root/mm/filemap.c

/* */

DEFINITIONS

1. multi_bmap
2. filemap_nopage
3. filemap_sync_page
4. filemap_swapout
5. filemap_swapin
6. filemap_sync_pte
7. filemap_sync_pte_range
8. filemap_sync_pmd_range
9. filemap_sync
10. filemap_unmap
11. filemap_close
12. 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 inline void multi_bmap(struct inode * inode, unsigned int block, unsigned int * nr, int shift)
     /*  */
  32 {
  33         int i = PAGE_SIZE >> shift;
  34         block >>= shift;
  35         do {
  36                 *nr = bmap(inode, block);
  37                 i--;
  38                 block++;
  39                 nr++;
  40         } while (i > 0);
  41 }
  42 
  43 static unsigned long filemap_nopage(struct vm_area_struct * area, unsigned long address,
     /*  */
  44         unsigned long page, int no_share)
  45 {
  46         struct inode * inode = area->vm_inode;
  47         int nr[PAGE_SIZE/512];
  48 
  49         multi_bmap(inode, (address & PAGE_MASK) - area->vm_start + area->vm_offset, nr,
  50                 inode->i_sb->s_blocksize_bits);
  51         return bread_page(page, inode->i_dev, nr, inode->i_sb->s_blocksize, no_share);
  52 }
  53 
  54 /*
  55  * NOTE! mmap sync doesn't really work yet. This is mainly a stub for it,
  56  * which only works if the buffers and the page were already sharing the
  57  * same physical page (that's actually pretty common, especially if the
  58  * file has been mmap'ed before being read the normal way).
  59  *
  60  * Todo:
  61  * - non-shared pages also need to be synced with the buffers.
  62  * - the "swapout()" function needs to swap out the page to
  63  *   the shared file instead of using the swap device.
  64  */
  65 static void filemap_sync_page(struct vm_area_struct * vma,
     /*  */
  66         unsigned long offset,
  67         unsigned long page)
  68 {
  69         struct inode * inode;
  70         int nr[PAGE_SIZE/512];
  71         struct buffer_head * bh;
  72 
  73         bh = buffer_pages[MAP_NR(page)];
  74         if (bh) {
  75                 /* whee.. just mark the buffer heads dirty */
  76                 struct buffer_head * tmp = bh;
  77                 do {
  78                         mark_buffer_dirty(tmp, 0);
  79                         tmp = tmp->b_this_page;
  80                 } while (tmp != bh);
  81                 return;
  82         }
  83         inode = vma->vm_inode;
  84         multi_bmap(inode, offset, nr, inode->i_sb->s_blocksize_bits);
  85         bwrite_page(page, inode->i_dev, nr, inode->i_sb->s_blocksize);
  86 }
  87 
  88 /*
  89  * Swapping to a shared file: while we're busy writing out the page
  90  * (and the page still exists in memory), we save the page information
  91  * in the page table, so that "filemap_swapin()" can re-use the page
  92  * immediately if it is called while we're busy swapping it out..
  93  *
  94  * Once we've written it all out, we mark the page entry "empty", which
  95  * will result in a normal page-in (instead of a swap-in) from the now
  96  * up-to-date shared file mapping.
  97  */
  98 void filemap_swapout(struct vm_area_struct * vma,
     /*  */
  99         unsigned long offset,
 100         pte_t *page_table)
 101 {
 102         unsigned long page = pte_page(*page_table);
 103         unsigned long entry = SWP_ENTRY(SHM_SWP_TYPE, MAP_NR(page));
 104 
 105         pte_val(*page_table) = entry;
 106         filemap_sync_page(vma, offset, page);
 107         if (pte_val(*page_table) == entry)
 108                 pte_clear(page_table);
 109 }
 110 
 111 /*
 112  * filemap_swapin() is called only if we have something in the page
 113  * tables that is non-zero (but not present), which we know to be the
 114  * page index of a page that is busy being swapped out (see above).
 115  * So we just use it directly..
 116  */
 117 static pte_t filemap_swapin(struct vm_area_struct * vma,
     /*  */
 118         unsigned long offset,
 119         unsigned long entry)
 120 {
 121         unsigned long page = SWP_OFFSET(entry);
 122 
 123         mem_map[page]++;
 124         page = (page << PAGE_SHIFT) + PAGE_OFFSET;
 125         return pte_mkdirty(mk_pte(page,vma->vm_page_prot));
 126 }
 127 
 128 static inline void filemap_sync_pte(pte_t * pte, struct vm_area_struct *vma,
     /*  */
 129         unsigned long address, unsigned int flags)
 130 {
 131         pte_t page = *pte;
 132 
 133         if (!pte_present(page))
 134                 return;
 135         if (!pte_dirty(page))
 136                 return;
 137         if (flags & MS_INVALIDATE) {
 138                 pte_clear(pte);
 139         } else {
 140                 mem_map[MAP_NR(pte_page(page))]++;
 141                 *pte = pte_mkclean(page);
 142         }
 143         filemap_sync_page(vma, address - vma->vm_start, pte_page(page));
 144         free_page(pte_page(page));
 145 }
 146 
 147 static inline void filemap_sync_pte_range(pmd_t * pmd,
     /*  */
 148         unsigned long address, unsigned long size, 
 149         struct vm_area_struct *vma, unsigned long offset, unsigned int flags)
 150 {
 151         pte_t * pte;
 152         unsigned long end;
 153 
 154         if (pmd_none(*pmd))
 155                 return;
 156         if (pmd_bad(*pmd)) {
 157                 printk("filemap_sync_pte_range: bad pmd (%08lx)\n", pmd_val(*pmd));
 158                 pmd_clear(pmd);
 159                 return;
 160         }
 161         pte = pte_offset(pmd, address);
 162         offset += address & PMD_MASK;
 163         address &= ~PMD_MASK;
 164         end = address + size;
 165         if (end > PMD_SIZE)
 166                 end = PMD_SIZE;
 167         do {
 168                 filemap_sync_pte(pte, vma, address + offset, flags);
 169                 address += PAGE_SIZE;
 170                 pte++;
 171         } while (address < end);
 172 }
 173 
 174 static inline void filemap_sync_pmd_range(pgd_t * pgd,
     /*  */
 175         unsigned long address, unsigned long size, 
 176         struct vm_area_struct *vma, unsigned int flags)
 177 {
 178         pmd_t * pmd;
 179         unsigned long offset, end;
 180 
 181         if (pgd_none(*pgd))
 182                 return;
 183         if (pgd_bad(*pgd)) {
 184                 printk("filemap_sync_pmd_range: bad pgd (%08lx)\n", pgd_val(*pgd));
 185                 pgd_clear(pgd);
 186                 return;
 187         }
 188         pmd = pmd_offset(pgd, address);
 189         offset = address & PMD_MASK;
 190         address &= ~PMD_MASK;
 191         end = address + size;
 192         if (end > PGDIR_SIZE)
 193                 end = PGDIR_SIZE;
 194         do {
 195                 filemap_sync_pte_range(pmd, address, end - address, vma, offset, flags);
 196                 address = (address + PMD_SIZE) & PMD_MASK;
 197                 pmd++;
 198         } while (address < end);
 199 }
 200 
 201 static void filemap_sync(struct vm_area_struct * vma, unsigned long address,
     /*  */
 202         size_t size, unsigned int flags)
 203 {
 204         pgd_t * dir;
 205         unsigned long end = address + size;
 206 
 207         dir = pgd_offset(current, address);
 208         while (address < end) {
 209                 filemap_sync_pmd_range(dir, address, end - address, vma, flags);
 210                 address = (address + PGDIR_SIZE) & PGDIR_MASK;
 211                 dir++;
 212         }
 213         invalidate();
 214         return;
 215 }
 216 
 217 /*
 218  * This handles area unmaps..
 219  */
 220 static void filemap_unmap(struct vm_area_struct *vma, unsigned long start, size_t len)
     /*  */
 221 {
 222         filemap_sync(vma, start, len, MS_ASYNC);
 223 }
 224 
 225 /*
 226  * This handles complete area closes..
 227  */
 228 static void filemap_close(struct vm_area_struct * vma)
     /*  */
 229 {
 230         filemap_sync(vma, vma->vm_start, vma->vm_end - vma->vm_start, MS_ASYNC);
 231 }
 232 
 233 /*
 234  * Shared mappings need to be able to do the right thing at
 235  * close/unmap/sync. They will also use the private file as
 236  * backing-store for swapping..
 237  */
 238 static struct vm_operations_struct file_shared_mmap = {
 239         NULL,                   /* open */
 240         filemap_close,          /* close */
 241         filemap_unmap,          /* unmap */
 242         NULL,                   /* protect */
 243         filemap_sync,           /* sync */
 244         NULL,                   /* advise */
 245         filemap_nopage,         /* nopage */
 246         NULL,                   /* wppage */
 247         filemap_swapout,        /* swapout */
 248         filemap_swapin,         /* swapin */
 249 };
 250 
 251 /*
 252  * Private mappings just need to be able to load in the map
 253  *
 254  * (this is actually used for shared mappings as well, if we
 255  * know they can't ever get write permissions..)
 256  */
 257 static struct vm_operations_struct file_private_mmap = {
 258         NULL,                   /* open */
 259         NULL,                   /* close */
 260         NULL,                   /* unmap */
 261         NULL,                   /* protect */
 262         NULL,                   /* sync */
 263         NULL,                   /* advise */
 264         filemap_nopage,         /* nopage */
 265         NULL,                   /* wppage */
 266         NULL,                   /* swapout */
 267         NULL,                   /* swapin */
 268 };
 269 
 270 /* This is used for a general mmap of a disk file */
 271 int generic_mmap(struct inode * inode, struct file * file, struct vm_area_struct * vma)
     /*  */
 272 {
 273         struct vm_operations_struct * ops;
 274 
 275         if (vma->vm_offset & (inode->i_sb->s_blocksize - 1))
 276                 return -EINVAL;
 277         if (!inode->i_sb || !S_ISREG(inode->i_mode))
 278                 return -EACCES;
 279         if (!inode->i_op || !inode->i_op->bmap)
 280                 return -ENOEXEC;
 281         ops = &file_private_mmap;
 282         if (vma->vm_flags & VM_SHARED) {
 283                 if (vma->vm_flags & (VM_WRITE | VM_MAYWRITE))
 284                         ops = &file_shared_mmap;
 285         }
 286         if (!IS_RDONLY(inode)) {
 287                 inode->i_atime = CURRENT_TIME;
 288                 inode->i_dirt = 1;
 289         }
 290         vma->vm_inode = inode;
 291         inode->i_count++;
 292         vma->vm_ops = ops;
 293         return 0;
 294 }

/* */