SAFE public projects git trees. - safe/jmp/linux-2.6/blob - mm/nommu.c

   1 /*
   2  *  linux/mm/nommu.c
   3  *
   4  *  Replacement code for mm functions to support CPU's that don't
   5  *  have any form of memory management unit (thus no virtual memory).
   6  *
   7  *  See Documentation/nommu-mmap.txt
   8  *
   9  *  Copyright (c) 2004-2005 David Howells <dhowells@redhat.com>
  10  *  Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
  11  *  Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
  12  *  Copyright (c) 2002      Greg Ungerer <gerg@snapgear.com>
  13  */
  14
  15 #include <linux/mm.h>
  16 #include <linux/mman.h>
  17 #include <linux/swap.h>
  18 #include <linux/file.h>
  19 #include <linux/highmem.h>
  20 #include <linux/pagemap.h>
  21 #include <linux/slab.h>
  22 #include <linux/vmalloc.h>
  23 #include <linux/ptrace.h>
  24 #include <linux/blkdev.h>
  25 #include <linux/backing-dev.h>
  26 #include <linux/mount.h>
  27 #include <linux/personality.h>
  28 #include <linux/security.h>
  29 #include <linux/syscalls.h>
  30
  31 #include <asm/uaccess.h>
  32 #include <asm/tlb.h>
  33 #include <asm/tlbflush.h>
  34
  35 void *high_memory;
  36 struct page *mem_map;
  37 unsigned long max_mapnr;
  38 unsigned long num_physpages;
  39 unsigned long askedalloc, realalloc;
  40 atomic_t vm_committed_space = ATOMIC_INIT(0);
  41 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
  42 int sysctl_overcommit_ratio = 50; /* default is 50% */
  43 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
  44 int heap_stack_gap = 0;
  45
  46 EXPORT_SYMBOL(mem_map);
  47 EXPORT_SYMBOL(sysctl_max_map_count);
  48 EXPORT_SYMBOL(sysctl_overcommit_memory);
  49 EXPORT_SYMBOL(sysctl_overcommit_ratio);
  50 EXPORT_SYMBOL(vm_committed_space);
  51 EXPORT_SYMBOL(__vm_enough_memory);
  52
  53 /* list of shareable VMAs */
  54 struct rb_root nommu_vma_tree = RB_ROOT;
  55 DECLARE_RWSEM(nommu_vma_sem);
  56
  57 struct vm_operations_struct generic_file_vm_ops = {
  58 };
  59
  60 /*
  61  * Handle all mappings that got truncated by a "truncate()"
  62  * system call.
  63  *
  64  * NOTE! We have to be ready to update the memory sharing
  65  * between the file and the memory map for a potential last
  66  * incomplete page.  Ugly, but necessary.
  67  */
  68 int vmtruncate(struct inode *inode, loff_t offset)
  69 {
  70         struct address_space *mapping = inode->i_mapping;
  71         unsigned long limit;
  72
  73         if (inode->i_size < offset)
  74                 goto do_expand;
  75         i_size_write(inode, offset);
  76
  77         truncate_inode_pages(mapping, offset);
  78         goto out_truncate;
  79
  80 do_expand:
  81         limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
  82         if (limit != RLIM_INFINITY && offset > limit)
  83                 goto out_sig;
  84         if (offset > inode->i_sb->s_maxbytes)
  85                 goto out;
  86         i_size_write(inode, offset);
  87
  88 out_truncate:
  89         if (inode->i_op && inode->i_op->truncate)
  90                 inode->i_op->truncate(inode);
  91         return 0;
  92 out_sig:
  93         send_sig(SIGXFSZ, current, 0);
  94 out:
  95         return -EFBIG;
  96 }
  97
  98 EXPORT_SYMBOL(vmtruncate);
  99
 100 /*
 101  * Return the total memory allocated for this pointer, not
 102  * just what the caller asked for.
 103  *
 104  * Doesn't have to be accurate, i.e. may have races.
 105  */
 106 unsigned int kobjsize(const void *objp)
 107 {
 108         struct page *page;
 109
 110         if (!objp || !((page = virt_to_page(objp))))
 111                 return 0;
 112
 113         if (PageSlab(page))
 114                 return ksize(objp);
 115
 116         BUG_ON(page->index < 0);
 117         BUG_ON(page->index >= MAX_ORDER);
 118
 119         return (PAGE_SIZE << page->index);
 120 }
 121
 122 /*
 123  * The nommu dodgy version :-)
 124  */
 125 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
 126         unsigned long start, int len, int write, int force,
 127         struct page **pages, struct vm_area_struct **vmas)
 128 {
 129         int i;
 130         static struct vm_area_struct dummy_vma;
 131
 132         for (i = 0; i < len; i++) {
 133                 if (pages) {
 134                         pages[i] = virt_to_page(start);
 135                         if (pages[i])
 136                                 page_cache_get(pages[i]);
 137                 }
 138                 if (vmas)
 139                         vmas[i] = &dummy_vma;
 140                 start += PAGE_SIZE;
 141         }
 142         return(i);
 143 }
 144
 145 DEFINE_RWLOCK(vmlist_lock);
 146 struct vm_struct *vmlist;
 147
 148 void vfree(void *addr)
 149 {
 150         kfree(addr);
 151 }
 152
 153 void *__vmalloc(unsigned long size, int gfp_mask, pgprot_t prot)
 154 {
 155         /*
 156          * kmalloc doesn't like __GFP_HIGHMEM for some reason
 157          */
 158         return kmalloc(size, gfp_mask & ~__GFP_HIGHMEM);
 159 }
 160
 161 struct page * vmalloc_to_page(void *addr)
 162 {
 163         return virt_to_page(addr);
 164 }
 165
 166 unsigned long vmalloc_to_pfn(void *addr)
 167 {
 168         return page_to_pfn(virt_to_page(addr));
 169 }
 170
 171
 172 long vread(char *buf, char *addr, unsigned long count)
 173 {
 174         memcpy(buf, addr, count);
 175         return count;
 176 }
 177
 178 long vwrite(char *buf, char *addr, unsigned long count)
 179 {
 180         /* Don't allow overflow */
 181         if ((unsigned long) addr + count < count)
 182                 count = -(unsigned long) addr;
 183
 184         memcpy(addr, buf, count);
 185         return(count);
 186 }
 187
 188 /*
 189  *      vmalloc  -  allocate virtually continguos memory
 190  *
 191  *      @size:          allocation size
 192  *
 193  *      Allocate enough pages to cover @size from the page level
 194  *      allocator and map them into continguos kernel virtual space.
 195  *
 196  *      For tight cotrol over page level allocator and protection flags
 197  *      use __vmalloc() instead.
 198  */
 199 void *vmalloc(unsigned long size)
 200 {
 201        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
 202 }
 203
 204 /*
 205  *      vmalloc_32  -  allocate virtually continguos memory (32bit addressable)
 206  *
 207  *      @size:          allocation size
 208  *
 209  *      Allocate enough 32bit PA addressable pages to cover @size from the
 210  *      page level allocator and map them into continguos kernel virtual space.
 211  */
 212 void *vmalloc_32(unsigned long size)
 213 {
 214         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
 215 }
 216
 217 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
 218 {
 219         BUG();
 220         return NULL;
 221 }
 222
 223 void vunmap(void *addr)
 224 {
 225         BUG();
 226 }
 227
 228 /*
 229  *  sys_brk() for the most part doesn't need the global kernel
 230  *  lock, except when an application is doing something nasty
 231  *  like trying to un-brk an area that has already been mapped
 232  *  to a regular file.  in this case, the unmapping will need
 233  *  to invoke file system routines that need the global lock.
 234  */
 235 asmlinkage unsigned long sys_brk(unsigned long brk)
 236 {
 237         struct mm_struct *mm = current->mm;
 238
 239         if (brk < mm->start_brk || brk > mm->context.end_brk)
 240                 return mm->brk;
 241
 242         if (mm->brk == brk)
 243                 return mm->brk;
 244
 245         /*
 246          * Always allow shrinking brk
 247          */
 248         if (brk <= mm->brk) {
 249                 mm->brk = brk;
 250                 return brk;
 251         }
 252
 253         /*
 254          * Ok, looks good - let it rip.
 255          */
 256         return mm->brk = brk;
 257 }
 258
 259 #ifdef DEBUG
 260 static void show_process_blocks(void)
 261 {
 262         struct vm_list_struct *vml;
 263
 264         printk("Process blocks %d:", current->pid);
 265
 266         for (vml = &current->mm->context.vmlist; vml; vml = vml->next) {
 267                 printk(" %p: %p", vml, vml->vma);
 268                 if (vml->vma)
 269                         printk(" (%d @%lx #%d)",
 270                                kobjsize((void *) vml->vma->vm_start),
 271                                vml->vma->vm_start,
 272                                atomic_read(&vml->vma->vm_usage));
 273                 printk(vml->next ? " ->" : ".\n");
 274         }
 275 }
 276 #endif /* DEBUG */
 277
 278 static inline struct vm_area_struct *find_nommu_vma(unsigned long start)
 279 {
 280         struct vm_area_struct *vma;
 281         struct rb_node *n = nommu_vma_tree.rb_node;
 282
 283         while (n) {
 284                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
 285
 286                 if (start < vma->vm_start)
 287                         n = n->rb_left;
 288                 else if (start > vma->vm_start)
 289                         n = n->rb_right;
 290                 else
 291                         return vma;
 292         }
 293
 294         return NULL;
 295 }
 296
 297 static void add_nommu_vma(struct vm_area_struct *vma)
 298 {
 299         struct vm_area_struct *pvma;
 300         struct address_space *mapping;
 301         struct rb_node **p = &nommu_vma_tree.rb_node;
 302         struct rb_node *parent = NULL;
 303
 304         /* add the VMA to the mapping */
 305         if (vma->vm_file) {
 306                 mapping = vma->vm_file->f_mapping;
 307
 308                 flush_dcache_mmap_lock(mapping);
 309                 vma_prio_tree_insert(vma, &mapping->i_mmap);
 310                 flush_dcache_mmap_unlock(mapping);
 311         }
 312
 313         /* add the VMA to the master list */
 314         while (*p) {
 315                 parent = *p;
 316                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
 317
 318                 if (vma->vm_start < pvma->vm_start) {
 319                         p = &(*p)->rb_left;
 320                 }
 321                 else if (vma->vm_start > pvma->vm_start) {
 322                         p = &(*p)->rb_right;
 323                 }
 324                 else {
 325                         /* mappings are at the same address - this can only
 326                          * happen for shared-mem chardevs and shared file
 327                          * mappings backed by ramfs/tmpfs */
 328                         BUG_ON(!(pvma->vm_flags & VM_SHARED));
 329
 330                         if (vma < pvma)
 331                                 p = &(*p)->rb_left;
 332                         else if (vma > pvma)
 333                                 p = &(*p)->rb_right;
 334                         else
 335                                 BUG();
 336                 }
 337         }
 338
 339         rb_link_node(&vma->vm_rb, parent, p);
 340         rb_insert_color(&vma->vm_rb, &nommu_vma_tree);
 341 }
 342
 343 static void delete_nommu_vma(struct vm_area_struct *vma)
 344 {
 345         struct address_space *mapping;
 346
 347         /* remove the VMA from the mapping */
 348         if (vma->vm_file) {
 349                 mapping = vma->vm_file->f_mapping;
 350
 351                 flush_dcache_mmap_lock(mapping);
 352                 vma_prio_tree_remove(vma, &mapping->i_mmap);
 353                 flush_dcache_mmap_unlock(mapping);
 354         }
 355
 356         /* remove from the master list */
 357         rb_erase(&vma->vm_rb, &nommu_vma_tree);
 358 }
 359
 360 /*
 361  * determine whether a mapping should be permitted and, if so, what sort of
 362  * mapping we're capable of supporting
 363  */
 364 static int validate_mmap_request(struct file *file,
 365                                  unsigned long addr,
 366                                  unsigned long len,
 367                                  unsigned long prot,
 368                                  unsigned long flags,
 369                                  unsigned long pgoff,
 370                                  unsigned long *_capabilities)
 371 {
 372         unsigned long capabilities;
 373         unsigned long reqprot = prot;
 374         int ret;
 375
 376         /* do the simple checks first */
 377         if (flags & MAP_FIXED || addr) {
 378                 printk(KERN_DEBUG
 379                        "%d: Can't do fixed-address/overlay mmap of RAM\n",
 380                        current->pid);
 381                 return -EINVAL;
 382         }
 383
 384         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
 385             (flags & MAP_TYPE) != MAP_SHARED)
 386                 return -EINVAL;
 387
 388         if (PAGE_ALIGN(len) == 0)
 389                 return addr;
 390
 391         if (len > TASK_SIZE)
 392                 return -EINVAL;
 393
 394         /* offset overflow? */
 395         if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
 396                 return -EINVAL;
 397
 398         if (file) {
 399                 /* validate file mapping requests */
 400                 struct address_space *mapping;
 401
 402                 /* files must support mmap */
 403                 if (!file->f_op || !file->f_op->mmap)
 404                         return -ENODEV;
 405
 406                 /* work out if what we've got could possibly be shared
 407                  * - we support chardevs that provide their own "memory"
 408                  * - we support files/blockdevs that are memory backed
 409                  */
 410                 mapping = file->f_mapping;
 411                 if (!mapping)
 412                         mapping = file->f_dentry->d_inode->i_mapping;
 413
 414                 capabilities = 0;
 415                 if (mapping && mapping->backing_dev_info)
 416                         capabilities = mapping->backing_dev_info->capabilities;
 417
 418                 if (!capabilities) {
 419                         /* no explicit capabilities set, so assume some
 420                          * defaults */
 421                         switch (file->f_dentry->d_inode->i_mode & S_IFMT) {
 422                         case S_IFREG:
 423                         case S_IFBLK:
 424                                 capabilities = BDI_CAP_MAP_COPY;
 425                                 break;
 426
 427                         case S_IFCHR:
 428                                 capabilities =
 429                                         BDI_CAP_MAP_DIRECT |
 430                                         BDI_CAP_READ_MAP |
 431                                         BDI_CAP_WRITE_MAP;
 432                                 break;
 433
 434                         default:
 435                                 return -EINVAL;
 436                         }
 437                 }
 438
 439                 /* eliminate any capabilities that we can't support on this
 440                  * device */
 441                 if (!file->f_op->get_unmapped_area)
 442                         capabilities &= ~BDI_CAP_MAP_DIRECT;
 443                 if (!file->f_op->read)
 444                         capabilities &= ~BDI_CAP_MAP_COPY;
 445
 446                 if (flags & MAP_SHARED) {
 447                         /* do checks for writing, appending and locking */
 448                         if ((prot & PROT_WRITE) &&
 449                             !(file->f_mode & FMODE_WRITE))
 450                                 return -EACCES;
 451
 452                         if (IS_APPEND(file->f_dentry->d_inode) &&
 453                             (file->f_mode & FMODE_WRITE))
 454                                 return -EACCES;
 455
 456                         if (locks_verify_locked(file->f_dentry->d_inode))
 457                                 return -EAGAIN;
 458
 459                         if (!(capabilities & BDI_CAP_MAP_DIRECT))
 460                                 return -ENODEV;
 461
 462                         if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
 463                             ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
 464                             ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
 465                             ) {
 466                                 printk("MAP_SHARED not completely supported on !MMU\n");
 467                                 return -EINVAL;
 468                         }
 469
 470                         /* we mustn't privatise shared mappings */
 471                         capabilities &= ~BDI_CAP_MAP_COPY;
 472                 }
 473                 else {
 474                         /* we're going to read the file into private memory we
 475                          * allocate */
 476                         if (!(capabilities & BDI_CAP_MAP_COPY))
 477                                 return -ENODEV;
 478
 479                         /* we don't permit a private writable mapping to be
 480                          * shared with the backing device */
 481                         if (prot & PROT_WRITE)
 482                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
 483                 }
 484
 485                 /* handle executable mappings and implied executable
 486                  * mappings */
 487                 if (file->f_vfsmnt->mnt_flags & MNT_NOEXEC) {
 488                         if (prot & PROT_EXEC)
 489                                 return -EPERM;
 490                 }
 491                 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
 492                         /* handle implication of PROT_EXEC by PROT_READ */
 493                         if (current->personality & READ_IMPLIES_EXEC) {
 494                                 if (capabilities & BDI_CAP_EXEC_MAP)
 495                                         prot |= PROT_EXEC;
 496                         }
 497                 }
 498                 else if ((prot & PROT_READ) &&
 499                          (prot & PROT_EXEC) &&
 500                          !(capabilities & BDI_CAP_EXEC_MAP)
 501                          ) {
 502                         /* backing file is not executable, try to copy */
 503                         capabilities &= ~BDI_CAP_MAP_DIRECT;
 504                 }
 505         }
 506         else {
 507                 /* anonymous mappings are always memory backed and can be
 508                  * privately mapped
 509                  */
 510                 capabilities = BDI_CAP_MAP_COPY;
 511
 512                 /* handle PROT_EXEC implication by PROT_READ */
 513                 if ((prot & PROT_READ) &&
 514                     (current->personality & READ_IMPLIES_EXEC))
 515                         prot |= PROT_EXEC;
 516         }
 517
 518         /* allow the security API to have its say */
 519         ret = security_file_mmap(file, reqprot, prot, flags);
 520         if (ret < 0)
 521                 return ret;
 522
 523         /* looks okay */
 524         *_capabilities = capabilities;
 525         return 0;
 526 }
 527
 528 /*
 529  * we've determined that we can make the mapping, now translate what we
 530  * now know into VMA flags
 531  */
 532 static unsigned long determine_vm_flags(struct file *file,
 533                                         unsigned long prot,
 534                                         unsigned long flags,
 535                                         unsigned long capabilities)
 536 {
 537         unsigned long vm_flags;
 538
 539         vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
 540         vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 541         /* vm_flags |= mm->def_flags; */
 542
 543         if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
 544                 /* attempt to share read-only copies of mapped file chunks */
 545                 if (file && !(prot & PROT_WRITE))
 546                         vm_flags |= VM_MAYSHARE;
 547         }
 548         else {
 549                 /* overlay a shareable mapping on the backing device or inode
 550                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
 551                  * romfs/cramfs */
 552                 if (flags & MAP_SHARED)
 553                         vm_flags |= VM_MAYSHARE | VM_SHARED;
 554                 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
 555                         vm_flags |= VM_MAYSHARE;
 556         }
 557
 558         /* refuse to let anyone share private mappings with this process if
 559          * it's being traced - otherwise breakpoints set in it may interfere
 560          * with another untraced process
 561          */
 562         if ((flags & MAP_PRIVATE) && (current->ptrace & PT_PTRACED))
 563                 vm_flags &= ~VM_MAYSHARE;
 564
 565         return vm_flags;
 566 }
 567
 568 /*
 569  * set up a shared mapping on a file
 570  */
 571 static int do_mmap_shared_file(struct vm_area_struct *vma, unsigned long len)
 572 {
 573         int ret;
 574
 575         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
 576         if (ret != -ENOSYS)
 577                 return ret;
 578
 579         /* getting an ENOSYS error indicates that direct mmap isn't
 580          * possible (as opposed to tried but failed) so we'll fall
 581          * through to making a private copy of the data and mapping
 582          * that if we can */
 583         return -ENODEV;
 584 }
 585
 586 /*
 587  * set up a private mapping or an anonymous shared mapping
 588  */
 589 static int do_mmap_private(struct vm_area_struct *vma, unsigned long len)
 590 {
 591         void *base;
 592         int ret;
 593
 594         /* invoke the file's mapping function so that it can keep track of
 595          * shared mappings on devices or memory
 596          * - VM_MAYSHARE will be set if it may attempt to share
 597          */
 598         if (vma->vm_file) {
 599                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
 600                 if (ret != -ENOSYS) {
 601                         /* shouldn't return success if we're not sharing */
 602                         BUG_ON(ret == 0 && !(vma->vm_flags & VM_MAYSHARE));
 603                         return ret; /* success or a real error */
 604                 }
 605
 606                 /* getting an ENOSYS error indicates that direct mmap isn't
 607                  * possible (as opposed to tried but failed) so we'll try to
 608                  * make a private copy of the data and map that instead */
 609         }
 610
 611         /* allocate some memory to hold the mapping
 612          * - note that this may not return a page-aligned address if the object
 613          *   we're allocating is smaller than a page
 614          */
 615         base = kmalloc(len, GFP_KERNEL);
 616         if (!base)
 617                 goto enomem;
 618
 619         vma->vm_start = (unsigned long) base;
 620         vma->vm_end = vma->vm_start + len;
 621         vma->vm_flags |= VM_MAPPED_COPY;
 622
 623 #ifdef WARN_ON_SLACK
 624         if (len + WARN_ON_SLACK <= kobjsize(result))
 625                 printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n",
 626                        len, current->pid, kobjsize(result) - len);
 627 #endif
 628
 629         if (vma->vm_file) {
 630                 /* read the contents of a file into the copy */
 631                 mm_segment_t old_fs;
 632                 loff_t fpos;
 633
 634                 fpos = vma->vm_pgoff;
 635                 fpos <<= PAGE_SHIFT;
 636
 637                 old_fs = get_fs();
 638                 set_fs(KERNEL_DS);
 639                 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
 640                 set_fs(old_fs);
 641
 642                 if (ret < 0)
 643                         goto error_free;
 644
 645                 /* clear the last little bit */
 646                 if (ret < len)
 647                         memset(base + ret, 0, len - ret);
 648
 649         } else {
 650                 /* if it's an anonymous mapping, then just clear it */
 651                 memset(base, 0, len);
 652         }
 653
 654         return 0;
 655
 656 error_free:
 657         kfree(base);
 658         vma->vm_start = 0;
 659         return ret;
 660
 661 enomem:
 662         printk("Allocation of length %lu from process %d failed\n",
 663                len, current->pid);
 664         show_free_areas();
 665         return -ENOMEM;
 666 }
 667
 668 /*
 669  * handle mapping creation for uClinux
 670  */
 671 unsigned long do_mmap_pgoff(struct file *file,
 672                             unsigned long addr,
 673                             unsigned long len,
 674                             unsigned long prot,
 675                             unsigned long flags,
 676                             unsigned long pgoff)
 677 {
 678         struct vm_list_struct *vml = NULL;
 679         struct vm_area_struct *vma = NULL;
 680         struct rb_node *rb;
 681         unsigned long capabilities, vm_flags;
 682         void *result;
 683         int ret;
 684
 685         /* decide whether we should attempt the mapping, and if so what sort of
 686          * mapping */
 687         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
 688                                     &capabilities);
 689         if (ret < 0)
 690                 return ret;
 691
 692         /* we've determined that we can make the mapping, now translate what we
 693          * now know into VMA flags */
 694         vm_flags = determine_vm_flags(file, prot, flags, capabilities);
 695
 696         /* we're going to need to record the mapping if it works */
 697         vml = kmalloc(sizeof(struct vm_list_struct), GFP_KERNEL);
 698         if (!vml)
 699                 goto error_getting_vml;
 700         memset(vml, 0, sizeof(*vml));
 701
 702         down_write(&nommu_vma_sem);
 703
 704         /* if we want to share, we need to check for VMAs created by other
 705          * mmap() calls that overlap with our proposed mapping
 706          * - we can only share with an exact match on most regular files
 707          * - shared mappings on character devices and memory backed files are
 708          *   permitted to overlap inexactly as far as we are concerned for in
 709          *   these cases, sharing is handled in the driver or filesystem rather
 710          *   than here
 711          */
 712         if (vm_flags & VM_MAYSHARE) {
 713                 unsigned long pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 714                 unsigned long vmpglen;
 715
 716                 for (rb = rb_first(&nommu_vma_tree); rb; rb = rb_next(rb)) {
 717                         vma = rb_entry(rb, struct vm_area_struct, vm_rb);
 718
 719                         if (!(vma->vm_flags & VM_MAYSHARE))
 720                                 continue;
 721
 722                         /* search for overlapping mappings on the same file */
 723                         if (vma->vm_file->f_dentry->d_inode != file->f_dentry->d_inode)
 724                                 continue;
 725
 726                         if (vma->vm_pgoff >= pgoff + pglen)
 727                                 continue;
 728
 729                         vmpglen = vma->vm_end - vma->vm_start + PAGE_SIZE - 1;
 730                         vmpglen >>= PAGE_SHIFT;
 731                         if (pgoff >= vma->vm_pgoff + vmpglen)
 732                                 continue;
 733
 734                         /* handle inexactly overlapping matches between mappings */
 735                         if (vma->vm_pgoff != pgoff || vmpglen != pglen) {
 736                                 if (!(capabilities & BDI_CAP_MAP_DIRECT))
 737                                         goto sharing_violation;
 738                                 continue;
 739                         }
 740
 741                         /* we've found a VMA we can share */
 742                         atomic_inc(&vma->vm_usage);
 743
 744                         vml->vma = vma;
 745                         result = (void *) vma->vm_start;
 746                         goto shared;
 747                 }
 748
 749                 vma = NULL;
 750
 751                 /* obtain the address at which to make a shared mapping
 752                  * - this is the hook for quasi-memory character devices to
 753                  *   tell us the location of a shared mapping
 754                  */
 755                 if (file && file->f_op->get_unmapped_area) {
 756                         addr = file->f_op->get_unmapped_area(file, addr, len,
 757                                                              pgoff, flags);
 758                         if (IS_ERR((void *) addr)) {
 759                                 ret = addr;
 760                                 if (ret != (unsigned long) -ENOSYS)
 761                                         goto error;
 762
 763                                 /* the driver refused to tell us where to site
 764                                  * the mapping so we'll have to attempt to copy
 765                                  * it */
 766                                 ret = (unsigned long) -ENODEV;
 767                                 if (!(capabilities & BDI_CAP_MAP_COPY))
 768                                         goto error;
 769
 770                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
 771                         }
 772                 }
 773         }
 774
 775         /* we're going to need a VMA struct as well */
 776         vma = kmalloc(sizeof(struct vm_area_struct), GFP_KERNEL);
 777         if (!vma)
 778                 goto error_getting_vma;
 779
 780         memset(vma, 0, sizeof(*vma));
 781         INIT_LIST_HEAD(&vma->anon_vma_node);
 782         atomic_set(&vma->vm_usage, 1);
 783         if (file)
 784                 get_file(file);
 785         vma->vm_file    = file;
 786         vma->vm_flags   = vm_flags;
 787         vma->vm_start   = addr;
 788         vma->vm_end     = addr + len;
 789         vma->vm_pgoff   = pgoff;
 790
 791         vml->vma = vma;
 792
 793         /* set up the mapping */
 794         if (file && vma->vm_flags & VM_SHARED)
 795                 ret = do_mmap_shared_file(vma, len);
 796         else
 797                 ret = do_mmap_private(vma, len);
 798         if (ret < 0)
 799                 goto error;
 800
 801         /* okay... we have a mapping; now we have to register it */
 802         result = (void *) vma->vm_start;
 803
 804         if (vma->vm_flags & VM_MAPPED_COPY) {
 805                 realalloc += kobjsize(result);
 806                 askedalloc += len;
 807         }
 808
 809         realalloc += kobjsize(vma);
 810         askedalloc += sizeof(*vma);
 811
 812         current->mm->total_vm += len >> PAGE_SHIFT;
 813
 814         add_nommu_vma(vma);
 815
 816  shared:
 817         realalloc += kobjsize(vml);
 818         askedalloc += sizeof(*vml);
 819
 820         vml->next = current->mm->context.vmlist;
 821         current->mm->context.vmlist = vml;
 822
 823         up_write(&nommu_vma_sem);
 824
 825         if (prot & PROT_EXEC)
 826                 flush_icache_range((unsigned long) result,
 827                                    (unsigned long) result + len);
 828
 829 #ifdef DEBUG
 830         printk("do_mmap:\n");
 831         show_process_blocks();
 832 #endif
 833
 834         return (unsigned long) result;
 835
 836  error:
 837         up_write(&nommu_vma_sem);
 838         kfree(vml);
 839         if (vma) {
 840                 fput(vma->vm_file);
 841                 kfree(vma);
 842         }
 843         return ret;
 844
 845  sharing_violation:
 846         up_write(&nommu_vma_sem);
 847         printk("Attempt to share mismatched mappings\n");
 848         kfree(vml);
 849         return -EINVAL;
 850
 851  error_getting_vma:
 852         up_write(&nommu_vma_sem);
 853         kfree(vml);
 854         printk("Allocation of vml for %lu byte allocation from process %d failed\n",
 855                len, current->pid);
 856         show_free_areas();
 857         return -ENOMEM;
 858
 859  error_getting_vml:
 860         printk("Allocation of vml for %lu byte allocation from process %d failed\n",
 861                len, current->pid);
 862         show_free_areas();
 863         return -ENOMEM;
 864 }
 865
 866 /*
 867  * handle mapping disposal for uClinux
 868  */
 869 static void put_vma(struct vm_area_struct *vma)
 870 {
 871         if (vma) {
 872                 down_write(&nommu_vma_sem);
 873
 874                 if (atomic_dec_and_test(&vma->vm_usage)) {
 875                         delete_nommu_vma(vma);
 876
 877                         if (vma->vm_ops && vma->vm_ops->close)
 878                                 vma->vm_ops->close(vma);
 879
 880                         /* IO memory and memory shared directly out of the pagecache from
 881                          * ramfs/tmpfs mustn't be released here */
 882                         if (vma->vm_flags & VM_MAPPED_COPY) {
 883                                 realalloc -= kobjsize((void *) vma->vm_start);
 884                                 askedalloc -= vma->vm_end - vma->vm_start;
 885                                 kfree((void *) vma->vm_start);
 886                         }
 887
 888                         realalloc -= kobjsize(vma);
 889                         askedalloc -= sizeof(*vma);
 890
 891                         if (vma->vm_file)
 892                                 fput(vma->vm_file);
 893                         kfree(vma);
 894                 }
 895
 896                 up_write(&nommu_vma_sem);
 897         }
 898 }
 899
 900 int do_munmap(struct mm_struct *mm, unsigned long addr, size_t len)
 901 {
 902         struct vm_list_struct *vml, **parent;
 903         unsigned long end = addr + len;
 904
 905 #ifdef DEBUG
 906         printk("do_munmap:\n");
 907 #endif
 908
 909         for (parent = &mm->context.vmlist; *parent; parent = &(*parent)->next)
 910                 if ((*parent)->vma->vm_start == addr &&
 911                     (*parent)->vma->vm_end == end)
 912                         goto found;
 913
 914         printk("munmap of non-mmaped memory by process %d (%s): %p\n",
 915                current->pid, current->comm, (void *) addr);
 916         return -EINVAL;
 917
 918  found:
 919         vml = *parent;
 920
 921         put_vma(vml->vma);
 922
 923         *parent = vml->next;
 924         realalloc -= kobjsize(vml);
 925         askedalloc -= sizeof(*vml);
 926         kfree(vml);
 927         mm->total_vm -= len >> PAGE_SHIFT;
 928
 929 #ifdef DEBUG
 930         show_process_blocks();
 931 #endif
 932
 933         return 0;
 934 }
 935
 936 /* Release all mmaps. */
 937 void exit_mmap(struct mm_struct * mm)
 938 {
 939         struct vm_list_struct *tmp;
 940
 941         if (mm) {
 942 #ifdef DEBUG
 943                 printk("Exit_mmap:\n");
 944 #endif
 945
 946                 mm->total_vm = 0;
 947
 948                 while ((tmp = mm->context.vmlist)) {
 949                         mm->context.vmlist = tmp->next;
 950                         put_vma(tmp->vma);
 951
 952                         realalloc -= kobjsize(tmp);
 953                         askedalloc -= sizeof(*tmp);
 954                         kfree(tmp);
 955                 }
 956
 957 #ifdef DEBUG
 958                 show_process_blocks();
 959 #endif
 960         }
 961 }
 962
 963 asmlinkage long sys_munmap(unsigned long addr, size_t len)
 964 {
 965         int ret;
 966         struct mm_struct *mm = current->mm;
 967
 968         down_write(&mm->mmap_sem);
 969         ret = do_munmap(mm, addr, len);
 970         up_write(&mm->mmap_sem);
 971         return ret;
 972 }
 973
 974 unsigned long do_brk(unsigned long addr, unsigned long len)
 975 {
 976         return -ENOMEM;
 977 }
 978
 979 /*
 980  * Expand (or shrink) an existing mapping, potentially moving it at the
 981  * same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
 982  *
 983  * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise
 984  * This option implies MREMAP_MAYMOVE.
 985  *
 986  * on uClinux, we only permit changing a mapping's size, and only as long as it stays within the
 987  * hole allocated by the kmalloc() call in do_mmap_pgoff() and the block is not shareable
 988  */
 989 unsigned long do_mremap(unsigned long addr,
 990                         unsigned long old_len, unsigned long new_len,
 991                         unsigned long flags, unsigned long new_addr)
 992 {
 993         struct vm_list_struct *vml = NULL;
 994
 995         /* insanity checks first */
 996         if (new_len == 0)
 997                 return (unsigned long) -EINVAL;
 998
 999         if (flags & MREMAP_FIXED && new_addr != addr)
1000                 return (unsigned long) -EINVAL;
1001
1002         for (vml = current->mm->context.vmlist; vml; vml = vml->next)
1003                 if (vml->vma->vm_start == addr)
1004                         goto found;
1005
1006         return (unsigned long) -EINVAL;
1007
1008  found:
1009         if (vml->vma->vm_end != vml->vma->vm_start + old_len)
1010                 return (unsigned long) -EFAULT;
1011
1012         if (vml->vma->vm_flags & VM_MAYSHARE)
1013                 return (unsigned long) -EPERM;
1014
1015         if (new_len > kobjsize((void *) addr))
1016                 return (unsigned long) -ENOMEM;
1017
1018         /* all checks complete - do it */
1019         vml->vma->vm_end = vml->vma->vm_start + new_len;
1020
1021         askedalloc -= old_len;
1022         askedalloc += new_len;
1023
1024         return vml->vma->vm_start;
1025 }
1026
1027 /*
1028  * Look up the first VMA which satisfies  addr < vm_end,  NULL if none
1029  */
1030 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
1031 {
1032         struct vm_list_struct *vml;
1033
1034         for (vml = mm->context.vmlist; vml; vml = vml->next)
1035                 if (addr >= vml->vma->vm_start && addr < vml->vma->vm_end)
1036                         return vml->vma;
1037
1038         return NULL;
1039 }
1040
1041 EXPORT_SYMBOL(find_vma);
1042
1043 struct page * follow_page(struct mm_struct *mm, unsigned long addr, int write)
1044 {
1045         return NULL;
1046 }
1047
1048 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
1049 {
1050         return NULL;
1051 }
1052
1053 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1054                 unsigned long to, unsigned long size, pgprot_t prot)
1055 {
1056         return -EPERM;
1057 }
1058
1059 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1060 {
1061 }
1062
1063 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1064         unsigned long len, unsigned long pgoff, unsigned long flags)
1065 {
1066         return -ENOMEM;
1067 }
1068
1069 void arch_unmap_area(struct vm_area_struct *area)
1070 {
1071 }
1072
1073 void update_mem_hiwater(struct task_struct *tsk)
1074 {
1075         unsigned long rss = get_mm_counter(tsk->mm, rss);
1076
1077         if (likely(tsk->mm)) {
1078                 if (tsk->mm->hiwater_rss < rss)
1079                         tsk->mm->hiwater_rss = rss;
1080                 if (tsk->mm->hiwater_vm < tsk->mm->total_vm)
1081                         tsk->mm->hiwater_vm = tsk->mm->total_vm;
1082         }
1083 }
1084
1085 void unmap_mapping_range(struct address_space *mapping,
1086                          loff_t const holebegin, loff_t const holelen,
1087                          int even_cows)
1088 {
1089 }
1090
1091 /*
1092  * Check that a process has enough memory to allocate a new virtual
1093  * mapping. 0 means there is enough memory for the allocation to
1094  * succeed and -ENOMEM implies there is not.
1095  *
1096  * We currently support three overcommit policies, which are set via the
1097  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1098  *
1099  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1100  * Additional code 2002 Jul 20 by Robert Love.
1101  *
1102  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1103  *
1104  * Note this is a helper function intended to be used by LSMs which
1105  * wish to use this logic.
1106  */
1107 int __vm_enough_memory(long pages, int cap_sys_admin)
1108 {
1109         unsigned long free, allowed;
1110
1111         vm_acct_memory(pages);
1112
1113         /*
1114          * Sometimes we want to use more memory than we have
1115          */
1116         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1117                 return 0;
1118
1119         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1120                 unsigned long n;
1121
1122                 free = get_page_cache_size();
1123                 free += nr_swap_pages;
1124
1125                 /*
1126                  * Any slabs which are created with the
1127                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1128                  * which are reclaimable, under pressure.  The dentry
1129                  * cache and most inode caches should fall into this
1130                  */
1131                 free += atomic_read(&slab_reclaim_pages);
1132
1133                 /*
1134                  * Leave the last 3% for root
1135                  */
1136                 if (!cap_sys_admin)
1137                         free -= free / 32;
1138
1139                 if (free > pages)
1140                         return 0;
1141
1142                 /*
1143                  * nr_free_pages() is very expensive on large systems,
1144                  * only call if we're about to fail.
1145                  */
1146                 n = nr_free_pages();
1147                 if (!cap_sys_admin)
1148                         n -= n / 32;
1149                 free += n;
1150
1151                 if (free > pages)
1152                         return 0;
1153                 vm_unacct_memory(pages);
1154                 return -ENOMEM;
1155         }
1156
1157         allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1158         /*
1159          * Leave the last 3% for root
1160          */
1161         if (!cap_sys_admin)
1162                 allowed -= allowed / 32;
1163         allowed += total_swap_pages;
1164
1165         /* Don't let a single process grow too big:
1166            leave 3% of the size of this process for other processes */
1167         allowed -= current->mm->total_vm / 32;
1168
1169         if (atomic_read(&vm_committed_space) < allowed)
1170                 return 0;
1171
1172         vm_unacct_memory(pages);
1173
1174         return -ENOMEM;
1175 }
1176
1177 int in_gate_area_no_task(unsigned long addr)
1178 {
1179         return 0;
1180 }