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 current handling of page faults in memory.c. 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) yes     r: (yes) yes    r: (no) yes     r: (no) no
  30  *              w: (no) yes     w: (no) copy    w: (yes) yes    w: (no) no
  31  *              x: (no) no      x: (no) no      x: (no) no      x: (yes) no
  32  *              
  33  * MAP_PRIVATE  r: (no) yes     r: (yes) yes    r: (no) yes     r: (no) no
  34  *              w: (no) copy    w: (no) copy    w: (copy) copy  w: (no) no
  35  *              x: (no) no      x: (no) no      x: (no) no      x: (yes) no
  36  *
  37  */
  38 
  39 #define CODE_SPACE(addr)        \
  40  (PAGE_ALIGN(addr) < current->start_code + current->end_code)
  41 
  42 int do_mmap(struct file * file, unsigned long addr, unsigned long len,
     /*  */
  43         unsigned long prot, unsigned long flags, unsigned long off)
  44 {
  45         int mask, error;
  46 
  47         if ((len = PAGE_ALIGN(len)) == 0)
  48                 return addr;
  49 
  50         if (addr > TASK_SIZE || len > TASK_SIZE || addr > TASK_SIZE-len)
  51                 return -EINVAL;
  52 
  53         /*
  54          * do simple checking here so the lower-level routines won't have
  55          * to. we assume access permissions have been handled by the open
  56          * of the memory object, so we don't do any here.
  57          */
  58 
  59         if (file != NULL)
  60                 switch (flags & MAP_TYPE) {
  61                 case MAP_SHARED:
  62                         if ((prot & PROT_WRITE) && !(file->f_mode & 2))
  63                                 return -EACCES;
  64                         /* fall through */
  65                 case MAP_PRIVATE:
  66                         if (!(file->f_mode & 1))
  67                                 return -EACCES;
  68                         break;
  69 
  70                 default:
  71                         return -EINVAL;
  72                 }
  73         /*
  74          * obtain the address to map to. we verify (or select) it and ensure
  75          * that it represents a valid section of the address space.
  76          */
  77 
  78         if (flags & MAP_FIXED) {
  79                 if (addr & ~PAGE_MASK)
  80                         return -EINVAL;
  81                 if (len > TASK_SIZE || addr > TASK_SIZE - len)
  82                         return -EINVAL;
  83         } else {
  84                 struct vm_area_struct * vmm;
  85 
  86                 /* Maybe this works.. Ugly it is. */
  87                 addr = SHM_RANGE_START;
  88                 while (addr+len < SHM_RANGE_END) {
  89                         for (vmm = current->mmap ; vmm ; vmm = vmm->vm_next) {
  90                                 if (addr >= vmm->vm_end)
  91                                         continue;
  92                                 if (addr + len <= vmm->vm_start)
  93                                         continue;
  94                                 addr = PAGE_ALIGN(vmm->vm_end);
  95                                 break;
  96                         }
  97                         if (!vmm)
  98                                 break;
  99                 }
 100                 if (addr+len >= SHM_RANGE_END)
 101                         return -ENOMEM;
 102         }
 103 
 104         /*
 105          * determine the object being mapped and call the appropriate
 106          * specific mapper. the address has already been validated, but
 107          * not unmapped, but the maps are removed from the list.
 108          */
 109         if (file && (!file->f_op || !file->f_op->mmap))
 110                 return -ENODEV;
 111         mask = 0;
 112         if (prot & (PROT_READ | PROT_EXEC))
 113                 mask |= PAGE_READONLY;
 114         if (prot & PROT_WRITE)
 115                 if ((flags & MAP_TYPE) == MAP_PRIVATE)
 116                         mask |= PAGE_COW;
 117                 else
 118                         mask |= PAGE_RW;
 119         if (!mask)
 120                 return -EINVAL;
 121 
 122         do_munmap(addr, len);   /* Clear old maps */
 123 
 124         if (file)
 125                 error = file->f_op->mmap(file->f_inode, file, addr, len, mask, off);
 126         else
 127                 error = anon_map(NULL, NULL, addr, len, mask, off);
 128         
 129         if (!error)
 130                 return addr;
 131 
 132         if (!current->errno)
 133                 current->errno = -error;
 134         return -1;
 135 }
 136 
 137 asmlinkage int sys_mmap(unsigned long *buffer)
     /*  */
 138 {
 139         int error;
 140         unsigned long fd;
 141         struct file * file;
 142 
 143         error = verify_area(VERIFY_READ, buffer, 6*4);
 144         if (error)
 145                 return error;
 146         fd = get_fs_long(buffer+4);
 147         if (fd >= NR_OPEN || !(file = current->filp[fd]))
 148                 return -EBADF;
 149         return do_mmap(file, get_fs_long(buffer), get_fs_long(buffer+1),
 150                 get_fs_long(buffer+2), get_fs_long(buffer+3), 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                 return;
 195         }
 196 
 197         /* Work out to one of the ends */
 198         if (addr >= area->vm_start && end == area->vm_end)
 199                 area->vm_end = addr;
 200         if (addr == area->vm_start && end <= area->vm_end) {
 201                 area->vm_offset += (end - area->vm_start);
 202                 area->vm_start = end;
 203         }
 204 
 205         /* Unmapping a hole */
 206         if (addr > area->vm_start && end < area->vm_end)
 207         {
 208                 /* Add end mapping -- leave beginning for below */
 209                 mpnt = (struct vm_area_struct *)kmalloc(sizeof(*mpnt), GFP_KERNEL);
 210 
 211                 *mpnt = *area;
 212                 mpnt->vm_offset += (end - area->vm_start);
 213                 mpnt->vm_start = end;
 214                 if (mpnt->vm_inode)
 215                         mpnt->vm_inode->i_count++;
 216                 insert_vm_struct(current, mpnt);
 217                 area->vm_end = addr;    /* Truncate area */
 218         }
 219 
 220         /* construct whatever mapping is needed */
 221         mpnt = (struct vm_area_struct *)kmalloc(sizeof(*mpnt), GFP_KERNEL);
 222         *mpnt = *area;
 223         insert_vm_struct(current, mpnt);
 224 }
 225 
 226 
 227 asmlinkage int sys_mprotect(unsigned long addr, size_t len, unsigned long prot)
     /*  */
 228 {
 229         return -EINVAL; /* Not implemented yet */
 230 }
 231 
 232 asmlinkage int sys_munmap(unsigned long addr, size_t len)
     /*  */
 233 {
 234         return do_munmap(addr, len);
 235 }
 236 
 237 /*
 238  * Munmap is split into 2 main parts -- this part which finds
 239  * what needs doing, and the areas themselves, which do the
 240  * work.  This now handles partial unmappings.
 241  * Jeremy Fitzhardine <jeremy@sw.oz.au>
 242  */
 243 int do_munmap(unsigned long addr, size_t len)
     /*  */
 244 {
 245         struct vm_area_struct *mpnt, **npp, *free;
 246 
 247         if ((addr & ~PAGE_MASK) || addr > TASK_SIZE || len > TASK_SIZE-addr)
 248                 return -EINVAL;
 249 
 250         if ((len = PAGE_ALIGN(len)) == 0)
 251                 return 0;
 252 
 253         /*
 254          * Check if this memory area is ok - put it on the temporary
 255          * list if so..  The checks here are pretty simple --
 256          * every area affected in some way (by any overlap) is put
 257          * on the list.  If nothing is put on, nothing is affected.
 258          */
 259         npp = &current->mmap;
 260         free = NULL;
 261         for (mpnt = *npp; mpnt != NULL; mpnt = *npp) {
 262                 unsigned long end = addr+len;
 263 
 264                 if ((addr < mpnt->vm_start && end <= mpnt->vm_start) ||
 265                     (addr >= mpnt->vm_end && end > mpnt->vm_end))
 266                 {
 267                         npp = &mpnt->vm_next;
 268                         continue;
 269                 }
 270 
 271                 *npp = mpnt->vm_next;
 272                 mpnt->vm_next = free;
 273                 free = mpnt;
 274         }
 275 
 276         if (free == NULL)
 277                 return 0;
 278 
 279         /*
 280          * Ok - we have the memory areas we should free on the 'free' list,
 281          * so release them, and unmap the page range..
 282          * If the one of the segments is only being partially unmapped,
 283          * it will put new vm_area_struct(s) into the address space.
 284          */
 285         while (free) {
 286                 unsigned long st, end;
 287 
 288                 mpnt = free;
 289                 free = free->vm_next;
 290 
 291                 st = addr < mpnt->vm_start ? mpnt->vm_start : addr;
 292                 end = addr+len;
 293                 end = end > mpnt->vm_end ? mpnt->vm_end : end;
 294 
 295                 if (mpnt->vm_ops && mpnt->vm_ops->unmap)
 296                         mpnt->vm_ops->unmap(mpnt, st, end-st);
 297                 else
 298                         unmap_fixup(mpnt, st, end-st);
 299 
 300                 kfree(mpnt);
 301         }
 302 
 303         unmap_page_range(addr, len);
 304         return 0;
 305 }
 306 
 307 /* This is used for a general mmap of a disk file */
 308 int generic_mmap(struct inode * inode, struct file * file,
     /*  */
 309         unsigned long addr, size_t len, int prot, unsigned long off)
 310 {
 311         struct vm_area_struct * mpnt;
 312         extern struct vm_operations_struct file_mmap;
 313         struct buffer_head * bh;
 314 
 315         if (prot & PAGE_RW)     /* only PAGE_COW or read-only supported right now */
 316                 return -EINVAL;
 317         if (off & (inode->i_sb->s_blocksize - 1))
 318                 return -EINVAL;
 319         if (!inode->i_sb || !S_ISREG(inode->i_mode))
 320                 return -EACCES;
 321         if (!inode->i_op || !inode->i_op->bmap)
 322                 return -ENOEXEC;
 323         if (!(bh = bread(inode->i_dev,bmap(inode,0),inode->i_sb->s_blocksize)))
 324                 return -EACCES;
 325         if (!IS_RDONLY(inode)) {
 326                 inode->i_atime = CURRENT_TIME;
 327                 inode->i_dirt = 1;
 328         }
 329         brelse(bh);
 330 
 331         mpnt = (struct vm_area_struct * ) kmalloc(sizeof(struct vm_area_struct), GFP_KERNEL);
 332         if (!mpnt)
 333                 return -ENOMEM;
 334 
 335         unmap_page_range(addr, len);    
 336         mpnt->vm_task = current;
 337         mpnt->vm_start = addr;
 338         mpnt->vm_end = addr + len;
 339         mpnt->vm_page_prot = prot;
 340         mpnt->vm_share = NULL;
 341         mpnt->vm_inode = inode;
 342         inode->i_count++;
 343         mpnt->vm_offset = off;
 344         mpnt->vm_ops = &file_mmap;
 345         insert_vm_struct(current, mpnt);
 346         merge_segments(current->mmap, NULL, NULL);
 347         
 348         return 0;
 349 }
 350 
 351 /*
 352  * Insert vm structure into process list
 353  * This makes sure the list is sorted by start address, and
 354  * some some simple overlap checking.
 355  * JSGF
 356  */
 357 void insert_vm_struct(struct task_struct *t, struct vm_area_struct *vmp)
     /*  */
 358 {
 359         struct vm_area_struct **nxtpp, *mpnt;
 360 
 361         nxtpp = &t->mmap;
 362         
 363         for(mpnt = t->mmap; mpnt != NULL; mpnt = mpnt->vm_next)
 364         {
 365                 if (mpnt->vm_start > vmp->vm_start)
 366                         break;
 367                 nxtpp = &mpnt->vm_next;
 368 
 369                 if ((vmp->vm_start >= mpnt->vm_start &&
 370                      vmp->vm_start < mpnt->vm_end) ||
 371                     (vmp->vm_end >= mpnt->vm_start &&
 372                      vmp->vm_end < mpnt->vm_end))
 373                         printk("insert_vm_struct: ins area %lx-%lx in area %lx-%lx\n",
 374                                vmp->vm_start, vmp->vm_end,
 375                                mpnt->vm_start, vmp->vm_end);
 376         }
 377         
 378         vmp->vm_next = mpnt;
 379 
 380         *nxtpp = vmp;
 381 }
 382 
 383 /*
 384  * Merge a list of memory segments if possible.
 385  * Redundant vm_area_structs are freed.
 386  * This assumes that the list is ordered by address.
 387  */
 388 void merge_segments(struct vm_area_struct *mpnt,
     /*  */
 389                     map_mergep_fnp mergep, void *mpd)
 390 {
 391         struct vm_area_struct *prev, *next;
 392 
 393         if (mpnt == NULL)
 394                 return;
 395         
 396         for(prev = mpnt, mpnt = mpnt->vm_next;
 397             mpnt != NULL;
 398             prev = mpnt, mpnt = next)
 399         {
 400                 int mp;
 401 
 402                 next = mpnt->vm_next;
 403                 
 404                 if (mergep == NULL)
 405                 {
 406                         unsigned long psz = prev->vm_end - prev->vm_start;
 407                         mp = prev->vm_offset + psz == mpnt->vm_offset;
 408                 }
 409                 else
 410                         mp = (*mergep)(prev, mpnt, mpd);
 411 
 412                 /*
 413                  * Check they are compatible.
 414                  * and the like...
 415                  * What does the share pointer mean?
 416                  */
 417                 if (prev->vm_ops != mpnt->vm_ops ||
 418                     prev->vm_page_prot != mpnt->vm_page_prot ||
 419                     prev->vm_inode != mpnt->vm_inode ||
 420                     prev->vm_end != mpnt->vm_start ||
 421                     !mp ||
 422                     prev->vm_share != mpnt->vm_share ||         /* ?? */
 423                     prev->vm_next != mpnt)                      /* !!! */
 424                         continue;
 425 
 426                 /*
 427                  * merge prev with mpnt and set up pointers so the new
 428                  * big segment can possibly merge with the next one.
 429                  * The old unused mpnt is freed.
 430                  */
 431                 prev->vm_end = mpnt->vm_end;
 432                 prev->vm_next = mpnt->vm_next;
 433                 kfree_s(mpnt, sizeof(*mpnt));
 434                 mpnt = prev;
 435         }
 436 }
 437 
 438 /*
 439  * Map memory not associated with any file into a process
 440  * address space.  Adjecent memory is merged.
 441  */
 442 static int anon_map(struct inode *ino, struct file * file,
     /*  */
 443                     unsigned long addr, size_t len, int mask,
 444                     unsigned long off)
 445 {
 446         struct vm_area_struct * mpnt;
 447 
 448         if (zeromap_page_range(addr, len, mask))
 449                 return -ENOMEM;
 450 
 451         mpnt = (struct vm_area_struct * ) kmalloc(sizeof(struct vm_area_struct), GFP_KERNEL);
 452         if (!mpnt)
 453                 return -ENOMEM;
 454 
 455         mpnt->vm_task = current;
 456         mpnt->vm_start = addr;
 457         mpnt->vm_end = addr + len;
 458         mpnt->vm_page_prot = mask;
 459         mpnt->vm_share = NULL;
 460         mpnt->vm_inode = NULL;
 461         mpnt->vm_offset = 0;
 462         mpnt->vm_ops = NULL;
 463         insert_vm_struct(current, mpnt);
 464         merge_segments(current->mmap, ignoff_mergep, NULL);
 465 
 466         return 0;
 467 }
 468 
 469 /* Merge, ignoring offsets */
 470 int ignoff_mergep(const struct vm_area_struct *m1,
     /*  */
 471                   const struct vm_area_struct *m2,
 472                   void *data)
 473 {
 474         if (m1->vm_inode != m2->vm_inode)       /* Just to be sure */
 475                 return 0;
 476 
 477         return (struct inode *)data == m1->vm_inode;
 478 }

/* */