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 static inline 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 freepages;
  61
  62         /* Sometimes we want to use more memory than we have. */
  63         if (sysctl_overcommit_memory)
  64             return 1;
  65
  66         freepages = buffermem >> PAGE_SHIFT;
  67         freepages += page_cache_size;
  68         freepages >>= 1;
  69         freepages += nr_free_pages;
  70         freepages += nr_swap_pages;
  71         freepages -= num_physpages >> 4;
  72         return freepages > 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         lock_kernel();
  96         if (brk < mm->end_code)
  97                 goto out;
  98         newbrk = PAGE_ALIGN(brk);
  99         oldbrk = PAGE_ALIGN(mm->brk);
 100         if (oldbrk == newbrk)
 101                 goto set_brk;
 102
 103         /* Always allow shrinking brk. */
 104         if (brk <= mm->brk) {
 105                 if (!do_munmap(newbrk, oldbrk-newbrk))
 106                         goto set_brk;
 107                 goto out;
 108         }
 109
 110         /* Check against rlimit and stack.. */
 111         rlim = current->rlim[RLIMIT_DATA].rlim_cur;
 112         if (rlim >= RLIM_INFINITY)
 113                 rlim = ~0;
 114         if (brk - mm->end_code > rlim)
 115                 goto out;
 116
 117         /* Check against existing mmap mappings. */
 118         if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
 119                 goto out;
 120
 121         /* Check if we have enough memory.. */
 122         if (!vm_enough_memory((newbrk-oldbrk) >> PAGE_SHIFT))
 123                 goto out;
 124
 125         /* Ok, looks good - let it rip. */
 126         if (do_mmap(NULL, oldbrk, newbrk-oldbrk,
 127                    PROT_READ|PROT_WRITE|PROT_EXEC,
 128                    MAP_FIXED|MAP_PRIVATE, 0) != oldbrk)
 129                 goto out;
 130 set_brk:
 131         mm->brk = brk;
 132 out:
 133         retval = mm->brk;
 134         unlock_kernel();
 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 there are no mandatory locks on the file. */
 200                         if (locks_verify_locked(file->f_dentry->d_inode))
 201                                 return -EAGAIN;
 202                         /* fall through */
 203                 case MAP_PRIVATE:
 204                         if (!(file->f_mode & 1))
 205                                 return -EACCES;
 206                         break;
 207
 208                 default:
 209                         return -EINVAL;
 210                 }
 211         } else if ((flags & MAP_TYPE) != MAP_PRIVATE)
 212                 return -EINVAL;
 213
 214         /* Obtain the address to map to. we verify (or select) it and ensure
 215          * that it represents a valid section of the address space.
 216          */
 217         if (flags & MAP_FIXED) {
 218                 if (addr & ~PAGE_MASK)
 219                         return -EINVAL;
 220         } else {
 221                 addr = get_unmapped_area(addr, len);
 222                 if (!addr)
 223                         return -ENOMEM;
 224         }
 225
 226         /* Determine the object being mapped and call the appropriate
 227          * specific mapper. the address has already been validated, but
 228          * not unmapped, but the maps are removed from the list.
 229          */
 230         if (file && (!file->f_op || !file->f_op->mmap))
 231                 return -ENODEV;
 232
 233         vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 234         if (!vma)
 235                 return -ENOMEM;
 236
 237         vma->vm_mm = mm;
 238         vma->vm_start = addr;
 239         vma->vm_end = addr + len;
 240         vma->vm_flags = vm_flags(prot,flags) | mm->def_flags;
 241
 242         if (file) {
 243                 if (file->f_mode & 1)
 244                         vma->vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 245                 if (flags & MAP_SHARED) {
 246                         vma->vm_flags |= VM_SHARED | VM_MAYSHARE;
 247
 248                         /* This looks strange, but when we don't have the file open
 249                          * for writing, we can demote the shared mapping to a simpler
 250                          * private mapping. That also takes care of a security hole
 251                          * with ptrace() writing to a shared mapping without write
 252                          * permissions.
 253                          *
 254                          * We leave the VM_MAYSHARE bit on, just to get correct output
 255                          * from /proc/xxx/maps..
 256                          */
 257                         if (!(file->f_mode & 2))
 258                                 vma->vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
 259                 }
 260         } else
 261                 vma->vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 262         vma->vm_page_prot = protection_map[vma->vm_flags & 0x0f];
 263         vma->vm_ops = NULL;
 264         vma->vm_offset = off;
 265         vma->vm_file = NULL;
 266         vma->vm_pte = 0;
 267
 268         /* Clear old maps */
 269         error = -ENOMEM;
 270         if (do_munmap(addr, len))
 271                 goto free_vma;
 272
 273         /* Check against address space limit. */
 274         if ((mm->total_vm << PAGE_SHIFT) + len
 275             > current->rlim[RLIMIT_AS].rlim_cur)
 276                 goto free_vma;
 277
 278         /* Private writable mapping? Check memory availability.. */
 279         if ((vma->vm_flags & (VM_SHARED | VM_WRITE)) == VM_WRITE &&
 280             !(flags & MAP_NORESERVE)                             &&
 281             !vm_enough_memory(len >> PAGE_SHIFT))
 282                 goto free_vma;
 283
 284         error = 0;
 285         if (file) {
 286                 if (vma->vm_flags & VM_DENYWRITE) {
 287                         if (file->f_dentry->d_inode->i_writecount > 0)
 288                                 error = -ETXTBSY;
 289                         else {
 290                                 /* f_op->mmap might possibly sleep
 291                                  * (generic_file_mmap doesn't, but other code
 292                                  * might). In any case, this takes care of any
 293                                  * race that this might cause.
 294                                  */
 295                                 file->f_dentry->d_inode->i_writecount--;
 296                                 correct_wcount = 1;
 297                         }
 298                 }
 299                 if (!error)
 300                         error = file->f_op->mmap(file, vma);
 301
 302         }
 303         /* Fix up the count if necessary, then check for an error */
 304         if (correct_wcount)
 305                 file->f_dentry->d_inode->i_writecount++;
 306         if (error)
 307                 goto free_vma;
 308
 309         /*
 310          * merge_segments may merge our vma, so we can't refer to it
 311          * after the call.  Save the values we need now ...
 312          */
 313         flags = vma->vm_flags;
 314         addr = vma->vm_start; /* can addr have changed?? */
 315         insert_vm_struct(mm, vma);
 316         merge_segments(mm, vma->vm_start, vma->vm_end);
 317
 318         mm->total_vm += len >> PAGE_SHIFT;
 319         if ((flags & VM_LOCKED) && !(flags & VM_IO)) {
 320                 unsigned long start = addr;
 321                 mm->locked_vm += len >> PAGE_SHIFT;
 322                 do {
 323                         char c;
 324                         get_user(c,(char *) start);
 325                         len -= PAGE_SIZE;
 326                         start += PAGE_SIZE;
 327                         __asm__ __volatile__("": :"r" (c));
 328                 } while (len > 0);
 329         }
 330         return addr;
 331
 332 free_vma:
 333         kmem_cache_free(vm_area_cachep, vma);
 334         return error;
 335 }
 336
 337 /* Get an address range which is currently unmapped.
 338  * For mmap() without MAP_FIXED and shmat() with addr=0.
 339  * Return value 0 means ENOMEM.
 340  */
 341 unsigned long get_unmapped_area(unsigned long addr, unsigned long len)
 342 {
 343         struct vm_area_struct * vmm;
 344
 345         if (len > TASK_SIZE)
 346                 return 0;
 347         if (!addr)
 348                 addr = TASK_UNMAPPED_BASE;
 349         addr = PAGE_ALIGN(addr);
 350
 351         for (vmm = find_vma(current->mm, addr); ; vmm = vmm->vm_next) {
 352                 /* At this point:  (!vmm || addr < vmm->vm_end). */
 353                 if (TASK_SIZE - len < addr)
 354                         return 0;
 355                 if (!vmm || addr + len <= vmm->vm_start)
 356                         return addr;
 357                 addr = vmm->vm_end;
 358         }
 359 }
 360
 361 /* Normal function to fix up a mapping
 362  * This function is the default for when an area has no specific
 363  * function.  This may be used as part of a more specific routine.
 364  * This function works out what part of an area is affected and
 365  * adjusts the mapping information.  Since the actual page
 366  * manipulation is done in do_mmap(), none need be done here,
 367  * though it would probably be more appropriate.
 368  *
 369  * By the time this function is called, the area struct has been
 370  * removed from the process mapping list, so it needs to be
 371  * reinserted if necessary.
 372  *
 373  * The 4 main cases are:
 374  *    Unmapping the whole area
 375  *    Unmapping from the start of the segment to a point in it
 376  *    Unmapping from an intermediate point to the end
 377  *    Unmapping between to intermediate points, making a hole.
 378  *
 379  * Case 4 involves the creation of 2 new areas, for each side of
 380  * the hole.  If possible, we reuse the existing area rather than
 381  * allocate a new one, and the return indicates whether the old
 382  * area was reused.
 383  */
 384 static int unmap_fixup(struct vm_area_struct *area, unsigned long addr,
 385                          size_t len, struct vm_area_struct **extra)
 386 {
 387         struct vm_area_struct *mpnt;
 388         unsigned long end = addr + len;
 389
 390         area->vm_mm->total_vm -= len >> PAGE_SHIFT;
 391         if (area->vm_flags & VM_LOCKED)
 392                 area->vm_mm->locked_vm -= len >> PAGE_SHIFT;
 393
 394         /* Unmapping the whole area. */
 395         if (addr == area->vm_start && end == area->vm_end) {
 396                 if (area->vm_ops && area->vm_ops->close)
 397                         area->vm_ops->close(area);
 398                 if (area->vm_file)
 399                         fput(area->vm_file);
 400                 return 0;
 401         }
 402
 403         /* Work out to one of the ends. */
 404         if (end == area->vm_end)
 405                 area->vm_end = addr;
 406         else if (addr == area->vm_start) {
 407                 area->vm_offset += (end - area->vm_start);
 408                 area->vm_start = end;
 409         } else {
 410         /* Unmapping a hole: area->vm_start < addr <= end < area->vm_end */
 411                 /* Add end mapping -- leave beginning for below */
 412                 mpnt = *extra;
 413                 *extra = NULL;
 414
 415                 mpnt->vm_mm = area->vm_mm;
 416                 mpnt->vm_start = end;
 417                 mpnt->vm_end = area->vm_end;
 418                 mpnt->vm_page_prot = area->vm_page_prot;
 419                 mpnt->vm_flags = area->vm_flags;
 420                 mpnt->vm_ops = area->vm_ops;
 421                 mpnt->vm_offset = area->vm_offset + (end - area->vm_start);
 422                 mpnt->vm_file = area->vm_file;
 423                 if (mpnt->vm_file)
 424                         mpnt->vm_file->f_count++;
 425                 if (mpnt->vm_ops && mpnt->vm_ops->open)
 426                         mpnt->vm_ops->open(mpnt);
 427                 area->vm_end = addr;    /* Truncate area */
 428                 insert_vm_struct(current->mm, mpnt);
 429         }
 430
 431         /* Close the current area ... */
 432         if (area->vm_ops && area->vm_ops->close) {
 433                 end = area->vm_end; /* save new end */
 434                 area->vm_end = area->vm_start;
 435                 area->vm_ops->close(area);
 436                 area->vm_end = end;
 437         }
 438         /* ... then reopen and reinsert. */
 439         if (area->vm_ops && area->vm_ops->open)
 440                 area->vm_ops->open(area);
 441         insert_vm_struct(current->mm, area);
 442         return 1;
 443 }
 444
 445 asmlinkage int sys_munmap(unsigned long addr, size_t len)
 446 {
 447         int ret;
 448
 449         lock_kernel();
 450         ret = do_munmap(addr, len);
 451         unlock_kernel();
 452         return ret;
 453 }
 454
 455 /* Munmap is split into 2 main parts -- this part which finds
 456  * what needs doing, and the areas themselves, which do the
 457  * work.  This now handles partial unmappings.
 458  * Jeremy Fitzhardine <jeremy@sw.oz.au>
 459  */
 460 int do_munmap(unsigned long addr, size_t len)
 461 {
 462         struct mm_struct * mm;
 463         struct vm_area_struct *mpnt, *next, *free, *extra;
 464         int freed;
 465
 466         if ((addr & ~PAGE_MASK) || addr > TASK_SIZE || len > TASK_SIZE-addr)
 467                 return -EINVAL;
 468
 469         if ((len = PAGE_ALIGN(len)) == 0)
 470                 return 0;
 471
 472         /* Check if this memory area is ok - put it on the temporary
 473          * list if so..  The checks here are pretty simple --
 474          * every area affected in some way (by any overlap) is put
 475          * on the list.  If nothing is put on, nothing is affected.
 476          */
 477         mm = current->mm;
 478         mpnt = mm->mmap;
 479         while(mpnt && mpnt->vm_end <= addr)
 480                 mpnt = mpnt->vm_next;
 481         if (!mpnt)
 482                 return 0;
 483
 484         /*
 485          * We may need one additional vma to fix up the mappings ...
 486          * and this is the last chance for an easy error exit.
 487          */
 488         extra = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 489         if (!extra)
 490                 return -ENOMEM;
 491
 492         next = mpnt->vm_next;
 493
 494         /* we have mpnt->vm_next = next and addr < mpnt->vm_end */
 495         free = NULL;
 496         for ( ; mpnt && mpnt->vm_start < addr+len; ) {
 497                 struct vm_area_struct *next = mpnt->vm_next;
 498
 499                 if(mpnt->vm_next)
 500                         mpnt->vm_next->vm_pprev = mpnt->vm_pprev;
 501                 *mpnt->vm_pprev = mpnt->vm_next;
 502
 503                 mpnt->vm_next = free;
 504                 free = mpnt;
 505                 mpnt = next;
 506         }
 507
 508         if (free && (free->vm_start < addr) && (free->vm_end > addr+len)) {
 509                 if (mm->map_count > MAX_MAP_COUNT) {
 510                         kmem_cache_free(vm_area_cachep, extra);
 511                         return -ENOMEM;
 512                 }
 513         }
 514
 515         /* Ok - we have the memory areas we should free on the 'free' list,
 516          * so release them, and unmap the page range..
 517          * If the one of the segments is only being partially unmapped,
 518          * it will put new vm_area_struct(s) into the address space.
 519          */
 520         freed = 0;
 521         while ((mpnt = free) != NULL) {
 522                 unsigned long st, end, size;
 523
 524                 free = free->vm_next;
 525                 freed = 1;
 526
 527                 mm->map_count--;
 528                 remove_shared_vm_struct(mpnt);
 529
 530                 st = addr < mpnt->vm_start ? mpnt->vm_start : addr;
 531                 end = addr+len;
 532                 end = end > mpnt->vm_end ? mpnt->vm_end : end;
 533                 size = end - st;
 534
 535                 if (mpnt->vm_ops && mpnt->vm_ops->unmap)
 536                         mpnt->vm_ops->unmap(mpnt, st, size);
 537
 538                 flush_cache_range(mm, st, end);
 539                 zap_page_range(mm, st, size);
 540                 flush_tlb_range(mm, st, end);
 541
 542                 /*
 543                  * Fix the mapping, and free the old area if it wasn't reused.
 544                  */
 545                 if (!unmap_fixup(mpnt, st, size, &extra))
 546                         kmem_cache_free(vm_area_cachep, mpnt);
 547         }
 548
 549         /* Release the extra vma struct if it wasn't used */
 550         if (extra)
 551                 kmem_cache_free(vm_area_cachep, extra);
 552
 553         if (freed)
 554                 mm->mmap_cache = NULL;  /* Kill the cache. */
 555         return 0;
 556 }
 557
 558 /* Release all mmaps. */
 559 void exit_mmap(struct mm_struct * mm)
 560 {
 561         struct vm_area_struct * mpnt;
 562
 563         mpnt = mm->mmap;
 564         mm->mmap = mm->mmap_cache = NULL;
 565         mm->rss = 0;
 566         mm->total_vm = 0;
 567         mm->locked_vm = 0;
 568         while (mpnt) {
 569                 struct vm_area_struct * next = mpnt->vm_next;
 570                 unsigned long start = mpnt->vm_start;
 571                 unsigned long end = mpnt->vm_end;
 572                 unsigned long size = end - start;
 573
 574                 if (mpnt->vm_ops) {
 575                         if (mpnt->vm_ops->unmap)
 576                                 mpnt->vm_ops->unmap(mpnt, start, size);
 577                         if (mpnt->vm_ops->close)
 578                                 mpnt->vm_ops->close(mpnt);
 579                 }
 580                 mm->map_count--;
 581                 remove_shared_vm_struct(mpnt);
 582                 zap_page_range(mm, start, size);
 583                 if (mpnt->vm_file)
 584                         fput(mpnt->vm_file);
 585                 kmem_cache_free(vm_area_cachep, mpnt);
 586                 mpnt = next;
 587         }
 588
 589         /* This is just debugging */
 590         if (mm->map_count)
 591                 printk("exit_mmap: map count is %d\n", mm->map_count);
 592 }
 593
 594 /* Insert vm structure into process list sorted by address
 595  * and into the inode's i_mmap ring.
 596  */
 597 void insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vmp)
 598 {
 599         struct vm_area_struct **pprev = &mm->mmap;
 600         struct file * file;
 601
 602         mm->map_count++;
 603
 604         /* Find where to link it in. */
 605         while(*pprev && (*pprev)->vm_start <= vmp->vm_start)
 606                 pprev = &(*pprev)->vm_next;
 607
 608         /* Insert it. */
 609         if((vmp->vm_next = *pprev) != NULL)
 610                 (*pprev)->vm_pprev = &vmp->vm_next;
 611         *pprev = vmp;
 612         vmp->vm_pprev = pprev;
 613
 614         file = vmp->vm_file;
 615         if (file) {
 616                 struct inode * inode = file->f_dentry->d_inode;
 617                 if (vmp->vm_flags & VM_DENYWRITE)
 618                         inode->i_writecount--;
 619
 620                 /* insert vmp into inode's share list */
 621                 if((vmp->vm_next_share = inode->i_mmap) != NULL)
 622                         inode->i_mmap->vm_pprev_share = &vmp->vm_next_share;
 623                 inode->i_mmap = vmp;
 624                 vmp->vm_pprev_share = &inode->i_mmap;
 625         }
 626 }
 627
 628 /* Merge the list of memory segments if possible.
 629  * Redundant vm_area_structs are freed.
 630  * This assumes that the list is ordered by address.
 631  * We don't need to traverse the entire list, only those segments
 632  * which intersect or are adjacent to a given interval.
 633  */
 634 void merge_segments (struct mm_struct * mm, unsigned long start_addr, unsigned long end_addr)
 635 {
 636         struct vm_area_struct *prev, *mpnt, *next;
 637
 638         down(&mm->mmap_sem);
 639
 640         prev = NULL;
 641         mpnt = mm->mmap;
 642         while(mpnt && mpnt->vm_end <= start_addr) {
 643                 prev = mpnt;
 644                 mpnt = mpnt->vm_next;
 645         }
 646         if (!mpnt)
 647                 goto no_vma;
 648
 649         next = mpnt->vm_next;
 650
 651         /* we have prev->vm_next == mpnt && mpnt->vm_next = next */
 652         if (!prev) {
 653                 prev = mpnt;
 654                 mpnt = next;
 655         }
 656
 657         /* prev and mpnt cycle through the list, as long as
 658          * start_addr < mpnt->vm_end && prev->vm_start < end_addr
 659          */
 660         for ( ; mpnt && prev->vm_start < end_addr ; prev = mpnt, mpnt = next) {
 661                 next = mpnt->vm_next;
 662
 663                 /* To share, we must have the same file, operations.. */
 664                 if ((mpnt->vm_file != prev->vm_file)||
 665                     (mpnt->vm_pte != prev->vm_pte)      ||
 666                     (mpnt->vm_ops != prev->vm_ops)      ||
 667                     (mpnt->vm_flags != prev->vm_flags)  ||
 668                     (prev->vm_end != mpnt->vm_start))
 669                         continue;
 670
 671                 /*
 672                  * If we have a file or it's a shared memory area
 673                  * the offsets must be contiguous..
 674                  */
 675                 if ((mpnt->vm_file != NULL) || (mpnt->vm_flags & VM_SHM)) {
 676                         unsigned long off = prev->vm_offset+prev->vm_end-prev->vm_start;
 677                         if (off != mpnt->vm_offset)
 678                                 continue;
 679                 }
 680
 681                 /* merge prev with mpnt and set up pointers so the new
 682                  * big segment can possibly merge with the next one.
 683                  * The old unused mpnt is freed.
 684                  */
 685                 if(mpnt->vm_next)
 686                         mpnt->vm_next->vm_pprev = mpnt->vm_pprev;
 687                 *mpnt->vm_pprev = mpnt->vm_next;
 688
 689                 prev->vm_end = mpnt->vm_end;
 690                 if (mpnt->vm_ops && mpnt->vm_ops->close) {
 691                         mpnt->vm_offset += mpnt->vm_end - mpnt->vm_start;
 692                         mpnt->vm_start = mpnt->vm_end;
 693                         mpnt->vm_ops->close(mpnt);
 694                 }
 695                 mm->map_count--;
 696                 remove_shared_vm_struct(mpnt);
 697                 if (mpnt->vm_file)
 698                         fput(mpnt->vm_file);
 699                 kmem_cache_free(vm_area_cachep, mpnt);
 700                 mpnt = prev;
 701         }
 702         mm->mmap_cache = NULL;          /* Kill the cache. */
 703 no_vma:
 704         up(&mm->mmap_sem);
 705 }
 706
 707 __initfunc(void vma_init(void))
 708 {
 709         vm_area_cachep = kmem_cache_create("vm_area_struct",
 710                                            sizeof(struct vm_area_struct),
 711                                            0, SLAB_HWCACHE_ALIGN,
 712                                            NULL, NULL);
 713         if(!vm_area_cachep)
 714                 panic("vma_init: Cannot alloc vm_area_struct cache.");
 715
 716         mm_cachep = kmem_cache_create("mm_struct",
 717                                       sizeof(struct mm_struct),
 718                                       0, SLAB_HWCACHE_ALIGN,
 719                                       NULL, NULL);
 720         if(!mm_cachep)
 721                 panic("vma_init: Cannot alloc mm_struct cache.");
 722 }