x86: hpet: allow force enable on ICH10 HPET
[linux-2.6/linux-2.6-openrd.git] / mm / swapfile.c
blobda422c47e2ee515bb68c131f0980c603b7d1b380
1 /*
2 * linux/mm/swapfile.c
4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 * Swap reorganised 29.12.95, Stephen Tweedie
6 */
8 #include <linux/mm.h>
9 #include <linux/hugetlb.h>
10 #include <linux/mman.h>
11 #include <linux/slab.h>
12 #include <linux/kernel_stat.h>
13 #include <linux/swap.h>
14 #include <linux/vmalloc.h>
15 #include <linux/pagemap.h>
16 #include <linux/namei.h>
17 #include <linux/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;
40 long nr_swap_pages;
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);
57 * We need this because the bdev->unplug_fn can sleep and we cannot
58 * hold swap_lock while calling the unplug_fn. And swap_lock
59 * cannot be turned into a mutex.
61 static DECLARE_RWSEM(swap_unplug_sem);
63 void swap_unplug_io_fn(struct backing_dev_info *unused_bdi, struct page *page)
65 swp_entry_t entry;
67 down_read(&swap_unplug_sem);
68 entry.val = page_private(page);
69 if (PageSwapCache(page)) {
70 struct block_device *bdev = swap_info[swp_type(entry)].bdev;
71 struct backing_dev_info *bdi;
74 * If the page is removed from swapcache from under us (with a
75 * racy try_to_unuse/swapoff) we need an additional reference
76 * count to avoid reading garbage from page_private(page) above.
77 * If the WARN_ON triggers during a swapoff it maybe the race
78 * condition and it's harmless. However if it triggers without
79 * swapoff it signals a problem.
81 WARN_ON(page_count(page) <= 1);
83 bdi = bdev->bd_inode->i_mapping->backing_dev_info;
84 blk_run_backing_dev(bdi, page);
86 up_read(&swap_unplug_sem);
90 * swapon tell device that all the old swap contents can be discarded,
91 * to allow the swap device to optimize its wear-levelling.
93 static int discard_swap(struct swap_info_struct *si)
95 struct swap_extent *se;
96 int err = 0;
98 list_for_each_entry(se, &si->extent_list, list) {
99 sector_t start_block = se->start_block << (PAGE_SHIFT - 9);
100 sector_t nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
102 if (se->start_page == 0) {
103 /* Do not discard the swap header page! */
104 start_block += 1 << (PAGE_SHIFT - 9);
105 nr_blocks -= 1 << (PAGE_SHIFT - 9);
106 if (!nr_blocks)
107 continue;
110 err = blkdev_issue_discard(si->bdev, start_block,
111 nr_blocks, GFP_KERNEL);
112 if (err)
113 break;
115 cond_resched();
117 return err; /* That will often be -EOPNOTSUPP */
121 * swap allocation tell device that a cluster of swap can now be discarded,
122 * to allow the swap device to optimize its wear-levelling.
124 static void discard_swap_cluster(struct swap_info_struct *si,
125 pgoff_t start_page, pgoff_t nr_pages)
127 struct swap_extent *se = si->curr_swap_extent;
128 int found_extent = 0;
130 while (nr_pages) {
131 struct list_head *lh;
133 if (se->start_page <= start_page &&
134 start_page < se->start_page + se->nr_pages) {
135 pgoff_t offset = start_page - se->start_page;
136 sector_t start_block = se->start_block + offset;
137 sector_t nr_blocks = se->nr_pages - offset;
139 if (nr_blocks > nr_pages)
140 nr_blocks = nr_pages;
141 start_page += nr_blocks;
142 nr_pages -= nr_blocks;
144 if (!found_extent++)
145 si->curr_swap_extent = se;
147 start_block <<= PAGE_SHIFT - 9;
148 nr_blocks <<= PAGE_SHIFT - 9;
149 if (blkdev_issue_discard(si->bdev, start_block,
150 nr_blocks, GFP_NOIO))
151 break;
154 lh = se->list.next;
155 if (lh == &si->extent_list)
156 lh = lh->next;
157 se = list_entry(lh, struct swap_extent, list);
161 static int wait_for_discard(void *word)
163 schedule();
164 return 0;
167 #define SWAPFILE_CLUSTER 256
168 #define LATENCY_LIMIT 256
170 static inline unsigned long scan_swap_map(struct swap_info_struct *si)
172 unsigned long offset;
173 unsigned long scan_base;
174 unsigned long last_in_cluster = 0;
175 int latency_ration = LATENCY_LIMIT;
176 int found_free_cluster = 0;
179 * We try to cluster swap pages by allocating them sequentially
180 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
181 * way, however, we resort to first-free allocation, starting
182 * a new cluster. This prevents us from scattering swap pages
183 * all over the entire swap partition, so that we reduce
184 * overall disk seek times between swap pages. -- sct
185 * But we do now try to find an empty cluster. -Andrea
186 * And we let swap pages go all over an SSD partition. Hugh
189 si->flags += SWP_SCANNING;
190 scan_base = offset = si->cluster_next;
192 if (unlikely(!si->cluster_nr--)) {
193 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
194 si->cluster_nr = SWAPFILE_CLUSTER - 1;
195 goto checks;
197 if (si->flags & SWP_DISCARDABLE) {
199 * Start range check on racing allocations, in case
200 * they overlap the cluster we eventually decide on
201 * (we scan without swap_lock to allow preemption).
202 * It's hardly conceivable that cluster_nr could be
203 * wrapped during our scan, but don't depend on it.
205 if (si->lowest_alloc)
206 goto checks;
207 si->lowest_alloc = si->max;
208 si->highest_alloc = 0;
210 spin_unlock(&swap_lock);
213 * If seek is expensive, start searching for new cluster from
214 * start of partition, to minimize the span of allocated swap.
215 * But if seek is cheap, search from our current position, so
216 * that swap is allocated from all over the partition: if the
217 * Flash Translation Layer only remaps within limited zones,
218 * we don't want to wear out the first zone too quickly.
220 if (!(si->flags & SWP_SOLIDSTATE))
221 scan_base = offset = si->lowest_bit;
222 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
224 /* Locate the first empty (unaligned) cluster */
225 for (; last_in_cluster <= si->highest_bit; offset++) {
226 if (si->swap_map[offset])
227 last_in_cluster = offset + SWAPFILE_CLUSTER;
228 else if (offset == last_in_cluster) {
229 spin_lock(&swap_lock);
230 offset -= SWAPFILE_CLUSTER - 1;
231 si->cluster_next = offset;
232 si->cluster_nr = SWAPFILE_CLUSTER - 1;
233 found_free_cluster = 1;
234 goto checks;
236 if (unlikely(--latency_ration < 0)) {
237 cond_resched();
238 latency_ration = LATENCY_LIMIT;
242 offset = si->lowest_bit;
243 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
245 /* Locate the first empty (unaligned) cluster */
246 for (; last_in_cluster < scan_base; offset++) {
247 if (si->swap_map[offset])
248 last_in_cluster = offset + SWAPFILE_CLUSTER;
249 else if (offset == last_in_cluster) {
250 spin_lock(&swap_lock);
251 offset -= SWAPFILE_CLUSTER - 1;
252 si->cluster_next = offset;
253 si->cluster_nr = SWAPFILE_CLUSTER - 1;
254 found_free_cluster = 1;
255 goto checks;
257 if (unlikely(--latency_ration < 0)) {
258 cond_resched();
259 latency_ration = LATENCY_LIMIT;
263 offset = scan_base;
264 spin_lock(&swap_lock);
265 si->cluster_nr = SWAPFILE_CLUSTER - 1;
266 si->lowest_alloc = 0;
269 checks:
270 if (!(si->flags & SWP_WRITEOK))
271 goto no_page;
272 if (!si->highest_bit)
273 goto no_page;
274 if (offset > si->highest_bit)
275 scan_base = offset = si->lowest_bit;
276 if (si->swap_map[offset])
277 goto scan;
279 if (offset == si->lowest_bit)
280 si->lowest_bit++;
281 if (offset == si->highest_bit)
282 si->highest_bit--;
283 si->inuse_pages++;
284 if (si->inuse_pages == si->pages) {
285 si->lowest_bit = si->max;
286 si->highest_bit = 0;
288 si->swap_map[offset] = 1;
289 si->cluster_next = offset + 1;
290 si->flags -= SWP_SCANNING;
292 if (si->lowest_alloc) {
294 * Only set when SWP_DISCARDABLE, and there's a scan
295 * for a free cluster in progress or just completed.
297 if (found_free_cluster) {
299 * To optimize wear-levelling, discard the
300 * old data of the cluster, taking care not to
301 * discard any of its pages that have already
302 * been allocated by racing tasks (offset has
303 * already stepped over any at the beginning).
305 if (offset < si->highest_alloc &&
306 si->lowest_alloc <= last_in_cluster)
307 last_in_cluster = si->lowest_alloc - 1;
308 si->flags |= SWP_DISCARDING;
309 spin_unlock(&swap_lock);
311 if (offset < last_in_cluster)
312 discard_swap_cluster(si, offset,
313 last_in_cluster - offset + 1);
315 spin_lock(&swap_lock);
316 si->lowest_alloc = 0;
317 si->flags &= ~SWP_DISCARDING;
319 smp_mb(); /* wake_up_bit advises this */
320 wake_up_bit(&si->flags, ilog2(SWP_DISCARDING));
322 } else if (si->flags & SWP_DISCARDING) {
324 * Delay using pages allocated by racing tasks
325 * until the whole discard has been issued. We
326 * could defer that delay until swap_writepage,
327 * but it's easier to keep this self-contained.
329 spin_unlock(&swap_lock);
330 wait_on_bit(&si->flags, ilog2(SWP_DISCARDING),
331 wait_for_discard, TASK_UNINTERRUPTIBLE);
332 spin_lock(&swap_lock);
333 } else {
335 * Note pages allocated by racing tasks while
336 * scan for a free cluster is in progress, so
337 * that its final discard can exclude them.
339 if (offset < si->lowest_alloc)
340 si->lowest_alloc = offset;
341 if (offset > si->highest_alloc)
342 si->highest_alloc = offset;
345 return offset;
347 scan:
348 spin_unlock(&swap_lock);
349 while (++offset <= si->highest_bit) {
350 if (!si->swap_map[offset]) {
351 spin_lock(&swap_lock);
352 goto checks;
354 if (unlikely(--latency_ration < 0)) {
355 cond_resched();
356 latency_ration = LATENCY_LIMIT;
359 offset = si->lowest_bit;
360 while (++offset < scan_base) {
361 if (!si->swap_map[offset]) {
362 spin_lock(&swap_lock);
363 goto checks;
365 if (unlikely(--latency_ration < 0)) {
366 cond_resched();
367 latency_ration = LATENCY_LIMIT;
370 spin_lock(&swap_lock);
372 no_page:
373 si->flags -= SWP_SCANNING;
374 return 0;
377 swp_entry_t get_swap_page(void)
379 struct swap_info_struct *si;
380 pgoff_t offset;
381 int type, next;
382 int wrapped = 0;
384 spin_lock(&swap_lock);
385 if (nr_swap_pages <= 0)
386 goto noswap;
387 nr_swap_pages--;
389 for (type = swap_list.next; type >= 0 && wrapped < 2; type = next) {
390 si = swap_info + type;
391 next = si->next;
392 if (next < 0 ||
393 (!wrapped && si->prio != swap_info[next].prio)) {
394 next = swap_list.head;
395 wrapped++;
398 if (!si->highest_bit)
399 continue;
400 if (!(si->flags & SWP_WRITEOK))
401 continue;
403 swap_list.next = next;
404 offset = scan_swap_map(si);
405 if (offset) {
406 spin_unlock(&swap_lock);
407 return swp_entry(type, offset);
409 next = swap_list.next;
412 nr_swap_pages++;
413 noswap:
414 spin_unlock(&swap_lock);
415 return (swp_entry_t) {0};
418 swp_entry_t get_swap_page_of_type(int type)
420 struct swap_info_struct *si;
421 pgoff_t offset;
423 spin_lock(&swap_lock);
424 si = swap_info + type;
425 if (si->flags & SWP_WRITEOK) {
426 nr_swap_pages--;
427 offset = scan_swap_map(si);
428 if (offset) {
429 spin_unlock(&swap_lock);
430 return swp_entry(type, offset);
432 nr_swap_pages++;
434 spin_unlock(&swap_lock);
435 return (swp_entry_t) {0};
438 static struct swap_info_struct * swap_info_get(swp_entry_t entry)
440 struct swap_info_struct * p;
441 unsigned long offset, type;
443 if (!entry.val)
444 goto out;
445 type = swp_type(entry);
446 if (type >= nr_swapfiles)
447 goto bad_nofile;
448 p = & swap_info[type];
449 if (!(p->flags & SWP_USED))
450 goto bad_device;
451 offset = swp_offset(entry);
452 if (offset >= p->max)
453 goto bad_offset;
454 if (!p->swap_map[offset])
455 goto bad_free;
456 spin_lock(&swap_lock);
457 return p;
459 bad_free:
460 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val);
461 goto out;
462 bad_offset:
463 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val);
464 goto out;
465 bad_device:
466 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val);
467 goto out;
468 bad_nofile:
469 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val);
470 out:
471 return NULL;
474 static int swap_entry_free(struct swap_info_struct *p, swp_entry_t ent)
476 unsigned long offset = swp_offset(ent);
477 int count = p->swap_map[offset];
479 if (count < SWAP_MAP_MAX) {
480 count--;
481 p->swap_map[offset] = count;
482 if (!count) {
483 if (offset < p->lowest_bit)
484 p->lowest_bit = offset;
485 if (offset > p->highest_bit)
486 p->highest_bit = offset;
487 if (p->prio > swap_info[swap_list.next].prio)
488 swap_list.next = p - swap_info;
489 nr_swap_pages++;
490 p->inuse_pages--;
491 mem_cgroup_uncharge_swap(ent);
494 return count;
498 * Caller has made sure that the swapdevice corresponding to entry
499 * is still around or has not been recycled.
501 void swap_free(swp_entry_t entry)
503 struct swap_info_struct * p;
505 p = swap_info_get(entry);
506 if (p) {
507 swap_entry_free(p, entry);
508 spin_unlock(&swap_lock);
513 * How many references to page are currently swapped out?
515 static inline int page_swapcount(struct page *page)
517 int count = 0;
518 struct swap_info_struct *p;
519 swp_entry_t entry;
521 entry.val = page_private(page);
522 p = swap_info_get(entry);
523 if (p) {
524 /* Subtract the 1 for the swap cache itself */
525 count = p->swap_map[swp_offset(entry)] - 1;
526 spin_unlock(&swap_lock);
528 return count;
532 * We can write to an anon page without COW if there are no other references
533 * to it. And as a side-effect, free up its swap: because the old content
534 * on disk will never be read, and seeking back there to write new content
535 * later would only waste time away from clustering.
537 int reuse_swap_page(struct page *page)
539 int count;
541 VM_BUG_ON(!PageLocked(page));
542 count = page_mapcount(page);
543 if (count <= 1 && PageSwapCache(page)) {
544 count += page_swapcount(page);
545 if (count == 1 && !PageWriteback(page)) {
546 delete_from_swap_cache(page);
547 SetPageDirty(page);
550 return count == 1;
554 * If swap is getting full, or if there are no more mappings of this page,
555 * then try_to_free_swap is called to free its swap space.
557 int try_to_free_swap(struct page *page)
559 VM_BUG_ON(!PageLocked(page));
561 if (!PageSwapCache(page))
562 return 0;
563 if (PageWriteback(page))
564 return 0;
565 if (page_swapcount(page))
566 return 0;
568 delete_from_swap_cache(page);
569 SetPageDirty(page);
570 return 1;
574 * Free the swap entry like above, but also try to
575 * free the page cache entry if it is the last user.
577 int free_swap_and_cache(swp_entry_t entry)
579 struct swap_info_struct *p;
580 struct page *page = NULL;
582 if (is_migration_entry(entry))
583 return 1;
585 p = swap_info_get(entry);
586 if (p) {
587 if (swap_entry_free(p, entry) == 1) {
588 page = find_get_page(&swapper_space, entry.val);
589 if (page && !trylock_page(page)) {
590 page_cache_release(page);
591 page = NULL;
594 spin_unlock(&swap_lock);
596 if (page) {
598 * Not mapped elsewhere, or swap space full? Free it!
599 * Also recheck PageSwapCache now page is locked (above).
601 if (PageSwapCache(page) && !PageWriteback(page) &&
602 (!page_mapped(page) || vm_swap_full())) {
603 delete_from_swap_cache(page);
604 SetPageDirty(page);
606 unlock_page(page);
607 page_cache_release(page);
609 return p != NULL;
612 #ifdef CONFIG_HIBERNATION
614 * Find the swap type that corresponds to given device (if any).
616 * @offset - number of the PAGE_SIZE-sized block of the device, starting
617 * from 0, in which the swap header is expected to be located.
619 * This is needed for the suspend to disk (aka swsusp).
621 int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
623 struct block_device *bdev = NULL;
624 int i;
626 if (device)
627 bdev = bdget(device);
629 spin_lock(&swap_lock);
630 for (i = 0; i < nr_swapfiles; i++) {
631 struct swap_info_struct *sis = swap_info + i;
633 if (!(sis->flags & SWP_WRITEOK))
634 continue;
636 if (!bdev) {
637 if (bdev_p)
638 *bdev_p = sis->bdev;
640 spin_unlock(&swap_lock);
641 return i;
643 if (bdev == sis->bdev) {
644 struct swap_extent *se;
646 se = list_entry(sis->extent_list.next,
647 struct swap_extent, list);
648 if (se->start_block == offset) {
649 if (bdev_p)
650 *bdev_p = sis->bdev;
652 spin_unlock(&swap_lock);
653 bdput(bdev);
654 return i;
658 spin_unlock(&swap_lock);
659 if (bdev)
660 bdput(bdev);
662 return -ENODEV;
666 * Return either the total number of swap pages of given type, or the number
667 * of free pages of that type (depending on @free)
669 * This is needed for software suspend
671 unsigned int count_swap_pages(int type, int free)
673 unsigned int n = 0;
675 if (type < nr_swapfiles) {
676 spin_lock(&swap_lock);
677 if (swap_info[type].flags & SWP_WRITEOK) {
678 n = swap_info[type].pages;
679 if (free)
680 n -= swap_info[type].inuse_pages;
682 spin_unlock(&swap_lock);
684 return n;
686 #endif
689 * No need to decide whether this PTE shares the swap entry with others,
690 * just let do_wp_page work it out if a write is requested later - to
691 * force COW, vm_page_prot omits write permission from any private vma.
693 static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
694 unsigned long addr, swp_entry_t entry, struct page *page)
696 struct mem_cgroup *ptr = NULL;
697 spinlock_t *ptl;
698 pte_t *pte;
699 int ret = 1;
701 if (mem_cgroup_try_charge_swapin(vma->vm_mm, page, GFP_KERNEL, &ptr))
702 ret = -ENOMEM;
704 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
705 if (unlikely(!pte_same(*pte, swp_entry_to_pte(entry)))) {
706 if (ret > 0)
707 mem_cgroup_cancel_charge_swapin(ptr);
708 ret = 0;
709 goto out;
712 inc_mm_counter(vma->vm_mm, anon_rss);
713 get_page(page);
714 set_pte_at(vma->vm_mm, addr, pte,
715 pte_mkold(mk_pte(page, vma->vm_page_prot)));
716 page_add_anon_rmap(page, vma, addr);
717 mem_cgroup_commit_charge_swapin(page, ptr);
718 swap_free(entry);
720 * Move the page to the active list so it is not
721 * immediately swapped out again after swapon.
723 activate_page(page);
724 out:
725 pte_unmap_unlock(pte, ptl);
726 return ret;
729 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
730 unsigned long addr, unsigned long end,
731 swp_entry_t entry, struct page *page)
733 pte_t swp_pte = swp_entry_to_pte(entry);
734 pte_t *pte;
735 int ret = 0;
738 * We don't actually need pte lock while scanning for swp_pte: since
739 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
740 * page table while we're scanning; though it could get zapped, and on
741 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
742 * of unmatched parts which look like swp_pte, so unuse_pte must
743 * recheck under pte lock. Scanning without pte lock lets it be
744 * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
746 pte = pte_offset_map(pmd, addr);
747 do {
749 * swapoff spends a _lot_ of time in this loop!
750 * Test inline before going to call unuse_pte.
752 if (unlikely(pte_same(*pte, swp_pte))) {
753 pte_unmap(pte);
754 ret = unuse_pte(vma, pmd, addr, entry, page);
755 if (ret)
756 goto out;
757 pte = pte_offset_map(pmd, addr);
759 } while (pte++, addr += PAGE_SIZE, addr != end);
760 pte_unmap(pte - 1);
761 out:
762 return ret;
765 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
766 unsigned long addr, unsigned long end,
767 swp_entry_t entry, struct page *page)
769 pmd_t *pmd;
770 unsigned long next;
771 int ret;
773 pmd = pmd_offset(pud, addr);
774 do {
775 next = pmd_addr_end(addr, end);
776 if (pmd_none_or_clear_bad(pmd))
777 continue;
778 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
779 if (ret)
780 return ret;
781 } while (pmd++, addr = next, addr != end);
782 return 0;
785 static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
786 unsigned long addr, unsigned long end,
787 swp_entry_t entry, struct page *page)
789 pud_t *pud;
790 unsigned long next;
791 int ret;
793 pud = pud_offset(pgd, addr);
794 do {
795 next = pud_addr_end(addr, end);
796 if (pud_none_or_clear_bad(pud))
797 continue;
798 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
799 if (ret)
800 return ret;
801 } while (pud++, addr = next, addr != end);
802 return 0;
805 static int unuse_vma(struct vm_area_struct *vma,
806 swp_entry_t entry, struct page *page)
808 pgd_t *pgd;
809 unsigned long addr, end, next;
810 int ret;
812 if (page->mapping) {
813 addr = page_address_in_vma(page, vma);
814 if (addr == -EFAULT)
815 return 0;
816 else
817 end = addr + PAGE_SIZE;
818 } else {
819 addr = vma->vm_start;
820 end = vma->vm_end;
823 pgd = pgd_offset(vma->vm_mm, addr);
824 do {
825 next = pgd_addr_end(addr, end);
826 if (pgd_none_or_clear_bad(pgd))
827 continue;
828 ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
829 if (ret)
830 return ret;
831 } while (pgd++, addr = next, addr != end);
832 return 0;
835 static int unuse_mm(struct mm_struct *mm,
836 swp_entry_t entry, struct page *page)
838 struct vm_area_struct *vma;
839 int ret = 0;
841 if (!down_read_trylock(&mm->mmap_sem)) {
843 * Activate page so shrink_inactive_list is unlikely to unmap
844 * its ptes while lock is dropped, so swapoff can make progress.
846 activate_page(page);
847 unlock_page(page);
848 down_read(&mm->mmap_sem);
849 lock_page(page);
851 for (vma = mm->mmap; vma; vma = vma->vm_next) {
852 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
853 break;
855 up_read(&mm->mmap_sem);
856 return (ret < 0)? ret: 0;
860 * Scan swap_map from current position to next entry still in use.
861 * Recycle to start on reaching the end, returning 0 when empty.
863 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
864 unsigned int prev)
866 unsigned int max = si->max;
867 unsigned int i = prev;
868 int count;
871 * No need for swap_lock here: we're just looking
872 * for whether an entry is in use, not modifying it; false
873 * hits are okay, and sys_swapoff() has already prevented new
874 * allocations from this area (while holding swap_lock).
876 for (;;) {
877 if (++i >= max) {
878 if (!prev) {
879 i = 0;
880 break;
883 * No entries in use at top of swap_map,
884 * loop back to start and recheck there.
886 max = prev + 1;
887 prev = 0;
888 i = 1;
890 count = si->swap_map[i];
891 if (count && count != SWAP_MAP_BAD)
892 break;
894 return i;
898 * We completely avoid races by reading each swap page in advance,
899 * and then search for the process using it. All the necessary
900 * page table adjustments can then be made atomically.
902 static int try_to_unuse(unsigned int type)
904 struct swap_info_struct * si = &swap_info[type];
905 struct mm_struct *start_mm;
906 unsigned short *swap_map;
907 unsigned short swcount;
908 struct page *page;
909 swp_entry_t entry;
910 unsigned int i = 0;
911 int retval = 0;
912 int reset_overflow = 0;
913 int shmem;
916 * When searching mms for an entry, a good strategy is to
917 * start at the first mm we freed the previous entry from
918 * (though actually we don't notice whether we or coincidence
919 * freed the entry). Initialize this start_mm with a hold.
921 * A simpler strategy would be to start at the last mm we
922 * freed the previous entry from; but that would take less
923 * advantage of mmlist ordering, which clusters forked mms
924 * together, child after parent. If we race with dup_mmap(), we
925 * prefer to resolve parent before child, lest we miss entries
926 * duplicated after we scanned child: using last mm would invert
927 * that. Though it's only a serious concern when an overflowed
928 * swap count is reset from SWAP_MAP_MAX, preventing a rescan.
930 start_mm = &init_mm;
931 atomic_inc(&init_mm.mm_users);
934 * Keep on scanning until all entries have gone. Usually,
935 * one pass through swap_map is enough, but not necessarily:
936 * there are races when an instance of an entry might be missed.
938 while ((i = find_next_to_unuse(si, i)) != 0) {
939 if (signal_pending(current)) {
940 retval = -EINTR;
941 break;
945 * Get a page for the entry, using the existing swap
946 * cache page if there is one. Otherwise, get a clean
947 * page and read the swap into it.
949 swap_map = &si->swap_map[i];
950 entry = swp_entry(type, i);
951 page = read_swap_cache_async(entry,
952 GFP_HIGHUSER_MOVABLE, NULL, 0);
953 if (!page) {
955 * Either swap_duplicate() failed because entry
956 * has been freed independently, and will not be
957 * reused since sys_swapoff() already disabled
958 * allocation from here, or alloc_page() failed.
960 if (!*swap_map)
961 continue;
962 retval = -ENOMEM;
963 break;
967 * Don't hold on to start_mm if it looks like exiting.
969 if (atomic_read(&start_mm->mm_users) == 1) {
970 mmput(start_mm);
971 start_mm = &init_mm;
972 atomic_inc(&init_mm.mm_users);
976 * Wait for and lock page. When do_swap_page races with
977 * try_to_unuse, do_swap_page can handle the fault much
978 * faster than try_to_unuse can locate the entry. This
979 * apparently redundant "wait_on_page_locked" lets try_to_unuse
980 * defer to do_swap_page in such a case - in some tests,
981 * do_swap_page and try_to_unuse repeatedly compete.
983 wait_on_page_locked(page);
984 wait_on_page_writeback(page);
985 lock_page(page);
986 wait_on_page_writeback(page);
989 * Remove all references to entry.
990 * Whenever we reach init_mm, there's no address space
991 * to search, but use it as a reminder to search shmem.
993 shmem = 0;
994 swcount = *swap_map;
995 if (swcount > 1) {
996 if (start_mm == &init_mm)
997 shmem = shmem_unuse(entry, page);
998 else
999 retval = unuse_mm(start_mm, entry, page);
1001 if (*swap_map > 1) {
1002 int set_start_mm = (*swap_map >= swcount);
1003 struct list_head *p = &start_mm->mmlist;
1004 struct mm_struct *new_start_mm = start_mm;
1005 struct mm_struct *prev_mm = start_mm;
1006 struct mm_struct *mm;
1008 atomic_inc(&new_start_mm->mm_users);
1009 atomic_inc(&prev_mm->mm_users);
1010 spin_lock(&mmlist_lock);
1011 while (*swap_map > 1 && !retval && !shmem &&
1012 (p = p->next) != &start_mm->mmlist) {
1013 mm = list_entry(p, struct mm_struct, mmlist);
1014 if (!atomic_inc_not_zero(&mm->mm_users))
1015 continue;
1016 spin_unlock(&mmlist_lock);
1017 mmput(prev_mm);
1018 prev_mm = mm;
1020 cond_resched();
1022 swcount = *swap_map;
1023 if (swcount <= 1)
1025 else if (mm == &init_mm) {
1026 set_start_mm = 1;
1027 shmem = shmem_unuse(entry, page);
1028 } else
1029 retval = unuse_mm(mm, entry, page);
1030 if (set_start_mm && *swap_map < swcount) {
1031 mmput(new_start_mm);
1032 atomic_inc(&mm->mm_users);
1033 new_start_mm = mm;
1034 set_start_mm = 0;
1036 spin_lock(&mmlist_lock);
1038 spin_unlock(&mmlist_lock);
1039 mmput(prev_mm);
1040 mmput(start_mm);
1041 start_mm = new_start_mm;
1043 if (shmem) {
1044 /* page has already been unlocked and released */
1045 if (shmem > 0)
1046 continue;
1047 retval = shmem;
1048 break;
1050 if (retval) {
1051 unlock_page(page);
1052 page_cache_release(page);
1053 break;
1057 * How could swap count reach 0x7fff when the maximum
1058 * pid is 0x7fff, and there's no way to repeat a swap
1059 * page within an mm (except in shmem, where it's the
1060 * shared object which takes the reference count)?
1061 * We believe SWAP_MAP_MAX cannot occur in Linux 2.4.
1063 * If that's wrong, then we should worry more about
1064 * exit_mmap() and do_munmap() cases described above:
1065 * we might be resetting SWAP_MAP_MAX too early here.
1066 * We know "Undead"s can happen, they're okay, so don't
1067 * report them; but do report if we reset SWAP_MAP_MAX.
1069 if (*swap_map == SWAP_MAP_MAX) {
1070 spin_lock(&swap_lock);
1071 *swap_map = 1;
1072 spin_unlock(&swap_lock);
1073 reset_overflow = 1;
1077 * If a reference remains (rare), we would like to leave
1078 * the page in the swap cache; but try_to_unmap could
1079 * then re-duplicate the entry once we drop page lock,
1080 * so we might loop indefinitely; also, that page could
1081 * not be swapped out to other storage meanwhile. So:
1082 * delete from cache even if there's another reference,
1083 * after ensuring that the data has been saved to disk -
1084 * since if the reference remains (rarer), it will be
1085 * read from disk into another page. Splitting into two
1086 * pages would be incorrect if swap supported "shared
1087 * private" pages, but they are handled by tmpfs files.
1089 if ((*swap_map > 1) && PageDirty(page) && PageSwapCache(page)) {
1090 struct writeback_control wbc = {
1091 .sync_mode = WB_SYNC_NONE,
1094 swap_writepage(page, &wbc);
1095 lock_page(page);
1096 wait_on_page_writeback(page);
1100 * It is conceivable that a racing task removed this page from
1101 * swap cache just before we acquired the page lock at the top,
1102 * or while we dropped it in unuse_mm(). The page might even
1103 * be back in swap cache on another swap area: that we must not
1104 * delete, since it may not have been written out to swap yet.
1106 if (PageSwapCache(page) &&
1107 likely(page_private(page) == entry.val))
1108 delete_from_swap_cache(page);
1111 * So we could skip searching mms once swap count went
1112 * to 1, we did not mark any present ptes as dirty: must
1113 * mark page dirty so shrink_page_list will preserve it.
1115 SetPageDirty(page);
1116 unlock_page(page);
1117 page_cache_release(page);
1120 * Make sure that we aren't completely killing
1121 * interactive performance.
1123 cond_resched();
1126 mmput(start_mm);
1127 if (reset_overflow) {
1128 printk(KERN_WARNING "swapoff: cleared swap entry overflow\n");
1129 swap_overflow = 0;
1131 return retval;
1135 * After a successful try_to_unuse, if no swap is now in use, we know
1136 * we can empty the mmlist. swap_lock must be held on entry and exit.
1137 * Note that mmlist_lock nests inside swap_lock, and an mm must be
1138 * added to the mmlist just after page_duplicate - before would be racy.
1140 static void drain_mmlist(void)
1142 struct list_head *p, *next;
1143 unsigned int i;
1145 for (i = 0; i < nr_swapfiles; i++)
1146 if (swap_info[i].inuse_pages)
1147 return;
1148 spin_lock(&mmlist_lock);
1149 list_for_each_safe(p, next, &init_mm.mmlist)
1150 list_del_init(p);
1151 spin_unlock(&mmlist_lock);
1155 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
1156 * corresponds to page offset `offset'.
1158 sector_t map_swap_page(struct swap_info_struct *sis, pgoff_t offset)
1160 struct swap_extent *se = sis->curr_swap_extent;
1161 struct swap_extent *start_se = se;
1163 for ( ; ; ) {
1164 struct list_head *lh;
1166 if (se->start_page <= offset &&
1167 offset < (se->start_page + se->nr_pages)) {
1168 return se->start_block + (offset - se->start_page);
1170 lh = se->list.next;
1171 if (lh == &sis->extent_list)
1172 lh = lh->next;
1173 se = list_entry(lh, struct swap_extent, list);
1174 sis->curr_swap_extent = se;
1175 BUG_ON(se == start_se); /* It *must* be present */
1179 #ifdef CONFIG_HIBERNATION
1181 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1182 * corresponding to given index in swap_info (swap type).
1184 sector_t swapdev_block(int swap_type, pgoff_t offset)
1186 struct swap_info_struct *sis;
1188 if (swap_type >= nr_swapfiles)
1189 return 0;
1191 sis = swap_info + swap_type;
1192 return (sis->flags & SWP_WRITEOK) ? map_swap_page(sis, offset) : 0;
1194 #endif /* CONFIG_HIBERNATION */
1197 * Free all of a swapdev's extent information
1199 static void destroy_swap_extents(struct swap_info_struct *sis)
1201 while (!list_empty(&sis->extent_list)) {
1202 struct swap_extent *se;
1204 se = list_entry(sis->extent_list.next,
1205 struct swap_extent, list);
1206 list_del(&se->list);
1207 kfree(se);
1212 * Add a block range (and the corresponding page range) into this swapdev's
1213 * extent list. The extent list is kept sorted in page order.
1215 * This function rather assumes that it is called in ascending page order.
1217 static int
1218 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1219 unsigned long nr_pages, sector_t start_block)
1221 struct swap_extent *se;
1222 struct swap_extent *new_se;
1223 struct list_head *lh;
1225 lh = sis->extent_list.prev; /* The highest page extent */
1226 if (lh != &sis->extent_list) {
1227 se = list_entry(lh, struct swap_extent, list);
1228 BUG_ON(se->start_page + se->nr_pages != start_page);
1229 if (se->start_block + se->nr_pages == start_block) {
1230 /* Merge it */
1231 se->nr_pages += nr_pages;
1232 return 0;
1237 * No merge. Insert a new extent, preserving ordering.
1239 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1240 if (new_se == NULL)
1241 return -ENOMEM;
1242 new_se->start_page = start_page;
1243 new_se->nr_pages = nr_pages;
1244 new_se->start_block = start_block;
1246 list_add_tail(&new_se->list, &sis->extent_list);
1247 return 1;
1251 * A `swap extent' is a simple thing which maps a contiguous range of pages
1252 * onto a contiguous range of disk blocks. An ordered list of swap extents
1253 * is built at swapon time and is then used at swap_writepage/swap_readpage
1254 * time for locating where on disk a page belongs.
1256 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1257 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1258 * swap files identically.
1260 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1261 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1262 * swapfiles are handled *identically* after swapon time.
1264 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1265 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1266 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1267 * requirements, they are simply tossed out - we will never use those blocks
1268 * for swapping.
1270 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
1271 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1272 * which will scribble on the fs.
1274 * The amount of disk space which a single swap extent represents varies.
1275 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1276 * extents in the list. To avoid much list walking, we cache the previous
1277 * search location in `curr_swap_extent', and start new searches from there.
1278 * This is extremely effective. The average number of iterations in
1279 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1281 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
1283 struct inode *inode;
1284 unsigned blocks_per_page;
1285 unsigned long page_no;
1286 unsigned blkbits;
1287 sector_t probe_block;
1288 sector_t last_block;
1289 sector_t lowest_block = -1;
1290 sector_t highest_block = 0;
1291 int nr_extents = 0;
1292 int ret;
1294 inode = sis->swap_file->f_mapping->host;
1295 if (S_ISBLK(inode->i_mode)) {
1296 ret = add_swap_extent(sis, 0, sis->max, 0);
1297 *span = sis->pages;
1298 goto done;
1301 blkbits = inode->i_blkbits;
1302 blocks_per_page = PAGE_SIZE >> blkbits;
1305 * Map all the blocks into the extent list. This code doesn't try
1306 * to be very smart.
1308 probe_block = 0;
1309 page_no = 0;
1310 last_block = i_size_read(inode) >> blkbits;
1311 while ((probe_block + blocks_per_page) <= last_block &&
1312 page_no < sis->max) {
1313 unsigned block_in_page;
1314 sector_t first_block;
1316 first_block = bmap(inode, probe_block);
1317 if (first_block == 0)
1318 goto bad_bmap;
1321 * It must be PAGE_SIZE aligned on-disk
1323 if (first_block & (blocks_per_page - 1)) {
1324 probe_block++;
1325 goto reprobe;
1328 for (block_in_page = 1; block_in_page < blocks_per_page;
1329 block_in_page++) {
1330 sector_t block;
1332 block = bmap(inode, probe_block + block_in_page);
1333 if (block == 0)
1334 goto bad_bmap;
1335 if (block != first_block + block_in_page) {
1336 /* Discontiguity */
1337 probe_block++;
1338 goto reprobe;
1342 first_block >>= (PAGE_SHIFT - blkbits);
1343 if (page_no) { /* exclude the header page */
1344 if (first_block < lowest_block)
1345 lowest_block = first_block;
1346 if (first_block > highest_block)
1347 highest_block = first_block;
1351 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1353 ret = add_swap_extent(sis, page_no, 1, first_block);
1354 if (ret < 0)
1355 goto out;
1356 nr_extents += ret;
1357 page_no++;
1358 probe_block += blocks_per_page;
1359 reprobe:
1360 continue;
1362 ret = nr_extents;
1363 *span = 1 + highest_block - lowest_block;
1364 if (page_no == 0)
1365 page_no = 1; /* force Empty message */
1366 sis->max = page_no;
1367 sis->pages = page_no - 1;
1368 sis->highest_bit = page_no - 1;
1369 done:
1370 sis->curr_swap_extent = list_entry(sis->extent_list.prev,
1371 struct swap_extent, list);
1372 goto out;
1373 bad_bmap:
1374 printk(KERN_ERR "swapon: swapfile has holes\n");
1375 ret = -EINVAL;
1376 out:
1377 return ret;
1380 asmlinkage long sys_swapoff(const char __user * specialfile)
1382 struct swap_info_struct * p = NULL;
1383 unsigned short *swap_map;
1384 struct file *swap_file, *victim;
1385 struct address_space *mapping;
1386 struct inode *inode;
1387 char * pathname;
1388 int i, type, prev;
1389 int err;
1391 if (!capable(CAP_SYS_ADMIN))
1392 return -EPERM;
1394 pathname = getname(specialfile);
1395 err = PTR_ERR(pathname);
1396 if (IS_ERR(pathname))
1397 goto out;
1399 victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0);
1400 putname(pathname);
1401 err = PTR_ERR(victim);
1402 if (IS_ERR(victim))
1403 goto out;
1405 mapping = victim->f_mapping;
1406 prev = -1;
1407 spin_lock(&swap_lock);
1408 for (type = swap_list.head; type >= 0; type = swap_info[type].next) {
1409 p = swap_info + type;
1410 if (p->flags & SWP_WRITEOK) {
1411 if (p->swap_file->f_mapping == mapping)
1412 break;
1414 prev = type;
1416 if (type < 0) {
1417 err = -EINVAL;
1418 spin_unlock(&swap_lock);
1419 goto out_dput;
1421 if (!security_vm_enough_memory(p->pages))
1422 vm_unacct_memory(p->pages);
1423 else {
1424 err = -ENOMEM;
1425 spin_unlock(&swap_lock);
1426 goto out_dput;
1428 if (prev < 0) {
1429 swap_list.head = p->next;
1430 } else {
1431 swap_info[prev].next = p->next;
1433 if (type == swap_list.next) {
1434 /* just pick something that's safe... */
1435 swap_list.next = swap_list.head;
1437 if (p->prio < 0) {
1438 for (i = p->next; i >= 0; i = swap_info[i].next)
1439 swap_info[i].prio = p->prio--;
1440 least_priority++;
1442 nr_swap_pages -= p->pages;
1443 total_swap_pages -= p->pages;
1444 p->flags &= ~SWP_WRITEOK;
1445 spin_unlock(&swap_lock);
1447 current->flags |= PF_SWAPOFF;
1448 err = try_to_unuse(type);
1449 current->flags &= ~PF_SWAPOFF;
1451 if (err) {
1452 /* re-insert swap space back into swap_list */
1453 spin_lock(&swap_lock);
1454 if (p->prio < 0)
1455 p->prio = --least_priority;
1456 prev = -1;
1457 for (i = swap_list.head; i >= 0; i = swap_info[i].next) {
1458 if (p->prio >= swap_info[i].prio)
1459 break;
1460 prev = i;
1462 p->next = i;
1463 if (prev < 0)
1464 swap_list.head = swap_list.next = p - swap_info;
1465 else
1466 swap_info[prev].next = p - swap_info;
1467 nr_swap_pages += p->pages;
1468 total_swap_pages += p->pages;
1469 p->flags |= SWP_WRITEOK;
1470 spin_unlock(&swap_lock);
1471 goto out_dput;
1474 /* wait for any unplug function to finish */
1475 down_write(&swap_unplug_sem);
1476 up_write(&swap_unplug_sem);
1478 destroy_swap_extents(p);
1479 mutex_lock(&swapon_mutex);
1480 spin_lock(&swap_lock);
1481 drain_mmlist();
1483 /* wait for anyone still in scan_swap_map */
1484 p->highest_bit = 0; /* cuts scans short */
1485 while (p->flags >= SWP_SCANNING) {
1486 spin_unlock(&swap_lock);
1487 schedule_timeout_uninterruptible(1);
1488 spin_lock(&swap_lock);
1491 swap_file = p->swap_file;
1492 p->swap_file = NULL;
1493 p->max = 0;
1494 swap_map = p->swap_map;
1495 p->swap_map = NULL;
1496 p->flags = 0;
1497 spin_unlock(&swap_lock);
1498 mutex_unlock(&swapon_mutex);
1499 vfree(swap_map);
1500 /* Destroy swap account informatin */
1501 swap_cgroup_swapoff(type);
1503 inode = mapping->host;
1504 if (S_ISBLK(inode->i_mode)) {
1505 struct block_device *bdev = I_BDEV(inode);
1506 set_blocksize(bdev, p->old_block_size);
1507 bd_release(bdev);
1508 } else {
1509 mutex_lock(&inode->i_mutex);
1510 inode->i_flags &= ~S_SWAPFILE;
1511 mutex_unlock(&inode->i_mutex);
1513 filp_close(swap_file, NULL);
1514 err = 0;
1516 out_dput:
1517 filp_close(victim, NULL);
1518 out:
1519 return err;
1522 #ifdef CONFIG_PROC_FS
1523 /* iterator */
1524 static void *swap_start(struct seq_file *swap, loff_t *pos)
1526 struct swap_info_struct *ptr = swap_info;
1527 int i;
1528 loff_t l = *pos;
1530 mutex_lock(&swapon_mutex);
1532 if (!l)
1533 return SEQ_START_TOKEN;
1535 for (i = 0; i < nr_swapfiles; i++, ptr++) {
1536 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1537 continue;
1538 if (!--l)
1539 return ptr;
1542 return NULL;
1545 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1547 struct swap_info_struct *ptr;
1548 struct swap_info_struct *endptr = swap_info + nr_swapfiles;
1550 if (v == SEQ_START_TOKEN)
1551 ptr = swap_info;
1552 else {
1553 ptr = v;
1554 ptr++;
1557 for (; ptr < endptr; ptr++) {
1558 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1559 continue;
1560 ++*pos;
1561 return ptr;
1564 return NULL;
1567 static void swap_stop(struct seq_file *swap, void *v)
1569 mutex_unlock(&swapon_mutex);
1572 static int swap_show(struct seq_file *swap, void *v)
1574 struct swap_info_struct *ptr = v;
1575 struct file *file;
1576 int len;
1578 if (ptr == SEQ_START_TOKEN) {
1579 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1580 return 0;
1583 file = ptr->swap_file;
1584 len = seq_path(swap, &file->f_path, " \t\n\\");
1585 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
1586 len < 40 ? 40 - len : 1, " ",
1587 S_ISBLK(file->f_path.dentry->d_inode->i_mode) ?
1588 "partition" : "file\t",
1589 ptr->pages << (PAGE_SHIFT - 10),
1590 ptr->inuse_pages << (PAGE_SHIFT - 10),
1591 ptr->prio);
1592 return 0;
1595 static const struct seq_operations swaps_op = {
1596 .start = swap_start,
1597 .next = swap_next,
1598 .stop = swap_stop,
1599 .show = swap_show
1602 static int swaps_open(struct inode *inode, struct file *file)
1604 return seq_open(file, &swaps_op);
1607 static const struct file_operations proc_swaps_operations = {
1608 .open = swaps_open,
1609 .read = seq_read,
1610 .llseek = seq_lseek,
1611 .release = seq_release,
1614 static int __init procswaps_init(void)
1616 proc_create("swaps", 0, NULL, &proc_swaps_operations);
1617 return 0;
1619 __initcall(procswaps_init);
1620 #endif /* CONFIG_PROC_FS */
1622 #ifdef MAX_SWAPFILES_CHECK
1623 static int __init max_swapfiles_check(void)
1625 MAX_SWAPFILES_CHECK();
1626 return 0;
1628 late_initcall(max_swapfiles_check);
1629 #endif
1632 * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
1634 * The swapon system call
1636 asmlinkage long sys_swapon(const char __user * specialfile, int swap_flags)
1638 struct swap_info_struct * p;
1639 char *name = NULL;
1640 struct block_device *bdev = NULL;
1641 struct file *swap_file = NULL;
1642 struct address_space *mapping;
1643 unsigned int type;
1644 int i, prev;
1645 int error;
1646 union swap_header *swap_header = NULL;
1647 unsigned int nr_good_pages = 0;
1648 int nr_extents = 0;
1649 sector_t span;
1650 unsigned long maxpages = 1;
1651 unsigned long swapfilepages;
1652 unsigned short *swap_map = NULL;
1653 struct page *page = NULL;
1654 struct inode *inode = NULL;
1655 int did_down = 0;
1657 if (!capable(CAP_SYS_ADMIN))
1658 return -EPERM;
1659 spin_lock(&swap_lock);
1660 p = swap_info;
1661 for (type = 0 ; type < nr_swapfiles ; type++,p++)
1662 if (!(p->flags & SWP_USED))
1663 break;
1664 error = -EPERM;
1665 if (type >= MAX_SWAPFILES) {
1666 spin_unlock(&swap_lock);
1667 goto out;
1669 if (type >= nr_swapfiles)
1670 nr_swapfiles = type+1;
1671 memset(p, 0, sizeof(*p));
1672 INIT_LIST_HEAD(&p->extent_list);
1673 p->flags = SWP_USED;
1674 p->next = -1;
1675 spin_unlock(&swap_lock);
1676 name = getname(specialfile);
1677 error = PTR_ERR(name);
1678 if (IS_ERR(name)) {
1679 name = NULL;
1680 goto bad_swap_2;
1682 swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0);
1683 error = PTR_ERR(swap_file);
1684 if (IS_ERR(swap_file)) {
1685 swap_file = NULL;
1686 goto bad_swap_2;
1689 p->swap_file = swap_file;
1690 mapping = swap_file->f_mapping;
1691 inode = mapping->host;
1693 error = -EBUSY;
1694 for (i = 0; i < nr_swapfiles; i++) {
1695 struct swap_info_struct *q = &swap_info[i];
1697 if (i == type || !q->swap_file)
1698 continue;
1699 if (mapping == q->swap_file->f_mapping)
1700 goto bad_swap;
1703 error = -EINVAL;
1704 if (S_ISBLK(inode->i_mode)) {
1705 bdev = I_BDEV(inode);
1706 error = bd_claim(bdev, sys_swapon);
1707 if (error < 0) {
1708 bdev = NULL;
1709 error = -EINVAL;
1710 goto bad_swap;
1712 p->old_block_size = block_size(bdev);
1713 error = set_blocksize(bdev, PAGE_SIZE);
1714 if (error < 0)
1715 goto bad_swap;
1716 p->bdev = bdev;
1717 } else if (S_ISREG(inode->i_mode)) {
1718 p->bdev = inode->i_sb->s_bdev;
1719 mutex_lock(&inode->i_mutex);
1720 did_down = 1;
1721 if (IS_SWAPFILE(inode)) {
1722 error = -EBUSY;
1723 goto bad_swap;
1725 } else {
1726 goto bad_swap;
1729 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
1732 * Read the swap header.
1734 if (!mapping->a_ops->readpage) {
1735 error = -EINVAL;
1736 goto bad_swap;
1738 page = read_mapping_page(mapping, 0, swap_file);
1739 if (IS_ERR(page)) {
1740 error = PTR_ERR(page);
1741 goto bad_swap;
1743 swap_header = kmap(page);
1745 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
1746 printk(KERN_ERR "Unable to find swap-space signature\n");
1747 error = -EINVAL;
1748 goto bad_swap;
1751 /* swap partition endianess hack... */
1752 if (swab32(swap_header->info.version) == 1) {
1753 swab32s(&swap_header->info.version);
1754 swab32s(&swap_header->info.last_page);
1755 swab32s(&swap_header->info.nr_badpages);
1756 for (i = 0; i < swap_header->info.nr_badpages; i++)
1757 swab32s(&swap_header->info.badpages[i]);
1759 /* Check the swap header's sub-version */
1760 if (swap_header->info.version != 1) {
1761 printk(KERN_WARNING
1762 "Unable to handle swap header version %d\n",
1763 swap_header->info.version);
1764 error = -EINVAL;
1765 goto bad_swap;
1768 p->lowest_bit = 1;
1769 p->cluster_next = 1;
1772 * Find out how many pages are allowed for a single swap
1773 * device. There are two limiting factors: 1) the number of
1774 * bits for the swap offset in the swp_entry_t type and
1775 * 2) the number of bits in the a swap pte as defined by
1776 * the different architectures. In order to find the
1777 * largest possible bit mask a swap entry with swap type 0
1778 * and swap offset ~0UL is created, encoded to a swap pte,
1779 * decoded to a swp_entry_t again and finally the swap
1780 * offset is extracted. This will mask all the bits from
1781 * the initial ~0UL mask that can't be encoded in either
1782 * the swp_entry_t or the architecture definition of a
1783 * swap pte.
1785 maxpages = swp_offset(pte_to_swp_entry(
1786 swp_entry_to_pte(swp_entry(0, ~0UL)))) - 1;
1787 if (maxpages > swap_header->info.last_page)
1788 maxpages = swap_header->info.last_page;
1789 p->highest_bit = maxpages - 1;
1791 error = -EINVAL;
1792 if (!maxpages)
1793 goto bad_swap;
1794 if (swapfilepages && maxpages > swapfilepages) {
1795 printk(KERN_WARNING
1796 "Swap area shorter than signature indicates\n");
1797 goto bad_swap;
1799 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
1800 goto bad_swap;
1801 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
1802 goto bad_swap;
1804 /* OK, set up the swap map and apply the bad block list */
1805 swap_map = vmalloc(maxpages * sizeof(short));
1806 if (!swap_map) {
1807 error = -ENOMEM;
1808 goto bad_swap;
1811 memset(swap_map, 0, maxpages * sizeof(short));
1812 for (i = 0; i < swap_header->info.nr_badpages; i++) {
1813 int page_nr = swap_header->info.badpages[i];
1814 if (page_nr <= 0 || page_nr >= swap_header->info.last_page) {
1815 error = -EINVAL;
1816 goto bad_swap;
1818 swap_map[page_nr] = SWAP_MAP_BAD;
1821 error = swap_cgroup_swapon(type, maxpages);
1822 if (error)
1823 goto bad_swap;
1825 nr_good_pages = swap_header->info.last_page -
1826 swap_header->info.nr_badpages -
1827 1 /* header page */;
1829 if (nr_good_pages) {
1830 swap_map[0] = SWAP_MAP_BAD;
1831 p->max = maxpages;
1832 p->pages = nr_good_pages;
1833 nr_extents = setup_swap_extents(p, &span);
1834 if (nr_extents < 0) {
1835 error = nr_extents;
1836 goto bad_swap;
1838 nr_good_pages = p->pages;
1840 if (!nr_good_pages) {
1841 printk(KERN_WARNING "Empty swap-file\n");
1842 error = -EINVAL;
1843 goto bad_swap;
1846 if (blk_queue_nonrot(bdev_get_queue(p->bdev))) {
1847 p->flags |= SWP_SOLIDSTATE;
1848 p->cluster_next = 1 + (random32() % p->highest_bit);
1850 if (discard_swap(p) == 0)
1851 p->flags |= SWP_DISCARDABLE;
1853 mutex_lock(&swapon_mutex);
1854 spin_lock(&swap_lock);
1855 if (swap_flags & SWAP_FLAG_PREFER)
1856 p->prio =
1857 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
1858 else
1859 p->prio = --least_priority;
1860 p->swap_map = swap_map;
1861 p->flags |= SWP_WRITEOK;
1862 nr_swap_pages += nr_good_pages;
1863 total_swap_pages += nr_good_pages;
1865 printk(KERN_INFO "Adding %uk swap on %s. "
1866 "Priority:%d extents:%d across:%lluk %s%s\n",
1867 nr_good_pages<<(PAGE_SHIFT-10), name, p->prio,
1868 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
1869 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
1870 (p->flags & SWP_DISCARDABLE) ? "D" : "");
1872 /* insert swap space into swap_list: */
1873 prev = -1;
1874 for (i = swap_list.head; i >= 0; i = swap_info[i].next) {
1875 if (p->prio >= swap_info[i].prio) {
1876 break;
1878 prev = i;
1880 p->next = i;
1881 if (prev < 0) {
1882 swap_list.head = swap_list.next = p - swap_info;
1883 } else {
1884 swap_info[prev].next = p - swap_info;
1886 spin_unlock(&swap_lock);
1887 mutex_unlock(&swapon_mutex);
1888 error = 0;
1889 goto out;
1890 bad_swap:
1891 if (bdev) {
1892 set_blocksize(bdev, p->old_block_size);
1893 bd_release(bdev);
1895 destroy_swap_extents(p);
1896 swap_cgroup_swapoff(type);
1897 bad_swap_2:
1898 spin_lock(&swap_lock);
1899 p->swap_file = NULL;
1900 p->flags = 0;
1901 spin_unlock(&swap_lock);
1902 vfree(swap_map);
1903 if (swap_file)
1904 filp_close(swap_file, NULL);
1905 out:
1906 if (page && !IS_ERR(page)) {
1907 kunmap(page);
1908 page_cache_release(page);
1910 if (name)
1911 putname(name);
1912 if (did_down) {
1913 if (!error)
1914 inode->i_flags |= S_SWAPFILE;
1915 mutex_unlock(&inode->i_mutex);
1917 return error;
1920 void si_swapinfo(struct sysinfo *val)
1922 unsigned int i;
1923 unsigned long nr_to_be_unused = 0;
1925 spin_lock(&swap_lock);
1926 for (i = 0; i < nr_swapfiles; i++) {
1927 if (!(swap_info[i].flags & SWP_USED) ||
1928 (swap_info[i].flags & SWP_WRITEOK))
1929 continue;
1930 nr_to_be_unused += swap_info[i].inuse_pages;
1932 val->freeswap = nr_swap_pages + nr_to_be_unused;
1933 val->totalswap = total_swap_pages + nr_to_be_unused;
1934 spin_unlock(&swap_lock);
1938 * Verify that a swap entry is valid and increment its swap map count.
1940 * Note: if swap_map[] reaches SWAP_MAP_MAX the entries are treated as
1941 * "permanent", but will be reclaimed by the next swapoff.
1943 int swap_duplicate(swp_entry_t entry)
1945 struct swap_info_struct * p;
1946 unsigned long offset, type;
1947 int result = 0;
1949 if (is_migration_entry(entry))
1950 return 1;
1952 type = swp_type(entry);
1953 if (type >= nr_swapfiles)
1954 goto bad_file;
1955 p = type + swap_info;
1956 offset = swp_offset(entry);
1958 spin_lock(&swap_lock);
1959 if (offset < p->max && p->swap_map[offset]) {
1960 if (p->swap_map[offset] < SWAP_MAP_MAX - 1) {
1961 p->swap_map[offset]++;
1962 result = 1;
1963 } else if (p->swap_map[offset] <= SWAP_MAP_MAX) {
1964 if (swap_overflow++ < 5)
1965 printk(KERN_WARNING "swap_dup: swap entry overflow\n");
1966 p->swap_map[offset] = SWAP_MAP_MAX;
1967 result = 1;
1970 spin_unlock(&swap_lock);
1971 out:
1972 return result;
1974 bad_file:
1975 printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
1976 goto out;
1979 struct swap_info_struct *
1980 get_swap_info_struct(unsigned type)
1982 return &swap_info[type];
1986 * swap_lock prevents swap_map being freed. Don't grab an extra
1987 * reference on the swaphandle, it doesn't matter if it becomes unused.
1989 int valid_swaphandles(swp_entry_t entry, unsigned long *offset)
1991 struct swap_info_struct *si;
1992 int our_page_cluster = page_cluster;
1993 pgoff_t target, toff;
1994 pgoff_t base, end;
1995 int nr_pages = 0;
1997 if (!our_page_cluster) /* no readahead */
1998 return 0;
2000 si = &swap_info[swp_type(entry)];
2001 target = swp_offset(entry);
2002 base = (target >> our_page_cluster) << our_page_cluster;
2003 end = base + (1 << our_page_cluster);
2004 if (!base) /* first page is swap header */
2005 base++;
2007 spin_lock(&swap_lock);
2008 if (end > si->max) /* don't go beyond end of map */
2009 end = si->max;
2011 /* Count contiguous allocated slots above our target */
2012 for (toff = target; ++toff < end; nr_pages++) {
2013 /* Don't read in free or bad pages */
2014 if (!si->swap_map[toff])
2015 break;
2016 if (si->swap_map[toff] == SWAP_MAP_BAD)
2017 break;
2019 /* Count contiguous allocated slots below our target */
2020 for (toff = target; --toff >= base; nr_pages++) {
2021 /* Don't read in free or bad pages */
2022 if (!si->swap_map[toff])
2023 break;
2024 if (si->swap_map[toff] == SWAP_MAP_BAD)
2025 break;
2027 spin_unlock(&swap_lock);
2030 * Indicate starting offset, and return number of pages to get:
2031 * if only 1, say 0, since there's then no readahead to be done.
2033 *offset = ++toff;
2034 return nr_pages? ++nr_pages: 0;