root/mm/mmap.c

/* */

DEFINITIONS

1. do_mmap
2. sys_mmap
3. unmap_fixup
4. sys_munmap
5. do_munmap
6. generic_mmap
7. insert_vm_struct
8. merge_segments
9. anon_map

   1 /*
   2  *      linux/mm/mmap.c
   3  *
   4  * Written by obz.
   5  */
   6 #include <linux/stat.h>
   7 #include <linux/sched.h>
   8 #include <linux/kernel.h>
   9 #include <linux/mm.h>
  10 #include <linux/shm.h>
  11 #include <linux/errno.h>
  12 #include <linux/mman.h>
  13 #include <linux/string.h>
  14 #include <linux/malloc.h>
  15 
  16 #include <asm/segment.h>
  17 #include <asm/system.h>
  18 
  19 static int anon_map(struct inode *, struct file *, struct vm_area_struct *);
  20 
  21 /*
  22  * description of effects of mapping type and prot in current implementation.
  23  * this is due to the limited x86 page protection hardware.  The expected
  24  * behavior is in parens:
  25  *
  26  * map_type     prot
  27  *              PROT_NONE       PROT_READ       PROT_WRITE      PROT_EXEC
  28  * MAP_SHARED   r: (no) no      r: (yes) yes    r: (no) yes     r: (no) yes
  29  *              w: (no) no      w: (no) no      w: (yes) yes    w: (no) no
  30  *              x: (no) no      x: (no) yes     x: (no) yes     x: (yes) yes
  31  *              
  32  * MAP_PRIVATE  r: (no) no      r: (yes) yes    r: (no) yes     r: (no) yes
  33  *              w: (no) no      w: (no) no      w: (copy) copy  w: (no) no
  34  *              x: (no) no      x: (no) yes     x: (no) yes     x: (yes) yes
  35  *
  36  */
  37 
  38 int do_mmap(struct file * file, unsigned long addr, unsigned long len,
     /*  */
  39         unsigned long prot, unsigned long flags, unsigned long off)
  40 {
  41         int mask, error;
  42         struct vm_area_struct * vma;
  43 
  44         if ((len = PAGE_ALIGN(len)) == 0)
  45                 return addr;
  46 
  47         if (addr > TASK_SIZE || len > TASK_SIZE || addr > TASK_SIZE-len)
  48                 return -EINVAL;
  49 
  50         /* offset overflow? */
  51         if (off + len < off)
  52                 return -EINVAL;
  53 
  54         /*
  55          * do simple checking here so the lower-level routines won't have
  56          * to. we assume access permissions have been handled by the open
  57          * of the memory object, so we don't do any here.
  58          */
  59 
  60         if (file != NULL) {
  61                 switch (flags & MAP_TYPE) {
  62                 case MAP_SHARED:
  63                         if ((prot & PROT_WRITE) && !(file->f_mode & 2))
  64                                 return -EACCES;
  65                         /* fall through */
  66                 case MAP_PRIVATE:
  67                         if (!(file->f_mode & 1))
  68                                 return -EACCES;
  69                         break;
  70 
  71                 default:
  72                         return -EINVAL;
  73                 }
  74         } else if ((flags & MAP_TYPE) == MAP_SHARED)
  75                 return -EINVAL;
  76 
  77         /*
  78          * obtain the address to map to. we verify (or select) it and ensure
  79          * that it represents a valid section of the address space.
  80          */
  81 
  82         if (flags & MAP_FIXED) {
  83                 if (addr & ~PAGE_MASK)
  84                         return -EINVAL;
  85                 if (len > TASK_SIZE || addr > TASK_SIZE - len)
  86                         return -EINVAL;
  87         } else {
  88                 struct vm_area_struct * vmm;
  89 
  90                 /* Maybe this works.. Ugly it is. */
  91                 addr = SHM_RANGE_START;
  92                 while (addr+len < SHM_RANGE_END) {
  93                         for (vmm = current->mm->mmap ; vmm ; vmm = vmm->vm_next) {
  94                                 if (addr >= vmm->vm_end)
  95                                         continue;
  96                                 if (addr + len <= vmm->vm_start)
  97                                         continue;
  98                                 addr = PAGE_ALIGN(vmm->vm_end);
  99                                 break;
 100                         }
 101                         if (!vmm)
 102                                 break;
 103                 }
 104                 if (addr+len >= SHM_RANGE_END)
 105                         return -ENOMEM;
 106         }
 107 
 108         /*
 109          * determine the object being mapped and call the appropriate
 110          * specific mapper. the address has already been validated, but
 111          * not unmapped, but the maps are removed from the list.
 112          */
 113         if (file && (!file->f_op || !file->f_op->mmap))
 114                 return -ENODEV;
 115         mask = PAGE_PRESENT;
 116         if (prot & (PROT_READ | PROT_EXEC))
 117                 mask |= PAGE_READONLY;
 118         if (prot & PROT_WRITE)
 119                 if ((flags & MAP_TYPE) == MAP_PRIVATE)
 120                         mask |= PAGE_COPY;
 121                 else
 122                         mask |= PAGE_SHARED;
 123 
 124         vma = kmalloc(sizeof(struct vm_area_struct), GFP_KERNEL);
 125         if (!vma)
 126                 return -ENOMEM;
 127 
 128         vma->vm_task = current;
 129         vma->vm_start = addr;
 130         vma->vm_end = addr + len;
 131         vma->vm_page_prot = mask;
 132         vma->vm_flags = prot & (VM_READ | VM_WRITE | VM_EXEC);
 133         vma->vm_flags |= flags & (VM_GROWSDOWN | VM_DENYWRITE | VM_EXECUTABLE);
 134 
 135         if (file) {
 136                 if (file->f_mode & 1)
 137                         vma->vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 138                 if (flags & MAP_SHARED) {
 139                         vma->vm_flags |= VM_SHARED | VM_MAYSHARE;
 140                         if (!(file->f_mode & 2))
 141                                 vma->vm_flags &= ~VM_MAYWRITE;
 142                 }
 143         } else
 144                 vma->vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 145         vma->vm_ops = NULL;
 146         vma->vm_offset = off;
 147         vma->vm_inode = NULL;
 148         vma->vm_pte = 0;
 149 
 150         do_munmap(addr, len);   /* Clear old maps */
 151 
 152         if (file)
 153                 error = file->f_op->mmap(file->f_inode, file, vma);
 154         else
 155                 error = anon_map(NULL, NULL, vma);
 156         
 157         if (error) {
 158                 kfree(vma);
 159                 if (!current->errno)
 160                         current->errno = -error;
 161                 return -1;
 162         }
 163         insert_vm_struct(current, vma);
 164         merge_segments(current->mm->mmap);
 165         return addr;
 166 }
 167 
 168 asmlinkage int sys_mmap(unsigned long *buffer)
     /*  */
 169 {
 170         int error;
 171         unsigned long flags;
 172         struct file * file = NULL;
 173 
 174         error = verify_area(VERIFY_READ, buffer, 6*4);
 175         if (error)
 176                 return error;
 177         flags = get_fs_long(buffer+3);
 178         if (!(flags & MAP_ANONYMOUS)) {
 179                 unsigned long fd = get_fs_long(buffer+4);
 180                 if (fd >= NR_OPEN || !(file = current->files->fd[fd]))
 181                         return -EBADF;
 182         }
 183         return do_mmap(file, get_fs_long(buffer), get_fs_long(buffer+1),
 184                 get_fs_long(buffer+2), flags, get_fs_long(buffer+5));
 185 }
 186 
 187 /*
 188  * Normal function to fix up a mapping
 189  * This function is the default for when an area has no specific
 190  * function.  This may be used as part of a more specific routine.
 191  * This function works out what part of an area is affected and
 192  * adjusts the mapping information.  Since the actual page
 193  * manipulation is done in do_mmap(), none need be done here,
 194  * though it would probably be more appropriate.
 195  *
 196  * By the time this function is called, the area struct has been
 197  * removed from the process mapping list, so it needs to be
 198  * reinserted if necessary.
 199  *
 200  * The 4 main cases are:
 201  *    Unmapping the whole area
 202  *    Unmapping from the start of the segment to a point in it
 203  *    Unmapping from an intermediate point to the end
 204  *    Unmapping between to intermediate points, making a hole.
 205  *
 206  * Case 4 involves the creation of 2 new areas, for each side of
 207  * the hole.
 208  */
 209 void unmap_fixup(struct vm_area_struct *area,
     /*  */
 210                  unsigned long addr, size_t len)
 211 {
 212         struct vm_area_struct *mpnt;
 213         unsigned long end = addr + len;
 214 
 215         if (addr < area->vm_start || addr >= area->vm_end ||
 216             end <= area->vm_start || end > area->vm_end ||
 217             end < addr)
 218         {
 219                 printk("unmap_fixup: area=%lx-%lx, unmap %lx-%lx!!\n",
 220                        area->vm_start, area->vm_end, addr, end);
 221                 return;
 222         }
 223 
 224         /* Unmapping the whole area */
 225         if (addr == area->vm_start && end == area->vm_end) {
 226                 if (area->vm_ops && area->vm_ops->close)
 227                         area->vm_ops->close(area);
 228                 if (area->vm_inode)
 229                         iput(area->vm_inode);
 230                 return;
 231         }
 232 
 233         /* Work out to one of the ends */
 234         if (addr >= area->vm_start && end == area->vm_end)
 235                 area->vm_end = addr;
 236         if (addr == area->vm_start && end <= area->vm_end) {
 237                 area->vm_offset += (end - area->vm_start);
 238                 area->vm_start = end;
 239         }
 240 
 241         /* Unmapping a hole */
 242         if (addr > area->vm_start && end < area->vm_end)
 243         {
 244                 /* Add end mapping -- leave beginning for below */
 245                 mpnt = (struct vm_area_struct *)kmalloc(sizeof(*mpnt), GFP_KERNEL);
 246 
 247                 *mpnt = *area;
 248                 mpnt->vm_offset += (end - area->vm_start);
 249                 mpnt->vm_start = end;
 250                 if (mpnt->vm_inode)
 251                         mpnt->vm_inode->i_count++;
 252                 area->vm_end = addr;    /* Truncate area */
 253                 insert_vm_struct(current, mpnt);
 254         }
 255 
 256         /* construct whatever mapping is needed */
 257         mpnt = (struct vm_area_struct *)kmalloc(sizeof(*mpnt), GFP_KERNEL);
 258         *mpnt = *area;
 259         insert_vm_struct(current, mpnt);
 260 }
 261 
 262 asmlinkage int sys_munmap(unsigned long addr, size_t len)
     /*  */
 263 {
 264         return do_munmap(addr, len);
 265 }
 266 
 267 /*
 268  * Munmap is split into 2 main parts -- this part which finds
 269  * what needs doing, and the areas themselves, which do the
 270  * work.  This now handles partial unmappings.
 271  * Jeremy Fitzhardine <jeremy@sw.oz.au>
 272  */
 273 int do_munmap(unsigned long addr, size_t len)
     /*  */
 274 {
 275         struct vm_area_struct *mpnt, **npp, *free;
 276 
 277         if ((addr & ~PAGE_MASK) || addr > TASK_SIZE || len > TASK_SIZE-addr)
 278                 return -EINVAL;
 279 
 280         if ((len = PAGE_ALIGN(len)) == 0)
 281                 return 0;
 282 
 283         /*
 284          * Check if this memory area is ok - put it on the temporary
 285          * list if so..  The checks here are pretty simple --
 286          * every area affected in some way (by any overlap) is put
 287          * on the list.  If nothing is put on, nothing is affected.
 288          */
 289         npp = &current->mm->mmap;
 290         free = NULL;
 291         for (mpnt = *npp; mpnt != NULL; mpnt = *npp) {
 292                 unsigned long end = addr+len;
 293 
 294                 if ((addr < mpnt->vm_start && end <= mpnt->vm_start) ||
 295                     (addr >= mpnt->vm_end && end > mpnt->vm_end))
 296                 {
 297                         npp = &mpnt->vm_next;
 298                         continue;
 299                 }
 300 
 301                 *npp = mpnt->vm_next;
 302                 mpnt->vm_next = free;
 303                 free = mpnt;
 304         }
 305 
 306         if (free == NULL)
 307                 return 0;
 308 
 309         /*
 310          * Ok - we have the memory areas we should free on the 'free' list,
 311          * so release them, and unmap the page range..
 312          * If the one of the segments is only being partially unmapped,
 313          * it will put new vm_area_struct(s) into the address space.
 314          */
 315         while (free) {
 316                 unsigned long st, end;
 317 
 318                 mpnt = free;
 319                 free = free->vm_next;
 320 
 321                 st = addr < mpnt->vm_start ? mpnt->vm_start : addr;
 322                 end = addr+len;
 323                 end = end > mpnt->vm_end ? mpnt->vm_end : end;
 324 
 325                 if (mpnt->vm_ops && mpnt->vm_ops->unmap)
 326                         mpnt->vm_ops->unmap(mpnt, st, end-st);
 327                 else
 328                         unmap_fixup(mpnt, st, end-st);
 329 
 330                 kfree(mpnt);
 331         }
 332 
 333         unmap_page_range(addr, len);
 334         return 0;
 335 }
 336 
 337 /* This is used for a general mmap of a disk file */
 338 int generic_mmap(struct inode * inode, struct file * file, struct vm_area_struct * vma)
     /*  */
 339 {
 340         extern struct vm_operations_struct file_mmap;
 341 
 342         if (vma->vm_page_prot & PAGE_RW)        /* only PAGE_COW or read-only supported right now */
 343                 return -EINVAL;
 344         if (vma->vm_offset & (inode->i_sb->s_blocksize - 1))
 345                 return -EINVAL;
 346         if (!inode->i_sb || !S_ISREG(inode->i_mode))
 347                 return -EACCES;
 348         if (!inode->i_op || !inode->i_op->bmap)
 349                 return -ENOEXEC;
 350         if (!IS_RDONLY(inode)) {
 351                 inode->i_atime = CURRENT_TIME;
 352                 inode->i_dirt = 1;
 353         }
 354         vma->vm_inode = inode;
 355         inode->i_count++;
 356         vma->vm_ops = &file_mmap;
 357         return 0;
 358 }
 359 
 360 /*
 361  * Insert vm structure into process list sorted by address.
 362  */
 363 void insert_vm_struct(struct task_struct *t, struct vm_area_struct *vmp)
     /*  */
 364 {
 365         struct vm_area_struct **p, *mpnt;
 366 
 367         p = &t->mm->mmap;
 368         while ((mpnt = *p) != NULL) {
 369                 if (mpnt->vm_start > vmp->vm_start)
 370                         break;
 371                 if (mpnt->vm_end > vmp->vm_start)
 372                         printk("insert_vm_struct: overlapping memory areas\n");
 373                 p = &mpnt->vm_next;
 374         }
 375         vmp->vm_next = mpnt;
 376         *p = vmp;
 377 }
 378 
 379 /*
 380  * Merge a list of memory segments if possible.
 381  * Redundant vm_area_structs are freed.
 382  * This assumes that the list is ordered by address.
 383  */
 384 void merge_segments(struct vm_area_struct *mpnt)
     /*  */
 385 {
 386         struct vm_area_struct *prev, *next;
 387 
 388         if (mpnt == NULL)
 389                 return;
 390         
 391         for(prev = mpnt, mpnt = mpnt->vm_next;
 392             mpnt != NULL;
 393             prev = mpnt, mpnt = next)
 394         {
 395                 next = mpnt->vm_next;
 396 
 397                 /*
 398                  * To share, we must have the same inode, operations.. 
 399                  */
 400                 if (mpnt->vm_inode != prev->vm_inode)
 401                         continue;
 402                 if (mpnt->vm_pte != prev->vm_pte)
 403                         continue;
 404                 if (mpnt->vm_ops != prev->vm_ops)
 405                         continue;
 406                 if (mpnt->vm_page_prot != prev->vm_page_prot ||
 407                     mpnt->vm_flags != prev->vm_flags)
 408                         continue;
 409                 if (prev->vm_end != mpnt->vm_start)
 410                         continue;
 411                 /*
 412                  * and if we have an inode, the offsets must be contiguous..
 413                  */
 414                 if (mpnt->vm_inode != NULL) {
 415                         if (prev->vm_offset + prev->vm_end - prev->vm_start != mpnt->vm_offset)
 416                                 continue;
 417                 }
 418 
 419                 /*
 420                  * merge prev with mpnt and set up pointers so the new
 421                  * big segment can possibly merge with the next one.
 422                  * The old unused mpnt is freed.
 423                  */
 424                 prev->vm_end = mpnt->vm_end;
 425                 prev->vm_next = mpnt->vm_next;
 426                 kfree_s(mpnt, sizeof(*mpnt));
 427                 mpnt = prev;
 428         }
 429 }
 430 
 431 /*
 432  * Map memory not associated with any file into a process
 433  * address space.  Adjecent memory is merged.
 434  */
 435 static int anon_map(struct inode *ino, struct file * file, struct vm_area_struct * vma)
     /*  */
 436 {
 437         if (zeromap_page_range(vma->vm_start, vma->vm_end - vma->vm_start, vma->vm_page_prot))
 438                 return -ENOMEM;
 439         return 0;
 440 }

/* */