mm/page_io.c

   1 /*
   2  *  linux/mm/page_io.c
   3  *
   4  *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
   5  *
   6  *  Swap reorganised 29.12.95,
   7  *  Asynchronous swapping added 30.12.95. Stephen Tweedie
   8  *  Removed race in async swapping. 14.4.1996. Bruno Haible
   9  *  Add swap of shared pages through the page cache. 20.2.1998. Stephen Tweedie
  10  *  Always use brw_page, life becomes simpler. 12 May 1998 Eric Biederman
  11  */
  12
  13 #include <linux/mm.h>
  14 #include <linux/kernel_stat.h>
  15 #include <linux/swap.h>
  16 #include <linux/locks.h>
  17 #include <linux/swapctl.h>
  18
  19 #include <asm/pgtable.h>
  20
  21 /*
  22  * Reads or writes a swap page.
  23  * wait=1: start I/O and wait for completion. wait=0: start asynchronous I/O.
  24  *
  25  * Important prevention of race condition: the caller *must* atomically
  26  * create a unique swap cache entry for this swap page before calling
  27  * rw_swap_page, and must lock that page.  By ensuring that there is a
  28  * single page of memory reserved for the swap entry, the normal VM page
  29  * lock on that page also doubles as a lock on swap entries.  Having only
  30  * one lock to deal with per swap entry (rather than locking swap and memory
  31  * independently) also makes it easier to make certain swapping operations
  32  * atomic, which is particularly important when we are trying to ensure
  33  * that shared pages stay shared while being swapped.
  34  */
  35
  36 static int rw_swap_page_base(int rw, swp_entry_t entry, struct page *page, int wait)
  37 {
  38         unsigned long offset;
  39         int zones[PAGE_SIZE/512];
  40         int zones_used;
  41         kdev_t dev = 0;
  42         int block_size;
  43         struct inode *swapf = 0;
  44
  45         /* Don't allow too many pending pages in flight.. */
  46         if (atomic_read(&nr_async_pages) > pager_daemon.swap_cluster)
  47                 wait = 1;
  48
  49         if (rw == READ) {
  50                 ClearPageUptodate(page);
  51                 kstat.pswpin++;
  52         } else
  53                 kstat.pswpout++;
  54
  55         get_swaphandle_info(entry, &offset, &dev, &swapf);
  56         if (dev) {
  57                 zones[0] = offset;
  58                 zones_used = 1;
  59                 block_size = PAGE_SIZE;
  60         } else if (swapf) {
  61                 int i, j;
  62                 unsigned int block = offset
  63                         << (PAGE_SHIFT - swapf->i_sb->s_blocksize_bits);
  64
  65                 block_size = swapf->i_sb->s_blocksize;
  66                 for (i=0, j=0; j< PAGE_SIZE ; i++, j += block_size)
  67                         if (!(zones[i] = bmap(swapf,block++))) {
  68                                 printk("rw_swap_page: bad swap file\n");
  69                                 return 0;
  70                         }
  71                 zones_used = i;
  72                 dev = swapf->i_dev;
  73         } else {
  74                 return 0;
  75         }
  76         if (!wait) {
  77                 SetPageDecrAfter(page);
  78                 atomic_inc(&nr_async_pages);
  79         }
  80
  81         /* block_size == PAGE_SIZE/zones_used */
  82         brw_page(rw, page, dev, zones, block_size);
  83
  84         /* Note! For consistency we do all of the logic,
  85          * decrementing the page count, and unlocking the page in the
  86          * swap lock map - in the IO completion handler.
  87          */
  88         if (!wait)
  89                 return 1;
  90
  91         wait_on_page(page);
  92         /* This shouldn't happen, but check to be sure. */
  93         if (page_count(page) == 0)
  94                 printk(KERN_ERR "rw_swap_page: page unused while waiting!\n");
  95
  96         return 1;
  97 }
  98
  99 /*
 100  * A simple wrapper so the base function doesn't need to enforce
 101  * that all swap pages go through the swap cache! We verify that:
 102  *  - the page is locked
 103  *  - it's marked as being swap-cache
 104  *  - it's associated with the swap inode
 105  */
 106 void rw_swap_page(int rw, struct page *page, int wait)
 107 {
 108         swp_entry_t entry;
 109
 110         entry.val = page->index;
 111
 112         if (!PageLocked(page))
 113                 PAGE_BUG(page);
 114         if (!PageSwapCache(page))
 115                 PAGE_BUG(page);
 116         if (page->mapping != &swapper_space)
 117                 PAGE_BUG(page);
 118         if (!rw_swap_page_base(rw, entry, page, wait))
 119                 UnlockPage(page);
 120 }
 121
 122 /*
 123  * The swap lock map insists that pages be in the page cache!
 124  * Therefore we can't use it.  Later when we can remove the need for the
 125  * lock map and we can reduce the number of functions exported.
 126  */
 127 void rw_swap_page_nolock(int rw, swp_entry_t entry, char *buf, int wait)
 128 {
 129         struct page *page = virt_to_page(buf);
 130
 131         if (!PageLocked(page))
 132                 PAGE_BUG(page);
 133         if (PageSwapCache(page))
 134                 PAGE_BUG(page);
 135         if (page->mapping)
 136                 PAGE_BUG(page);
 137         /* needs sync_page to wait I/O completation */
 138         page->mapping = &swapper_space;
 139         if (!rw_swap_page_base(rw, entry, page, wait))
 140                 UnlockPage(page);
 141         page->mapping = NULL;
 142 }