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 #include <linux/vs_base.h>
  31 #include <linux/vs_memory.h>
  32
  33 #include <asm/uaccess.h>
  34 #include <asm/tlb.h>
  35 #include <asm/tlbflush.h>
  36
  37 void *high_memory;
  38 struct page *mem_map;
  39 unsigned long max_mapnr;
  40 unsigned long num_physpages;
  41 unsigned long askedalloc, realalloc;
  42 atomic_t vm_committed_space = ATOMIC_INIT(0);
  43 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
  44 int sysctl_overcommit_ratio = 50; /* default is 50% */
  45 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
  46 int heap_stack_gap = 0;
  47
  48 EXPORT_SYMBOL(mem_map);
  49 EXPORT_SYMBOL(__vm_enough_memory);
  50
  51 /* list of shareable VMAs */
  52 struct rb_root nommu_vma_tree = RB_ROOT;
  53 DECLARE_RWSEM(nommu_vma_sem);
  54
  55 struct vm_operations_struct generic_file_vm_ops = {
  56 };
  57
  58 EXPORT_SYMBOL(vfree);
  59 EXPORT_SYMBOL(vmalloc_to_page);
  60 EXPORT_SYMBOL(vmalloc_32);
  61 EXPORT_SYMBOL(vmap);
  62 EXPORT_SYMBOL(vunmap);
  63
  64 /*
  65  * Handle all mappings that got truncated by a "truncate()"
  66  * system call.
  67  *
  68  * NOTE! We have to be ready to update the memory sharing
  69  * between the file and the memory map for a potential last
  70  * incomplete page.  Ugly, but necessary.
  71  */
  72 int vmtruncate(struct inode *inode, loff_t offset)
  73 {
  74         struct address_space *mapping = inode->i_mapping;
  75         unsigned long limit;
  76
  77         if (inode->i_size < offset)
  78                 goto do_expand;
  79         i_size_write(inode, offset);
  80
  81         truncate_inode_pages(mapping, offset);
  82         goto out_truncate;
  83
  84 do_expand:
  85         limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
  86         if (limit != RLIM_INFINITY && offset > limit)
  87                 goto out_sig;
  88         if (offset > inode->i_sb->s_maxbytes)
  89                 goto out;
  90         i_size_write(inode, offset);
  91
  92 out_truncate:
  93         if (inode->i_op && inode->i_op->truncate)
  94                 inode->i_op->truncate(inode);
  95         return 0;
  96 out_sig:
  97         send_sig(SIGXFSZ, current, 0);
  98 out:
  99         return -EFBIG;
 100 }
 101
 102 EXPORT_SYMBOL(vmtruncate);
 103
 104 /*
 105  * Return the total memory allocated for this pointer, not
 106  * just what the caller asked for.
 107  *
 108  * Doesn't have to be accurate, i.e. may have races.
 109  */
 110 unsigned int kobjsize(const void *objp)
 111 {
 112         struct page *page;
 113
 114         if (!objp || !((page = virt_to_page(objp))))
 115                 return 0;
 116
 117         if (PageSlab(page))
 118                 return ksize(objp);
 119
 120         BUG_ON(page->index < 0);
 121         BUG_ON(page->index >= MAX_ORDER);
 122
 123         return (PAGE_SIZE << page->index);
 124 }
 125
 126 /*
 127  * The nommu dodgy version :-)
 128  */
 129 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
 130         unsigned long start, int len, int write, int force,
 131         struct page **pages, struct vm_area_struct **vmas)
 132 {
 133         int i;
 134         static struct vm_area_struct dummy_vma;
 135
 136         for (i = 0; i < len; i++) {
 137                 if (pages) {
 138                         pages[i] = virt_to_page(start);
 139                         if (pages[i])
 140                                 page_cache_get(pages[i]);
 141                 }
 142                 if (vmas)
 143                         vmas[i] = &dummy_vma;
 144                 start += PAGE_SIZE;
 145         }
 146         return(i);
 147 }
 148
 149 EXPORT_SYMBOL(get_user_pages);
 150
 151 DEFINE_RWLOCK(vmlist_lock);
 152 struct vm_struct *vmlist;
 153
 154 void vfree(void *addr)
 155 {
 156         kfree(addr);
 157 }
 158
 159 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
 160 {
 161         /*
 162          * kmalloc doesn't like __GFP_HIGHMEM for some reason
 163          */
 164         return kmalloc(size, gfp_mask & ~__GFP_HIGHMEM);
 165 }
 166
 167 struct page * vmalloc_to_page(void *addr)
 168 {
 169         return virt_to_page(addr);
 170 }
 171
 172 unsigned long vmalloc_to_pfn(void *addr)
 173 {
 174         return page_to_pfn(virt_to_page(addr));
 175 }
 176
 177
 178 long vread(char *buf, char *addr, unsigned long count)
 179 {
 180         memcpy(buf, addr, count);
 181         return count;
 182 }
 183
 184 long vwrite(char *buf, char *addr, unsigned long count)
 185 {
 186         /* Don't allow overflow */
 187         if ((unsigned long) addr + count < count)
 188                 count = -(unsigned long) addr;
 189
 190         memcpy(addr, buf, count);
 191         return(count);
 192 }
 193
 194 /*
 195  *      vmalloc  -  allocate virtually continguos memory
 196  *
 197  *      @size:          allocation size
 198  *
 199  *      Allocate enough pages to cover @size from the page level
 200  *      allocator and map them into continguos kernel virtual space.
 201  *
 202  *      For tight cotrol over page level allocator and protection flags
 203  *      use __vmalloc() instead.
 204  */
 205 void *vmalloc(unsigned long size)
 206 {
 207        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
 208 }
 209 EXPORT_SYMBOL(vmalloc);
 210
 211 void *vmalloc_node(unsigned long size, int node)
 212 {
 213         return vmalloc(size);
 214 }
 215 EXPORT_SYMBOL(vmalloc_node);
 216
 217 /*
 218  *      vmalloc_32  -  allocate virtually continguos memory (32bit addressable)
 219  *
 220  *      @size:          allocation size
 221  *
 222  *      Allocate enough 32bit PA addressable pages to cover @size from the
 223  *      page level allocator and map them into continguos kernel virtual space.
 224  */
 225 void *vmalloc_32(unsigned long size)
 226 {
 227         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
 228 }
 229
 230 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
 231 {
 232         BUG();
 233         return NULL;
 234 }
 235
 236 void vunmap(void *addr)
 237 {
 238         BUG();
 239 }
 240
 241 /*
 242  *  sys_brk() for the most part doesn't need the global kernel
 243  *  lock, except when an application is doing something nasty
 244  *  like trying to un-brk an area that has already been mapped
 245  *  to a regular file.  in this case, the unmapping will need
 246  *  to invoke file system routines that need the global lock.
 247  */
 248 asmlinkage unsigned long sys_brk(unsigned long brk)
 249 {
 250         struct mm_struct *mm = current->mm;
 251
 252         if (brk < mm->start_brk || brk > mm->context.end_brk)
 253                 return mm->brk;
 254
 255         if (mm->brk == brk)
 256                 return mm->brk;
 257
 258         /*
 259          * Always allow shrinking brk
 260          */
 261         if (brk <= mm->brk) {
 262                 mm->brk = brk;
 263                 return brk;
 264         }
 265
 266         /*
 267          * Ok, looks good - let it rip.
 268          */
 269         return mm->brk = brk;
 270 }
 271
 272 #ifdef DEBUG
 273 static void show_process_blocks(void)
 274 {
 275         struct vm_list_struct *vml;
 276
 277         printk("Process blocks %d:", current->pid);
 278
 279         for (vml = &current->mm->context.vmlist; vml; vml = vml->next) {
 280                 printk(" %p: %p", vml, vml->vma);
 281                 if (vml->vma)
 282                         printk(" (%d @%lx #%d)",
 283                                kobjsize((void *) vml->vma->vm_start),
 284                                vml->vma->vm_start,
 285                                atomic_read(&vml->vma->vm_usage));
 286                 printk(vml->next ? " ->" : ".\n");
 287         }
 288 }
 289 #endif /* DEBUG */
 290
 291 static inline struct vm_area_struct *find_nommu_vma(unsigned long start)
 292 {
 293         struct vm_area_struct *vma;
 294         struct rb_node *n = nommu_vma_tree.rb_node;
 295
 296         while (n) {
 297                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
 298
 299                 if (start < vma->vm_start)
 300                         n = n->rb_left;
 301                 else if (start > vma->vm_start)
 302                         n = n->rb_right;
 303                 else
 304                         return vma;
 305         }
 306
 307         return NULL;
 308 }
 309
 310 static void add_nommu_vma(struct vm_area_struct *vma)
 311 {
 312         struct vm_area_struct *pvma;
 313         struct address_space *mapping;
 314         struct rb_node **p = &nommu_vma_tree.rb_node;
 315         struct rb_node *parent = NULL;
 316
 317         /* add the VMA to the mapping */
 318         if (vma->vm_file) {
 319                 mapping = vma->vm_file->f_mapping;
 320
 321                 flush_dcache_mmap_lock(mapping);
 322                 vma_prio_tree_insert(vma, &mapping->i_mmap);
 323                 flush_dcache_mmap_unlock(mapping);
 324         }
 325
 326         /* add the VMA to the master list */
 327         while (*p) {
 328                 parent = *p;
 329                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
 330
 331                 if (vma->vm_start < pvma->vm_start) {
 332                         p = &(*p)->rb_left;
 333                 }
 334                 else if (vma->vm_start > pvma->vm_start) {
 335                         p = &(*p)->rb_right;
 336                 }
 337                 else {
 338                         /* mappings are at the same address - this can only
 339                          * happen for shared-mem chardevs and shared file
 340                          * mappings backed by ramfs/tmpfs */
 341                         BUG_ON(!(pvma->vm_flags & VM_SHARED));
 342
 343                         if (vma < pvma)
 344                                 p = &(*p)->rb_left;
 345                         else if (vma > pvma)
 346                                 p = &(*p)->rb_right;
 347                         else
 348                                 BUG();
 349                 }
 350         }
 351
 352         rb_link_node(&vma->vm_rb, parent, p);
 353         rb_insert_color(&vma->vm_rb, &nommu_vma_tree);
 354 }
 355
 356 static void delete_nommu_vma(struct vm_area_struct *vma)
 357 {
 358         struct address_space *mapping;
 359
 360         /* remove the VMA from the mapping */
 361         if (vma->vm_file) {
 362                 mapping = vma->vm_file->f_mapping;
 363
 364                 flush_dcache_mmap_lock(mapping);
 365                 vma_prio_tree_remove(vma, &mapping->i_mmap);
 366                 flush_dcache_mmap_unlock(mapping);
 367         }
 368
 369         /* remove from the master list */
 370         rb_erase(&vma->vm_rb, &nommu_vma_tree);
 371 }
 372
 373 /*
 374  * determine whether a mapping should be permitted and, if so, what sort of
 375  * mapping we're capable of supporting
 376  */
 377 static int validate_mmap_request(struct file *file,
 378                                  unsigned long addr,
 379                                  unsigned long len,
 380                                  unsigned long prot,
 381                                  unsigned long flags,
 382                                  unsigned long pgoff,
 383                                  unsigned long *_capabilities)
 384 {
 385         unsigned long capabilities;
 386         unsigned long reqprot = prot;
 387         int ret;
 388
 389         /* do the simple checks first */
 390         if (flags & MAP_FIXED || addr) {
 391                 printk(KERN_DEBUG
 392                        "%d: Can't do fixed-address/overlay mmap of RAM\n",
 393                        current->pid);
 394                 return -EINVAL;
 395         }
 396
 397         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
 398             (flags & MAP_TYPE) != MAP_SHARED)
 399                 return -EINVAL;
 400
 401         if (PAGE_ALIGN(len) == 0)
 402                 return addr;
 403
 404         if (len > TASK_SIZE)
 405                 return -EINVAL;
 406
 407         /* offset overflow? */
 408         if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
 409                 return -EINVAL;
 410
 411         if (file) {
 412                 /* validate file mapping requests */
 413                 struct address_space *mapping;
 414
 415                 /* files must support mmap */
 416                 if (!file->f_op || !file->f_op->mmap)
 417                         return -ENODEV;
 418
 419                 /* work out if what we've got could possibly be shared
 420                  * - we support chardevs that provide their own "memory"
 421                  * - we support files/blockdevs that are memory backed
 422                  */
 423                 mapping = file->f_mapping;
 424                 if (!mapping)
 425                         mapping = file->f_dentry->d_inode->i_mapping;
 426
 427                 capabilities = 0;
 428                 if (mapping && mapping->backing_dev_info)
 429                         capabilities = mapping->backing_dev_info->capabilities;
 430
 431                 if (!capabilities) {
 432                         /* no explicit capabilities set, so assume some
 433                          * defaults */
 434                         switch (file->f_dentry->d_inode->i_mode & S_IFMT) {
 435                         case S_IFREG:
 436                         case S_IFBLK:
 437                                 capabilities = BDI_CAP_MAP_COPY;
 438                                 break;
 439
 440                         case S_IFCHR:
 441                                 capabilities =
 442                                         BDI_CAP_MAP_DIRECT |
 443                                         BDI_CAP_READ_MAP |
 444                                         BDI_CAP_WRITE_MAP;
 445                                 break;
 446
 447                         default:
 448                                 return -EINVAL;
 449                         }
 450                 }
 451
 452                 /* eliminate any capabilities that we can't support on this
 453                  * device */
 454                 if (!file->f_op->get_unmapped_area)
 455                         capabilities &= ~BDI_CAP_MAP_DIRECT;
 456                 if (!file->f_op->read)
 457                         capabilities &= ~BDI_CAP_MAP_COPY;
 458
 459                 if (flags & MAP_SHARED) {
 460                         /* do checks for writing, appending and locking */
 461                         if ((prot & PROT_WRITE) &&
 462                             !(file->f_mode & FMODE_WRITE))
 463                                 return -EACCES;
 464
 465                         if (IS_APPEND(file->f_dentry->d_inode) &&
 466                             (file->f_mode & FMODE_WRITE))
 467                                 return -EACCES;
 468
 469                         if (locks_verify_locked(file->f_dentry->d_inode))
 470                                 return -EAGAIN;
 471
 472                         if (!(capabilities & BDI_CAP_MAP_DIRECT))
 473                                 return -ENODEV;
 474
 475                         if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
 476                             ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
 477                             ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
 478                             ) {
 479                                 printk("MAP_SHARED not completely supported on !MMU\n");
 480                                 return -EINVAL;
 481                         }
 482
 483                         /* we mustn't privatise shared mappings */
 484                         capabilities &= ~BDI_CAP_MAP_COPY;
 485                 }
 486                 else {
 487                         /* we're going to read the file into private memory we
 488                          * allocate */
 489                         if (!(capabilities & BDI_CAP_MAP_COPY))
 490                                 return -ENODEV;
 491
 492                         /* we don't permit a private writable mapping to be
 493                          * shared with the backing device */
 494                         if (prot & PROT_WRITE)
 495                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
 496                 }
 497
 498                 /* handle executable mappings and implied executable
 499                  * mappings */
 500                 if (file->f_vfsmnt->mnt_flags & MNT_NOEXEC) {
 501                         if (prot & PROT_EXEC)
 502                                 return -EPERM;
 503                 }
 504                 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
 505                         /* handle implication of PROT_EXEC by PROT_READ */
 506                         if (current->personality & READ_IMPLIES_EXEC) {
 507                                 if (capabilities & BDI_CAP_EXEC_MAP)
 508                                         prot |= PROT_EXEC;
 509                         }
 510                 }
 511                 else if ((prot & PROT_READ) &&
 512                          (prot & PROT_EXEC) &&
 513                          !(capabilities & BDI_CAP_EXEC_MAP)
 514                          ) {
 515                         /* backing file is not executable, try to copy */
 516                         capabilities &= ~BDI_CAP_MAP_DIRECT;
 517                 }
 518         }
 519         else {
 520                 /* anonymous mappings are always memory backed and can be
 521                  * privately mapped
 522                  */
 523                 capabilities = BDI_CAP_MAP_COPY;
 524
 525                 /* handle PROT_EXEC implication by PROT_READ */
 526                 if ((prot & PROT_READ) &&
 527                     (current->personality & READ_IMPLIES_EXEC))
 528                         prot |= PROT_EXEC;
 529         }
 530
 531         /* allow the security API to have its say */
 532         ret = security_file_mmap(file, reqprot, prot, flags);
 533         if (ret < 0)
 534                 return ret;
 535
 536         /* looks okay */
 537         *_capabilities = capabilities;
 538         return 0;
 539 }
 540
 541 /*
 542  * we've determined that we can make the mapping, now translate what we
 543  * now know into VMA flags
 544  */
 545 static unsigned long determine_vm_flags(struct file *file,
 546                                         unsigned long prot,
 547                                         unsigned long flags,
 548                                         unsigned long capabilities)
 549 {
 550         unsigned long vm_flags;
 551
 552         vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
 553         vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 554         /* vm_flags |= mm->def_flags; */
 555
 556         if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
 557                 /* attempt to share read-only copies of mapped file chunks */
 558                 if (file && !(prot & PROT_WRITE))
 559                         vm_flags |= VM_MAYSHARE;
 560         }
 561         else {
 562                 /* overlay a shareable mapping on the backing device or inode
 563                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
 564                  * romfs/cramfs */
 565                 if (flags & MAP_SHARED)
 566                         vm_flags |= VM_MAYSHARE | VM_SHARED;
 567                 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
 568                         vm_flags |= VM_MAYSHARE;
 569         }
 570
 571         /* refuse to let anyone share private mappings with this process if
 572          * it's being traced - otherwise breakpoints set in it may interfere
 573          * with another untraced process
 574          */
 575         if ((flags & MAP_PRIVATE) && (current->ptrace & PT_PTRACED))
 576                 vm_flags &= ~VM_MAYSHARE;
 577
 578         return vm_flags;
 579 }
 580
 581 /*
 582  * set up a shared mapping on a file
 583  */
 584 static int do_mmap_shared_file(struct vm_area_struct *vma, unsigned long len)
 585 {
 586         int ret;
 587
 588         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
 589         if (ret != -ENOSYS)
 590                 return ret;
 591
 592         /* getting an ENOSYS error indicates that direct mmap isn't
 593          * possible (as opposed to tried but failed) so we'll fall
 594          * through to making a private copy of the data and mapping
 595          * that if we can */
 596         return -ENODEV;
 597 }
 598
 599 /*
 600  * set up a private mapping or an anonymous shared mapping
 601  */
 602 static int do_mmap_private(struct vm_area_struct *vma, unsigned long len)
 603 {
 604         void *base;
 605         int ret;
 606
 607         /* invoke the file's mapping function so that it can keep track of
 608          * shared mappings on devices or memory
 609          * - VM_MAYSHARE will be set if it may attempt to share
 610          */
 611         if (vma->vm_file) {
 612                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
 613                 if (ret != -ENOSYS) {
 614                         /* shouldn't return success if we're not sharing */
 615                         BUG_ON(ret == 0 && !(vma->vm_flags & VM_MAYSHARE));
 616                         return ret; /* success or a real error */
 617                 }
 618
 619                 /* getting an ENOSYS error indicates that direct mmap isn't
 620                  * possible (as opposed to tried but failed) so we'll try to
 621                  * make a private copy of the data and map that instead */
 622         }
 623
 624         /* allocate some memory to hold the mapping
 625          * - note that this may not return a page-aligned address if the object
 626          *   we're allocating is smaller than a page
 627          */
 628         base = kmalloc(len, GFP_KERNEL);
 629         if (!base)
 630                 goto enomem;
 631
 632         vma->vm_start = (unsigned long) base;
 633         vma->vm_end = vma->vm_start + len;
 634         vma->vm_flags |= VM_MAPPED_COPY;
 635
 636 #ifdef WARN_ON_SLACK
 637         if (len + WARN_ON_SLACK <= kobjsize(result))
 638                 printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n",
 639                        len, current->pid, kobjsize(result) - len);
 640 #endif
 641
 642         if (vma->vm_file) {
 643                 /* read the contents of a file into the copy */
 644                 mm_segment_t old_fs;
 645                 loff_t fpos;
 646
 647                 fpos = vma->vm_pgoff;
 648                 fpos <<= PAGE_SHIFT;
 649
 650                 old_fs = get_fs();
 651                 set_fs(KERNEL_DS);
 652                 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
 653                 set_fs(old_fs);
 654
 655                 if (ret < 0)
 656                         goto error_free;
 657
 658                 /* clear the last little bit */
 659                 if (ret < len)
 660                         memset(base + ret, 0, len - ret);
 661
 662         } else {
 663                 /* if it's an anonymous mapping, then just clear it */
 664                 memset(base, 0, len);
 665         }
 666
 667         return 0;
 668
 669 error_free:
 670         kfree(base);
 671         vma->vm_start = 0;
 672         return ret;
 673
 674 enomem:
 675         printk("Allocation of length %lu from process %d failed\n",
 676                len, current->pid);
 677         show_free_areas();
 678         return -ENOMEM;
 679 }
 680
 681 /*
 682  * handle mapping creation for uClinux
 683  */
 684 unsigned long do_mmap_pgoff(struct file *file,
 685                             unsigned long addr,
 686                             unsigned long len,
 687                             unsigned long prot,
 688                             unsigned long flags,
 689                             unsigned long pgoff)
 690 {
 691         struct vm_list_struct *vml = NULL;
 692         struct vm_area_struct *vma = NULL;
 693         struct rb_node *rb;
 694         unsigned long capabilities, vm_flags;
 695         void *result;
 696         int ret;
 697
 698         /* decide whether we should attempt the mapping, and if so what sort of
 699          * mapping */
 700         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
 701                                     &capabilities);
 702         if (ret < 0)
 703                 return ret;
 704
 705         /* we've determined that we can make the mapping, now translate what we
 706          * now know into VMA flags */
 707         vm_flags = determine_vm_flags(file, prot, flags, capabilities);
 708
 709         /* we're going to need to record the mapping if it works */
 710         vml = kmalloc(sizeof(struct vm_list_struct), GFP_KERNEL);
 711         if (!vml)
 712                 goto error_getting_vml;
 713         memset(vml, 0, sizeof(*vml));
 714
 715         down_write(&nommu_vma_sem);
 716
 717         /* if we want to share, we need to check for VMAs created by other
 718          * mmap() calls that overlap with our proposed mapping
 719          * - we can only share with an exact match on most regular files
 720          * - shared mappings on character devices and memory backed files are
 721          *   permitted to overlap inexactly as far as we are concerned for in
 722          *   these cases, sharing is handled in the driver or filesystem rather
 723          *   than here
 724          */
 725         if (vm_flags & VM_MAYSHARE) {
 726                 unsigned long pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 727                 unsigned long vmpglen;
 728
 729                 for (rb = rb_first(&nommu_vma_tree); rb; rb = rb_next(rb)) {
 730                         vma = rb_entry(rb, struct vm_area_struct, vm_rb);
 731
 732                         if (!(vma->vm_flags & VM_MAYSHARE))
 733                                 continue;
 734
 735                         /* search for overlapping mappings on the same file */
 736                         if (vma->vm_file->f_dentry->d_inode != file->f_dentry->d_inode)
 737                                 continue;
 738
 739                         if (vma->vm_pgoff >= pgoff + pglen)
 740                                 continue;
 741
 742                         vmpglen = vma->vm_end - vma->vm_start + PAGE_SIZE - 1;
 743                         vmpglen >>= PAGE_SHIFT;
 744                         if (pgoff >= vma->vm_pgoff + vmpglen)
 745                                 continue;
 746
 747                         /* handle inexactly overlapping matches between mappings */
 748                         if (vma->vm_pgoff != pgoff || vmpglen != pglen) {
 749                                 if (!(capabilities & BDI_CAP_MAP_DIRECT))
 750                                         goto sharing_violation;
 751                                 continue;
 752                         }
 753
 754                         /* we've found a VMA we can share */
 755                         atomic_inc(&vma->vm_usage);
 756
 757                         vml->vma = vma;
 758                         result = (void *) vma->vm_start;
 759                         goto shared;
 760                 }
 761
 762                 vma = NULL;
 763
 764                 /* obtain the address at which to make a shared mapping
 765                  * - this is the hook for quasi-memory character devices to
 766                  *   tell us the location of a shared mapping
 767                  */
 768                 if (file && file->f_op->get_unmapped_area) {
 769                         addr = file->f_op->get_unmapped_area(file, addr, len,
 770                                                              pgoff, flags);
 771                         if (IS_ERR((void *) addr)) {
 772                                 ret = addr;
 773                                 if (ret != (unsigned long) -ENOSYS)
 774                                         goto error;
 775
 776                                 /* the driver refused to tell us where to site
 777                                  * the mapping so we'll have to attempt to copy
 778                                  * it */
 779                                 ret = (unsigned long) -ENODEV;
 780                                 if (!(capabilities & BDI_CAP_MAP_COPY))
 781                                         goto error;
 782
 783                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
 784                         }
 785                 }
 786         }
 787
 788         /* we're going to need a VMA struct as well */
 789         vma = kmalloc(sizeof(struct vm_area_struct), GFP_KERNEL);
 790         if (!vma)
 791                 goto error_getting_vma;
 792
 793         memset(vma, 0, sizeof(*vma));
 794         INIT_LIST_HEAD(&vma->anon_vma_node);
 795         atomic_set(&vma->vm_usage, 1);
 796         if (file)
 797                 get_file(file);
 798         vma->vm_file    = file;
 799         vma->vm_flags   = vm_flags;
 800         vma->vm_start   = addr;
 801         vma->vm_end     = addr + len;
 802         vma->vm_pgoff   = pgoff;
 803
 804         vml->vma = vma;
 805
 806         /* set up the mapping */
 807         if (file && vma->vm_flags & VM_SHARED)
 808                 ret = do_mmap_shared_file(vma, len);
 809         else
 810                 ret = do_mmap_private(vma, len);
 811         if (ret < 0)
 812                 goto error;
 813
 814         /* okay... we have a mapping; now we have to register it */
 815         result = (void *) vma->vm_start;
 816
 817         if (vma->vm_flags & VM_MAPPED_COPY) {
 818                 realalloc += kobjsize(result);
 819                 askedalloc += len;
 820         }
 821
 822         realalloc += kobjsize(vma);
 823         askedalloc += sizeof(*vma);
 824
 825         vx_vmpages_add(current->mm, len >> PAGE_SHIFT);
 826
 827         add_nommu_vma(vma);
 828
 829  shared:
 830         realalloc += kobjsize(vml);
 831         askedalloc += sizeof(*vml);
 832
 833         vml->next = current->mm->context.vmlist;
 834         current->mm->context.vmlist = vml;
 835
 836         up_write(&nommu_vma_sem);
 837
 838         if (prot & PROT_EXEC)
 839                 flush_icache_range((unsigned long) result,
 840                                    (unsigned long) result + len);
 841
 842 #ifdef DEBUG
 843         printk("do_mmap:\n");
 844         show_process_blocks();
 845 #endif
 846
 847         return (unsigned long) result;
 848
 849  error:
 850         up_write(&nommu_vma_sem);
 851         kfree(vml);
 852         if (vma) {
 853                 fput(vma->vm_file);
 854                 kfree(vma);
 855         }
 856         return ret;
 857
 858  sharing_violation:
 859         up_write(&nommu_vma_sem);
 860         printk("Attempt to share mismatched mappings\n");
 861         kfree(vml);
 862         return -EINVAL;
 863
 864  error_getting_vma:
 865         up_write(&nommu_vma_sem);
 866         kfree(vml);
 867         printk("Allocation of vma for %lu byte allocation from process %d failed\n",
 868                len, current->pid);
 869         show_free_areas();
 870         return -ENOMEM;
 871
 872  error_getting_vml:
 873         printk("Allocation of vml for %lu byte allocation from process %d failed\n",
 874                len, current->pid);
 875         show_free_areas();
 876         return -ENOMEM;
 877 }
 878
 879 /*
 880  * handle mapping disposal for uClinux
 881  */
 882 static void put_vma(struct vm_area_struct *vma)
 883 {
 884         if (vma) {
 885                 down_write(&nommu_vma_sem);
 886
 887                 if (atomic_dec_and_test(&vma->vm_usage)) {
 888                         delete_nommu_vma(vma);
 889
 890                         if (vma->vm_ops && vma->vm_ops->close)
 891                                 vma->vm_ops->close(vma);
 892
 893                         /* IO memory and memory shared directly out of the pagecache from
 894                          * ramfs/tmpfs mustn't be released here */
 895                         if (vma->vm_flags & VM_MAPPED_COPY) {
 896                                 realalloc -= kobjsize((void *) vma->vm_start);
 897                                 askedalloc -= vma->vm_end - vma->vm_start;
 898                                 kfree((void *) vma->vm_start);
 899                         }
 900
 901                         realalloc -= kobjsize(vma);
 902                         askedalloc -= sizeof(*vma);
 903
 904                         if (vma->vm_file)
 905                                 fput(vma->vm_file);
 906                         kfree(vma);
 907                 }
 908
 909                 up_write(&nommu_vma_sem);
 910         }
 911 }
 912
 913 int do_munmap(struct mm_struct *mm, unsigned long addr, size_t len)
 914 {
 915         struct vm_list_struct *vml, **parent;
 916         unsigned long end = addr + len;
 917
 918 #ifdef DEBUG
 919         printk("do_munmap:\n");
 920 #endif
 921
 922         for (parent = &mm->context.vmlist; *parent; parent = &(*parent)->next)
 923                 if ((*parent)->vma->vm_start == addr &&
 924                     ((len == 0) || ((*parent)->vma->vm_end == end)))
 925                         goto found;
 926
 927         printk("munmap of non-mmaped memory by process %d (%s): %p\n",
 928                current->pid, current->comm, (void *) addr);
 929         return -EINVAL;
 930
 931  found:
 932         vml = *parent;
 933
 934         put_vma(vml->vma);
 935
 936         *parent = vml->next;
 937         realalloc -= kobjsize(vml);
 938         askedalloc -= sizeof(*vml);
 939         kfree(vml);
 940
 941         update_hiwater_vm(mm);
 942         vx_vmpages_sub(mm, len >> PAGE_SHIFT);
 943
 944 #ifdef DEBUG
 945         show_process_blocks();
 946 #endif
 947
 948         return 0;
 949 }
 950
 951 /* Release all mmaps. */
 952 void exit_mmap(struct mm_struct * mm)
 953 {
 954         struct vm_list_struct *tmp;
 955
 956         if (mm) {
 957 #ifdef DEBUG
 958                 printk("Exit_mmap:\n");
 959 #endif
 960
 961                 vx_vmpages_sub(mm, mm->total_vm);
 962
 963                 while ((tmp = mm->context.vmlist)) {
 964                         mm->context.vmlist = tmp->next;
 965                         put_vma(tmp->vma);
 966
 967                         realalloc -= kobjsize(tmp);
 968                         askedalloc -= sizeof(*tmp);
 969                         kfree(tmp);
 970                 }
 971
 972 #ifdef DEBUG
 973                 show_process_blocks();
 974 #endif
 975         }
 976 }
 977
 978 asmlinkage long sys_munmap(unsigned long addr, size_t len)
 979 {
 980         int ret;
 981         struct mm_struct *mm = current->mm;
 982
 983         down_write(&mm->mmap_sem);
 984         ret = do_munmap(mm, addr, len);
 985         up_write(&mm->mmap_sem);
 986         return ret;
 987 }
 988
 989 unsigned long do_brk(unsigned long addr, unsigned long len)
 990 {
 991         return -ENOMEM;
 992 }
 993
 994 /*
 995  * Expand (or shrink) an existing mapping, potentially moving it at the
 996  * same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
 997  *
 998  * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise
 999  * This option implies MREMAP_MAYMOVE.
1000  *
1001  * on uClinux, we only permit changing a mapping's size, and only as long as it stays within the
1002  * hole allocated by the kmalloc() call in do_mmap_pgoff() and the block is not shareable
1003  */
1004 unsigned long do_mremap(unsigned long addr,
1005                         unsigned long old_len, unsigned long new_len,
1006                         unsigned long flags, unsigned long new_addr)
1007 {
1008         struct vm_list_struct *vml = NULL;
1009
1010         /* insanity checks first */
1011         if (new_len == 0)
1012                 return (unsigned long) -EINVAL;
1013
1014         if (flags & MREMAP_FIXED && new_addr != addr)
1015                 return (unsigned long) -EINVAL;
1016
1017         for (vml = current->mm->context.vmlist; vml; vml = vml->next)
1018                 if (vml->vma->vm_start == addr)
1019                         goto found;
1020
1021         return (unsigned long) -EINVAL;
1022
1023  found:
1024         if (vml->vma->vm_end != vml->vma->vm_start + old_len)
1025                 return (unsigned long) -EFAULT;
1026
1027         if (vml->vma->vm_flags & VM_MAYSHARE)
1028                 return (unsigned long) -EPERM;
1029
1030         if (new_len > kobjsize((void *) addr))
1031                 return (unsigned long) -ENOMEM;
1032
1033         /* all checks complete - do it */
1034         vml->vma->vm_end = vml->vma->vm_start + new_len;
1035
1036         askedalloc -= old_len;
1037         askedalloc += new_len;
1038
1039         return vml->vma->vm_start;
1040 }
1041
1042 /*
1043  * Look up the first VMA which satisfies  addr < vm_end,  NULL if none
1044  */
1045 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
1046 {
1047         struct vm_list_struct *vml;
1048
1049         for (vml = mm->context.vmlist; vml; vml = vml->next)
1050                 if (addr >= vml->vma->vm_start && addr < vml->vma->vm_end)
1051                         return vml->vma;
1052
1053         return NULL;
1054 }
1055
1056 EXPORT_SYMBOL(find_vma);
1057
1058 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1059                         unsigned int foll_flags)
1060 {
1061         return NULL;
1062 }
1063
1064 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
1065 {
1066         return NULL;
1067 }
1068
1069 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1070                 unsigned long to, unsigned long size, pgprot_t prot)
1071 {
1072         vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1073         return 0;
1074 }
1075
1076 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1077 {
1078 }
1079
1080 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1081         unsigned long len, unsigned long pgoff, unsigned long flags)
1082 {
1083         return -ENOMEM;
1084 }
1085
1086 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1087 {
1088 }
1089
1090 void unmap_mapping_range(struct address_space *mapping,
1091                          loff_t const holebegin, loff_t const holelen,
1092                          int even_cows)
1093 {
1094 }
1095
1096 /*
1097  * Check that a process has enough memory to allocate a new virtual
1098  * mapping. 0 means there is enough memory for the allocation to
1099  * succeed and -ENOMEM implies there is not.
1100  *
1101  * We currently support three overcommit policies, which are set via the
1102  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1103  *
1104  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1105  * Additional code 2002 Jul 20 by Robert Love.
1106  *
1107  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1108  *
1109  * Note this is a helper function intended to be used by LSMs which
1110  * wish to use this logic.
1111  */
1112 int __vm_enough_memory(long pages, int cap_sys_admin)
1113 {
1114         unsigned long free, allowed;
1115
1116         vm_acct_memory(pages);
1117
1118         /*
1119          * Sometimes we want to use more memory than we have
1120          */
1121         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1122                 return 0;
1123
1124         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1125                 unsigned long n;
1126
1127                 free = get_page_cache_size();
1128                 free += nr_swap_pages;
1129
1130                 /*
1131                  * Any slabs which are created with the
1132                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1133                  * which are reclaimable, under pressure.  The dentry
1134                  * cache and most inode caches should fall into this
1135                  */
1136                 free += atomic_read(&slab_reclaim_pages);
1137
1138                 /*
1139                  * Leave the last 3% for root
1140                  */
1141                 if (!cap_sys_admin)
1142                         free -= free / 32;
1143
1144                 if (free > pages)
1145                         return 0;
1146
1147                 /*
1148                  * nr_free_pages() is very expensive on large systems,
1149                  * only call if we're about to fail.
1150                  */
1151                 n = nr_free_pages();
1152                 if (!cap_sys_admin)
1153                         n -= n / 32;
1154                 free += n;
1155
1156                 if (free > pages)
1157                         return 0;
1158                 vm_unacct_memory(pages);
1159                 return -ENOMEM;
1160         }
1161
1162         allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1163         /*
1164          * Leave the last 3% for root
1165          */
1166         if (!cap_sys_admin)
1167                 allowed -= allowed / 32;
1168         allowed += total_swap_pages;
1169
1170         /* Don't let a single process grow too big:
1171            leave 3% of the size of this process for other processes */
1172         allowed -= current->mm->total_vm / 32;
1173
1174         /*
1175          * cast `allowed' as a signed long because vm_committed_space
1176          * sometimes has a negative value
1177          */
1178         if (atomic_read(&vm_committed_space) < (long)allowed)
1179                 return 0;
1180
1181         vm_unacct_memory(pages);
1182
1183         return -ENOMEM;
1184 }
1185
1186 int in_gate_area_no_task(unsigned long addr)
1187 {
1188         return 0;
1189 }
1190
1191 struct page *filemap_nopage(struct vm_area_struct *area,
1192                         unsigned long address, int *type)
1193 {
1194         BUG();
1195         return NULL;
1196 }