4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 * Swap reorganised 29.12.95, Stephen Tweedie
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/shm.h>
18 #include <linux/blkdev.h>
19 #include <linux/random.h>
20 #include <linux/writeback.h>
21 #include <linux/proc_fs.h>
22 #include <linux/seq_file.h>
23 #include <linux/init.h>
24 #include <linux/module.h>
25 #include <linux/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>
33 #include <asm/pgtable.h>
34 #include <asm/tlbflush.h>
35 #include <linux/swapops.h>
36 #include <linux/page_cgroup.h>
38 static DEFINE_SPINLOCK(swap_lock);
39 static unsigned int nr_swapfiles;
41 long total_swap_pages;
42 static int swap_overflow;
43 static int least_priority;
45 static const char Bad_file[] = "Bad swap file entry ";
46 static const char Unused_file[] = "Unused swap file entry ";
47 static const char Bad_offset[] = "Bad swap offset entry ";
48 static const char Unused_offset[] = "Unused swap offset entry ";
50 static struct swap_list_t swap_list = {-1, -1};
52 static struct swap_info_struct *swap_info[MAX_SWAPFILES];
54 static DEFINE_MUTEX(swapon_mutex);
56 /* For reference count accounting in swap_map */
57 /* enum for swap_map[] handling. internal use only */
59 SWAP_MAP = 0, /* ops for reference from swap users */
60 SWAP_CACHE, /* ops for reference from swap cache */
63 static inline int swap_count(unsigned short ent)
65 return ent & SWAP_COUNT_MASK;
68 static inline bool swap_has_cache(unsigned short ent)
70 return !!(ent & SWAP_HAS_CACHE);
73 static inline unsigned short encode_swapmap(int count, bool has_cache)
75 unsigned short ret = count;
78 return SWAP_HAS_CACHE | ret;
82 /* returns 1 if swap entry is freed */
84 __try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset)
86 swp_entry_t entry = swp_entry(si->type, offset);
90 page = find_get_page(&swapper_space, entry.val);
94 * This function is called from scan_swap_map() and it's called
95 * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
96 * We have to use trylock for avoiding deadlock. This is a special
97 * case and you should use try_to_free_swap() with explicit lock_page()
98 * in usual operations.
100 if (trylock_page(page)) {
101 ret = try_to_free_swap(page);
104 page_cache_release(page);
109 * We need this because the bdev->unplug_fn can sleep and we cannot
110 * hold swap_lock while calling the unplug_fn. And swap_lock
111 * cannot be turned into a mutex.
113 static DECLARE_RWSEM(swap_unplug_sem);
115 void swap_unplug_io_fn(struct backing_dev_info *unused_bdi, struct page *page)
119 down_read(&swap_unplug_sem);
120 entry.val = page_private(page);
121 if (PageSwapCache(page)) {
122 struct block_device *bdev = swap_info[swp_type(entry)]->bdev;
123 struct backing_dev_info *bdi;
126 * If the page is removed from swapcache from under us (with a
127 * racy try_to_unuse/swapoff) we need an additional reference
128 * count to avoid reading garbage from page_private(page) above.
129 * If the WARN_ON triggers during a swapoff it maybe the race
130 * condition and it's harmless. However if it triggers without
131 * swapoff it signals a problem.
133 WARN_ON(page_count(page) <= 1);
135 bdi = bdev->bd_inode->i_mapping->backing_dev_info;
136 blk_run_backing_dev(bdi, page);
138 up_read(&swap_unplug_sem);
142 * swapon tell device that all the old swap contents can be discarded,
143 * to allow the swap device to optimize its wear-levelling.
145 static int discard_swap(struct swap_info_struct *si)
147 struct swap_extent *se;
148 sector_t start_block;
152 /* Do not discard the swap header page! */
153 se = &si->first_swap_extent;
154 start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
155 nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
157 err = blkdev_issue_discard(si->bdev, start_block,
158 nr_blocks, GFP_KERNEL, DISCARD_FL_BARRIER);
164 list_for_each_entry(se, &si->first_swap_extent.list, list) {
165 start_block = se->start_block << (PAGE_SHIFT - 9);
166 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
168 err = blkdev_issue_discard(si->bdev, start_block,
169 nr_blocks, GFP_KERNEL, DISCARD_FL_BARRIER);
175 return err; /* That will often be -EOPNOTSUPP */
179 * swap allocation tell device that a cluster of swap can now be discarded,
180 * to allow the swap device to optimize its wear-levelling.
182 static void discard_swap_cluster(struct swap_info_struct *si,
183 pgoff_t start_page, pgoff_t nr_pages)
185 struct swap_extent *se = si->curr_swap_extent;
186 int found_extent = 0;
189 struct list_head *lh;
191 if (se->start_page <= start_page &&
192 start_page < se->start_page + se->nr_pages) {
193 pgoff_t offset = start_page - se->start_page;
194 sector_t start_block = se->start_block + offset;
195 sector_t nr_blocks = se->nr_pages - offset;
197 if (nr_blocks > nr_pages)
198 nr_blocks = nr_pages;
199 start_page += nr_blocks;
200 nr_pages -= nr_blocks;
203 si->curr_swap_extent = se;
205 start_block <<= PAGE_SHIFT - 9;
206 nr_blocks <<= PAGE_SHIFT - 9;
207 if (blkdev_issue_discard(si->bdev, start_block,
208 nr_blocks, GFP_NOIO, DISCARD_FL_BARRIER))
213 se = list_entry(lh, struct swap_extent, list);
217 static int wait_for_discard(void *word)
223 #define SWAPFILE_CLUSTER 256
224 #define LATENCY_LIMIT 256
226 static inline unsigned long scan_swap_map(struct swap_info_struct *si,
229 unsigned long offset;
230 unsigned long scan_base;
231 unsigned long last_in_cluster = 0;
232 int latency_ration = LATENCY_LIMIT;
233 int found_free_cluster = 0;
236 * We try to cluster swap pages by allocating them sequentially
237 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
238 * way, however, we resort to first-free allocation, starting
239 * a new cluster. This prevents us from scattering swap pages
240 * all over the entire swap partition, so that we reduce
241 * overall disk seek times between swap pages. -- sct
242 * But we do now try to find an empty cluster. -Andrea
243 * And we let swap pages go all over an SSD partition. Hugh
246 si->flags += SWP_SCANNING;
247 scan_base = offset = si->cluster_next;
249 if (unlikely(!si->cluster_nr--)) {
250 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
251 si->cluster_nr = SWAPFILE_CLUSTER - 1;
254 if (si->flags & SWP_DISCARDABLE) {
256 * Start range check on racing allocations, in case
257 * they overlap the cluster we eventually decide on
258 * (we scan without swap_lock to allow preemption).
259 * It's hardly conceivable that cluster_nr could be
260 * wrapped during our scan, but don't depend on it.
262 if (si->lowest_alloc)
264 si->lowest_alloc = si->max;
265 si->highest_alloc = 0;
267 spin_unlock(&swap_lock);
270 * If seek is expensive, start searching for new cluster from
271 * start of partition, to minimize the span of allocated swap.
272 * But if seek is cheap, search from our current position, so
273 * that swap is allocated from all over the partition: if the
274 * Flash Translation Layer only remaps within limited zones,
275 * we don't want to wear out the first zone too quickly.
277 if (!(si->flags & SWP_SOLIDSTATE))
278 scan_base = offset = si->lowest_bit;
279 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
281 /* Locate the first empty (unaligned) cluster */
282 for (; last_in_cluster <= si->highest_bit; offset++) {
283 if (si->swap_map[offset])
284 last_in_cluster = offset + SWAPFILE_CLUSTER;
285 else if (offset == last_in_cluster) {
286 spin_lock(&swap_lock);
287 offset -= SWAPFILE_CLUSTER - 1;
288 si->cluster_next = offset;
289 si->cluster_nr = SWAPFILE_CLUSTER - 1;
290 found_free_cluster = 1;
293 if (unlikely(--latency_ration < 0)) {
295 latency_ration = LATENCY_LIMIT;
299 offset = si->lowest_bit;
300 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
302 /* Locate the first empty (unaligned) cluster */
303 for (; last_in_cluster < scan_base; offset++) {
304 if (si->swap_map[offset])
305 last_in_cluster = offset + SWAPFILE_CLUSTER;
306 else if (offset == last_in_cluster) {
307 spin_lock(&swap_lock);
308 offset -= SWAPFILE_CLUSTER - 1;
309 si->cluster_next = offset;
310 si->cluster_nr = SWAPFILE_CLUSTER - 1;
311 found_free_cluster = 1;
314 if (unlikely(--latency_ration < 0)) {
316 latency_ration = LATENCY_LIMIT;
321 spin_lock(&swap_lock);
322 si->cluster_nr = SWAPFILE_CLUSTER - 1;
323 si->lowest_alloc = 0;
327 if (!(si->flags & SWP_WRITEOK))
329 if (!si->highest_bit)
331 if (offset > si->highest_bit)
332 scan_base = offset = si->lowest_bit;
334 /* reuse swap entry of cache-only swap if not busy. */
335 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
337 spin_unlock(&swap_lock);
338 swap_was_freed = __try_to_reclaim_swap(si, offset);
339 spin_lock(&swap_lock);
340 /* entry was freed successfully, try to use this again */
343 goto scan; /* check next one */
346 if (si->swap_map[offset])
349 if (offset == si->lowest_bit)
351 if (offset == si->highest_bit)
354 if (si->inuse_pages == si->pages) {
355 si->lowest_bit = si->max;
358 if (cache == SWAP_CACHE) /* at usual swap-out via vmscan.c */
359 si->swap_map[offset] = encode_swapmap(0, true);
360 else /* at suspend */
361 si->swap_map[offset] = encode_swapmap(1, false);
362 si->cluster_next = offset + 1;
363 si->flags -= SWP_SCANNING;
365 if (si->lowest_alloc) {
367 * Only set when SWP_DISCARDABLE, and there's a scan
368 * for a free cluster in progress or just completed.
370 if (found_free_cluster) {
372 * To optimize wear-levelling, discard the
373 * old data of the cluster, taking care not to
374 * discard any of its pages that have already
375 * been allocated by racing tasks (offset has
376 * already stepped over any at the beginning).
378 if (offset < si->highest_alloc &&
379 si->lowest_alloc <= last_in_cluster)
380 last_in_cluster = si->lowest_alloc - 1;
381 si->flags |= SWP_DISCARDING;
382 spin_unlock(&swap_lock);
384 if (offset < last_in_cluster)
385 discard_swap_cluster(si, offset,
386 last_in_cluster - offset + 1);
388 spin_lock(&swap_lock);
389 si->lowest_alloc = 0;
390 si->flags &= ~SWP_DISCARDING;
392 smp_mb(); /* wake_up_bit advises this */
393 wake_up_bit(&si->flags, ilog2(SWP_DISCARDING));
395 } else if (si->flags & SWP_DISCARDING) {
397 * Delay using pages allocated by racing tasks
398 * until the whole discard has been issued. We
399 * could defer that delay until swap_writepage,
400 * but it's easier to keep this self-contained.
402 spin_unlock(&swap_lock);
403 wait_on_bit(&si->flags, ilog2(SWP_DISCARDING),
404 wait_for_discard, TASK_UNINTERRUPTIBLE);
405 spin_lock(&swap_lock);
408 * Note pages allocated by racing tasks while
409 * scan for a free cluster is in progress, so
410 * that its final discard can exclude them.
412 if (offset < si->lowest_alloc)
413 si->lowest_alloc = offset;
414 if (offset > si->highest_alloc)
415 si->highest_alloc = offset;
421 spin_unlock(&swap_lock);
422 while (++offset <= si->highest_bit) {
423 if (!si->swap_map[offset]) {
424 spin_lock(&swap_lock);
427 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
428 spin_lock(&swap_lock);
431 if (unlikely(--latency_ration < 0)) {
433 latency_ration = LATENCY_LIMIT;
436 offset = si->lowest_bit;
437 while (++offset < scan_base) {
438 if (!si->swap_map[offset]) {
439 spin_lock(&swap_lock);
442 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
443 spin_lock(&swap_lock);
446 if (unlikely(--latency_ration < 0)) {
448 latency_ration = LATENCY_LIMIT;
451 spin_lock(&swap_lock);
454 si->flags -= SWP_SCANNING;
458 swp_entry_t get_swap_page(void)
460 struct swap_info_struct *si;
465 spin_lock(&swap_lock);
466 if (nr_swap_pages <= 0)
470 for (type = swap_list.next; type >= 0 && wrapped < 2; type = next) {
471 si = swap_info[type];
474 (!wrapped && si->prio != swap_info[next]->prio)) {
475 next = swap_list.head;
479 if (!si->highest_bit)
481 if (!(si->flags & SWP_WRITEOK))
484 swap_list.next = next;
485 /* This is called for allocating swap entry for cache */
486 offset = scan_swap_map(si, SWAP_CACHE);
488 spin_unlock(&swap_lock);
489 return swp_entry(type, offset);
491 next = swap_list.next;
496 spin_unlock(&swap_lock);
497 return (swp_entry_t) {0};
500 /* The only caller of this function is now susupend routine */
501 swp_entry_t get_swap_page_of_type(int type)
503 struct swap_info_struct *si;
506 spin_lock(&swap_lock);
507 si = swap_info[type];
508 if (si && (si->flags & SWP_WRITEOK)) {
510 /* This is called for allocating swap entry, not cache */
511 offset = scan_swap_map(si, SWAP_MAP);
513 spin_unlock(&swap_lock);
514 return swp_entry(type, offset);
518 spin_unlock(&swap_lock);
519 return (swp_entry_t) {0};
522 static struct swap_info_struct * swap_info_get(swp_entry_t entry)
524 struct swap_info_struct * p;
525 unsigned long offset, type;
529 type = swp_type(entry);
530 if (type >= nr_swapfiles)
533 if (!(p->flags & SWP_USED))
535 offset = swp_offset(entry);
536 if (offset >= p->max)
538 if (!p->swap_map[offset])
540 spin_lock(&swap_lock);
544 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val);
547 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val);
550 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val);
553 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val);
558 static int swap_entry_free(struct swap_info_struct *p,
559 swp_entry_t ent, int cache)
561 unsigned long offset = swp_offset(ent);
562 int count = swap_count(p->swap_map[offset]);
565 has_cache = swap_has_cache(p->swap_map[offset]);
567 if (cache == SWAP_MAP) { /* dropping usage count of swap */
568 if (count < SWAP_MAP_MAX) {
570 p->swap_map[offset] = encode_swapmap(count, has_cache);
572 } else { /* dropping swap cache flag */
573 VM_BUG_ON(!has_cache);
574 p->swap_map[offset] = encode_swapmap(count, false);
578 count = p->swap_map[offset];
579 /* free if no reference */
581 if (offset < p->lowest_bit)
582 p->lowest_bit = offset;
583 if (offset > p->highest_bit)
584 p->highest_bit = offset;
585 if (swap_list.next >= 0 &&
586 p->prio > swap_info[swap_list.next]->prio)
587 swap_list.next = p->type;
591 if (!swap_count(count))
592 mem_cgroup_uncharge_swap(ent);
597 * Caller has made sure that the swapdevice corresponding to entry
598 * is still around or has not been recycled.
600 void swap_free(swp_entry_t entry)
602 struct swap_info_struct * p;
604 p = swap_info_get(entry);
606 swap_entry_free(p, entry, SWAP_MAP);
607 spin_unlock(&swap_lock);
612 * Called after dropping swapcache to decrease refcnt to swap entries.
614 void swapcache_free(swp_entry_t entry, struct page *page)
616 struct swap_info_struct *p;
619 p = swap_info_get(entry);
621 ret = swap_entry_free(p, entry, SWAP_CACHE);
625 swapout = true; /* the end of swap out */
627 swapout = false; /* no more swap users! */
628 mem_cgroup_uncharge_swapcache(page, entry, swapout);
630 spin_unlock(&swap_lock);
636 * How many references to page are currently swapped out?
638 static inline int page_swapcount(struct page *page)
641 struct swap_info_struct *p;
644 entry.val = page_private(page);
645 p = swap_info_get(entry);
647 count = swap_count(p->swap_map[swp_offset(entry)]);
648 spin_unlock(&swap_lock);
654 * We can write to an anon page without COW if there are no other references
655 * to it. And as a side-effect, free up its swap: because the old content
656 * on disk will never be read, and seeking back there to write new content
657 * later would only waste time away from clustering.
659 int reuse_swap_page(struct page *page)
663 VM_BUG_ON(!PageLocked(page));
664 count = page_mapcount(page);
665 if (count <= 1 && PageSwapCache(page)) {
666 count += page_swapcount(page);
667 if (count == 1 && !PageWriteback(page)) {
668 delete_from_swap_cache(page);
676 * If swap is getting full, or if there are no more mappings of this page,
677 * then try_to_free_swap is called to free its swap space.
679 int try_to_free_swap(struct page *page)
681 VM_BUG_ON(!PageLocked(page));
683 if (!PageSwapCache(page))
685 if (PageWriteback(page))
687 if (page_swapcount(page))
690 delete_from_swap_cache(page);
696 * Free the swap entry like above, but also try to
697 * free the page cache entry if it is the last user.
699 int free_swap_and_cache(swp_entry_t entry)
701 struct swap_info_struct *p;
702 struct page *page = NULL;
704 if (non_swap_entry(entry))
707 p = swap_info_get(entry);
709 if (swap_entry_free(p, entry, SWAP_MAP) == SWAP_HAS_CACHE) {
710 page = find_get_page(&swapper_space, entry.val);
711 if (page && !trylock_page(page)) {
712 page_cache_release(page);
716 spin_unlock(&swap_lock);
720 * Not mapped elsewhere, or swap space full? Free it!
721 * Also recheck PageSwapCache now page is locked (above).
723 if (PageSwapCache(page) && !PageWriteback(page) &&
724 (!page_mapped(page) || vm_swap_full())) {
725 delete_from_swap_cache(page);
729 page_cache_release(page);
734 #ifdef CONFIG_HIBERNATION
736 * Find the swap type that corresponds to given device (if any).
738 * @offset - number of the PAGE_SIZE-sized block of the device, starting
739 * from 0, in which the swap header is expected to be located.
741 * This is needed for the suspend to disk (aka swsusp).
743 int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
745 struct block_device *bdev = NULL;
749 bdev = bdget(device);
751 spin_lock(&swap_lock);
752 for (type = 0; type < nr_swapfiles; type++) {
753 struct swap_info_struct *sis = swap_info[type];
755 if (!(sis->flags & SWP_WRITEOK))
760 *bdev_p = bdgrab(sis->bdev);
762 spin_unlock(&swap_lock);
765 if (bdev == sis->bdev) {
766 struct swap_extent *se = &sis->first_swap_extent;
768 if (se->start_block == offset) {
770 *bdev_p = bdgrab(sis->bdev);
772 spin_unlock(&swap_lock);
778 spin_unlock(&swap_lock);
786 * Return either the total number of swap pages of given type, or the number
787 * of free pages of that type (depending on @free)
789 * This is needed for software suspend
791 unsigned int count_swap_pages(int type, int free)
795 spin_lock(&swap_lock);
796 if ((unsigned int)type < nr_swapfiles) {
797 struct swap_info_struct *sis = swap_info[type];
799 if (sis->flags & SWP_WRITEOK) {
802 n -= sis->inuse_pages;
805 spin_unlock(&swap_lock);
811 * No need to decide whether this PTE shares the swap entry with others,
812 * just let do_wp_page work it out if a write is requested later - to
813 * force COW, vm_page_prot omits write permission from any private vma.
815 static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
816 unsigned long addr, swp_entry_t entry, struct page *page)
818 struct mem_cgroup *ptr = NULL;
823 if (mem_cgroup_try_charge_swapin(vma->vm_mm, page, GFP_KERNEL, &ptr)) {
828 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
829 if (unlikely(!pte_same(*pte, swp_entry_to_pte(entry)))) {
831 mem_cgroup_cancel_charge_swapin(ptr);
836 inc_mm_counter(vma->vm_mm, anon_rss);
838 set_pte_at(vma->vm_mm, addr, pte,
839 pte_mkold(mk_pte(page, vma->vm_page_prot)));
840 page_add_anon_rmap(page, vma, addr);
841 mem_cgroup_commit_charge_swapin(page, ptr);
844 * Move the page to the active list so it is not
845 * immediately swapped out again after swapon.
849 pte_unmap_unlock(pte, ptl);
854 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
855 unsigned long addr, unsigned long end,
856 swp_entry_t entry, struct page *page)
858 pte_t swp_pte = swp_entry_to_pte(entry);
863 * We don't actually need pte lock while scanning for swp_pte: since
864 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
865 * page table while we're scanning; though it could get zapped, and on
866 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
867 * of unmatched parts which look like swp_pte, so unuse_pte must
868 * recheck under pte lock. Scanning without pte lock lets it be
869 * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
871 pte = pte_offset_map(pmd, addr);
874 * swapoff spends a _lot_ of time in this loop!
875 * Test inline before going to call unuse_pte.
877 if (unlikely(pte_same(*pte, swp_pte))) {
879 ret = unuse_pte(vma, pmd, addr, entry, page);
882 pte = pte_offset_map(pmd, addr);
884 } while (pte++, addr += PAGE_SIZE, addr != end);
890 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
891 unsigned long addr, unsigned long end,
892 swp_entry_t entry, struct page *page)
898 pmd = pmd_offset(pud, addr);
900 next = pmd_addr_end(addr, end);
901 if (pmd_none_or_clear_bad(pmd))
903 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
906 } while (pmd++, addr = next, addr != end);
910 static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
911 unsigned long addr, unsigned long end,
912 swp_entry_t entry, struct page *page)
918 pud = pud_offset(pgd, addr);
920 next = pud_addr_end(addr, end);
921 if (pud_none_or_clear_bad(pud))
923 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
926 } while (pud++, addr = next, addr != end);
930 static int unuse_vma(struct vm_area_struct *vma,
931 swp_entry_t entry, struct page *page)
934 unsigned long addr, end, next;
938 addr = page_address_in_vma(page, vma);
942 end = addr + PAGE_SIZE;
944 addr = vma->vm_start;
948 pgd = pgd_offset(vma->vm_mm, addr);
950 next = pgd_addr_end(addr, end);
951 if (pgd_none_or_clear_bad(pgd))
953 ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
956 } while (pgd++, addr = next, addr != end);
960 static int unuse_mm(struct mm_struct *mm,
961 swp_entry_t entry, struct page *page)
963 struct vm_area_struct *vma;
966 if (!down_read_trylock(&mm->mmap_sem)) {
968 * Activate page so shrink_inactive_list is unlikely to unmap
969 * its ptes while lock is dropped, so swapoff can make progress.
973 down_read(&mm->mmap_sem);
976 for (vma = mm->mmap; vma; vma = vma->vm_next) {
977 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
980 up_read(&mm->mmap_sem);
981 return (ret < 0)? ret: 0;
985 * Scan swap_map from current position to next entry still in use.
986 * Recycle to start on reaching the end, returning 0 when empty.
988 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
991 unsigned int max = si->max;
992 unsigned int i = prev;
996 * No need for swap_lock here: we're just looking
997 * for whether an entry is in use, not modifying it; false
998 * hits are okay, and sys_swapoff() has already prevented new
999 * allocations from this area (while holding swap_lock).
1008 * No entries in use at top of swap_map,
1009 * loop back to start and recheck there.
1015 count = si->swap_map[i];
1016 if (count && swap_count(count) != SWAP_MAP_BAD)
1023 * We completely avoid races by reading each swap page in advance,
1024 * and then search for the process using it. All the necessary
1025 * page table adjustments can then be made atomically.
1027 static int try_to_unuse(unsigned int type)
1029 struct swap_info_struct *si = swap_info[type];
1030 struct mm_struct *start_mm;
1031 unsigned short *swap_map;
1032 unsigned short swcount;
1037 int reset_overflow = 0;
1041 * When searching mms for an entry, a good strategy is to
1042 * start at the first mm we freed the previous entry from
1043 * (though actually we don't notice whether we or coincidence
1044 * freed the entry). Initialize this start_mm with a hold.
1046 * A simpler strategy would be to start at the last mm we
1047 * freed the previous entry from; but that would take less
1048 * advantage of mmlist ordering, which clusters forked mms
1049 * together, child after parent. If we race with dup_mmap(), we
1050 * prefer to resolve parent before child, lest we miss entries
1051 * duplicated after we scanned child: using last mm would invert
1052 * that. Though it's only a serious concern when an overflowed
1053 * swap count is reset from SWAP_MAP_MAX, preventing a rescan.
1055 start_mm = &init_mm;
1056 atomic_inc(&init_mm.mm_users);
1059 * Keep on scanning until all entries have gone. Usually,
1060 * one pass through swap_map is enough, but not necessarily:
1061 * there are races when an instance of an entry might be missed.
1063 while ((i = find_next_to_unuse(si, i)) != 0) {
1064 if (signal_pending(current)) {
1070 * Get a page for the entry, using the existing swap
1071 * cache page if there is one. Otherwise, get a clean
1072 * page and read the swap into it.
1074 swap_map = &si->swap_map[i];
1075 entry = swp_entry(type, i);
1076 page = read_swap_cache_async(entry,
1077 GFP_HIGHUSER_MOVABLE, NULL, 0);
1080 * Either swap_duplicate() failed because entry
1081 * has been freed independently, and will not be
1082 * reused since sys_swapoff() already disabled
1083 * allocation from here, or alloc_page() failed.
1092 * Don't hold on to start_mm if it looks like exiting.
1094 if (atomic_read(&start_mm->mm_users) == 1) {
1096 start_mm = &init_mm;
1097 atomic_inc(&init_mm.mm_users);
1101 * Wait for and lock page. When do_swap_page races with
1102 * try_to_unuse, do_swap_page can handle the fault much
1103 * faster than try_to_unuse can locate the entry. This
1104 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1105 * defer to do_swap_page in such a case - in some tests,
1106 * do_swap_page and try_to_unuse repeatedly compete.
1108 wait_on_page_locked(page);
1109 wait_on_page_writeback(page);
1111 wait_on_page_writeback(page);
1114 * Remove all references to entry.
1115 * Whenever we reach init_mm, there's no address space
1116 * to search, but use it as a reminder to search shmem.
1119 swcount = *swap_map;
1120 if (swap_count(swcount)) {
1121 if (start_mm == &init_mm)
1122 shmem = shmem_unuse(entry, page);
1124 retval = unuse_mm(start_mm, entry, page);
1126 if (swap_count(*swap_map)) {
1127 int set_start_mm = (*swap_map >= swcount);
1128 struct list_head *p = &start_mm->mmlist;
1129 struct mm_struct *new_start_mm = start_mm;
1130 struct mm_struct *prev_mm = start_mm;
1131 struct mm_struct *mm;
1133 atomic_inc(&new_start_mm->mm_users);
1134 atomic_inc(&prev_mm->mm_users);
1135 spin_lock(&mmlist_lock);
1136 while (swap_count(*swap_map) && !retval && !shmem &&
1137 (p = p->next) != &start_mm->mmlist) {
1138 mm = list_entry(p, struct mm_struct, mmlist);
1139 if (!atomic_inc_not_zero(&mm->mm_users))
1141 spin_unlock(&mmlist_lock);
1147 swcount = *swap_map;
1148 if (!swap_count(swcount)) /* any usage ? */
1150 else if (mm == &init_mm) {
1152 shmem = shmem_unuse(entry, page);
1154 retval = unuse_mm(mm, entry, page);
1156 if (set_start_mm && *swap_map < swcount) {
1157 mmput(new_start_mm);
1158 atomic_inc(&mm->mm_users);
1162 spin_lock(&mmlist_lock);
1164 spin_unlock(&mmlist_lock);
1167 start_mm = new_start_mm;
1170 /* page has already been unlocked and released */
1178 page_cache_release(page);
1183 * How could swap count reach 0x7ffe ?
1184 * There's no way to repeat a swap page within an mm
1185 * (except in shmem, where it's the shared object which takes
1186 * the reference count)?
1187 * We believe SWAP_MAP_MAX cannot occur.(if occur, unsigned
1188 * short is too small....)
1189 * If that's wrong, then we should worry more about
1190 * exit_mmap() and do_munmap() cases described above:
1191 * we might be resetting SWAP_MAP_MAX too early here.
1192 * We know "Undead"s can happen, they're okay, so don't
1193 * report them; but do report if we reset SWAP_MAP_MAX.
1195 /* We might release the lock_page() in unuse_mm(). */
1196 if (!PageSwapCache(page) || page_private(page) != entry.val)
1199 if (swap_count(*swap_map) == SWAP_MAP_MAX) {
1200 spin_lock(&swap_lock);
1201 *swap_map = encode_swapmap(0, true);
1202 spin_unlock(&swap_lock);
1207 * If a reference remains (rare), we would like to leave
1208 * the page in the swap cache; but try_to_unmap could
1209 * then re-duplicate the entry once we drop page lock,
1210 * so we might loop indefinitely; also, that page could
1211 * not be swapped out to other storage meanwhile. So:
1212 * delete from cache even if there's another reference,
1213 * after ensuring that the data has been saved to disk -
1214 * since if the reference remains (rarer), it will be
1215 * read from disk into another page. Splitting into two
1216 * pages would be incorrect if swap supported "shared
1217 * private" pages, but they are handled by tmpfs files.
1219 if (swap_count(*swap_map) &&
1220 PageDirty(page) && PageSwapCache(page)) {
1221 struct writeback_control wbc = {
1222 .sync_mode = WB_SYNC_NONE,
1225 swap_writepage(page, &wbc);
1227 wait_on_page_writeback(page);
1231 * It is conceivable that a racing task removed this page from
1232 * swap cache just before we acquired the page lock at the top,
1233 * or while we dropped it in unuse_mm(). The page might even
1234 * be back in swap cache on another swap area: that we must not
1235 * delete, since it may not have been written out to swap yet.
1237 if (PageSwapCache(page) &&
1238 likely(page_private(page) == entry.val))
1239 delete_from_swap_cache(page);
1242 * So we could skip searching mms once swap count went
1243 * to 1, we did not mark any present ptes as dirty: must
1244 * mark page dirty so shrink_page_list will preserve it.
1249 page_cache_release(page);
1252 * Make sure that we aren't completely killing
1253 * interactive performance.
1259 if (reset_overflow) {
1260 printk(KERN_WARNING "swapoff: cleared swap entry overflow\n");
1267 * After a successful try_to_unuse, if no swap is now in use, we know
1268 * we can empty the mmlist. swap_lock must be held on entry and exit.
1269 * Note that mmlist_lock nests inside swap_lock, and an mm must be
1270 * added to the mmlist just after page_duplicate - before would be racy.
1272 static void drain_mmlist(void)
1274 struct list_head *p, *next;
1277 for (type = 0; type < nr_swapfiles; type++)
1278 if (swap_info[type]->inuse_pages)
1280 spin_lock(&mmlist_lock);
1281 list_for_each_safe(p, next, &init_mm.mmlist)
1283 spin_unlock(&mmlist_lock);
1287 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
1288 * corresponds to page offset `offset'. Note that the type of this function
1289 * is sector_t, but it returns page offset into the bdev, not sector offset.
1291 sector_t map_swap_page(swp_entry_t entry, struct block_device **bdev)
1293 struct swap_info_struct *sis;
1294 struct swap_extent *start_se;
1295 struct swap_extent *se;
1298 sis = swap_info[swp_type(entry)];
1301 offset = swp_offset(entry);
1302 start_se = sis->curr_swap_extent;
1306 struct list_head *lh;
1308 if (se->start_page <= offset &&
1309 offset < (se->start_page + se->nr_pages)) {
1310 return se->start_block + (offset - se->start_page);
1313 se = list_entry(lh, struct swap_extent, list);
1314 sis->curr_swap_extent = se;
1315 BUG_ON(se == start_se); /* It *must* be present */
1319 #ifdef CONFIG_HIBERNATION
1321 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1322 * corresponding to given index in swap_info (swap type).
1324 sector_t swapdev_block(int type, pgoff_t offset)
1326 struct block_device *bdev;
1328 if ((unsigned int)type >= nr_swapfiles)
1330 if (!(swap_info[type]->flags & SWP_WRITEOK))
1332 return map_swap_page(swp_entry(type, offset), &bdev);
1334 #endif /* CONFIG_HIBERNATION */
1337 * Free all of a swapdev's extent information
1339 static void destroy_swap_extents(struct swap_info_struct *sis)
1341 while (!list_empty(&sis->first_swap_extent.list)) {
1342 struct swap_extent *se;
1344 se = list_entry(sis->first_swap_extent.list.next,
1345 struct swap_extent, list);
1346 list_del(&se->list);
1352 * Add a block range (and the corresponding page range) into this swapdev's
1353 * extent list. The extent list is kept sorted in page order.
1355 * This function rather assumes that it is called in ascending page order.
1358 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1359 unsigned long nr_pages, sector_t start_block)
1361 struct swap_extent *se;
1362 struct swap_extent *new_se;
1363 struct list_head *lh;
1365 if (start_page == 0) {
1366 se = &sis->first_swap_extent;
1367 sis->curr_swap_extent = se;
1369 se->nr_pages = nr_pages;
1370 se->start_block = start_block;
1373 lh = sis->first_swap_extent.list.prev; /* Highest extent */
1374 se = list_entry(lh, struct swap_extent, list);
1375 BUG_ON(se->start_page + se->nr_pages != start_page);
1376 if (se->start_block + se->nr_pages == start_block) {
1378 se->nr_pages += nr_pages;
1384 * No merge. Insert a new extent, preserving ordering.
1386 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1389 new_se->start_page = start_page;
1390 new_se->nr_pages = nr_pages;
1391 new_se->start_block = start_block;
1393 list_add_tail(&new_se->list, &sis->first_swap_extent.list);
1398 * A `swap extent' is a simple thing which maps a contiguous range of pages
1399 * onto a contiguous range of disk blocks. An ordered list of swap extents
1400 * is built at swapon time and is then used at swap_writepage/swap_readpage
1401 * time for locating where on disk a page belongs.
1403 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1404 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1405 * swap files identically.
1407 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1408 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1409 * swapfiles are handled *identically* after swapon time.
1411 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1412 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1413 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1414 * requirements, they are simply tossed out - we will never use those blocks
1417 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
1418 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1419 * which will scribble on the fs.
1421 * The amount of disk space which a single swap extent represents varies.
1422 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1423 * extents in the list. To avoid much list walking, we cache the previous
1424 * search location in `curr_swap_extent', and start new searches from there.
1425 * This is extremely effective. The average number of iterations in
1426 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1428 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
1430 struct inode *inode;
1431 unsigned blocks_per_page;
1432 unsigned long page_no;
1434 sector_t probe_block;
1435 sector_t last_block;
1436 sector_t lowest_block = -1;
1437 sector_t highest_block = 0;
1441 inode = sis->swap_file->f_mapping->host;
1442 if (S_ISBLK(inode->i_mode)) {
1443 ret = add_swap_extent(sis, 0, sis->max, 0);
1448 blkbits = inode->i_blkbits;
1449 blocks_per_page = PAGE_SIZE >> blkbits;
1452 * Map all the blocks into the extent list. This code doesn't try
1457 last_block = i_size_read(inode) >> blkbits;
1458 while ((probe_block + blocks_per_page) <= last_block &&
1459 page_no < sis->max) {
1460 unsigned block_in_page;
1461 sector_t first_block;
1463 first_block = bmap(inode, probe_block);
1464 if (first_block == 0)
1468 * It must be PAGE_SIZE aligned on-disk
1470 if (first_block & (blocks_per_page - 1)) {
1475 for (block_in_page = 1; block_in_page < blocks_per_page;
1479 block = bmap(inode, probe_block + block_in_page);
1482 if (block != first_block + block_in_page) {
1489 first_block >>= (PAGE_SHIFT - blkbits);
1490 if (page_no) { /* exclude the header page */
1491 if (first_block < lowest_block)
1492 lowest_block = first_block;
1493 if (first_block > highest_block)
1494 highest_block = first_block;
1498 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1500 ret = add_swap_extent(sis, page_no, 1, first_block);
1505 probe_block += blocks_per_page;
1510 *span = 1 + highest_block - lowest_block;
1512 page_no = 1; /* force Empty message */
1514 sis->pages = page_no - 1;
1515 sis->highest_bit = page_no - 1;
1519 printk(KERN_ERR "swapon: swapfile has holes\n");
1524 SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
1526 struct swap_info_struct * p = NULL;
1527 unsigned short *swap_map;
1528 struct file *swap_file, *victim;
1529 struct address_space *mapping;
1530 struct inode *inode;
1535 if (!capable(CAP_SYS_ADMIN))
1538 pathname = getname(specialfile);
1539 err = PTR_ERR(pathname);
1540 if (IS_ERR(pathname))
1543 victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0);
1545 err = PTR_ERR(victim);
1549 mapping = victim->f_mapping;
1551 spin_lock(&swap_lock);
1552 for (type = swap_list.head; type >= 0; type = swap_info[type]->next) {
1553 p = swap_info[type];
1554 if (p->flags & SWP_WRITEOK) {
1555 if (p->swap_file->f_mapping == mapping)
1562 spin_unlock(&swap_lock);
1565 if (!security_vm_enough_memory(p->pages))
1566 vm_unacct_memory(p->pages);
1569 spin_unlock(&swap_lock);
1573 swap_list.head = p->next;
1575 swap_info[prev]->next = p->next;
1576 if (type == swap_list.next) {
1577 /* just pick something that's safe... */
1578 swap_list.next = swap_list.head;
1581 for (i = p->next; i >= 0; i = swap_info[i]->next)
1582 swap_info[i]->prio = p->prio--;
1585 nr_swap_pages -= p->pages;
1586 total_swap_pages -= p->pages;
1587 p->flags &= ~SWP_WRITEOK;
1588 spin_unlock(&swap_lock);
1590 current->flags |= PF_OOM_ORIGIN;
1591 err = try_to_unuse(type);
1592 current->flags &= ~PF_OOM_ORIGIN;
1595 /* re-insert swap space back into swap_list */
1596 spin_lock(&swap_lock);
1598 p->prio = --least_priority;
1600 for (i = swap_list.head; i >= 0; i = swap_info[i]->next) {
1601 if (p->prio >= swap_info[i]->prio)
1607 swap_list.head = swap_list.next = type;
1609 swap_info[prev]->next = type;
1610 nr_swap_pages += p->pages;
1611 total_swap_pages += p->pages;
1612 p->flags |= SWP_WRITEOK;
1613 spin_unlock(&swap_lock);
1617 /* wait for any unplug function to finish */
1618 down_write(&swap_unplug_sem);
1619 up_write(&swap_unplug_sem);
1621 destroy_swap_extents(p);
1622 mutex_lock(&swapon_mutex);
1623 spin_lock(&swap_lock);
1626 /* wait for anyone still in scan_swap_map */
1627 p->highest_bit = 0; /* cuts scans short */
1628 while (p->flags >= SWP_SCANNING) {
1629 spin_unlock(&swap_lock);
1630 schedule_timeout_uninterruptible(1);
1631 spin_lock(&swap_lock);
1634 swap_file = p->swap_file;
1635 p->swap_file = NULL;
1637 swap_map = p->swap_map;
1640 spin_unlock(&swap_lock);
1641 mutex_unlock(&swapon_mutex);
1643 /* Destroy swap account informatin */
1644 swap_cgroup_swapoff(type);
1646 inode = mapping->host;
1647 if (S_ISBLK(inode->i_mode)) {
1648 struct block_device *bdev = I_BDEV(inode);
1649 set_blocksize(bdev, p->old_block_size);
1652 mutex_lock(&inode->i_mutex);
1653 inode->i_flags &= ~S_SWAPFILE;
1654 mutex_unlock(&inode->i_mutex);
1656 filp_close(swap_file, NULL);
1660 filp_close(victim, NULL);
1665 #ifdef CONFIG_PROC_FS
1667 static void *swap_start(struct seq_file *swap, loff_t *pos)
1669 struct swap_info_struct *si;
1673 mutex_lock(&swapon_mutex);
1676 return SEQ_START_TOKEN;
1678 for (type = 0; type < nr_swapfiles; type++) {
1679 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1680 si = swap_info[type];
1681 if (!(si->flags & SWP_USED) || !si->swap_map)
1690 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1692 struct swap_info_struct *si = v;
1695 if (v == SEQ_START_TOKEN)
1698 type = si->type + 1;
1700 for (; type < nr_swapfiles; type++) {
1701 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1702 si = swap_info[type];
1703 if (!(si->flags & SWP_USED) || !si->swap_map)
1712 static void swap_stop(struct seq_file *swap, void *v)
1714 mutex_unlock(&swapon_mutex);
1717 static int swap_show(struct seq_file *swap, void *v)
1719 struct swap_info_struct *si = v;
1723 if (si == SEQ_START_TOKEN) {
1724 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1728 file = si->swap_file;
1729 len = seq_path(swap, &file->f_path, " \t\n\\");
1730 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
1731 len < 40 ? 40 - len : 1, " ",
1732 S_ISBLK(file->f_path.dentry->d_inode->i_mode) ?
1733 "partition" : "file\t",
1734 si->pages << (PAGE_SHIFT - 10),
1735 si->inuse_pages << (PAGE_SHIFT - 10),
1740 static const struct seq_operations swaps_op = {
1741 .start = swap_start,
1747 static int swaps_open(struct inode *inode, struct file *file)
1749 return seq_open(file, &swaps_op);
1752 static const struct file_operations proc_swaps_operations = {
1755 .llseek = seq_lseek,
1756 .release = seq_release,
1759 static int __init procswaps_init(void)
1761 proc_create("swaps", 0, NULL, &proc_swaps_operations);
1764 __initcall(procswaps_init);
1765 #endif /* CONFIG_PROC_FS */
1767 #ifdef MAX_SWAPFILES_CHECK
1768 static int __init max_swapfiles_check(void)
1770 MAX_SWAPFILES_CHECK();
1773 late_initcall(max_swapfiles_check);
1777 * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
1779 * The swapon system call
1781 SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
1783 struct swap_info_struct * p;
1785 struct block_device *bdev = NULL;
1786 struct file *swap_file = NULL;
1787 struct address_space *mapping;
1791 union swap_header *swap_header = NULL;
1792 unsigned int nr_good_pages = 0;
1795 unsigned long maxpages = 1;
1796 unsigned long swapfilepages;
1797 unsigned short *swap_map = NULL;
1798 struct page *page = NULL;
1799 struct inode *inode = NULL;
1802 if (!capable(CAP_SYS_ADMIN))
1805 p = kzalloc(sizeof(*p), GFP_KERNEL);
1809 spin_lock(&swap_lock);
1810 for (type = 0; type < nr_swapfiles; type++) {
1811 if (!(swap_info[type]->flags & SWP_USED))
1815 if (type >= MAX_SWAPFILES) {
1816 spin_unlock(&swap_lock);
1820 if (type >= nr_swapfiles) {
1822 swap_info[type] = p;
1824 * Write swap_info[type] before nr_swapfiles, in case a
1825 * racing procfs swap_start() or swap_next() is reading them.
1826 * (We never shrink nr_swapfiles, we never free this entry.)
1832 p = swap_info[type];
1834 * Do not memset this entry: a racing procfs swap_next()
1835 * would be relying on p->type to remain valid.
1838 INIT_LIST_HEAD(&p->first_swap_extent.list);
1839 p->flags = SWP_USED;
1841 spin_unlock(&swap_lock);
1843 name = getname(specialfile);
1844 error = PTR_ERR(name);
1849 swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0);
1850 error = PTR_ERR(swap_file);
1851 if (IS_ERR(swap_file)) {
1856 p->swap_file = swap_file;
1857 mapping = swap_file->f_mapping;
1858 inode = mapping->host;
1861 for (i = 0; i < nr_swapfiles; i++) {
1862 struct swap_info_struct *q = swap_info[i];
1864 if (i == type || !q->swap_file)
1866 if (mapping == q->swap_file->f_mapping)
1871 if (S_ISBLK(inode->i_mode)) {
1872 bdev = I_BDEV(inode);
1873 error = bd_claim(bdev, sys_swapon);
1879 p->old_block_size = block_size(bdev);
1880 error = set_blocksize(bdev, PAGE_SIZE);
1884 } else if (S_ISREG(inode->i_mode)) {
1885 p->bdev = inode->i_sb->s_bdev;
1886 mutex_lock(&inode->i_mutex);
1888 if (IS_SWAPFILE(inode)) {
1896 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
1899 * Read the swap header.
1901 if (!mapping->a_ops->readpage) {
1905 page = read_mapping_page(mapping, 0, swap_file);
1907 error = PTR_ERR(page);
1910 swap_header = kmap(page);
1912 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
1913 printk(KERN_ERR "Unable to find swap-space signature\n");
1918 /* swap partition endianess hack... */
1919 if (swab32(swap_header->info.version) == 1) {
1920 swab32s(&swap_header->info.version);
1921 swab32s(&swap_header->info.last_page);
1922 swab32s(&swap_header->info.nr_badpages);
1923 for (i = 0; i < swap_header->info.nr_badpages; i++)
1924 swab32s(&swap_header->info.badpages[i]);
1926 /* Check the swap header's sub-version */
1927 if (swap_header->info.version != 1) {
1929 "Unable to handle swap header version %d\n",
1930 swap_header->info.version);
1936 p->cluster_next = 1;
1940 * Find out how many pages are allowed for a single swap
1941 * device. There are two limiting factors: 1) the number of
1942 * bits for the swap offset in the swp_entry_t type and
1943 * 2) the number of bits in the a swap pte as defined by
1944 * the different architectures. In order to find the
1945 * largest possible bit mask a swap entry with swap type 0
1946 * and swap offset ~0UL is created, encoded to a swap pte,
1947 * decoded to a swp_entry_t again and finally the swap
1948 * offset is extracted. This will mask all the bits from
1949 * the initial ~0UL mask that can't be encoded in either
1950 * the swp_entry_t or the architecture definition of a
1953 maxpages = swp_offset(pte_to_swp_entry(
1954 swp_entry_to_pte(swp_entry(0, ~0UL)))) - 1;
1955 if (maxpages > swap_header->info.last_page)
1956 maxpages = swap_header->info.last_page;
1957 p->highest_bit = maxpages - 1;
1962 if (swapfilepages && maxpages > swapfilepages) {
1964 "Swap area shorter than signature indicates\n");
1967 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
1969 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
1972 /* OK, set up the swap map and apply the bad block list */
1973 swap_map = vmalloc(maxpages * sizeof(short));
1979 memset(swap_map, 0, maxpages * sizeof(short));
1980 for (i = 0; i < swap_header->info.nr_badpages; i++) {
1981 int page_nr = swap_header->info.badpages[i];
1982 if (page_nr <= 0 || page_nr >= swap_header->info.last_page) {
1986 swap_map[page_nr] = SWAP_MAP_BAD;
1989 error = swap_cgroup_swapon(type, maxpages);
1993 nr_good_pages = swap_header->info.last_page -
1994 swap_header->info.nr_badpages -
1995 1 /* header page */;
1997 if (nr_good_pages) {
1998 swap_map[0] = SWAP_MAP_BAD;
2000 p->pages = nr_good_pages;
2001 nr_extents = setup_swap_extents(p, &span);
2002 if (nr_extents < 0) {
2006 nr_good_pages = p->pages;
2008 if (!nr_good_pages) {
2009 printk(KERN_WARNING "Empty swap-file\n");
2015 if (blk_queue_nonrot(bdev_get_queue(p->bdev))) {
2016 p->flags |= SWP_SOLIDSTATE;
2017 p->cluster_next = 1 + (random32() % p->highest_bit);
2019 if (discard_swap(p) == 0)
2020 p->flags |= SWP_DISCARDABLE;
2023 mutex_lock(&swapon_mutex);
2024 spin_lock(&swap_lock);
2025 if (swap_flags & SWAP_FLAG_PREFER)
2027 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
2029 p->prio = --least_priority;
2030 p->swap_map = swap_map;
2031 p->flags |= SWP_WRITEOK;
2032 nr_swap_pages += nr_good_pages;
2033 total_swap_pages += nr_good_pages;
2035 printk(KERN_INFO "Adding %uk swap on %s. "
2036 "Priority:%d extents:%d across:%lluk %s%s\n",
2037 nr_good_pages<<(PAGE_SHIFT-10), name, p->prio,
2038 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
2039 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
2040 (p->flags & SWP_DISCARDABLE) ? "D" : "");
2042 /* insert swap space into swap_list: */
2044 for (i = swap_list.head; i >= 0; i = swap_info[i]->next) {
2045 if (p->prio >= swap_info[i]->prio)
2051 swap_list.head = swap_list.next = type;
2053 swap_info[prev]->next = type;
2054 spin_unlock(&swap_lock);
2055 mutex_unlock(&swapon_mutex);
2060 set_blocksize(bdev, p->old_block_size);
2063 destroy_swap_extents(p);
2064 swap_cgroup_swapoff(type);
2066 spin_lock(&swap_lock);
2067 p->swap_file = NULL;
2069 spin_unlock(&swap_lock);
2072 filp_close(swap_file, NULL);
2074 if (page && !IS_ERR(page)) {
2076 page_cache_release(page);
2082 inode->i_flags |= S_SWAPFILE;
2083 mutex_unlock(&inode->i_mutex);
2088 void si_swapinfo(struct sysinfo *val)
2091 unsigned long nr_to_be_unused = 0;
2093 spin_lock(&swap_lock);
2094 for (type = 0; type < nr_swapfiles; type++) {
2095 struct swap_info_struct *si = swap_info[type];
2097 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
2098 nr_to_be_unused += si->inuse_pages;
2100 val->freeswap = nr_swap_pages + nr_to_be_unused;
2101 val->totalswap = total_swap_pages + nr_to_be_unused;
2102 spin_unlock(&swap_lock);
2106 * Verify that a swap entry is valid and increment its swap map count.
2108 * Note: if swap_map[] reaches SWAP_MAP_MAX the entries are treated as
2109 * "permanent", but will be reclaimed by the next swapoff.
2110 * Returns error code in following case.
2112 * - swp_entry is invalid -> EINVAL
2113 * - swp_entry is migration entry -> EINVAL
2114 * - swap-cache reference is requested but there is already one. -> EEXIST
2115 * - swap-cache reference is requested but the entry is not used. -> ENOENT
2117 static int __swap_duplicate(swp_entry_t entry, bool cache)
2119 struct swap_info_struct * p;
2120 unsigned long offset, type;
2121 int result = -EINVAL;
2125 if (non_swap_entry(entry))
2128 type = swp_type(entry);
2129 if (type >= nr_swapfiles)
2131 p = swap_info[type];
2132 offset = swp_offset(entry);
2134 spin_lock(&swap_lock);
2136 if (unlikely(offset >= p->max))
2139 count = swap_count(p->swap_map[offset]);
2140 has_cache = swap_has_cache(p->swap_map[offset]);
2142 if (cache == SWAP_CACHE) { /* called for swapcache/swapin-readahead */
2144 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
2145 if (!has_cache && count) {
2146 p->swap_map[offset] = encode_swapmap(count, true);
2148 } else if (has_cache) /* someone added cache */
2150 else if (!count) /* no users */
2153 } else if (count || has_cache) {
2154 if (count < SWAP_MAP_MAX - 1) {
2155 p->swap_map[offset] = encode_swapmap(count + 1,
2158 } else if (count <= SWAP_MAP_MAX) {
2159 if (swap_overflow++ < 5)
2161 "swap_dup: swap entry overflow\n");
2162 p->swap_map[offset] = encode_swapmap(SWAP_MAP_MAX,
2167 result = -ENOENT; /* unused swap entry */
2169 spin_unlock(&swap_lock);
2174 printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
2178 * increase reference count of swap entry by 1.
2180 void swap_duplicate(swp_entry_t entry)
2182 __swap_duplicate(entry, SWAP_MAP);
2186 * @entry: swap entry for which we allocate swap cache.
2188 * Called when allocating swap cache for exising swap entry,
2189 * This can return error codes. Returns 0 at success.
2190 * -EBUSY means there is a swap cache.
2191 * Note: return code is different from swap_duplicate().
2193 int swapcache_prepare(swp_entry_t entry)
2195 return __swap_duplicate(entry, SWAP_CACHE);
2199 * swap_lock prevents swap_map being freed. Don't grab an extra
2200 * reference on the swaphandle, it doesn't matter if it becomes unused.
2202 int valid_swaphandles(swp_entry_t entry, unsigned long *offset)
2204 struct swap_info_struct *si;
2205 int our_page_cluster = page_cluster;
2206 pgoff_t target, toff;
2210 if (!our_page_cluster) /* no readahead */
2213 si = swap_info[swp_type(entry)];
2214 target = swp_offset(entry);
2215 base = (target >> our_page_cluster) << our_page_cluster;
2216 end = base + (1 << our_page_cluster);
2217 if (!base) /* first page is swap header */
2220 spin_lock(&swap_lock);
2221 if (end > si->max) /* don't go beyond end of map */
2224 /* Count contiguous allocated slots above our target */
2225 for (toff = target; ++toff < end; nr_pages++) {
2226 /* Don't read in free or bad pages */
2227 if (!si->swap_map[toff])
2229 if (swap_count(si->swap_map[toff]) == SWAP_MAP_BAD)
2232 /* Count contiguous allocated slots below our target */
2233 for (toff = target; --toff >= base; nr_pages++) {
2234 /* Don't read in free or bad pages */
2235 if (!si->swap_map[toff])
2237 if (swap_count(si->swap_map[toff]) == SWAP_MAP_BAD)
2240 spin_unlock(&swap_lock);
2243 * Indicate starting offset, and return number of pages to get:
2244 * if only 1, say 0, since there's then no readahead to be done.
2247 return nr_pages? ++nr_pages: 0;