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  * get a list of pages in an address range belonging to the specified process
 128  * and indicate the VMA that covers each page
 129  * - this is potentially dodgy as we may end incrementing the page count of a
 130  *   slab page or a secondary page from a compound page
 131  * - don't permit access to VMAs that don't support it, such as I/O mappings
 132  */
 133 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
 134         unsigned long start, int len, int write, int force,
 135         struct page **pages, struct vm_area_struct **vmas)
 136 {
 137         struct vm_area_struct *vma;
 138         unsigned long vm_flags;
 139         int i;
 140
 141         /* calculate required read or write permissions.
 142          * - if 'force' is set, we only require the "MAY" flags.
 143          */
 144         vm_flags  = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
 145         vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
 146
 147         for (i = 0; i < len; i++) {
 148                 vma = find_vma(mm, start);
 149                 if (!vma)
 150                         goto finish_or_fault;
 151
 152                 /* protect what we can, including chardevs */
 153                 if (vma->vm_flags & (VM_IO | VM_PFNMAP) ||
 154                     !(vm_flags & vma->vm_flags))
 155                         goto finish_or_fault;
 156
 157                 if (pages) {
 158                         pages[i] = virt_to_page(start);
 159                         if (pages[i])
 160                                 page_cache_get(pages[i]);
 161                 }
 162                 if (vmas)
 163                         vmas[i] = vma;
 164                 start += PAGE_SIZE;
 165         }
 166
 167         return i;
 168
 169 finish_or_fault:
 170         return i ? : -EFAULT;
 171 }
 172
 173 EXPORT_SYMBOL(get_user_pages);
 174
 175 DEFINE_RWLOCK(vmlist_lock);
 176 struct vm_struct *vmlist;
 177
 178 void vfree(void *addr)
 179 {
 180         kfree(addr);
 181 }
 182
 183 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
 184 {
 185         /*
 186          * kmalloc doesn't like __GFP_HIGHMEM for some reason
 187          */
 188         return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
 189 }
 190
 191 struct page * vmalloc_to_page(void *addr)
 192 {
 193         return virt_to_page(addr);
 194 }
 195
 196 unsigned long vmalloc_to_pfn(void *addr)
 197 {
 198         return page_to_pfn(virt_to_page(addr));
 199 }
 200
 201
 202 long vread(char *buf, char *addr, unsigned long count)
 203 {
 204         memcpy(buf, addr, count);
 205         return count;
 206 }
 207
 208 long vwrite(char *buf, char *addr, unsigned long count)
 209 {
 210         /* Don't allow overflow */
 211         if ((unsigned long) addr + count < count)
 212                 count = -(unsigned long) addr;
 213
 214         memcpy(addr, buf, count);
 215         return(count);
 216 }
 217
 218 /*
 219  *      vmalloc  -  allocate virtually continguos memory
 220  *
 221  *      @size:          allocation size
 222  *
 223  *      Allocate enough pages to cover @size from the page level
 224  *      allocator and map them into continguos kernel virtual space.
 225  *
 226  *      For tight control over page level allocator and protection flags
 227  *      use __vmalloc() instead.
 228  */
 229 void *vmalloc(unsigned long size)
 230 {
 231        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
 232 }
 233 EXPORT_SYMBOL(vmalloc);
 234
 235 void *vmalloc_node(unsigned long size, int node)
 236 {
 237         return vmalloc(size);
 238 }
 239 EXPORT_SYMBOL(vmalloc_node);
 240
 241 /*
 242  *      vmalloc_32  -  allocate virtually continguos memory (32bit addressable)
 243  *
 244  *      @size:          allocation size
 245  *
 246  *      Allocate enough 32bit PA addressable pages to cover @size from the
 247  *      page level allocator and map them into continguos kernel virtual space.
 248  */
 249 void *vmalloc_32(unsigned long size)
 250 {
 251         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
 252 }
 253
 254 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
 255 {
 256         BUG();
 257         return NULL;
 258 }
 259
 260 void vunmap(void *addr)
 261 {
 262         BUG();
 263 }
 264
 265 /*
 266  *  sys_brk() for the most part doesn't need the global kernel
 267  *  lock, except when an application is doing something nasty
 268  *  like trying to un-brk an area that has already been mapped
 269  *  to a regular file.  in this case, the unmapping will need
 270  *  to invoke file system routines that need the global lock.
 271  */
 272 asmlinkage unsigned long sys_brk(unsigned long brk)
 273 {
 274         struct mm_struct *mm = current->mm;
 275
 276         if (brk < mm->start_brk || brk > mm->context.end_brk)
 277                 return mm->brk;
 278
 279         if (mm->brk == brk)
 280                 return mm->brk;
 281
 282         /*
 283          * Always allow shrinking brk
 284          */
 285         if (brk <= mm->brk) {
 286                 mm->brk = brk;
 287                 return brk;
 288         }
 289
 290         /*
 291          * Ok, looks good - let it rip.
 292          */
 293         return mm->brk = brk;
 294 }
 295
 296 #ifdef DEBUG
 297 static void show_process_blocks(void)
 298 {
 299         struct vm_list_struct *vml;
 300
 301         printk("Process blocks %d:", current->pid);
 302
 303         for (vml = &current->mm->context.vmlist; vml; vml = vml->next) {
 304                 printk(" %p: %p", vml, vml->vma);
 305                 if (vml->vma)
 306                         printk(" (%d @%lx #%d)",
 307                                kobjsize((void *) vml->vma->vm_start),
 308                                vml->vma->vm_start,
 309                                atomic_read(&vml->vma->vm_usage));
 310                 printk(vml->next ? " ->" : ".\n");
 311         }
 312 }
 313 #endif /* DEBUG */
 314
 315 /*
 316  * add a VMA into a process's mm_struct in the appropriate place in the list
 317  * - should be called with mm->mmap_sem held writelocked
 318  */
 319 static void add_vma_to_mm(struct mm_struct *mm, struct vm_list_struct *vml)
 320 {
 321         struct vm_list_struct **ppv;
 322
 323         for (ppv = &current->mm->context.vmlist; *ppv; ppv = &(*ppv)->next)
 324                 if ((*ppv)->vma->vm_start > vml->vma->vm_start)
 325                         break;
 326
 327         vml->next = *ppv;
 328         *ppv = vml;
 329 }
 330
 331 /*
 332  * look up the first VMA in which addr resides, NULL if none
 333  * - should be called with mm->mmap_sem at least held readlocked
 334  */
 335 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
 336 {
 337         struct vm_list_struct *loop, *vml;
 338
 339         /* search the vm_start ordered list */
 340         vml = NULL;
 341         for (loop = mm->context.vmlist; loop; loop = loop->next) {
 342                 if (loop->vma->vm_start > addr)
 343                         break;
 344                 vml = loop;
 345         }
 346
 347         if (vml && vml->vma->vm_end > addr)
 348                 return vml->vma;
 349
 350         return NULL;
 351 }
 352 EXPORT_SYMBOL(find_vma);
 353
 354 /*
 355  * find a VMA
 356  * - we don't extend stack VMAs under NOMMU conditions
 357  */
 358 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
 359 {
 360         return find_vma(mm, addr);
 361 }
 362
 363 /*
 364  * look up the first VMA exactly that exactly matches addr
 365  * - should be called with mm->mmap_sem at least held readlocked
 366  */
 367 static inline struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
 368                                                     unsigned long addr)
 369 {
 370         struct vm_list_struct *vml;
 371
 372         /* search the vm_start ordered list */
 373         for (vml = mm->context.vmlist; vml; vml = vml->next) {
 374                 if (vml->vma->vm_start == addr)
 375                         return vml->vma;
 376                 if (vml->vma->vm_start > addr)
 377                         break;
 378         }
 379
 380         return NULL;
 381 }
 382
 383 /*
 384  * find a VMA in the global tree
 385  */
 386 static inline struct vm_area_struct *find_nommu_vma(unsigned long start)
 387 {
 388         struct vm_area_struct *vma;
 389         struct rb_node *n = nommu_vma_tree.rb_node;
 390
 391         while (n) {
 392                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
 393
 394                 if (start < vma->vm_start)
 395                         n = n->rb_left;
 396                 else if (start > vma->vm_start)
 397                         n = n->rb_right;
 398                 else
 399                         return vma;
 400         }
 401
 402         return NULL;
 403 }
 404
 405 /*
 406  * add a VMA in the global tree
 407  */
 408 static void add_nommu_vma(struct vm_area_struct *vma)
 409 {
 410         struct vm_area_struct *pvma;
 411         struct address_space *mapping;
 412         struct rb_node **p = &nommu_vma_tree.rb_node;
 413         struct rb_node *parent = NULL;
 414
 415         /* add the VMA to the mapping */
 416         if (vma->vm_file) {
 417                 mapping = vma->vm_file->f_mapping;
 418
 419                 flush_dcache_mmap_lock(mapping);
 420                 vma_prio_tree_insert(vma, &mapping->i_mmap);
 421                 flush_dcache_mmap_unlock(mapping);
 422         }
 423
 424         /* add the VMA to the master list */
 425         while (*p) {
 426                 parent = *p;
 427                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
 428
 429                 if (vma->vm_start < pvma->vm_start) {
 430                         p = &(*p)->rb_left;
 431                 }
 432                 else if (vma->vm_start > pvma->vm_start) {
 433                         p = &(*p)->rb_right;
 434                 }
 435                 else {
 436                         /* mappings are at the same address - this can only
 437                          * happen for shared-mem chardevs and shared file
 438                          * mappings backed by ramfs/tmpfs */
 439                         BUG_ON(!(pvma->vm_flags & VM_SHARED));
 440
 441                         if (vma < pvma)
 442                                 p = &(*p)->rb_left;
 443                         else if (vma > pvma)
 444                                 p = &(*p)->rb_right;
 445                         else
 446                                 BUG();
 447                 }
 448         }
 449
 450         rb_link_node(&vma->vm_rb, parent, p);
 451         rb_insert_color(&vma->vm_rb, &nommu_vma_tree);
 452 }
 453
 454 /*
 455  * delete a VMA from the global list
 456  */
 457 static void delete_nommu_vma(struct vm_area_struct *vma)
 458 {
 459         struct address_space *mapping;
 460
 461         /* remove the VMA from the mapping */
 462         if (vma->vm_file) {
 463                 mapping = vma->vm_file->f_mapping;
 464
 465                 flush_dcache_mmap_lock(mapping);
 466                 vma_prio_tree_remove(vma, &mapping->i_mmap);
 467                 flush_dcache_mmap_unlock(mapping);
 468         }
 469
 470         /* remove from the master list */
 471         rb_erase(&vma->vm_rb, &nommu_vma_tree);
 472 }
 473
 474 /*
 475  * determine whether a mapping should be permitted and, if so, what sort of
 476  * mapping we're capable of supporting
 477  */
 478 static int validate_mmap_request(struct file *file,
 479                                  unsigned long addr,
 480                                  unsigned long len,
 481                                  unsigned long prot,
 482                                  unsigned long flags,
 483                                  unsigned long pgoff,
 484                                  unsigned long *_capabilities)
 485 {
 486         unsigned long capabilities;
 487         unsigned long reqprot = prot;
 488         int ret;
 489
 490         /* do the simple checks first */
 491         if (flags & MAP_FIXED || addr) {
 492                 printk(KERN_DEBUG
 493                        "%d: Can't do fixed-address/overlay mmap of RAM\n",
 494                        current->pid);
 495                 return -EINVAL;
 496         }
 497
 498         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
 499             (flags & MAP_TYPE) != MAP_SHARED)
 500                 return -EINVAL;
 501
 502         if (!len)
 503                 return -EINVAL;
 504
 505         /* Careful about overflows.. */
 506         len = PAGE_ALIGN(len);
 507         if (!len || len > TASK_SIZE)
 508                 return -ENOMEM;
 509
 510         /* offset overflow? */
 511         if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
 512                 return -EOVERFLOW;
 513
 514         if (file) {
 515                 /* validate file mapping requests */
 516                 struct address_space *mapping;
 517
 518                 /* files must support mmap */
 519                 if (!file->f_op || !file->f_op->mmap)
 520                         return -ENODEV;
 521
 522                 /* work out if what we've got could possibly be shared
 523                  * - we support chardevs that provide their own "memory"
 524                  * - we support files/blockdevs that are memory backed
 525                  */
 526                 mapping = file->f_mapping;
 527                 if (!mapping)
 528                         mapping = file->f_path.dentry->d_inode->i_mapping;
 529
 530                 capabilities = 0;
 531                 if (mapping && mapping->backing_dev_info)
 532                         capabilities = mapping->backing_dev_info->capabilities;
 533
 534                 if (!capabilities) {
 535                         /* no explicit capabilities set, so assume some
 536                          * defaults */
 537                         switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
 538                         case S_IFREG:
 539                         case S_IFBLK:
 540                                 capabilities = BDI_CAP_MAP_COPY;
 541                                 break;
 542
 543                         case S_IFCHR:
 544                                 capabilities =
 545                                         BDI_CAP_MAP_DIRECT |
 546                                         BDI_CAP_READ_MAP |
 547                                         BDI_CAP_WRITE_MAP;
 548                                 break;
 549
 550                         default:
 551                                 return -EINVAL;
 552                         }
 553                 }
 554
 555                 /* eliminate any capabilities that we can't support on this
 556                  * device */
 557                 if (!file->f_op->get_unmapped_area)
 558                         capabilities &= ~BDI_CAP_MAP_DIRECT;
 559                 if (!file->f_op->read)
 560                         capabilities &= ~BDI_CAP_MAP_COPY;
 561
 562                 if (flags & MAP_SHARED) {
 563                         /* do checks for writing, appending and locking */
 564                         if ((prot & PROT_WRITE) &&
 565                             !(file->f_mode & FMODE_WRITE))
 566                                 return -EACCES;
 567
 568                         if (IS_APPEND(file->f_path.dentry->d_inode) &&
 569                             (file->f_mode & FMODE_WRITE))
 570                                 return -EACCES;
 571
 572                         if (locks_verify_locked(file->f_path.dentry->d_inode))
 573                                 return -EAGAIN;
 574
 575                         if (!(capabilities & BDI_CAP_MAP_DIRECT))
 576                                 return -ENODEV;
 577
 578                         if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
 579                             ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
 580                             ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
 581                             ) {
 582                                 printk("MAP_SHARED not completely supported on !MMU\n");
 583                                 return -EINVAL;
 584                         }
 585
 586                         /* we mustn't privatise shared mappings */
 587                         capabilities &= ~BDI_CAP_MAP_COPY;
 588                 }
 589                 else {
 590                         /* we're going to read the file into private memory we
 591                          * allocate */
 592                         if (!(capabilities & BDI_CAP_MAP_COPY))
 593                                 return -ENODEV;
 594
 595                         /* we don't permit a private writable mapping to be
 596                          * shared with the backing device */
 597                         if (prot & PROT_WRITE)
 598                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
 599                 }
 600
 601                 /* handle executable mappings and implied executable
 602                  * mappings */
 603                 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
 604                         if (prot & PROT_EXEC)
 605                                 return -EPERM;
 606                 }
 607                 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
 608                         /* handle implication of PROT_EXEC by PROT_READ */
 609                         if (current->personality & READ_IMPLIES_EXEC) {
 610                                 if (capabilities & BDI_CAP_EXEC_MAP)
 611                                         prot |= PROT_EXEC;
 612                         }
 613                 }
 614                 else if ((prot & PROT_READ) &&
 615                          (prot & PROT_EXEC) &&
 616                          !(capabilities & BDI_CAP_EXEC_MAP)
 617                          ) {
 618                         /* backing file is not executable, try to copy */
 619                         capabilities &= ~BDI_CAP_MAP_DIRECT;
 620                 }
 621         }
 622         else {
 623                 /* anonymous mappings are always memory backed and can be
 624                  * privately mapped
 625                  */
 626                 capabilities = BDI_CAP_MAP_COPY;
 627
 628                 /* handle PROT_EXEC implication by PROT_READ */
 629                 if ((prot & PROT_READ) &&
 630                     (current->personality & READ_IMPLIES_EXEC))
 631                         prot |= PROT_EXEC;
 632         }
 633
 634         /* allow the security API to have its say */
 635         ret = security_file_mmap(file, reqprot, prot, flags);
 636         if (ret < 0)
 637                 return ret;
 638
 639         /* looks okay */
 640         *_capabilities = capabilities;
 641         return 0;
 642 }
 643
 644 /*
 645  * we've determined that we can make the mapping, now translate what we
 646  * now know into VMA flags
 647  */
 648 static unsigned long determine_vm_flags(struct file *file,
 649                                         unsigned long prot,
 650                                         unsigned long flags,
 651                                         unsigned long capabilities)
 652 {
 653         unsigned long vm_flags;
 654
 655         vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
 656         vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 657         /* vm_flags |= mm->def_flags; */
 658
 659         if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
 660                 /* attempt to share read-only copies of mapped file chunks */
 661                 if (file && !(prot & PROT_WRITE))
 662                         vm_flags |= VM_MAYSHARE;
 663         }
 664         else {
 665                 /* overlay a shareable mapping on the backing device or inode
 666                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
 667                  * romfs/cramfs */
 668                 if (flags & MAP_SHARED)
 669                         vm_flags |= VM_MAYSHARE | VM_SHARED;
 670                 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
 671                         vm_flags |= VM_MAYSHARE;
 672         }
 673
 674         /* refuse to let anyone share private mappings with this process if
 675          * it's being traced - otherwise breakpoints set in it may interfere
 676          * with another untraced process
 677          */
 678         if ((flags & MAP_PRIVATE) && (current->ptrace & PT_PTRACED))
 679                 vm_flags &= ~VM_MAYSHARE;
 680
 681         return vm_flags;
 682 }
 683
 684 /*
 685  * set up a shared mapping on a file
 686  */
 687 static int do_mmap_shared_file(struct vm_area_struct *vma, unsigned long len)
 688 {
 689         int ret;
 690
 691         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
 692         if (ret != -ENOSYS)
 693                 return ret;
 694
 695         /* getting an ENOSYS error indicates that direct mmap isn't
 696          * possible (as opposed to tried but failed) so we'll fall
 697          * through to making a private copy of the data and mapping
 698          * that if we can */
 699         return -ENODEV;
 700 }
 701
 702 /*
 703  * set up a private mapping or an anonymous shared mapping
 704  */
 705 static int do_mmap_private(struct vm_area_struct *vma, unsigned long len)
 706 {
 707         void *base;
 708         int ret;
 709
 710         /* invoke the file's mapping function so that it can keep track of
 711          * shared mappings on devices or memory
 712          * - VM_MAYSHARE will be set if it may attempt to share
 713          */
 714         if (vma->vm_file) {
 715                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
 716                 if (ret != -ENOSYS) {
 717                         /* shouldn't return success if we're not sharing */
 718                         BUG_ON(ret == 0 && !(vma->vm_flags & VM_MAYSHARE));
 719                         return ret; /* success or a real error */
 720                 }
 721
 722                 /* getting an ENOSYS error indicates that direct mmap isn't
 723                  * possible (as opposed to tried but failed) so we'll try to
 724                  * make a private copy of the data and map that instead */
 725         }
 726
 727         /* allocate some memory to hold the mapping
 728          * - note that this may not return a page-aligned address if the object
 729          *   we're allocating is smaller than a page
 730          */
 731         base = kmalloc(len, GFP_KERNEL|__GFP_COMP);
 732         if (!base)
 733                 goto enomem;
 734
 735         vma->vm_start = (unsigned long) base;
 736         vma->vm_end = vma->vm_start + len;
 737         vma->vm_flags |= VM_MAPPED_COPY;
 738
 739 #ifdef WARN_ON_SLACK
 740         if (len + WARN_ON_SLACK <= kobjsize(result))
 741                 printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n",
 742                        len, current->pid, kobjsize(result) - len);
 743 #endif
 744
 745         if (vma->vm_file) {
 746                 /* read the contents of a file into the copy */
 747                 mm_segment_t old_fs;
 748                 loff_t fpos;
 749
 750                 fpos = vma->vm_pgoff;
 751                 fpos <<= PAGE_SHIFT;
 752
 753                 old_fs = get_fs();
 754                 set_fs(KERNEL_DS);
 755                 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
 756                 set_fs(old_fs);
 757
 758                 if (ret < 0)
 759                         goto error_free;
 760
 761                 /* clear the last little bit */
 762                 if (ret < len)
 763                         memset(base + ret, 0, len - ret);
 764
 765         } else {
 766                 /* if it's an anonymous mapping, then just clear it */
 767                 memset(base, 0, len);
 768         }
 769
 770         return 0;
 771
 772 error_free:
 773         kfree(base);
 774         vma->vm_start = 0;
 775         return ret;
 776
 777 enomem:
 778         printk("Allocation of length %lu from process %d failed\n",
 779                len, current->pid);
 780         show_free_areas();
 781         return -ENOMEM;
 782 }
 783
 784 /*
 785  * handle mapping creation for uClinux
 786  */
 787 unsigned long do_mmap_pgoff(struct file *file,
 788                             unsigned long addr,
 789                             unsigned long len,
 790                             unsigned long prot,
 791                             unsigned long flags,
 792                             unsigned long pgoff)
 793 {
 794         struct vm_list_struct *vml = NULL;
 795         struct vm_area_struct *vma = NULL;
 796         struct rb_node *rb;
 797         unsigned long capabilities, vm_flags;
 798         void *result;
 799         int ret;
 800
 801         /* decide whether we should attempt the mapping, and if so what sort of
 802          * mapping */
 803         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
 804                                     &capabilities);
 805         if (ret < 0)
 806                 return ret;
 807
 808         /* we've determined that we can make the mapping, now translate what we
 809          * now know into VMA flags */
 810         vm_flags = determine_vm_flags(file, prot, flags, capabilities);
 811
 812         /* we're going to need to record the mapping if it works */
 813         vml = kzalloc(sizeof(struct vm_list_struct), GFP_KERNEL);
 814         if (!vml)
 815                 goto error_getting_vml;
 816
 817         down_write(&nommu_vma_sem);
 818
 819         /* if we want to share, we need to check for VMAs created by other
 820          * mmap() calls that overlap with our proposed mapping
 821          * - we can only share with an exact match on most regular files
 822          * - shared mappings on character devices and memory backed files are
 823          *   permitted to overlap inexactly as far as we are concerned for in
 824          *   these cases, sharing is handled in the driver or filesystem rather
 825          *   than here
 826          */
 827         if (vm_flags & VM_MAYSHARE) {
 828                 unsigned long pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 829                 unsigned long vmpglen;
 830
 831                 for (rb = rb_first(&nommu_vma_tree); rb; rb = rb_next(rb)) {
 832                         vma = rb_entry(rb, struct vm_area_struct, vm_rb);
 833
 834                         if (!(vma->vm_flags & VM_MAYSHARE))
 835                                 continue;
 836
 837                         /* search for overlapping mappings on the same file */
 838                         if (vma->vm_file->f_path.dentry->d_inode != file->f_path.dentry->d_inode)
 839                                 continue;
 840
 841                         if (vma->vm_pgoff >= pgoff + pglen)
 842                                 continue;
 843
 844                         vmpglen = vma->vm_end - vma->vm_start + PAGE_SIZE - 1;
 845                         vmpglen >>= PAGE_SHIFT;
 846                         if (pgoff >= vma->vm_pgoff + vmpglen)
 847                                 continue;
 848
 849                         /* handle inexactly overlapping matches between mappings */
 850                         if (vma->vm_pgoff != pgoff || vmpglen != pglen) {
 851                                 if (!(capabilities & BDI_CAP_MAP_DIRECT))
 852                                         goto sharing_violation;
 853                                 continue;
 854                         }
 855
 856                         /* we've found a VMA we can share */
 857                         atomic_inc(&vma->vm_usage);
 858
 859                         vml->vma = vma;
 860                         result = (void *) vma->vm_start;
 861                         goto shared;
 862                 }
 863
 864                 vma = NULL;
 865
 866                 /* obtain the address at which to make a shared mapping
 867                  * - this is the hook for quasi-memory character devices to
 868                  *   tell us the location of a shared mapping
 869                  */
 870                 if (file && file->f_op->get_unmapped_area) {
 871                         addr = file->f_op->get_unmapped_area(file, addr, len,
 872                                                              pgoff, flags);
 873                         if (IS_ERR((void *) addr)) {
 874                                 ret = addr;
 875                                 if (ret != (unsigned long) -ENOSYS)
 876                                         goto error;
 877
 878                                 /* the driver refused to tell us where to site
 879                                  * the mapping so we'll have to attempt to copy
 880                                  * it */
 881                                 ret = (unsigned long) -ENODEV;
 882                                 if (!(capabilities & BDI_CAP_MAP_COPY))
 883                                         goto error;
 884
 885                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
 886                         }
 887                 }
 888         }
 889
 890         /* we're going to need a VMA struct as well */
 891         vma = kzalloc(sizeof(struct vm_area_struct), GFP_KERNEL);
 892         if (!vma)
 893                 goto error_getting_vma;
 894
 895         INIT_LIST_HEAD(&vma->anon_vma_node);
 896         atomic_set(&vma->vm_usage, 1);
 897         if (file)
 898                 get_file(file);
 899         vma->vm_file    = file;
 900         vma->vm_flags   = vm_flags;
 901         vma->vm_start   = addr;
 902         vma->vm_end     = addr + len;
 903         vma->vm_pgoff   = pgoff;
 904
 905         vml->vma = vma;
 906
 907         /* set up the mapping */
 908         if (file && vma->vm_flags & VM_SHARED)
 909                 ret = do_mmap_shared_file(vma, len);
 910         else
 911                 ret = do_mmap_private(vma, len);
 912         if (ret < 0)
 913                 goto error;
 914
 915         /* okay... we have a mapping; now we have to register it */
 916         result = (void *) vma->vm_start;
 917
 918         if (vma->vm_flags & VM_MAPPED_COPY) {
 919                 realalloc += kobjsize(result);
 920                 askedalloc += len;
 921         }
 922
 923         realalloc += kobjsize(vma);
 924         askedalloc += sizeof(*vma);
 925
 926         vx_vmpages_add(current->mm, len >> PAGE_SHIFT);
 927
 928         add_nommu_vma(vma);
 929
 930  shared:
 931         realalloc += kobjsize(vml);
 932         askedalloc += sizeof(*vml);
 933
 934         add_vma_to_mm(current->mm, vml);
 935
 936         up_write(&nommu_vma_sem);
 937
 938         if (prot & PROT_EXEC)
 939                 flush_icache_range((unsigned long) result,
 940                                    (unsigned long) result + len);
 941
 942 #ifdef DEBUG
 943         printk("do_mmap:\n");
 944         show_process_blocks();
 945 #endif
 946
 947         return (unsigned long) result;
 948
 949  error:
 950         up_write(&nommu_vma_sem);
 951         kfree(vml);
 952         if (vma) {
 953                 if (vma->vm_file)
 954                         fput(vma->vm_file);
 955                 kfree(vma);
 956         }
 957         return ret;
 958
 959  sharing_violation:
 960         up_write(&nommu_vma_sem);
 961         printk("Attempt to share mismatched mappings\n");
 962         kfree(vml);
 963         return -EINVAL;
 964
 965  error_getting_vma:
 966         up_write(&nommu_vma_sem);
 967         kfree(vml);
 968         printk("Allocation of vma for %lu byte allocation from process %d failed\n",
 969                len, current->pid);
 970         show_free_areas();
 971         return -ENOMEM;
 972
 973  error_getting_vml:
 974         printk("Allocation of vml for %lu byte allocation from process %d failed\n",
 975                len, current->pid);
 976         show_free_areas();
 977         return -ENOMEM;
 978 }
 979
 980 /*
 981  * handle mapping disposal for uClinux
 982  */
 983 static void put_vma(struct vm_area_struct *vma)
 984 {
 985         if (vma) {
 986                 down_write(&nommu_vma_sem);
 987
 988                 if (atomic_dec_and_test(&vma->vm_usage)) {
 989                         delete_nommu_vma(vma);
 990
 991                         if (vma->vm_ops && vma->vm_ops->close)
 992                                 vma->vm_ops->close(vma);
 993
 994                         /* IO memory and memory shared directly out of the pagecache from
 995                          * ramfs/tmpfs mustn't be released here */
 996                         if (vma->vm_flags & VM_MAPPED_COPY) {
 997                                 realalloc -= kobjsize((void *) vma->vm_start);
 998                                 askedalloc -= vma->vm_end - vma->vm_start;
 999                                 kfree((void *) vma->vm_start);
1000                         }
1001
1002                         realalloc -= kobjsize(vma);
1003                         askedalloc -= sizeof(*vma);
1004
1005                         if (vma->vm_file)
1006                                 fput(vma->vm_file);
1007                         kfree(vma);
1008                 }
1009
1010                 up_write(&nommu_vma_sem);
1011         }
1012 }
1013
1014 /*
1015  * release a mapping
1016  * - under NOMMU conditions the parameters must match exactly to the mapping to
1017  *   be removed
1018  */
1019 int do_munmap(struct mm_struct *mm, unsigned long addr, size_t len)
1020 {
1021         struct vm_list_struct *vml, **parent;
1022         unsigned long end = addr + len;
1023
1024 #ifdef DEBUG
1025         printk("do_munmap:\n");
1026 #endif
1027
1028         for (parent = &mm->context.vmlist; *parent; parent = &(*parent)->next) {
1029                 if ((*parent)->vma->vm_start > addr)
1030                         break;
1031                 if ((*parent)->vma->vm_start == addr &&
1032                     ((len == 0) || ((*parent)->vma->vm_end == end)))
1033                         goto found;
1034         }
1035
1036         printk("munmap of non-mmaped memory by process %d (%s): %p\n",
1037                current->pid, current->comm, (void *) addr);
1038         return -EINVAL;
1039
1040  found:
1041         vml = *parent;
1042
1043         put_vma(vml->vma);
1044
1045         *parent = vml->next;
1046         realalloc -= kobjsize(vml);
1047         askedalloc -= sizeof(*vml);
1048         kfree(vml);
1049
1050         update_hiwater_vm(mm);
1051         vx_vmpages_sub(mm, len >> PAGE_SHIFT);
1052
1053 #ifdef DEBUG
1054         show_process_blocks();
1055 #endif
1056
1057         return 0;
1058 }
1059
1060 asmlinkage long sys_munmap(unsigned long addr, size_t len)
1061 {
1062         int ret;
1063         struct mm_struct *mm = current->mm;
1064
1065         down_write(&mm->mmap_sem);
1066         ret = do_munmap(mm, addr, len);
1067         up_write(&mm->mmap_sem);
1068         return ret;
1069 }
1070
1071 /*
1072  * Release all mappings
1073  */
1074 void exit_mmap(struct mm_struct * mm)
1075 {
1076         struct vm_list_struct *tmp;
1077
1078         if (mm) {
1079 #ifdef DEBUG
1080                 printk("Exit_mmap:\n");
1081 #endif
1082
1083                 vx_vmpages_sub(mm, mm->total_vm);
1084
1085                 while ((tmp = mm->context.vmlist)) {
1086                         mm->context.vmlist = tmp->next;
1087                         put_vma(tmp->vma);
1088
1089                         realalloc -= kobjsize(tmp);
1090                         askedalloc -= sizeof(*tmp);
1091                         kfree(tmp);
1092                 }
1093
1094 #ifdef DEBUG
1095                 show_process_blocks();
1096 #endif
1097         }
1098 }
1099
1100 unsigned long do_brk(unsigned long addr, unsigned long len)
1101 {
1102         return -ENOMEM;
1103 }
1104
1105 /*
1106  * expand (or shrink) an existing mapping, potentially moving it at the same
1107  * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1108  *
1109  * under NOMMU conditions, we only permit changing a mapping's size, and only
1110  * as long as it stays within the hole allocated by the kmalloc() call in
1111  * do_mmap_pgoff() and the block is not shareable
1112  *
1113  * MREMAP_FIXED is not supported under NOMMU conditions
1114  */
1115 unsigned long do_mremap(unsigned long addr,
1116                         unsigned long old_len, unsigned long new_len,
1117                         unsigned long flags, unsigned long new_addr)
1118 {
1119         struct vm_area_struct *vma;
1120
1121         /* insanity checks first */
1122         if (new_len == 0)
1123                 return (unsigned long) -EINVAL;
1124
1125         if (flags & MREMAP_FIXED && new_addr != addr)
1126                 return (unsigned long) -EINVAL;
1127
1128         vma = find_vma_exact(current->mm, addr);
1129         if (!vma)
1130                 return (unsigned long) -EINVAL;
1131
1132         if (vma->vm_end != vma->vm_start + old_len)
1133                 return (unsigned long) -EFAULT;
1134
1135         if (vma->vm_flags & VM_MAYSHARE)
1136                 return (unsigned long) -EPERM;
1137
1138         if (new_len > kobjsize((void *) addr))
1139                 return (unsigned long) -ENOMEM;
1140
1141         /* all checks complete - do it */
1142         vma->vm_end = vma->vm_start + new_len;
1143
1144         askedalloc -= old_len;
1145         askedalloc += new_len;
1146
1147         return vma->vm_start;
1148 }
1149
1150 asmlinkage unsigned long sys_mremap(unsigned long addr,
1151         unsigned long old_len, unsigned long new_len,
1152         unsigned long flags, unsigned long new_addr)
1153 {
1154         unsigned long ret;
1155
1156         down_write(&current->mm->mmap_sem);
1157         ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1158         up_write(&current->mm->mmap_sem);
1159         return ret;
1160 }
1161
1162 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1163                         unsigned int foll_flags)
1164 {
1165         return NULL;
1166 }
1167
1168 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1169                 unsigned long to, unsigned long size, pgprot_t prot)
1170 {
1171         vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1172         return 0;
1173 }
1174 EXPORT_SYMBOL(remap_pfn_range);
1175
1176 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1177 {
1178 }
1179
1180 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1181         unsigned long len, unsigned long pgoff, unsigned long flags)
1182 {
1183         return -ENOMEM;
1184 }
1185
1186 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1187 {
1188 }
1189
1190 void unmap_mapping_range(struct address_space *mapping,
1191                          loff_t const holebegin, loff_t const holelen,
1192                          int even_cows)
1193 {
1194 }
1195 EXPORT_SYMBOL(unmap_mapping_range);
1196
1197 /*
1198  * Check that a process has enough memory to allocate a new virtual
1199  * mapping. 0 means there is enough memory for the allocation to
1200  * succeed and -ENOMEM implies there is not.
1201  *
1202  * We currently support three overcommit policies, which are set via the
1203  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1204  *
1205  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1206  * Additional code 2002 Jul 20 by Robert Love.
1207  *
1208  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1209  *
1210  * Note this is a helper function intended to be used by LSMs which
1211  * wish to use this logic.
1212  */
1213 int __vm_enough_memory(long pages, int cap_sys_admin)
1214 {
1215         unsigned long free, allowed;
1216
1217         vm_acct_memory(pages);
1218
1219         /*
1220          * Sometimes we want to use more memory than we have
1221          */
1222         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1223                 return 0;
1224
1225         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1226                 unsigned long n;
1227
1228                 free = global_page_state(NR_FILE_PAGES);
1229                 free += nr_swap_pages;
1230
1231                 /*
1232                  * Any slabs which are created with the
1233                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1234                  * which are reclaimable, under pressure.  The dentry
1235                  * cache and most inode caches should fall into this
1236                  */
1237                 free += global_page_state(NR_SLAB_RECLAIMABLE);
1238
1239                 /*
1240                  * Leave the last 3% for root
1241                  */
1242                 if (!cap_sys_admin)
1243                         free -= free / 32;
1244
1245                 if (free > pages)
1246                         return 0;
1247
1248                 /*
1249                  * nr_free_pages() is very expensive on large systems,
1250                  * only call if we're about to fail.
1251                  */
1252                 n = nr_free_pages();
1253
1254                 /*
1255                  * Leave reserved pages. The pages are not for anonymous pages.
1256                  */
1257                 if (n <= totalreserve_pages)
1258                         goto error;
1259                 else
1260                         n -= totalreserve_pages;
1261
1262                 /*
1263                  * Leave the last 3% for root
1264                  */
1265                 if (!cap_sys_admin)
1266                         n -= n / 32;
1267                 free += n;
1268
1269                 if (free > pages)
1270                         return 0;
1271
1272                 goto error;
1273         }
1274
1275         allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1276         /*
1277          * Leave the last 3% for root
1278          */
1279         if (!cap_sys_admin)
1280                 allowed -= allowed / 32;
1281         allowed += total_swap_pages;
1282
1283         /* Don't let a single process grow too big:
1284            leave 3% of the size of this process for other processes */
1285         allowed -= current->mm->total_vm / 32;
1286
1287         /*
1288          * cast `allowed' as a signed long because vm_committed_space
1289          * sometimes has a negative value
1290          */
1291         if (atomic_read(&vm_committed_space) < (long)allowed)
1292                 return 0;
1293 error:
1294         vm_unacct_memory(pages);
1295
1296         return -ENOMEM;
1297 }
1298
1299 int in_gate_area_no_task(unsigned long addr)
1300 {
1301         return 0;
1302 }
1303
1304 struct page *filemap_nopage(struct vm_area_struct *area,
1305                         unsigned long address, int *type)
1306 {
1307         BUG();
1308         return NULL;
1309 }
1310
1311 /*
1312  * Access another process' address space.
1313  * - source/target buffer must be kernel space
1314  */
1315 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1316 {
1317         struct vm_area_struct *vma;
1318         struct mm_struct *mm;
1319
1320         if (addr + len < addr)
1321                 return 0;
1322
1323         mm = get_task_mm(tsk);
1324         if (!mm)
1325                 return 0;
1326
1327         down_read(&mm->mmap_sem);
1328
1329         /* the access must start within one of the target process's mappings */
1330         vma = find_vma(mm, addr);
1331         if (vma) {
1332                 /* don't overrun this mapping */
1333                 if (addr + len >= vma->vm_end)
1334                         len = vma->vm_end - addr;
1335
1336                 /* only read or write mappings where it is permitted */
1337                 if (write && vma->vm_flags & VM_MAYWRITE)
1338                         len -= copy_to_user((void *) addr, buf, len);
1339                 else if (!write && vma->vm_flags & VM_MAYREAD)
1340                         len -= copy_from_user(buf, (void *) addr, len);
1341                 else
1342                         len = 0;
1343         } else {
1344                 len = 0;
1345         }
1346
1347         up_read(&mm->mmap_sem);
1348         mmput(mm);
1349         return len;
1350 }