swapfile.c revision 11d31886dbcb61039ed3789e583d21c6e70960fd
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/backing-dev.h> 28#include <linux/syscalls.h> 29 30#include <asm/pgtable.h> 31#include <asm/tlbflush.h> 32#include <linux/swapops.h> 33 34DEFINE_SPINLOCK(swaplock); 35unsigned int nr_swapfiles; 36long total_swap_pages; 37static int swap_overflow; 38 39EXPORT_SYMBOL(total_swap_pages); 40 41static const char Bad_file[] = "Bad swap file entry "; 42static const char Unused_file[] = "Unused swap file entry "; 43static const char Bad_offset[] = "Bad swap offset entry "; 44static const char Unused_offset[] = "Unused swap offset entry "; 45 46struct swap_list_t swap_list = {-1, -1}; 47 48struct swap_info_struct swap_info[MAX_SWAPFILES]; 49 50static DECLARE_MUTEX(swapon_sem); 51 52/* 53 * We need this because the bdev->unplug_fn can sleep and we cannot 54 * hold swap_list_lock while calling the unplug_fn. And swap_list_lock 55 * cannot be turned into a semaphore. 56 */ 57static DECLARE_RWSEM(swap_unplug_sem); 58 59#define SWAPFILE_CLUSTER 256 60 61void swap_unplug_io_fn(struct backing_dev_info *unused_bdi, struct page *page) 62{ 63 swp_entry_t entry; 64 65 down_read(&swap_unplug_sem); 66 entry.val = page->private; 67 if (PageSwapCache(page)) { 68 struct block_device *bdev = swap_info[swp_type(entry)].bdev; 69 struct backing_dev_info *bdi; 70 71 /* 72 * If the page is removed from swapcache from under us (with a 73 * racy try_to_unuse/swapoff) we need an additional reference 74 * count to avoid reading garbage from page->private above. If 75 * the WARN_ON triggers during a swapoff it maybe the race 76 * condition and it's harmless. However if it triggers without 77 * swapoff it signals a problem. 78 */ 79 WARN_ON(page_count(page) <= 1); 80 81 bdi = bdev->bd_inode->i_mapping->backing_dev_info; 82 blk_run_backing_dev(bdi, page); 83 } 84 up_read(&swap_unplug_sem); 85} 86 87static inline int scan_swap_map(struct swap_info_struct *si) 88{ 89 unsigned long offset; 90 /* 91 * We try to cluster swap pages by allocating them 92 * sequentially in swap. Once we've allocated 93 * SWAPFILE_CLUSTER pages this way, however, we resort to 94 * first-free allocation, starting a new cluster. This 95 * prevents us from scattering swap pages all over the entire 96 * swap partition, so that we reduce overall disk seek times 97 * between swap pages. -- sct */ 98 if (si->cluster_nr) { 99 while (si->cluster_next <= si->highest_bit) { 100 offset = si->cluster_next++; 101 if (si->swap_map[offset]) 102 continue; 103 si->cluster_nr--; 104 goto got_page; 105 } 106 } 107 si->cluster_nr = SWAPFILE_CLUSTER; 108 109 /* try to find an empty (even not aligned) cluster. */ 110 offset = si->lowest_bit; 111 check_next_cluster: 112 if (offset+SWAPFILE_CLUSTER-1 <= si->highest_bit) 113 { 114 unsigned long nr; 115 for (nr = offset; nr < offset+SWAPFILE_CLUSTER; nr++) 116 if (si->swap_map[nr]) 117 { 118 offset = nr+1; 119 goto check_next_cluster; 120 } 121 /* We found a completly empty cluster, so start 122 * using it. 123 */ 124 goto got_page; 125 } 126 /* No luck, so now go finegrined as usual. -Andrea */ 127 for (offset = si->lowest_bit; offset <= si->highest_bit ; offset++) { 128 if (si->swap_map[offset]) 129 continue; 130 si->lowest_bit = offset+1; 131 got_page: 132 if (offset == si->lowest_bit) 133 si->lowest_bit++; 134 if (offset == si->highest_bit) 135 si->highest_bit--; 136 if (si->lowest_bit > si->highest_bit) { 137 si->lowest_bit = si->max; 138 si->highest_bit = 0; 139 } 140 si->swap_map[offset] = 1; 141 si->inuse_pages++; 142 nr_swap_pages--; 143 si->cluster_next = offset+1; 144 return offset; 145 } 146 si->lowest_bit = si->max; 147 si->highest_bit = 0; 148 return 0; 149} 150 151swp_entry_t get_swap_page(void) 152{ 153 struct swap_info_struct * p; 154 unsigned long offset; 155 swp_entry_t entry; 156 int type, wrapped = 0; 157 158 entry.val = 0; /* Out of memory */ 159 swap_list_lock(); 160 type = swap_list.next; 161 if (type < 0) 162 goto out; 163 if (nr_swap_pages <= 0) 164 goto out; 165 166 while (1) { 167 p = &swap_info[type]; 168 if ((p->flags & SWP_ACTIVE) == SWP_ACTIVE) { 169 swap_device_lock(p); 170 offset = scan_swap_map(p); 171 swap_device_unlock(p); 172 if (offset) { 173 entry = swp_entry(type,offset); 174 type = swap_info[type].next; 175 if (type < 0 || 176 p->prio != swap_info[type].prio) { 177 swap_list.next = swap_list.head; 178 } else { 179 swap_list.next = type; 180 } 181 goto out; 182 } 183 } 184 type = p->next; 185 if (!wrapped) { 186 if (type < 0 || p->prio != swap_info[type].prio) { 187 type = swap_list.head; 188 wrapped = 1; 189 } 190 } else 191 if (type < 0) 192 goto out; /* out of swap space */ 193 } 194out: 195 swap_list_unlock(); 196 return entry; 197} 198 199static struct swap_info_struct * swap_info_get(swp_entry_t entry) 200{ 201 struct swap_info_struct * p; 202 unsigned long offset, type; 203 204 if (!entry.val) 205 goto out; 206 type = swp_type(entry); 207 if (type >= nr_swapfiles) 208 goto bad_nofile; 209 p = & swap_info[type]; 210 if (!(p->flags & SWP_USED)) 211 goto bad_device; 212 offset = swp_offset(entry); 213 if (offset >= p->max) 214 goto bad_offset; 215 if (!p->swap_map[offset]) 216 goto bad_free; 217 swap_list_lock(); 218 if (p->prio > swap_info[swap_list.next].prio) 219 swap_list.next = type; 220 swap_device_lock(p); 221 return p; 222 223bad_free: 224 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val); 225 goto out; 226bad_offset: 227 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val); 228 goto out; 229bad_device: 230 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val); 231 goto out; 232bad_nofile: 233 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val); 234out: 235 return NULL; 236} 237 238static void swap_info_put(struct swap_info_struct * p) 239{ 240 swap_device_unlock(p); 241 swap_list_unlock(); 242} 243 244static int swap_entry_free(struct swap_info_struct *p, unsigned long offset) 245{ 246 int count = p->swap_map[offset]; 247 248 if (count < SWAP_MAP_MAX) { 249 count--; 250 p->swap_map[offset] = count; 251 if (!count) { 252 if (offset < p->lowest_bit) 253 p->lowest_bit = offset; 254 if (offset > p->highest_bit) 255 p->highest_bit = offset; 256 nr_swap_pages++; 257 p->inuse_pages--; 258 } 259 } 260 return count; 261} 262 263/* 264 * Caller has made sure that the swapdevice corresponding to entry 265 * is still around or has not been recycled. 266 */ 267void swap_free(swp_entry_t entry) 268{ 269 struct swap_info_struct * p; 270 271 p = swap_info_get(entry); 272 if (p) { 273 swap_entry_free(p, swp_offset(entry)); 274 swap_info_put(p); 275 } 276} 277 278/* 279 * How many references to page are currently swapped out? 280 */ 281static inline int page_swapcount(struct page *page) 282{ 283 int count = 0; 284 struct swap_info_struct *p; 285 swp_entry_t entry; 286 287 entry.val = page->private; 288 p = swap_info_get(entry); 289 if (p) { 290 /* Subtract the 1 for the swap cache itself */ 291 count = p->swap_map[swp_offset(entry)] - 1; 292 swap_info_put(p); 293 } 294 return count; 295} 296 297/* 298 * We can use this swap cache entry directly 299 * if there are no other references to it. 300 */ 301int can_share_swap_page(struct page *page) 302{ 303 int count; 304 305 BUG_ON(!PageLocked(page)); 306 count = page_mapcount(page); 307 if (count <= 1 && PageSwapCache(page)) 308 count += page_swapcount(page); 309 return count == 1; 310} 311 312/* 313 * Work out if there are any other processes sharing this 314 * swap cache page. Free it if you can. Return success. 315 */ 316int remove_exclusive_swap_page(struct page *page) 317{ 318 int retval; 319 struct swap_info_struct * p; 320 swp_entry_t entry; 321 322 BUG_ON(PagePrivate(page)); 323 BUG_ON(!PageLocked(page)); 324 325 if (!PageSwapCache(page)) 326 return 0; 327 if (PageWriteback(page)) 328 return 0; 329 if (page_count(page) != 2) /* 2: us + cache */ 330 return 0; 331 332 entry.val = page->private; 333 p = swap_info_get(entry); 334 if (!p) 335 return 0; 336 337 /* Is the only swap cache user the cache itself? */ 338 retval = 0; 339 if (p->swap_map[swp_offset(entry)] == 1) { 340 /* Recheck the page count with the swapcache lock held.. */ 341 write_lock_irq(&swapper_space.tree_lock); 342 if ((page_count(page) == 2) && !PageWriteback(page)) { 343 __delete_from_swap_cache(page); 344 SetPageDirty(page); 345 retval = 1; 346 } 347 write_unlock_irq(&swapper_space.tree_lock); 348 } 349 swap_info_put(p); 350 351 if (retval) { 352 swap_free(entry); 353 page_cache_release(page); 354 } 355 356 return retval; 357} 358 359/* 360 * Free the swap entry like above, but also try to 361 * free the page cache entry if it is the last user. 362 */ 363void free_swap_and_cache(swp_entry_t entry) 364{ 365 struct swap_info_struct * p; 366 struct page *page = NULL; 367 368 p = swap_info_get(entry); 369 if (p) { 370 if (swap_entry_free(p, swp_offset(entry)) == 1) 371 page = find_trylock_page(&swapper_space, entry.val); 372 swap_info_put(p); 373 } 374 if (page) { 375 int one_user; 376 377 BUG_ON(PagePrivate(page)); 378 page_cache_get(page); 379 one_user = (page_count(page) == 2); 380 /* Only cache user (+us), or swap space full? Free it! */ 381 if (!PageWriteback(page) && (one_user || vm_swap_full())) { 382 delete_from_swap_cache(page); 383 SetPageDirty(page); 384 } 385 unlock_page(page); 386 page_cache_release(page); 387 } 388} 389 390/* 391 * Always set the resulting pte to be nowrite (the same as COW pages 392 * after one process has exited). We don't know just how many PTEs will 393 * share this swap entry, so be cautious and let do_wp_page work out 394 * what to do if a write is requested later. 395 * 396 * vma->vm_mm->page_table_lock is held. 397 */ 398static void unuse_pte(struct vm_area_struct *vma, pte_t *pte, 399 unsigned long addr, swp_entry_t entry, struct page *page) 400{ 401 inc_mm_counter(vma->vm_mm, rss); 402 get_page(page); 403 set_pte_at(vma->vm_mm, addr, pte, 404 pte_mkold(mk_pte(page, vma->vm_page_prot))); 405 page_add_anon_rmap(page, vma, addr); 406 swap_free(entry); 407 /* 408 * Move the page to the active list so it is not 409 * immediately swapped out again after swapon. 410 */ 411 activate_page(page); 412} 413 414static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd, 415 unsigned long addr, unsigned long end, 416 swp_entry_t entry, struct page *page) 417{ 418 pte_t *pte; 419 pte_t swp_pte = swp_entry_to_pte(entry); 420 421 pte = pte_offset_map(pmd, addr); 422 do { 423 /* 424 * swapoff spends a _lot_ of time in this loop! 425 * Test inline before going to call unuse_pte. 426 */ 427 if (unlikely(pte_same(*pte, swp_pte))) { 428 unuse_pte(vma, pte, addr, entry, page); 429 pte_unmap(pte); 430 return 1; 431 } 432 } while (pte++, addr += PAGE_SIZE, addr != end); 433 pte_unmap(pte - 1); 434 return 0; 435} 436 437static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud, 438 unsigned long addr, unsigned long end, 439 swp_entry_t entry, struct page *page) 440{ 441 pmd_t *pmd; 442 unsigned long next; 443 444 pmd = pmd_offset(pud, addr); 445 do { 446 next = pmd_addr_end(addr, end); 447 if (pmd_none_or_clear_bad(pmd)) 448 continue; 449 if (unuse_pte_range(vma, pmd, addr, next, entry, page)) 450 return 1; 451 } while (pmd++, addr = next, addr != end); 452 return 0; 453} 454 455static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd, 456 unsigned long addr, unsigned long end, 457 swp_entry_t entry, struct page *page) 458{ 459 pud_t *pud; 460 unsigned long next; 461 462 pud = pud_offset(pgd, addr); 463 do { 464 next = pud_addr_end(addr, end); 465 if (pud_none_or_clear_bad(pud)) 466 continue; 467 if (unuse_pmd_range(vma, pud, addr, next, entry, page)) 468 return 1; 469 } while (pud++, addr = next, addr != end); 470 return 0; 471} 472 473static int unuse_vma(struct vm_area_struct *vma, 474 swp_entry_t entry, struct page *page) 475{ 476 pgd_t *pgd; 477 unsigned long addr, end, next; 478 479 if (page->mapping) { 480 addr = page_address_in_vma(page, vma); 481 if (addr == -EFAULT) 482 return 0; 483 else 484 end = addr + PAGE_SIZE; 485 } else { 486 addr = vma->vm_start; 487 end = vma->vm_end; 488 } 489 490 pgd = pgd_offset(vma->vm_mm, addr); 491 do { 492 next = pgd_addr_end(addr, end); 493 if (pgd_none_or_clear_bad(pgd)) 494 continue; 495 if (unuse_pud_range(vma, pgd, addr, next, entry, page)) 496 return 1; 497 } while (pgd++, addr = next, addr != end); 498 return 0; 499} 500 501static int unuse_mm(struct mm_struct *mm, 502 swp_entry_t entry, struct page *page) 503{ 504 struct vm_area_struct *vma; 505 506 if (!down_read_trylock(&mm->mmap_sem)) { 507 /* 508 * Activate page so shrink_cache is unlikely to unmap its 509 * ptes while lock is dropped, so swapoff can make progress. 510 */ 511 activate_page(page); 512 unlock_page(page); 513 down_read(&mm->mmap_sem); 514 lock_page(page); 515 } 516 spin_lock(&mm->page_table_lock); 517 for (vma = mm->mmap; vma; vma = vma->vm_next) { 518 if (vma->anon_vma && unuse_vma(vma, entry, page)) 519 break; 520 } 521 spin_unlock(&mm->page_table_lock); 522 up_read(&mm->mmap_sem); 523 /* 524 * Currently unuse_mm cannot fail, but leave error handling 525 * at call sites for now, since we change it from time to time. 526 */ 527 return 0; 528} 529 530/* 531 * Scan swap_map from current position to next entry still in use. 532 * Recycle to start on reaching the end, returning 0 when empty. 533 */ 534static int find_next_to_unuse(struct swap_info_struct *si, int prev) 535{ 536 int max = si->max; 537 int i = prev; 538 int count; 539 540 /* 541 * No need for swap_device_lock(si) here: we're just looking 542 * for whether an entry is in use, not modifying it; false 543 * hits are okay, and sys_swapoff() has already prevented new 544 * allocations from this area (while holding swap_list_lock()). 545 */ 546 for (;;) { 547 if (++i >= max) { 548 if (!prev) { 549 i = 0; 550 break; 551 } 552 /* 553 * No entries in use at top of swap_map, 554 * loop back to start and recheck there. 555 */ 556 max = prev + 1; 557 prev = 0; 558 i = 1; 559 } 560 count = si->swap_map[i]; 561 if (count && count != SWAP_MAP_BAD) 562 break; 563 } 564 return i; 565} 566 567/* 568 * We completely avoid races by reading each swap page in advance, 569 * and then search for the process using it. All the necessary 570 * page table adjustments can then be made atomically. 571 */ 572static int try_to_unuse(unsigned int type) 573{ 574 struct swap_info_struct * si = &swap_info[type]; 575 struct mm_struct *start_mm; 576 unsigned short *swap_map; 577 unsigned short swcount; 578 struct page *page; 579 swp_entry_t entry; 580 int i = 0; 581 int retval = 0; 582 int reset_overflow = 0; 583 int shmem; 584 585 /* 586 * When searching mms for an entry, a good strategy is to 587 * start at the first mm we freed the previous entry from 588 * (though actually we don't notice whether we or coincidence 589 * freed the entry). Initialize this start_mm with a hold. 590 * 591 * A simpler strategy would be to start at the last mm we 592 * freed the previous entry from; but that would take less 593 * advantage of mmlist ordering, which clusters forked mms 594 * together, child after parent. If we race with dup_mmap(), we 595 * prefer to resolve parent before child, lest we miss entries 596 * duplicated after we scanned child: using last mm would invert 597 * that. Though it's only a serious concern when an overflowed 598 * swap count is reset from SWAP_MAP_MAX, preventing a rescan. 599 */ 600 start_mm = &init_mm; 601 atomic_inc(&init_mm.mm_users); 602 603 /* 604 * Keep on scanning until all entries have gone. Usually, 605 * one pass through swap_map is enough, but not necessarily: 606 * there are races when an instance of an entry might be missed. 607 */ 608 while ((i = find_next_to_unuse(si, i)) != 0) { 609 if (signal_pending(current)) { 610 retval = -EINTR; 611 break; 612 } 613 614 /* 615 * Get a page for the entry, using the existing swap 616 * cache page if there is one. Otherwise, get a clean 617 * page and read the swap into it. 618 */ 619 swap_map = &si->swap_map[i]; 620 entry = swp_entry(type, i); 621 page = read_swap_cache_async(entry, NULL, 0); 622 if (!page) { 623 /* 624 * Either swap_duplicate() failed because entry 625 * has been freed independently, and will not be 626 * reused since sys_swapoff() already disabled 627 * allocation from here, or alloc_page() failed. 628 */ 629 if (!*swap_map) 630 continue; 631 retval = -ENOMEM; 632 break; 633 } 634 635 /* 636 * Don't hold on to start_mm if it looks like exiting. 637 */ 638 if (atomic_read(&start_mm->mm_users) == 1) { 639 mmput(start_mm); 640 start_mm = &init_mm; 641 atomic_inc(&init_mm.mm_users); 642 } 643 644 /* 645 * Wait for and lock page. When do_swap_page races with 646 * try_to_unuse, do_swap_page can handle the fault much 647 * faster than try_to_unuse can locate the entry. This 648 * apparently redundant "wait_on_page_locked" lets try_to_unuse 649 * defer to do_swap_page in such a case - in some tests, 650 * do_swap_page and try_to_unuse repeatedly compete. 651 */ 652 wait_on_page_locked(page); 653 wait_on_page_writeback(page); 654 lock_page(page); 655 wait_on_page_writeback(page); 656 657 /* 658 * Remove all references to entry. 659 * Whenever we reach init_mm, there's no address space 660 * to search, but use it as a reminder to search shmem. 661 */ 662 shmem = 0; 663 swcount = *swap_map; 664 if (swcount > 1) { 665 if (start_mm == &init_mm) 666 shmem = shmem_unuse(entry, page); 667 else 668 retval = unuse_mm(start_mm, entry, page); 669 } 670 if (*swap_map > 1) { 671 int set_start_mm = (*swap_map >= swcount); 672 struct list_head *p = &start_mm->mmlist; 673 struct mm_struct *new_start_mm = start_mm; 674 struct mm_struct *prev_mm = start_mm; 675 struct mm_struct *mm; 676 677 atomic_inc(&new_start_mm->mm_users); 678 atomic_inc(&prev_mm->mm_users); 679 spin_lock(&mmlist_lock); 680 while (*swap_map > 1 && !retval && 681 (p = p->next) != &start_mm->mmlist) { 682 mm = list_entry(p, struct mm_struct, mmlist); 683 if (atomic_inc_return(&mm->mm_users) == 1) { 684 atomic_dec(&mm->mm_users); 685 continue; 686 } 687 spin_unlock(&mmlist_lock); 688 mmput(prev_mm); 689 prev_mm = mm; 690 691 cond_resched(); 692 693 swcount = *swap_map; 694 if (swcount <= 1) 695 ; 696 else if (mm == &init_mm) { 697 set_start_mm = 1; 698 shmem = shmem_unuse(entry, page); 699 } else 700 retval = unuse_mm(mm, entry, page); 701 if (set_start_mm && *swap_map < swcount) { 702 mmput(new_start_mm); 703 atomic_inc(&mm->mm_users); 704 new_start_mm = mm; 705 set_start_mm = 0; 706 } 707 spin_lock(&mmlist_lock); 708 } 709 spin_unlock(&mmlist_lock); 710 mmput(prev_mm); 711 mmput(start_mm); 712 start_mm = new_start_mm; 713 } 714 if (retval) { 715 unlock_page(page); 716 page_cache_release(page); 717 break; 718 } 719 720 /* 721 * How could swap count reach 0x7fff when the maximum 722 * pid is 0x7fff, and there's no way to repeat a swap 723 * page within an mm (except in shmem, where it's the 724 * shared object which takes the reference count)? 725 * We believe SWAP_MAP_MAX cannot occur in Linux 2.4. 726 * 727 * If that's wrong, then we should worry more about 728 * exit_mmap() and do_munmap() cases described above: 729 * we might be resetting SWAP_MAP_MAX too early here. 730 * We know "Undead"s can happen, they're okay, so don't 731 * report them; but do report if we reset SWAP_MAP_MAX. 732 */ 733 if (*swap_map == SWAP_MAP_MAX) { 734 swap_device_lock(si); 735 *swap_map = 1; 736 swap_device_unlock(si); 737 reset_overflow = 1; 738 } 739 740 /* 741 * If a reference remains (rare), we would like to leave 742 * the page in the swap cache; but try_to_unmap could 743 * then re-duplicate the entry once we drop page lock, 744 * so we might loop indefinitely; also, that page could 745 * not be swapped out to other storage meanwhile. So: 746 * delete from cache even if there's another reference, 747 * after ensuring that the data has been saved to disk - 748 * since if the reference remains (rarer), it will be 749 * read from disk into another page. Splitting into two 750 * pages would be incorrect if swap supported "shared 751 * private" pages, but they are handled by tmpfs files. 752 * 753 * Note shmem_unuse already deleted a swappage from 754 * the swap cache, unless the move to filepage failed: 755 * in which case it left swappage in cache, lowered its 756 * swap count to pass quickly through the loops above, 757 * and now we must reincrement count to try again later. 758 */ 759 if ((*swap_map > 1) && PageDirty(page) && PageSwapCache(page)) { 760 struct writeback_control wbc = { 761 .sync_mode = WB_SYNC_NONE, 762 }; 763 764 swap_writepage(page, &wbc); 765 lock_page(page); 766 wait_on_page_writeback(page); 767 } 768 if (PageSwapCache(page)) { 769 if (shmem) 770 swap_duplicate(entry); 771 else 772 delete_from_swap_cache(page); 773 } 774 775 /* 776 * So we could skip searching mms once swap count went 777 * to 1, we did not mark any present ptes as dirty: must 778 * mark page dirty so shrink_list will preserve it. 779 */ 780 SetPageDirty(page); 781 unlock_page(page); 782 page_cache_release(page); 783 784 /* 785 * Make sure that we aren't completely killing 786 * interactive performance. 787 */ 788 cond_resched(); 789 } 790 791 mmput(start_mm); 792 if (reset_overflow) { 793 printk(KERN_WARNING "swapoff: cleared swap entry overflow\n"); 794 swap_overflow = 0; 795 } 796 return retval; 797} 798 799/* 800 * After a successful try_to_unuse, if no swap is now in use, we know we 801 * can empty the mmlist. swap_list_lock must be held on entry and exit. 802 * Note that mmlist_lock nests inside swap_list_lock, and an mm must be 803 * added to the mmlist just after page_duplicate - before would be racy. 804 */ 805static void drain_mmlist(void) 806{ 807 struct list_head *p, *next; 808 unsigned int i; 809 810 for (i = 0; i < nr_swapfiles; i++) 811 if (swap_info[i].inuse_pages) 812 return; 813 spin_lock(&mmlist_lock); 814 list_for_each_safe(p, next, &init_mm.mmlist) 815 list_del_init(p); 816 spin_unlock(&mmlist_lock); 817} 818 819/* 820 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which 821 * corresponds to page offset `offset'. 822 */ 823sector_t map_swap_page(struct swap_info_struct *sis, pgoff_t offset) 824{ 825 struct swap_extent *se = sis->curr_swap_extent; 826 struct swap_extent *start_se = se; 827 828 for ( ; ; ) { 829 struct list_head *lh; 830 831 if (se->start_page <= offset && 832 offset < (se->start_page + se->nr_pages)) { 833 return se->start_block + (offset - se->start_page); 834 } 835 lh = se->list.next; 836 if (lh == &sis->extent_list) 837 lh = lh->next; 838 se = list_entry(lh, struct swap_extent, list); 839 sis->curr_swap_extent = se; 840 BUG_ON(se == start_se); /* It *must* be present */ 841 } 842} 843 844/* 845 * Free all of a swapdev's extent information 846 */ 847static void destroy_swap_extents(struct swap_info_struct *sis) 848{ 849 while (!list_empty(&sis->extent_list)) { 850 struct swap_extent *se; 851 852 se = list_entry(sis->extent_list.next, 853 struct swap_extent, list); 854 list_del(&se->list); 855 kfree(se); 856 } 857 sis->nr_extents = 0; 858} 859 860/* 861 * Add a block range (and the corresponding page range) into this swapdev's 862 * extent list. The extent list is kept sorted in page order. 863 * 864 * This function rather assumes that it is called in ascending page order. 865 */ 866static int 867add_swap_extent(struct swap_info_struct *sis, unsigned long start_page, 868 unsigned long nr_pages, sector_t start_block) 869{ 870 struct swap_extent *se; 871 struct swap_extent *new_se; 872 struct list_head *lh; 873 874 lh = sis->extent_list.prev; /* The highest page extent */ 875 if (lh != &sis->extent_list) { 876 se = list_entry(lh, struct swap_extent, list); 877 BUG_ON(se->start_page + se->nr_pages != start_page); 878 if (se->start_block + se->nr_pages == start_block) { 879 /* Merge it */ 880 se->nr_pages += nr_pages; 881 return 0; 882 } 883 } 884 885 /* 886 * No merge. Insert a new extent, preserving ordering. 887 */ 888 new_se = kmalloc(sizeof(*se), GFP_KERNEL); 889 if (new_se == NULL) 890 return -ENOMEM; 891 new_se->start_page = start_page; 892 new_se->nr_pages = nr_pages; 893 new_se->start_block = start_block; 894 895 list_add_tail(&new_se->list, &sis->extent_list); 896 sis->nr_extents++; 897 return 0; 898} 899 900/* 901 * A `swap extent' is a simple thing which maps a contiguous range of pages 902 * onto a contiguous range of disk blocks. An ordered list of swap extents 903 * is built at swapon time and is then used at swap_writepage/swap_readpage 904 * time for locating where on disk a page belongs. 905 * 906 * If the swapfile is an S_ISBLK block device, a single extent is installed. 907 * This is done so that the main operating code can treat S_ISBLK and S_ISREG 908 * swap files identically. 909 * 910 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap 911 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK 912 * swapfiles are handled *identically* after swapon time. 913 * 914 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks 915 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If 916 * some stray blocks are found which do not fall within the PAGE_SIZE alignment 917 * requirements, they are simply tossed out - we will never use those blocks 918 * for swapping. 919 * 920 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This 921 * prevents root from shooting her foot off by ftruncating an in-use swapfile, 922 * which will scribble on the fs. 923 * 924 * The amount of disk space which a single swap extent represents varies. 925 * Typically it is in the 1-4 megabyte range. So we can have hundreds of 926 * extents in the list. To avoid much list walking, we cache the previous 927 * search location in `curr_swap_extent', and start new searches from there. 928 * This is extremely effective. The average number of iterations in 929 * map_swap_page() has been measured at about 0.3 per page. - akpm. 930 */ 931static int setup_swap_extents(struct swap_info_struct *sis) 932{ 933 struct inode *inode; 934 unsigned blocks_per_page; 935 unsigned long page_no; 936 unsigned blkbits; 937 sector_t probe_block; 938 sector_t last_block; 939 int ret; 940 941 inode = sis->swap_file->f_mapping->host; 942 if (S_ISBLK(inode->i_mode)) { 943 ret = add_swap_extent(sis, 0, sis->max, 0); 944 goto done; 945 } 946 947 blkbits = inode->i_blkbits; 948 blocks_per_page = PAGE_SIZE >> blkbits; 949 950 /* 951 * Map all the blocks into the extent list. This code doesn't try 952 * to be very smart. 953 */ 954 probe_block = 0; 955 page_no = 0; 956 last_block = i_size_read(inode) >> blkbits; 957 while ((probe_block + blocks_per_page) <= last_block && 958 page_no < sis->max) { 959 unsigned block_in_page; 960 sector_t first_block; 961 962 first_block = bmap(inode, probe_block); 963 if (first_block == 0) 964 goto bad_bmap; 965 966 /* 967 * It must be PAGE_SIZE aligned on-disk 968 */ 969 if (first_block & (blocks_per_page - 1)) { 970 probe_block++; 971 goto reprobe; 972 } 973 974 for (block_in_page = 1; block_in_page < blocks_per_page; 975 block_in_page++) { 976 sector_t block; 977 978 block = bmap(inode, probe_block + block_in_page); 979 if (block == 0) 980 goto bad_bmap; 981 if (block != first_block + block_in_page) { 982 /* Discontiguity */ 983 probe_block++; 984 goto reprobe; 985 } 986 } 987 988 /* 989 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks 990 */ 991 ret = add_swap_extent(sis, page_no, 1, 992 first_block >> (PAGE_SHIFT - blkbits)); 993 if (ret) 994 goto out; 995 page_no++; 996 probe_block += blocks_per_page; 997reprobe: 998 continue; 999 } 1000 ret = 0; 1001 if (page_no == 0) 1002 page_no = 1; /* force Empty message */ 1003 sis->max = page_no; 1004 sis->pages = page_no - 1; 1005 sis->highest_bit = page_no - 1; 1006done: 1007 sis->curr_swap_extent = list_entry(sis->extent_list.prev, 1008 struct swap_extent, list); 1009 goto out; 1010bad_bmap: 1011 printk(KERN_ERR "swapon: swapfile has holes\n"); 1012 ret = -EINVAL; 1013out: 1014 return ret; 1015} 1016 1017#if 0 /* We don't need this yet */ 1018#include <linux/backing-dev.h> 1019int page_queue_congested(struct page *page) 1020{ 1021 struct backing_dev_info *bdi; 1022 1023 BUG_ON(!PageLocked(page)); /* It pins the swap_info_struct */ 1024 1025 if (PageSwapCache(page)) { 1026 swp_entry_t entry = { .val = page->private }; 1027 struct swap_info_struct *sis; 1028 1029 sis = get_swap_info_struct(swp_type(entry)); 1030 bdi = sis->bdev->bd_inode->i_mapping->backing_dev_info; 1031 } else 1032 bdi = page->mapping->backing_dev_info; 1033 return bdi_write_congested(bdi); 1034} 1035#endif 1036 1037asmlinkage long sys_swapoff(const char __user * specialfile) 1038{ 1039 struct swap_info_struct * p = NULL; 1040 unsigned short *swap_map; 1041 struct file *swap_file, *victim; 1042 struct address_space *mapping; 1043 struct inode *inode; 1044 char * pathname; 1045 int i, type, prev; 1046 int err; 1047 1048 if (!capable(CAP_SYS_ADMIN)) 1049 return -EPERM; 1050 1051 pathname = getname(specialfile); 1052 err = PTR_ERR(pathname); 1053 if (IS_ERR(pathname)) 1054 goto out; 1055 1056 victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0); 1057 putname(pathname); 1058 err = PTR_ERR(victim); 1059 if (IS_ERR(victim)) 1060 goto out; 1061 1062 mapping = victim->f_mapping; 1063 prev = -1; 1064 swap_list_lock(); 1065 for (type = swap_list.head; type >= 0; type = swap_info[type].next) { 1066 p = swap_info + type; 1067 if ((p->flags & SWP_ACTIVE) == SWP_ACTIVE) { 1068 if (p->swap_file->f_mapping == mapping) 1069 break; 1070 } 1071 prev = type; 1072 } 1073 if (type < 0) { 1074 err = -EINVAL; 1075 swap_list_unlock(); 1076 goto out_dput; 1077 } 1078 if (!security_vm_enough_memory(p->pages)) 1079 vm_unacct_memory(p->pages); 1080 else { 1081 err = -ENOMEM; 1082 swap_list_unlock(); 1083 goto out_dput; 1084 } 1085 if (prev < 0) { 1086 swap_list.head = p->next; 1087 } else { 1088 swap_info[prev].next = p->next; 1089 } 1090 if (type == swap_list.next) { 1091 /* just pick something that's safe... */ 1092 swap_list.next = swap_list.head; 1093 } 1094 nr_swap_pages -= p->pages; 1095 total_swap_pages -= p->pages; 1096 p->flags &= ~SWP_WRITEOK; 1097 swap_list_unlock(); 1098 current->flags |= PF_SWAPOFF; 1099 err = try_to_unuse(type); 1100 current->flags &= ~PF_SWAPOFF; 1101 1102 /* wait for any unplug function to finish */ 1103 down_write(&swap_unplug_sem); 1104 up_write(&swap_unplug_sem); 1105 1106 if (err) { 1107 /* re-insert swap space back into swap_list */ 1108 swap_list_lock(); 1109 for (prev = -1, i = swap_list.head; i >= 0; prev = i, i = swap_info[i].next) 1110 if (p->prio >= swap_info[i].prio) 1111 break; 1112 p->next = i; 1113 if (prev < 0) 1114 swap_list.head = swap_list.next = p - swap_info; 1115 else 1116 swap_info[prev].next = p - swap_info; 1117 nr_swap_pages += p->pages; 1118 total_swap_pages += p->pages; 1119 p->flags |= SWP_WRITEOK; 1120 swap_list_unlock(); 1121 goto out_dput; 1122 } 1123 destroy_swap_extents(p); 1124 down(&swapon_sem); 1125 swap_list_lock(); 1126 drain_mmlist(); 1127 swap_device_lock(p); 1128 swap_file = p->swap_file; 1129 p->swap_file = NULL; 1130 p->max = 0; 1131 swap_map = p->swap_map; 1132 p->swap_map = NULL; 1133 p->flags = 0; 1134 swap_device_unlock(p); 1135 swap_list_unlock(); 1136 up(&swapon_sem); 1137 vfree(swap_map); 1138 inode = mapping->host; 1139 if (S_ISBLK(inode->i_mode)) { 1140 struct block_device *bdev = I_BDEV(inode); 1141 set_blocksize(bdev, p->old_block_size); 1142 bd_release(bdev); 1143 } else { 1144 down(&inode->i_sem); 1145 inode->i_flags &= ~S_SWAPFILE; 1146 up(&inode->i_sem); 1147 } 1148 filp_close(swap_file, NULL); 1149 err = 0; 1150 1151out_dput: 1152 filp_close(victim, NULL); 1153out: 1154 return err; 1155} 1156 1157#ifdef CONFIG_PROC_FS 1158/* iterator */ 1159static void *swap_start(struct seq_file *swap, loff_t *pos) 1160{ 1161 struct swap_info_struct *ptr = swap_info; 1162 int i; 1163 loff_t l = *pos; 1164 1165 down(&swapon_sem); 1166 1167 for (i = 0; i < nr_swapfiles; i++, ptr++) { 1168 if (!(ptr->flags & SWP_USED) || !ptr->swap_map) 1169 continue; 1170 if (!l--) 1171 return ptr; 1172 } 1173 1174 return NULL; 1175} 1176 1177static void *swap_next(struct seq_file *swap, void *v, loff_t *pos) 1178{ 1179 struct swap_info_struct *ptr = v; 1180 struct swap_info_struct *endptr = swap_info + nr_swapfiles; 1181 1182 for (++ptr; ptr < endptr; ptr++) { 1183 if (!(ptr->flags & SWP_USED) || !ptr->swap_map) 1184 continue; 1185 ++*pos; 1186 return ptr; 1187 } 1188 1189 return NULL; 1190} 1191 1192static void swap_stop(struct seq_file *swap, void *v) 1193{ 1194 up(&swapon_sem); 1195} 1196 1197static int swap_show(struct seq_file *swap, void *v) 1198{ 1199 struct swap_info_struct *ptr = v; 1200 struct file *file; 1201 int len; 1202 1203 if (v == swap_info) 1204 seq_puts(swap, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n"); 1205 1206 file = ptr->swap_file; 1207 len = seq_path(swap, file->f_vfsmnt, file->f_dentry, " \t\n\\"); 1208 seq_printf(swap, "%*s%s\t%d\t%ld\t%d\n", 1209 len < 40 ? 40 - len : 1, " ", 1210 S_ISBLK(file->f_dentry->d_inode->i_mode) ? 1211 "partition" : "file\t", 1212 ptr->pages << (PAGE_SHIFT - 10), 1213 ptr->inuse_pages << (PAGE_SHIFT - 10), 1214 ptr->prio); 1215 return 0; 1216} 1217 1218static struct seq_operations swaps_op = { 1219 .start = swap_start, 1220 .next = swap_next, 1221 .stop = swap_stop, 1222 .show = swap_show 1223}; 1224 1225static int swaps_open(struct inode *inode, struct file *file) 1226{ 1227 return seq_open(file, &swaps_op); 1228} 1229 1230static struct file_operations proc_swaps_operations = { 1231 .open = swaps_open, 1232 .read = seq_read, 1233 .llseek = seq_lseek, 1234 .release = seq_release, 1235}; 1236 1237static int __init procswaps_init(void) 1238{ 1239 struct proc_dir_entry *entry; 1240 1241 entry = create_proc_entry("swaps", 0, NULL); 1242 if (entry) 1243 entry->proc_fops = &proc_swaps_operations; 1244 return 0; 1245} 1246__initcall(procswaps_init); 1247#endif /* CONFIG_PROC_FS */ 1248 1249/* 1250 * Written 01/25/92 by Simmule Turner, heavily changed by Linus. 1251 * 1252 * The swapon system call 1253 */ 1254asmlinkage long sys_swapon(const char __user * specialfile, int swap_flags) 1255{ 1256 struct swap_info_struct * p; 1257 char *name = NULL; 1258 struct block_device *bdev = NULL; 1259 struct file *swap_file = NULL; 1260 struct address_space *mapping; 1261 unsigned int type; 1262 int i, prev; 1263 int error; 1264 static int least_priority; 1265 union swap_header *swap_header = NULL; 1266 int swap_header_version; 1267 int nr_good_pages = 0; 1268 unsigned long maxpages = 1; 1269 int swapfilesize; 1270 unsigned short *swap_map; 1271 struct page *page = NULL; 1272 struct inode *inode = NULL; 1273 int did_down = 0; 1274 1275 if (!capable(CAP_SYS_ADMIN)) 1276 return -EPERM; 1277 swap_list_lock(); 1278 p = swap_info; 1279 for (type = 0 ; type < nr_swapfiles ; type++,p++) 1280 if (!(p->flags & SWP_USED)) 1281 break; 1282 error = -EPERM; 1283 /* 1284 * Test if adding another swap device is possible. There are 1285 * two limiting factors: 1) the number of bits for the swap 1286 * type swp_entry_t definition and 2) the number of bits for 1287 * the swap type in the swap ptes as defined by the different 1288 * architectures. To honor both limitations a swap entry 1289 * with swap offset 0 and swap type ~0UL is created, encoded 1290 * to a swap pte, decoded to a swp_entry_t again and finally 1291 * the swap type part is extracted. This will mask all bits 1292 * from the initial ~0UL that can't be encoded in either the 1293 * swp_entry_t or the architecture definition of a swap pte. 1294 */ 1295 if (type > swp_type(pte_to_swp_entry(swp_entry_to_pte(swp_entry(~0UL,0))))) { 1296 swap_list_unlock(); 1297 goto out; 1298 } 1299 if (type >= nr_swapfiles) 1300 nr_swapfiles = type+1; 1301 INIT_LIST_HEAD(&p->extent_list); 1302 p->flags = SWP_USED; 1303 p->nr_extents = 0; 1304 p->swap_file = NULL; 1305 p->old_block_size = 0; 1306 p->swap_map = NULL; 1307 p->lowest_bit = 0; 1308 p->highest_bit = 0; 1309 p->cluster_nr = 0; 1310 p->inuse_pages = 0; 1311 spin_lock_init(&p->sdev_lock); 1312 p->next = -1; 1313 if (swap_flags & SWAP_FLAG_PREFER) { 1314 p->prio = 1315 (swap_flags & SWAP_FLAG_PRIO_MASK)>>SWAP_FLAG_PRIO_SHIFT; 1316 } else { 1317 p->prio = --least_priority; 1318 } 1319 swap_list_unlock(); 1320 name = getname(specialfile); 1321 error = PTR_ERR(name); 1322 if (IS_ERR(name)) { 1323 name = NULL; 1324 goto bad_swap_2; 1325 } 1326 swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0); 1327 error = PTR_ERR(swap_file); 1328 if (IS_ERR(swap_file)) { 1329 swap_file = NULL; 1330 goto bad_swap_2; 1331 } 1332 1333 p->swap_file = swap_file; 1334 mapping = swap_file->f_mapping; 1335 inode = mapping->host; 1336 1337 error = -EBUSY; 1338 for (i = 0; i < nr_swapfiles; i++) { 1339 struct swap_info_struct *q = &swap_info[i]; 1340 1341 if (i == type || !q->swap_file) 1342 continue; 1343 if (mapping == q->swap_file->f_mapping) 1344 goto bad_swap; 1345 } 1346 1347 error = -EINVAL; 1348 if (S_ISBLK(inode->i_mode)) { 1349 bdev = I_BDEV(inode); 1350 error = bd_claim(bdev, sys_swapon); 1351 if (error < 0) { 1352 bdev = NULL; 1353 goto bad_swap; 1354 } 1355 p->old_block_size = block_size(bdev); 1356 error = set_blocksize(bdev, PAGE_SIZE); 1357 if (error < 0) 1358 goto bad_swap; 1359 p->bdev = bdev; 1360 } else if (S_ISREG(inode->i_mode)) { 1361 p->bdev = inode->i_sb->s_bdev; 1362 down(&inode->i_sem); 1363 did_down = 1; 1364 if (IS_SWAPFILE(inode)) { 1365 error = -EBUSY; 1366 goto bad_swap; 1367 } 1368 } else { 1369 goto bad_swap; 1370 } 1371 1372 swapfilesize = i_size_read(inode) >> PAGE_SHIFT; 1373 1374 /* 1375 * Read the swap header. 1376 */ 1377 if (!mapping->a_ops->readpage) { 1378 error = -EINVAL; 1379 goto bad_swap; 1380 } 1381 page = read_cache_page(mapping, 0, 1382 (filler_t *)mapping->a_ops->readpage, swap_file); 1383 if (IS_ERR(page)) { 1384 error = PTR_ERR(page); 1385 goto bad_swap; 1386 } 1387 wait_on_page_locked(page); 1388 if (!PageUptodate(page)) 1389 goto bad_swap; 1390 kmap(page); 1391 swap_header = page_address(page); 1392 1393 if (!memcmp("SWAP-SPACE",swap_header->magic.magic,10)) 1394 swap_header_version = 1; 1395 else if (!memcmp("SWAPSPACE2",swap_header->magic.magic,10)) 1396 swap_header_version = 2; 1397 else { 1398 printk("Unable to find swap-space signature\n"); 1399 error = -EINVAL; 1400 goto bad_swap; 1401 } 1402 1403 switch (swap_header_version) { 1404 case 1: 1405 printk(KERN_ERR "version 0 swap is no longer supported. " 1406 "Use mkswap -v1 %s\n", name); 1407 error = -EINVAL; 1408 goto bad_swap; 1409 case 2: 1410 /* Check the swap header's sub-version and the size of 1411 the swap file and bad block lists */ 1412 if (swap_header->info.version != 1) { 1413 printk(KERN_WARNING 1414 "Unable to handle swap header version %d\n", 1415 swap_header->info.version); 1416 error = -EINVAL; 1417 goto bad_swap; 1418 } 1419 1420 p->lowest_bit = 1; 1421 /* 1422 * Find out how many pages are allowed for a single swap 1423 * device. There are two limiting factors: 1) the number of 1424 * bits for the swap offset in the swp_entry_t type and 1425 * 2) the number of bits in the a swap pte as defined by 1426 * the different architectures. In order to find the 1427 * largest possible bit mask a swap entry with swap type 0 1428 * and swap offset ~0UL is created, encoded to a swap pte, 1429 * decoded to a swp_entry_t again and finally the swap 1430 * offset is extracted. This will mask all the bits from 1431 * the initial ~0UL mask that can't be encoded in either 1432 * the swp_entry_t or the architecture definition of a 1433 * swap pte. 1434 */ 1435 maxpages = swp_offset(pte_to_swp_entry(swp_entry_to_pte(swp_entry(0,~0UL)))) - 1; 1436 if (maxpages > swap_header->info.last_page) 1437 maxpages = swap_header->info.last_page; 1438 p->highest_bit = maxpages - 1; 1439 1440 error = -EINVAL; 1441 if (!maxpages) 1442 goto bad_swap; 1443 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode)) 1444 goto bad_swap; 1445 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) 1446 goto bad_swap; 1447 1448 /* OK, set up the swap map and apply the bad block list */ 1449 if (!(p->swap_map = vmalloc(maxpages * sizeof(short)))) { 1450 error = -ENOMEM; 1451 goto bad_swap; 1452 } 1453 1454 error = 0; 1455 memset(p->swap_map, 0, maxpages * sizeof(short)); 1456 for (i=0; i<swap_header->info.nr_badpages; i++) { 1457 int page = swap_header->info.badpages[i]; 1458 if (page <= 0 || page >= swap_header->info.last_page) 1459 error = -EINVAL; 1460 else 1461 p->swap_map[page] = SWAP_MAP_BAD; 1462 } 1463 nr_good_pages = swap_header->info.last_page - 1464 swap_header->info.nr_badpages - 1465 1 /* header page */; 1466 if (error) 1467 goto bad_swap; 1468 } 1469 1470 if (swapfilesize && maxpages > swapfilesize) { 1471 printk(KERN_WARNING 1472 "Swap area shorter than signature indicates\n"); 1473 error = -EINVAL; 1474 goto bad_swap; 1475 } 1476 if (nr_good_pages) { 1477 p->swap_map[0] = SWAP_MAP_BAD; 1478 p->max = maxpages; 1479 p->pages = nr_good_pages; 1480 error = setup_swap_extents(p); 1481 if (error) 1482 goto bad_swap; 1483 nr_good_pages = p->pages; 1484 } 1485 if (!nr_good_pages) { 1486 printk(KERN_WARNING "Empty swap-file\n"); 1487 error = -EINVAL; 1488 goto bad_swap; 1489 } 1490 1491 down(&swapon_sem); 1492 swap_list_lock(); 1493 swap_device_lock(p); 1494 p->flags = SWP_ACTIVE; 1495 nr_swap_pages += nr_good_pages; 1496 total_swap_pages += nr_good_pages; 1497 printk(KERN_INFO "Adding %dk swap on %s. Priority:%d extents:%d\n", 1498 nr_good_pages<<(PAGE_SHIFT-10), name, 1499 p->prio, p->nr_extents); 1500 1501 /* insert swap space into swap_list: */ 1502 prev = -1; 1503 for (i = swap_list.head; i >= 0; i = swap_info[i].next) { 1504 if (p->prio >= swap_info[i].prio) { 1505 break; 1506 } 1507 prev = i; 1508 } 1509 p->next = i; 1510 if (prev < 0) { 1511 swap_list.head = swap_list.next = p - swap_info; 1512 } else { 1513 swap_info[prev].next = p - swap_info; 1514 } 1515 swap_device_unlock(p); 1516 swap_list_unlock(); 1517 up(&swapon_sem); 1518 error = 0; 1519 goto out; 1520bad_swap: 1521 if (bdev) { 1522 set_blocksize(bdev, p->old_block_size); 1523 bd_release(bdev); 1524 } 1525 destroy_swap_extents(p); 1526bad_swap_2: 1527 swap_list_lock(); 1528 swap_map = p->swap_map; 1529 p->swap_file = NULL; 1530 p->swap_map = NULL; 1531 p->flags = 0; 1532 if (!(swap_flags & SWAP_FLAG_PREFER)) 1533 ++least_priority; 1534 swap_list_unlock(); 1535 vfree(swap_map); 1536 if (swap_file) 1537 filp_close(swap_file, NULL); 1538out: 1539 if (page && !IS_ERR(page)) { 1540 kunmap(page); 1541 page_cache_release(page); 1542 } 1543 if (name) 1544 putname(name); 1545 if (did_down) { 1546 if (!error) 1547 inode->i_flags |= S_SWAPFILE; 1548 up(&inode->i_sem); 1549 } 1550 return error; 1551} 1552 1553void si_swapinfo(struct sysinfo *val) 1554{ 1555 unsigned int i; 1556 unsigned long nr_to_be_unused = 0; 1557 1558 swap_list_lock(); 1559 for (i = 0; i < nr_swapfiles; i++) { 1560 if (!(swap_info[i].flags & SWP_USED) || 1561 (swap_info[i].flags & SWP_WRITEOK)) 1562 continue; 1563 nr_to_be_unused += swap_info[i].inuse_pages; 1564 } 1565 val->freeswap = nr_swap_pages + nr_to_be_unused; 1566 val->totalswap = total_swap_pages + nr_to_be_unused; 1567 swap_list_unlock(); 1568} 1569 1570/* 1571 * Verify that a swap entry is valid and increment its swap map count. 1572 * 1573 * Note: if swap_map[] reaches SWAP_MAP_MAX the entries are treated as 1574 * "permanent", but will be reclaimed by the next swapoff. 1575 */ 1576int swap_duplicate(swp_entry_t entry) 1577{ 1578 struct swap_info_struct * p; 1579 unsigned long offset, type; 1580 int result = 0; 1581 1582 type = swp_type(entry); 1583 if (type >= nr_swapfiles) 1584 goto bad_file; 1585 p = type + swap_info; 1586 offset = swp_offset(entry); 1587 1588 swap_device_lock(p); 1589 if (offset < p->max && p->swap_map[offset]) { 1590 if (p->swap_map[offset] < SWAP_MAP_MAX - 1) { 1591 p->swap_map[offset]++; 1592 result = 1; 1593 } else if (p->swap_map[offset] <= SWAP_MAP_MAX) { 1594 if (swap_overflow++ < 5) 1595 printk(KERN_WARNING "swap_dup: swap entry overflow\n"); 1596 p->swap_map[offset] = SWAP_MAP_MAX; 1597 result = 1; 1598 } 1599 } 1600 swap_device_unlock(p); 1601out: 1602 return result; 1603 1604bad_file: 1605 printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val); 1606 goto out; 1607} 1608 1609struct swap_info_struct * 1610get_swap_info_struct(unsigned type) 1611{ 1612 return &swap_info[type]; 1613} 1614 1615/* 1616 * swap_device_lock prevents swap_map being freed. Don't grab an extra 1617 * reference on the swaphandle, it doesn't matter if it becomes unused. 1618 */ 1619int valid_swaphandles(swp_entry_t entry, unsigned long *offset) 1620{ 1621 int ret = 0, i = 1 << page_cluster; 1622 unsigned long toff; 1623 struct swap_info_struct *swapdev = swp_type(entry) + swap_info; 1624 1625 if (!page_cluster) /* no readahead */ 1626 return 0; 1627 toff = (swp_offset(entry) >> page_cluster) << page_cluster; 1628 if (!toff) /* first page is swap header */ 1629 toff++, i--; 1630 *offset = toff; 1631 1632 swap_device_lock(swapdev); 1633 do { 1634 /* Don't read-ahead past the end of the swap area */ 1635 if (toff >= swapdev->max) 1636 break; 1637 /* Don't read in free or bad pages */ 1638 if (!swapdev->swap_map[toff]) 1639 break; 1640 if (swapdev->swap_map[toff] == SWAP_MAP_BAD) 1641 break; 1642 toff++; 1643 ret++; 1644 } while (--i); 1645 swap_device_unlock(swapdev); 1646 return ret; 1647} 1648