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 = (struct vm_area_struct *)kmalloc(sizeof(struct vm_area_struct),
 125                 GFP_KERNEL);
 126         if (!vma)
 127                 return -ENOMEM;
 128 
 129         vma->vm_task = current;
 130         vma->vm_start = addr;
 131         vma->vm_end = addr + len;
 132         vma->vm_page_prot = mask;
 133         vma->vm_flags = prot & (VM_READ | VM_WRITE | VM_EXEC);
 134         vma->vm_flags |= flags & (VM_GROWSDOWN | VM_DENYWRITE | VM_EXECUTABLE);
 135 
 136         if (file) {
 137                 if (file->f_mode & 1)
 138                         vma->vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 139                 if (flags & MAP_SHARED) {
 140                         vma->vm_flags |= VM_SHARED | VM_MAYSHARE;
 141                         if (!(file->f_mode & 2))
 142                                 vma->vm_flags &= ~VM_MAYWRITE;
 143                 }
 144         } else
 145                 vma->vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 146         vma->vm_ops = NULL;
 147         vma->vm_offset = off;
 148         vma->vm_inode = NULL;
 149         vma->vm_pte = 0;
 150 
 151         do_munmap(addr, len);   /* Clear old maps */
 152 
 153         if (file)
 154                 error = file->f_op->mmap(file->f_inode, file, vma);
 155         else
 156                 error = anon_map(NULL, NULL, vma);
 157         
 158         if (error) {
 159                 kfree(vma);
 160                 if (!current->errno)
 161                         current->errno = -error;
 162                 return -1;
 163         }
 164         insert_vm_struct(current, vma);
 165         merge_segments(current->mm->mmap);
 166         return addr;
 167 }
 168 
 169 asmlinkage int sys_mmap(unsigned long *buffer)
     /*  */
 170 {
 171         int error;
 172         unsigned long flags;
 173         struct file * file = NULL;
 174 
 175         error = verify_area(VERIFY_READ, buffer, 6*4);
 176         if (error)
 177                 return error;
 178         flags = get_fs_long(buffer+3);
 179         if (!(flags & MAP_ANONYMOUS)) {
 180                 unsigned long fd = get_fs_long(buffer+4);
 181                 if (fd >= NR_OPEN || !(file = current->files->fd[fd]))
 182                         return -EBADF;
 183         }
 184         return do_mmap(file, get_fs_long(buffer), get_fs_long(buffer+1),
 185                 get_fs_long(buffer+2), flags, get_fs_long(buffer+5));
 186 }
 187 
 188 /*
 189  * Normal function to fix up a mapping
 190  * This function is the default for when an area has no specific
 191  * function.  This may be used as part of a more specific routine.
 192  * This function works out what part of an area is affected and
 193  * adjusts the mapping information.  Since the actual page
 194  * manipulation is done in do_mmap(), none need be done here,
 195  * though it would probably be more appropriate.
 196  *
 197  * By the time this function is called, the area struct has been
 198  * removed from the process mapping list, so it needs to be
 199  * reinserted if necessary.
 200  *
 201  * The 4 main cases are:
 202  *    Unmapping the whole area
 203  *    Unmapping from the start of the segment to a point in it
 204  *    Unmapping from an intermediate point to the end
 205  *    Unmapping between to intermediate points, making a hole.
 206  *
 207  * Case 4 involves the creation of 2 new areas, for each side of
 208  * the hole.
 209  */
 210 void unmap_fixup(struct vm_area_struct *area,
     /*  */
 211                  unsigned long addr, size_t len)
 212 {
 213         struct vm_area_struct *mpnt;
 214         unsigned long end = addr + len;
 215 
 216         if (addr < area->vm_start || addr >= area->vm_end ||
 217             end <= area->vm_start || end > area->vm_end ||
 218             end < addr)
 219         {
 220                 printk("unmap_fixup: area=%lx-%lx, unmap %lx-%lx!!\n",
 221                        area->vm_start, area->vm_end, addr, end);
 222                 return;
 223         }
 224 
 225         /* Unmapping the whole area */
 226         if (addr == area->vm_start && end == area->vm_end) {
 227                 if (area->vm_ops && area->vm_ops->close)
 228                         area->vm_ops->close(area);
 229                 if (area->vm_inode)
 230                         iput(area->vm_inode);
 231                 return;
 232         }
 233 
 234         /* Work out to one of the ends */
 235         if (addr >= area->vm_start && end == area->vm_end)
 236                 area->vm_end = addr;
 237         if (addr == area->vm_start && end <= area->vm_end) {
 238                 area->vm_offset += (end - area->vm_start);
 239                 area->vm_start = end;
 240         }
 241 
 242         /* Unmapping a hole */
 243         if (addr > area->vm_start && end < area->vm_end)
 244         {
 245                 /* Add end mapping -- leave beginning for below */
 246                 mpnt = (struct vm_area_struct *)kmalloc(sizeof(*mpnt), GFP_KERNEL);
 247 
 248                 *mpnt = *area;
 249                 mpnt->vm_offset += (end - area->vm_start);
 250                 mpnt->vm_start = end;
 251                 if (mpnt->vm_inode)
 252                         mpnt->vm_inode->i_count++;
 253                 area->vm_end = addr;    /* Truncate area */
 254                 insert_vm_struct(current, mpnt);
 255         }
 256 
 257         /* construct whatever mapping is needed */
 258         mpnt = (struct vm_area_struct *)kmalloc(sizeof(*mpnt), GFP_KERNEL);
 259         *mpnt = *area;
 260         insert_vm_struct(current, mpnt);
 261 }
 262 
 263 asmlinkage int sys_munmap(unsigned long addr, size_t len)
     /*  */
 264 {
 265         return do_munmap(addr, len);
 266 }
 267 
 268 /*
 269  * Munmap is split into 2 main parts -- this part which finds
 270  * what needs doing, and the areas themselves, which do the
 271  * work.  This now handles partial unmappings.
 272  * Jeremy Fitzhardine <jeremy@sw.oz.au>
 273  */
 274 int do_munmap(unsigned long addr, size_t len)
     /*  */
 275 {
 276         struct vm_area_struct *mpnt, **npp, *free;
 277 
 278         if ((addr & ~PAGE_MASK) || addr > TASK_SIZE || len > TASK_SIZE-addr)
 279                 return -EINVAL;
 280 
 281         if ((len = PAGE_ALIGN(len)) == 0)
 282                 return 0;
 283 
 284         /*
 285          * Check if this memory area is ok - put it on the temporary
 286          * list if so..  The checks here are pretty simple --
 287          * every area affected in some way (by any overlap) is put
 288          * on the list.  If nothing is put on, nothing is affected.
 289          */
 290         npp = &current->mm->mmap;
 291         free = NULL;
 292         for (mpnt = *npp; mpnt != NULL; mpnt = *npp) {
 293                 unsigned long end = addr+len;
 294 
 295                 if ((addr < mpnt->vm_start && end <= mpnt->vm_start) ||
 296                     (addr >= mpnt->vm_end && end > mpnt->vm_end))
 297                 {
 298                         npp = &mpnt->vm_next;
 299                         continue;
 300                 }
 301 
 302                 *npp = mpnt->vm_next;
 303                 mpnt->vm_next = free;
 304                 free = mpnt;
 305         }
 306 
 307         if (free == NULL)
 308                 return 0;
 309 
 310         /*
 311          * Ok - we have the memory areas we should free on the 'free' list,
 312          * so release them, and unmap the page range..
 313          * If the one of the segments is only being partially unmapped,
 314          * it will put new vm_area_struct(s) into the address space.
 315          */
 316         while (free) {
 317                 unsigned long st, end;
 318 
 319                 mpnt = free;
 320                 free = free->vm_next;
 321 
 322                 st = addr < mpnt->vm_start ? mpnt->vm_start : addr;
 323                 end = addr+len;
 324                 end = end > mpnt->vm_end ? mpnt->vm_end : end;
 325 
 326                 if (mpnt->vm_ops && mpnt->vm_ops->unmap)
 327                         mpnt->vm_ops->unmap(mpnt, st, end-st);
 328                 else
 329                         unmap_fixup(mpnt, st, end-st);
 330 
 331                 kfree(mpnt);
 332         }
 333 
 334         unmap_page_range(addr, len);
 335         return 0;
 336 }
 337 
 338 /* This is used for a general mmap of a disk file */
 339 int generic_mmap(struct inode * inode, struct file * file, struct vm_area_struct * vma)
     /*  */
 340 {
 341         extern struct vm_operations_struct file_mmap;
 342 
 343         if (vma->vm_page_prot & PAGE_RW)        /* only PAGE_COW or read-only supported right now */
 344                 return -EINVAL;
 345         if (vma->vm_offset & (inode->i_sb->s_blocksize - 1))
 346                 return -EINVAL;
 347         if (!inode->i_sb || !S_ISREG(inode->i_mode))
 348                 return -EACCES;
 349         if (!inode->i_op || !inode->i_op->bmap)
 350                 return -ENOEXEC;
 351         if (!IS_RDONLY(inode)) {
 352                 inode->i_atime = CURRENT_TIME;
 353                 inode->i_dirt = 1;
 354         }
 355         vma->vm_inode = inode;
 356         inode->i_count++;
 357         vma->vm_ops = &file_mmap;
 358         return 0;
 359 }
 360 
 361 /*
 362  * Insert vm structure into process list sorted by address.
 363  */
 364 void insert_vm_struct(struct task_struct *t, struct vm_area_struct *vmp)
     /*  */
 365 {
 366         struct vm_area_struct **p, *mpnt;
 367 
 368         p = &t->mm->mmap;
 369         while ((mpnt = *p) != NULL) {
 370                 if (mpnt->vm_start > vmp->vm_start)
 371                         break;
 372                 if (mpnt->vm_end > vmp->vm_start)
 373                         printk("insert_vm_struct: overlapping memory areas\n");
 374                 p = &mpnt->vm_next;
 375         }
 376         vmp->vm_next = mpnt;
 377         *p = vmp;
 378 }
 379 
 380 /*
 381  * Merge a list of memory segments if possible.
 382  * Redundant vm_area_structs are freed.
 383  * This assumes that the list is ordered by address.
 384  */
 385 void merge_segments(struct vm_area_struct *mpnt)
     /*  */
 386 {
 387         struct vm_area_struct *prev, *next;
 388 
 389         if (mpnt == NULL)
 390                 return;
 391         
 392         for(prev = mpnt, mpnt = mpnt->vm_next;
 393             mpnt != NULL;
 394             prev = mpnt, mpnt = next)
 395         {
 396                 next = mpnt->vm_next;
 397 
 398                 /*
 399                  * To share, we must have the same inode, operations.. 
 400                  */
 401                 if (mpnt->vm_inode != prev->vm_inode)
 402                         continue;
 403                 if (mpnt->vm_pte != prev->vm_pte)
 404                         continue;
 405                 if (mpnt->vm_ops != prev->vm_ops)
 406                         continue;
 407                 if (mpnt->vm_page_prot != prev->vm_page_prot ||
 408                     mpnt->vm_flags != prev->vm_flags)
 409                         continue;
 410                 if (prev->vm_end != mpnt->vm_start)
 411                         continue;
 412                 /*
 413                  * and if we have an inode, the offsets must be contiguous..
 414                  */
 415                 if (mpnt->vm_inode != NULL) {
 416                         if (prev->vm_offset + prev->vm_end - prev->vm_start != mpnt->vm_offset)
 417                                 continue;
 418                 }
 419 
 420                 /*
 421                  * merge prev with mpnt and set up pointers so the new
 422                  * big segment can possibly merge with the next one.
 423                  * The old unused mpnt is freed.
 424                  */
 425                 prev->vm_end = mpnt->vm_end;
 426                 prev->vm_next = mpnt->vm_next;
 427                 kfree_s(mpnt, sizeof(*mpnt));
 428                 mpnt = prev;
 429         }
 430 }
 431 
 432 /*
 433  * Map memory not associated with any file into a process
 434  * address space.  Adjacent memory is merged.
 435  */
 436 static int anon_map(struct inode *ino, struct file * file, struct vm_area_struct * vma)
     /*  */
 437 {
 438         if (zeromap_page_range(vma->vm_start, vma->vm_end - vma->vm_start, vma->vm_page_prot))
 439                 return -ENOMEM;
 440         return 0;
 441 }

/* */