1 /*
2 * linux/mm/swapfile.c
3 *
4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 * Swap reorganised 29.12.95, Stephen Tweedie
6 */
7
8 #include <linux/config.h>
9 #include <linux/mm.h>
10 #include <linux/hugetlb.h>
11 #include <linux/mman.h>
12 #include <linux/slab.h>
13 #include <linux/kernel_stat.h>
14 #include <linux/swap.h>
15 #include <linux/vmalloc.h>
16 #include <linux/pagemap.h>
17 #include <linux/namei.h>
18 #include <linux/shm.h>
19 #include <linux/blkdev.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/acct.h>
28 #include <linux/backing-dev.h>
29 #include <linux/syscalls.h>
30
31 #include <asm/pgtable.h>
32 #include <asm/tlbflush.h>
33 #include <linux/swapops.h>
34
35 DEFINE_SPINLOCK(swaplock);
36 unsigned int nr_swapfiles;
37 long total_swap_pages;
38 static int swap_overflow;
39
40 EXPORT_SYMBOL(total_swap_pages);
41
42 static const char Bad_file[] = "Bad swap file entry ";
43 static const char Unused_file[] = "Unused swap file entry ";
44 static const char Bad_offset[] = "Bad swap offset entry ";
45 static const char Unused_offset[] = "Unused swap offset entry ";
46
47 struct swap_list_t swap_list = {-1, -1};
48
49 struct swap_info_struct swap_info[MAX_SWAPFILES];
50
51 static DECLARE_MUTEX(swapon_sem);
52
53 /*
54 * We need this because the bdev->unplug_fn can sleep and we cannot
55 * hold swap_list_lock while calling the unplug_fn. And swap_list_lock
56 * cannot be turned into a semaphore.
57 */
58 static DECLARE_RWSEM(swap_unplug_sem);
59
60 #define SWAPFILE_CLUSTER 256
61
62 void swap_unplug_io_fn(struct backing_dev_info *unused_bdi, struct page *page)
63 {
64 swp_entry_t entry;
65
66 down_read(&swap_unplug_sem);
67 entry.val = page->private;
68 if (PageSwapCache(page)) {
69 struct block_device *bdev = swap_info[swp_type(entry)].bdev;
70 struct backing_dev_info *bdi;
71
72 /*
73 * If the page is removed from swapcache from under us (with a
74 * racy try_to_unuse/swapoff) we need an additional reference
75 * count to avoid reading garbage from page->private above. If
76 * the WARN_ON triggers during a swapoff it maybe the race
77 * condition and it's harmless. However if it triggers without
78 * swapoff it signals a problem.
79 */
80 WARN_ON(page_count(page) <= 1);
81
82 bdi = bdev->bd_inode->i_mapping->backing_dev_info;
83 bdi->unplug_io_fn(bdi, page);
84 }
85 up_read(&swap_unplug_sem);
86 }
87
88 static inline int scan_swap_map(struct swap_info_struct *si)
89 {
90 unsigned long offset;
91 /*
92 * We try to cluster swap pages by allocating them
93 * sequentially in swap. Once we've allocated
94 * SWAPFILE_CLUSTER pages this way, however, we resort to
95 * first-free allocation, starting a new cluster. This
96 * prevents us from scattering swap pages all over the entire
97 * swap partition, so that we reduce overall disk seek times
98 * between swap pages. -- sct */
99 if (si->cluster_nr) {
100 while (si->cluster_next <= si->highest_bit) {
101 offset = si->cluster_next++;
102 if (si->swap_map[offset])
103 continue;
104 si->cluster_nr--;
105 goto got_page;
106 }
107 }
108 si->cluster_nr = SWAPFILE_CLUSTER;
109
110 /* try to find an empty (even not aligned) cluster. */
111 offset = si->lowest_bit;
112 check_next_cluster:
113 if (offset+SWAPFILE_CLUSTER-1 <= si->highest_bit)
114 {
115 unsigned long nr;
116 for (nr = offset; nr < offset+SWAPFILE_CLUSTER; nr++)
117 if (si->swap_map[nr])
118 {
119 offset = nr+1;
120 goto check_next_cluster;
121 }
122 /* We found a completly empty cluster, so start
123 * using it.
124 */
125 goto got_page;
126 }
127 /* No luck, so now go finegrined as usual. -Andrea */
128 for (offset = si->lowest_bit; offset <= si->highest_bit ; offset++) {
129 if (si->swap_map[offset])
130 continue;
131 si->lowest_bit = offset+1;
132 got_page:
133 if (offset == si->lowest_bit)
134 si->lowest_bit++;
135 if (offset == si->highest_bit)
136 si->highest_bit--;
137 if (si->lowest_bit > si->highest_bit) {
138 si->lowest_bit = si->max;
139 si->highest_bit = 0;
140 }
141 si->swap_map[offset] = 1;
142 si->inuse_pages++;
143 nr_swap_pages--;
144 si->cluster_next = offset+1;
145 return offset;
146 }
147 si->lowest_bit = si->max;
148 si->highest_bit = 0;
149 return 0;
150 }
151
152 swp_entry_t get_swap_page(void)
153 {
154 struct swap_info_struct * p;
155 unsigned long offset;
156 swp_entry_t entry;
157 int type, wrapped = 0;
158
159 entry.val = 0; /* Out of memory */
160 swap_list_lock();
161 type = swap_list.next;
162 if (type < 0)
163 goto out;
164 if (nr_swap_pages <= 0)
165 goto out;
166
167 while (1) {
168 p = &swap_info[type];
169 if ((p->flags & SWP_ACTIVE) == SWP_ACTIVE) {
170 swap_device_lock(p);
171 offset = scan_swap_map(p);
172 swap_device_unlock(p);
173 if (offset) {
174 entry = swp_entry(type,offset);
175 type = swap_info[type].next;
176 if (type < 0 ||
177 p->prio != swap_info[type].prio) {
178 swap_list.next = swap_list.head;
179 } else {
180 swap_list.next = type;
181 }
182 goto out;
183 }
184 }
185 type = p->next;
186 if (!wrapped) {
187 if (type < 0 || p->prio != swap_info[type].prio) {
188 type = swap_list.head;
189 wrapped = 1;
190 }
191 } else
192 if (type < 0)
193 goto out; /* out of swap space */
194 }
195 out:
196 swap_list_unlock();
197 return entry;
198 }
199
200 static struct swap_info_struct * swap_info_get(swp_entry_t entry)
201 {
202 struct swap_info_struct * p;
203 unsigned long offset, type;
204
205 if (!entry.val)
206 goto out;
207 type = swp_type(entry);
208 if (type >= nr_swapfiles)
209 goto bad_nofile;
210 p = & swap_info[type];
211 if (!(p->flags & SWP_USED))
212 goto bad_device;
213 offset = swp_offset(entry);
214 if (offset >= p->max)
215 goto bad_offset;
216 if (!p->swap_map[offset])
217 goto bad_free;
218 swap_list_lock();
219 if (p->prio > swap_info[swap_list.next].prio)
220 swap_list.next = type;
221 swap_device_lock(p);
222 return p;
223
224 bad_free:
225 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val);
226 goto out;
227 bad_offset:
228 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val);
229 goto out;
230 bad_device:
231 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val);
232 goto out;
233 bad_nofile:
234 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val);
235 out:
236 return NULL;
237 }
238
239 static void swap_info_put(struct swap_info_struct * p)
240 {
241 swap_device_unlock(p);
242 swap_list_unlock();
243 }
244
245 static int swap_entry_free(struct swap_info_struct *p, unsigned long offset)
246 {
247 int count = p->swap_map[offset];
248
249 if (count < SWAP_MAP_MAX) {
250 count--;
251 p->swap_map[offset] = count;
252 if (!count) {
253 if (offset < p->lowest_bit)
254 p->lowest_bit = offset;
255 if (offset > p->highest_bit)
256 p->highest_bit = offset;
257 nr_swap_pages++;
258 p->inuse_pages--;
259 }
260 }
261 return count;
262 }
263
264 /*
265 * Caller has made sure that the swapdevice corresponding to entry
266 * is still around or has not been recycled.
267 */
268 void swap_free(swp_entry_t entry)
269 {
270 struct swap_info_struct * p;
271
272 p = swap_info_get(entry);
273 if (p) {
274 swap_entry_free(p, swp_offset(entry));
275 swap_info_put(p);
276 }
277 }
278
279 /*
280 * Check if we're the only user of a swap page,
281 * when the page is locked.
282 */
283 static int exclusive_swap_page(struct page *page)
284 {
285 int retval = 0;
286 struct swap_info_struct * p;
287 swp_entry_t entry;
288
289 entry.val = page->private;
290 p = swap_info_get(entry);
291 if (p) {
292 /* Is the only swap cache user the cache itself? */
293 if (p->swap_map[swp_offset(entry)] == 1) {
294 /* Recheck the page count with the swapcache lock held.. */
295 spin_lock_irq(&swapper_space.tree_lock);
296 if (page_count(page) == 2)
297 retval = 1;
298 spin_unlock_irq(&swapper_space.tree_lock);
299 }
300 swap_info_put(p);
301 }
302 return retval;
303 }
304
305 /*
306 * We can use this swap cache entry directly
307 * if there are no other references to it.
308 *
309 * Here "exclusive_swap_page()" does the real
310 * work, but we opportunistically check whether
311 * we need to get all the locks first..
312 */
313 int can_share_swap_page(struct page *page)
314 {
315 int retval = 0;
316
317 if (!PageLocked(page))
318 BUG();
319 switch (page_count(page)) {
320 case 3:
321 if (!PagePrivate(page))
322 break;
323 /* Fallthrough */
324 case 2:
325 if (!PageSwapCache(page))
326 break;
327 retval = exclusive_swap_page(page);
328 break;
329 case 1:
330 if (PageReserved(page))
331 break;
332 retval = 1;
333 }
334 return retval;
335 }
336
337 /*
338 * Work out if there are any other processes sharing this
339 * swap cache page. Free it if you can. Return success.
340 */
341 int remove_exclusive_swap_page(struct page *page)
342 {
343 int retval;
344 struct swap_info_struct * p;
345 swp_entry_t entry;
346
347 BUG_ON(PagePrivate(page));
348 BUG_ON(!PageLocked(page));
349
350 if (!PageSwapCache(page))
351 return 0;
352 if (PageWriteback(page))
353 return 0;
354 if (page_count(page) != 2) /* 2: us + cache */
355 return 0;
356
357 entry.val = page->private;
358 p = swap_info_get(entry);
359 if (!p)
360 return 0;
361
362 /* Is the only swap cache user the cache itself? */
363 retval = 0;
364 if (p->swap_map[swp_offset(entry)] == 1) {
365 /* Recheck the page count with the swapcache lock held.. */
366 spin_lock_irq(&swapper_space.tree_lock);
367 if ((page_count(page) == 2) && !PageWriteback(page)) {
368 __delete_from_swap_cache(page);
369 SetPageDirty(page);
370 retval = 1;
371 }
372 spin_unlock_irq(&swapper_space.tree_lock);
373 }
374 swap_info_put(p);
375
376 if (retval) {
377 swap_free(entry);
378 page_cache_release(page);
379 }
380
381 return retval;
382 }
383
384 /*
385 * Free the swap entry like above, but also try to
386 * free the page cache entry if it is the last user.
387 */
388 void free_swap_and_cache(swp_entry_t entry)
389 {
390 struct swap_info_struct * p;
391 struct page *page = NULL;
392
393 p = swap_info_get(entry);
394 if (p) {
395 if (swap_entry_free(p, swp_offset(entry)) == 1) {
396 spin_lock_irq(&swapper_space.tree_lock);
397 page = radix_tree_lookup(&swapper_space.page_tree,
398 entry.val);
399 if (page && TestSetPageLocked(page))
400 page = NULL;
401 spin_unlock_irq(&swapper_space.tree_lock);
402 }
403 swap_info_put(p);
404 }
405 if (page) {
406 int one_user;
407
408 BUG_ON(PagePrivate(page));
409 page_cache_get(page);
410 one_user = (page_count(page) == 2);
411 /* Only cache user (+us), or swap space full? Free it! */
412 if (!PageWriteback(page) && (one_user || vm_swap_full())) {
413 delete_from_swap_cache(page);
414 SetPageDirty(page);
415 }
416 unlock_page(page);
417 page_cache_release(page);
418 }
419 }
420
421 /*
422 * The swap entry has been read in advance, and we return 1 to indicate
423 * that the page has been used or is no longer needed.
424 *
425 * Always set the resulting pte to be nowrite (the same as COW pages
426 * after one process has exited). We don't know just how many PTEs will
427 * share this swap entry, so be cautious and let do_wp_page work out
428 * what to do if a write is requested later.
429 */
430 /* vma->vm_mm->page_table_lock is held */
431 static void
432 unuse_pte(struct vm_area_struct *vma, unsigned long address, pte_t *dir,
433 swp_entry_t entry, struct page *page)
434 {
435 vma->vm_mm->rss++;
436 get_page(page);
437 set_pte(dir, pte_mkold(mk_pte(page, vma->vm_page_prot)));
438 page_add_anon_rmap(page, vma, address);
439 swap_free(entry);
440 acct_update_integrals();
441 update_mem_hiwater();
442 }
443
444 /* vma->vm_mm->page_table_lock is held */
445 static unsigned long unuse_pmd(struct vm_area_struct *vma, pmd_t *dir,
446 unsigned long address, unsigned long end,
447 swp_entry_t entry, struct page *page)
448 {
449 pte_t *pte;
450 pte_t swp_pte = swp_entry_to_pte(entry);
451
452 if (pmd_none(*dir))
453 return 0;
454 if (pmd_bad(*dir)) {
455 pmd_ERROR(*dir);
456 pmd_clear(dir);
457 return 0;
458 }
459 pte = pte_offset_map(dir, address);
460 do {
461 /*
462 * swapoff spends a _lot_ of time in this loop!
463 * Test inline before going to call unuse_pte.
464 */
465 if (unlikely(pte_same(*pte, swp_pte))) {
466 unuse_pte(vma, address, pte, entry, page);
467 pte_unmap(pte);
468
469 /*
470 * Move the page to the active list so it is not
471 * immediately swapped out again after swapon.
472 */
473 activate_page(page);
474
475 /* add 1 since address may be 0 */
476 return 1 + address;
477 }
478 address += PAGE_SIZE;
479 pte++;
480 } while (address < end);
481 pte_unmap(pte - 1);
482 return 0;
483 }
484
485 /* vma->vm_mm->page_table_lock is held */
486 static unsigned long unuse_pud(struct vm_area_struct *vma, pud_t *pud,
487 unsigned long address, unsigned long end,
488 swp_entry_t entry, struct page *page)
489 {
490 pmd_t *pmd;
491 unsigned long next;
492 unsigned long foundaddr;
493
494 if (pud_none(*pud))
495 return 0;
496 if (pud_bad(*pud)) {
497 pud_ERROR(*pud);
498 pud_clear(pud);
499 return 0;
500 }
501 pmd = pmd_offset(pud, address);
502 do {
503 next = (address + PMD_SIZE) & PMD_MASK;
504 if (next > end || !next)
505 next = end;
506 foundaddr = unuse_pmd(vma, pmd, address, next, entry, page);
507 if (foundaddr)
508 return foundaddr;
509 address = next;
510 pmd++;
511 } while (address < end);
512 return 0;
513 }
514
515 /* vma->vm_mm->page_table_lock is held */
516 static unsigned long unuse_pgd(struct vm_area_struct *vma, pgd_t *pgd,
517 unsigned long address, unsigned long end,
518 swp_entry_t entry, struct page *page)
519 {
520 pud_t *pud;
521 unsigned long next;
522 unsigned long foundaddr;
523
524 if (pgd_none(*pgd))
525 return 0;
526 if (pgd_bad(*pgd)) {
527 pgd_ERROR(*pgd);
528 pgd_clear(pgd);
529 return 0;
530 }
531 pud = pud_offset(pgd, address);
532 do {
533 next = (address + PUD_SIZE) & PUD_MASK;
534 if (next > end || !next)
535 next = end;
536 foundaddr = unuse_pud(vma, pud, address, next, entry, page);
537 if (foundaddr)
538 return foundaddr;
539 address = next;
540 pud++;
541 } while (address < end);
542 return 0;
543 }
544
545 /* vma->vm_mm->page_table_lock is held */
546 static unsigned long unuse_vma(struct vm_area_struct *vma,
547 swp_entry_t entry, struct page *page)
548 {
549 pgd_t *pgd;
550 unsigned long address, next, end;
551 unsigned long foundaddr;
552
553 if (page->mapping) {
554 address = page_address_in_vma(page, vma);
555 if (address == -EFAULT)
556 return 0;
557 else
558 end = address + PAGE_SIZE;
559 } else {
560 address = vma->vm_start;
561 end = vma->vm_end;
562 }
563 pgd = pgd_offset(vma->vm_mm, address);
564 do {
565 next = (address + PGDIR_SIZE) & PGDIR_MASK;
566 if (next > end || !next)
567 next = end;
568 foundaddr = unuse_pgd(vma, pgd, address, next, entry, page);
569 if (foundaddr)
570 return foundaddr;
571 address = next;
572 pgd++;
573 } while (address < end);
574 return 0;
575 }
576
577 static int unuse_process(struct mm_struct * mm,
578 swp_entry_t entry, struct page* page)
579 {
580 struct vm_area_struct* vma;
581 unsigned long foundaddr = 0;
582
583 /*
584 * Go through process' page directory.
585 */
586 if (!down_read_trylock(&mm->mmap_sem)) {
587 /*
588 * Our reference to the page stops try_to_unmap_one from
589 * unmapping its ptes, so swapoff can make progress.
590 */
591 unlock_page(page);
592 down_read(&mm->mmap_sem);
593 lock_page(page);
594 }
595 spin_lock(&mm->page_table_lock);
596 for (vma = mm->mmap; vma; vma = vma->vm_next) {
597 if (vma->anon_vma) {
598 foundaddr = unuse_vma(vma, entry, page);
599 if (foundaddr)
600 break;
601 }
602 }
603 spin_unlock(&mm->page_table_lock);
604 up_read(&mm->mmap_sem);
605 /*
606 * Currently unuse_process cannot fail, but leave error handling
607 * at call sites for now, since we change it from time to time.
608 */
609 return 0;
610 }
611
612 /*
613 * Scan swap_map from current position to next entry still in use.
614 * Recycle to start on reaching the end, returning 0 when empty.
615 */
616 static int find_next_to_unuse(struct swap_info_struct *si, int prev)
617 {
618 int max = si->max;
619 int i = prev;
620 int count;
621
622 /*
623 * No need for swap_device_lock(si) here: we're just looking
624 * for whether an entry is in use, not modifying it; false
625 * hits are okay, and sys_swapoff() has already prevented new
626 * allocations from this area (while holding swap_list_lock()).
627 */
628 for (;;) {
629 if (++i >= max) {
630 if (!prev) {
631 i = 0;
632 break;
633 }
634 /*
635 * No entries in use at top of swap_map,
636 * loop back to start and recheck there.
637 */
638 max = prev + 1;
639 prev = 0;
640 i = 1;
641 }
642 count = si->swap_map[i];
643 if (count && count != SWAP_MAP_BAD)
644 break;
645 }
646 return i;
647 }
648
649 /*
650 * We completely avoid races by reading each swap page in advance,
651 * and then search for the process using it. All the necessary
652 * page table adjustments can then be made atomically.
653 */
654 static int try_to_unuse(unsigned int type)
655 {
656 struct swap_info_struct * si = &swap_info[type];
657 struct mm_struct *start_mm;
658 unsigned short *swap_map;
659 unsigned short swcount;
660 struct page *page;
661 swp_entry_t entry;
662 int i = 0;
663 int retval = 0;
664 int reset_overflow = 0;
665 int shmem;
666
667 /*
668 * When searching mms for an entry, a good strategy is to
669 * start at the first mm we freed the previous entry from
670 * (though actually we don't notice whether we or coincidence
671 * freed the entry). Initialize this start_mm with a hold.
672 *
673 * A simpler strategy would be to start at the last mm we
674 * freed the previous entry from; but that would take less
675 * advantage of mmlist ordering, which clusters forked mms
676 * together, child after parent. If we race with dup_mmap(), we
677 * prefer to resolve parent before child, lest we miss entries
678 * duplicated after we scanned child: using last mm would invert
679 * that. Though it's only a serious concern when an overflowed
680 * swap count is reset from SWAP_MAP_MAX, preventing a rescan.
681 */
682 start_mm = &init_mm;
683 atomic_inc(&init_mm.mm_users);
684
685 /*
686 * Keep on scanning until all entries have gone. Usually,
687 * one pass through swap_map is enough, but not necessarily:
688 * there are races when an instance of an entry might be missed.
689 */
690 while ((i = find_next_to_unuse(si, i)) != 0) {
691 if (signal_pending(current)) {
692 retval = -EINTR;
693 break;
694 }
695
696 /*
697 * Get a page for the entry, using the existing swap
698 * cache page if there is one. Otherwise, get a clean
699 * page and read the swap into it.
700 */
701 swap_map = &si->swap_map[i];
702 entry = swp_entry(type, i);
703 page = read_swap_cache_async(entry, NULL, 0);
704 if (!page) {
705 /*
706 * Either swap_duplicate() failed because entry
707 * has been freed independently, and will not be
708 * reused since sys_swapoff() already disabled
709 * allocation from here, or alloc_page() failed.
710 */
711 if (!*swap_map)
712 continue;
713 retval = -ENOMEM;
714 break;
715 }
716
717 /*
718 * Don't hold on to start_mm if it looks like exiting.
719 */
720 if (atomic_read(&start_mm->mm_users) == 1) {
721 mmput(start_mm);
722 start_mm = &init_mm;
723 atomic_inc(&init_mm.mm_users);
724 }
725
726 /*
727 * Wait for and lock page. When do_swap_page races with
728 * try_to_unuse, do_swap_page can handle the fault much
729 * faster than try_to_unuse can locate the entry. This
730 * apparently redundant "wait_on_page_locked" lets try_to_unuse
731 * defer to do_swap_page in such a case - in some tests,
732 * do_swap_page and try_to_unuse repeatedly compete.
733 */
734 wait_on_page_locked(page);
735 wait_on_page_writeback(page);
736 lock_page(page);
737 wait_on_page_writeback(page);
738
739 /*
740 * Remove all references to entry.
741 * Whenever we reach init_mm, there's no address space
742 * to search, but use it as a reminder to search shmem.
743 */
744 shmem = 0;
745 swcount = *swap_map;
746 if (swcount > 1) {
747 if (start_mm == &init_mm)
748 shmem = shmem_unuse(entry, page);
749 else
750 retval = unuse_process(start_mm, entry, page);
751 }
752 if (*swap_map > 1) {
753 int set_start_mm = (*swap_map >= swcount);
754 struct list_head *p = &start_mm->mmlist;
755 struct mm_struct *new_start_mm = start_mm;
756 struct mm_struct *prev_mm = start_mm;
757 struct mm_struct *mm;
758
759 atomic_inc(&new_start_mm->mm_users);
760 atomic_inc(&prev_mm->mm_users);
761 spin_lock(&mmlist_lock);
762 while (*swap_map > 1 && !retval &&
763 (p = p->next) != &start_mm->mmlist) {
764 mm = list_entry(p, struct mm_struct, mmlist);
765 if (atomic_inc_return(&mm->mm_users) == 1) {
766 atomic_dec(&mm->mm_users);
767 continue;
768 }
769 spin_unlock(&mmlist_lock);
770 mmput(prev_mm);
771 prev_mm = mm;
772
773 cond_resched();
774
775 swcount = *swap_map;
776 if (swcount <= 1)
777 ;
778 else if (mm == &init_mm) {
779 set_start_mm = 1;
780 shmem = shmem_unuse(entry, page);
781 } else
782 retval = unuse_process(mm, entry, page);
783 if (set_start_mm && *swap_map < swcount) {
784 mmput(new_start_mm);
785 atomic_inc(&mm->mm_users);
786 new_start_mm = mm;
787 set_start_mm = 0;
788 }
789 spin_lock(&mmlist_lock);
790 }
791 spin_unlock(&mmlist_lock);
792 mmput(prev_mm);
793 mmput(start_mm);
794 start_mm = new_start_mm;
795 }
796 if (retval) {
797 unlock_page(page);
798 page_cache_release(page);
799 break;
800 }
801
802 /*
803 * How could swap count reach 0x7fff when the maximum
804 * pid is 0x7fff, and there's no way to repeat a swap
805 * page within an mm (except in shmem, where it's the
806 * shared object which takes the reference count)?
807 * We believe SWAP_MAP_MAX cannot occur in Linux 2.4.
808 *
809 * If that's wrong, then we should worry more about
810 * exit_mmap() and do_munmap() cases described above:
811 * we might be resetting SWAP_MAP_MAX too early here.
812 * We know "Undead"s can happen, they're okay, so don't
813 * report them; but do report if we reset SWAP_MAP_MAX.
814 */
815 if (*swap_map == SWAP_MAP_MAX) {
816 swap_device_lock(si);
817 *swap_map = 1;
818 swap_device_unlock(si);
819 reset_overflow = 1;
820 }
821
822 /*
823 * If a reference remains (rare), we would like to leave
824 * the page in the swap cache; but try_to_unmap could
825 * then re-duplicate the entry once we drop page lock,
826 * so we might loop indefinitely; also, that page could
827 * not be swapped out to other storage meanwhile. So:
828 * delete from cache even if there's another reference,
829 * after ensuring that the data has been saved to disk -
830 * since if the reference remains (rarer), it will be
831 * read from disk into another page. Splitting into two
832 * pages would be incorrect if swap supported "shared
833 * private" pages, but they are handled by tmpfs files.
834 *
835 * Note shmem_unuse already deleted a swappage from
836 * the swap cache, unless the move to filepage failed:
837 * in which case it left swappage in cache, lowered its
838 * swap count to pass quickly through the loops above,
839 * and now we must reincrement count to try again later.
840 */
841 if ((*swap_map > 1) && PageDirty(page) && PageSwapCache(page)) {
842 struct writeback_control wbc = {
843 .sync_mode = WB_SYNC_NONE,
844 };
845
846 swap_writepage(page, &wbc);
847 lock_page(page);
848 wait_on_page_writeback(page);
849 }
850 if (PageSwapCache(page)) {
851 if (shmem)
852 swap_duplicate(entry);
853 else
854 delete_from_swap_cache(page);
855 }
856
857 /*
858 * So we could skip searching mms once swap count went
859 * to 1, we did not mark any present ptes as dirty: must
860 * mark page dirty so shrink_list will preserve it.
861 */
862 SetPageDirty(page);
863 unlock_page(page);
864 page_cache_release(page);
865
866 /*
867 * Make sure that we aren't completely killing
868 * interactive performance.
869 */
870 cond_resched();
871 }
872
873 mmput(start_mm);
874 if (reset_overflow) {
875 printk(KERN_WARNING "swapoff: cleared swap entry overflow\n");
876 swap_overflow = 0;
877 }
878 return retval;
879 }
880
881 /*
882 * After a successful try_to_unuse, if no swap is now in use, we know we
883 * can empty the mmlist. swap_list_lock must be held on entry and exit.
884 * Note that mmlist_lock nests inside swap_list_lock, and an mm must be
885 * added to the mmlist just after page_duplicate - before would be racy.
886 */
887 static void drain_mmlist(void)
888 {
889 struct list_head *p, *next;
890 unsigned int i;
891
892 for (i = 0; i < nr_swapfiles; i++)
893 if (swap_info[i].inuse_pages)
894 return;
895 spin_lock(&mmlist_lock);
896 list_for_each_safe(p, next, &init_mm.mmlist)
897 list_del_init(p);
898 spin_unlock(&mmlist_lock);
899 }
900
901 /*
902 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
903 * corresponds to page offset `offset'.
904 */
905 sector_t map_swap_page(struct swap_info_struct *sis, pgoff_t offset)
906 {
907 struct swap_extent *se = sis->curr_swap_extent;
908 struct swap_extent *start_se = se;
909
910 for ( ; ; ) {
911 struct list_head *lh;
912
913 if (se->start_page <= offset &&
914 offset < (se->start_page + se->nr_pages)) {
915 return se->start_block + (offset - se->start_page);
916 }
917 lh = se->list.prev;
918 if (lh == &sis->extent_list)
919 lh = lh->prev;
920 se = list_entry(lh, struct swap_extent, list);
921 sis->curr_swap_extent = se;
922 BUG_ON(se == start_se); /* It *must* be present */
923 }
924 }
925
926 /*
927 * Free all of a swapdev's extent information
928 */
929 static void destroy_swap_extents(struct swap_info_struct *sis)
930 {
931 while (!list_empty(&sis->extent_list)) {
932 struct swap_extent *se;
933
934 se = list_entry(sis->extent_list.next,
935 struct swap_extent, list);
936 list_del(&se->list);
937 kfree(se);
938 }
939 sis->nr_extents = 0;
940 }
941
942 /*
943 * Add a block range (and the corresponding page range) into this swapdev's
944 * extent list. The extent list is kept sorted in block order.
945 *
946 * This function rather assumes that it is called in ascending sector_t order.
947 * It doesn't look for extent coalescing opportunities.
948 */
949 static int
950 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
951 unsigned long nr_pages, sector_t start_block)
952 {
953 struct swap_extent *se;
954 struct swap_extent *new_se;
955 struct list_head *lh;
956
957 lh = sis->extent_list.next; /* The highest-addressed block */
958 while (lh != &sis->extent_list) {
959 se = list_entry(lh, struct swap_extent, list);
960 if (se->start_block + se->nr_pages == start_block &&
961 se->start_page + se->nr_pages == start_page) {
962 /* Merge it */
963 se->nr_pages += nr_pages;
964 return 0;
965 }
966 lh = lh->next;
967 }
968
969 /*
970 * No merge. Insert a new extent, preserving ordering.
971 */
972 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
973 if (new_se == NULL)
974 return -ENOMEM;
975 new_se->start_page = start_page;
976 new_se->nr_pages = nr_pages;
977 new_se->start_block = start_block;
978
979 lh = sis->extent_list.prev; /* The lowest block */
980 while (lh != &sis->extent_list) {
981 se = list_entry(lh, struct swap_extent, list);
982 if (se->start_block > start_block)
983 break;
984 lh = lh->prev;
985 }
986 list_add_tail(&new_se->list, lh);
987 sis->nr_extents++;
988 return 0;
989 }
990
991 /*
992 * A `swap extent' is a simple thing which maps a contiguous range of pages
993 * onto a contiguous range of disk blocks. An ordered list of swap extents
994 * is built at swapon time and is then used at swap_writepage/swap_readpage
995 * time for locating where on disk a page belongs.
996 *
997 * If the swapfile is an S_ISBLK block device, a single extent is installed.
998 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
999 * swap files identically.
1000 *
1001 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1002 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1003 * swapfiles are handled *identically* after swapon time.
1004 *
1005 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1006 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1007 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1008 * requirements, they are simply tossed out - we will never use those blocks
1009 * for swapping.
1010 *
1011 * For S_ISREG swapfiles we hold i_sem across the life of the swapon. This
1012 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1013 * which will scribble on the fs.
1014 *
1015 * The amount of disk space which a single swap extent represents varies.
1016 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1017 * extents in the list. To avoid much list walking, we cache the previous
1018 * search location in `curr_swap_extent', and start new searches from there.
1019 * This is extremely effective. The average number of iterations in
1020 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1021 */
1022 static int setup_swap_extents(struct swap_info_struct *sis)
1023 {
1024 struct inode *inode;
1025 unsigned blocks_per_page;
1026 unsigned long page_no;
1027 unsigned blkbits;
1028 sector_t probe_block;
1029 sector_t last_block;
1030 int ret;
1031
1032 inode = sis->swap_file->f_mapping->host;
1033 if (S_ISBLK(inode->i_mode)) {
1034 ret = add_swap_extent(sis, 0, sis->max, 0);
1035 goto done;
1036 }
1037
1038 blkbits = inode->i_blkbits;
1039 blocks_per_page = PAGE_SIZE >> blkbits;
1040
1041 /*
1042 * Map all the blocks into the extent list. This code doesn't try
1043 * to be very smart.
1044 */
1045 probe_block = 0;
1046 page_no = 0;
1047 last_block = i_size_read(inode) >> blkbits;
1048 while ((probe_block + blocks_per_page) <= last_block &&
1049 page_no < sis->max) {
1050 unsigned block_in_page;
1051 sector_t first_block;
1052
1053 first_block = bmap(inode, probe_block);
1054 if (first_block == 0)
1055 goto bad_bmap;
1056
1057 /*
1058 * It must be PAGE_SIZE aligned on-disk
1059 */
1060 if (first_block & (blocks_per_page - 1)) {
1061 probe_block++;
1062 goto reprobe;
1063 }
1064
1065 for (block_in_page = 1; block_in_page < blocks_per_page;
1066 block_in_page++) {
1067 sector_t block;
1068
1069 block = bmap(inode, probe_block + block_in_page);
1070 if (block == 0)
1071 goto bad_bmap;
1072 if (block != first_block + block_in_page) {
1073 /* Discontiguity */
1074 probe_block++;
1075 goto reprobe;
1076 }
1077 }
1078
1079 /*
1080 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1081 */
1082 ret = add_swap_extent(sis, page_no, 1,
1083 first_block >> (PAGE_SHIFT - blkbits));
1084 if (ret)
1085 goto out;
1086 page_no++;
1087 probe_block += blocks_per_page;
1088 reprobe:
1089 continue;
1090 }
1091 ret = 0;
1092 if (page_no == 0)
1093 ret = -EINVAL;
1094 sis->max = page_no;
1095 sis->highest_bit = page_no - 1;
1096 done:
1097 sis->curr_swap_extent = list_entry(sis->extent_list.prev,
1098 struct swap_extent, list);
1099 goto out;
1100 bad_bmap:
1101 printk(KERN_ERR "swapon: swapfile has holes\n");
1102 ret = -EINVAL;
1103 out:
1104 return ret;
1105 }
1106
1107 #if 0 /* We don't need this yet */
1108 #include <linux/backing-dev.h>
1109 int page_queue_congested(struct page *page)
1110 {
1111 struct backing_dev_info *bdi;
1112
1113 BUG_ON(!PageLocked(page)); /* It pins the swap_info_struct */
1114
1115 if (PageSwapCache(page)) {
1116 swp_entry_t entry = { .val = page->private };
1117 struct swap_info_struct *sis;
1118
1119 sis = get_swap_info_struct(swp_type(entry));
1120 bdi = sis->bdev->bd_inode->i_mapping->backing_dev_info;
1121 } else
1122 bdi = page->mapping->backing_dev_info;
1123 return bdi_write_congested(bdi);
1124 }
1125 #endif
1126
1127 asmlinkage long sys_swapoff(const char __user * specialfile)
1128 {
1129 struct swap_info_struct * p = NULL;
1130 unsigned short *swap_map;
1131 struct file *swap_file, *victim;
1132 struct address_space *mapping;
1133 struct inode *inode;
1134 char * pathname;
1135 int i, type, prev;
1136 int err;
1137
1138 if (!capable(CAP_SYS_ADMIN))
1139 return -EPERM;
1140
1141 pathname = getname(specialfile);
1142 err = PTR_ERR(pathname);
1143 if (IS_ERR(pathname))
1144 goto out;
1145
1146 victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0);
1147 putname(pathname);
1148 err = PTR_ERR(victim);
1149 if (IS_ERR(victim))
1150 goto out;
1151
1152 mapping = victim->f_mapping;
1153 prev = -1;
1154 swap_list_lock();
1155 for (type = swap_list.head; type >= 0; type = swap_info[type].next) {
1156 p = swap_info + type;
1157 if ((p->flags & SWP_ACTIVE) == SWP_ACTIVE) {
1158 if (p->swap_file->f_mapping == mapping)
1159 break;
1160 }
1161 prev = type;
1162 }
1163 if (type < 0) {
1164 err = -EINVAL;
1165 swap_list_unlock();
1166 goto out_dput;
1167 }
1168 if (!security_vm_enough_memory(p->pages))
1169 vm_unacct_memory(p->pages);
1170 else {
1171 err = -ENOMEM;
1172 swap_list_unlock();
1173 goto out_dput;
1174 }
1175 if (prev < 0) {
1176 swap_list.head = p->next;
1177 } else {
1178 swap_info[prev].next = p->next;
1179 }
1180 if (type == swap_list.next) {
1181 /* just pick something that's safe... */
1182 swap_list.next = swap_list.head;
1183 }
1184 nr_swap_pages -= p->pages;
1185 total_swap_pages -= p->pages;
1186 p->flags &= ~SWP_WRITEOK;
1187 swap_list_unlock();
1188 current->flags |= PF_SWAPOFF;
1189 err = try_to_unuse(type);
1190 current->flags &= ~PF_SWAPOFF;
1191
1192 /* wait for any unplug function to finish */
1193 down_write(&swap_unplug_sem);
1194 up_write(&swap_unplug_sem);
1195
1196 if (err) {
1197 /* re-insert swap space back into swap_list */
1198 swap_list_lock();
1199 for (prev = -1, i = swap_list.head; i >= 0; prev = i, i = swap_info[i].next)
1200 if (p->prio >= swap_info[i].prio)
1201 break;
1202 p->next = i;
1203 if (prev < 0)
1204 swap_list.head = swap_list.next = p - swap_info;
1205 else
1206 swap_info[prev].next = p - swap_info;
1207 nr_swap_pages += p->pages;
1208 total_swap_pages += p->pages;
1209 p->flags |= SWP_WRITEOK;
1210 swap_list_unlock();
1211 goto out_dput;
1212 }
1213 down(&swapon_sem);
1214 swap_list_lock();
1215 drain_mmlist();
1216 swap_device_lock(p);
1217 swap_file = p->swap_file;
1218 p->swap_file = NULL;
1219 p->max = 0;
1220 swap_map = p->swap_map;
1221 p->swap_map = NULL;
1222 p->flags = 0;
1223 destroy_swap_extents(p);
1224 swap_device_unlock(p);
1225 swap_list_unlock();
1226 up(&swapon_sem);
1227 vfree(swap_map);
1228 inode = mapping->host;
1229 if (S_ISBLK(inode->i_mode)) {
1230 struct block_device *bdev = I_BDEV(inode);
1231 set_blocksize(bdev, p->old_block_size);
1232 bd_release(bdev);
1233 } else {
1234 down(&inode->i_sem);
1235 inode->i_flags &= ~S_SWAPFILE;
1236 up(&inode->i_sem);
1237 }
1238 filp_close(swap_file, NULL);
1239 err = 0;
1240
1241 out_dput:
1242 filp_close(victim, NULL);
1243 out:
1244 return err;
1245 }
1246
1247 #ifdef CONFIG_PROC_FS
1248 /* iterator */
1249 static void *swap_start(struct seq_file *swap, loff_t *pos)
1250 {
1251 struct swap_info_struct *ptr = swap_info;
1252 int i;
1253 loff_t l = *pos;
1254
1255 down(&swapon_sem);
1256
1257 for (i = 0; i < nr_swapfiles; i++, ptr++) {
1258 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1259 continue;
1260 if (!l--)
1261 return ptr;
1262 }
1263
1264 return NULL;
1265 }
1266
1267 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1268 {
1269 struct swap_info_struct *ptr = v;
1270 struct swap_info_struct *endptr = swap_info + nr_swapfiles;
1271
1272 for (++ptr; ptr < endptr; ptr++) {
1273 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1274 continue;
1275 ++*pos;
1276 return ptr;
1277 }
1278
1279 return NULL;
1280 }
1281
1282 static void swap_stop(struct seq_file *swap, void *v)
1283 {
1284 up(&swapon_sem);
1285 }
1286
1287 static int swap_show(struct seq_file *swap, void *v)
1288 {
1289 struct swap_info_struct *ptr = v;
1290 struct file *file;
1291 int len;
1292
1293 if (v == swap_info)
1294 seq_puts(swap, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1295
1296 file = ptr->swap_file;
1297 len = seq_path(swap, file->f_vfsmnt, file->f_dentry, " \t\n\\");
1298 seq_printf(swap, "%*s%s\t%d\t%ld\t%d\n",
1299 len < 40 ? 40 - len : 1, " ",
1300 S_ISBLK(file->f_dentry->d_inode->i_mode) ?
1301 "partition" : "file\t",
1302 ptr->pages << (PAGE_SHIFT - 10),
1303 ptr->inuse_pages << (PAGE_SHIFT - 10),
1304 ptr->prio);
1305 return 0;
1306 }
1307
1308 static struct seq_operations swaps_op = {
1309 .start = swap_start,
1310 .next = swap_next,
1311 .stop = swap_stop,
1312 .show = swap_show
1313 };
1314
1315 static int swaps_open(struct inode *inode, struct file *file)
1316 {
1317 return seq_open(file, &swaps_op);
1318 }
1319
1320 static struct file_operations proc_swaps_operations = {
1321 .open = swaps_open,
1322 .read = seq_read,
1323 .llseek = seq_lseek,
1324 .release = seq_release,
1325 };
1326
1327 static int __init procswaps_init(void)
1328 {
1329 struct proc_dir_entry *entry;
1330
1331 entry = create_proc_entry("swaps", 0, NULL);
1332 if (entry)
1333 entry->proc_fops = &proc_swaps_operations;
1334 return 0;
1335 }
1336 __initcall(procswaps_init);
1337 #endif /* CONFIG_PROC_FS */
1338
1339 /*
1340 * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
1341 *
1342 * The swapon system call
1343 */
1344 asmlinkage long sys_swapon(const char __user * specialfile, int swap_flags)
1345 {
1346 struct swap_info_struct * p;
1347 char *name = NULL;
1348 struct block_device *bdev = NULL;
1349 struct file *swap_file = NULL;
1350 struct address_space *mapping;
1351 unsigned int type;
1352 int i, prev;
1353 int error;
1354 static int least_priority;
1355 union swap_header *swap_header = NULL;
1356 int swap_header_version;
1357 int nr_good_pages = 0;
1358 unsigned long maxpages = 1;
1359 int swapfilesize;
1360 unsigned short *swap_map;
1361 struct page *page = NULL;
1362 struct inode *inode = NULL;
1363 int did_down = 0;
1364
1365 if (!capable(CAP_SYS_ADMIN))
1366 return -EPERM;
1367 swap_list_lock();
1368 p = swap_info;
1369 for (type = 0 ; type < nr_swapfiles ; type++,p++)
1370 if (!(p->flags & SWP_USED))
1371 break;
1372 error = -EPERM;
1373 /*
1374 * Test if adding another swap device is possible. There are
1375 * two limiting factors: 1) the number of bits for the swap
1376 * type swp_entry_t definition and 2) the number of bits for
1377 * the swap type in the swap ptes as defined by the different
1378 * architectures. To honor both limitations a swap entry
1379 * with swap offset 0 and swap type ~0UL is created, encoded
1380 * to a swap pte, decoded to a swp_entry_t again and finally
1381 * the swap type part is extracted. This will mask all bits
1382 * from the initial ~0UL that can't be encoded in either the
1383 * swp_entry_t or the architecture definition of a swap pte.
1384 */
1385 if (type > swp_type(pte_to_swp_entry(swp_entry_to_pte(swp_entry(~0UL,0))))) {
1386 swap_list_unlock();
1387 goto out;
1388 }
1389 if (type >= nr_swapfiles)
1390 nr_swapfiles = type+1;
1391 INIT_LIST_HEAD(&p->extent_list);
1392 p->flags = SWP_USED;
1393 p->nr_extents = 0;
1394 p->swap_file = NULL;
1395 p->old_block_size = 0;
1396 p->swap_map = NULL;
1397 p->lowest_bit = 0;
1398 p->highest_bit = 0;
1399 p->cluster_nr = 0;
1400 p->inuse_pages = 0;
1401 spin_lock_init(&p->sdev_lock);
1402 p->next = -1;
1403 if (swap_flags & SWAP_FLAG_PREFER) {
1404 p->prio =
1405 (swap_flags & SWAP_FLAG_PRIO_MASK)>>SWAP_FLAG_PRIO_SHIFT;
1406 } else {
1407 p->prio = --least_priority;
1408 }
1409 swap_list_unlock();
1410 name = getname(specialfile);
1411 error = PTR_ERR(name);
1412 if (IS_ERR(name)) {
1413 name = NULL;
1414 goto bad_swap_2;
1415 }
1416 swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0);
1417 error = PTR_ERR(swap_file);
1418 if (IS_ERR(swap_file)) {
1419 swap_file = NULL;
1420 goto bad_swap_2;
1421 }
1422
1423 p->swap_file = swap_file;
1424 mapping = swap_file->f_mapping;
1425 inode = mapping->host;
1426
1427 error = -EBUSY;
1428 for (i = 0; i < nr_swapfiles; i++) {
1429 struct swap_info_struct *q = &swap_info[i];
1430
1431 if (i == type || !q->swap_file)
1432 continue;
1433 if (mapping == q->swap_file->f_mapping)
1434 goto bad_swap;
1435 }
1436
1437 error = -EINVAL;
1438 if (S_ISBLK(inode->i_mode)) {
1439 bdev = I_BDEV(inode);
1440 error = bd_claim(bdev, sys_swapon);
1441 if (error < 0) {
1442 bdev = NULL;
1443 goto bad_swap;
1444 }
1445 p->old_block_size = block_size(bdev);
1446 error = set_blocksize(bdev, PAGE_SIZE);
1447 if (error < 0)
1448 goto bad_swap;
1449 p->bdev = bdev;
1450 } else if (S_ISREG(inode->i_mode)) {
1451 p->bdev = inode->i_sb->s_bdev;
1452 down(&inode->i_sem);
1453 did_down = 1;
1454 if (IS_SWAPFILE(inode)) {
1455 error = -EBUSY;
1456 goto bad_swap;
1457 }
1458 } else {
1459 goto bad_swap;
1460 }
1461
1462 swapfilesize = i_size_read(inode) >> PAGE_SHIFT;
1463
1464 /*
1465 * Read the swap header.
1466 */
1467 if (!mapping->a_ops->readpage) {
1468 error = -EINVAL;
1469 goto bad_swap;
1470 }
1471 page = read_cache_page(mapping, 0,
1472 (filler_t *)mapping->a_ops->readpage, swap_file);
1473 if (IS_ERR(page)) {
1474 error = PTR_ERR(page);
1475 goto bad_swap;
1476 }
1477 wait_on_page_locked(page);
1478 if (!PageUptodate(page))
1479 goto bad_swap;
1480 kmap(page);
1481 swap_header = page_address(page);
1482
1483 if (!memcmp("SWAP-SPACE",swap_header->magic.magic,10))
1484 swap_header_version = 1;
1485 else if (!memcmp("SWAPSPACE2",swap_header->magic.magic,10))
1486 swap_header_version = 2;
1487 else {
1488 printk("Unable to find swap-space signature\n");
1489 error = -EINVAL;
1490 goto bad_swap;
1491 }
1492
1493 switch (swap_header_version) {
1494 case 1:
1495 printk(KERN_ERR "version 0 swap is no longer supported. "
1496 "Use mkswap -v1 %s\n", name);
1497 error = -EINVAL;
1498 goto bad_swap;
1499 case 2:
1500 /* Check the swap header's sub-version and the size of
1501 the swap file and bad block lists */
1502 if (swap_header->info.version != 1) {
1503 printk(KERN_WARNING
1504 "Unable to handle swap header version %d\n",
1505 swap_header->info.version);
1506 error = -EINVAL;
1507 goto bad_swap;
1508 }
1509
1510 p->lowest_bit = 1;
1511 /*
1512 * Find out how many pages are allowed for a single swap
1513 * device. There are two limiting factors: 1) the number of
1514 * bits for the swap offset in the swp_entry_t type and
1515 * 2) the number of bits in the a swap pte as defined by
1516 * the different architectures. In order to find the
1517 * largest possible bit mask a swap entry with swap type 0
1518 * and swap offset ~0UL is created, encoded to a swap pte,
1519 * decoded to a swp_entry_t again and finally the swap
1520 * offset is extracted. This will mask all the bits from
1521 * the initial ~0UL mask that can't be encoded in either
1522 * the swp_entry_t or the architecture definition of a
1523 * swap pte.
1524 */
1525 maxpages = swp_offset(pte_to_swp_entry(swp_entry_to_pte(swp_entry(0,~0UL)))) - 1;
1526 if (maxpages > swap_header->info.last_page)
1527 maxpages = swap_header->info.last_page;
1528 p->highest_bit = maxpages - 1;
1529
1530 error = -EINVAL;
1531 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
1532 goto bad_swap;
1533
1534 /* OK, set up the swap map and apply the bad block list */
1535 if (!(p->swap_map = vmalloc(maxpages * sizeof(short)))) {
1536 error = -ENOMEM;
1537 goto bad_swap;
1538 }
1539
1540 error = 0;
1541 memset(p->swap_map, 0, maxpages * sizeof(short));
1542 for (i=0; i<swap_header->info.nr_badpages; i++) {
1543 int page = swap_header->info.badpages[i];
1544 if (page <= 0 || page >= swap_header->info.last_page)
1545 error = -EINVAL;
1546 else
1547 p->swap_map[page] = SWAP_MAP_BAD;
1548 }
1549 nr_good_pages = swap_header->info.last_page -
1550 swap_header->info.nr_badpages -
1551 1 /* header page */;
1552 if (error)
1553 goto bad_swap;
1554 }
1555
1556 if (swapfilesize && maxpages > swapfilesize) {
1557 printk(KERN_WARNING
1558 "Swap area shorter than signature indicates\n");
1559 error = -EINVAL;
1560 goto bad_swap;
1561 }
1562 if (!nr_good_pages) {
1563 printk(KERN_WARNING "Empty swap-file\n");
1564 error = -EINVAL;
1565 goto bad_swap;
1566 }
1567 p->swap_map[0] = SWAP_MAP_BAD;
1568 p->max = maxpages;
1569 p->pages = nr_good_pages;
1570
1571 error = setup_swap_extents(p);
1572 if (error)
1573 goto bad_swap;
1574
1575 down(&swapon_sem);
1576 swap_list_lock();
1577 swap_device_lock(p);
1578 p->flags = SWP_ACTIVE;
1579 nr_swap_pages += nr_good_pages;
1580 total_swap_pages += nr_good_pages;
1581 printk(KERN_INFO "Adding %dk swap on %s. Priority:%d extents:%d\n",
1582 nr_good_pages<<(PAGE_SHIFT-10), name,
1583 p->prio, p->nr_extents);
1584
1585 /* insert swap space into swap_list: */
1586 prev = -1;
1587 for (i = swap_list.head; i >= 0; i = swap_info[i].next) {
1588 if (p->prio >= swap_info[i].prio) {
1589 break;
1590 }
1591 prev = i;
1592 }
1593 p->next = i;
1594 if (prev < 0) {
1595 swap_list.head = swap_list.next = p - swap_info;
1596 } else {
1597 swap_info[prev].next = p - swap_info;
1598 }
1599 swap_device_unlock(p);
1600 swap_list_unlock();
1601 up(&swapon_sem);
1602 error = 0;
1603 goto out;
1604 bad_swap:
1605 if (bdev) {
1606 set_blocksize(bdev, p->old_block_size);
1607 bd_release(bdev);
1608 }
1609 bad_swap_2:
1610 swap_list_lock();
1611 swap_map = p->swap_map;
1612 p->swap_file = NULL;
1613 p->swap_map = NULL;
1614 p->flags = 0;
1615 if (!(swap_flags & SWAP_FLAG_PREFER))
1616 ++least_priority;
1617 swap_list_unlock();
1618 destroy_swap_extents(p);
1619 vfree(swap_map);
1620 if (swap_file)
1621 filp_close(swap_file, NULL);
1622 out:
1623 if (page && !IS_ERR(page)) {
1624 kunmap(page);
1625 page_cache_release(page);
1626 }
1627 if (name)
1628 putname(name);
1629 if (did_down) {
1630 if (!error)
1631 inode->i_flags |= S_SWAPFILE;
1632 up(&inode->i_sem);
1633 }
1634 return error;
1635 }
1636
1637 void si_swapinfo(struct sysinfo *val)
1638 {
1639 unsigned int i;
1640 unsigned long nr_to_be_unused = 0;
1641
1642 swap_list_lock();
1643 for (i = 0; i < nr_swapfiles; i++) {
1644 if (!(swap_info[i].flags & SWP_USED) ||
1645 (swap_info[i].flags & SWP_WRITEOK))
1646 continue;
1647 nr_to_be_unused += swap_info[i].inuse_pages;
1648 }
1649 val->freeswap = nr_swap_pages + nr_to_be_unused;
1650 val->totalswap = total_swap_pages + nr_to_be_unused;
1651 swap_list_unlock();
1652 }
1653
1654 /*
1655 * Verify that a swap entry is valid and increment its swap map count.
1656 *
1657 * Note: if swap_map[] reaches SWAP_MAP_MAX the entries are treated as
1658 * "permanent", but will be reclaimed by the next swapoff.
1659 */
1660 int swap_duplicate(swp_entry_t entry)
1661 {
1662 struct swap_info_struct * p;
1663 unsigned long offset, type;
1664 int result = 0;
1665
1666 type = swp_type(entry);
1667 if (type >= nr_swapfiles)
1668 goto bad_file;
1669 p = type + swap_info;
1670 offset = swp_offset(entry);
1671
1672 swap_device_lock(p);
1673 if (offset < p->max && p->swap_map[offset]) {
1674 if (p->swap_map[offset] < SWAP_MAP_MAX - 1) {
1675 p->swap_map[offset]++;
1676 result = 1;
1677 } else if (p->swap_map[offset] <= SWAP_MAP_MAX) {
1678 if (swap_overflow++ < 5)
1679 printk(KERN_WARNING "swap_dup: swap entry overflow\n");
1680 p->swap_map[offset] = SWAP_MAP_MAX;
1681 result = 1;
1682 }
1683 }
1684 swap_device_unlock(p);
1685 out:
1686 return result;
1687
1688 bad_file:
1689 printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
1690 goto out;
1691 }
1692
1693 struct swap_info_struct *
1694 get_swap_info_struct(unsigned type)
1695 {
1696 return &swap_info[type];
1697 }
1698
1699 /*
1700 * swap_device_lock prevents swap_map being freed. Don't grab an extra
1701 * reference on the swaphandle, it doesn't matter if it becomes unused.
1702 */
1703 int valid_swaphandles(swp_entry_t entry, unsigned long *offset)
1704 {
1705 int ret = 0, i = 1 << page_cluster;
1706 unsigned long toff;
1707 struct swap_info_struct *swapdev = swp_type(entry) + swap_info;
1708
1709 if (!page_cluster) /* no readahead */
1710 return 0;
1711 toff = (swp_offset(entry) >> page_cluster) << page_cluster;
1712 if (!toff) /* first page is swap header */
1713 toff++, i--;
1714 *offset = toff;
1715
1716 swap_device_lock(swapdev);
1717 do {
1718 /* Don't read-ahead past the end of the swap area */
1719 if (toff >= swapdev->max)
1720 break;
1721 /* Don't read in free or bad pages */
1722 if (!swapdev->swap_map[toff])
1723 break;
1724 if (swapdev->swap_map[toff] == SWAP_MAP_BAD)
1725 break;
1726 toff++;
1727 ret++;
1728 } while (--i);
1729 swap_device_unlock(swapdev);
1730 return ret;
1731 }
1732
|
This page was automatically generated by the
LXR engine.
|