root/mm/mmap.c

/* */

DEFINITIONS

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

   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 *,
  20                     unsigned long, size_t, int,
  21                     unsigned long);
  22 /*
  23  * description of effects of mapping type and prot in current implementation.
  24  * this is due to the limited x86 page protection hardware.  The expected
  25  * behavior is in parens:
  26  *
  27  * map_type     prot
  28  *              PROT_NONE       PROT_READ       PROT_WRITE      PROT_EXEC
  29  * MAP_SHARED   r: (no) no      r: (yes) yes    r: (no) yes     r: (no) yes
  30  *              w: (no) no      w: (no) no      w: (yes) yes    w: (no) no
  31  *              x: (no) no      x: (no) yes     x: (no) yes     x: (yes) yes
  32  *              
  33  * MAP_PRIVATE  r: (no) no      r: (yes) yes    r: (no) yes     r: (no) yes
  34  *              w: (no) no      w: (no) no      w: (copy) copy  w: (no) no
  35  *              x: (no) no      x: (no) yes     x: (no) yes     x: (yes) yes
  36  *
  37  */
  38 
  39 int do_mmap(struct file * file, unsigned long addr, unsigned long len,
     /*  */
  40         unsigned long prot, unsigned long flags, unsigned long off)
  41 {
  42         int mask, error;
  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         /*
  51          * do simple checking here so the lower-level routines won't have
  52          * to. we assume access permissions have been handled by the open
  53          * of the memory object, so we don't do any here.
  54          */
  55 
  56         if (file != NULL)
  57                 switch (flags & MAP_TYPE) {
  58                 case MAP_SHARED:
  59                         if ((prot & PROT_WRITE) && !(file->f_mode & 2))
  60                                 return -EACCES;
  61                         /* fall through */
  62                 case MAP_PRIVATE:
  63                         if (!(file->f_mode & 1))
  64                                 return -EACCES;
  65                         break;
  66 
  67                 default:
  68                         return -EINVAL;
  69                 }
  70         /*
  71          * obtain the address to map to. we verify (or select) it and ensure
  72          * that it represents a valid section of the address space.
  73          */
  74 
  75         if (flags & MAP_FIXED) {
  76                 if (addr & ~PAGE_MASK)
  77                         return -EINVAL;
  78                 if (len > TASK_SIZE || addr > TASK_SIZE - len)
  79                         return -EINVAL;
  80         } else {
  81                 struct vm_area_struct * vmm;
  82 
  83                 /* Maybe this works.. Ugly it is. */
  84                 addr = SHM_RANGE_START;
  85                 while (addr+len < SHM_RANGE_END) {
  86                         for (vmm = current->mm->mmap ; vmm ; vmm = vmm->vm_next) {
  87                                 if (addr >= vmm->vm_end)
  88                                         continue;
  89                                 if (addr + len <= vmm->vm_start)
  90                                         continue;
  91                                 addr = PAGE_ALIGN(vmm->vm_end);
  92                                 break;
  93                         }
  94                         if (!vmm)
  95                                 break;
  96                 }
  97                 if (addr+len >= SHM_RANGE_END)
  98                         return -ENOMEM;
  99         }
 100 
 101         /*
 102          * determine the object being mapped and call the appropriate
 103          * specific mapper. the address has already been validated, but
 104          * not unmapped, but the maps are removed from the list.
 105          */
 106         if (file && (!file->f_op || !file->f_op->mmap))
 107                 return -ENODEV;
 108         mask = 0;
 109         if (prot & (PROT_READ | PROT_EXEC))
 110                 mask |= PAGE_READONLY;
 111         if (prot & PROT_WRITE)
 112                 if ((flags & MAP_TYPE) == MAP_PRIVATE)
 113                         mask |= PAGE_COPY;
 114                 else
 115                         mask |= PAGE_SHARED;
 116         if (!mask) /* PROT_NONE */
 117                 mask = PAGE_PRESENT;    /* none of PAGE_USER, PAGE_RW, PAGE_COW */
 118 
 119         do_munmap(addr, len);   /* Clear old maps */
 120 
 121         if (file)
 122                 error = file->f_op->mmap(file->f_inode, file, addr, len, mask, off);
 123         else
 124                 error = anon_map(NULL, NULL, addr, len, mask, off);
 125         
 126         if (!error)
 127                 return addr;
 128 
 129         if (!current->errno)
 130                 current->errno = -error;
 131         return -1;
 132 }
 133 
 134 asmlinkage int sys_mmap(unsigned long *buffer)
     /*  */
 135 {
 136         int error;
 137         unsigned long flags;
 138         struct file * file = NULL;
 139 
 140         error = verify_area(VERIFY_READ, buffer, 6*4);
 141         if (error)
 142                 return error;
 143         flags = get_fs_long(buffer+3);
 144         if (!(flags & MAP_ANONYMOUS)) {
 145                 unsigned long fd = get_fs_long(buffer+4);
 146                 if (fd >= NR_OPEN || !(file = current->files->fd[fd]))
 147                         return -EBADF;
 148         }
 149         return do_mmap(file, get_fs_long(buffer), get_fs_long(buffer+1),
 150                 get_fs_long(buffer+2), flags, get_fs_long(buffer+5));
 151 }
 152 
 153 /*
 154  * Normal function to fix up a mapping
 155  * This function is the default for when an area has no specific
 156  * function.  This may be used as part of a more specific routine.
 157  * This function works out what part of an area is affected and
 158  * adjusts the mapping information.  Since the actual page
 159  * manipulation is done in do_mmap(), none need be done here,
 160  * though it would probably be more appropriate.
 161  *
 162  * By the time this function is called, the area struct has been
 163  * removed from the process mapping list, so it needs to be
 164  * reinserted if necessary.
 165  *
 166  * The 4 main cases are:
 167  *    Unmapping the whole area
 168  *    Unmapping from the start of the segment to a point in it
 169  *    Unmapping from an intermediate point to the end
 170  *    Unmapping between to intermediate points, making a hole.
 171  *
 172  * Case 4 involves the creation of 2 new areas, for each side of
 173  * the hole.
 174  */
 175 void unmap_fixup(struct vm_area_struct *area,
     /*  */
 176                  unsigned long addr, size_t len)
 177 {
 178         struct vm_area_struct *mpnt;
 179         unsigned long end = addr + len;
 180 
 181         if (addr < area->vm_start || addr >= area->vm_end ||
 182             end <= area->vm_start || end > area->vm_end ||
 183             end < addr)
 184         {
 185                 printk("unmap_fixup: area=%lx-%lx, unmap %lx-%lx!!\n",
 186                        area->vm_start, area->vm_end, addr, end);
 187                 return;
 188         }
 189 
 190         /* Unmapping the whole area */
 191         if (addr == area->vm_start && end == area->vm_end) {
 192                 if (area->vm_ops && area->vm_ops->close)
 193                         area->vm_ops->close(area);
 194                 if (area->vm_inode)
 195                         iput(area->vm_inode);
 196                 return;
 197         }
 198 
 199         /* Work out to one of the ends */
 200         if (addr >= area->vm_start && end == area->vm_end)
 201                 area->vm_end = addr;
 202         if (addr == area->vm_start && end <= area->vm_end) {
 203                 area->vm_offset += (end - area->vm_start);
 204                 area->vm_start = end;
 205         }
 206 
 207         /* Unmapping a hole */
 208         if (addr > area->vm_start && end < area->vm_end)
 209         {
 210                 /* Add end mapping -- leave beginning for below */
 211                 mpnt = (struct vm_area_struct *)kmalloc(sizeof(*mpnt), GFP_KERNEL);
 212 
 213                 *mpnt = *area;
 214                 mpnt->vm_offset += (end - area->vm_start);
 215                 mpnt->vm_start = end;
 216                 if (mpnt->vm_inode)
 217                         mpnt->vm_inode->i_count++;
 218                 area->vm_end = addr;    /* Truncate area */
 219                 insert_vm_struct(current, mpnt);
 220         }
 221 
 222         /* construct whatever mapping is needed */
 223         mpnt = (struct vm_area_struct *)kmalloc(sizeof(*mpnt), GFP_KERNEL);
 224         *mpnt = *area;
 225         insert_vm_struct(current, mpnt);
 226 }
 227 
 228 
 229 asmlinkage int sys_mprotect(unsigned long addr, size_t len, unsigned long prot)
     /*  */
 230 {
 231         return -EINVAL; /* Not implemented yet */
 232 }
 233 
 234 asmlinkage int sys_munmap(unsigned long addr, size_t len)
     /*  */
 235 {
 236         return do_munmap(addr, len);
 237 }
 238 
 239 /*
 240  * Munmap is split into 2 main parts -- this part which finds
 241  * what needs doing, and the areas themselves, which do the
 242  * work.  This now handles partial unmappings.
 243  * Jeremy Fitzhardine <jeremy@sw.oz.au>
 244  */
 245 int do_munmap(unsigned long addr, size_t len)
     /*  */
 246 {
 247         struct vm_area_struct *mpnt, **npp, *free;
 248 
 249         if ((addr & ~PAGE_MASK) || addr > TASK_SIZE || len > TASK_SIZE-addr)
 250                 return -EINVAL;
 251 
 252         if ((len = PAGE_ALIGN(len)) == 0)
 253                 return 0;
 254 
 255         /*
 256          * Check if this memory area is ok - put it on the temporary
 257          * list if so..  The checks here are pretty simple --
 258          * every area affected in some way (by any overlap) is put
 259          * on the list.  If nothing is put on, nothing is affected.
 260          */
 261         npp = &current->mm->mmap;
 262         free = NULL;
 263         for (mpnt = *npp; mpnt != NULL; mpnt = *npp) {
 264                 unsigned long end = addr+len;
 265 
 266                 if ((addr < mpnt->vm_start && end <= mpnt->vm_start) ||
 267                     (addr >= mpnt->vm_end && end > mpnt->vm_end))
 268                 {
 269                         npp = &mpnt->vm_next;
 270                         continue;
 271                 }
 272 
 273                 *npp = mpnt->vm_next;
 274                 mpnt->vm_next = free;
 275                 free = mpnt;
 276         }
 277 
 278         if (free == NULL)
 279                 return 0;
 280 
 281         /*
 282          * Ok - we have the memory areas we should free on the 'free' list,
 283          * so release them, and unmap the page range..
 284          * If the one of the segments is only being partially unmapped,
 285          * it will put new vm_area_struct(s) into the address space.
 286          */
 287         while (free) {
 288                 unsigned long st, end;
 289 
 290                 mpnt = free;
 291                 free = free->vm_next;
 292 
 293                 st = addr < mpnt->vm_start ? mpnt->vm_start : addr;
 294                 end = addr+len;
 295                 end = end > mpnt->vm_end ? mpnt->vm_end : end;
 296 
 297                 if (mpnt->vm_ops && mpnt->vm_ops->unmap)
 298                         mpnt->vm_ops->unmap(mpnt, st, end-st);
 299                 else
 300                         unmap_fixup(mpnt, st, end-st);
 301 
 302                 kfree(mpnt);
 303         }
 304 
 305         unmap_page_range(addr, len);
 306         return 0;
 307 }
 308 
 309 /* This is used for a general mmap of a disk file */
 310 int generic_mmap(struct inode * inode, struct file * file,
     /*  */
 311         unsigned long addr, size_t len, int prot, unsigned long off)
 312 {
 313         struct vm_area_struct * mpnt;
 314         extern struct vm_operations_struct file_mmap;
 315         struct buffer_head * bh;
 316 
 317         if (prot & PAGE_RW)     /* only PAGE_COW or read-only supported right now */
 318                 return -EINVAL;
 319         if (off & (inode->i_sb->s_blocksize - 1))
 320                 return -EINVAL;
 321         if (!inode->i_sb || !S_ISREG(inode->i_mode))
 322                 return -EACCES;
 323         if (!inode->i_op || !inode->i_op->bmap)
 324                 return -ENOEXEC;
 325         if (!(bh = bread(inode->i_dev,bmap(inode,0),inode->i_sb->s_blocksize)))
 326                 return -EACCES;
 327         if (!IS_RDONLY(inode)) {
 328                 inode->i_atime = CURRENT_TIME;
 329                 inode->i_dirt = 1;
 330         }
 331         brelse(bh);
 332 
 333         mpnt = (struct vm_area_struct * ) kmalloc(sizeof(struct vm_area_struct), GFP_KERNEL);
 334         if (!mpnt)
 335                 return -ENOMEM;
 336 
 337         unmap_page_range(addr, len);    
 338         mpnt->vm_task = current;
 339         mpnt->vm_start = addr;
 340         mpnt->vm_end = addr + len;
 341         mpnt->vm_page_prot = prot;
 342         mpnt->vm_flags = 0;
 343         mpnt->vm_share = NULL;
 344         mpnt->vm_inode = inode;
 345         inode->i_count++;
 346         mpnt->vm_offset = off;
 347         mpnt->vm_ops = &file_mmap;
 348         insert_vm_struct(current, mpnt);
 349         merge_segments(current->mm->mmap, NULL, NULL);
 350         
 351         return 0;
 352 }
 353 
 354 /*
 355  * Insert vm structure into process list
 356  * This makes sure the list is sorted by start address, and
 357  * some some simple overlap checking.
 358  * JSGF
 359  */
 360 void insert_vm_struct(struct task_struct *t, struct vm_area_struct *vmp)
     /*  */
 361 {
 362         struct vm_area_struct **nxtpp, *mpnt;
 363 
 364         nxtpp = &t->mm->mmap;
 365         
 366         for(mpnt = t->mm->mmap; mpnt != NULL; mpnt = mpnt->vm_next)
 367         {
 368                 if (mpnt->vm_start > vmp->vm_start)
 369                         break;
 370                 nxtpp = &mpnt->vm_next;
 371 
 372                 if ((vmp->vm_start >= mpnt->vm_start &&
 373                      vmp->vm_start < mpnt->vm_end) ||
 374                     (vmp->vm_end >= mpnt->vm_start &&
 375                      vmp->vm_end < mpnt->vm_end))
 376                         printk("insert_vm_struct: ins area %lx-%lx in area %lx-%lx\n",
 377                                vmp->vm_start, vmp->vm_end,
 378                                mpnt->vm_start, vmp->vm_end);
 379         }
 380         
 381         vmp->vm_next = mpnt;
 382 
 383         *nxtpp = vmp;
 384 }
 385 
 386 /*
 387  * Merge a list of memory segments if possible.
 388  * Redundant vm_area_structs are freed.
 389  * This assumes that the list is ordered by address.
 390  */
 391 void merge_segments(struct vm_area_struct *mpnt,
     /*  */
 392                     map_mergep_fnp mergep, void *mpd)
 393 {
 394         struct vm_area_struct *prev, *next;
 395 
 396         if (mpnt == NULL)
 397                 return;
 398         
 399         for(prev = mpnt, mpnt = mpnt->vm_next;
 400             mpnt != NULL;
 401             prev = mpnt, mpnt = next)
 402         {
 403                 int mp;
 404 
 405                 next = mpnt->vm_next;
 406                 
 407                 if (mergep == NULL)
 408                 {
 409                         unsigned long psz = prev->vm_end - prev->vm_start;
 410                         mp = prev->vm_offset + psz == mpnt->vm_offset;
 411                 }
 412                 else
 413                         mp = (*mergep)(prev, mpnt, mpd);
 414 
 415                 /*
 416                  * Check they are compatible.
 417                  * and the like...
 418                  * What does the share pointer mean?
 419                  */
 420                 if (prev->vm_ops != mpnt->vm_ops ||
 421                     prev->vm_page_prot != mpnt->vm_page_prot ||
 422                     prev->vm_inode != mpnt->vm_inode ||
 423                     prev->vm_end != mpnt->vm_start ||
 424                     !mp ||
 425                     prev->vm_flags != mpnt->vm_flags ||
 426                     prev->vm_share != mpnt->vm_share ||         /* ?? */
 427                     prev->vm_next != mpnt)                      /* !!! */
 428                         continue;
 429 
 430                 /*
 431                  * merge prev with mpnt and set up pointers so the new
 432                  * big segment can possibly merge with the next one.
 433                  * The old unused mpnt is freed.
 434                  */
 435                 prev->vm_end = mpnt->vm_end;
 436                 prev->vm_next = mpnt->vm_next;
 437                 kfree_s(mpnt, sizeof(*mpnt));
 438                 mpnt = prev;
 439         }
 440 }
 441 
 442 /*
 443  * Map memory not associated with any file into a process
 444  * address space.  Adjecent memory is merged.
 445  */
 446 static int anon_map(struct inode *ino, struct file * file,
     /*  */
 447                     unsigned long addr, size_t len, int mask,
 448                     unsigned long off)
 449 {
 450         struct vm_area_struct * mpnt;
 451 
 452         if (zeromap_page_range(addr, len, mask))
 453                 return -ENOMEM;
 454 
 455         mpnt = (struct vm_area_struct * ) kmalloc(sizeof(struct vm_area_struct), GFP_KERNEL);
 456         if (!mpnt)
 457                 return -ENOMEM;
 458 
 459         mpnt->vm_task = current;
 460         mpnt->vm_start = addr;
 461         mpnt->vm_end = addr + len;
 462         mpnt->vm_page_prot = mask;
 463         mpnt->vm_flags = 0;
 464         mpnt->vm_share = NULL;
 465         mpnt->vm_inode = NULL;
 466         mpnt->vm_offset = 0;
 467         mpnt->vm_ops = NULL;
 468         insert_vm_struct(current, mpnt);
 469         merge_segments(current->mm->mmap, ignoff_mergep, NULL);
 470 
 471         return 0;
 472 }
 473 
 474 /* Merge, ignoring offsets */
 475 int ignoff_mergep(const struct vm_area_struct *m1,
     /*  */
 476                   const struct vm_area_struct *m2,
 477                   void *data)
 478 {
 479         if (m1->vm_inode != m2->vm_inode)       /* Just to be sure */
 480                 return 0;
 481 
 482         return (struct inode *)data == m1->vm_inode;
 483 }

/* */