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;
84 * We need this because the bdev->unplug_fn can sleep and we cannot
85 * hold swap_lock while calling the unplug_fn. And swap_lock
86 * cannot be turned into a mutex.
88 static DECLARE_RWSEM(swap_unplug_sem);
90 void swap_unplug_io_fn(struct backing_dev_info *unused_bdi, struct page *page)
94 down_read(&swap_unplug_sem);
95 entry.val = page_private(page);
96 if (PageSwapCache(page)) {
97 struct block_device *bdev = swap_info[swp_type(entry)].bdev;
98 struct backing_dev_info *bdi;
101 * If the page is removed from swapcache from under us (with a
102 * racy try_to_unuse/swapoff) we need an additional reference
103 * count to avoid reading garbage from page_private(page) above.
104 * If the WARN_ON triggers during a swapoff it maybe the race
105 * condition and it's harmless. However if it triggers without
106 * swapoff it signals a problem.
108 WARN_ON(page_count(page) <= 1);
110 bdi = bdev->bd_inode->i_mapping->backing_dev_info;
111 blk_run_backing_dev(bdi, page);
113 up_read(&swap_unplug_sem);
117 * swapon tell device that all the old swap contents can be discarded,
118 * to allow the swap device to optimize its wear-levelling.
120 static int discard_swap(struct swap_info_struct *si)
122 struct swap_extent *se;
125 list_for_each_entry(se, &si->extent_list, list) {
126 sector_t start_block = se->start_block << (PAGE_SHIFT - 9);
127 sector_t nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
129 if (se->start_page == 0) {
130 /* Do not discard the swap header page! */
131 start_block += 1 << (PAGE_SHIFT - 9);
132 nr_blocks -= 1 << (PAGE_SHIFT - 9);
137 err = blkdev_issue_discard(si->bdev, start_block,
138 nr_blocks, GFP_KERNEL);
144 return err; /* That will often be -EOPNOTSUPP */
148 * swap allocation tell device that a cluster of swap can now be discarded,
149 * to allow the swap device to optimize its wear-levelling.
151 static void discard_swap_cluster(struct swap_info_struct *si,
152 pgoff_t start_page, pgoff_t nr_pages)
154 struct swap_extent *se = si->curr_swap_extent;
155 int found_extent = 0;
158 struct list_head *lh;
160 if (se->start_page <= start_page &&
161 start_page < se->start_page + se->nr_pages) {
162 pgoff_t offset = start_page - se->start_page;
163 sector_t start_block = se->start_block + offset;
164 sector_t nr_blocks = se->nr_pages - offset;
166 if (nr_blocks > nr_pages)
167 nr_blocks = nr_pages;
168 start_page += nr_blocks;
169 nr_pages -= nr_blocks;
172 si->curr_swap_extent = se;
174 start_block <<= PAGE_SHIFT - 9;
175 nr_blocks <<= PAGE_SHIFT - 9;
176 if (blkdev_issue_discard(si->bdev, start_block,
177 nr_blocks, GFP_NOIO))
182 if (lh == &si->extent_list)
184 se = list_entry(lh, struct swap_extent, list);
188 static int wait_for_discard(void *word)
194 #define SWAPFILE_CLUSTER 256
195 #define LATENCY_LIMIT 256
197 static inline unsigned long scan_swap_map(struct swap_info_struct *si,
200 unsigned long offset;
201 unsigned long scan_base;
202 unsigned long last_in_cluster = 0;
203 int latency_ration = LATENCY_LIMIT;
204 int found_free_cluster = 0;
207 * We try to cluster swap pages by allocating them sequentially
208 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
209 * way, however, we resort to first-free allocation, starting
210 * a new cluster. This prevents us from scattering swap pages
211 * all over the entire swap partition, so that we reduce
212 * overall disk seek times between swap pages. -- sct
213 * But we do now try to find an empty cluster. -Andrea
214 * And we let swap pages go all over an SSD partition. Hugh
217 si->flags += SWP_SCANNING;
218 scan_base = offset = si->cluster_next;
220 if (unlikely(!si->cluster_nr--)) {
221 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
222 si->cluster_nr = SWAPFILE_CLUSTER - 1;
225 if (si->flags & SWP_DISCARDABLE) {
227 * Start range check on racing allocations, in case
228 * they overlap the cluster we eventually decide on
229 * (we scan without swap_lock to allow preemption).
230 * It's hardly conceivable that cluster_nr could be
231 * wrapped during our scan, but don't depend on it.
233 if (si->lowest_alloc)
235 si->lowest_alloc = si->max;
236 si->highest_alloc = 0;
238 spin_unlock(&swap_lock);
241 * If seek is expensive, start searching for new cluster from
242 * start of partition, to minimize the span of allocated swap.
243 * But if seek is cheap, search from our current position, so
244 * that swap is allocated from all over the partition: if the
245 * Flash Translation Layer only remaps within limited zones,
246 * we don't want to wear out the first zone too quickly.
248 if (!(si->flags & SWP_SOLIDSTATE))
249 scan_base = offset = si->lowest_bit;
250 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
252 /* Locate the first empty (unaligned) cluster */
253 for (; last_in_cluster <= si->highest_bit; offset++) {
254 if (si->swap_map[offset])
255 last_in_cluster = offset + SWAPFILE_CLUSTER;
256 else if (offset == last_in_cluster) {
257 spin_lock(&swap_lock);
258 offset -= SWAPFILE_CLUSTER - 1;
259 si->cluster_next = offset;
260 si->cluster_nr = SWAPFILE_CLUSTER - 1;
261 found_free_cluster = 1;
264 if (unlikely(--latency_ration < 0)) {
266 latency_ration = LATENCY_LIMIT;
270 offset = si->lowest_bit;
271 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
273 /* Locate the first empty (unaligned) cluster */
274 for (; last_in_cluster < scan_base; offset++) {
275 if (si->swap_map[offset])
276 last_in_cluster = offset + SWAPFILE_CLUSTER;
277 else if (offset == last_in_cluster) {
278 spin_lock(&swap_lock);
279 offset -= SWAPFILE_CLUSTER - 1;
280 si->cluster_next = offset;
281 si->cluster_nr = SWAPFILE_CLUSTER - 1;
282 found_free_cluster = 1;
285 if (unlikely(--latency_ration < 0)) {
287 latency_ration = LATENCY_LIMIT;
292 spin_lock(&swap_lock);
293 si->cluster_nr = SWAPFILE_CLUSTER - 1;
294 si->lowest_alloc = 0;
298 if (!(si->flags & SWP_WRITEOK))
300 if (!si->highest_bit)
302 if (offset > si->highest_bit)
303 scan_base = offset = si->lowest_bit;
304 if (si->swap_map[offset])
307 if (offset == si->lowest_bit)
309 if (offset == si->highest_bit)
312 if (si->inuse_pages == si->pages) {
313 si->lowest_bit = si->max;
316 if (cache == SWAP_CACHE) /* at usual swap-out via vmscan.c */
317 si->swap_map[offset] = encode_swapmap(0, true);
318 else /* at suspend */
319 si->swap_map[offset] = encode_swapmap(1, false);
320 si->cluster_next = offset + 1;
321 si->flags -= SWP_SCANNING;
323 if (si->lowest_alloc) {
325 * Only set when SWP_DISCARDABLE, and there's a scan
326 * for a free cluster in progress or just completed.
328 if (found_free_cluster) {
330 * To optimize wear-levelling, discard the
331 * old data of the cluster, taking care not to
332 * discard any of its pages that have already
333 * been allocated by racing tasks (offset has
334 * already stepped over any at the beginning).
336 if (offset < si->highest_alloc &&
337 si->lowest_alloc <= last_in_cluster)
338 last_in_cluster = si->lowest_alloc - 1;
339 si->flags |= SWP_DISCARDING;
340 spin_unlock(&swap_lock);
342 if (offset < last_in_cluster)
343 discard_swap_cluster(si, offset,
344 last_in_cluster - offset + 1);
346 spin_lock(&swap_lock);
347 si->lowest_alloc = 0;
348 si->flags &= ~SWP_DISCARDING;
350 smp_mb(); /* wake_up_bit advises this */
351 wake_up_bit(&si->flags, ilog2(SWP_DISCARDING));
353 } else if (si->flags & SWP_DISCARDING) {
355 * Delay using pages allocated by racing tasks
356 * until the whole discard has been issued. We
357 * could defer that delay until swap_writepage,
358 * but it's easier to keep this self-contained.
360 spin_unlock(&swap_lock);
361 wait_on_bit(&si->flags, ilog2(SWP_DISCARDING),
362 wait_for_discard, TASK_UNINTERRUPTIBLE);
363 spin_lock(&swap_lock);
366 * Note pages allocated by racing tasks while
367 * scan for a free cluster is in progress, so
368 * that its final discard can exclude them.
370 if (offset < si->lowest_alloc)
371 si->lowest_alloc = offset;
372 if (offset > si->highest_alloc)
373 si->highest_alloc = offset;
379 spin_unlock(&swap_lock);
380 while (++offset <= si->highest_bit) {
381 if (!si->swap_map[offset]) {
382 spin_lock(&swap_lock);
385 if (unlikely(--latency_ration < 0)) {
387 latency_ration = LATENCY_LIMIT;
390 offset = si->lowest_bit;
391 while (++offset < scan_base) {
392 if (!si->swap_map[offset]) {
393 spin_lock(&swap_lock);
396 if (unlikely(--latency_ration < 0)) {
398 latency_ration = LATENCY_LIMIT;
401 spin_lock(&swap_lock);
404 si->flags -= SWP_SCANNING;
408 swp_entry_t get_swap_page(void)
410 struct swap_info_struct *si;
415 spin_lock(&swap_lock);
416 if (nr_swap_pages <= 0)
420 for (type = swap_list.next; type >= 0 && wrapped < 2; type = next) {
421 si = swap_info + type;
424 (!wrapped && si->prio != swap_info[next].prio)) {
425 next = swap_list.head;
429 if (!si->highest_bit)
431 if (!(si->flags & SWP_WRITEOK))
434 swap_list.next = next;
435 /* This is called for allocating swap entry for cache */
436 offset = scan_swap_map(si, SWAP_CACHE);
438 spin_unlock(&swap_lock);
439 return swp_entry(type, offset);
441 next = swap_list.next;
446 spin_unlock(&swap_lock);
447 return (swp_entry_t) {0};
450 /* The only caller of this function is now susupend routine */
451 swp_entry_t get_swap_page_of_type(int type)
453 struct swap_info_struct *si;
456 spin_lock(&swap_lock);
457 si = swap_info + type;
458 if (si->flags & SWP_WRITEOK) {
460 /* This is called for allocating swap entry, not cache */
461 offset = scan_swap_map(si, SWAP_MAP);
463 spin_unlock(&swap_lock);
464 return swp_entry(type, offset);
468 spin_unlock(&swap_lock);
469 return (swp_entry_t) {0};
472 static struct swap_info_struct * swap_info_get(swp_entry_t entry)
474 struct swap_info_struct * p;
475 unsigned long offset, type;
479 type = swp_type(entry);
480 if (type >= nr_swapfiles)
482 p = & swap_info[type];
483 if (!(p->flags & SWP_USED))
485 offset = swp_offset(entry);
486 if (offset >= p->max)
488 if (!p->swap_map[offset])
490 spin_lock(&swap_lock);
494 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val);
497 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val);
500 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val);
503 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val);
508 static int swap_entry_free(struct swap_info_struct *p,
509 swp_entry_t ent, int cache)
511 unsigned long offset = swp_offset(ent);
512 int count = swap_count(p->swap_map[offset]);
515 has_cache = swap_has_cache(p->swap_map[offset]);
517 if (cache == SWAP_MAP) { /* dropping usage count of swap */
518 if (count < SWAP_MAP_MAX) {
520 p->swap_map[offset] = encode_swapmap(count, has_cache);
522 } else { /* dropping swap cache flag */
523 VM_BUG_ON(!has_cache);
524 p->swap_map[offset] = encode_swapmap(count, false);
528 count = p->swap_map[offset];
529 /* free if no reference */
531 if (offset < p->lowest_bit)
532 p->lowest_bit = offset;
533 if (offset > p->highest_bit)
534 p->highest_bit = offset;
535 if (p->prio > swap_info[swap_list.next].prio)
536 swap_list.next = p - swap_info;
539 mem_cgroup_uncharge_swap(ent);
545 * Caller has made sure that the swapdevice corresponding to entry
546 * is still around or has not been recycled.
548 void swap_free(swp_entry_t entry)
550 struct swap_info_struct * p;
552 p = swap_info_get(entry);
554 swap_entry_free(p, entry, SWAP_MAP);
555 spin_unlock(&swap_lock);
560 * Called after dropping swapcache to decrease refcnt to swap entries.
562 void swapcache_free(swp_entry_t entry, struct page *page)
564 struct swap_info_struct *p;
567 mem_cgroup_uncharge_swapcache(page, entry);
568 p = swap_info_get(entry);
570 swap_entry_free(p, entry, SWAP_CACHE);
571 spin_unlock(&swap_lock);
577 * How many references to page are currently swapped out?
579 static inline int page_swapcount(struct page *page)
582 struct swap_info_struct *p;
585 entry.val = page_private(page);
586 p = swap_info_get(entry);
588 count = swap_count(p->swap_map[swp_offset(entry)]);
589 spin_unlock(&swap_lock);
595 * We can write to an anon page without COW if there are no other references
596 * to it. And as a side-effect, free up its swap: because the old content
597 * on disk will never be read, and seeking back there to write new content
598 * later would only waste time away from clustering.
600 int reuse_swap_page(struct page *page)
604 VM_BUG_ON(!PageLocked(page));
605 count = page_mapcount(page);
606 if (count <= 1 && PageSwapCache(page)) {
607 count += page_swapcount(page);
608 if (count == 1 && !PageWriteback(page)) {
609 delete_from_swap_cache(page);
617 * If swap is getting full, or if there are no more mappings of this page,
618 * then try_to_free_swap is called to free its swap space.
620 int try_to_free_swap(struct page *page)
622 VM_BUG_ON(!PageLocked(page));
624 if (!PageSwapCache(page))
626 if (PageWriteback(page))
628 if (page_swapcount(page))
631 delete_from_swap_cache(page);
637 * Free the swap entry like above, but also try to
638 * free the page cache entry if it is the last user.
640 int free_swap_and_cache(swp_entry_t entry)
642 struct swap_info_struct *p;
643 struct page *page = NULL;
645 if (is_migration_entry(entry))
648 p = swap_info_get(entry);
650 if (swap_entry_free(p, entry, SWAP_MAP) == SWAP_HAS_CACHE) {
651 page = find_get_page(&swapper_space, entry.val);
652 if (page && !trylock_page(page)) {
653 page_cache_release(page);
657 spin_unlock(&swap_lock);
661 * Not mapped elsewhere, or swap space full? Free it!
662 * Also recheck PageSwapCache now page is locked (above).
664 if (PageSwapCache(page) && !PageWriteback(page) &&
665 (!page_mapped(page) || vm_swap_full())) {
666 delete_from_swap_cache(page);
670 page_cache_release(page);
675 #ifdef CONFIG_HIBERNATION
677 * Find the swap type that corresponds to given device (if any).
679 * @offset - number of the PAGE_SIZE-sized block of the device, starting
680 * from 0, in which the swap header is expected to be located.
682 * This is needed for the suspend to disk (aka swsusp).
684 int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
686 struct block_device *bdev = NULL;
690 bdev = bdget(device);
692 spin_lock(&swap_lock);
693 for (i = 0; i < nr_swapfiles; i++) {
694 struct swap_info_struct *sis = swap_info + i;
696 if (!(sis->flags & SWP_WRITEOK))
701 *bdev_p = bdget(sis->bdev->bd_dev);
703 spin_unlock(&swap_lock);
706 if (bdev == sis->bdev) {
707 struct swap_extent *se;
709 se = list_entry(sis->extent_list.next,
710 struct swap_extent, list);
711 if (se->start_block == offset) {
713 *bdev_p = bdget(sis->bdev->bd_dev);
715 spin_unlock(&swap_lock);
721 spin_unlock(&swap_lock);
729 * Return either the total number of swap pages of given type, or the number
730 * of free pages of that type (depending on @free)
732 * This is needed for software suspend
734 unsigned int count_swap_pages(int type, int free)
738 if (type < nr_swapfiles) {
739 spin_lock(&swap_lock);
740 if (swap_info[type].flags & SWP_WRITEOK) {
741 n = swap_info[type].pages;
743 n -= swap_info[type].inuse_pages;
745 spin_unlock(&swap_lock);
752 * No need to decide whether this PTE shares the swap entry with others,
753 * just let do_wp_page work it out if a write is requested later - to
754 * force COW, vm_page_prot omits write permission from any private vma.
756 static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
757 unsigned long addr, swp_entry_t entry, struct page *page)
759 struct mem_cgroup *ptr = NULL;
764 if (mem_cgroup_try_charge_swapin(vma->vm_mm, page, GFP_KERNEL, &ptr)) {
769 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
770 if (unlikely(!pte_same(*pte, swp_entry_to_pte(entry)))) {
772 mem_cgroup_cancel_charge_swapin(ptr);
777 inc_mm_counter(vma->vm_mm, anon_rss);
779 set_pte_at(vma->vm_mm, addr, pte,
780 pte_mkold(mk_pte(page, vma->vm_page_prot)));
781 page_add_anon_rmap(page, vma, addr);
782 mem_cgroup_commit_charge_swapin(page, ptr);
785 * Move the page to the active list so it is not
786 * immediately swapped out again after swapon.
790 pte_unmap_unlock(pte, ptl);
795 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
796 unsigned long addr, unsigned long end,
797 swp_entry_t entry, struct page *page)
799 pte_t swp_pte = swp_entry_to_pte(entry);
804 * We don't actually need pte lock while scanning for swp_pte: since
805 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
806 * page table while we're scanning; though it could get zapped, and on
807 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
808 * of unmatched parts which look like swp_pte, so unuse_pte must
809 * recheck under pte lock. Scanning without pte lock lets it be
810 * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
812 pte = pte_offset_map(pmd, addr);
815 * swapoff spends a _lot_ of time in this loop!
816 * Test inline before going to call unuse_pte.
818 if (unlikely(pte_same(*pte, swp_pte))) {
820 ret = unuse_pte(vma, pmd, addr, entry, page);
823 pte = pte_offset_map(pmd, addr);
825 } while (pte++, addr += PAGE_SIZE, addr != end);
831 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
832 unsigned long addr, unsigned long end,
833 swp_entry_t entry, struct page *page)
839 pmd = pmd_offset(pud, addr);
841 next = pmd_addr_end(addr, end);
842 if (pmd_none_or_clear_bad(pmd))
844 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
847 } while (pmd++, addr = next, addr != end);
851 static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
852 unsigned long addr, unsigned long end,
853 swp_entry_t entry, struct page *page)
859 pud = pud_offset(pgd, addr);
861 next = pud_addr_end(addr, end);
862 if (pud_none_or_clear_bad(pud))
864 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
867 } while (pud++, addr = next, addr != end);
871 static int unuse_vma(struct vm_area_struct *vma,
872 swp_entry_t entry, struct page *page)
875 unsigned long addr, end, next;
879 addr = page_address_in_vma(page, vma);
883 end = addr + PAGE_SIZE;
885 addr = vma->vm_start;
889 pgd = pgd_offset(vma->vm_mm, addr);
891 next = pgd_addr_end(addr, end);
892 if (pgd_none_or_clear_bad(pgd))
894 ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
897 } while (pgd++, addr = next, addr != end);
901 static int unuse_mm(struct mm_struct *mm,
902 swp_entry_t entry, struct page *page)
904 struct vm_area_struct *vma;
907 if (!down_read_trylock(&mm->mmap_sem)) {
909 * Activate page so shrink_inactive_list is unlikely to unmap
910 * its ptes while lock is dropped, so swapoff can make progress.
914 down_read(&mm->mmap_sem);
917 for (vma = mm->mmap; vma; vma = vma->vm_next) {
918 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
921 up_read(&mm->mmap_sem);
922 return (ret < 0)? ret: 0;
926 * Scan swap_map from current position to next entry still in use.
927 * Recycle to start on reaching the end, returning 0 when empty.
929 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
932 unsigned int max = si->max;
933 unsigned int i = prev;
937 * No need for swap_lock here: we're just looking
938 * for whether an entry is in use, not modifying it; false
939 * hits are okay, and sys_swapoff() has already prevented new
940 * allocations from this area (while holding swap_lock).
949 * No entries in use at top of swap_map,
950 * loop back to start and recheck there.
956 count = si->swap_map[i];
957 if (count && swap_count(count) != SWAP_MAP_BAD)
964 * We completely avoid races by reading each swap page in advance,
965 * and then search for the process using it. All the necessary
966 * page table adjustments can then be made atomically.
968 static int try_to_unuse(unsigned int type)
970 struct swap_info_struct * si = &swap_info[type];
971 struct mm_struct *start_mm;
972 unsigned short *swap_map;
973 unsigned short swcount;
978 int reset_overflow = 0;
982 * When searching mms for an entry, a good strategy is to
983 * start at the first mm we freed the previous entry from
984 * (though actually we don't notice whether we or coincidence
985 * freed the entry). Initialize this start_mm with a hold.
987 * A simpler strategy would be to start at the last mm we
988 * freed the previous entry from; but that would take less
989 * advantage of mmlist ordering, which clusters forked mms
990 * together, child after parent. If we race with dup_mmap(), we
991 * prefer to resolve parent before child, lest we miss entries
992 * duplicated after we scanned child: using last mm would invert
993 * that. Though it's only a serious concern when an overflowed
994 * swap count is reset from SWAP_MAP_MAX, preventing a rescan.
997 atomic_inc(&init_mm.mm_users);
1000 * Keep on scanning until all entries have gone. Usually,
1001 * one pass through swap_map is enough, but not necessarily:
1002 * there are races when an instance of an entry might be missed.
1004 while ((i = find_next_to_unuse(si, i)) != 0) {
1005 if (signal_pending(current)) {
1011 * Get a page for the entry, using the existing swap
1012 * cache page if there is one. Otherwise, get a clean
1013 * page and read the swap into it.
1015 swap_map = &si->swap_map[i];
1016 entry = swp_entry(type, i);
1017 page = read_swap_cache_async(entry,
1018 GFP_HIGHUSER_MOVABLE, NULL, 0);
1021 * Either swap_duplicate() failed because entry
1022 * has been freed independently, and will not be
1023 * reused since sys_swapoff() already disabled
1024 * allocation from here, or alloc_page() failed.
1033 * Don't hold on to start_mm if it looks like exiting.
1035 if (atomic_read(&start_mm->mm_users) == 1) {
1037 start_mm = &init_mm;
1038 atomic_inc(&init_mm.mm_users);
1042 * Wait for and lock page. When do_swap_page races with
1043 * try_to_unuse, do_swap_page can handle the fault much
1044 * faster than try_to_unuse can locate the entry. This
1045 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1046 * defer to do_swap_page in such a case - in some tests,
1047 * do_swap_page and try_to_unuse repeatedly compete.
1049 wait_on_page_locked(page);
1050 wait_on_page_writeback(page);
1052 wait_on_page_writeback(page);
1055 * Remove all references to entry.
1056 * Whenever we reach init_mm, there's no address space
1057 * to search, but use it as a reminder to search shmem.
1060 swcount = *swap_map;
1061 if (swap_count(swcount)) {
1062 if (start_mm == &init_mm)
1063 shmem = shmem_unuse(entry, page);
1065 retval = unuse_mm(start_mm, entry, page);
1067 if (swap_count(*swap_map)) {
1068 int set_start_mm = (*swap_map >= swcount);
1069 struct list_head *p = &start_mm->mmlist;
1070 struct mm_struct *new_start_mm = start_mm;
1071 struct mm_struct *prev_mm = start_mm;
1072 struct mm_struct *mm;
1074 atomic_inc(&new_start_mm->mm_users);
1075 atomic_inc(&prev_mm->mm_users);
1076 spin_lock(&mmlist_lock);
1077 while (swap_count(*swap_map) && !retval && !shmem &&
1078 (p = p->next) != &start_mm->mmlist) {
1079 mm = list_entry(p, struct mm_struct, mmlist);
1080 if (!atomic_inc_not_zero(&mm->mm_users))
1082 spin_unlock(&mmlist_lock);
1088 swcount = *swap_map;
1089 if (!swap_count(swcount)) /* any usage ? */
1091 else if (mm == &init_mm) {
1093 shmem = shmem_unuse(entry, page);
1095 retval = unuse_mm(mm, entry, page);
1098 swap_count(*swap_map) < swcount) {
1099 mmput(new_start_mm);
1100 atomic_inc(&mm->mm_users);
1104 spin_lock(&mmlist_lock);
1106 spin_unlock(&mmlist_lock);
1109 start_mm = new_start_mm;
1112 /* page has already been unlocked and released */
1120 page_cache_release(page);
1125 * How could swap count reach 0x7ffe ?
1126 * There's no way to repeat a swap page within an mm
1127 * (except in shmem, where it's the shared object which takes
1128 * the reference count)?
1129 * We believe SWAP_MAP_MAX cannot occur.(if occur, unsigned
1130 * short is too small....)
1131 * If that's wrong, then we should worry more about
1132 * exit_mmap() and do_munmap() cases described above:
1133 * we might be resetting SWAP_MAP_MAX too early here.
1134 * We know "Undead"s can happen, they're okay, so don't
1135 * report them; but do report if we reset SWAP_MAP_MAX.
1137 /* We might release the lock_page() in unuse_mm(). */
1138 if (!PageSwapCache(page) || page_private(page) != entry.val)
1141 if (swap_count(*swap_map) == SWAP_MAP_MAX) {
1142 spin_lock(&swap_lock);
1143 *swap_map = encode_swapmap(0, true);
1144 spin_unlock(&swap_lock);
1149 * If a reference remains (rare), we would like to leave
1150 * the page in the swap cache; but try_to_unmap could
1151 * then re-duplicate the entry once we drop page lock,
1152 * so we might loop indefinitely; also, that page could
1153 * not be swapped out to other storage meanwhile. So:
1154 * delete from cache even if there's another reference,
1155 * after ensuring that the data has been saved to disk -
1156 * since if the reference remains (rarer), it will be
1157 * read from disk into another page. Splitting into two
1158 * pages would be incorrect if swap supported "shared
1159 * private" pages, but they are handled by tmpfs files.
1161 if (swap_count(*swap_map) &&
1162 PageDirty(page) && PageSwapCache(page)) {
1163 struct writeback_control wbc = {
1164 .sync_mode = WB_SYNC_NONE,
1167 swap_writepage(page, &wbc);
1169 wait_on_page_writeback(page);
1173 * It is conceivable that a racing task removed this page from
1174 * swap cache just before we acquired the page lock at the top,
1175 * or while we dropped it in unuse_mm(). The page might even
1176 * be back in swap cache on another swap area: that we must not
1177 * delete, since it may not have been written out to swap yet.
1179 if (PageSwapCache(page) &&
1180 likely(page_private(page) == entry.val))
1181 delete_from_swap_cache(page);
1184 * So we could skip searching mms once swap count went
1185 * to 1, we did not mark any present ptes as dirty: must
1186 * mark page dirty so shrink_page_list will preserve it.
1191 page_cache_release(page);
1194 * Make sure that we aren't completely killing
1195 * interactive performance.
1201 if (reset_overflow) {
1202 printk(KERN_WARNING "swapoff: cleared swap entry overflow\n");
1209 * After a successful try_to_unuse, if no swap is now in use, we know
1210 * we can empty the mmlist. swap_lock must be held on entry and exit.
1211 * Note that mmlist_lock nests inside swap_lock, and an mm must be
1212 * added to the mmlist just after page_duplicate - before would be racy.
1214 static void drain_mmlist(void)
1216 struct list_head *p, *next;
1219 for (i = 0; i < nr_swapfiles; i++)
1220 if (swap_info[i].inuse_pages)
1222 spin_lock(&mmlist_lock);
1223 list_for_each_safe(p, next, &init_mm.mmlist)
1225 spin_unlock(&mmlist_lock);
1229 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
1230 * corresponds to page offset `offset'.
1232 sector_t map_swap_page(struct swap_info_struct *sis, pgoff_t offset)
1234 struct swap_extent *se = sis->curr_swap_extent;
1235 struct swap_extent *start_se = se;
1238 struct list_head *lh;
1240 if (se->start_page <= offset &&
1241 offset < (se->start_page + se->nr_pages)) {
1242 return se->start_block + (offset - se->start_page);
1245 if (lh == &sis->extent_list)
1247 se = list_entry(lh, struct swap_extent, list);
1248 sis->curr_swap_extent = se;
1249 BUG_ON(se == start_se); /* It *must* be present */
1253 #ifdef CONFIG_HIBERNATION
1255 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1256 * corresponding to given index in swap_info (swap type).
1258 sector_t swapdev_block(int swap_type, pgoff_t offset)
1260 struct swap_info_struct *sis;
1262 if (swap_type >= nr_swapfiles)
1265 sis = swap_info + swap_type;
1266 return (sis->flags & SWP_WRITEOK) ? map_swap_page(sis, offset) : 0;
1268 #endif /* CONFIG_HIBERNATION */
1271 * Free all of a swapdev's extent information
1273 static void destroy_swap_extents(struct swap_info_struct *sis)
1275 while (!list_empty(&sis->extent_list)) {
1276 struct swap_extent *se;
1278 se = list_entry(sis->extent_list.next,
1279 struct swap_extent, list);
1280 list_del(&se->list);
1286 * Add a block range (and the corresponding page range) into this swapdev's
1287 * extent list. The extent list is kept sorted in page order.
1289 * This function rather assumes that it is called in ascending page order.
1292 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1293 unsigned long nr_pages, sector_t start_block)
1295 struct swap_extent *se;
1296 struct swap_extent *new_se;
1297 struct list_head *lh;
1299 lh = sis->extent_list.prev; /* The highest page extent */
1300 if (lh != &sis->extent_list) {
1301 se = list_entry(lh, struct swap_extent, list);
1302 BUG_ON(se->start_page + se->nr_pages != start_page);
1303 if (se->start_block + se->nr_pages == start_block) {
1305 se->nr_pages += nr_pages;
1311 * No merge. Insert a new extent, preserving ordering.
1313 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1316 new_se->start_page = start_page;
1317 new_se->nr_pages = nr_pages;
1318 new_se->start_block = start_block;
1320 list_add_tail(&new_se->list, &sis->extent_list);
1325 * A `swap extent' is a simple thing which maps a contiguous range of pages
1326 * onto a contiguous range of disk blocks. An ordered list of swap extents
1327 * is built at swapon time and is then used at swap_writepage/swap_readpage
1328 * time for locating where on disk a page belongs.
1330 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1331 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1332 * swap files identically.
1334 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1335 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1336 * swapfiles are handled *identically* after swapon time.
1338 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1339 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1340 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1341 * requirements, they are simply tossed out - we will never use those blocks
1344 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
1345 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1346 * which will scribble on the fs.
1348 * The amount of disk space which a single swap extent represents varies.
1349 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1350 * extents in the list. To avoid much list walking, we cache the previous
1351 * search location in `curr_swap_extent', and start new searches from there.
1352 * This is extremely effective. The average number of iterations in
1353 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1355 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
1357 struct inode *inode;
1358 unsigned blocks_per_page;
1359 unsigned long page_no;
1361 sector_t probe_block;
1362 sector_t last_block;
1363 sector_t lowest_block = -1;
1364 sector_t highest_block = 0;
1368 inode = sis->swap_file->f_mapping->host;
1369 if (S_ISBLK(inode->i_mode)) {
1370 ret = add_swap_extent(sis, 0, sis->max, 0);
1375 blkbits = inode->i_blkbits;
1376 blocks_per_page = PAGE_SIZE >> blkbits;
1379 * Map all the blocks into the extent list. This code doesn't try
1384 last_block = i_size_read(inode) >> blkbits;
1385 while ((probe_block + blocks_per_page) <= last_block &&
1386 page_no < sis->max) {
1387 unsigned block_in_page;
1388 sector_t first_block;
1390 first_block = bmap(inode, probe_block);
1391 if (first_block == 0)
1395 * It must be PAGE_SIZE aligned on-disk
1397 if (first_block & (blocks_per_page - 1)) {
1402 for (block_in_page = 1; block_in_page < blocks_per_page;
1406 block = bmap(inode, probe_block + block_in_page);
1409 if (block != first_block + block_in_page) {
1416 first_block >>= (PAGE_SHIFT - blkbits);
1417 if (page_no) { /* exclude the header page */
1418 if (first_block < lowest_block)
1419 lowest_block = first_block;
1420 if (first_block > highest_block)
1421 highest_block = first_block;
1425 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1427 ret = add_swap_extent(sis, page_no, 1, first_block);
1432 probe_block += blocks_per_page;
1437 *span = 1 + highest_block - lowest_block;
1439 page_no = 1; /* force Empty message */
1441 sis->pages = page_no - 1;
1442 sis->highest_bit = page_no - 1;
1444 sis->curr_swap_extent = list_entry(sis->extent_list.prev,
1445 struct swap_extent, list);
1448 printk(KERN_ERR "swapon: swapfile has holes\n");
1454 SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
1456 struct swap_info_struct * p = NULL;
1457 unsigned short *swap_map;
1458 struct file *swap_file, *victim;
1459 struct address_space *mapping;
1460 struct inode *inode;
1465 if (!capable(CAP_SYS_ADMIN))
1468 pathname = getname(specialfile);
1469 err = PTR_ERR(pathname);
1470 if (IS_ERR(pathname))
1473 victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0);
1475 err = PTR_ERR(victim);
1479 mapping = victim->f_mapping;
1481 spin_lock(&swap_lock);
1482 for (type = swap_list.head; type >= 0; type = swap_info[type].next) {
1483 p = swap_info + type;
1484 if (p->flags & SWP_WRITEOK) {
1485 if (p->swap_file->f_mapping == mapping)
1492 spin_unlock(&swap_lock);
1495 if (!security_vm_enough_memory(p->pages))
1496 vm_unacct_memory(p->pages);
1499 spin_unlock(&swap_lock);
1503 swap_list.head = p->next;
1505 swap_info[prev].next = p->next;
1507 if (type == swap_list.next) {
1508 /* just pick something that's safe... */
1509 swap_list.next = swap_list.head;
1512 for (i = p->next; i >= 0; i = swap_info[i].next)
1513 swap_info[i].prio = p->prio--;
1516 nr_swap_pages -= p->pages;
1517 total_swap_pages -= p->pages;
1518 p->flags &= ~SWP_WRITEOK;
1519 spin_unlock(&swap_lock);
1521 current->flags |= PF_SWAPOFF;
1522 err = try_to_unuse(type);
1523 current->flags &= ~PF_SWAPOFF;
1526 /* re-insert swap space back into swap_list */
1527 spin_lock(&swap_lock);
1529 p->prio = --least_priority;
1531 for (i = swap_list.head; i >= 0; i = swap_info[i].next) {
1532 if (p->prio >= swap_info[i].prio)
1538 swap_list.head = swap_list.next = p - swap_info;
1540 swap_info[prev].next = p - swap_info;
1541 nr_swap_pages += p->pages;
1542 total_swap_pages += p->pages;
1543 p->flags |= SWP_WRITEOK;
1544 spin_unlock(&swap_lock);
1548 /* wait for any unplug function to finish */
1549 down_write(&swap_unplug_sem);
1550 up_write(&swap_unplug_sem);
1552 destroy_swap_extents(p);
1553 mutex_lock(&swapon_mutex);
1554 spin_lock(&swap_lock);
1557 /* wait for anyone still in scan_swap_map */
1558 p->highest_bit = 0; /* cuts scans short */
1559 while (p->flags >= SWP_SCANNING) {
1560 spin_unlock(&swap_lock);
1561 schedule_timeout_uninterruptible(1);
1562 spin_lock(&swap_lock);
1565 swap_file = p->swap_file;
1566 p->swap_file = NULL;
1568 swap_map = p->swap_map;
1571 spin_unlock(&swap_lock);
1572 mutex_unlock(&swapon_mutex);
1574 /* Destroy swap account informatin */
1575 swap_cgroup_swapoff(type);
1577 inode = mapping->host;
1578 if (S_ISBLK(inode->i_mode)) {
1579 struct block_device *bdev = I_BDEV(inode);
1580 set_blocksize(bdev, p->old_block_size);
1583 mutex_lock(&inode->i_mutex);
1584 inode->i_flags &= ~S_SWAPFILE;
1585 mutex_unlock(&inode->i_mutex);
1587 filp_close(swap_file, NULL);
1591 filp_close(victim, NULL);
1596 #ifdef CONFIG_PROC_FS
1598 static void *swap_start(struct seq_file *swap, loff_t *pos)
1600 struct swap_info_struct *ptr = swap_info;
1604 mutex_lock(&swapon_mutex);
1607 return SEQ_START_TOKEN;
1609 for (i = 0; i < nr_swapfiles; i++, ptr++) {
1610 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1619 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1621 struct swap_info_struct *ptr;
1622 struct swap_info_struct *endptr = swap_info + nr_swapfiles;
1624 if (v == SEQ_START_TOKEN)
1631 for (; ptr < endptr; ptr++) {
1632 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1641 static void swap_stop(struct seq_file *swap, void *v)
1643 mutex_unlock(&swapon_mutex);
1646 static int swap_show(struct seq_file *swap, void *v)
1648 struct swap_info_struct *ptr = v;
1652 if (ptr == SEQ_START_TOKEN) {
1653 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1657 file = ptr->swap_file;
1658 len = seq_path(swap, &file->f_path, " \t\n\\");
1659 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
1660 len < 40 ? 40 - len : 1, " ",
1661 S_ISBLK(file->f_path.dentry->d_inode->i_mode) ?
1662 "partition" : "file\t",
1663 ptr->pages << (PAGE_SHIFT - 10),
1664 ptr->inuse_pages << (PAGE_SHIFT - 10),
1669 static const struct seq_operations swaps_op = {
1670 .start = swap_start,
1676 static int swaps_open(struct inode *inode, struct file *file)
1678 return seq_open(file, &swaps_op);
1681 static const struct file_operations proc_swaps_operations = {
1684 .llseek = seq_lseek,
1685 .release = seq_release,
1688 static int __init procswaps_init(void)
1690 proc_create("swaps", 0, NULL, &proc_swaps_operations);
1693 __initcall(procswaps_init);
1694 #endif /* CONFIG_PROC_FS */
1696 #ifdef MAX_SWAPFILES_CHECK
1697 static int __init max_swapfiles_check(void)
1699 MAX_SWAPFILES_CHECK();
1702 late_initcall(max_swapfiles_check);
1706 * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
1708 * The swapon system call
1710 SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
1712 struct swap_info_struct * p;
1714 struct block_device *bdev = NULL;
1715 struct file *swap_file = NULL;
1716 struct address_space *mapping;
1720 union swap_header *swap_header = NULL;
1721 unsigned int nr_good_pages = 0;
1724 unsigned long maxpages = 1;
1725 unsigned long swapfilepages;
1726 unsigned short *swap_map = NULL;
1727 struct page *page = NULL;
1728 struct inode *inode = NULL;
1731 if (!capable(CAP_SYS_ADMIN))
1733 spin_lock(&swap_lock);
1735 for (type = 0 ; type < nr_swapfiles ; type++,p++)
1736 if (!(p->flags & SWP_USED))
1739 if (type >= MAX_SWAPFILES) {
1740 spin_unlock(&swap_lock);
1743 if (type >= nr_swapfiles)
1744 nr_swapfiles = type+1;
1745 memset(p, 0, sizeof(*p));
1746 INIT_LIST_HEAD(&p->extent_list);
1747 p->flags = SWP_USED;
1749 spin_unlock(&swap_lock);
1750 name = getname(specialfile);
1751 error = PTR_ERR(name);
1756 swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0);
1757 error = PTR_ERR(swap_file);
1758 if (IS_ERR(swap_file)) {
1763 p->swap_file = swap_file;
1764 mapping = swap_file->f_mapping;
1765 inode = mapping->host;
1768 for (i = 0; i < nr_swapfiles; i++) {
1769 struct swap_info_struct *q = &swap_info[i];
1771 if (i == type || !q->swap_file)
1773 if (mapping == q->swap_file->f_mapping)
1778 if (S_ISBLK(inode->i_mode)) {
1779 bdev = I_BDEV(inode);
1780 error = bd_claim(bdev, sys_swapon);
1786 p->old_block_size = block_size(bdev);
1787 error = set_blocksize(bdev, PAGE_SIZE);
1791 } else if (S_ISREG(inode->i_mode)) {
1792 p->bdev = inode->i_sb->s_bdev;
1793 mutex_lock(&inode->i_mutex);
1795 if (IS_SWAPFILE(inode)) {
1803 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
1806 * Read the swap header.
1808 if (!mapping->a_ops->readpage) {
1812 page = read_mapping_page(mapping, 0, swap_file);
1814 error = PTR_ERR(page);
1817 swap_header = kmap(page);
1819 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
1820 printk(KERN_ERR "Unable to find swap-space signature\n");
1825 /* swap partition endianess hack... */
1826 if (swab32(swap_header->info.version) == 1) {
1827 swab32s(&swap_header->info.version);
1828 swab32s(&swap_header->info.last_page);
1829 swab32s(&swap_header->info.nr_badpages);
1830 for (i = 0; i < swap_header->info.nr_badpages; i++)
1831 swab32s(&swap_header->info.badpages[i]);
1833 /* Check the swap header's sub-version */
1834 if (swap_header->info.version != 1) {
1836 "Unable to handle swap header version %d\n",
1837 swap_header->info.version);
1843 p->cluster_next = 1;
1846 * Find out how many pages are allowed for a single swap
1847 * device. There are two limiting factors: 1) the number of
1848 * bits for the swap offset in the swp_entry_t type and
1849 * 2) the number of bits in the a swap pte as defined by
1850 * the different architectures. In order to find the
1851 * largest possible bit mask a swap entry with swap type 0
1852 * and swap offset ~0UL is created, encoded to a swap pte,
1853 * decoded to a swp_entry_t again and finally the swap
1854 * offset is extracted. This will mask all the bits from
1855 * the initial ~0UL mask that can't be encoded in either
1856 * the swp_entry_t or the architecture definition of a
1859 maxpages = swp_offset(pte_to_swp_entry(
1860 swp_entry_to_pte(swp_entry(0, ~0UL)))) - 1;
1861 if (maxpages > swap_header->info.last_page)
1862 maxpages = swap_header->info.last_page;
1863 p->highest_bit = maxpages - 1;
1868 if (swapfilepages && maxpages > swapfilepages) {
1870 "Swap area shorter than signature indicates\n");
1873 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
1875 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
1878 /* OK, set up the swap map and apply the bad block list */
1879 swap_map = vmalloc(maxpages * sizeof(short));
1885 memset(swap_map, 0, maxpages * sizeof(short));
1886 for (i = 0; i < swap_header->info.nr_badpages; i++) {
1887 int page_nr = swap_header->info.badpages[i];
1888 if (page_nr <= 0 || page_nr >= swap_header->info.last_page) {
1892 swap_map[page_nr] = SWAP_MAP_BAD;
1895 error = swap_cgroup_swapon(type, maxpages);
1899 nr_good_pages = swap_header->info.last_page -
1900 swap_header->info.nr_badpages -
1901 1 /* header page */;
1903 if (nr_good_pages) {
1904 swap_map[0] = SWAP_MAP_BAD;
1906 p->pages = nr_good_pages;
1907 nr_extents = setup_swap_extents(p, &span);
1908 if (nr_extents < 0) {
1912 nr_good_pages = p->pages;
1914 if (!nr_good_pages) {
1915 printk(KERN_WARNING "Empty swap-file\n");
1920 if (blk_queue_nonrot(bdev_get_queue(p->bdev))) {
1921 p->flags |= SWP_SOLIDSTATE;
1922 p->cluster_next = 1 + (random32() % p->highest_bit);
1924 if (discard_swap(p) == 0)
1925 p->flags |= SWP_DISCARDABLE;
1927 mutex_lock(&swapon_mutex);
1928 spin_lock(&swap_lock);
1929 if (swap_flags & SWAP_FLAG_PREFER)
1931 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
1933 p->prio = --least_priority;
1934 p->swap_map = swap_map;
1935 p->flags |= SWP_WRITEOK;
1936 nr_swap_pages += nr_good_pages;
1937 total_swap_pages += nr_good_pages;
1939 printk(KERN_INFO "Adding %uk swap on %s. "
1940 "Priority:%d extents:%d across:%lluk %s%s\n",
1941 nr_good_pages<<(PAGE_SHIFT-10), name, p->prio,
1942 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
1943 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
1944 (p->flags & SWP_DISCARDABLE) ? "D" : "");
1946 /* insert swap space into swap_list: */
1948 for (i = swap_list.head; i >= 0; i = swap_info[i].next) {
1949 if (p->prio >= swap_info[i].prio) {
1956 swap_list.head = swap_list.next = p - swap_info;
1958 swap_info[prev].next = p - swap_info;
1960 spin_unlock(&swap_lock);
1961 mutex_unlock(&swapon_mutex);
1966 set_blocksize(bdev, p->old_block_size);
1969 destroy_swap_extents(p);
1970 swap_cgroup_swapoff(type);
1972 spin_lock(&swap_lock);
1973 p->swap_file = NULL;
1975 spin_unlock(&swap_lock);
1978 filp_close(swap_file, NULL);
1980 if (page && !IS_ERR(page)) {
1982 page_cache_release(page);
1988 inode->i_flags |= S_SWAPFILE;
1989 mutex_unlock(&inode->i_mutex);
1994 void si_swapinfo(struct sysinfo *val)
1997 unsigned long nr_to_be_unused = 0;
1999 spin_lock(&swap_lock);
2000 for (i = 0; i < nr_swapfiles; i++) {
2001 if (!(swap_info[i].flags & SWP_USED) ||
2002 (swap_info[i].flags & SWP_WRITEOK))
2004 nr_to_be_unused += swap_info[i].inuse_pages;
2006 val->freeswap = nr_swap_pages + nr_to_be_unused;
2007 val->totalswap = total_swap_pages + nr_to_be_unused;
2008 spin_unlock(&swap_lock);
2012 * Verify that a swap entry is valid and increment its swap map count.
2014 * Note: if swap_map[] reaches SWAP_MAP_MAX the entries are treated as
2015 * "permanent", but will be reclaimed by the next swapoff.
2016 * Returns error code in following case.
2018 * - swp_entry is invalid -> EINVAL
2019 * - swp_entry is migration entry -> EINVAL
2020 * - swap-cache reference is requested but there is already one. -> EEXIST
2021 * - swap-cache reference is requested but the entry is not used. -> ENOENT
2023 static int __swap_duplicate(swp_entry_t entry, bool cache)
2025 struct swap_info_struct * p;
2026 unsigned long offset, type;
2027 int result = -EINVAL;
2031 if (is_migration_entry(entry))
2034 type = swp_type(entry);
2035 if (type >= nr_swapfiles)
2037 p = type + swap_info;
2038 offset = swp_offset(entry);
2040 spin_lock(&swap_lock);
2042 if (unlikely(offset >= p->max))
2045 count = swap_count(p->swap_map[offset]);
2046 has_cache = swap_has_cache(p->swap_map[offset]);
2048 if (cache == SWAP_CACHE) { /* called for swapcache/swapin-readahead */
2050 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
2051 if (!has_cache && count) {
2052 p->swap_map[offset] = encode_swapmap(count, true);
2054 } else if (has_cache) /* someone added cache */
2056 else if (!count) /* no users */
2059 } else if (count || has_cache) {
2060 if (count < SWAP_MAP_MAX - 1) {
2061 p->swap_map[offset] = encode_swapmap(count + 1,
2064 } else if (count <= SWAP_MAP_MAX) {
2065 if (swap_overflow++ < 5)
2067 "swap_dup: swap entry overflow\n");
2068 p->swap_map[offset] = encode_swapmap(SWAP_MAP_MAX,
2073 result = -ENOENT; /* unused swap entry */
2075 spin_unlock(&swap_lock);
2080 printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
2084 * increase reference count of swap entry by 1.
2086 void swap_duplicate(swp_entry_t entry)
2088 __swap_duplicate(entry, SWAP_MAP);
2092 * @entry: swap entry for which we allocate swap cache.
2094 * Called when allocating swap cache for exising swap entry,
2095 * This can return error codes. Returns 0 at success.
2096 * -EBUSY means there is a swap cache.
2097 * Note: return code is different from swap_duplicate().
2099 int swapcache_prepare(swp_entry_t entry)
2101 return __swap_duplicate(entry, SWAP_CACHE);
2105 struct swap_info_struct *
2106 get_swap_info_struct(unsigned type)
2108 return &swap_info[type];
2112 * swap_lock prevents swap_map being freed. Don't grab an extra
2113 * reference on the swaphandle, it doesn't matter if it becomes unused.
2115 int valid_swaphandles(swp_entry_t entry, unsigned long *offset)
2117 struct swap_info_struct *si;
2118 int our_page_cluster = page_cluster;
2119 pgoff_t target, toff;
2123 if (!our_page_cluster) /* no readahead */
2126 si = &swap_info[swp_type(entry)];
2127 target = swp_offset(entry);
2128 base = (target >> our_page_cluster) << our_page_cluster;
2129 end = base + (1 << our_page_cluster);
2130 if (!base) /* first page is swap header */
2133 spin_lock(&swap_lock);
2134 if (end > si->max) /* don't go beyond end of map */
2137 /* Count contiguous allocated slots above our target */
2138 for (toff = target; ++toff < end; nr_pages++) {
2139 /* Don't read in free or bad pages */
2140 if (!si->swap_map[toff])
2142 if (swap_count(si->swap_map[toff]) == SWAP_MAP_BAD)
2145 /* Count contiguous allocated slots below our target */
2146 for (toff = target; --toff >= base; nr_pages++) {
2147 /* Don't read in free or bad pages */
2148 if (!si->swap_map[toff])
2150 if (swap_count(si->swap_map[toff]) == SWAP_MAP_BAD)
2153 spin_unlock(&swap_lock);
2156 * Indicate starting offset, and return number of pages to get:
2157 * if only 1, say 0, since there's then no readahead to be done.
2160 return nr_pages? ++nr_pages: 0;