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 /* returnes 1 if swap entry is freed */
84 __try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset)
86 int type = si - swap_info;
87 swp_entry_t entry = swp_entry(type, offset);
91 page = find_get_page(&swapper_space, entry.val);
95 * This function is called from scan_swap_map() and it's called
96 * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
97 * We have to use trylock for avoiding deadlock. This is a special
98 * case and you should use try_to_free_swap() with explicit lock_page()
99 * in usual operations.
101 if (trylock_page(page)) {
102 ret = try_to_free_swap(page);
105 page_cache_release(page);
110 * We need this because the bdev->unplug_fn can sleep and we cannot
111 * hold swap_lock while calling the unplug_fn. And swap_lock
112 * cannot be turned into a mutex.
114 static DECLARE_RWSEM(swap_unplug_sem);
116 void swap_unplug_io_fn(struct backing_dev_info *unused_bdi, struct page *page)
120 down_read(&swap_unplug_sem);
121 entry.val = page_private(page);
122 if (PageSwapCache(page)) {
123 struct block_device *bdev = swap_info[swp_type(entry)].bdev;
124 struct backing_dev_info *bdi;
127 * If the page is removed from swapcache from under us (with a
128 * racy try_to_unuse/swapoff) we need an additional reference
129 * count to avoid reading garbage from page_private(page) above.
130 * If the WARN_ON triggers during a swapoff it maybe the race
131 * condition and it's harmless. However if it triggers without
132 * swapoff it signals a problem.
134 WARN_ON(page_count(page) <= 1);
136 bdi = bdev->bd_inode->i_mapping->backing_dev_info;
137 blk_run_backing_dev(bdi, page);
139 up_read(&swap_unplug_sem);
143 * swapon tell device that all the old swap contents can be discarded,
144 * to allow the swap device to optimize its wear-levelling.
146 static int discard_swap(struct swap_info_struct *si)
148 struct swap_extent *se;
151 list_for_each_entry(se, &si->extent_list, list) {
152 sector_t start_block = se->start_block << (PAGE_SHIFT - 9);
153 sector_t nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
155 if (se->start_page == 0) {
156 /* Do not discard the swap header page! */
157 start_block += 1 << (PAGE_SHIFT - 9);
158 nr_blocks -= 1 << (PAGE_SHIFT - 9);
163 err = blkdev_issue_discard(si->bdev, start_block,
164 nr_blocks, GFP_KERNEL);
170 return err; /* That will often be -EOPNOTSUPP */
174 * swap allocation tell device that a cluster of swap can now be discarded,
175 * to allow the swap device to optimize its wear-levelling.
177 static void discard_swap_cluster(struct swap_info_struct *si,
178 pgoff_t start_page, pgoff_t nr_pages)
180 struct swap_extent *se = si->curr_swap_extent;
181 int found_extent = 0;
184 struct list_head *lh;
186 if (se->start_page <= start_page &&
187 start_page < se->start_page + se->nr_pages) {
188 pgoff_t offset = start_page - se->start_page;
189 sector_t start_block = se->start_block + offset;
190 sector_t nr_blocks = se->nr_pages - offset;
192 if (nr_blocks > nr_pages)
193 nr_blocks = nr_pages;
194 start_page += nr_blocks;
195 nr_pages -= nr_blocks;
198 si->curr_swap_extent = se;
200 start_block <<= PAGE_SHIFT - 9;
201 nr_blocks <<= PAGE_SHIFT - 9;
202 if (blkdev_issue_discard(si->bdev, start_block,
203 nr_blocks, GFP_NOIO))
208 if (lh == &si->extent_list)
210 se = list_entry(lh, struct swap_extent, list);
214 static int wait_for_discard(void *word)
220 #define SWAPFILE_CLUSTER 256
221 #define LATENCY_LIMIT 256
223 static inline unsigned long scan_swap_map(struct swap_info_struct *si,
226 unsigned long offset;
227 unsigned long scan_base;
228 unsigned long last_in_cluster = 0;
229 int latency_ration = LATENCY_LIMIT;
230 int found_free_cluster = 0;
233 * We try to cluster swap pages by allocating them sequentially
234 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
235 * way, however, we resort to first-free allocation, starting
236 * a new cluster. This prevents us from scattering swap pages
237 * all over the entire swap partition, so that we reduce
238 * overall disk seek times between swap pages. -- sct
239 * But we do now try to find an empty cluster. -Andrea
240 * And we let swap pages go all over an SSD partition. Hugh
243 si->flags += SWP_SCANNING;
244 scan_base = offset = si->cluster_next;
246 if (unlikely(!si->cluster_nr--)) {
247 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
248 si->cluster_nr = SWAPFILE_CLUSTER - 1;
251 if (si->flags & SWP_DISCARDABLE) {
253 * Start range check on racing allocations, in case
254 * they overlap the cluster we eventually decide on
255 * (we scan without swap_lock to allow preemption).
256 * It's hardly conceivable that cluster_nr could be
257 * wrapped during our scan, but don't depend on it.
259 if (si->lowest_alloc)
261 si->lowest_alloc = si->max;
262 si->highest_alloc = 0;
264 spin_unlock(&swap_lock);
267 * If seek is expensive, start searching for new cluster from
268 * start of partition, to minimize the span of allocated swap.
269 * But if seek is cheap, search from our current position, so
270 * that swap is allocated from all over the partition: if the
271 * Flash Translation Layer only remaps within limited zones,
272 * we don't want to wear out the first zone too quickly.
274 if (!(si->flags & SWP_SOLIDSTATE))
275 scan_base = offset = si->lowest_bit;
276 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
278 /* Locate the first empty (unaligned) cluster */
279 for (; last_in_cluster <= si->highest_bit; offset++) {
280 if (si->swap_map[offset])
281 last_in_cluster = offset + SWAPFILE_CLUSTER;
282 else if (offset == last_in_cluster) {
283 spin_lock(&swap_lock);
284 offset -= SWAPFILE_CLUSTER - 1;
285 si->cluster_next = offset;
286 si->cluster_nr = SWAPFILE_CLUSTER - 1;
287 found_free_cluster = 1;
290 if (unlikely(--latency_ration < 0)) {
292 latency_ration = LATENCY_LIMIT;
296 offset = si->lowest_bit;
297 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
299 /* Locate the first empty (unaligned) cluster */
300 for (; last_in_cluster < scan_base; offset++) {
301 if (si->swap_map[offset])
302 last_in_cluster = offset + SWAPFILE_CLUSTER;
303 else if (offset == last_in_cluster) {
304 spin_lock(&swap_lock);
305 offset -= SWAPFILE_CLUSTER - 1;
306 si->cluster_next = offset;
307 si->cluster_nr = SWAPFILE_CLUSTER - 1;
308 found_free_cluster = 1;
311 if (unlikely(--latency_ration < 0)) {
313 latency_ration = LATENCY_LIMIT;
318 spin_lock(&swap_lock);
319 si->cluster_nr = SWAPFILE_CLUSTER - 1;
320 si->lowest_alloc = 0;
324 if (!(si->flags & SWP_WRITEOK))
326 if (!si->highest_bit)
328 if (offset > si->highest_bit)
329 scan_base = offset = si->lowest_bit;
331 /* reuse swap entry of cache-only swap if not busy. */
332 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
334 spin_unlock(&swap_lock);
335 swap_was_freed = __try_to_reclaim_swap(si, offset);
336 spin_lock(&swap_lock);
337 /* entry was freed successfully, try to use this again */
340 goto scan; /* check next one */
343 if (si->swap_map[offset])
346 if (offset == si->lowest_bit)
348 if (offset == si->highest_bit)
351 if (si->inuse_pages == si->pages) {
352 si->lowest_bit = si->max;
355 if (cache == SWAP_CACHE) /* at usual swap-out via vmscan.c */
356 si->swap_map[offset] = encode_swapmap(0, true);
357 else /* at suspend */
358 si->swap_map[offset] = encode_swapmap(1, false);
359 si->cluster_next = offset + 1;
360 si->flags -= SWP_SCANNING;
362 if (si->lowest_alloc) {
364 * Only set when SWP_DISCARDABLE, and there's a scan
365 * for a free cluster in progress or just completed.
367 if (found_free_cluster) {
369 * To optimize wear-levelling, discard the
370 * old data of the cluster, taking care not to
371 * discard any of its pages that have already
372 * been allocated by racing tasks (offset has
373 * already stepped over any at the beginning).
375 if (offset < si->highest_alloc &&
376 si->lowest_alloc <= last_in_cluster)
377 last_in_cluster = si->lowest_alloc - 1;
378 si->flags |= SWP_DISCARDING;
379 spin_unlock(&swap_lock);
381 if (offset < last_in_cluster)
382 discard_swap_cluster(si, offset,
383 last_in_cluster - offset + 1);
385 spin_lock(&swap_lock);
386 si->lowest_alloc = 0;
387 si->flags &= ~SWP_DISCARDING;
389 smp_mb(); /* wake_up_bit advises this */
390 wake_up_bit(&si->flags, ilog2(SWP_DISCARDING));
392 } else if (si->flags & SWP_DISCARDING) {
394 * Delay using pages allocated by racing tasks
395 * until the whole discard has been issued. We
396 * could defer that delay until swap_writepage,
397 * but it's easier to keep this self-contained.
399 spin_unlock(&swap_lock);
400 wait_on_bit(&si->flags, ilog2(SWP_DISCARDING),
401 wait_for_discard, TASK_UNINTERRUPTIBLE);
402 spin_lock(&swap_lock);
405 * Note pages allocated by racing tasks while
406 * scan for a free cluster is in progress, so
407 * that its final discard can exclude them.
409 if (offset < si->lowest_alloc)
410 si->lowest_alloc = offset;
411 if (offset > si->highest_alloc)
412 si->highest_alloc = offset;
418 spin_unlock(&swap_lock);
419 while (++offset <= si->highest_bit) {
420 if (!si->swap_map[offset]) {
421 spin_lock(&swap_lock);
424 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
425 spin_lock(&swap_lock);
428 if (unlikely(--latency_ration < 0)) {
430 latency_ration = LATENCY_LIMIT;
433 offset = si->lowest_bit;
434 while (++offset < scan_base) {
435 if (!si->swap_map[offset]) {
436 spin_lock(&swap_lock);
439 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
440 spin_lock(&swap_lock);
443 if (unlikely(--latency_ration < 0)) {
445 latency_ration = LATENCY_LIMIT;
448 spin_lock(&swap_lock);
451 si->flags -= SWP_SCANNING;
455 swp_entry_t get_swap_page(void)
457 struct swap_info_struct *si;
462 spin_lock(&swap_lock);
463 if (nr_swap_pages <= 0)
467 for (type = swap_list.next; type >= 0 && wrapped < 2; type = next) {
468 si = swap_info + type;
471 (!wrapped && si->prio != swap_info[next].prio)) {
472 next = swap_list.head;
476 if (!si->highest_bit)
478 if (!(si->flags & SWP_WRITEOK))
481 swap_list.next = next;
482 /* This is called for allocating swap entry for cache */
483 offset = scan_swap_map(si, SWAP_CACHE);
485 spin_unlock(&swap_lock);
486 return swp_entry(type, offset);
488 next = swap_list.next;
493 spin_unlock(&swap_lock);
494 return (swp_entry_t) {0};
497 /* The only caller of this function is now susupend routine */
498 swp_entry_t get_swap_page_of_type(int type)
500 struct swap_info_struct *si;
503 spin_lock(&swap_lock);
504 si = swap_info + type;
505 if (si->flags & SWP_WRITEOK) {
507 /* This is called for allocating swap entry, not cache */
508 offset = scan_swap_map(si, SWAP_MAP);
510 spin_unlock(&swap_lock);
511 return swp_entry(type, offset);
515 spin_unlock(&swap_lock);
516 return (swp_entry_t) {0};
519 static struct swap_info_struct * swap_info_get(swp_entry_t entry)
521 struct swap_info_struct * p;
522 unsigned long offset, type;
526 type = swp_type(entry);
527 if (type >= nr_swapfiles)
529 p = & swap_info[type];
530 if (!(p->flags & SWP_USED))
532 offset = swp_offset(entry);
533 if (offset >= p->max)
535 if (!p->swap_map[offset])
537 spin_lock(&swap_lock);
541 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val);
544 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val);
547 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val);
550 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val);
555 static int swap_entry_free(struct swap_info_struct *p,
556 swp_entry_t ent, int cache)
558 unsigned long offset = swp_offset(ent);
559 int count = swap_count(p->swap_map[offset]);
562 has_cache = swap_has_cache(p->swap_map[offset]);
564 if (cache == SWAP_MAP) { /* dropping usage count of swap */
565 if (count < SWAP_MAP_MAX) {
567 p->swap_map[offset] = encode_swapmap(count, has_cache);
569 } else { /* dropping swap cache flag */
570 VM_BUG_ON(!has_cache);
571 p->swap_map[offset] = encode_swapmap(count, false);
575 count = p->swap_map[offset];
576 /* free if no reference */
578 if (offset < p->lowest_bit)
579 p->lowest_bit = offset;
580 if (offset > p->highest_bit)
581 p->highest_bit = offset;
582 if (p->prio > swap_info[swap_list.next].prio)
583 swap_list.next = p - swap_info;
587 if (!swap_count(count))
588 mem_cgroup_uncharge_swap(ent);
593 * Caller has made sure that the swapdevice corresponding to entry
594 * is still around or has not been recycled.
596 void swap_free(swp_entry_t entry)
598 struct swap_info_struct * p;
600 p = swap_info_get(entry);
602 swap_entry_free(p, entry, SWAP_MAP);
603 spin_unlock(&swap_lock);
608 * Called after dropping swapcache to decrease refcnt to swap entries.
610 void swapcache_free(swp_entry_t entry, struct page *page)
612 struct swap_info_struct *p;
615 p = swap_info_get(entry);
617 ret = swap_entry_free(p, entry, SWAP_CACHE);
621 swapout = true; /* the end of swap out */
623 swapout = false; /* no more swap users! */
624 mem_cgroup_uncharge_swapcache(page, entry, swapout);
626 spin_unlock(&swap_lock);
632 * How many references to page are currently swapped out?
634 static inline int page_swapcount(struct page *page)
637 struct swap_info_struct *p;
640 entry.val = page_private(page);
641 p = swap_info_get(entry);
643 count = swap_count(p->swap_map[swp_offset(entry)]);
644 spin_unlock(&swap_lock);
650 * We can write to an anon page without COW if there are no other references
651 * to it. And as a side-effect, free up its swap: because the old content
652 * on disk will never be read, and seeking back there to write new content
653 * later would only waste time away from clustering.
655 int reuse_swap_page(struct page *page)
659 VM_BUG_ON(!PageLocked(page));
660 count = page_mapcount(page);
661 if (count <= 1 && PageSwapCache(page)) {
662 count += page_swapcount(page);
663 if (count == 1 && !PageWriteback(page)) {
664 delete_from_swap_cache(page);
672 * If swap is getting full, or if there are no more mappings of this page,
673 * then try_to_free_swap is called to free its swap space.
675 int try_to_free_swap(struct page *page)
677 VM_BUG_ON(!PageLocked(page));
679 if (!PageSwapCache(page))
681 if (PageWriteback(page))
683 if (page_swapcount(page))
686 delete_from_swap_cache(page);
692 * Free the swap entry like above, but also try to
693 * free the page cache entry if it is the last user.
695 int free_swap_and_cache(swp_entry_t entry)
697 struct swap_info_struct *p;
698 struct page *page = NULL;
700 if (is_migration_entry(entry))
703 p = swap_info_get(entry);
705 if (swap_entry_free(p, entry, SWAP_MAP) == SWAP_HAS_CACHE) {
706 page = find_get_page(&swapper_space, entry.val);
707 if (page && !trylock_page(page)) {
708 page_cache_release(page);
712 spin_unlock(&swap_lock);
716 * Not mapped elsewhere, or swap space full? Free it!
717 * Also recheck PageSwapCache now page is locked (above).
719 if (PageSwapCache(page) && !PageWriteback(page) &&
720 (!page_mapped(page) || vm_swap_full())) {
721 delete_from_swap_cache(page);
725 page_cache_release(page);
730 #ifdef CONFIG_HIBERNATION
732 * Find the swap type that corresponds to given device (if any).
734 * @offset - number of the PAGE_SIZE-sized block of the device, starting
735 * from 0, in which the swap header is expected to be located.
737 * This is needed for the suspend to disk (aka swsusp).
739 int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
741 struct block_device *bdev = NULL;
745 bdev = bdget(device);
747 spin_lock(&swap_lock);
748 for (i = 0; i < nr_swapfiles; i++) {
749 struct swap_info_struct *sis = swap_info + i;
751 if (!(sis->flags & SWP_WRITEOK))
756 *bdev_p = bdget(sis->bdev->bd_dev);
758 spin_unlock(&swap_lock);
761 if (bdev == sis->bdev) {
762 struct swap_extent *se;
764 se = list_entry(sis->extent_list.next,
765 struct swap_extent, list);
766 if (se->start_block == offset) {
768 *bdev_p = bdget(sis->bdev->bd_dev);
770 spin_unlock(&swap_lock);
776 spin_unlock(&swap_lock);
784 * Return either the total number of swap pages of given type, or the number
785 * of free pages of that type (depending on @free)
787 * This is needed for software suspend
789 unsigned int count_swap_pages(int type, int free)
793 if (type < nr_swapfiles) {
794 spin_lock(&swap_lock);
795 if (swap_info[type].flags & SWP_WRITEOK) {
796 n = swap_info[type].pages;
798 n -= swap_info[type].inuse_pages;
800 spin_unlock(&swap_lock);
807 * No need to decide whether this PTE shares the swap entry with others,
808 * just let do_wp_page work it out if a write is requested later - to
809 * force COW, vm_page_prot omits write permission from any private vma.
811 static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
812 unsigned long addr, swp_entry_t entry, struct page *page)
814 struct mem_cgroup *ptr = NULL;
819 if (mem_cgroup_try_charge_swapin(vma->vm_mm, page, GFP_KERNEL, &ptr)) {
824 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
825 if (unlikely(!pte_same(*pte, swp_entry_to_pte(entry)))) {
827 mem_cgroup_cancel_charge_swapin(ptr);
832 inc_mm_counter(vma->vm_mm, anon_rss);
834 set_pte_at(vma->vm_mm, addr, pte,
835 pte_mkold(mk_pte(page, vma->vm_page_prot)));
836 page_add_anon_rmap(page, vma, addr);
837 mem_cgroup_commit_charge_swapin(page, ptr);
840 * Move the page to the active list so it is not
841 * immediately swapped out again after swapon.
845 pte_unmap_unlock(pte, ptl);
850 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
851 unsigned long addr, unsigned long end,
852 swp_entry_t entry, struct page *page)
854 pte_t swp_pte = swp_entry_to_pte(entry);
859 * We don't actually need pte lock while scanning for swp_pte: since
860 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
861 * page table while we're scanning; though it could get zapped, and on
862 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
863 * of unmatched parts which look like swp_pte, so unuse_pte must
864 * recheck under pte lock. Scanning without pte lock lets it be
865 * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
867 pte = pte_offset_map(pmd, addr);
870 * swapoff spends a _lot_ of time in this loop!
871 * Test inline before going to call unuse_pte.
873 if (unlikely(pte_same(*pte, swp_pte))) {
875 ret = unuse_pte(vma, pmd, addr, entry, page);
878 pte = pte_offset_map(pmd, addr);
880 } while (pte++, addr += PAGE_SIZE, addr != end);
886 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
887 unsigned long addr, unsigned long end,
888 swp_entry_t entry, struct page *page)
894 pmd = pmd_offset(pud, addr);
896 next = pmd_addr_end(addr, end);
897 if (pmd_none_or_clear_bad(pmd))
899 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
902 } while (pmd++, addr = next, addr != end);
906 static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
907 unsigned long addr, unsigned long end,
908 swp_entry_t entry, struct page *page)
914 pud = pud_offset(pgd, addr);
916 next = pud_addr_end(addr, end);
917 if (pud_none_or_clear_bad(pud))
919 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
922 } while (pud++, addr = next, addr != end);
926 static int unuse_vma(struct vm_area_struct *vma,
927 swp_entry_t entry, struct page *page)
930 unsigned long addr, end, next;
934 addr = page_address_in_vma(page, vma);
938 end = addr + PAGE_SIZE;
940 addr = vma->vm_start;
944 pgd = pgd_offset(vma->vm_mm, addr);
946 next = pgd_addr_end(addr, end);
947 if (pgd_none_or_clear_bad(pgd))
949 ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
952 } while (pgd++, addr = next, addr != end);
956 static int unuse_mm(struct mm_struct *mm,
957 swp_entry_t entry, struct page *page)
959 struct vm_area_struct *vma;
962 if (!down_read_trylock(&mm->mmap_sem)) {
964 * Activate page so shrink_inactive_list is unlikely to unmap
965 * its ptes while lock is dropped, so swapoff can make progress.
969 down_read(&mm->mmap_sem);
972 for (vma = mm->mmap; vma; vma = vma->vm_next) {
973 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
976 up_read(&mm->mmap_sem);
977 return (ret < 0)? ret: 0;
981 * Scan swap_map from current position to next entry still in use.
982 * Recycle to start on reaching the end, returning 0 when empty.
984 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
987 unsigned int max = si->max;
988 unsigned int i = prev;
992 * No need for swap_lock here: we're just looking
993 * for whether an entry is in use, not modifying it; false
994 * hits are okay, and sys_swapoff() has already prevented new
995 * allocations from this area (while holding swap_lock).
1004 * No entries in use at top of swap_map,
1005 * loop back to start and recheck there.
1011 count = si->swap_map[i];
1012 if (count && swap_count(count) != SWAP_MAP_BAD)
1019 * We completely avoid races by reading each swap page in advance,
1020 * and then search for the process using it. All the necessary
1021 * page table adjustments can then be made atomically.
1023 static int try_to_unuse(unsigned int type)
1025 struct swap_info_struct * si = &swap_info[type];
1026 struct mm_struct *start_mm;
1027 unsigned short *swap_map;
1028 unsigned short swcount;
1033 int reset_overflow = 0;
1037 * When searching mms for an entry, a good strategy is to
1038 * start at the first mm we freed the previous entry from
1039 * (though actually we don't notice whether we or coincidence
1040 * freed the entry). Initialize this start_mm with a hold.
1042 * A simpler strategy would be to start at the last mm we
1043 * freed the previous entry from; but that would take less
1044 * advantage of mmlist ordering, which clusters forked mms
1045 * together, child after parent. If we race with dup_mmap(), we
1046 * prefer to resolve parent before child, lest we miss entries
1047 * duplicated after we scanned child: using last mm would invert
1048 * that. Though it's only a serious concern when an overflowed
1049 * swap count is reset from SWAP_MAP_MAX, preventing a rescan.
1051 start_mm = &init_mm;
1052 atomic_inc(&init_mm.mm_users);
1055 * Keep on scanning until all entries have gone. Usually,
1056 * one pass through swap_map is enough, but not necessarily:
1057 * there are races when an instance of an entry might be missed.
1059 while ((i = find_next_to_unuse(si, i)) != 0) {
1060 if (signal_pending(current)) {
1066 * Get a page for the entry, using the existing swap
1067 * cache page if there is one. Otherwise, get a clean
1068 * page and read the swap into it.
1070 swap_map = &si->swap_map[i];
1071 entry = swp_entry(type, i);
1072 page = read_swap_cache_async(entry,
1073 GFP_HIGHUSER_MOVABLE, NULL, 0);
1076 * Either swap_duplicate() failed because entry
1077 * has been freed independently, and will not be
1078 * reused since sys_swapoff() already disabled
1079 * allocation from here, or alloc_page() failed.
1088 * Don't hold on to start_mm if it looks like exiting.
1090 if (atomic_read(&start_mm->mm_users) == 1) {
1092 start_mm = &init_mm;
1093 atomic_inc(&init_mm.mm_users);
1097 * Wait for and lock page. When do_swap_page races with
1098 * try_to_unuse, do_swap_page can handle the fault much
1099 * faster than try_to_unuse can locate the entry. This
1100 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1101 * defer to do_swap_page in such a case - in some tests,
1102 * do_swap_page and try_to_unuse repeatedly compete.
1104 wait_on_page_locked(page);
1105 wait_on_page_writeback(page);
1107 wait_on_page_writeback(page);
1110 * Remove all references to entry.
1111 * Whenever we reach init_mm, there's no address space
1112 * to search, but use it as a reminder to search shmem.
1115 swcount = *swap_map;
1116 if (swap_count(swcount)) {
1117 if (start_mm == &init_mm)
1118 shmem = shmem_unuse(entry, page);
1120 retval = unuse_mm(start_mm, entry, page);
1122 if (swap_count(*swap_map)) {
1123 int set_start_mm = (*swap_map >= swcount);
1124 struct list_head *p = &start_mm->mmlist;
1125 struct mm_struct *new_start_mm = start_mm;
1126 struct mm_struct *prev_mm = start_mm;
1127 struct mm_struct *mm;
1129 atomic_inc(&new_start_mm->mm_users);
1130 atomic_inc(&prev_mm->mm_users);
1131 spin_lock(&mmlist_lock);
1132 while (swap_count(*swap_map) && !retval && !shmem &&
1133 (p = p->next) != &start_mm->mmlist) {
1134 mm = list_entry(p, struct mm_struct, mmlist);
1135 if (!atomic_inc_not_zero(&mm->mm_users))
1137 spin_unlock(&mmlist_lock);
1143 swcount = *swap_map;
1144 if (!swap_count(swcount)) /* any usage ? */
1146 else if (mm == &init_mm) {
1148 shmem = shmem_unuse(entry, page);
1150 retval = unuse_mm(mm, entry, page);
1153 swap_count(*swap_map) < swcount) {
1154 mmput(new_start_mm);
1155 atomic_inc(&mm->mm_users);
1159 spin_lock(&mmlist_lock);
1161 spin_unlock(&mmlist_lock);
1164 start_mm = new_start_mm;
1167 /* page has already been unlocked and released */
1175 page_cache_release(page);
1180 * How could swap count reach 0x7ffe ?
1181 * There's no way to repeat a swap page within an mm
1182 * (except in shmem, where it's the shared object which takes
1183 * the reference count)?
1184 * We believe SWAP_MAP_MAX cannot occur.(if occur, unsigned
1185 * short is too small....)
1186 * If that's wrong, then we should worry more about
1187 * exit_mmap() and do_munmap() cases described above:
1188 * we might be resetting SWAP_MAP_MAX too early here.
1189 * We know "Undead"s can happen, they're okay, so don't
1190 * report them; but do report if we reset SWAP_MAP_MAX.
1192 /* We might release the lock_page() in unuse_mm(). */
1193 if (!PageSwapCache(page) || page_private(page) != entry.val)
1196 if (swap_count(*swap_map) == SWAP_MAP_MAX) {
1197 spin_lock(&swap_lock);
1198 *swap_map = encode_swapmap(0, true);
1199 spin_unlock(&swap_lock);
1204 * If a reference remains (rare), we would like to leave
1205 * the page in the swap cache; but try_to_unmap could
1206 * then re-duplicate the entry once we drop page lock,
1207 * so we might loop indefinitely; also, that page could
1208 * not be swapped out to other storage meanwhile. So:
1209 * delete from cache even if there's another reference,
1210 * after ensuring that the data has been saved to disk -
1211 * since if the reference remains (rarer), it will be
1212 * read from disk into another page. Splitting into two
1213 * pages would be incorrect if swap supported "shared
1214 * private" pages, but they are handled by tmpfs files.
1216 if (swap_count(*swap_map) &&
1217 PageDirty(page) && PageSwapCache(page)) {
1218 struct writeback_control wbc = {
1219 .sync_mode = WB_SYNC_NONE,
1222 swap_writepage(page, &wbc);
1224 wait_on_page_writeback(page);
1228 * It is conceivable that a racing task removed this page from
1229 * swap cache just before we acquired the page lock at the top,
1230 * or while we dropped it in unuse_mm(). The page might even
1231 * be back in swap cache on another swap area: that we must not
1232 * delete, since it may not have been written out to swap yet.
1234 if (PageSwapCache(page) &&
1235 likely(page_private(page) == entry.val))
1236 delete_from_swap_cache(page);
1239 * So we could skip searching mms once swap count went
1240 * to 1, we did not mark any present ptes as dirty: must
1241 * mark page dirty so shrink_page_list will preserve it.
1246 page_cache_release(page);
1249 * Make sure that we aren't completely killing
1250 * interactive performance.
1256 if (reset_overflow) {
1257 printk(KERN_WARNING "swapoff: cleared swap entry overflow\n");
1264 * After a successful try_to_unuse, if no swap is now in use, we know
1265 * we can empty the mmlist. swap_lock must be held on entry and exit.
1266 * Note that mmlist_lock nests inside swap_lock, and an mm must be
1267 * added to the mmlist just after page_duplicate - before would be racy.
1269 static void drain_mmlist(void)
1271 struct list_head *p, *next;
1274 for (i = 0; i < nr_swapfiles; i++)
1275 if (swap_info[i].inuse_pages)
1277 spin_lock(&mmlist_lock);
1278 list_for_each_safe(p, next, &init_mm.mmlist)
1280 spin_unlock(&mmlist_lock);
1284 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
1285 * corresponds to page offset `offset'.
1287 sector_t map_swap_page(struct swap_info_struct *sis, pgoff_t offset)
1289 struct swap_extent *se = sis->curr_swap_extent;
1290 struct swap_extent *start_se = se;
1293 struct list_head *lh;
1295 if (se->start_page <= offset &&
1296 offset < (se->start_page + se->nr_pages)) {
1297 return se->start_block + (offset - se->start_page);
1300 if (lh == &sis->extent_list)
1302 se = list_entry(lh, struct swap_extent, list);
1303 sis->curr_swap_extent = se;
1304 BUG_ON(se == start_se); /* It *must* be present */
1308 #ifdef CONFIG_HIBERNATION
1310 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1311 * corresponding to given index in swap_info (swap type).
1313 sector_t swapdev_block(int swap_type, pgoff_t offset)
1315 struct swap_info_struct *sis;
1317 if (swap_type >= nr_swapfiles)
1320 sis = swap_info + swap_type;
1321 return (sis->flags & SWP_WRITEOK) ? map_swap_page(sis, offset) : 0;
1323 #endif /* CONFIG_HIBERNATION */
1326 * Free all of a swapdev's extent information
1328 static void destroy_swap_extents(struct swap_info_struct *sis)
1330 while (!list_empty(&sis->extent_list)) {
1331 struct swap_extent *se;
1333 se = list_entry(sis->extent_list.next,
1334 struct swap_extent, list);
1335 list_del(&se->list);
1341 * Add a block range (and the corresponding page range) into this swapdev's
1342 * extent list. The extent list is kept sorted in page order.
1344 * This function rather assumes that it is called in ascending page order.
1347 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1348 unsigned long nr_pages, sector_t start_block)
1350 struct swap_extent *se;
1351 struct swap_extent *new_se;
1352 struct list_head *lh;
1354 lh = sis->extent_list.prev; /* The highest page extent */
1355 if (lh != &sis->extent_list) {
1356 se = list_entry(lh, struct swap_extent, list);
1357 BUG_ON(se->start_page + se->nr_pages != start_page);
1358 if (se->start_block + se->nr_pages == start_block) {
1360 se->nr_pages += nr_pages;
1366 * No merge. Insert a new extent, preserving ordering.
1368 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1371 new_se->start_page = start_page;
1372 new_se->nr_pages = nr_pages;
1373 new_se->start_block = start_block;
1375 list_add_tail(&new_se->list, &sis->extent_list);
1380 * A `swap extent' is a simple thing which maps a contiguous range of pages
1381 * onto a contiguous range of disk blocks. An ordered list of swap extents
1382 * is built at swapon time and is then used at swap_writepage/swap_readpage
1383 * time for locating where on disk a page belongs.
1385 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1386 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1387 * swap files identically.
1389 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1390 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1391 * swapfiles are handled *identically* after swapon time.
1393 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1394 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1395 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1396 * requirements, they are simply tossed out - we will never use those blocks
1399 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
1400 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1401 * which will scribble on the fs.
1403 * The amount of disk space which a single swap extent represents varies.
1404 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1405 * extents in the list. To avoid much list walking, we cache the previous
1406 * search location in `curr_swap_extent', and start new searches from there.
1407 * This is extremely effective. The average number of iterations in
1408 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1410 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
1412 struct inode *inode;
1413 unsigned blocks_per_page;
1414 unsigned long page_no;
1416 sector_t probe_block;
1417 sector_t last_block;
1418 sector_t lowest_block = -1;
1419 sector_t highest_block = 0;
1423 inode = sis->swap_file->f_mapping->host;
1424 if (S_ISBLK(inode->i_mode)) {
1425 ret = add_swap_extent(sis, 0, sis->max, 0);
1430 blkbits = inode->i_blkbits;
1431 blocks_per_page = PAGE_SIZE >> blkbits;
1434 * Map all the blocks into the extent list. This code doesn't try
1439 last_block = i_size_read(inode) >> blkbits;
1440 while ((probe_block + blocks_per_page) <= last_block &&
1441 page_no < sis->max) {
1442 unsigned block_in_page;
1443 sector_t first_block;
1445 first_block = bmap(inode, probe_block);
1446 if (first_block == 0)
1450 * It must be PAGE_SIZE aligned on-disk
1452 if (first_block & (blocks_per_page - 1)) {
1457 for (block_in_page = 1; block_in_page < blocks_per_page;
1461 block = bmap(inode, probe_block + block_in_page);
1464 if (block != first_block + block_in_page) {
1471 first_block >>= (PAGE_SHIFT - blkbits);
1472 if (page_no) { /* exclude the header page */
1473 if (first_block < lowest_block)
1474 lowest_block = first_block;
1475 if (first_block > highest_block)
1476 highest_block = first_block;
1480 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1482 ret = add_swap_extent(sis, page_no, 1, first_block);
1487 probe_block += blocks_per_page;
1492 *span = 1 + highest_block - lowest_block;
1494 page_no = 1; /* force Empty message */
1496 sis->pages = page_no - 1;
1497 sis->highest_bit = page_no - 1;
1499 sis->curr_swap_extent = list_entry(sis->extent_list.prev,
1500 struct swap_extent, list);
1503 printk(KERN_ERR "swapon: swapfile has holes\n");
1509 SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
1511 struct swap_info_struct * p = NULL;
1512 unsigned short *swap_map;
1513 struct file *swap_file, *victim;
1514 struct address_space *mapping;
1515 struct inode *inode;
1520 if (!capable(CAP_SYS_ADMIN))
1523 pathname = getname(specialfile);
1524 err = PTR_ERR(pathname);
1525 if (IS_ERR(pathname))
1528 victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0);
1530 err = PTR_ERR(victim);
1534 mapping = victim->f_mapping;
1536 spin_lock(&swap_lock);
1537 for (type = swap_list.head; type >= 0; type = swap_info[type].next) {
1538 p = swap_info + type;
1539 if (p->flags & SWP_WRITEOK) {
1540 if (p->swap_file->f_mapping == mapping)
1547 spin_unlock(&swap_lock);
1550 if (!security_vm_enough_memory(p->pages))
1551 vm_unacct_memory(p->pages);
1554 spin_unlock(&swap_lock);
1558 swap_list.head = p->next;
1560 swap_info[prev].next = p->next;
1562 if (type == swap_list.next) {
1563 /* just pick something that's safe... */
1564 swap_list.next = swap_list.head;
1567 for (i = p->next; i >= 0; i = swap_info[i].next)
1568 swap_info[i].prio = p->prio--;
1571 nr_swap_pages -= p->pages;
1572 total_swap_pages -= p->pages;
1573 p->flags &= ~SWP_WRITEOK;
1574 spin_unlock(&swap_lock);
1576 current->flags |= PF_SWAPOFF;
1577 err = try_to_unuse(type);
1578 current->flags &= ~PF_SWAPOFF;
1581 /* re-insert swap space back into swap_list */
1582 spin_lock(&swap_lock);
1584 p->prio = --least_priority;
1586 for (i = swap_list.head; i >= 0; i = swap_info[i].next) {
1587 if (p->prio >= swap_info[i].prio)
1593 swap_list.head = swap_list.next = p - swap_info;
1595 swap_info[prev].next = p - swap_info;
1596 nr_swap_pages += p->pages;
1597 total_swap_pages += p->pages;
1598 p->flags |= SWP_WRITEOK;
1599 spin_unlock(&swap_lock);
1603 /* wait for any unplug function to finish */
1604 down_write(&swap_unplug_sem);
1605 up_write(&swap_unplug_sem);
1607 destroy_swap_extents(p);
1608 mutex_lock(&swapon_mutex);
1609 spin_lock(&swap_lock);
1612 /* wait for anyone still in scan_swap_map */
1613 p->highest_bit = 0; /* cuts scans short */
1614 while (p->flags >= SWP_SCANNING) {
1615 spin_unlock(&swap_lock);
1616 schedule_timeout_uninterruptible(1);
1617 spin_lock(&swap_lock);
1620 swap_file = p->swap_file;
1621 p->swap_file = NULL;
1623 swap_map = p->swap_map;
1626 spin_unlock(&swap_lock);
1627 mutex_unlock(&swapon_mutex);
1629 /* Destroy swap account informatin */
1630 swap_cgroup_swapoff(type);
1632 inode = mapping->host;
1633 if (S_ISBLK(inode->i_mode)) {
1634 struct block_device *bdev = I_BDEV(inode);
1635 set_blocksize(bdev, p->old_block_size);
1638 mutex_lock(&inode->i_mutex);
1639 inode->i_flags &= ~S_SWAPFILE;
1640 mutex_unlock(&inode->i_mutex);
1642 filp_close(swap_file, NULL);
1646 filp_close(victim, NULL);
1651 #ifdef CONFIG_PROC_FS
1653 static void *swap_start(struct seq_file *swap, loff_t *pos)
1655 struct swap_info_struct *ptr = swap_info;
1659 mutex_lock(&swapon_mutex);
1662 return SEQ_START_TOKEN;
1664 for (i = 0; i < nr_swapfiles; i++, ptr++) {
1665 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1674 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1676 struct swap_info_struct *ptr;
1677 struct swap_info_struct *endptr = swap_info + nr_swapfiles;
1679 if (v == SEQ_START_TOKEN)
1686 for (; ptr < endptr; ptr++) {
1687 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1696 static void swap_stop(struct seq_file *swap, void *v)
1698 mutex_unlock(&swapon_mutex);
1701 static int swap_show(struct seq_file *swap, void *v)
1703 struct swap_info_struct *ptr = v;
1707 if (ptr == SEQ_START_TOKEN) {
1708 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1712 file = ptr->swap_file;
1713 len = seq_path(swap, &file->f_path, " \t\n\\");
1714 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
1715 len < 40 ? 40 - len : 1, " ",
1716 S_ISBLK(file->f_path.dentry->d_inode->i_mode) ?
1717 "partition" : "file\t",
1718 ptr->pages << (PAGE_SHIFT - 10),
1719 ptr->inuse_pages << (PAGE_SHIFT - 10),
1724 static const struct seq_operations swaps_op = {
1725 .start = swap_start,
1731 static int swaps_open(struct inode *inode, struct file *file)
1733 return seq_open(file, &swaps_op);
1736 static const struct file_operations proc_swaps_operations = {
1739 .llseek = seq_lseek,
1740 .release = seq_release,
1743 static int __init procswaps_init(void)
1745 proc_create("swaps", 0, NULL, &proc_swaps_operations);
1748 __initcall(procswaps_init);
1749 #endif /* CONFIG_PROC_FS */
1751 #ifdef MAX_SWAPFILES_CHECK
1752 static int __init max_swapfiles_check(void)
1754 MAX_SWAPFILES_CHECK();
1757 late_initcall(max_swapfiles_check);
1761 * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
1763 * The swapon system call
1765 SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
1767 struct swap_info_struct * p;
1769 struct block_device *bdev = NULL;
1770 struct file *swap_file = NULL;
1771 struct address_space *mapping;
1775 union swap_header *swap_header = NULL;
1776 unsigned int nr_good_pages = 0;
1779 unsigned long maxpages = 1;
1780 unsigned long swapfilepages;
1781 unsigned short *swap_map = NULL;
1782 struct page *page = NULL;
1783 struct inode *inode = NULL;
1786 if (!capable(CAP_SYS_ADMIN))
1788 spin_lock(&swap_lock);
1790 for (type = 0 ; type < nr_swapfiles ; type++,p++)
1791 if (!(p->flags & SWP_USED))
1794 if (type >= MAX_SWAPFILES) {
1795 spin_unlock(&swap_lock);
1798 if (type >= nr_swapfiles)
1799 nr_swapfiles = type+1;
1800 memset(p, 0, sizeof(*p));
1801 INIT_LIST_HEAD(&p->extent_list);
1802 p->flags = SWP_USED;
1804 spin_unlock(&swap_lock);
1805 name = getname(specialfile);
1806 error = PTR_ERR(name);
1811 swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0);
1812 error = PTR_ERR(swap_file);
1813 if (IS_ERR(swap_file)) {
1818 p->swap_file = swap_file;
1819 mapping = swap_file->f_mapping;
1820 inode = mapping->host;
1823 for (i = 0; i < nr_swapfiles; i++) {
1824 struct swap_info_struct *q = &swap_info[i];
1826 if (i == type || !q->swap_file)
1828 if (mapping == q->swap_file->f_mapping)
1833 if (S_ISBLK(inode->i_mode)) {
1834 bdev = I_BDEV(inode);
1835 error = bd_claim(bdev, sys_swapon);
1841 p->old_block_size = block_size(bdev);
1842 error = set_blocksize(bdev, PAGE_SIZE);
1846 } else if (S_ISREG(inode->i_mode)) {
1847 p->bdev = inode->i_sb->s_bdev;
1848 mutex_lock(&inode->i_mutex);
1850 if (IS_SWAPFILE(inode)) {
1858 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
1861 * Read the swap header.
1863 if (!mapping->a_ops->readpage) {
1867 page = read_mapping_page(mapping, 0, swap_file);
1869 error = PTR_ERR(page);
1872 swap_header = kmap(page);
1874 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
1875 printk(KERN_ERR "Unable to find swap-space signature\n");
1880 /* swap partition endianess hack... */
1881 if (swab32(swap_header->info.version) == 1) {
1882 swab32s(&swap_header->info.version);
1883 swab32s(&swap_header->info.last_page);
1884 swab32s(&swap_header->info.nr_badpages);
1885 for (i = 0; i < swap_header->info.nr_badpages; i++)
1886 swab32s(&swap_header->info.badpages[i]);
1888 /* Check the swap header's sub-version */
1889 if (swap_header->info.version != 1) {
1891 "Unable to handle swap header version %d\n",
1892 swap_header->info.version);
1898 p->cluster_next = 1;
1901 * Find out how many pages are allowed for a single swap
1902 * device. There are two limiting factors: 1) the number of
1903 * bits for the swap offset in the swp_entry_t type and
1904 * 2) the number of bits in the a swap pte as defined by
1905 * the different architectures. In order to find the
1906 * largest possible bit mask a swap entry with swap type 0
1907 * and swap offset ~0UL is created, encoded to a swap pte,
1908 * decoded to a swp_entry_t again and finally the swap
1909 * offset is extracted. This will mask all the bits from
1910 * the initial ~0UL mask that can't be encoded in either
1911 * the swp_entry_t or the architecture definition of a
1914 maxpages = swp_offset(pte_to_swp_entry(
1915 swp_entry_to_pte(swp_entry(0, ~0UL)))) - 1;
1916 if (maxpages > swap_header->info.last_page)
1917 maxpages = swap_header->info.last_page;
1918 p->highest_bit = maxpages - 1;
1923 if (swapfilepages && maxpages > swapfilepages) {
1925 "Swap area shorter than signature indicates\n");
1928 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
1930 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
1933 /* OK, set up the swap map and apply the bad block list */
1934 swap_map = vmalloc(maxpages * sizeof(short));
1940 memset(swap_map, 0, maxpages * sizeof(short));
1941 for (i = 0; i < swap_header->info.nr_badpages; i++) {
1942 int page_nr = swap_header->info.badpages[i];
1943 if (page_nr <= 0 || page_nr >= swap_header->info.last_page) {
1947 swap_map[page_nr] = SWAP_MAP_BAD;
1950 error = swap_cgroup_swapon(type, maxpages);
1954 nr_good_pages = swap_header->info.last_page -
1955 swap_header->info.nr_badpages -
1956 1 /* header page */;
1958 if (nr_good_pages) {
1959 swap_map[0] = SWAP_MAP_BAD;
1961 p->pages = nr_good_pages;
1962 nr_extents = setup_swap_extents(p, &span);
1963 if (nr_extents < 0) {
1967 nr_good_pages = p->pages;
1969 if (!nr_good_pages) {
1970 printk(KERN_WARNING "Empty swap-file\n");
1975 if (blk_queue_nonrot(bdev_get_queue(p->bdev))) {
1976 p->flags |= SWP_SOLIDSTATE;
1977 p->cluster_next = 1 + (random32() % p->highest_bit);
1979 if (discard_swap(p) == 0)
1980 p->flags |= SWP_DISCARDABLE;
1982 mutex_lock(&swapon_mutex);
1983 spin_lock(&swap_lock);
1984 if (swap_flags & SWAP_FLAG_PREFER)
1986 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
1988 p->prio = --least_priority;
1989 p->swap_map = swap_map;
1990 p->flags |= SWP_WRITEOK;
1991 nr_swap_pages += nr_good_pages;
1992 total_swap_pages += nr_good_pages;
1994 printk(KERN_INFO "Adding %uk swap on %s. "
1995 "Priority:%d extents:%d across:%lluk %s%s\n",
1996 nr_good_pages<<(PAGE_SHIFT-10), name, p->prio,
1997 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
1998 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
1999 (p->flags & SWP_DISCARDABLE) ? "D" : "");
2001 /* insert swap space into swap_list: */
2003 for (i = swap_list.head; i >= 0; i = swap_info[i].next) {
2004 if (p->prio >= swap_info[i].prio) {
2011 swap_list.head = swap_list.next = p - swap_info;
2013 swap_info[prev].next = p - swap_info;
2015 spin_unlock(&swap_lock);
2016 mutex_unlock(&swapon_mutex);
2021 set_blocksize(bdev, p->old_block_size);
2024 destroy_swap_extents(p);
2025 swap_cgroup_swapoff(type);
2027 spin_lock(&swap_lock);
2028 p->swap_file = NULL;
2030 spin_unlock(&swap_lock);
2033 filp_close(swap_file, NULL);
2035 if (page && !IS_ERR(page)) {
2037 page_cache_release(page);
2043 inode->i_flags |= S_SWAPFILE;
2044 mutex_unlock(&inode->i_mutex);
2049 void si_swapinfo(struct sysinfo *val)
2052 unsigned long nr_to_be_unused = 0;
2054 spin_lock(&swap_lock);
2055 for (i = 0; i < nr_swapfiles; i++) {
2056 if (!(swap_info[i].flags & SWP_USED) ||
2057 (swap_info[i].flags & SWP_WRITEOK))
2059 nr_to_be_unused += swap_info[i].inuse_pages;
2061 val->freeswap = nr_swap_pages + nr_to_be_unused;
2062 val->totalswap = total_swap_pages + nr_to_be_unused;
2063 spin_unlock(&swap_lock);
2067 * Verify that a swap entry is valid and increment its swap map count.
2069 * Note: if swap_map[] reaches SWAP_MAP_MAX the entries are treated as
2070 * "permanent", but will be reclaimed by the next swapoff.
2071 * Returns error code in following case.
2073 * - swp_entry is invalid -> EINVAL
2074 * - swp_entry is migration entry -> EINVAL
2075 * - swap-cache reference is requested but there is already one. -> EEXIST
2076 * - swap-cache reference is requested but the entry is not used. -> ENOENT
2078 static int __swap_duplicate(swp_entry_t entry, bool cache)
2080 struct swap_info_struct * p;
2081 unsigned long offset, type;
2082 int result = -EINVAL;
2086 if (is_migration_entry(entry))
2089 type = swp_type(entry);
2090 if (type >= nr_swapfiles)
2092 p = type + swap_info;
2093 offset = swp_offset(entry);
2095 spin_lock(&swap_lock);
2097 if (unlikely(offset >= p->max))
2100 count = swap_count(p->swap_map[offset]);
2101 has_cache = swap_has_cache(p->swap_map[offset]);
2103 if (cache == SWAP_CACHE) { /* called for swapcache/swapin-readahead */
2105 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
2106 if (!has_cache && count) {
2107 p->swap_map[offset] = encode_swapmap(count, true);
2109 } else if (has_cache) /* someone added cache */
2111 else if (!count) /* no users */
2114 } else if (count || has_cache) {
2115 if (count < SWAP_MAP_MAX - 1) {
2116 p->swap_map[offset] = encode_swapmap(count + 1,
2119 } else if (count <= SWAP_MAP_MAX) {
2120 if (swap_overflow++ < 5)
2122 "swap_dup: swap entry overflow\n");
2123 p->swap_map[offset] = encode_swapmap(SWAP_MAP_MAX,
2128 result = -ENOENT; /* unused swap entry */
2130 spin_unlock(&swap_lock);
2135 printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
2139 * increase reference count of swap entry by 1.
2141 void swap_duplicate(swp_entry_t entry)
2143 __swap_duplicate(entry, SWAP_MAP);
2147 * @entry: swap entry for which we allocate swap cache.
2149 * Called when allocating swap cache for exising swap entry,
2150 * This can return error codes. Returns 0 at success.
2151 * -EBUSY means there is a swap cache.
2152 * Note: return code is different from swap_duplicate().
2154 int swapcache_prepare(swp_entry_t entry)
2156 return __swap_duplicate(entry, SWAP_CACHE);
2160 struct swap_info_struct *
2161 get_swap_info_struct(unsigned type)
2163 return &swap_info[type];
2167 * swap_lock prevents swap_map being freed. Don't grab an extra
2168 * reference on the swaphandle, it doesn't matter if it becomes unused.
2170 int valid_swaphandles(swp_entry_t entry, unsigned long *offset)
2172 struct swap_info_struct *si;
2173 int our_page_cluster = page_cluster;
2174 pgoff_t target, toff;
2178 if (!our_page_cluster) /* no readahead */
2181 si = &swap_info[swp_type(entry)];
2182 target = swp_offset(entry);
2183 base = (target >> our_page_cluster) << our_page_cluster;
2184 end = base + (1 << our_page_cluster);
2185 if (!base) /* first page is swap header */
2188 spin_lock(&swap_lock);
2189 if (end > si->max) /* don't go beyond end of map */
2192 /* Count contiguous allocated slots above our target */
2193 for (toff = target; ++toff < end; nr_pages++) {
2194 /* Don't read in free or bad pages */
2195 if (!si->swap_map[toff])
2197 if (swap_count(si->swap_map[toff]) == SWAP_MAP_BAD)
2200 /* Count contiguous allocated slots below our target */
2201 for (toff = target; --toff >= base; nr_pages++) {
2202 /* Don't read in free or bad pages */
2203 if (!si->swap_map[toff])
2205 if (swap_count(si->swap_map[toff]) == SWAP_MAP_BAD)
2208 spin_unlock(&swap_lock);
2211 * Indicate starting offset, and return number of pages to get:
2212 * if only 1, say 0, since there's then no readahead to be done.
2215 return nr_pages? ++nr_pages: 0;