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(__vm_enough_memory);
  48
  49 /* list of shareable VMAs */
  50 struct rb_root nommu_vma_tree = RB_ROOT;
  51 DECLARE_RWSEM(nommu_vma_sem);
  52
  53 struct vm_operations_struct generic_file_vm_ops = {
  54 };
  55
  56 EXPORT_SYMBOL(vmalloc);
  57 EXPORT_SYMBOL(vfree);
  58 EXPORT_SYMBOL(vmalloc_to_page);
  59 EXPORT_SYMBOL(vmalloc_32);
  60 EXPORT_SYMBOL(vmap);
  61 EXPORT_SYMBOL(vunmap);
  62
  63 /*
  64  * Handle all mappings that got truncated by a "truncate()"
  65  * system call.
  66  *
  67  * NOTE! We have to be ready to update the memory sharing
  68  * between the file and the memory map for a potential last
  69  * incomplete page.  Ugly, but necessary.
  70  */
  71 int vmtruncate(struct inode *inode, loff_t offset)
  72 {
  73         struct address_space *mapping = inode->i_mapping;
  74         unsigned long limit;
  75
  76         if (inode->i_size < offset)
  77                 goto do_expand;
  78         i_size_write(inode, offset);
  79
  80         truncate_inode_pages(mapping, offset);
  81         goto out_truncate;
  82
  83 do_expand:
  84         limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
  85         if (limit != RLIM_INFINITY && offset > limit)
  86                 goto out_sig;
  87         if (offset > inode->i_sb->s_maxbytes)
  88                 goto out;
  89         i_size_write(inode, offset);
  90
  91 out_truncate:
  92         if (inode->i_op && inode->i_op->truncate)
  93                 inode->i_op->truncate(inode);
  94         return 0;
  95 out_sig:
  96         send_sig(SIGXFSZ, current, 0);
  97 out:
  98         return -EFBIG;
  99 }
 100
 101 EXPORT_SYMBOL(vmtruncate);
 102
 103 /*
 104  * Return the total memory allocated for this pointer, not
 105  * just what the caller asked for.
 106  *
 107  * Doesn't have to be accurate, i.e. may have races.
 108  */
 109 unsigned int kobjsize(const void *objp)
 110 {
 111         struct page *page;
 112
 113         if (!objp || !((page = virt_to_page(objp))))
 114                 return 0;
 115
 116         if (PageSlab(page))
 117                 return ksize(objp);
 118
 119         BUG_ON(page->index < 0);
 120         BUG_ON(page->index >= MAX_ORDER);
 121
 122         return (PAGE_SIZE << page->index);
 123 }
 124
 125 /*
 126  * The nommu dodgy version :-)
 127  */
 128 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
 129         unsigned long start, int len, int write, int force,
 130         struct page **pages, struct vm_area_struct **vmas)
 131 {
 132         int i;
 133         static struct vm_area_struct dummy_vma;
 134
 135         for (i = 0; i < len; i++) {
 136                 if (pages) {
 137                         pages[i] = virt_to_page(start);
 138                         if (pages[i])
 139                                 page_cache_get(pages[i]);
 140                 }
 141                 if (vmas)
 142                         vmas[i] = &dummy_vma;
 143                 start += PAGE_SIZE;
 144         }
 145         return(i);
 146 }
 147
 148 EXPORT_SYMBOL(get_user_pages);
 149
 150 DEFINE_RWLOCK(vmlist_lock);
 151 struct vm_struct *vmlist;
 152
 153 void vfree(void *addr)
 154 {
 155         kfree(addr);
 156 }
 157
 158 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
 159 {
 160         /*
 161          * kmalloc doesn't like __GFP_HIGHMEM for some reason
 162          */
 163         return kmalloc(size, gfp_mask & ~__GFP_HIGHMEM);
 164 }
 165
 166 struct page * vmalloc_to_page(void *addr)
 167 {
 168         return virt_to_page(addr);
 169 }
 170
 171 unsigned long vmalloc_to_pfn(void *addr)
 172 {
 173         return page_to_pfn(virt_to_page(addr));
 174 }
 175
 176
 177 long vread(char *buf, char *addr, unsigned long count)
 178 {
 179         memcpy(buf, addr, count);
 180         return count;
 181 }
 182
 183 long vwrite(char *buf, char *addr, unsigned long count)
 184 {
 185         /* Don't allow overflow */
 186         if ((unsigned long) addr + count < count)
 187                 count = -(unsigned long) addr;
 188
 189         memcpy(addr, buf, count);
 190         return(count);
 191 }
 192
 193 /*
 194  *      vmalloc  -  allocate virtually continguos memory
 195  *
 196  *      @size:          allocation size
 197  *
 198  *      Allocate enough pages to cover @size from the page level
 199  *      allocator and map them into continguos kernel virtual space.
 200  *
 201  *      For tight cotrol over page level allocator and protection flags
 202  *      use __vmalloc() instead.
 203  */
 204 void *vmalloc(unsigned long size)
 205 {
 206        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
 207 }
 208
 209 /*
 210  *      vmalloc_32  -  allocate virtually continguos memory (32bit addressable)
 211  *
 212  *      @size:          allocation size
 213  *
 214  *      Allocate enough 32bit PA addressable pages to cover @size from the
 215  *      page level allocator and map them into continguos kernel virtual space.
 216  */
 217 void *vmalloc_32(unsigned long size)
 218 {
 219         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
 220 }
 221
 222 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
 223 {
 224         BUG();
 225         return NULL;
 226 }
 227
 228 void vunmap(void *addr)
 229 {
 230         BUG();
 231 }
 232
 233 /*
 234  *  sys_brk() for the most part doesn't need the global kernel
 235  *  lock, except when an application is doing something nasty
 236  *  like trying to un-brk an area that has already been mapped
 237  *  to a regular file.  in this case, the unmapping will need
 238  *  to invoke file system routines that need the global lock.
 239  */
 240 asmlinkage unsigned long sys_brk(unsigned long brk)
 241 {
 242         struct mm_struct *mm = current->mm;
 243
 244         if (brk < mm->start_brk || brk > mm->context.end_brk)
 245                 return mm->brk;
 246
 247         if (mm->brk == brk)
 248                 return mm->brk;
 249
 250         /*
 251          * Always allow shrinking brk
 252          */
 253         if (brk <= mm->brk) {
 254                 mm->brk = brk;
 255                 return brk;
 256         }
 257
 258         /*
 259          * Ok, looks good - let it rip.
 260          */
 261         return mm->brk = brk;
 262 }
 263
 264 #ifdef DEBUG
 265 static void show_process_blocks(void)
 266 {
 267         struct vm_list_struct *vml;
 268
 269         printk("Process blocks %d:", current->pid);
 270
 271         for (vml = &current->mm->context.vmlist; vml; vml = vml->next) {
 272                 printk(" %p: %p", vml, vml->vma);
 273                 if (vml->vma)
 274                         printk(" (%d @%lx #%d)",
 275                                kobjsize((void *) vml->vma->vm_start),
 276                                vml->vma->vm_start,
 277                                atomic_read(&vml->vma->vm_usage));
 278                 printk(vml->next ? " ->" : ".\n");
 279         }
 280 }
 281 #endif /* DEBUG */
 282
 283 static inline struct vm_area_struct *find_nommu_vma(unsigned long start)
 284 {
 285         struct vm_area_struct *vma;
 286         struct rb_node *n = nommu_vma_tree.rb_node;
 287
 288         while (n) {
 289                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
 290
 291                 if (start < vma->vm_start)
 292                         n = n->rb_left;
 293                 else if (start > vma->vm_start)
 294                         n = n->rb_right;
 295                 else
 296                         return vma;
 297         }
 298
 299         return NULL;
 300 }
 301
 302 static void add_nommu_vma(struct vm_area_struct *vma)
 303 {
 304         struct vm_area_struct *pvma;
 305         struct address_space *mapping;
 306         struct rb_node **p = &nommu_vma_tree.rb_node;
 307         struct rb_node *parent = NULL;
 308
 309         /* add the VMA to the mapping */
 310         if (vma->vm_file) {
 311                 mapping = vma->vm_file->f_mapping;
 312
 313                 flush_dcache_mmap_lock(mapping);
 314                 vma_prio_tree_insert(vma, &mapping->i_mmap);
 315                 flush_dcache_mmap_unlock(mapping);
 316         }
 317
 318         /* add the VMA to the master list */
 319         while (*p) {
 320                 parent = *p;
 321                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
 322
 323                 if (vma->vm_start < pvma->vm_start) {
 324                         p = &(*p)->rb_left;
 325                 }
 326                 else if (vma->vm_start > pvma->vm_start) {
 327                         p = &(*p)->rb_right;
 328                 }
 329                 else {
 330                         /* mappings are at the same address - this can only
 331                          * happen for shared-mem chardevs and shared file
 332                          * mappings backed by ramfs/tmpfs */
 333                         BUG_ON(!(pvma->vm_flags & VM_SHARED));
 334
 335                         if (vma < pvma)
 336                                 p = &(*p)->rb_left;
 337                         else if (vma > pvma)
 338                                 p = &(*p)->rb_right;
 339                         else
 340                                 BUG();
 341                 }
 342         }
 343
 344         rb_link_node(&vma->vm_rb, parent, p);
 345         rb_insert_color(&vma->vm_rb, &nommu_vma_tree);
 346 }
 347
 348 static void delete_nommu_vma(struct vm_area_struct *vma)
 349 {
 350         struct address_space *mapping;
 351
 352         /* remove the VMA from the mapping */
 353         if (vma->vm_file) {
 354                 mapping = vma->vm_file->f_mapping;
 355
 356                 flush_dcache_mmap_lock(mapping);
 357                 vma_prio_tree_remove(vma, &mapping->i_mmap);
 358                 flush_dcache_mmap_unlock(mapping);
 359         }
 360
 361         /* remove from the master list */
 362         rb_erase(&vma->vm_rb, &nommu_vma_tree);
 363 }
 364
 365 /*
 366  * determine whether a mapping should be permitted and, if so, what sort of
 367  * mapping we're capable of supporting
 368  */
 369 static int validate_mmap_request(struct file *file,
 370                                  unsigned long addr,
 371                                  unsigned long len,
 372                                  unsigned long prot,
 373                                  unsigned long flags,
 374                                  unsigned long pgoff,
 375                                  unsigned long *_capabilities)
 376 {
 377         unsigned long capabilities;
 378         unsigned long reqprot = prot;
 379         int ret;
 380
 381         /* do the simple checks first */
 382         if (flags & MAP_FIXED || addr) {
 383                 printk(KERN_DEBUG
 384                        "%d: Can't do fixed-address/overlay mmap of RAM\n",
 385                        current->pid);
 386                 return -EINVAL;
 387         }
 388
 389         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
 390             (flags & MAP_TYPE) != MAP_SHARED)
 391                 return -EINVAL;
 392
 393         if (PAGE_ALIGN(len) == 0)
 394                 return addr;
 395
 396         if (len > TASK_SIZE)
 397                 return -EINVAL;
 398
 399         /* offset overflow? */
 400         if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
 401                 return -EINVAL;
 402
 403         if (file) {
 404                 /* validate file mapping requests */
 405                 struct address_space *mapping;
 406
 407                 /* files must support mmap */
 408                 if (!file->f_op || !file->f_op->mmap)
 409                         return -ENODEV;
 410
 411                 /* work out if what we've got could possibly be shared
 412                  * - we support chardevs that provide their own "memory"
 413                  * - we support files/blockdevs that are memory backed
 414                  */
 415                 mapping = file->f_mapping;
 416                 if (!mapping)
 417                         mapping = file->f_dentry->d_inode->i_mapping;
 418
 419                 capabilities = 0;
 420                 if (mapping && mapping->backing_dev_info)
 421                         capabilities = mapping->backing_dev_info->capabilities;
 422
 423                 if (!capabilities) {
 424                         /* no explicit capabilities set, so assume some
 425                          * defaults */
 426                         switch (file->f_dentry->d_inode->i_mode & S_IFMT) {
 427                         case S_IFREG:
 428                         case S_IFBLK:
 429                                 capabilities = BDI_CAP_MAP_COPY;
 430                                 break;
 431
 432                         case S_IFCHR:
 433                                 capabilities =
 434                                         BDI_CAP_MAP_DIRECT |
 435                                         BDI_CAP_READ_MAP |
 436                                         BDI_CAP_WRITE_MAP;
 437                                 break;
 438
 439                         default:
 440                                 return -EINVAL;
 441                         }
 442                 }
 443
 444                 /* eliminate any capabilities that we can't support on this
 445                  * device */
 446                 if (!file->f_op->get_unmapped_area)
 447                         capabilities &= ~BDI_CAP_MAP_DIRECT;
 448                 if (!file->f_op->read)
 449                         capabilities &= ~BDI_CAP_MAP_COPY;
 450
 451                 if (flags & MAP_SHARED) {
 452                         /* do checks for writing, appending and locking */
 453                         if ((prot & PROT_WRITE) &&
 454                             !(file->f_mode & FMODE_WRITE))
 455                                 return -EACCES;
 456
 457                         if (IS_APPEND(file->f_dentry->d_inode) &&
 458                             (file->f_mode & FMODE_WRITE))
 459                                 return -EACCES;
 460
 461                         if (locks_verify_locked(file->f_dentry->d_inode))
 462                                 return -EAGAIN;
 463
 464                         if (!(capabilities & BDI_CAP_MAP_DIRECT))
 465                                 return -ENODEV;
 466
 467                         if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
 468                             ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
 469                             ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
 470                             ) {
 471                                 printk("MAP_SHARED not completely supported on !MMU\n");
 472                                 return -EINVAL;
 473                         }
 474
 475                         /* we mustn't privatise shared mappings */
 476                         capabilities &= ~BDI_CAP_MAP_COPY;
 477                 }
 478                 else {
 479                         /* we're going to read the file into private memory we
 480                          * allocate */
 481                         if (!(capabilities & BDI_CAP_MAP_COPY))
 482                                 return -ENODEV;
 483
 484                         /* we don't permit a private writable mapping to be
 485                          * shared with the backing device */
 486                         if (prot & PROT_WRITE)
 487                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
 488                 }
 489
 490                 /* handle executable mappings and implied executable
 491                  * mappings */
 492                 if (file->f_vfsmnt->mnt_flags & MNT_NOEXEC) {
 493                         if (prot & PROT_EXEC)
 494                                 return -EPERM;
 495                 }
 496                 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
 497                         /* handle implication of PROT_EXEC by PROT_READ */
 498                         if (current->personality & READ_IMPLIES_EXEC) {
 499                                 if (capabilities & BDI_CAP_EXEC_MAP)
 500                                         prot |= PROT_EXEC;
 501                         }
 502                 }
 503                 else if ((prot & PROT_READ) &&
 504                          (prot & PROT_EXEC) &&
 505                          !(capabilities & BDI_CAP_EXEC_MAP)
 506                          ) {
 507                         /* backing file is not executable, try to copy */
 508                         capabilities &= ~BDI_CAP_MAP_DIRECT;
 509                 }
 510         }
 511         else {
 512                 /* anonymous mappings are always memory backed and can be
 513                  * privately mapped
 514                  */
 515                 capabilities = BDI_CAP_MAP_COPY;
 516
 517                 /* handle PROT_EXEC implication by PROT_READ */
 518                 if ((prot & PROT_READ) &&
 519                     (current->personality & READ_IMPLIES_EXEC))
 520                         prot |= PROT_EXEC;
 521         }
 522
 523         /* allow the security API to have its say */
 524         ret = security_file_mmap(file, reqprot, prot, flags);
 525         if (ret < 0)
 526                 return ret;
 527
 528         /* looks okay */
 529         *_capabilities = capabilities;
 530         return 0;
 531 }
 532
 533 /*
 534  * we've determined that we can make the mapping, now translate what we
 535  * now know into VMA flags
 536  */
 537 static unsigned long determine_vm_flags(struct file *file,
 538                                         unsigned long prot,
 539                                         unsigned long flags,
 540                                         unsigned long capabilities)
 541 {
 542         unsigned long vm_flags;
 543
 544         vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
 545         vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 546         /* vm_flags |= mm->def_flags; */
 547
 548         if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
 549                 /* attempt to share read-only copies of mapped file chunks */
 550                 if (file && !(prot & PROT_WRITE))
 551                         vm_flags |= VM_MAYSHARE;
 552         }
 553         else {
 554                 /* overlay a shareable mapping on the backing device or inode
 555                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
 556                  * romfs/cramfs */
 557                 if (flags & MAP_SHARED)
 558                         vm_flags |= VM_MAYSHARE | VM_SHARED;
 559                 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
 560                         vm_flags |= VM_MAYSHARE;
 561         }
 562
 563         /* refuse to let anyone share private mappings with this process if
 564          * it's being traced - otherwise breakpoints set in it may interfere
 565          * with another untraced process
 566          */
 567         if ((flags & MAP_PRIVATE) && (current->ptrace & PT_PTRACED))
 568                 vm_flags &= ~VM_MAYSHARE;
 569
 570         return vm_flags;
 571 }
 572
 573 /*
 574  * set up a shared mapping on a file
 575  */
 576 static int do_mmap_shared_file(struct vm_area_struct *vma, unsigned long len)
 577 {
 578         int ret;
 579
 580         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
 581         if (ret != -ENOSYS)
 582                 return ret;
 583
 584         /* getting an ENOSYS error indicates that direct mmap isn't
 585          * possible (as opposed to tried but failed) so we'll fall
 586          * through to making a private copy of the data and mapping
 587          * that if we can */
 588         return -ENODEV;
 589 }
 590
 591 /*
 592  * set up a private mapping or an anonymous shared mapping
 593  */
 594 static int do_mmap_private(struct vm_area_struct *vma, unsigned long len)
 595 {
 596         void *base;
 597         int ret;
 598
 599         /* invoke the file's mapping function so that it can keep track of
 600          * shared mappings on devices or memory
 601          * - VM_MAYSHARE will be set if it may attempt to share
 602          */
 603         if (vma->vm_file) {
 604                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
 605                 if (ret != -ENOSYS) {
 606                         /* shouldn't return success if we're not sharing */
 607                         BUG_ON(ret == 0 && !(vma->vm_flags & VM_MAYSHARE));
 608                         return ret; /* success or a real error */
 609                 }
 610
 611                 /* getting an ENOSYS error indicates that direct mmap isn't
 612                  * possible (as opposed to tried but failed) so we'll try to
 613                  * make a private copy of the data and map that instead */
 614         }
 615
 616         /* allocate some memory to hold the mapping
 617          * - note that this may not return a page-aligned address if the object
 618          *   we're allocating is smaller than a page
 619          */
 620         base = kmalloc(len, GFP_KERNEL);
 621         if (!base)
 622                 goto enomem;
 623
 624         vma->vm_start = (unsigned long) base;
 625         vma->vm_end = vma->vm_start + len;
 626         vma->vm_flags |= VM_MAPPED_COPY;
 627
 628 #ifdef WARN_ON_SLACK
 629         if (len + WARN_ON_SLACK <= kobjsize(result))
 630                 printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n",
 631                        len, current->pid, kobjsize(result) - len);
 632 #endif
 633
 634         if (vma->vm_file) {
 635                 /* read the contents of a file into the copy */
 636                 mm_segment_t old_fs;
 637                 loff_t fpos;
 638
 639                 fpos = vma->vm_pgoff;
 640                 fpos <<= PAGE_SHIFT;
 641
 642                 old_fs = get_fs();
 643                 set_fs(KERNEL_DS);
 644                 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
 645                 set_fs(old_fs);
 646
 647                 if (ret < 0)
 648                         goto error_free;
 649
 650                 /* clear the last little bit */
 651                 if (ret < len)
 652                         memset(base + ret, 0, len - ret);
 653
 654         } else {
 655                 /* if it's an anonymous mapping, then just clear it */
 656                 memset(base, 0, len);
 657         }
 658
 659         return 0;
 660
 661 error_free:
 662         kfree(base);
 663         vma->vm_start = 0;
 664         return ret;
 665
 666 enomem:
 667         printk("Allocation of length %lu from process %d failed\n",
 668                len, current->pid);
 669         show_free_areas();
 670         return -ENOMEM;
 671 }
 672
 673 /*
 674  * handle mapping creation for uClinux
 675  */
 676 unsigned long do_mmap_pgoff(struct file *file,
 677                             unsigned long addr,
 678                             unsigned long len,
 679                             unsigned long prot,
 680                             unsigned long flags,
 681                             unsigned long pgoff)
 682 {
 683         struct vm_list_struct *vml = NULL;
 684         struct vm_area_struct *vma = NULL;
 685         struct rb_node *rb;
 686         unsigned long capabilities, vm_flags;
 687         void *result;
 688         int ret;
 689
 690         /* decide whether we should attempt the mapping, and if so what sort of
 691          * mapping */
 692         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
 693                                     &capabilities);
 694         if (ret < 0)
 695                 return ret;
 696
 697         /* we've determined that we can make the mapping, now translate what we
 698          * now know into VMA flags */
 699         vm_flags = determine_vm_flags(file, prot, flags, capabilities);
 700
 701         /* we're going to need to record the mapping if it works */
 702         vml = kmalloc(sizeof(struct vm_list_struct), GFP_KERNEL);
 703         if (!vml)
 704                 goto error_getting_vml;
 705         memset(vml, 0, sizeof(*vml));
 706
 707         down_write(&nommu_vma_sem);
 708
 709         /* if we want to share, we need to check for VMAs created by other
 710          * mmap() calls that overlap with our proposed mapping
 711          * - we can only share with an exact match on most regular files
 712          * - shared mappings on character devices and memory backed files are
 713          *   permitted to overlap inexactly as far as we are concerned for in
 714          *   these cases, sharing is handled in the driver or filesystem rather
 715          *   than here
 716          */
 717         if (vm_flags & VM_MAYSHARE) {
 718                 unsigned long pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 719                 unsigned long vmpglen;
 720
 721                 for (rb = rb_first(&nommu_vma_tree); rb; rb = rb_next(rb)) {
 722                         vma = rb_entry(rb, struct vm_area_struct, vm_rb);
 723
 724                         if (!(vma->vm_flags & VM_MAYSHARE))
 725                                 continue;
 726
 727                         /* search for overlapping mappings on the same file */
 728                         if (vma->vm_file->f_dentry->d_inode != file->f_dentry->d_inode)
 729                                 continue;
 730
 731                         if (vma->vm_pgoff >= pgoff + pglen)
 732                                 continue;
 733
 734                         vmpglen = vma->vm_end - vma->vm_start + PAGE_SIZE - 1;
 735                         vmpglen >>= PAGE_SHIFT;
 736                         if (pgoff >= vma->vm_pgoff + vmpglen)
 737                                 continue;
 738
 739                         /* handle inexactly overlapping matches between mappings */
 740                         if (vma->vm_pgoff != pgoff || vmpglen != pglen) {
 741                                 if (!(capabilities & BDI_CAP_MAP_DIRECT))
 742                                         goto sharing_violation;
 743                                 continue;
 744                         }
 745
 746                         /* we've found a VMA we can share */
 747                         atomic_inc(&vma->vm_usage);
 748
 749                         vml->vma = vma;
 750                         result = (void *) vma->vm_start;
 751                         goto shared;
 752                 }
 753
 754                 vma = NULL;
 755
 756                 /* obtain the address at which to make a shared mapping
 757                  * - this is the hook for quasi-memory character devices to
 758                  *   tell us the location of a shared mapping
 759                  */
 760                 if (file && file->f_op->get_unmapped_area) {
 761                         addr = file->f_op->get_unmapped_area(file, addr, len,
 762                                                              pgoff, flags);
 763                         if (IS_ERR((void *) addr)) {
 764                                 ret = addr;
 765                                 if (ret != (unsigned long) -ENOSYS)
 766                                         goto error;
 767
 768                                 /* the driver refused to tell us where to site
 769                                  * the mapping so we'll have to attempt to copy
 770                                  * it */
 771                                 ret = (unsigned long) -ENODEV;
 772                                 if (!(capabilities & BDI_CAP_MAP_COPY))
 773                                         goto error;
 774
 775                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
 776                         }
 777                 }
 778         }
 779
 780         /* we're going to need a VMA struct as well */
 781         vma = kmalloc(sizeof(struct vm_area_struct), GFP_KERNEL);
 782         if (!vma)
 783                 goto error_getting_vma;
 784
 785         memset(vma, 0, sizeof(*vma));
 786         INIT_LIST_HEAD(&vma->anon_vma_node);
 787         atomic_set(&vma->vm_usage, 1);
 788         if (file)
 789                 get_file(file);
 790         vma->vm_file    = file;
 791         vma->vm_flags   = vm_flags;
 792         vma->vm_start   = addr;
 793         vma->vm_end     = addr + len;
 794         vma->vm_pgoff   = pgoff;
 795
 796         vml->vma = vma;
 797
 798         /* set up the mapping */
 799         if (file && vma->vm_flags & VM_SHARED)
 800                 ret = do_mmap_shared_file(vma, len);
 801         else
 802                 ret = do_mmap_private(vma, len);
 803         if (ret < 0)
 804                 goto error;
 805
 806         /* okay... we have a mapping; now we have to register it */
 807         result = (void *) vma->vm_start;
 808
 809         if (vma->vm_flags & VM_MAPPED_COPY) {
 810                 realalloc += kobjsize(result);
 811                 askedalloc += len;
 812         }
 813
 814         realalloc += kobjsize(vma);
 815         askedalloc += sizeof(*vma);
 816
 817         current->mm->total_vm += len >> PAGE_SHIFT;
 818
 819         add_nommu_vma(vma);
 820
 821  shared:
 822         realalloc += kobjsize(vml);
 823         askedalloc += sizeof(*vml);
 824
 825         vml->next = current->mm->context.vmlist;
 826         current->mm->context.vmlist = vml;
 827
 828         up_write(&nommu_vma_sem);
 829
 830         if (prot & PROT_EXEC)
 831                 flush_icache_range((unsigned long) result,
 832                                    (unsigned long) result + len);
 833
 834 #ifdef DEBUG
 835         printk("do_mmap:\n");
 836         show_process_blocks();
 837 #endif
 838
 839         return (unsigned long) result;
 840
 841  error:
 842         up_write(&nommu_vma_sem);
 843         kfree(vml);
 844         if (vma) {
 845                 fput(vma->vm_file);
 846                 kfree(vma);
 847         }
 848         return ret;
 849
 850  sharing_violation:
 851         up_write(&nommu_vma_sem);
 852         printk("Attempt to share mismatched mappings\n");
 853         kfree(vml);
 854         return -EINVAL;
 855
 856  error_getting_vma:
 857         up_write(&nommu_vma_sem);
 858         kfree(vml);
 859         printk("Allocation of vma for %lu byte allocation from process %d failed\n",
 860                len, current->pid);
 861         show_free_areas();
 862         return -ENOMEM;
 863
 864  error_getting_vml:
 865         printk("Allocation of vml for %lu byte allocation from process %d failed\n",
 866                len, current->pid);
 867         show_free_areas();
 868         return -ENOMEM;
 869 }
 870
 871 /*
 872  * handle mapping disposal for uClinux
 873  */
 874 static void put_vma(struct vm_area_struct *vma)
 875 {
 876         if (vma) {
 877                 down_write(&nommu_vma_sem);
 878
 879                 if (atomic_dec_and_test(&vma->vm_usage)) {
 880                         delete_nommu_vma(vma);
 881
 882                         if (vma->vm_ops && vma->vm_ops->close)
 883                                 vma->vm_ops->close(vma);
 884
 885                         /* IO memory and memory shared directly out of the pagecache from
 886                          * ramfs/tmpfs mustn't be released here */
 887                         if (vma->vm_flags & VM_MAPPED_COPY) {
 888                                 realalloc -= kobjsize((void *) vma->vm_start);
 889                                 askedalloc -= vma->vm_end - vma->vm_start;
 890                                 kfree((void *) vma->vm_start);
 891                         }
 892
 893                         realalloc -= kobjsize(vma);
 894                         askedalloc -= sizeof(*vma);
 895
 896                         if (vma->vm_file)
 897                                 fput(vma->vm_file);
 898                         kfree(vma);
 899                 }
 900
 901                 up_write(&nommu_vma_sem);
 902         }
 903 }
 904
 905 int do_munmap(struct mm_struct *mm, unsigned long addr, size_t len)
 906 {
 907         struct vm_list_struct *vml, **parent;
 908         unsigned long end = addr + len;
 909
 910 #ifdef DEBUG
 911         printk("do_munmap:\n");
 912 #endif
 913
 914         for (parent = &mm->context.vmlist; *parent; parent = &(*parent)->next)
 915                 if ((*parent)->vma->vm_start == addr &&
 916                     ((len == 0) || ((*parent)->vma->vm_end == end)))
 917                         goto found;
 918
 919         printk("munmap of non-mmaped memory by process %d (%s): %p\n",
 920                current->pid, current->comm, (void *) addr);
 921         return -EINVAL;
 922
 923  found:
 924         vml = *parent;
 925
 926         put_vma(vml->vma);
 927
 928         *parent = vml->next;
 929         realalloc -= kobjsize(vml);
 930         askedalloc -= sizeof(*vml);
 931         kfree(vml);
 932
 933         update_hiwater_vm(mm);
 934         mm->total_vm -= len >> PAGE_SHIFT;
 935
 936 #ifdef DEBUG
 937         show_process_blocks();
 938 #endif
 939
 940         return 0;
 941 }
 942
 943 /* Release all mmaps. */
 944 void exit_mmap(struct mm_struct * mm)
 945 {
 946         struct vm_list_struct *tmp;
 947
 948         if (mm) {
 949 #ifdef DEBUG
 950                 printk("Exit_mmap:\n");
 951 #endif
 952
 953                 mm->total_vm = 0;
 954
 955                 while ((tmp = mm->context.vmlist)) {
 956                         mm->context.vmlist = tmp->next;
 957                         put_vma(tmp->vma);
 958
 959                         realalloc -= kobjsize(tmp);
 960                         askedalloc -= sizeof(*tmp);
 961                         kfree(tmp);
 962                 }
 963
 964 #ifdef DEBUG
 965                 show_process_blocks();
 966 #endif
 967         }
 968 }
 969
 970 asmlinkage long sys_munmap(unsigned long addr, size_t len)
 971 {
 972         int ret;
 973         struct mm_struct *mm = current->mm;
 974
 975         down_write(&mm->mmap_sem);
 976         ret = do_munmap(mm, addr, len);
 977         up_write(&mm->mmap_sem);
 978         return ret;
 979 }
 980
 981 unsigned long do_brk(unsigned long addr, unsigned long len)
 982 {
 983         return -ENOMEM;
 984 }
 985
 986 /*
 987  * Expand (or shrink) an existing mapping, potentially moving it at the
 988  * same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
 989  *
 990  * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise
 991  * This option implies MREMAP_MAYMOVE.
 992  *
 993  * on uClinux, we only permit changing a mapping's size, and only as long as it stays within the
 994  * hole allocated by the kmalloc() call in do_mmap_pgoff() and the block is not shareable
 995  */
 996 unsigned long do_mremap(unsigned long addr,
 997                         unsigned long old_len, unsigned long new_len,
 998                         unsigned long flags, unsigned long new_addr)
 999 {
1000         struct vm_list_struct *vml = NULL;
1001
1002         /* insanity checks first */
1003         if (new_len == 0)
1004                 return (unsigned long) -EINVAL;
1005
1006         if (flags & MREMAP_FIXED && new_addr != addr)
1007                 return (unsigned long) -EINVAL;
1008
1009         for (vml = current->mm->context.vmlist; vml; vml = vml->next)
1010                 if (vml->vma->vm_start == addr)
1011                         goto found;
1012
1013         return (unsigned long) -EINVAL;
1014
1015  found:
1016         if (vml->vma->vm_end != vml->vma->vm_start + old_len)
1017                 return (unsigned long) -EFAULT;
1018
1019         if (vml->vma->vm_flags & VM_MAYSHARE)
1020                 return (unsigned long) -EPERM;
1021
1022         if (new_len > kobjsize((void *) addr))
1023                 return (unsigned long) -ENOMEM;
1024
1025         /* all checks complete - do it */
1026         vml->vma->vm_end = vml->vma->vm_start + new_len;
1027
1028         askedalloc -= old_len;
1029         askedalloc += new_len;
1030
1031         return vml->vma->vm_start;
1032 }
1033
1034 /*
1035  * Look up the first VMA which satisfies  addr < vm_end,  NULL if none
1036  */
1037 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
1038 {
1039         struct vm_list_struct *vml;
1040
1041         for (vml = mm->context.vmlist; vml; vml = vml->next)
1042                 if (addr >= vml->vma->vm_start && addr < vml->vma->vm_end)
1043                         return vml->vma;
1044
1045         return NULL;
1046 }
1047
1048 EXPORT_SYMBOL(find_vma);
1049
1050 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1051                         unsigned int foll_flags)
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 unmap_mapping_range(struct address_space *mapping,
1083                          loff_t const holebegin, loff_t const holelen,
1084                          int even_cows)
1085 {
1086 }
1087
1088 /*
1089  * Check that a process has enough memory to allocate a new virtual
1090  * mapping. 0 means there is enough memory for the allocation to
1091  * succeed and -ENOMEM implies there is not.
1092  *
1093  * We currently support three overcommit policies, which are set via the
1094  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1095  *
1096  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1097  * Additional code 2002 Jul 20 by Robert Love.
1098  *
1099  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1100  *
1101  * Note this is a helper function intended to be used by LSMs which
1102  * wish to use this logic.
1103  */
1104 int __vm_enough_memory(long pages, int cap_sys_admin)
1105 {
1106         unsigned long free, allowed;
1107
1108         vm_acct_memory(pages);
1109
1110         /*
1111          * Sometimes we want to use more memory than we have
1112          */
1113         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1114                 return 0;
1115
1116         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1117                 unsigned long n;
1118
1119                 free = get_page_cache_size();
1120                 free += nr_swap_pages;
1121
1122                 /*
1123                  * Any slabs which are created with the
1124                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1125                  * which are reclaimable, under pressure.  The dentry
1126                  * cache and most inode caches should fall into this
1127                  */
1128                 free += atomic_read(&slab_reclaim_pages);
1129
1130                 /*
1131                  * Leave the last 3% for root
1132                  */
1133                 if (!cap_sys_admin)
1134                         free -= free / 32;
1135
1136                 if (free > pages)
1137                         return 0;
1138
1139                 /*
1140                  * nr_free_pages() is very expensive on large systems,
1141                  * only call if we're about to fail.
1142                  */
1143                 n = nr_free_pages();
1144                 if (!cap_sys_admin)
1145                         n -= n / 32;
1146                 free += n;
1147
1148                 if (free > pages)
1149                         return 0;
1150                 vm_unacct_memory(pages);
1151                 return -ENOMEM;
1152         }
1153
1154         allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1155         /*
1156          * Leave the last 3% for root
1157          */
1158         if (!cap_sys_admin)
1159                 allowed -= allowed / 32;
1160         allowed += total_swap_pages;
1161
1162         /* Don't let a single process grow too big:
1163            leave 3% of the size of this process for other processes */
1164         allowed -= current->mm->total_vm / 32;
1165
1166         /*
1167          * cast `allowed' as a signed long because vm_committed_space
1168          * sometimes has a negative value
1169          */
1170         if (atomic_read(&vm_committed_space) < (long)allowed)
1171                 return 0;
1172
1173         vm_unacct_memory(pages);
1174
1175         return -ENOMEM;
1176 }
1177
1178 int in_gate_area_no_task(unsigned long addr)
1179 {
1180         return 0;
1181 }
1182
1183 struct page *filemap_nopage(struct vm_area_struct *area,
1184                         unsigned long address, int *type)
1185 {
1186         BUG();
1187         return NULL;
1188 }