mm/filemap.c

/* */
This source file includes following definitions.
multi_bmap
filemap_nopage
filemap_write_page
filemap_swapout
filemap_swapin
filemap_sync_pte
filemap_sync_pte_range
filemap_sync_pmd_range
filemap_sync
filemap_unmap
filemap_close
generic_mmap
msync_interval
sys_msync
   1 /*
   2  *      linux/mm/filemap.c
   3  *
   4  * Copyright (C) 1994, 1995  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 #include <linux/fs.h>
  22 #include <linux/locks.h>
  23 
  24 #include <asm/segment.h>
  25 #include <asm/system.h>
  26 #include <asm/pgtable.h>
  27 
  28 /*
  29  * Shared mappings implemented 30.11.1994. It's not fully working yet,
  30  * though.
  31  *
  32  * Shared mappings now work. 15.8.1995  Bruno.
  33  */
  34 
  35 /*
  36  * Simple routines for both non-shared and shared mappings.
  37  */
  38 
  39 static inline void multi_bmap(struct inode * inode, unsigned long block, unsigned int * nr, int shift)
     /*  */
  40 {
  41         int i = PAGE_SIZE >> shift;
  42         block >>= shift;
  43         do {
  44                 *nr = bmap(inode, block);
  45                 i--;
  46                 block++;
  47                 nr++;
  48         } while (i > 0);
  49 }
  50 
  51 /*
  52  * Semantics for shared and private memory areas are different past the end
  53  * of the file. A shared mapping past the last page of the file is an error
  54  * and results in a SIBGUS, while a private mapping just maps in a zero page.
  55  */
  56 static unsigned long filemap_nopage(struct vm_area_struct * area, unsigned long address,
     /*  */
  57         unsigned long page, int no_share)
  58 {
  59         struct inode * inode = area->vm_inode;
  60         int nr[PAGE_SIZE/512];
  61 
  62         address = (address & PAGE_MASK) - area->vm_start + area->vm_offset;
  63         if (address >= inode->i_size && (area->vm_flags & VM_SHARED) && area->vm_mm == current->mm)
  64                 send_sig(SIGBUS, current, 1);
  65         multi_bmap(inode, address, nr, inode->i_sb->s_blocksize_bits);
  66         return bread_page(page, inode->i_dev, nr, inode->i_sb->s_blocksize, no_share);
  67 }
  68 
  69 
  70 /*
  71  * Tries to write a shared mapped page to its backing store. May return -EIO
  72  * if the disk is full.
  73  */
  74 static int filemap_write_page(struct vm_area_struct * vma,
     /*  */
  75         unsigned long offset,
  76         unsigned long page)
  77 {
  78         int old_fs;
  79         unsigned long size, result;
  80         struct file file;
  81         struct inode * inode;
  82         struct buffer_head * bh;
  83 
  84         bh = buffer_pages[MAP_NR(page)];
  85         if (bh) {
  86                 /* whee.. just mark the buffer heads dirty */
  87                 struct buffer_head * tmp = bh;
  88                 do {
  89                         mark_buffer_dirty(tmp, 0);
  90                         tmp = tmp->b_this_page;
  91                 } while (tmp != bh);
  92                 return 0;
  93         }
  94 
  95         inode = vma->vm_inode;
  96         file.f_op = inode->i_op->default_file_ops;
  97         if (!file.f_op->write)
  98                 return -EIO;
  99         size = offset + PAGE_SIZE;
 100         /* refuse to extend file size.. */
 101         if (S_ISREG(inode->i_mode)) {
 102                 if (size > inode->i_size)
 103                         size = inode->i_size;
 104                 /* Ho humm.. We should have tested for this earlier */
 105                 if (size < offset)
 106                         return -EIO;
 107         }
 108         size -= offset;
 109         file.f_mode = 3;
 110         file.f_flags = 0;
 111         file.f_count = 1;
 112         file.f_inode = inode;
 113         file.f_pos = offset;
 114         file.f_reada = 0;
 115         old_fs = get_fs();
 116         set_fs(KERNEL_DS);
 117         result = file.f_op->write(inode, &file, (const char *) page, size);
 118         set_fs(old_fs);
 119         if (result != size)
 120                 return -EIO;
 121         return 0;
 122 }
 123 
 124 
 125 /*
 126  * Swapping to a shared file: while we're busy writing out the page
 127  * (and the page still exists in memory), we save the page information
 128  * in the page table, so that "filemap_swapin()" can re-use the page
 129  * immediately if it is called while we're busy swapping it out..
 130  *
 131  * Once we've written it all out, we mark the page entry "empty", which
 132  * will result in a normal page-in (instead of a swap-in) from the now
 133  * up-to-date disk file.
 134  */
 135 int filemap_swapout(struct vm_area_struct * vma,
     /*  */
 136         unsigned long offset,
 137         pte_t *page_table)
 138 {
 139         int error;
 140         unsigned long page = pte_page(*page_table);
 141         unsigned long entry = SWP_ENTRY(SHM_SWP_TYPE, MAP_NR(page));
 142 
 143         set_pte(page_table, __pte(entry));
 144         invalidate();
 145         error = filemap_write_page(vma, offset, page);
 146         if (pte_val(*page_table) == entry)
 147                 pte_clear(page_table);
 148         return error;
 149 }
 150 
 151 /*
 152  * filemap_swapin() is called only if we have something in the page
 153  * tables that is non-zero (but not present), which we know to be the
 154  * page index of a page that is busy being swapped out (see above).
 155  * So we just use it directly..
 156  */
 157 static pte_t filemap_swapin(struct vm_area_struct * vma,
     /*  */
 158         unsigned long offset,
 159         unsigned long entry)
 160 {
 161         unsigned long page = SWP_OFFSET(entry);
 162 
 163         mem_map[page].count++;
 164         page = (page << PAGE_SHIFT) + PAGE_OFFSET;
 165         return mk_pte(page,vma->vm_page_prot);
 166 }
 167 
 168 
 169 static inline int filemap_sync_pte(pte_t * ptep, struct vm_area_struct *vma,
     /*  */
 170         unsigned long address, unsigned int flags)
 171 {
 172         pte_t pte = *ptep;
 173         unsigned long page;
 174         int error;
 175 
 176         if (!(flags & MS_INVALIDATE)) {
 177                 if (!pte_present(pte))
 178                         return 0;
 179                 if (!pte_dirty(pte))
 180                         return 0;
 181                 set_pte(ptep, pte_mkclean(pte));
 182                 page = pte_page(pte);
 183                 mem_map[MAP_NR(page)].count++;
 184         } else {
 185                 if (pte_none(pte))
 186                         return 0;
 187                 pte_clear(ptep);
 188                 invalidate();
 189                 if (!pte_present(pte)) {
 190                         swap_free(pte_val(pte));
 191                         return 0;
 192                 }
 193                 page = pte_page(pte);
 194                 if (!pte_dirty(pte) || flags == MS_INVALIDATE) {
 195                         free_page(page);
 196                         return 0;
 197                 }
 198         }
 199         error = filemap_write_page(vma, address - vma->vm_start + vma->vm_offset, page);
 200         free_page(page);
 201         return error;
 202 }
 203 
 204 static inline int filemap_sync_pte_range(pmd_t * pmd,
     /*  */
 205         unsigned long address, unsigned long size, 
 206         struct vm_area_struct *vma, unsigned long offset, unsigned int flags)
 207 {
 208         pte_t * pte;
 209         unsigned long end;
 210         int error;
 211 
 212         if (pmd_none(*pmd))
 213                 return 0;
 214         if (pmd_bad(*pmd)) {
 215                 printk("filemap_sync_pte_range: bad pmd (%08lx)\n", pmd_val(*pmd));
 216                 pmd_clear(pmd);
 217                 return 0;
 218         }
 219         pte = pte_offset(pmd, address);
 220         offset += address & PMD_MASK;
 221         address &= ~PMD_MASK;
 222         end = address + size;
 223         if (end > PMD_SIZE)
 224                 end = PMD_SIZE;
 225         error = 0;
 226         do {
 227                 error |= filemap_sync_pte(pte, vma, address + offset, flags);
 228                 address += PAGE_SIZE;
 229                 pte++;
 230         } while (address < end);
 231         return error;
 232 }
 233 
 234 static inline int filemap_sync_pmd_range(pgd_t * pgd,
     /*  */
 235         unsigned long address, unsigned long size, 
 236         struct vm_area_struct *vma, unsigned int flags)
 237 {
 238         pmd_t * pmd;
 239         unsigned long offset, end;
 240         int error;
 241 
 242         if (pgd_none(*pgd))
 243                 return 0;
 244         if (pgd_bad(*pgd)) {
 245                 printk("filemap_sync_pmd_range: bad pgd (%08lx)\n", pgd_val(*pgd));
 246                 pgd_clear(pgd);
 247                 return 0;
 248         }
 249         pmd = pmd_offset(pgd, address);
 250         offset = address & PMD_MASK;
 251         address &= ~PMD_MASK;
 252         end = address + size;
 253         if (end > PGDIR_SIZE)
 254                 end = PGDIR_SIZE;
 255         error = 0;
 256         do {
 257                 error |= filemap_sync_pte_range(pmd, address, end - address, vma, offset, flags);
 258                 address = (address + PMD_SIZE) & PMD_MASK;
 259                 pmd++;
 260         } while (address < end);
 261         return error;
 262 }
 263 
 264 static int filemap_sync(struct vm_area_struct * vma, unsigned long address,
     /*  */
 265         size_t size, unsigned int flags)
 266 {
 267         pgd_t * dir;
 268         unsigned long end = address + size;
 269         int error = 0;
 270 
 271         dir = pgd_offset(current->mm, address);
 272         while (address < end) {
 273                 error |= filemap_sync_pmd_range(dir, address, end - address, vma, flags);
 274                 address = (address + PGDIR_SIZE) & PGDIR_MASK;
 275                 dir++;
 276         }
 277         invalidate();
 278         return error;
 279 }
 280 
 281 /*
 282  * This handles partial area unmaps..
 283  */
 284 static void filemap_unmap(struct vm_area_struct *vma, unsigned long start, size_t len)
     /*  */
 285 {
 286         filemap_sync(vma, start, len, MS_ASYNC);
 287 }
 288 
 289 /*
 290  * This handles complete area closes..
 291  */
 292 static void filemap_close(struct vm_area_struct * vma)
     /*  */
 293 {
 294         filemap_sync(vma, vma->vm_start, vma->vm_end - vma->vm_start, MS_ASYNC);
 295 }
 296 
 297 /*
 298  * Shared mappings need to be able to do the right thing at
 299  * close/unmap/sync. They will also use the private file as
 300  * backing-store for swapping..
 301  */
 302 static struct vm_operations_struct file_shared_mmap = {
 303         NULL,                   /* open */
 304         filemap_close,          /* close */
 305         filemap_unmap,          /* unmap */
 306         NULL,                   /* protect */
 307         filemap_sync,           /* sync */
 308         NULL,                   /* advise */
 309         filemap_nopage,         /* nopage */
 310         NULL,                   /* wppage */
 311         filemap_swapout,        /* swapout */
 312         filemap_swapin,         /* swapin */
 313 };
 314 
 315 /*
 316  * Private mappings just need to be able to load in the map.
 317  *
 318  * (This is actually used for shared mappings as well, if we
 319  * know they can't ever get write permissions..)
 320  */
 321 static struct vm_operations_struct file_private_mmap = {
 322         NULL,                   /* open */
 323         NULL,                   /* close */
 324         NULL,                   /* unmap */
 325         NULL,                   /* protect */
 326         NULL,                   /* sync */
 327         NULL,                   /* advise */
 328         filemap_nopage,         /* nopage */
 329         NULL,                   /* wppage */
 330         NULL,                   /* swapout */
 331         NULL,                   /* swapin */
 332 };
 333 
 334 /* This is used for a general mmap of a disk file */
 335 int generic_mmap(struct inode * inode, struct file * file, struct vm_area_struct * vma)
     /*  */
 336 {
 337         struct vm_operations_struct * ops;
 338 
 339         if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) {
 340                 ops = &file_shared_mmap;
 341                 /* share_page() can only guarantee proper page sharing if
 342                  * the offsets are all page aligned. */
 343                 if (vma->vm_offset & (PAGE_SIZE - 1))
 344                         return -EINVAL;
 345         } else {
 346                 ops = &file_private_mmap;
 347                 if (vma->vm_offset & (inode->i_sb->s_blocksize - 1))
 348                         return -EINVAL;
 349         }
 350         if (!inode->i_sb || !S_ISREG(inode->i_mode))
 351                 return -EACCES;
 352         if (!inode->i_op || !inode->i_op->bmap)
 353                 return -ENOEXEC;
 354         if (!IS_RDONLY(inode)) {
 355                 inode->i_atime = CURRENT_TIME;
 356                 inode->i_dirt = 1;
 357         }
 358         vma->vm_inode = inode;
 359         inode->i_count++;
 360         vma->vm_ops = ops;
 361         return 0;
 362 }
 363 
 364 
 365 /*
 366  * The msync() system call.
 367  */
 368 
 369 static int msync_interval(struct vm_area_struct * vma,
     /*  */
 370         unsigned long start, unsigned long end, int flags)
 371 {
 372         if (!vma->vm_inode)
 373                 return 0;
 374         if (vma->vm_ops->sync) {
 375                 int error;
 376                 error = vma->vm_ops->sync(vma, start, end-start, flags);
 377                 if (error)
 378                         return error;
 379                 if (flags & MS_SYNC)
 380                         return file_fsync(vma->vm_inode, NULL);
 381                 return 0;
 382         }
 383         return 0;
 384 }
 385 
 386 asmlinkage int sys_msync(unsigned long start, size_t len, int flags)
     /*  */
 387 {
 388         unsigned long end;
 389         struct vm_area_struct * vma;
 390         int unmapped_error, error;
 391 
 392         if (start & ~PAGE_MASK)
 393                 return -EINVAL;
 394         len = (len + ~PAGE_MASK) & PAGE_MASK;
 395         end = start + len;
 396         if (end < start)
 397                 return -EINVAL;
 398         if (flags & ~(MS_ASYNC | MS_INVALIDATE | MS_SYNC))
 399                 return -EINVAL;
 400         if (end == start)
 401                 return 0;
 402         /*
 403          * If the interval [start,end) covers some unmapped address ranges,
 404          * just ignore them, but return -EFAULT at the end.
 405          */
 406         vma = find_vma(current, start);
 407         unmapped_error = 0;
 408         for (;;) {
 409                 /* Still start < end. */
 410                 if (!vma)
 411                         return -EFAULT;
 412                 /* Here start < vma->vm_end. */
 413                 if (start < vma->vm_start) {
 414                         unmapped_error = -EFAULT;
 415                         start = vma->vm_start;
 416                 }
 417                 /* Here vma->vm_start <= start < vma->vm_end. */
 418                 if (end <= vma->vm_end) {
 419                         if (start < end) {
 420                                 error = msync_interval(vma, start, end, flags);
 421                                 if (error)
 422                                         return error;
 423                         }
 424                         return unmapped_error;
 425                 }
 426                 /* Here vma->vm_start <= start < vma->vm_end < end. */
 427                 error = msync_interval(vma, start, vma->vm_end, flags);
 428                 if (error)
 429                         return error;
 430                 start = vma->vm_end;
 431                 vma = vma->vm_next;
 432         }
 433 }
/* */
root/mm/filemap.c

DEFINITIONS
