mm/mmap.c

   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/slab.h>
  11 #include <linux/shm.h>
  12 #include <linux/errno.h>
  13 #include <linux/mman.h>
  14 #include <linux/string.h>
  15 #include <linux/pagemap.h>
  16 #include <linux/swap.h>
  17 #include <linux/smp.h>
  18 #include <linux/smp_lock.h>
  19 #include <linux/init.h>
  20 #include <linux/file.h>
  21
  22 #include <asm/uaccess.h>
  23 #include <asm/system.h>
  24 #include <asm/pgtable.h>
  25
  26 /* description of effects of mapping type and prot in current implementation.
  27  * this is due to the limited x86 page protection hardware.  The expected
  28  * behavior is in parens:
  29  *
  30  * map_type     prot
  31  *              PROT_NONE       PROT_READ       PROT_WRITE      PROT_EXEC
  32  * MAP_SHARED   r: (no) no      r: (yes) yes    r: (no) yes     r: (no) yes
  33  *              w: (no) no      w: (no) no      w: (yes) yes    w: (no) no
  34  *              x: (no) no      x: (no) yes     x: (no) yes     x: (yes) yes
  35  *
  36  * MAP_PRIVATE  r: (no) no      r: (yes) yes    r: (no) yes     r: (no) yes
  37  *              w: (no) no      w: (no) no      w: (copy) copy  w: (no) no
  38  *              x: (no) no      x: (no) yes     x: (no) yes     x: (yes) yes
  39  *
  40  */
  41 pgprot_t protection_map[16] = {
  42         __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
  43         __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
  44 };
  45
  46 /* SLAB cache for vm_area_struct's. */
  47 kmem_cache_t *vm_area_cachep;
  48
  49 int sysctl_overcommit_memory;
  50
  51 /* Check that a process has enough memory to allocate a
  52  * new virtual mapping.
  53  */
  54 int vm_enough_memory(long pages)
  55 {
  56         /* Stupid algorithm to decide if we have enough memory: while
  57          * simple, it hopefully works in most obvious cases.. Easy to
  58          * fool it, but this should catch most mistakes.
  59          */
  60         long free;
  61
  62         /* Sometimes we want to use more memory than we have. */
  63         if (sysctl_overcommit_memory)
  64             return 1;
  65
  66         free = buffermem >> PAGE_SHIFT;
  67         free += page_cache_size;
  68         free >>= 1;
  69         free += nr_free_pages;
  70         free += nr_swap_pages;
  71         free -= num_physpages >> 4;
  72         return free > pages;
  73 }
  74
  75 /* Remove one vm structure from the inode's i_mmap ring. */
  76 static inline void remove_shared_vm_struct(struct vm_area_struct *vma)
  77 {
  78         struct file * file = vma->vm_file;
  79
  80         if (file) {
  81                 if (vma->vm_flags & VM_DENYWRITE)
  82                         file->f_dentry->d_inode->i_writecount++;
  83                 if(vma->vm_next_share)
  84                         vma->vm_next_share->vm_pprev_share = vma->vm_pprev_share;
  85                 *vma->vm_pprev_share = vma->vm_next_share;
  86         }
  87 }
  88
  89 asmlinkage unsigned long sys_brk(unsigned long brk)
  90 {
  91         unsigned long rlim, retval;
  92         unsigned long newbrk, oldbrk;
  93         struct mm_struct *mm = current->mm;
  94
  95         down(&mm->mmap_sem);
  96         lock_kernel();
  97         if (brk < mm->end_code)
  98                 goto out;
  99         newbrk = PAGE_ALIGN(brk);
 100         oldbrk = PAGE_ALIGN(mm->brk);
 101         if (oldbrk == newbrk)
 102                 goto set_brk;
 103
 104         /* Always allow shrinking brk. */
 105         if (brk <= mm->brk) {
 106                 if (!do_munmap(newbrk, oldbrk-newbrk))
 107                         goto set_brk;
 108                 goto out;
 109         }
 110
 111         /* Check against rlimit and stack.. */
 112         rlim = current->rlim[RLIMIT_DATA].rlim_cur;
 113         if (rlim < RLIM_INFINITY && brk - mm->end_code > rlim)
 114                 goto out;
 115
 116         /* Check against existing mmap mappings. */
 117         if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
 118                 goto out;
 119
 120         /* Check if we have enough memory.. */
 121         if (!vm_enough_memory((newbrk-oldbrk) >> PAGE_SHIFT))
 122                 goto out;
 123
 124         /* Ok, looks good - let it rip. */
 125         if (do_mmap(NULL, oldbrk, newbrk-oldbrk,
 126                    PROT_READ|PROT_WRITE|PROT_EXEC,
 127                    MAP_FIXED|MAP_PRIVATE, 0) != oldbrk)
 128                 goto out;
 129 set_brk:
 130         mm->brk = brk;
 131 out:
 132         retval = mm->brk;
 133         unlock_kernel();
 134         up(&mm->mmap_sem);
 135         return retval;
 136 }
 137
 138 /* Combine the mmap "prot" and "flags" argument into one "vm_flags" used
 139  * internally. Essentially, translate the "PROT_xxx" and "MAP_xxx" bits
 140  * into "VM_xxx".
 141  */
 142 static inline unsigned long vm_flags(unsigned long prot, unsigned long flags)
 143 {
 144 #define _trans(x,bit1,bit2) \
 145 ((bit1==bit2)?(x&bit1):(x&bit1)?bit2:0)
 146
 147         unsigned long prot_bits, flag_bits;
 148         prot_bits =
 149                 _trans(prot, PROT_READ, VM_READ) |
 150                 _trans(prot, PROT_WRITE, VM_WRITE) |
 151                 _trans(prot, PROT_EXEC, VM_EXEC);
 152         flag_bits =
 153                 _trans(flags, MAP_GROWSDOWN, VM_GROWSDOWN) |
 154                 _trans(flags, MAP_DENYWRITE, VM_DENYWRITE) |
 155                 _trans(flags, MAP_EXECUTABLE, VM_EXECUTABLE);
 156         return prot_bits | flag_bits;
 157 #undef _trans
 158 }
 159
 160 unsigned long do_mmap(struct file * file, unsigned long addr, unsigned long len,
 161         unsigned long prot, unsigned long flags, unsigned long off)
 162 {
 163         struct mm_struct * mm = current->mm;
 164         struct vm_area_struct * vma;
 165         int correct_wcount = 0, error;
 166
 167         if ((len = PAGE_ALIGN(len)) == 0)
 168                 return addr;
 169
 170         if (len > TASK_SIZE || addr > TASK_SIZE-len)
 171                 return -EINVAL;
 172
 173         /* offset overflow? */
 174         if (off + len < off)
 175                 return -EINVAL;
 176
 177         /* Too many mappings? */
 178         if (mm->map_count > MAX_MAP_COUNT)
 179                 return -ENOMEM;
 180
 181         /* mlock MCL_FUTURE? */
 182         if (mm->def_flags & VM_LOCKED) {
 183                 unsigned long locked = mm->locked_vm << PAGE_SHIFT;
 184                 locked += len;
 185                 if (locked > current->rlim[RLIMIT_MEMLOCK].rlim_cur)
 186                         return -EAGAIN;
 187         }
 188
 189         /* Do simple checking here so the lower-level routines won't have
 190          * to. we assume access permissions have been handled by the open
 191          * of the memory object, so we don't do any here.
 192          */
 193         if (file != NULL) {
 194                 switch (flags & MAP_TYPE) {
 195                 case MAP_SHARED:
 196                         if ((prot & PROT_WRITE) && !(file->f_mode & 2))
 197                                 return -EACCES;
 198
 199                         /* Make sure we don't allow writing to an append-only file.. */
 200                         if (IS_APPEND(file->f_dentry->d_inode) && (file->f_mode & 2))
 201                                 return -EACCES;
 202
 203                         /* make sure there are no mandatory locks on the file. */
 204                         if (locks_verify_locked(file->f_dentry->d_inode))
 205                                 return -EAGAIN;
 206
 207                         /* fall through */
 208                 case MAP_PRIVATE:
 209                         if (!(file->f_mode & 1))
 210                                 return -EACCES;
 211                         break;
 212
 213                 default:
 214                         return -EINVAL;
 215                 }
 216         } else if ((flags & MAP_TYPE) != MAP_PRIVATE)
 217                 return -EINVAL;
 218
 219         /* Obtain the address to map to. we verify (or select) it and ensure
 220          * that it represents a valid section of the address space.
 221          */
 222         if (flags & MAP_FIXED) {
 223                 if (addr & ~PAGE_MASK)
 224                         return -EINVAL;
 225         } else {
 226                 addr = get_unmapped_area(addr, len);
 227                 if (!addr)
 228                         return -ENOMEM;
 229         }
 230
 231         /* Determine the object being mapped and call the appropriate
 232          * specific mapper. the address has already been validated, but
 233          * not unmapped, but the maps are removed from the list.
 234          */
 235         if (file && (!file->f_op || !file->f_op->mmap))
 236                 return -ENODEV;
 237
 238         vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 239         if (!vma)
 240                 return -ENOMEM;
 241
 242         vma->vm_mm = mm;
 243         vma->vm_start = addr;
 244         vma->vm_end = addr + len;
 245         vma->vm_flags = vm_flags(prot,flags) | mm->def_flags;
 246
 247         if (file) {
 248                 if (file->f_mode & 1)
 249                         vma->vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 250                 if (flags & MAP_SHARED) {
 251                         vma->vm_flags |= VM_SHARED | VM_MAYSHARE;
 252
 253                         /* This looks strange, but when we don't have the file open
 254                          * for writing, we can demote the shared mapping to a simpler
 255                          * private mapping. That also takes care of a security hole
 256                          * with ptrace() writing to a shared mapping without write
 257                          * permissions.
 258                          *
 259                          * We leave the VM_MAYSHARE bit on, just to get correct output
 260                          * from /proc/xxx/maps..
 261                          */
 262                         if (!(file->f_mode & 2))
 263                                 vma->vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
 264                 }
 265         } else
 266                 vma->vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 267         vma->vm_page_prot = protection_map[vma->vm_flags & 0x0f];
 268         vma->vm_ops = NULL;
 269         vma->vm_offset = off;
 270         vma->vm_file = NULL;
 271         vma->vm_pte = 0;
 272
 273         /* Clear old maps */
 274         error = -ENOMEM;
 275         if (do_munmap(addr, len))
 276                 goto free_vma;
 277
 278         /* Check against address space limit. */
 279         if ((mm->total_vm << PAGE_SHIFT) + len
 280             > current->rlim[RLIMIT_AS].rlim_cur)
 281                 goto free_vma;
 282
 283         /* Private writable mapping? Check memory availability.. */
 284         if ((vma->vm_flags & (VM_SHARED | VM_WRITE)) == VM_WRITE &&
 285             !(flags & MAP_NORESERVE)                             &&
 286             !vm_enough_memory(len >> PAGE_SHIFT))
 287                 goto free_vma;
 288
 289         error = 0;
 290         if (file) {
 291                 if (vma->vm_flags & VM_DENYWRITE) {
 292                         if (file->f_dentry->d_inode->i_writecount > 0)
 293                                 error = -ETXTBSY;
 294                         else {
 295                                 /* f_op->mmap might possibly sleep
 296                                  * (generic_file_mmap doesn't, but other code
 297                                  * might). In any case, this takes care of any
 298                                  * race that this might cause.
 299                                  */
 300                                 file->f_dentry->d_inode->i_writecount--;
 301                                 correct_wcount = 1;
 302                         }
 303                 }
 304                 if (!error)
 305                         error = file->f_op->mmap(file, vma);
 306
 307         }
 308         /* Fix up the count if necessary, then check for an error */
 309         if (correct_wcount)
 310                 file->f_dentry->d_inode->i_writecount++;
 311         if (error)
 312                 goto free_vma;
 313
 314         /*
 315          * merge_segments may merge our vma, so we can't refer to it
 316          * after the call.  Save the values we need now ...
 317          */
 318         flags = vma->vm_flags;
 319         addr = vma->vm_start; /* can addr have changed?? */
 320         insert_vm_struct(mm, vma);
 321         merge_segments(mm, vma->vm_start, vma->vm_end);
 322
 323         mm->total_vm += len >> PAGE_SHIFT;
 324         if (flags & VM_LOCKED) {
 325                 mm->locked_vm += len >> PAGE_SHIFT;
 326                 make_pages_present(addr, addr + len);
 327         }
 328         return addr;
 329
 330 free_vma:
 331         kmem_cache_free(vm_area_cachep, vma);
 332         return error;
 333 }
 334
 335 /* Get an address range which is currently unmapped.
 336  * For mmap() without MAP_FIXED and shmat() with addr=0.
 337  * Return value 0 means ENOMEM.
 338  */
 339 unsigned long get_unmapped_area(unsigned long addr, unsigned long len)
 340 {
 341         struct vm_area_struct * vmm;
 342
 343         if (len > TASK_SIZE)
 344                 return 0;
 345         if (!addr)
 346                 addr = TASK_UNMAPPED_BASE;
 347         addr = PAGE_ALIGN(addr);
 348
 349         for (vmm = find_vma(current->mm, addr); ; vmm = vmm->vm_next) {
 350                 /* At this point:  (!vmm || addr < vmm->vm_end). */
 351                 if (TASK_SIZE - len < addr)
 352                         return 0;
 353                 if (!vmm || addr + len <= vmm->vm_start)
 354                         return addr;
 355                 addr = vmm->vm_end;
 356         }
 357 }
 358
 359 /* Normal function to fix up a mapping
 360  * This function is the default for when an area has no specific
 361  * function.  This may be used as part of a more specific routine.
 362  * This function works out what part of an area is affected and
 363  * adjusts the mapping information.  Since the actual page
 364  * manipulation is done in do_mmap(), none need be done here,
 365  * though it would probably be more appropriate.
 366  *
 367  * By the time this function is called, the area struct has been
 368  * removed from the process mapping list, so it needs to be
 369  * reinserted if necessary.
 370  *
 371  * The 4 main cases are:
 372  *    Unmapping the whole area
 373  *    Unmapping from the start of the segment to a point in it
 374  *    Unmapping from an intermediate point to the end
 375  *    Unmapping between to intermediate points, making a hole.
 376  *
 377  * Case 4 involves the creation of 2 new areas, for each side of
 378  * the hole.  If possible, we reuse the existing area rather than
 379  * allocate a new one, and the return indicates whether the old
 380  * area was reused.
 381  */
 382 static int unmap_fixup(struct vm_area_struct *area, unsigned long addr,
 383                          size_t len, struct vm_area_struct **extra)
 384 {
 385         struct vm_area_struct *mpnt;
 386         unsigned long end = addr + len;
 387
 388         area->vm_mm->total_vm -= len >> PAGE_SHIFT;
 389         if (area->vm_flags & VM_LOCKED)
 390                 area->vm_mm->locked_vm -= len >> PAGE_SHIFT;
 391
 392         /* Unmapping the whole area. */
 393         if (addr == area->vm_start && end == area->vm_end) {
 394                 if (area->vm_ops && area->vm_ops->close)
 395                         area->vm_ops->close(area);
 396                 if (area->vm_file)
 397                         fput(area->vm_file);
 398                 return 0;
 399         }
 400
 401         /* Work out to one of the ends. */
 402         if (end == area->vm_end)
 403                 area->vm_end = addr;
 404         else if (addr == area->vm_start) {
 405                 area->vm_offset += (end - area->vm_start);
 406                 area->vm_start = end;
 407         } else {
 408         /* Unmapping a hole: area->vm_start < addr <= end < area->vm_end */
 409                 /* Add end mapping -- leave beginning for below */
 410                 mpnt = *extra;
 411                 *extra = NULL;
 412
 413                 mpnt->vm_mm = area->vm_mm;
 414                 mpnt->vm_start = end;
 415                 mpnt->vm_end = area->vm_end;
 416                 mpnt->vm_page_prot = area->vm_page_prot;
 417                 mpnt->vm_flags = area->vm_flags;
 418                 mpnt->vm_ops = area->vm_ops;
 419                 mpnt->vm_offset = area->vm_offset + (end - area->vm_start);
 420                 mpnt->vm_file = area->vm_file;
 421                 if (mpnt->vm_file)
 422                         mpnt->vm_file->f_count++;
 423                 if (mpnt->vm_ops && mpnt->vm_ops->open)
 424                         mpnt->vm_ops->open(mpnt);
 425                 area->vm_end = addr;    /* Truncate area */
 426                 insert_vm_struct(current->mm, mpnt);
 427         }
 428
 429         insert_vm_struct(current->mm, area);
 430         return 1;
 431 }
 432
 433 /* Munmap is split into 2 main parts -- this part which finds
 434  * what needs doing, and the areas themselves, which do the
 435  * work.  This now handles partial unmappings.
 436  * Jeremy Fitzhardine <jeremy@sw.oz.au>
 437  */
 438 int do_munmap(unsigned long addr, size_t len)
 439 {
 440         struct mm_struct * mm;
 441         struct vm_area_struct *mpnt, *free, *extra;
 442         int freed;
 443
 444         if ((addr & ~PAGE_MASK) || addr > TASK_SIZE || len > TASK_SIZE-addr)
 445                 return -EINVAL;
 446
 447         if ((len = PAGE_ALIGN(len)) == 0)
 448                 return 0;
 449
 450         /* Check if this memory area is ok - put it on the temporary
 451          * list if so..  The checks here are pretty simple --
 452          * every area affected in some way (by any overlap) is put
 453          * on the list.  If nothing is put on, nothing is affected.
 454          */
 455         mm = current->mm;
 456         mpnt = mm->mmap;
 457         while(mpnt && mpnt->vm_end <= addr)
 458                 mpnt = mpnt->vm_next;
 459         if (!mpnt)
 460                 return 0;
 461
 462         /* If we'll make "hole", check the vm areas limit */
 463         if ((mpnt->vm_start < addr && mpnt->vm_end > addr+len) &&
 464             mm->map_count > MAX_MAP_COUNT)
 465                 return -ENOMEM;
 466
 467         /*
 468          * We may need one additional vma to fix up the mappings ...
 469          * and this is the last chance for an easy error exit.
 470          */
 471         extra = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 472         if (!extra)
 473                 return -ENOMEM;
 474
 475         /* we have addr < mpnt->vm_end */
 476         free = NULL;
 477         for ( ; mpnt && mpnt->vm_start < addr+len; ) {
 478                 struct vm_area_struct *next = mpnt->vm_next;
 479
 480                 if(mpnt->vm_next)
 481                         mpnt->vm_next->vm_pprev = mpnt->vm_pprev;
 482                 *mpnt->vm_pprev = mpnt->vm_next;
 483
 484                 mpnt->vm_next = free;
 485                 free = mpnt;
 486                 mpnt = next;
 487         }
 488
 489         /* Ok - we have the memory areas we should free on the 'free' list,
 490          * so release them, and unmap the page range..
 491          * If the one of the segments is only being partially unmapped,
 492          * it will put new vm_area_struct(s) into the address space.
 493          */
 494         freed = 0;
 495         while ((mpnt = free) != NULL) {
 496                 unsigned long st, end, size;
 497
 498                 free = free->vm_next;
 499                 freed = 1;
 500
 501                 mm->map_count--;
 502                 remove_shared_vm_struct(mpnt);
 503
 504                 st = addr < mpnt->vm_start ? mpnt->vm_start : addr;
 505                 end = addr+len;
 506                 end = end > mpnt->vm_end ? mpnt->vm_end : end;
 507                 size = end - st;
 508
 509                 if (mpnt->vm_ops && mpnt->vm_ops->unmap)
 510                         mpnt->vm_ops->unmap(mpnt, st, size);
 511
 512                 flush_cache_range(mm, st, end);
 513                 zap_page_range(mm, st, size);
 514                 flush_tlb_range(mm, st, end);
 515
 516                 /*
 517                  * Fix the mapping, and free the old area if it wasn't reused.
 518                  */
 519                 if (!unmap_fixup(mpnt, st, size, &extra))
 520                         kmem_cache_free(vm_area_cachep, mpnt);
 521         }
 522
 523         /* Release the extra vma struct if it wasn't used */
 524         if (extra)
 525                 kmem_cache_free(vm_area_cachep, extra);
 526
 527         if (freed)
 528                 mm->mmap_cache = NULL;  /* Kill the cache. */
 529         return 0;
 530 }
 531
 532 asmlinkage int sys_munmap(unsigned long addr, size_t len)
 533 {
 534         int ret;
 535
 536         down(&current->mm->mmap_sem);
 537         lock_kernel();
 538         ret = do_munmap(addr, len);
 539         unlock_kernel();
 540         up(&current->mm->mmap_sem);
 541         return ret;
 542 }
 543
 544 /* Release all mmaps. */
 545 void exit_mmap(struct mm_struct * mm)
 546 {
 547         struct vm_area_struct * mpnt;
 548
 549         mpnt = mm->mmap;
 550         mm->mmap = mm->mmap_cache = NULL;
 551         mm->rss = 0;
 552         mm->total_vm = 0;
 553         mm->locked_vm = 0;
 554         while (mpnt) {
 555                 struct vm_area_struct * next = mpnt->vm_next;
 556                 unsigned long start = mpnt->vm_start;
 557                 unsigned long end = mpnt->vm_end;
 558                 unsigned long size = end - start;
 559
 560                 if (mpnt->vm_ops) {
 561                         if (mpnt->vm_ops->unmap)
 562                                 mpnt->vm_ops->unmap(mpnt, start, size);
 563                         if (mpnt->vm_ops->close)
 564                                 mpnt->vm_ops->close(mpnt);
 565                 }
 566                 mm->map_count--;
 567                 remove_shared_vm_struct(mpnt);
 568                 zap_page_range(mm, start, size);
 569                 if (mpnt->vm_file)
 570                         fput(mpnt->vm_file);
 571                 kmem_cache_free(vm_area_cachep, mpnt);
 572                 mpnt = next;
 573         }
 574
 575         /* This is just debugging */
 576         if (mm->map_count)
 577                 printk("exit_mmap: map count is %d\n", mm->map_count);
 578 }
 579
 580 /* Insert vm structure into process list sorted by address
 581  * and into the inode's i_mmap ring.
 582  */
 583 void insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vmp)
 584 {
 585         struct vm_area_struct **pprev = &mm->mmap;
 586         struct file * file;
 587
 588         mm->map_count++;
 589
 590         /* Find where to link it in. */
 591         while(*pprev && (*pprev)->vm_start <= vmp->vm_start)
 592                 pprev = &(*pprev)->vm_next;
 593
 594         /* Insert it. */
 595         if((vmp->vm_next = *pprev) != NULL)
 596                 (*pprev)->vm_pprev = &vmp->vm_next;
 597         *pprev = vmp;
 598         vmp->vm_pprev = pprev;
 599
 600         file = vmp->vm_file;
 601         if (file) {
 602                 struct inode * inode = file->f_dentry->d_inode;
 603                 if (vmp->vm_flags & VM_DENYWRITE)
 604                         inode->i_writecount--;
 605
 606                 /* insert vmp into inode's share list */
 607                 if((vmp->vm_next_share = inode->i_mmap) != NULL)
 608                         inode->i_mmap->vm_pprev_share = &vmp->vm_next_share;
 609                 inode->i_mmap = vmp;
 610                 vmp->vm_pprev_share = &inode->i_mmap;
 611         }
 612 }
 613
 614 /* Merge the list of memory segments if possible.
 615  * Redundant vm_area_structs are freed.
 616  * This assumes that the list is ordered by address.
 617  * We don't need to traverse the entire list, only those segments
 618  * which intersect or are adjacent to a given interval.
 619  *
 620  * We must already hold the mm semaphore when we get here..
 621  */
 622 void merge_segments (struct mm_struct * mm, unsigned long start_addr, unsigned long end_addr)
 623 {
 624         struct vm_area_struct *prev, *mpnt, *next;
 625
 626         prev = NULL;
 627         mpnt = mm->mmap;
 628         while(mpnt && mpnt->vm_end <= start_addr) {
 629                 prev = mpnt;
 630                 mpnt = mpnt->vm_next;
 631         }
 632         if (!mpnt)
 633                 return;
 634
 635         next = mpnt->vm_next;
 636
 637         /* we have prev->vm_next == mpnt && mpnt->vm_next = next */
 638         if (!prev) {
 639                 prev = mpnt;
 640                 mpnt = next;
 641         }
 642
 643         /* prev and mpnt cycle through the list, as long as
 644          * start_addr < mpnt->vm_end && prev->vm_start < end_addr
 645          */
 646         for ( ; mpnt && prev->vm_start < end_addr ; prev = mpnt, mpnt = next) {
 647                 next = mpnt->vm_next;
 648
 649                 /* To share, we must have the same file, operations.. */
 650                 if ((mpnt->vm_file != prev->vm_file)||
 651                     (mpnt->vm_pte != prev->vm_pte)      ||
 652                     (mpnt->vm_ops != prev->vm_ops)      ||
 653                     (mpnt->vm_flags != prev->vm_flags)  ||
 654                     (prev->vm_end != mpnt->vm_start))
 655                         continue;
 656
 657                 /*
 658                  * If we have a file or it's a shared memory area
 659                  * the offsets must be contiguous..
 660                  */
 661                 if ((mpnt->vm_file != NULL) || (mpnt->vm_flags & VM_SHM)) {
 662                         unsigned long off = prev->vm_offset+prev->vm_end-prev->vm_start;
 663                         if (off != mpnt->vm_offset)
 664                                 continue;
 665                 }
 666
 667                 /* merge prev with mpnt and set up pointers so the new
 668                  * big segment can possibly merge with the next one.
 669                  * The old unused mpnt is freed.
 670                  */
 671                 if(mpnt->vm_next)
 672                         mpnt->vm_next->vm_pprev = mpnt->vm_pprev;
 673                 *mpnt->vm_pprev = mpnt->vm_next;
 674
 675                 prev->vm_end = mpnt->vm_end;
 676                 if (mpnt->vm_ops && mpnt->vm_ops->close) {
 677                         mpnt->vm_offset += mpnt->vm_end - mpnt->vm_start;
 678                         mpnt->vm_start = mpnt->vm_end;
 679                         mpnt->vm_ops->close(mpnt);
 680                 }
 681                 mm->map_count--;
 682                 remove_shared_vm_struct(mpnt);
 683                 if (mpnt->vm_file)
 684                         fput(mpnt->vm_file);
 685                 kmem_cache_free(vm_area_cachep, mpnt);
 686                 mpnt = prev;
 687         }
 688         mm->mmap_cache = NULL;          /* Kill the cache. */
 689 }
 690
 691 void __init vma_init(void)
 692 {
 693         vm_area_cachep = kmem_cache_create("vm_area_struct",
 694                                            sizeof(struct vm_area_struct),
 695                                            0, SLAB_HWCACHE_ALIGN,
 696                                            NULL, NULL);
 697         if(!vm_area_cachep)
 698                 panic("vma_init: Cannot alloc vm_area_struct cache.");
 699
 700         mm_cachep = kmem_cache_create("mm_struct",
 701                                       sizeof(struct mm_struct),
 702                                       0, SLAB_HWCACHE_ALIGN,
 703                                       NULL, NULL);
 704         if(!mm_cachep)
 705                 panic("vma_init: Cannot alloc mm_struct cache.");
 706 }