root/mm/filemap.c

/* */

DEFINITIONS

1. filemap_nopage
2. filemap_sync_page
3. filemap_sync_pte
4. filemap_sync_pte_range
5. filemap_sync_pmd_range
6. filemap_sync
7. filemap_unmap
8. filemap_close
9. filemap_swapout
10. generic_mmap

   1 /*
   2  *      linux/mm/filemmap.c
   3  *
   4  * Copyright (C) 1994 Linus Torvalds
   5  */
   6 
   7 /*
   8  * This file handles the generic file mmap semantics used by
   9  * most "normal" filesystems (but you don't /have/ to use this:
  10  * the NFS filesystem does this differently, for example)
  11  */
  12 #include <linux/stat.h>
  13 #include <linux/sched.h>
  14 #include <linux/kernel.h>
  15 #include <linux/mm.h>
  16 #include <linux/shm.h>
  17 #include <linux/errno.h>
  18 #include <linux/mman.h>
  19 #include <linux/string.h>
  20 #include <linux/malloc.h>
  21 
  22 #include <asm/segment.h>
  23 #include <asm/system.h>
  24 #include <asm/pgtable.h>
  25 
  26 /*
  27  * Shared mappings implemented 30.11.1994. It's not fully working yet,
  28  * though.
  29  */
  30 
  31 static unsigned long filemap_nopage(struct vm_area_struct * area, unsigned long address,
     /*  */
  32         unsigned long page, int no_share)
  33 {
  34         struct inode * inode = area->vm_inode;
  35         unsigned int block;
  36         int nr[8];
  37         int i, *p;
  38 
  39         address &= PAGE_MASK;
  40         block = address - area->vm_start + area->vm_offset;
  41         block >>= inode->i_sb->s_blocksize_bits;
  42         i = PAGE_SIZE >> inode->i_sb->s_blocksize_bits;
  43         p = nr;
  44         do {
  45                 *p = bmap(inode,block);
  46                 i--;
  47                 block++;
  48                 p++;
  49         } while (i > 0);
  50         return bread_page(page, inode->i_dev, nr, inode->i_sb->s_blocksize, no_share);
  51 }
  52 
  53 /*
  54  * NOTE! mmap sync doesn't really work yet. This is mainly a stub for it,
  55  * which only works if the buffers and the page were already sharing the
  56  * same physical page (that's actually pretty common, especially if the
  57  * file has been mmap'ed before being read the normal way).
  58  *
  59  * Todo:
  60  * - non-shared pages also need to be synced with the buffers.
  61  * - the "swapout()" function needs to swap out the page to
  62  *   the shared file instead of using the swap device.
  63  */
  64 static void filemap_sync_page(struct vm_area_struct * vma,
     /*  */
  65         unsigned long offset,
  66         unsigned long page)
  67 {
  68         struct buffer_head * bh;
  69 
  70         printk("msync: %ld: [%08lx]\n", offset, page);
  71         bh = buffer_pages[MAP_NR(page)];
  72         if (bh) {
  73                 /* whee.. just mark the buffer heads dirty */
  74                 struct buffer_head * tmp = bh;
  75                 do {
  76                         mark_buffer_dirty(tmp, 0);
  77                         tmp = tmp->b_this_page;
  78                 } while (tmp != bh);
  79                 return;
  80         }
  81         /* we'll need to go fetch the buffer heads etc.. RSN */
  82         printk("Can't handle non-shared page yet\n");
  83         return;
  84 }
  85 
  86 static inline void filemap_sync_pte(pte_t * pte, struct vm_area_struct *vma,
     /*  */
  87         unsigned long address, unsigned int flags)
  88 {
  89         pte_t page = *pte;
  90 
  91         if (!pte_present(page))
  92                 return;
  93         if (!pte_dirty(page))
  94                 return;
  95         if (flags & MS_INVALIDATE) {
  96                 pte_clear(pte);
  97         } else {
  98                 mem_map[MAP_NR(pte_page(page))]++;
  99                 *pte = pte_mkclean(page);
 100         }
 101         filemap_sync_page(vma, address - vma->vm_start, pte_page(page));
 102         free_page(pte_page(page));
 103 }
 104 
 105 static inline void filemap_sync_pte_range(pmd_t * pmd,
     /*  */
 106         unsigned long address, unsigned long size, 
 107         struct vm_area_struct *vma, unsigned long offset, unsigned int flags)
 108 {
 109         pte_t * pte;
 110         unsigned long end;
 111 
 112         if (pmd_none(*pmd))
 113                 return;
 114         if (pmd_bad(*pmd)) {
 115                 printk("filemap_sync_pte_range: bad pmd (%08lx)\n", pmd_val(*pmd));
 116                 pmd_clear(pmd);
 117                 return;
 118         }
 119         pte = pte_offset(pmd, address);
 120         offset += address & PMD_MASK;
 121         address &= ~PMD_MASK;
 122         end = address + size;
 123         if (end > PMD_SIZE)
 124                 end = PMD_SIZE;
 125         do {
 126                 filemap_sync_pte(pte, vma, address + offset, flags);
 127                 address += PAGE_SIZE;
 128                 pte++;
 129         } while (address < end);
 130 }
 131 
 132 static inline void filemap_sync_pmd_range(pgd_t * pgd,
     /*  */
 133         unsigned long address, unsigned long size, 
 134         struct vm_area_struct *vma, unsigned int flags)
 135 {
 136         pmd_t * pmd;
 137         unsigned long offset, end;
 138 
 139         if (pgd_none(*pgd))
 140                 return;
 141         if (pgd_bad(*pgd)) {
 142                 printk("filemap_sync_pmd_range: bad pgd (%08lx)\n", pgd_val(*pgd));
 143                 pgd_clear(pgd);
 144                 return;
 145         }
 146         pmd = pmd_offset(pgd, address);
 147         offset = address & PMD_MASK;
 148         address &= ~PMD_MASK;
 149         end = address + size;
 150         if (end > PGDIR_SIZE)
 151                 end = PGDIR_SIZE;
 152         do {
 153                 filemap_sync_pte_range(pmd, address, end - address, vma, offset, flags);
 154                 address = (address + PMD_SIZE) & PMD_MASK;
 155                 pmd++;
 156         } while (address < end);
 157 }
 158 
 159 static void filemap_sync(struct vm_area_struct * vma, unsigned long address,
     /*  */
 160         size_t size, unsigned int flags)
 161 {
 162         pgd_t * dir;
 163         unsigned long end = address + size;
 164 
 165         dir = pgd_offset(current, address);
 166         while (address < end) {
 167                 filemap_sync_pmd_range(dir, address, end - address, vma, flags);
 168                 address = (address + PGDIR_SIZE) & PGDIR_MASK;
 169                 dir++;
 170         }
 171         invalidate();
 172         return;
 173 }
 174 
 175 /*
 176  * This handles area unmaps..
 177  */
 178 static void filemap_unmap(struct vm_area_struct *vma, unsigned long start, size_t len)
     /*  */
 179 {
 180         filemap_sync(vma, start, len, MS_ASYNC);
 181 }
 182 
 183 /*
 184  * This handles complete area closes..
 185  */
 186 static void filemap_close(struct vm_area_struct * vma)
     /*  */
 187 {
 188         filemap_sync(vma, vma->vm_start, vma->vm_end - vma->vm_start, MS_ASYNC);
 189 }
 190 
 191 /*
 192  * This isn't implemented yet: you'll get a warning and incorrect behaviour.
 193  *
 194  * Note that the page is free'd by the higher-level after return,
 195  * so we have to either write it out or just forget it. We currently
 196  * forget it..
 197  */
 198 void filemap_swapout(struct vm_area_struct * vma,
     /*  */
 199         unsigned long offset,
 200         pte_t *page_table)
 201 {
 202         printk("swapout not implemented on shared files..\n");
 203         pte_clear(page_table);
 204 }
 205 
 206 /*
 207  * Shared mappings need to be able to do the right thing at
 208  * close/unmap/sync. They will also use the private file as
 209  * backing-store for swapping..
 210  */
 211 static struct vm_operations_struct file_shared_mmap = {
 212         NULL,                   /* open */
 213         filemap_close,          /* close */
 214         filemap_unmap,          /* unmap */
 215         NULL,                   /* protect */
 216         filemap_sync,           /* sync */
 217         NULL,                   /* advise */
 218         filemap_nopage,         /* nopage */
 219         NULL,                   /* wppage */
 220         filemap_swapout,        /* swapout */
 221         NULL,                   /* swapin */
 222 };
 223 
 224 /*
 225  * Private mappings just need to be able to load in the map
 226  *
 227  * (this is actually used for shared mappings as well, if we
 228  * know they can't ever get write permissions..)
 229  */
 230 static struct vm_operations_struct file_private_mmap = {
 231         NULL,                   /* open */
 232         NULL,                   /* close */
 233         NULL,                   /* unmap */
 234         NULL,                   /* protect */
 235         NULL,                   /* sync */
 236         NULL,                   /* advise */
 237         filemap_nopage,         /* nopage */
 238         NULL,                   /* wppage */
 239         NULL,                   /* swapout */
 240         NULL,                   /* swapin */
 241 };
 242 
 243 /* This is used for a general mmap of a disk file */
 244 int generic_mmap(struct inode * inode, struct file * file, struct vm_area_struct * vma)
     /*  */
 245 {
 246         struct vm_operations_struct * ops;
 247 
 248         if (vma->vm_offset & (inode->i_sb->s_blocksize - 1))
 249                 return -EINVAL;
 250         if (!inode->i_sb || !S_ISREG(inode->i_mode))
 251                 return -EACCES;
 252         if (!inode->i_op || !inode->i_op->bmap)
 253                 return -ENOEXEC;
 254         ops = &file_private_mmap;
 255         if (vma->vm_flags & VM_SHARED) {
 256                 if (vma->vm_flags & (VM_WRITE | VM_MAYWRITE)) {
 257                         static int nr = 0;
 258                         ops = &file_shared_mmap;
 259 #ifndef SHARED_MMAP_REALLY_WORKS /* it doesn't, yet */
 260                         if (nr++ < 5)
 261                                 printk("%s tried to do a shared writeable mapping\n", current->comm);
 262                         return -EINVAL;
 263 #endif
 264                 }
 265         }
 266         if (!IS_RDONLY(inode)) {
 267                 inode->i_atime = CURRENT_TIME;
 268                 inode->i_dirt = 1;
 269         }
 270         vma->vm_inode = inode;
 271         inode->i_count++;
 272         vma->vm_ops = ops;
 273         return 0;
 274 }

/* */