git.oblomov.eu Git - 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 EXPORT_SYMBOL(vmalloc);
  61 EXPORT_SYMBOL(vfree);
  62 EXPORT_SYMBOL(vmalloc_to_page);
  63 EXPORT_SYMBOL(vmalloc_32);
  64
  65 /*
  66  * Handle all mappings that got truncated by a "truncate()"
  67  * system call.
  68  *
  69  * NOTE! We have to be ready to update the memory sharing
  70  * between the file and the memory map for a potential last
  71  * incomplete page.  Ugly, but necessary.
  72  */
  73 int vmtruncate(struct inode *inode, loff_t offset)
  74 {
  75         struct address_space *mapping = inode->i_mapping;
  76         unsigned long limit;
  77
  78         if (inode->i_size < offset)
  79                 goto do_expand;
  80         i_size_write(inode, offset);
  81
  82         truncate_inode_pages(mapping, offset);
  83         goto out_truncate;
  84
  85 do_expand:
  86         limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
  87         if (limit != RLIM_INFINITY && offset > limit)
  88                 goto out_sig;
  89         if (offset > inode->i_sb->s_maxbytes)
  90                 goto out;
  91         i_size_write(inode, offset);
  92
  93 out_truncate:
  94         if (inode->i_op && inode->i_op->truncate)
  95                 inode->i_op->truncate(inode);
  96         return 0;
  97 out_sig:
  98         send_sig(SIGXFSZ, current, 0);
  99 out:
 100         return -EFBIG;
 101 }
 102
 103 EXPORT_SYMBOL(vmtruncate);
 104
 105 /*
 106  * Return the total memory allocated for this pointer, not
 107  * just what the caller asked for.
 108  *
 109  * Doesn't have to be accurate, i.e. may have races.
 110  */
 111 unsigned int kobjsize(const void *objp)
 112 {
 113         struct page *page;
 114
 115         if (!objp || !((page = virt_to_page(objp))))
 116                 return 0;
 117
 118         if (PageSlab(page))
 119                 return ksize(objp);
 120
 121         BUG_ON(page->index < 0);
 122         BUG_ON(page->index >= MAX_ORDER);
 123
 124         return (PAGE_SIZE << page->index);
 125 }
 126
 127 /*
 128  * The nommu dodgy version :-)
 129  */
 130 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
 131         unsigned long start, int len, int write, int force,
 132         struct page **pages, struct vm_area_struct **vmas)
 133 {
 134         int i;
 135         static struct vm_area_struct dummy_vma;
 136
 137         for (i = 0; i < len; i++) {
 138                 if (pages) {
 139                         pages[i] = virt_to_page(start);
 140                         if (pages[i])
 141                                 page_cache_get(pages[i]);
 142                 }
 143                 if (vmas)
 144                         vmas[i] = &dummy_vma;
 145                 start += PAGE_SIZE;
 146         }
 147         return(i);
 148 }
 149
 150 EXPORT_SYMBOL(get_user_pages);
 151
 152 DEFINE_RWLOCK(vmlist_lock);
 153 struct vm_struct *vmlist;
 154
 155 void vfree(void *addr)
 156 {
 157         kfree(addr);
 158 }
 159
 160 void *__vmalloc(unsigned long size, unsigned int __nocast gfp_mask,
 161                         pgprot_t prot)
 162 {
 163         /*
 164          * kmalloc doesn't like __GFP_HIGHMEM for some reason
 165          */
 166         return kmalloc(size, gfp_mask & ~__GFP_HIGHMEM);
 167 }
 168
 169 struct page * vmalloc_to_page(void *addr)
 170 {
 171         return virt_to_page(addr);
 172 }
 173
 174 unsigned long vmalloc_to_pfn(void *addr)
 175 {
 176         return page_to_pfn(virt_to_page(addr));
 177 }
 178
 179
 180 long vread(char *buf, char *addr, unsigned long count)
 181 {
 182         memcpy(buf, addr, count);
 183         return count;
 184 }
 185
 186 long vwrite(char *buf, char *addr, unsigned long count)
 187 {
 188         /* Don't allow overflow */
 189         if ((unsigned long) addr + count < count)
 190                 count = -(unsigned long) addr;
 191
 192         memcpy(addr, buf, count);
 193         return(count);
 194 }
 195
 196 /*
 197  *      vmalloc  -  allocate virtually continguos memory
 198  *
 199  *      @size:          allocation size
 200  *
 201  *      Allocate enough pages to cover @size from the page level
 202  *      allocator and map them into continguos kernel virtual space.
 203  *
 204  *      For tight cotrol over page level allocator and protection flags
 205  *      use __vmalloc() instead.
 206  */
 207 void *vmalloc(unsigned long size)
 208 {
 209        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
 210 }
 211
 212 /*
 213  *      vmalloc_32  -  allocate virtually continguos memory (32bit addressable)
 214  *
 215  *      @size:          allocation size
 216  *
 217  *      Allocate enough 32bit PA addressable pages to cover @size from the
 218  *      page level allocator and map them into continguos kernel virtual space.
 219  */
 220 void *vmalloc_32(unsigned long size)
 221 {
 222         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
 223 }
 224
 225 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
 226 {
 227         BUG();
 228         return NULL;
 229 }
 230
 231 void vunmap(void *addr)
 232 {
 233         BUG();
 234 }
 235
 236 /*
 237  *  sys_brk() for the most part doesn't need the global kernel
 238  *  lock, except when an application is doing something nasty
 239  *  like trying to un-brk an area that has already been mapped
 240  *  to a regular file.  in this case, the unmapping will need
 241  *  to invoke file system routines that need the global lock.
 242  */
 243 asmlinkage unsigned long sys_brk(unsigned long brk)
 244 {
 245         struct mm_struct *mm = current->mm;
 246
 247         if (brk < mm->start_brk || brk > mm->context.end_brk)
 248                 return mm->brk;
 249
 250         if (mm->brk == brk)
 251                 return mm->brk;
 252
 253         /*
 254          * Always allow shrinking brk
 255          */
 256         if (brk <= mm->brk) {
 257                 mm->brk = brk;
 258                 return brk;
 259         }
 260
 261         /*
 262          * Ok, looks good - let it rip.
 263          */
 264         return mm->brk = brk;
 265 }
 266
 267 #ifdef DEBUG
 268 static void show_process_blocks(void)
 269 {
 270         struct vm_list_struct *vml;
 271
 272         printk("Process blocks %d:", current->pid);
 273
 274         for (vml = &current->mm->context.vmlist; vml; vml = vml->next) {
 275                 printk(" %p: %p", vml, vml->vma);
 276                 if (vml->vma)
 277                         printk(" (%d @%lx #%d)",
 278                                kobjsize((void *) vml->vma->vm_start),
 279                                vml->vma->vm_start,
 280                                atomic_read(&vml->vma->vm_usage));
 281                 printk(vml->next ? " ->" : ".\n");
 282         }
 283 }
 284 #endif /* DEBUG */
 285
 286 static inline struct vm_area_struct *find_nommu_vma(unsigned long start)
 287 {
 288         struct vm_area_struct *vma;
 289         struct rb_node *n = nommu_vma_tree.rb_node;
 290
 291         while (n) {
 292                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
 293
 294                 if (start < vma->vm_start)
 295                         n = n->rb_left;
 296                 else if (start > vma->vm_start)
 297                         n = n->rb_right;
 298                 else
 299                         return vma;
 300         }
 301
 302         return NULL;
 303 }
 304
 305 static void add_nommu_vma(struct vm_area_struct *vma)
 306 {
 307         struct vm_area_struct *pvma;
 308         struct address_space *mapping;
 309         struct rb_node **p = &nommu_vma_tree.rb_node;
 310         struct rb_node *parent = NULL;
 311
 312         /* add the VMA to the mapping */
 313         if (vma->vm_file) {
 314                 mapping = vma->vm_file->f_mapping;
 315
 316                 flush_dcache_mmap_lock(mapping);
 317                 vma_prio_tree_insert(vma, &mapping->i_mmap);
 318                 flush_dcache_mmap_unlock(mapping);
 319         }
 320
 321         /* add the VMA to the master list */
 322         while (*p) {
 323                 parent = *p;
 324                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
 325
 326                 if (vma->vm_start < pvma->vm_start) {
 327                         p = &(*p)->rb_left;
 328                 }
 329                 else if (vma->vm_start > pvma->vm_start) {
 330                         p = &(*p)->rb_right;
 331                 }
 332                 else {
 333                         /* mappings are at the same address - this can only
 334                          * happen for shared-mem chardevs and shared file
 335                          * mappings backed by ramfs/tmpfs */
 336                         BUG_ON(!(pvma->vm_flags & VM_SHARED));
 337
 338                         if (vma < pvma)
 339                                 p = &(*p)->rb_left;
 340                         else if (vma > pvma)
 341                                 p = &(*p)->rb_right;
 342                         else
 343                                 BUG();
 344                 }
 345         }
 346
 347         rb_link_node(&vma->vm_rb, parent, p);
 348         rb_insert_color(&vma->vm_rb, &nommu_vma_tree);
 349 }
 350
 351 static void delete_nommu_vma(struct vm_area_struct *vma)
 352 {
 353         struct address_space *mapping;
 354
 355         /* remove the VMA from the mapping */
 356         if (vma->vm_file) {
 357                 mapping = vma->vm_file->f_mapping;
 358
 359                 flush_dcache_mmap_lock(mapping);
 360                 vma_prio_tree_remove(vma, &mapping->i_mmap);
 361                 flush_dcache_mmap_unlock(mapping);
 362         }
 363
 364         /* remove from the master list */
 365         rb_erase(&vma->vm_rb, &nommu_vma_tree);
 366 }
 367
 368 /*
 369  * determine whether a mapping should be permitted and, if so, what sort of
 370  * mapping we're capable of supporting
 371  */
 372 static int validate_mmap_request(struct file *file,
 373                                  unsigned long addr,
 374                                  unsigned long len,
 375                                  unsigned long prot,
 376                                  unsigned long flags,
 377                                  unsigned long pgoff,
 378                                  unsigned long *_capabilities)
 379 {
 380         unsigned long capabilities;
 381         unsigned long reqprot = prot;
 382         int ret;
 383
 384         /* do the simple checks first */
 385         if (flags & MAP_FIXED || addr) {
 386                 printk(KERN_DEBUG
 387                        "%d: Can't do fixed-address/overlay mmap of RAM\n",
 388                        current->pid);
 389                 return -EINVAL;
 390         }
 391
 392         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
 393             (flags & MAP_TYPE) != MAP_SHARED)
 394                 return -EINVAL;
 395
 396         if (PAGE_ALIGN(len) == 0)
 397                 return addr;
 398
 399         if (len > TASK_SIZE)
 400                 return -EINVAL;
 401
 402         /* offset overflow? */
 403         if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
 404                 return -EINVAL;
 405
 406         if (file) {
 407                 /* validate file mapping requests */
 408                 struct address_space *mapping;
 409
 410                 /* files must support mmap */
 411                 if (!file->f_op || !file->f_op->mmap)
 412                         return -ENODEV;
 413
 414                 /* work out if what we've got could possibly be shared
 415                  * - we support chardevs that provide their own "memory"
 416                  * - we support files/blockdevs that are memory backed
 417                  */
 418                 mapping = file->f_mapping;
 419                 if (!mapping)
 420                         mapping = file->f_dentry->d_inode->i_mapping;
 421
 422                 capabilities = 0;
 423                 if (mapping && mapping->backing_dev_info)
 424                         capabilities = mapping->backing_dev_info->capabilities;
 425
 426                 if (!capabilities) {
 427                         /* no explicit capabilities set, so assume some
 428                          * defaults */
 429                         switch (file->f_dentry->d_inode->i_mode & S_IFMT) {
 430                         case S_IFREG:
 431                         case S_IFBLK:
 432                                 capabilities = BDI_CAP_MAP_COPY;
 433                                 break;
 434
 435                         case S_IFCHR:
 436                                 capabilities =
 437                                         BDI_CAP_MAP_DIRECT |
 438                                         BDI_CAP_READ_MAP |
 439                                         BDI_CAP_WRITE_MAP;
 440                                 break;
 441
 442                         default:
 443                                 return -EINVAL;
 444                         }
 445                 }
 446
 447                 /* eliminate any capabilities that we can't support on this
 448                  * device */
 449                 if (!file->f_op->get_unmapped_area)
 450                         capabilities &= ~BDI_CAP_MAP_DIRECT;
 451                 if (!file->f_op->read)
 452                         capabilities &= ~BDI_CAP_MAP_COPY;
 453
 454                 if (flags & MAP_SHARED) {
 455                         /* do checks for writing, appending and locking */
 456                         if ((prot & PROT_WRITE) &&
 457                             !(file->f_mode & FMODE_WRITE))
 458                                 return -EACCES;
 459
 460                         if (IS_APPEND(file->f_dentry->d_inode) &&
 461                             (file->f_mode & FMODE_WRITE))
 462                                 return -EACCES;
 463
 464                         if (locks_verify_locked(file->f_dentry->d_inode))
 465                                 return -EAGAIN;
 466
 467                         if (!(capabilities & BDI_CAP_MAP_DIRECT))
 468                                 return -ENODEV;
 469
 470                         if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
 471                             ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
 472                             ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
 473                             ) {
 474                                 printk("MAP_SHARED not completely supported on !MMU\n");
 475                                 return -EINVAL;
 476                         }
 477
 478                         /* we mustn't privatise shared mappings */
 479                         capabilities &= ~BDI_CAP_MAP_COPY;
 480                 }
 481                 else {
 482                         /* we're going to read the file into private memory we
 483                          * allocate */
 484                         if (!(capabilities & BDI_CAP_MAP_COPY))
 485                                 return -ENODEV;
 486
 487                         /* we don't permit a private writable mapping to be
 488                          * shared with the backing device */
 489                         if (prot & PROT_WRITE)
 490                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
 491                 }
 492
 493                 /* handle executable mappings and implied executable
 494                  * mappings */
 495                 if (file->f_vfsmnt->mnt_flags & MNT_NOEXEC) {
 496                         if (prot & PROT_EXEC)
 497                                 return -EPERM;
 498                 }
 499                 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
 500                         /* handle implication of PROT_EXEC by PROT_READ */
 501                         if (current->personality & READ_IMPLIES_EXEC) {
 502                                 if (capabilities & BDI_CAP_EXEC_MAP)
 503                                         prot |= PROT_EXEC;
 504                         }
 505                 }
 506                 else if ((prot & PROT_READ) &&
 507                          (prot & PROT_EXEC) &&
 508                          !(capabilities & BDI_CAP_EXEC_MAP)
 509                          ) {
 510                         /* backing file is not executable, try to copy */
 511                         capabilities &= ~BDI_CAP_MAP_DIRECT;
 512                 }
 513         }
 514         else {
 515                 /* anonymous mappings are always memory backed and can be
 516                  * privately mapped
 517                  */
 518                 capabilities = BDI_CAP_MAP_COPY;
 519
 520                 /* handle PROT_EXEC implication by PROT_READ */
 521                 if ((prot & PROT_READ) &&
 522                     (current->personality & READ_IMPLIES_EXEC))
 523                         prot |= PROT_EXEC;
 524         }
 525
 526         /* allow the security API to have its say */
 527         ret = security_file_mmap(file, reqprot, prot, flags);
 528         if (ret < 0)
 529                 return ret;
 530
 531         /* looks okay */
 532         *_capabilities = capabilities;
 533         return 0;
 534 }
 535
 536 /*
 537  * we've determined that we can make the mapping, now translate what we
 538  * now know into VMA flags
 539  */
 540 static unsigned long determine_vm_flags(struct file *file,
 541                                         unsigned long prot,
 542                                         unsigned long flags,
 543                                         unsigned long capabilities)
 544 {
 545         unsigned long vm_flags;
 546
 547         vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
 548         vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 549         /* vm_flags |= mm->def_flags; */
 550
 551         if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
 552                 /* attempt to share read-only copies of mapped file chunks */
 553                 if (file && !(prot & PROT_WRITE))
 554                         vm_flags |= VM_MAYSHARE;
 555         }
 556         else {
 557                 /* overlay a shareable mapping on the backing device or inode
 558                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
 559                  * romfs/cramfs */
 560                 if (flags & MAP_SHARED)
 561                         vm_flags |= VM_MAYSHARE | VM_SHARED;
 562                 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
 563                         vm_flags |= VM_MAYSHARE;
 564         }
 565
 566         /* refuse to let anyone share private mappings with this process if
 567          * it's being traced - otherwise breakpoints set in it may interfere
 568          * with another untraced process
 569          */
 570         if ((flags & MAP_PRIVATE) && (current->ptrace & PT_PTRACED))
 571                 vm_flags &= ~VM_MAYSHARE;
 572
 573         return vm_flags;
 574 }
 575
 576 /*
 577  * set up a shared mapping on a file
 578  */
 579 static int do_mmap_shared_file(struct vm_area_struct *vma, unsigned long len)
 580 {
 581         int ret;
 582
 583         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
 584         if (ret != -ENOSYS)
 585                 return ret;
 586
 587         /* getting an ENOSYS error indicates that direct mmap isn't
 588          * possible (as opposed to tried but failed) so we'll fall
 589          * through to making a private copy of the data and mapping
 590          * that if we can */
 591         return -ENODEV;
 592 }
 593
 594 /*
 595  * set up a private mapping or an anonymous shared mapping
 596  */
 597 static int do_mmap_private(struct vm_area_struct *vma, unsigned long len)
 598 {
 599         void *base;
 600         int ret;
 601
 602         /* invoke the file's mapping function so that it can keep track of
 603          * shared mappings on devices or memory
 604          * - VM_MAYSHARE will be set if it may attempt to share
 605          */
 606         if (vma->vm_file) {
 607                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
 608                 if (ret != -ENOSYS) {
 609                         /* shouldn't return success if we're not sharing */
 610                         BUG_ON(ret == 0 && !(vma->vm_flags & VM_MAYSHARE));
 611                         return ret; /* success or a real error */
 612                 }
 613
 614                 /* getting an ENOSYS error indicates that direct mmap isn't
 615                  * possible (as opposed to tried but failed) so we'll try to
 616                  * make a private copy of the data and map that instead */
 617         }
 618
 619         /* allocate some memory to hold the mapping
 620          * - note that this may not return a page-aligned address if the object
 621          *   we're allocating is smaller than a page
 622          */
 623         base = kmalloc(len, GFP_KERNEL);
 624         if (!base)
 625                 goto enomem;
 626
 627         vma->vm_start = (unsigned long) base;
 628         vma->vm_end = vma->vm_start + len;
 629         vma->vm_flags |= VM_MAPPED_COPY;
 630
 631 #ifdef WARN_ON_SLACK
 632         if (len + WARN_ON_SLACK <= kobjsize(result))
 633                 printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n",
 634                        len, current->pid, kobjsize(result) - len);
 635 #endif
 636
 637         if (vma->vm_file) {
 638                 /* read the contents of a file into the copy */
 639                 mm_segment_t old_fs;
 640                 loff_t fpos;
 641
 642                 fpos = vma->vm_pgoff;
 643                 fpos <<= PAGE_SHIFT;
 644
 645                 old_fs = get_fs();
 646                 set_fs(KERNEL_DS);
 647                 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
 648                 set_fs(old_fs);
 649
 650                 if (ret < 0)
 651                         goto error_free;
 652
 653                 /* clear the last little bit */
 654                 if (ret < len)
 655                         memset(base + ret, 0, len - ret);
 656
 657         } else {
 658                 /* if it's an anonymous mapping, then just clear it */
 659                 memset(base, 0, len);
 660         }
 661
 662         return 0;
 663
 664 error_free:
 665         kfree(base);
 666         vma->vm_start = 0;
 667         return ret;
 668
 669 enomem:
 670         printk("Allocation of length %lu from process %d failed\n",
 671                len, current->pid);
 672         show_free_areas();
 673         return -ENOMEM;
 674 }
 675
 676 /*
 677  * handle mapping creation for uClinux
 678  */
 679 unsigned long do_mmap_pgoff(struct file *file,
 680                             unsigned long addr,
 681                             unsigned long len,
 682                             unsigned long prot,
 683                             unsigned long flags,
 684                             unsigned long pgoff)
 685 {
 686         struct vm_list_struct *vml = NULL;
 687         struct vm_area_struct *vma = NULL;
 688         struct rb_node *rb;
 689         unsigned long capabilities, vm_flags;
 690         void *result;
 691         int ret;
 692
 693         /* decide whether we should attempt the mapping, and if so what sort of
 694          * mapping */
 695         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
 696                                     &capabilities);
 697         if (ret < 0)
 698                 return ret;
 699
 700         /* we've determined that we can make the mapping, now translate what we
 701          * now know into VMA flags */
 702         vm_flags = determine_vm_flags(file, prot, flags, capabilities);
 703
 704         /* we're going to need to record the mapping if it works */
 705         vml = kmalloc(sizeof(struct vm_list_struct), GFP_KERNEL);
 706         if (!vml)
 707                 goto error_getting_vml;
 708         memset(vml, 0, sizeof(*vml));
 709
 710         down_write(&nommu_vma_sem);
 711
 712         /* if we want to share, we need to check for VMAs created by other
 713          * mmap() calls that overlap with our proposed mapping
 714          * - we can only share with an exact match on most regular files
 715          * - shared mappings on character devices and memory backed files are
 716          *   permitted to overlap inexactly as far as we are concerned for in
 717          *   these cases, sharing is handled in the driver or filesystem rather
 718          *   than here
 719          */
 720         if (vm_flags & VM_MAYSHARE) {
 721                 unsigned long pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 722                 unsigned long vmpglen;
 723
 724                 for (rb = rb_first(&nommu_vma_tree); rb; rb = rb_next(rb)) {
 725                         vma = rb_entry(rb, struct vm_area_struct, vm_rb);
 726
 727                         if (!(vma->vm_flags & VM_MAYSHARE))
 728                                 continue;
 729
 730                         /* search for overlapping mappings on the same file */
 731                         if (vma->vm_file->f_dentry->d_inode != file->f_dentry->d_inode)
 732                                 continue;
 733
 734                         if (vma->vm_pgoff >= pgoff + pglen)
 735                                 continue;
 736
 737                         vmpglen = vma->vm_end - vma->vm_start + PAGE_SIZE - 1;
 738                         vmpglen >>= PAGE_SHIFT;
 739                         if (pgoff >= vma->vm_pgoff + vmpglen)
 740                                 continue;
 741
 742                         /* handle inexactly overlapping matches between mappings */
 743                         if (vma->vm_pgoff != pgoff || vmpglen != pglen) {
 744                                 if (!(capabilities & BDI_CAP_MAP_DIRECT))
 745                                         goto sharing_violation;
 746                                 continue;
 747                         }
 748
 749                         /* we've found a VMA we can share */
 750                         atomic_inc(&vma->vm_usage);
 751
 752                         vml->vma = vma;
 753                         result = (void *) vma->vm_start;
 754                         goto shared;
 755                 }
 756
 757                 vma = NULL;
 758
 759                 /* obtain the address at which to make a shared mapping
 760                  * - this is the hook for quasi-memory character devices to
 761                  *   tell us the location of a shared mapping
 762                  */
 763                 if (file && file->f_op->get_unmapped_area) {
 764                         addr = file->f_op->get_unmapped_area(file, addr, len,
 765                                                              pgoff, flags);
 766                         if (IS_ERR((void *) addr)) {
 767                                 ret = addr;
 768                                 if (ret != (unsigned long) -ENOSYS)
 769                                         goto error;
 770
 771                                 /* the driver refused to tell us where to site
 772                                  * the mapping so we'll have to attempt to copy
 773                                  * it */
 774                                 ret = (unsigned long) -ENODEV;
 775                                 if (!(capabilities & BDI_CAP_MAP_COPY))
 776                                         goto error;
 777
 778                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
 779                         }
 780                 }
 781         }
 782
 783         /* we're going to need a VMA struct as well */
 784         vma = kmalloc(sizeof(struct vm_area_struct), GFP_KERNEL);
 785         if (!vma)
 786                 goto error_getting_vma;
 787
 788         memset(vma, 0, sizeof(*vma));
 789         INIT_LIST_HEAD(&vma->anon_vma_node);
 790         atomic_set(&vma->vm_usage, 1);
 791         if (file)
 792                 get_file(file);
 793         vma->vm_file    = file;
 794         vma->vm_flags   = vm_flags;
 795         vma->vm_start   = addr;
 796         vma->vm_end     = addr + len;
 797         vma->vm_pgoff   = pgoff;
 798
 799         vml->vma = vma;
 800
 801         /* set up the mapping */
 802         if (file && vma->vm_flags & VM_SHARED)
 803                 ret = do_mmap_shared_file(vma, len);
 804         else
 805                 ret = do_mmap_private(vma, len);
 806         if (ret < 0)
 807                 goto error;
 808
 809         /* okay... we have a mapping; now we have to register it */
 810         result = (void *) vma->vm_start;
 811
 812         if (vma->vm_flags & VM_MAPPED_COPY) {
 813                 realalloc += kobjsize(result);
 814                 askedalloc += len;
 815         }
 816
 817         realalloc += kobjsize(vma);
 818         askedalloc += sizeof(*vma);
 819
 820         current->mm->total_vm += len >> PAGE_SHIFT;
 821
 822         add_nommu_vma(vma);
 823
 824  shared:
 825         realalloc += kobjsize(vml);
 826         askedalloc += sizeof(*vml);
 827
 828         vml->next = current->mm->context.vmlist;
 829         current->mm->context.vmlist = vml;
 830
 831         up_write(&nommu_vma_sem);
 832
 833         if (prot & PROT_EXEC)
 834                 flush_icache_range((unsigned long) result,
 835                                    (unsigned long) result + len);
 836
 837 #ifdef DEBUG
 838         printk("do_mmap:\n");
 839         show_process_blocks();
 840 #endif
 841
 842         return (unsigned long) result;
 843
 844  error:
 845         up_write(&nommu_vma_sem);
 846         kfree(vml);
 847         if (vma) {
 848                 fput(vma->vm_file);
 849                 kfree(vma);
 850         }
 851         return ret;
 852
 853  sharing_violation:
 854         up_write(&nommu_vma_sem);
 855         printk("Attempt to share mismatched mappings\n");
 856         kfree(vml);
 857         return -EINVAL;
 858
 859  error_getting_vma:
 860         up_write(&nommu_vma_sem);
 861         kfree(vml);
 862         printk("Allocation of vma for %lu byte allocation from process %d failed\n",
 863                len, current->pid);
 864         show_free_areas();
 865         return -ENOMEM;
 866
 867  error_getting_vml:
 868         printk("Allocation of vml for %lu byte allocation from process %d failed\n",
 869                len, current->pid);
 870         show_free_areas();
 871         return -ENOMEM;
 872 }
 873
 874 /*
 875  * handle mapping disposal for uClinux
 876  */
 877 static void put_vma(struct vm_area_struct *vma)
 878 {
 879         if (vma) {
 880                 down_write(&nommu_vma_sem);
 881
 882                 if (atomic_dec_and_test(&vma->vm_usage)) {
 883                         delete_nommu_vma(vma);
 884
 885                         if (vma->vm_ops && vma->vm_ops->close)
 886                                 vma->vm_ops->close(vma);
 887
 888                         /* IO memory and memory shared directly out of the pagecache from
 889                          * ramfs/tmpfs mustn't be released here */
 890                         if (vma->vm_flags & VM_MAPPED_COPY) {
 891                                 realalloc -= kobjsize((void *) vma->vm_start);
 892                                 askedalloc -= vma->vm_end - vma->vm_start;
 893                                 kfree((void *) vma->vm_start);
 894                         }
 895
 896                         realalloc -= kobjsize(vma);
 897                         askedalloc -= sizeof(*vma);
 898
 899                         if (vma->vm_file)
 900                                 fput(vma->vm_file);
 901                         kfree(vma);
 902                 }
 903
 904                 up_write(&nommu_vma_sem);
 905         }
 906 }
 907
 908 int do_munmap(struct mm_struct *mm, unsigned long addr, size_t len)
 909 {
 910         struct vm_list_struct *vml, **parent;
 911         unsigned long end = addr + len;
 912
 913 #ifdef DEBUG
 914         printk("do_munmap:\n");
 915 #endif
 916
 917         for (parent = &mm->context.vmlist; *parent; parent = &(*parent)->next)
 918                 if ((*parent)->vma->vm_start == addr &&
 919                     ((len == 0) || ((*parent)->vma->vm_end == end)))
 920                         goto found;
 921
 922         printk("munmap of non-mmaped memory by process %d (%s): %p\n",
 923                current->pid, current->comm, (void *) addr);
 924         return -EINVAL;
 925
 926  found:
 927         vml = *parent;
 928
 929         put_vma(vml->vma);
 930
 931         *parent = vml->next;
 932         realalloc -= kobjsize(vml);
 933         askedalloc -= sizeof(*vml);
 934         kfree(vml);
 935         mm->total_vm -= len >> PAGE_SHIFT;
 936
 937 #ifdef DEBUG
 938         show_process_blocks();
 939 #endif
 940
 941         return 0;
 942 }
 943
 944 /* Release all mmaps. */
 945 void exit_mmap(struct mm_struct * mm)
 946 {
 947         struct vm_list_struct *tmp;
 948
 949         if (mm) {
 950 #ifdef DEBUG
 951                 printk("Exit_mmap:\n");
 952 #endif
 953
 954                 mm->total_vm = 0;
 955
 956                 while ((tmp = mm->context.vmlist)) {
 957                         mm->context.vmlist = tmp->next;
 958                         put_vma(tmp->vma);
 959
 960                         realalloc -= kobjsize(tmp);
 961                         askedalloc -= sizeof(*tmp);
 962                         kfree(tmp);
 963                 }
 964
 965 #ifdef DEBUG
 966                 show_process_blocks();
 967 #endif
 968         }
 969 }
 970
 971 asmlinkage long sys_munmap(unsigned long addr, size_t len)
 972 {
 973         int ret;
 974         struct mm_struct *mm = current->mm;
 975
 976         down_write(&mm->mmap_sem);
 977         ret = do_munmap(mm, addr, len);
 978         up_write(&mm->mmap_sem);
 979         return ret;
 980 }
 981
 982 unsigned long do_brk(unsigned long addr, unsigned long len)
 983 {
 984         return -ENOMEM;
 985 }
 986
 987 /*
 988  * Expand (or shrink) an existing mapping, potentially moving it at the
 989  * same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
 990  *
 991  * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise
 992  * This option implies MREMAP_MAYMOVE.
 993  *
 994  * on uClinux, we only permit changing a mapping's size, and only as long as it stays within the
 995  * hole allocated by the kmalloc() call in do_mmap_pgoff() and the block is not shareable
 996  */
 997 unsigned long do_mremap(unsigned long addr,
 998                         unsigned long old_len, unsigned long new_len,
 999                         unsigned long flags, unsigned long new_addr)
1000 {
1001         struct vm_list_struct *vml = NULL;
1002
1003         /* insanity checks first */
1004         if (new_len == 0)
1005                 return (unsigned long) -EINVAL;
1006
1007         if (flags & MREMAP_FIXED && new_addr != addr)
1008                 return (unsigned long) -EINVAL;
1009
1010         for (vml = current->mm->context.vmlist; vml; vml = vml->next)
1011                 if (vml->vma->vm_start == addr)
1012                         goto found;
1013
1014         return (unsigned long) -EINVAL;
1015
1016  found:
1017         if (vml->vma->vm_end != vml->vma->vm_start + old_len)
1018                 return (unsigned long) -EFAULT;
1019
1020         if (vml->vma->vm_flags & VM_MAYSHARE)
1021                 return (unsigned long) -EPERM;
1022
1023         if (new_len > kobjsize((void *) addr))
1024                 return (unsigned long) -ENOMEM;
1025
1026         /* all checks complete - do it */
1027         vml->vma->vm_end = vml->vma->vm_start + new_len;
1028
1029         askedalloc -= old_len;
1030         askedalloc += new_len;
1031
1032         return vml->vma->vm_start;
1033 }
1034
1035 /*
1036  * Look up the first VMA which satisfies  addr < vm_end,  NULL if none
1037  */
1038 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
1039 {
1040         struct vm_list_struct *vml;
1041
1042         for (vml = mm->context.vmlist; vml; vml = vml->next)
1043                 if (addr >= vml->vma->vm_start && addr < vml->vma->vm_end)
1044                         return vml->vma;
1045
1046         return NULL;
1047 }
1048
1049 EXPORT_SYMBOL(find_vma);
1050
1051 struct page * follow_page(struct mm_struct *mm, unsigned long addr, int write)
1052 {
1053         return NULL;
1054 }
1055
1056 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
1057 {
1058         return NULL;
1059 }
1060
1061 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1062                 unsigned long to, unsigned long size, pgprot_t prot)
1063 {
1064         vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1065         return 0;
1066 }
1067
1068 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1069 {
1070 }
1071
1072 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1073         unsigned long len, unsigned long pgoff, unsigned long flags)
1074 {
1075         return -ENOMEM;
1076 }
1077
1078 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1079 {
1080 }
1081
1082 void update_mem_hiwater(struct task_struct *tsk)
1083 {
1084         unsigned long rss;
1085
1086         if (likely(tsk->mm)) {
1087                 rss = get_mm_counter(tsk->mm, rss);
1088                 if (tsk->mm->hiwater_rss < rss)
1089                         tsk->mm->hiwater_rss = rss;
1090                 if (tsk->mm->hiwater_vm < tsk->mm->total_vm)
1091                         tsk->mm->hiwater_vm = tsk->mm->total_vm;
1092         }
1093 }
1094
1095 void unmap_mapping_range(struct address_space *mapping,
1096                          loff_t const holebegin, loff_t const holelen,
1097                          int even_cows)
1098 {
1099 }
1100
1101 /*
1102  * Check that a process has enough memory to allocate a new virtual
1103  * mapping. 0 means there is enough memory for the allocation to
1104  * succeed and -ENOMEM implies there is not.
1105  *
1106  * We currently support three overcommit policies, which are set via the
1107  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1108  *
1109  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1110  * Additional code 2002 Jul 20 by Robert Love.
1111  *
1112  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1113  *
1114  * Note this is a helper function intended to be used by LSMs which
1115  * wish to use this logic.
1116  */
1117 int __vm_enough_memory(long pages, int cap_sys_admin)
1118 {
1119         unsigned long free, allowed;
1120
1121         vm_acct_memory(pages);
1122
1123         /*
1124          * Sometimes we want to use more memory than we have
1125          */
1126         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1127                 return 0;
1128
1129         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1130                 unsigned long n;
1131
1132                 free = get_page_cache_size();
1133                 free += nr_swap_pages;
1134
1135                 /*
1136                  * Any slabs which are created with the
1137                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1138                  * which are reclaimable, under pressure.  The dentry
1139                  * cache and most inode caches should fall into this
1140                  */
1141                 free += atomic_read(&slab_reclaim_pages);
1142
1143                 /*
1144                  * Leave the last 3% for root
1145                  */
1146                 if (!cap_sys_admin)
1147                         free -= free / 32;
1148
1149                 if (free > pages)
1150                         return 0;
1151
1152                 /*
1153                  * nr_free_pages() is very expensive on large systems,
1154                  * only call if we're about to fail.
1155                  */
1156                 n = nr_free_pages();
1157                 if (!cap_sys_admin)
1158                         n -= n / 32;
1159                 free += n;
1160
1161                 if (free > pages)
1162                         return 0;
1163                 vm_unacct_memory(pages);
1164                 return -ENOMEM;
1165         }
1166
1167         allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1168         /*
1169          * Leave the last 3% for root
1170          */
1171         if (!cap_sys_admin)
1172                 allowed -= allowed / 32;
1173         allowed += total_swap_pages;
1174
1175         /* Don't let a single process grow too big:
1176            leave 3% of the size of this process for other processes */
1177         allowed -= current->mm->total_vm / 32;
1178
1179         /*
1180          * cast `allowed' as a signed long because vm_committed_space
1181          * sometimes has a negative value
1182          */
1183         if (atomic_read(&vm_committed_space) < (long)allowed)
1184                 return 0;
1185
1186         vm_unacct_memory(pages);
1187
1188         return -ENOMEM;
1189 }
1190
1191 int in_gate_area_no_task(unsigned long addr)
1192 {
1193         return 0;
1194 }