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(vfree);
  57 EXPORT_SYMBOL(vmalloc_to_page);
  58 EXPORT_SYMBOL(vmalloc_32);
  59 EXPORT_SYMBOL(vmap);
  60 EXPORT_SYMBOL(vunmap);
  61
  62 /*
  63  * Handle all mappings that got truncated by a "truncate()"
  64  * system call.
  65  *
  66  * NOTE! We have to be ready to update the memory sharing
  67  * between the file and the memory map for a potential last
  68  * incomplete page.  Ugly, but necessary.
  69  */
  70 int vmtruncate(struct inode *inode, loff_t offset)
  71 {
  72         struct address_space *mapping = inode->i_mapping;
  73         unsigned long limit;
  74
  75         if (inode->i_size < offset)
  76                 goto do_expand;
  77         i_size_write(inode, offset);
  78
  79         truncate_inode_pages(mapping, offset);
  80         goto out_truncate;
  81
  82 do_expand:
  83         limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
  84         if (limit != RLIM_INFINITY && offset > limit)
  85                 goto out_sig;
  86         if (offset > inode->i_sb->s_maxbytes)
  87                 goto out;
  88         i_size_write(inode, offset);
  89
  90 out_truncate:
  91         if (inode->i_op && inode->i_op->truncate)
  92                 inode->i_op->truncate(inode);
  93         return 0;
  94 out_sig:
  95         send_sig(SIGXFSZ, current, 0);
  96 out:
  97         return -EFBIG;
  98 }
  99
 100 EXPORT_SYMBOL(vmtruncate);
 101
 102 /*
 103  * Return the total memory allocated for this pointer, not
 104  * just what the caller asked for.
 105  *
 106  * Doesn't have to be accurate, i.e. may have races.
 107  */
 108 unsigned int kobjsize(const void *objp)
 109 {
 110         struct page *page;
 111
 112         if (!objp || !((page = virt_to_page(objp))))
 113                 return 0;
 114
 115         if (PageSlab(page))
 116                 return ksize(objp);
 117
 118         BUG_ON(page->index < 0);
 119         BUG_ON(page->index >= MAX_ORDER);
 120
 121         return (PAGE_SIZE << page->index);
 122 }
 123
 124 /*
 125  * The nommu dodgy version :-)
 126  */
 127 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
 128         unsigned long start, int len, int write, int force,
 129         struct page **pages, struct vm_area_struct **vmas)
 130 {
 131         int i;
 132         struct vm_area_struct *vma;
 133
 134         for (i = 0; i < len; i++) {
 135                 vma = find_vma(mm, start);
 136                 if(!vma)
 137                         return i ? : -EFAULT;
 138
 139                 if (pages) {
 140                         pages[i] = virt_to_page(start);
 141                         if (pages[i])
 142                                 page_cache_get(pages[i]);
 143                 }
 144                 if (vmas)
 145                         vmas[i] = vma;
 146                 start += PAGE_SIZE;
 147         }
 148         return(i);
 149 }
 150
 151 EXPORT_SYMBOL(get_user_pages);
 152
 153 DEFINE_RWLOCK(vmlist_lock);
 154 struct vm_struct *vmlist;
 155
 156 void vfree(void *addr)
 157 {
 158         kfree(addr);
 159 }
 160
 161 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
 162 {
 163         /*
 164          * kmalloc doesn't like __GFP_HIGHMEM for some reason
 165          */
 166         return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__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 EXPORT_SYMBOL(vmalloc);
 212
 213 void *vmalloc_node(unsigned long size, int node)
 214 {
 215         return vmalloc(size);
 216 }
 217 EXPORT_SYMBOL(vmalloc_node);
 218
 219 /*
 220  *      vmalloc_32  -  allocate virtually continguos memory (32bit addressable)
 221  *
 222  *      @size:          allocation size
 223  *
 224  *      Allocate enough 32bit PA addressable pages to cover @size from the
 225  *      page level allocator and map them into continguos kernel virtual space.
 226  */
 227 void *vmalloc_32(unsigned long size)
 228 {
 229         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
 230 }
 231
 232 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
 233 {
 234         BUG();
 235         return NULL;
 236 }
 237
 238 void vunmap(void *addr)
 239 {
 240         BUG();
 241 }
 242
 243 /*
 244  *  sys_brk() for the most part doesn't need the global kernel
 245  *  lock, except when an application is doing something nasty
 246  *  like trying to un-brk an area that has already been mapped
 247  *  to a regular file.  in this case, the unmapping will need
 248  *  to invoke file system routines that need the global lock.
 249  */
 250 asmlinkage unsigned long sys_brk(unsigned long brk)
 251 {
 252         struct mm_struct *mm = current->mm;
 253
 254         if (brk < mm->start_brk || brk > mm->context.end_brk)
 255                 return mm->brk;
 256
 257         if (mm->brk == brk)
 258                 return mm->brk;
 259
 260         /*
 261          * Always allow shrinking brk
 262          */
 263         if (brk <= mm->brk) {
 264                 mm->brk = brk;
 265                 return brk;
 266         }
 267
 268         /*
 269          * Ok, looks good - let it rip.
 270          */
 271         return mm->brk = brk;
 272 }
 273
 274 #ifdef DEBUG
 275 static void show_process_blocks(void)
 276 {
 277         struct vm_list_struct *vml;
 278
 279         printk("Process blocks %d:", current->pid);
 280
 281         for (vml = &current->mm->context.vmlist; vml; vml = vml->next) {
 282                 printk(" %p: %p", vml, vml->vma);
 283                 if (vml->vma)
 284                         printk(" (%d @%lx #%d)",
 285                                kobjsize((void *) vml->vma->vm_start),
 286                                vml->vma->vm_start,
 287                                atomic_read(&vml->vma->vm_usage));
 288                 printk(vml->next ? " ->" : ".\n");
 289         }
 290 }
 291 #endif /* DEBUG */
 292
 293 static inline struct vm_area_struct *find_nommu_vma(unsigned long start)
 294 {
 295         struct vm_area_struct *vma;
 296         struct rb_node *n = nommu_vma_tree.rb_node;
 297
 298         while (n) {
 299                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
 300
 301                 if (start < vma->vm_start)
 302                         n = n->rb_left;
 303                 else if (start > vma->vm_start)
 304                         n = n->rb_right;
 305                 else
 306                         return vma;
 307         }
 308
 309         return NULL;
 310 }
 311
 312 static void add_nommu_vma(struct vm_area_struct *vma)
 313 {
 314         struct vm_area_struct *pvma;
 315         struct address_space *mapping;
 316         struct rb_node **p = &nommu_vma_tree.rb_node;
 317         struct rb_node *parent = NULL;
 318
 319         /* add the VMA to the mapping */
 320         if (vma->vm_file) {
 321                 mapping = vma->vm_file->f_mapping;
 322
 323                 flush_dcache_mmap_lock(mapping);
 324                 vma_prio_tree_insert(vma, &mapping->i_mmap);
 325                 flush_dcache_mmap_unlock(mapping);
 326         }
 327
 328         /* add the VMA to the master list */
 329         while (*p) {
 330                 parent = *p;
 331                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
 332
 333                 if (vma->vm_start < pvma->vm_start) {
 334                         p = &(*p)->rb_left;
 335                 }
 336                 else if (vma->vm_start > pvma->vm_start) {
 337                         p = &(*p)->rb_right;
 338                 }
 339                 else {
 340                         /* mappings are at the same address - this can only
 341                          * happen for shared-mem chardevs and shared file
 342                          * mappings backed by ramfs/tmpfs */
 343                         BUG_ON(!(pvma->vm_flags & VM_SHARED));
 344
 345                         if (vma < pvma)
 346                                 p = &(*p)->rb_left;
 347                         else if (vma > pvma)
 348                                 p = &(*p)->rb_right;
 349                         else
 350                                 BUG();
 351                 }
 352         }
 353
 354         rb_link_node(&vma->vm_rb, parent, p);
 355         rb_insert_color(&vma->vm_rb, &nommu_vma_tree);
 356 }
 357
 358 static void delete_nommu_vma(struct vm_area_struct *vma)
 359 {
 360         struct address_space *mapping;
 361
 362         /* remove the VMA from the mapping */
 363         if (vma->vm_file) {
 364                 mapping = vma->vm_file->f_mapping;
 365
 366                 flush_dcache_mmap_lock(mapping);
 367                 vma_prio_tree_remove(vma, &mapping->i_mmap);
 368                 flush_dcache_mmap_unlock(mapping);
 369         }
 370
 371         /* remove from the master list */
 372         rb_erase(&vma->vm_rb, &nommu_vma_tree);
 373 }
 374
 375 /*
 376  * determine whether a mapping should be permitted and, if so, what sort of
 377  * mapping we're capable of supporting
 378  */
 379 static int validate_mmap_request(struct file *file,
 380                                  unsigned long addr,
 381                                  unsigned long len,
 382                                  unsigned long prot,
 383                                  unsigned long flags,
 384                                  unsigned long pgoff,
 385                                  unsigned long *_capabilities)
 386 {
 387         unsigned long capabilities;
 388         unsigned long reqprot = prot;
 389         int ret;
 390
 391         /* do the simple checks first */
 392         if (flags & MAP_FIXED || addr) {
 393                 printk(KERN_DEBUG
 394                        "%d: Can't do fixed-address/overlay mmap of RAM\n",
 395                        current->pid);
 396                 return -EINVAL;
 397         }
 398
 399         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
 400             (flags & MAP_TYPE) != MAP_SHARED)
 401                 return -EINVAL;
 402
 403         if (PAGE_ALIGN(len) == 0)
 404                 return addr;
 405
 406         if (len > TASK_SIZE)
 407                 return -EINVAL;
 408
 409         /* offset overflow? */
 410         if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
 411                 return -EINVAL;
 412
 413         if (file) {
 414                 /* validate file mapping requests */
 415                 struct address_space *mapping;
 416
 417                 /* files must support mmap */
 418                 if (!file->f_op || !file->f_op->mmap)
 419                         return -ENODEV;
 420
 421                 /* work out if what we've got could possibly be shared
 422                  * - we support chardevs that provide their own "memory"
 423                  * - we support files/blockdevs that are memory backed
 424                  */
 425                 mapping = file->f_mapping;
 426                 if (!mapping)
 427                         mapping = file->f_dentry->d_inode->i_mapping;
 428
 429                 capabilities = 0;
 430                 if (mapping && mapping->backing_dev_info)
 431                         capabilities = mapping->backing_dev_info->capabilities;
 432
 433                 if (!capabilities) {
 434                         /* no explicit capabilities set, so assume some
 435                          * defaults */
 436                         switch (file->f_dentry->d_inode->i_mode & S_IFMT) {
 437                         case S_IFREG:
 438                         case S_IFBLK:
 439                                 capabilities = BDI_CAP_MAP_COPY;
 440                                 break;
 441
 442                         case S_IFCHR:
 443                                 capabilities =
 444                                         BDI_CAP_MAP_DIRECT |
 445                                         BDI_CAP_READ_MAP |
 446                                         BDI_CAP_WRITE_MAP;
 447                                 break;
 448
 449                         default:
 450                                 return -EINVAL;
 451                         }
 452                 }
 453
 454                 /* eliminate any capabilities that we can't support on this
 455                  * device */
 456                 if (!file->f_op->get_unmapped_area)
 457                         capabilities &= ~BDI_CAP_MAP_DIRECT;
 458                 if (!file->f_op->read)
 459                         capabilities &= ~BDI_CAP_MAP_COPY;
 460
 461                 if (flags & MAP_SHARED) {
 462                         /* do checks for writing, appending and locking */
 463                         if ((prot & PROT_WRITE) &&
 464                             !(file->f_mode & FMODE_WRITE))
 465                                 return -EACCES;
 466
 467                         if (IS_APPEND(file->f_dentry->d_inode) &&
 468                             (file->f_mode & FMODE_WRITE))
 469                                 return -EACCES;
 470
 471                         if (locks_verify_locked(file->f_dentry->d_inode))
 472                                 return -EAGAIN;
 473
 474                         if (!(capabilities & BDI_CAP_MAP_DIRECT))
 475                                 return -ENODEV;
 476
 477                         if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
 478                             ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
 479                             ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
 480                             ) {
 481                                 printk("MAP_SHARED not completely supported on !MMU\n");
 482                                 return -EINVAL;
 483                         }
 484
 485                         /* we mustn't privatise shared mappings */
 486                         capabilities &= ~BDI_CAP_MAP_COPY;
 487                 }
 488                 else {
 489                         /* we're going to read the file into private memory we
 490                          * allocate */
 491                         if (!(capabilities & BDI_CAP_MAP_COPY))
 492                                 return -ENODEV;
 493
 494                         /* we don't permit a private writable mapping to be
 495                          * shared with the backing device */
 496                         if (prot & PROT_WRITE)
 497                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
 498                 }
 499
 500                 /* handle executable mappings and implied executable
 501                  * mappings */
 502                 if (file->f_vfsmnt->mnt_flags & MNT_NOEXEC) {
 503                         if (prot & PROT_EXEC)
 504                                 return -EPERM;
 505                 }
 506                 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
 507                         /* handle implication of PROT_EXEC by PROT_READ */
 508                         if (current->personality & READ_IMPLIES_EXEC) {
 509                                 if (capabilities & BDI_CAP_EXEC_MAP)
 510                                         prot |= PROT_EXEC;
 511                         }
 512                 }
 513                 else if ((prot & PROT_READ) &&
 514                          (prot & PROT_EXEC) &&
 515                          !(capabilities & BDI_CAP_EXEC_MAP)
 516                          ) {
 517                         /* backing file is not executable, try to copy */
 518                         capabilities &= ~BDI_CAP_MAP_DIRECT;
 519                 }
 520         }
 521         else {
 522                 /* anonymous mappings are always memory backed and can be
 523                  * privately mapped
 524                  */
 525                 capabilities = BDI_CAP_MAP_COPY;
 526
 527                 /* handle PROT_EXEC implication by PROT_READ */
 528                 if ((prot & PROT_READ) &&
 529                     (current->personality & READ_IMPLIES_EXEC))
 530                         prot |= PROT_EXEC;
 531         }
 532
 533         /* allow the security API to have its say */
 534         ret = security_file_mmap(file, reqprot, prot, flags);
 535         if (ret < 0)
 536                 return ret;
 537
 538         /* looks okay */
 539         *_capabilities = capabilities;
 540         return 0;
 541 }
 542
 543 /*
 544  * we've determined that we can make the mapping, now translate what we
 545  * now know into VMA flags
 546  */
 547 static unsigned long determine_vm_flags(struct file *file,
 548                                         unsigned long prot,
 549                                         unsigned long flags,
 550                                         unsigned long capabilities)
 551 {
 552         unsigned long vm_flags;
 553
 554         vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
 555         vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 556         /* vm_flags |= mm->def_flags; */
 557
 558         if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
 559                 /* attempt to share read-only copies of mapped file chunks */
 560                 if (file && !(prot & PROT_WRITE))
 561                         vm_flags |= VM_MAYSHARE;
 562         }
 563         else {
 564                 /* overlay a shareable mapping on the backing device or inode
 565                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
 566                  * romfs/cramfs */
 567                 if (flags & MAP_SHARED)
 568                         vm_flags |= VM_MAYSHARE | VM_SHARED;
 569                 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
 570                         vm_flags |= VM_MAYSHARE;
 571         }
 572
 573         /* refuse to let anyone share private mappings with this process if
 574          * it's being traced - otherwise breakpoints set in it may interfere
 575          * with another untraced process
 576          */
 577         if ((flags & MAP_PRIVATE) && (current->ptrace & PT_PTRACED))
 578                 vm_flags &= ~VM_MAYSHARE;
 579
 580         return vm_flags;
 581 }
 582
 583 /*
 584  * set up a shared mapping on a file
 585  */
 586 static int do_mmap_shared_file(struct vm_area_struct *vma, unsigned long len)
 587 {
 588         int ret;
 589
 590         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
 591         if (ret != -ENOSYS)
 592                 return ret;
 593
 594         /* getting an ENOSYS error indicates that direct mmap isn't
 595          * possible (as opposed to tried but failed) so we'll fall
 596          * through to making a private copy of the data and mapping
 597          * that if we can */
 598         return -ENODEV;
 599 }
 600
 601 /*
 602  * set up a private mapping or an anonymous shared mapping
 603  */
 604 static int do_mmap_private(struct vm_area_struct *vma, unsigned long len)
 605 {
 606         void *base;
 607         int ret;
 608
 609         /* invoke the file's mapping function so that it can keep track of
 610          * shared mappings on devices or memory
 611          * - VM_MAYSHARE will be set if it may attempt to share
 612          */
 613         if (vma->vm_file) {
 614                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
 615                 if (ret != -ENOSYS) {
 616                         /* shouldn't return success if we're not sharing */
 617                         BUG_ON(ret == 0 && !(vma->vm_flags & VM_MAYSHARE));
 618                         return ret; /* success or a real error */
 619                 }
 620
 621                 /* getting an ENOSYS error indicates that direct mmap isn't
 622                  * possible (as opposed to tried but failed) so we'll try to
 623                  * make a private copy of the data and map that instead */
 624         }
 625
 626         /* allocate some memory to hold the mapping
 627          * - note that this may not return a page-aligned address if the object
 628          *   we're allocating is smaller than a page
 629          */
 630         base = kmalloc(len, GFP_KERNEL|__GFP_COMP);
 631         if (!base)
 632                 goto enomem;
 633
 634         vma->vm_start = (unsigned long) base;
 635         vma->vm_end = vma->vm_start + len;
 636         vma->vm_flags |= VM_MAPPED_COPY;
 637
 638 #ifdef WARN_ON_SLACK
 639         if (len + WARN_ON_SLACK <= kobjsize(result))
 640                 printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n",
 641                        len, current->pid, kobjsize(result) - len);
 642 #endif
 643
 644         if (vma->vm_file) {
 645                 /* read the contents of a file into the copy */
 646                 mm_segment_t old_fs;
 647                 loff_t fpos;
 648
 649                 fpos = vma->vm_pgoff;
 650                 fpos <<= PAGE_SHIFT;
 651
 652                 old_fs = get_fs();
 653                 set_fs(KERNEL_DS);
 654                 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
 655                 set_fs(old_fs);
 656
 657                 if (ret < 0)
 658                         goto error_free;
 659
 660                 /* clear the last little bit */
 661                 if (ret < len)
 662                         memset(base + ret, 0, len - ret);
 663
 664         } else {
 665                 /* if it's an anonymous mapping, then just clear it */
 666                 memset(base, 0, len);
 667         }
 668
 669         return 0;
 670
 671 error_free:
 672         kfree(base);
 673         vma->vm_start = 0;
 674         return ret;
 675
 676 enomem:
 677         printk("Allocation of length %lu from process %d failed\n",
 678                len, current->pid);
 679         show_free_areas();
 680         return -ENOMEM;
 681 }
 682
 683 /*
 684  * handle mapping creation for uClinux
 685  */
 686 unsigned long do_mmap_pgoff(struct file *file,
 687                             unsigned long addr,
 688                             unsigned long len,
 689                             unsigned long prot,
 690                             unsigned long flags,
 691                             unsigned long pgoff)
 692 {
 693         struct vm_list_struct *vml = NULL;
 694         struct vm_area_struct *vma = NULL;
 695         struct rb_node *rb;
 696         unsigned long capabilities, vm_flags;
 697         void *result;
 698         int ret;
 699
 700         /* decide whether we should attempt the mapping, and if so what sort of
 701          * mapping */
 702         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
 703                                     &capabilities);
 704         if (ret < 0)
 705                 return ret;
 706
 707         /* we've determined that we can make the mapping, now translate what we
 708          * now know into VMA flags */
 709         vm_flags = determine_vm_flags(file, prot, flags, capabilities);
 710
 711         /* we're going to need to record the mapping if it works */
 712         vml = kmalloc(sizeof(struct vm_list_struct), GFP_KERNEL);
 713         if (!vml)
 714                 goto error_getting_vml;
 715         memset(vml, 0, sizeof(*vml));
 716
 717         down_write(&nommu_vma_sem);
 718
 719         /* if we want to share, we need to check for VMAs created by other
 720          * mmap() calls that overlap with our proposed mapping
 721          * - we can only share with an exact match on most regular files
 722          * - shared mappings on character devices and memory backed files are
 723          *   permitted to overlap inexactly as far as we are concerned for in
 724          *   these cases, sharing is handled in the driver or filesystem rather
 725          *   than here
 726          */
 727         if (vm_flags & VM_MAYSHARE) {
 728                 unsigned long pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 729                 unsigned long vmpglen;
 730
 731                 for (rb = rb_first(&nommu_vma_tree); rb; rb = rb_next(rb)) {
 732                         vma = rb_entry(rb, struct vm_area_struct, vm_rb);
 733
 734                         if (!(vma->vm_flags & VM_MAYSHARE))
 735                                 continue;
 736
 737                         /* search for overlapping mappings on the same file */
 738                         if (vma->vm_file->f_dentry->d_inode != file->f_dentry->d_inode)
 739                                 continue;
 740
 741                         if (vma->vm_pgoff >= pgoff + pglen)
 742                                 continue;
 743
 744                         vmpglen = vma->vm_end - vma->vm_start + PAGE_SIZE - 1;
 745                         vmpglen >>= PAGE_SHIFT;
 746                         if (pgoff >= vma->vm_pgoff + vmpglen)
 747                                 continue;
 748
 749                         /* handle inexactly overlapping matches between mappings */
 750                         if (vma->vm_pgoff != pgoff || vmpglen != pglen) {
 751                                 if (!(capabilities & BDI_CAP_MAP_DIRECT))
 752                                         goto sharing_violation;
 753                                 continue;
 754                         }
 755
 756                         /* we've found a VMA we can share */
 757                         atomic_inc(&vma->vm_usage);
 758
 759                         vml->vma = vma;
 760                         result = (void *) vma->vm_start;
 761                         goto shared;
 762                 }
 763
 764                 vma = NULL;
 765
 766                 /* obtain the address at which to make a shared mapping
 767                  * - this is the hook for quasi-memory character devices to
 768                  *   tell us the location of a shared mapping
 769                  */
 770                 if (file && file->f_op->get_unmapped_area) {
 771                         addr = file->f_op->get_unmapped_area(file, addr, len,
 772                                                              pgoff, flags);
 773                         if (IS_ERR((void *) addr)) {
 774                                 ret = addr;
 775                                 if (ret != (unsigned long) -ENOSYS)
 776                                         goto error;
 777
 778                                 /* the driver refused to tell us where to site
 779                                  * the mapping so we'll have to attempt to copy
 780                                  * it */
 781                                 ret = (unsigned long) -ENODEV;
 782                                 if (!(capabilities & BDI_CAP_MAP_COPY))
 783                                         goto error;
 784
 785                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
 786                         }
 787                 }
 788         }
 789
 790         /* we're going to need a VMA struct as well */
 791         vma = kmalloc(sizeof(struct vm_area_struct), GFP_KERNEL);
 792         if (!vma)
 793                 goto error_getting_vma;
 794
 795         memset(vma, 0, sizeof(*vma));
 796         INIT_LIST_HEAD(&vma->anon_vma_node);
 797         atomic_set(&vma->vm_usage, 1);
 798         if (file)
 799                 get_file(file);
 800         vma->vm_file    = file;
 801         vma->vm_flags   = vm_flags;
 802         vma->vm_start   = addr;
 803         vma->vm_end     = addr + len;
 804         vma->vm_pgoff   = pgoff;
 805
 806         vml->vma = vma;
 807
 808         /* set up the mapping */
 809         if (file && vma->vm_flags & VM_SHARED)
 810                 ret = do_mmap_shared_file(vma, len);
 811         else
 812                 ret = do_mmap_private(vma, len);
 813         if (ret < 0)
 814                 goto error;
 815
 816         /* okay... we have a mapping; now we have to register it */
 817         result = (void *) vma->vm_start;
 818
 819         if (vma->vm_flags & VM_MAPPED_COPY) {
 820                 realalloc += kobjsize(result);
 821                 askedalloc += len;
 822         }
 823
 824         realalloc += kobjsize(vma);
 825         askedalloc += sizeof(*vma);
 826
 827         current->mm->total_vm += len >> PAGE_SHIFT;
 828
 829         add_nommu_vma(vma);
 830
 831  shared:
 832         realalloc += kobjsize(vml);
 833         askedalloc += sizeof(*vml);
 834
 835         vml->next = current->mm->context.vmlist;
 836         current->mm->context.vmlist = vml;
 837
 838         up_write(&nommu_vma_sem);
 839
 840         if (prot & PROT_EXEC)
 841                 flush_icache_range((unsigned long) result,
 842                                    (unsigned long) result + len);
 843
 844 #ifdef DEBUG
 845         printk("do_mmap:\n");
 846         show_process_blocks();
 847 #endif
 848
 849         return (unsigned long) result;
 850
 851  error:
 852         up_write(&nommu_vma_sem);
 853         kfree(vml);
 854         if (vma) {
 855                 fput(vma->vm_file);
 856                 kfree(vma);
 857         }
 858         return ret;
 859
 860  sharing_violation:
 861         up_write(&nommu_vma_sem);
 862         printk("Attempt to share mismatched mappings\n");
 863         kfree(vml);
 864         return -EINVAL;
 865
 866  error_getting_vma:
 867         up_write(&nommu_vma_sem);
 868         kfree(vml);
 869         printk("Allocation of vma for %lu byte allocation from process %d failed\n",
 870                len, current->pid);
 871         show_free_areas();
 872         return -ENOMEM;
 873
 874  error_getting_vml:
 875         printk("Allocation of vml for %lu byte allocation from process %d failed\n",
 876                len, current->pid);
 877         show_free_areas();
 878         return -ENOMEM;
 879 }
 880
 881 /*
 882  * handle mapping disposal for uClinux
 883  */
 884 static void put_vma(struct vm_area_struct *vma)
 885 {
 886         if (vma) {
 887                 down_write(&nommu_vma_sem);
 888
 889                 if (atomic_dec_and_test(&vma->vm_usage)) {
 890                         delete_nommu_vma(vma);
 891
 892                         if (vma->vm_ops && vma->vm_ops->close)
 893                                 vma->vm_ops->close(vma);
 894
 895                         /* IO memory and memory shared directly out of the pagecache from
 896                          * ramfs/tmpfs mustn't be released here */
 897                         if (vma->vm_flags & VM_MAPPED_COPY) {
 898                                 realalloc -= kobjsize((void *) vma->vm_start);
 899                                 askedalloc -= vma->vm_end - vma->vm_start;
 900                                 kfree((void *) vma->vm_start);
 901                         }
 902
 903                         realalloc -= kobjsize(vma);
 904                         askedalloc -= sizeof(*vma);
 905
 906                         if (vma->vm_file)
 907                                 fput(vma->vm_file);
 908                         kfree(vma);
 909                 }
 910
 911                 up_write(&nommu_vma_sem);
 912         }
 913 }
 914
 915 int do_munmap(struct mm_struct *mm, unsigned long addr, size_t len)
 916 {
 917         struct vm_list_struct *vml, **parent;
 918         unsigned long end = addr + len;
 919
 920 #ifdef DEBUG
 921         printk("do_munmap:\n");
 922 #endif
 923
 924         for (parent = &mm->context.vmlist; *parent; parent = &(*parent)->next)
 925                 if ((*parent)->vma->vm_start == addr &&
 926                     ((len == 0) || ((*parent)->vma->vm_end == end)))
 927                         goto found;
 928
 929         printk("munmap of non-mmaped memory by process %d (%s): %p\n",
 930                current->pid, current->comm, (void *) addr);
 931         return -EINVAL;
 932
 933  found:
 934         vml = *parent;
 935
 936         put_vma(vml->vma);
 937
 938         *parent = vml->next;
 939         realalloc -= kobjsize(vml);
 940         askedalloc -= sizeof(*vml);
 941         kfree(vml);
 942
 943         update_hiwater_vm(mm);
 944         mm->total_vm -= len >> PAGE_SHIFT;
 945
 946 #ifdef DEBUG
 947         show_process_blocks();
 948 #endif
 949
 950         return 0;
 951 }
 952
 953 /* Release all mmaps. */
 954 void exit_mmap(struct mm_struct * mm)
 955 {
 956         struct vm_list_struct *tmp;
 957
 958         if (mm) {
 959 #ifdef DEBUG
 960                 printk("Exit_mmap:\n");
 961 #endif
 962
 963                 mm->total_vm = 0;
 964
 965                 while ((tmp = mm->context.vmlist)) {
 966                         mm->context.vmlist = tmp->next;
 967                         put_vma(tmp->vma);
 968
 969                         realalloc -= kobjsize(tmp);
 970                         askedalloc -= sizeof(*tmp);
 971                         kfree(tmp);
 972                 }
 973
 974 #ifdef DEBUG
 975                 show_process_blocks();
 976 #endif
 977         }
 978 }
 979
 980 asmlinkage long sys_munmap(unsigned long addr, size_t len)
 981 {
 982         int ret;
 983         struct mm_struct *mm = current->mm;
 984
 985         down_write(&mm->mmap_sem);
 986         ret = do_munmap(mm, addr, len);
 987         up_write(&mm->mmap_sem);
 988         return ret;
 989 }
 990
 991 unsigned long do_brk(unsigned long addr, unsigned long len)
 992 {
 993         return -ENOMEM;
 994 }
 995
 996 /*
 997  * Expand (or shrink) an existing mapping, potentially moving it at the
 998  * same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
 999  *
1000  * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise
1001  * This option implies MREMAP_MAYMOVE.
1002  *
1003  * on uClinux, we only permit changing a mapping's size, and only as long as it stays within the
1004  * hole allocated by the kmalloc() call in do_mmap_pgoff() and the block is not shareable
1005  */
1006 unsigned long do_mremap(unsigned long addr,
1007                         unsigned long old_len, unsigned long new_len,
1008                         unsigned long flags, unsigned long new_addr)
1009 {
1010         struct vm_list_struct *vml = NULL;
1011
1012         /* insanity checks first */
1013         if (new_len == 0)
1014                 return (unsigned long) -EINVAL;
1015
1016         if (flags & MREMAP_FIXED && new_addr != addr)
1017                 return (unsigned long) -EINVAL;
1018
1019         for (vml = current->mm->context.vmlist; vml; vml = vml->next)
1020                 if (vml->vma->vm_start == addr)
1021                         goto found;
1022
1023         return (unsigned long) -EINVAL;
1024
1025  found:
1026         if (vml->vma->vm_end != vml->vma->vm_start + old_len)
1027                 return (unsigned long) -EFAULT;
1028
1029         if (vml->vma->vm_flags & VM_MAYSHARE)
1030                 return (unsigned long) -EPERM;
1031
1032         if (new_len > kobjsize((void *) addr))
1033                 return (unsigned long) -ENOMEM;
1034
1035         /* all checks complete - do it */
1036         vml->vma->vm_end = vml->vma->vm_start + new_len;
1037
1038         askedalloc -= old_len;
1039         askedalloc += new_len;
1040
1041         return vml->vma->vm_start;
1042 }
1043
1044 /*
1045  * Look up the first VMA which satisfies  addr < vm_end,  NULL if none
1046  * - should be called with mm->mmap_sem at least readlocked
1047  */
1048 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
1049 {
1050         struct vm_list_struct *vml;
1051
1052         for (vml = mm->context.vmlist; vml; vml = vml->next)
1053                 if (addr >= vml->vma->vm_start && addr < vml->vma->vm_end)
1054                         return vml->vma;
1055
1056         return NULL;
1057 }
1058
1059 EXPORT_SYMBOL(find_vma);
1060
1061 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1062                         unsigned int foll_flags)
1063 {
1064         return NULL;
1065 }
1066
1067 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
1068 {
1069         return NULL;
1070 }
1071
1072 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1073                 unsigned long to, unsigned long size, pgprot_t prot)
1074 {
1075         vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1076         return 0;
1077 }
1078 EXPORT_SYMBOL(remap_pfn_range);
1079
1080 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1081 {
1082 }
1083
1084 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1085         unsigned long len, unsigned long pgoff, unsigned long flags)
1086 {
1087         return -ENOMEM;
1088 }
1089
1090 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1091 {
1092 }
1093
1094 void unmap_mapping_range(struct address_space *mapping,
1095                          loff_t const holebegin, loff_t const holelen,
1096                          int even_cows)
1097 {
1098 }
1099 EXPORT_SYMBOL(unmap_mapping_range);
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 = global_page_state(NR_FILE_PAGES);
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 += global_page_state(NR_SLAB_RECLAIMABLE);
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
1158                 /*
1159                  * Leave reserved pages. The pages are not for anonymous pages.
1160                  */
1161                 if (n <= totalreserve_pages)
1162                         goto error;
1163                 else
1164                         n -= totalreserve_pages;
1165
1166                 /*
1167                  * Leave the last 3% for root
1168                  */
1169                 if (!cap_sys_admin)
1170                         n -= n / 32;
1171                 free += n;
1172
1173                 if (free > pages)
1174                         return 0;
1175
1176                 goto error;
1177         }
1178
1179         allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1180         /*
1181          * Leave the last 3% for root
1182          */
1183         if (!cap_sys_admin)
1184                 allowed -= allowed / 32;
1185         allowed += total_swap_pages;
1186
1187         /* Don't let a single process grow too big:
1188            leave 3% of the size of this process for other processes */
1189         allowed -= current->mm->total_vm / 32;
1190
1191         /*
1192          * cast `allowed' as a signed long because vm_committed_space
1193          * sometimes has a negative value
1194          */
1195         if (atomic_read(&vm_committed_space) < (long)allowed)
1196                 return 0;
1197 error:
1198         vm_unacct_memory(pages);
1199
1200         return -ENOMEM;
1201 }
1202
1203 int in_gate_area_no_task(unsigned long addr)
1204 {
1205         return 0;
1206 }
1207
1208 struct page *filemap_nopage(struct vm_area_struct *area,
1209                         unsigned long address, int *type)
1210 {
1211         BUG();
1212         return NULL;
1213 }
1214
1215 /*
1216  * Access another process' address space.
1217  * - source/target buffer must be kernel space
1218  */
1219 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1220 {
1221         struct vm_area_struct *vma;
1222         struct mm_struct *mm;
1223
1224         if (addr + len < addr)
1225                 return 0;
1226
1227         mm = get_task_mm(tsk);
1228         if (!mm)
1229                 return 0;
1230
1231         down_read(&mm->mmap_sem);
1232
1233         /* the access must start within one of the target process's mappings */
1234         vma = find_vma(mm, addr);
1235         if (vma) {
1236                 /* don't overrun this mapping */
1237                 if (addr + len >= vma->vm_end)
1238                         len = vma->vm_end - addr;
1239
1240                 /* only read or write mappings where it is permitted */
1241                 if (write && vma->vm_flags & VM_WRITE)
1242                         len -= copy_to_user((void *) addr, buf, len);
1243                 else if (!write && vma->vm_flags & VM_READ)
1244                         len -= copy_from_user(buf, (void *) addr, len);
1245                 else
1246                         len = 0;
1247         } else {
1248                 len = 0;
1249         }
1250
1251         up_read(&mm->mmap_sem);
1252         mmput(mm);
1253         return len;
1254 }