache;
2600        int err = -EINVAL;
2601
2602        if (non_swap_entry(entry))
2603                goto out;
2604
2605        type = swp_type(entry);
2606        if (type >= nr_swapfiles)
2607                goto bad_file;
2608        p = swap_info[type];
2609        offset = swp_offset(entry);
2610
2611        spin_lock(&p->lock);
2612        if (unlikely(offset >= p->max))
2613                goto unlock_out;
2614
2615        count = p->swap_map[offset];
2616
2617        /*
2618         * swapin_readahead() doesn't check if a swap entry is valid, so the
2619         * swap entry could be SWAP_MAP_BAD. Check here with lock held.
2620         */
2621        if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
2622                err = -ENOENT;
2623                goto unlock_out;
2624        }
2625
2626        has_cache = count & SWAP_HAS_CACHE;
2627        count &= ~SWAP_HAS_CACHE;
2628        err = 0;
2629
2630        if (usage == SWAP_HAS_CACHE) {
2631
2632                /* set SWAP_HAS_CACHE if there is no cache and entry is used */
2633                if (!has_cache && count)
2634                        has_cache = SWAP_HAS_CACHE;
2635                else if (has_cache)             /* someone else added cache */
2636                        err = -EEXIST;
2637                else                            /* no users remaining */
2638                        err = -ENOENT;
2639
2640        } else if (count || has_cache) {
2641
2642                if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
2643                        count += usage;
2644                else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
2645                        err = -EINVAL;
2646                else if (swap_count_continued(p, offset, count))
2647                        count = COUNT_CONTINUED;
2648                else
2649                        err = -ENOMEM;
2650        } else
2651                err = -ENOENT;                  /* unused swap entry */
2652
2653        p->swap_map[offset] = count | has_cache;
2654
2655unlock_out:
2656        spin_unlock(&p->lock);
2657out:
2658        return err;
2659
2660bad_file:
2661        pr_err("swap_dup: %s%08lx\n", Bad_file, entry.val);
2662        goto out;
2663}
2664
2665/*
2666 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2667 * (in which case its reference count is never incremented).
2668 */
2669void swap_shmem_alloc(swp_entry_t entry)
2670{
2671        __swap_duplicate(entry, SWAP_MAP_SHMEM);
2672}
2673
2674/*
2675 * Increase reference count of swap entry by 1.
2676 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2677 * but could not be atomically allocated.  Returns 0, just as if it succeeded,
2678 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2679 * might occur if a page table entry has got corrupted.
2680 */
2681int swap_duplicate(swp_entry_t entry)
2682{
2683        int err = 0;
2684
2685        while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
2686                err = add_swap_count_continuation(entry, GFP_ATOMIC);
2687        return err;
2688}
2689
2690/*
2691 * @entry: swap entry for which we allocate swap cache.
2692 *
2693 * Called when allocating swap cache for existing swap entry,
2694 * This can return error codes. Returns 0 at success.
2695 * -EBUSY means there is a swap cache.
2696 * Note: return code is different from swap_duplicate().
2697 */
2698int swapcache_prepare(swp_entry_t entry)
2699{
2700        return __swap_duplicate(entry, SWAP_HAS_CACHE);
2701}
2702
2703struct swap_info_struct *page_swap_info(struct page *page)
2704{
2705        swp_entry_t swap = { .val = page_private(page) };
2706        BUG_ON(!PageSwapCache(page));
2707        return swap_info[swp_type(swap)];
2708}
2709
2710/*
2711 * out-of-line __page_file_ methods to avoid include hell.
2712 */
2713struct address_space *__page_file_mapping(struct page *page)
2714{
2715        VM_BUG_ON_PAGE(!PageSwapCache(page), page);
2716        return page_swap_info(page)->swap_file->f_mapping;
2717}
2718EXPORT_SYMBOL_GPL(__page_file_mapping);
2719
2720pgoff_t __page_file_index(struct page *page)
2721{
2722        swp_entry_t swap = { .val = page_private(page) };
2723        VM_BUG_ON_PAGE(!PageSwapCache(page), page);
2724        return swp_offset(swap);
2725}
2726EXPORT_SYMBOL_GPL(__page_file_index);
2727
2728/*
2729 * add_swap_count_continuation - called when a swap count is duplicated
2730 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2731 * page of the original vmalloc'ed swap_map, to hold the continuation count
2732 * (for that entry and for its neighbouring PAGE_SIZE swap entries).  Called
2733 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2734 *
2735 * These continuation pages are seldom referenced: the common paths all work
2736 * on the original swap_map, only referring to a continuation page when the
2737 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2738 *
2739 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2740 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2741 * can be called after dropping locks.
2742 */
2743int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
2744{
2745        struct swap_info_struct *si;
2746        struct page *head;
2747        struct page *page;
2748        struct page *list_page;
2749        pgoff_t offset;
2750        unsigned char count;
2751
2752        /*
2753         * When debugging, it's easier to use __GFP_ZERO here; but it's better
2754         * for latency not to zero a page while GFP_ATOMIC and holding locks.
2755         */
2756        page = alloc_page(gfp_mask | __GFP_HIGHMEM);
2757
2758        si = swap_info_get(entry);
2759        if (!si) {
2760                /*
2761                 * An acceptable race has occurred since the failing
2762                 * __swap_duplicate(): the swap entry has been freed,
2763                 * perhaps even the whole swap_map cleared for swapoff.
2764                 */
2765                goto outer;
2766        }
2767
2768        offset = swp_offset(entry);
2769        count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
2770
2771        if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
2772                /*
2773                 * The higher the swap count, the more likely it is that tasks
2774                 * will race to add swap count continuation: we need to avoid
2775                 * over-provisioning.
2776                 */
2777                goto out;
2778        }
2779
2780        if (!page) {
2781                spin_unlock(&si->lock);
2782                return -ENOMEM;
2783        }
2784
2785        /*
2786         * We are fortunate that although vmalloc_to_page uses pte_offset_map,
2787         * no architecture is using highmem pages for kernel page tables: so it
2788         * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
2789         */
2790        head = vmalloc_to_page(si->swap_map + offset);
2791        offset &= ~PAGE_MASK;
2792
2793        /*
2794         * Page allocation does not initialize the page's lru field,
2795         * but it does always reset its private field.
2796         */
2797        if (!page_private(head)) {
2798                BUG_ON(count & COUNT_CONTINUED);
2799                INIT_LIST_HEAD(&head->lru);
2800                set_page_private(head, SWP_CONTINUED);
2801                si->flags |= SWP_CONTINUED;
2802        }
2803
2804        list_for_each_entry(list_page, &head->lru, lru) {
2805                unsigned char *map;
2806
2807                /*
2808                 * If the previous map said no continuation, but we've found
2809                 * a continuation page, free our allocation and use this one.
2810                 */
2811                if (!(count & COUNT_CONTINUED))
2812                        goto out;
2813
2814                map = kmap_atomic(list_page) + offset;
2815                count = *map;
2816                kunmap_atomic(map);
2817
2818                /*
2819                 * If this continuation count now has some space in it,
2820                 * free our allocation and use this one.
2821                 */
2822                if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
2823                        goto out;
2824        }
2825
2826        list_add_tail(&page->lru, &head->lru);
2827        page = NULL;                    /* now it's attached, don't free it */
2828out:
2829        spin_unlock(&si->lock);
2830outer:
2831        if (page)
2832                __free_page(page);
2833        return 0;
2834}
2835
2836/*
2837 * swap_count_continued - when the original swap_map count is incremented
2838 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2839 * into, carry if so, or else fail until a new continuation page is allocated;
2840 * when the original swap_map count is decremented from 0 with continuation,
2841 * borrow from the continuation and report whether it still holds more.
2842 * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2843 */
2844static bool swap_count_continued(struct swap_info_struct *si,
2845                                 pgoff_t offset, unsigned char count)
2846{
2847        struct page *head;
2848        struct page *page;
2849        unsigned char *map;
2850
2851        head = vmalloc_to_page(si->swap_map + offset);
2852        if (page_private(head) != SWP_CONTINUED) {
2853                BUG_ON(count & COUNT_CONTINUED);
2854                return false;           /* need to add count continuation */
2855        }
2856
2857        offset &= ~PAGE_MASK;
2858        page = list_entry(head->lru.next, struct page, lru);
2859        map = kmap_atomic(page) + offset;
2860
2861        if (count == SWAP_MAP_MAX)      /* initial increment from swap_map */
2862                goto init_map;          /* jump over SWAP_CONT_MAX checks */
2863
2864        if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
2865                /*
2866                 * Think of how you add 1 to 999
2867                 */
2868                while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
2869                        kunmap_atomic(map);
2870                        page = list_entry(page->lru.next, struct page, lru);
2871                        BUG_ON(page == head);
2872                        map = kmap_atomic(page) + offset;
2873                }
2874                if (*map == SWAP_CONT_MAX) {
2875                        kunmap_atomic(map);
2876                        page = list_entry(page->lru.next, struct page, lru);
2877                        if (page == head)
2878                                return false;   /* add count continuation */
2879                        map = kmap_atomic(page) + offset;
2880init_map:               *map = 0;               /* we didn't zero the page */
2881                }
2882                *map += 1;
2883                kunmap_atomic(map);
2884                page = list_entry(page->lru.prev, struct page, lru);
2885                while (page != head) {
2886                        map = kmap_atomic(page) + offset;
2887                        *map = COUNT_CONTINUED;
2888                        kunmap_atomic(map);
2889                        page = list_entry(page->lru.prev, struct page, lru);
2890                }
2891                return true;                    /* incremented */
2892
2893        } else {                                /* decrementing */
2894                /*
2895                 * Think of how you subtract 1 from 1000
2896                 */
2897                BUG_ON(count != COUNT_CONTINUED);
2898                while (*map == COUNT_CONTINUED) {
2899                        kunmap_atomic(map);
2900                        page = list_entry(page->lru.next, struct page, lru);
2901                        BUG_ON(page == head);
2902                        map = kmap_atomic(page) + offset;
2903                }
2904                BUG_ON(*map == 0);
2905                *map -= 1;
2906                if (*map == 0)
2907                        count = 0;
2908                kunmap_atomic(map);
2909                page = list_entry(page->lru.prev, struct page, lru);
2910                while (page != head) {
2911                        map = kmap_atomic(page) + offset;
2912                        *map = SWAP_CONT_MAX | count;
2913                        count = COUNT_CONTINUED;
2914                        kunmap_atomic(map);
2915                        page = list_entry(page->lru.prev, struct page, lru);
2916                }
2917                return count == COUNT_CONTINUED;
2918        }
2919}
2920
2921/*
2922 * free_swap_count_continuations - swapoff free all the continuation pages
2923 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
2924 */
2925static void free_swap_count_continuations(struct swap_info_struct *si)
2926{
2927        pgoff_t offset;
2928
2929        for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
2930                struct page *head;
2931                head = vmalloc_to_page(si->swap_map + offset);
2932                if (page_private(head)) {
2933                        struct list_head *this, *next;
2934                        list_for_each_safe(this, next, &head->lru) {
2935                                struct page *page;
2936                                page = list_entry(this, struct page, lru);
2937                                list_del(this);
2938                                __free_page(page);
2939                        }
2940                }
2941        }
2942}
Note: See TracBrowser for help on using the repository browser.