mm/nommu.c

   1 /*
   2  *  linux/mm/nommu.c
   3  *
   4  *  Replacement code for mm functions to support CPU's that don't
   5  *  have any form of memory management unit (thus no virtual memory).
   6  *
   7  *  See Documentation/nommu-mmap.txt
   8  *
   9  *  Copyright (c) 2004-2005 David Howells <dhowells@redhat.com>
  10  *  Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
  11  *  Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
  12  *  Copyright (c) 2002      Greg Ungerer <gerg@snapgear.com>
  13  */
  14
  15 #include <linux/mm.h>
  16 #include <linux/mman.h>
  17 #include <linux/swap.h>
  18 #include <linux/file.h>
  19 #include <linux/highmem.h>
  20 #include <linux/pagemap.h>
  21 #include <linux/slab.h>
  22 #include <linux/vmalloc.h>
  23 #include <linux/ptrace.h>
  24 #include <linux/blkdev.h>
  25 #include <linux/backing-dev.h>
  26 #include <linux/mount.h>
  27 #include <linux/personality.h>
  28 #include <linux/security.h>
  29 #include <linux/syscalls.h>
  30
  31 #include <asm/uaccess.h>
  32 #include <asm/tlb.h>
  33 #include <asm/tlbflush.h>
  34
  35 void *high_memory;
  36 struct page *mem_map;
  37 unsigned long max_mapnr;
  38 unsigned long num_physpages;
  39 unsigned long askedalloc, realalloc;
  40 atomic_t vm_committed_space = ATOMIC_INIT(0);
  41 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
  42 int sysctl_overcommit_ratio = 50; /* default is 50% */
  43 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
  44 int heap_stack_gap = 0;
  45
  46 EXPORT_SYMBOL(mem_map);
  47 EXPORT_SYMBOL(sysctl_max_map_count);
  48 EXPORT_SYMBOL(sysctl_overcommit_memory);
  49 EXPORT_SYMBOL(sysctl_overcommit_ratio);
  50 EXPORT_SYMBOL(vm_committed_space);
  51 EXPORT_SYMBOL(__vm_enough_memory);
  52
  53 /* list of shareable VMAs */
  54 struct rb_root nommu_vma_tree = RB_ROOT;
  55 DECLARE_RWSEM(nommu_vma_sem);
  56
  57 struct vm_operations_struct generic_file_vm_ops = {
  58 };
  59
  60 /*
  61  * Handle all mappings that got truncated by a "truncate()"
  62  * system call.
  63  *
  64  * NOTE! We have to be ready to update the memory sharing
  65  * between the file and the memory map for a potential last
  66  * incomplete page.  Ugly, but necessary.
  67  */
  68 int vmtruncate(struct inode *inode, loff_t offset)
  69 {
  70         struct address_space *mapping = inode->i_mapping;
  71         unsigned long limit;
  72
  73         if (inode->i_size < offset)
  74                 goto do_expand;
  75         i_size_write(inode, offset);
  76
  77         truncate_inode_pages(mapping, offset);
  78         goto out_truncate;
  79
  80 do_expand:
  81         limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
  82         if (limit != RLIM_INFINITY && offset > limit)
  83                 goto out_sig;
  84         if (offset > inode->i_sb->s_maxbytes)
  85                 goto out;
  86         i_size_write(inode, offset);
  87
  88 out_truncate:
  89         if (inode->i_op && inode->i_op->truncate)
  90                 inode->i_op->truncate(inode);
  91         return 0;
  92 out_sig:
  93         send_sig(SIGXFSZ, current, 0);
  94 out:
  95         return -EFBIG;
  96 }
  97
  98 EXPORT_SYMBOL(vmtruncate);
  99
 100 /*
 101  * Return the total memory allocated for this pointer, not
 102  * just what the caller asked for.
 103  *
 104  * Doesn't have to be accurate, i.e. may have races.
 105  */
 106 unsigned int kobjsize(const void *objp)
 107 {
 108         struct page *page;
 109
 110         if (!objp || !((page = virt_to_page(objp))))
 111                 return 0;
 112
 113         if (PageSlab(page))
 114                 return ksize(objp);
 115
 116         BUG_ON(page->index < 0);
 117         BUG_ON(page->index >= MAX_ORDER);
 118
 119         return (PAGE_SIZE << page->index);
 120 }
 121
 122 /*
 123  * The nommu dodgy version :-)
 124  */
 125 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
 126         unsigned long start, int len, int write, int force,
 127         struct page **pages, struct vm_area_struct **vmas)
 128 {
 129         int i;
 130         static struct vm_area_struct dummy_vma;
 131
 132         for (i = 0; i < len; i++) {
 133                 if (pages) {
 134                         pages[i] = virt_to_page(start);
 135                         if (pages[i])
 136                                 page_cache_get(pages[i]);
 137                 }
 138                 if (vmas)
 139                         vmas[i] = &dummy_vma;
 140                 start += PAGE_SIZE;
 141         }
 142         return(i);
 143 }
 144
 145 DEFINE_RWLOCK(vmlist_lock);
 146 struct vm_struct *vmlist;
 147
 148 void vfree(void *addr)
 149 {
 150         kfree(addr);
 151 }
 152
 153 void *__vmalloc(unsigned long size, unsigned int __nocast gfp_mask,
 154                         pgprot_t prot)
 155 {
 156         /*
 157          * kmalloc doesn't like __GFP_HIGHMEM for some reason
 158          */
 159         return kmalloc(size, gfp_mask & ~__GFP_HIGHMEM);
 160 }
 161
 162 struct page * vmalloc_to_page(void *addr)
 163 {
 164         return virt_to_page(addr);
 165 }
 166
 167 unsigned long vmalloc_to_pfn(void *addr)
 168 {
 169         return page_to_pfn(virt_to_page(addr));
 170 }
 171
 172
 173 long vread(char *buf, char *addr, unsigned long count)
 174 {
 175         memcpy(buf, addr, count);
 176         return count;
 177 }
 178
 179 long vwrite(char *buf, char *addr, unsigned long count)
 180 {
 181         /* Don't allow overflow */
 182         if ((unsigned long) addr + count < count)
 183                 count = -(unsigned long) addr;
 184
 185         memcpy(addr, buf, count);
 186         return(count);
 187 }
 188
 189 /*
 190  *      vmalloc  -  allocate virtually continguos memory
 191  *
 192  *      @size:          allocation size
 193  *
 194  *      Allocate enough pages to cover @size from the page level
 195  *      allocator and map them into continguos kernel virtual space.
 196  *
 197  *      For tight cotrol over page level allocator and protection flags
 198  *      use __vmalloc() instead.
 199  */
 200 void *vmalloc(unsigned long size)
 201 {
 202        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
 203 }
 204
 205 /*
 206  *      vmalloc_32  -  allocate virtually continguos memory (32bit addressable)
 207  *
 208  *      @size:          allocation size
 209  *
 210  *      Allocate enough 32bit PA addressable pages to cover @size from the
 211  *      page level allocator and map them into continguos kernel virtual space.
 212  */
 213 void *vmalloc_32(unsigned long size)
 214 {
 215         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
 216 }
 217
 218 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
 219 {
 220         BUG();
 221         return NULL;
 222 }
 223
 224 void vunmap(void *addr)
 225 {
 226         BUG();
 227 }
 228
 229 /*
 230  *  sys_brk() for the most part doesn't need the global kernel
 231  *  lock, except when an application is doing something nasty
 232  *  like trying to un-brk an area that has already been mapped
 233  *  to a regular file.  in this case, the unmapping will need
 234  *  to invoke file system routines that need the global lock.
 235  */
 236 asmlinkage unsigned long sys_brk(unsigned long brk)
 237 {
 238         struct mm_struct *mm = current->mm;
 239
 240         if (brk < mm->start_brk || brk > mm->context.end_brk)
 241                 return mm->brk;
 242
 243         if (mm->brk == brk)
 244                 return mm->brk;
 245
 246         /*
 247          * Always allow shrinking brk
 248          */
 249         if (brk <= mm->brk) {
 250                 mm->brk = brk;
 251                 return brk;
 252         }
 253
 254         /*
 255          * Ok, looks good - let it rip.
 256          */
 257         return mm->brk = brk;
 258 }
 259
 260 #ifdef DEBUG
 261 static void show_process_blocks(void)
 262 {
 263         struct vm_list_struct *vml;
 264
 265         printk("Process blocks %d:", current->pid);
 266
 267         for (vml = &current->mm->context.vmlist; vml; vml = vml->next) {
 268                 printk(" %p: %p", vml, vml->vma);
 269                 if (vml->vma)
 270                         printk(" (%d @%lx #%d)",
 271                                kobjsize((void *) vml->vma->vm_start),
 272                                vml->vma->vm_start,
 273                                atomic_read(&vml->vma->vm_usage));
 274                 printk(vml->next ? " ->" : ".\n");
 275         }
 276 }
 277 #endif /* DEBUG */
 278
 279 static inline struct vm_area_struct *find_nommu_vma(unsigned long start)
 280 {
 281         struct vm_area_struct *vma;
 282         struct rb_node *n = nommu_vma_tree.rb_node;
 283
 284         while (n) {
 285                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
 286
 287                 if (start < vma->vm_start)
 288                         n = n->rb_left;
 289                 else if (start > vma->vm_start)
 290                         n = n->rb_right;
 291                 else
 292                         return vma;
 293         }
 294
 295         return NULL;
 296 }
 297
 298 static void add_nommu_vma(struct vm_area_struct *vma)
 299 {
 300         struct vm_area_struct *pvma;
 301         struct address_space *mapping;
 302         struct rb_node **p = &nommu_vma_tree.rb_node;
 303         struct rb_node *parent = NULL;
 304
 305         /* add the VMA to the mapping */
 306         if (vma->vm_file) {
 307                 mapping = vma->vm_file->f_mapping;
 308
 309                 flush_dcache_mmap_lock(mapping);
 310                 vma_prio_tree_insert(vma, &mapping->i_mmap);
 311                 flush_dcache_mmap_unlock(mapping);
 312         }
 313
 314         /* add the VMA to the master list */
 315         while (*p) {
 316                 parent = *p;
 317                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
 318
 319                 if (vma->vm_start < pvma->vm_start) {
 320                         p = &(*p)->rb_left;
 321                 }
 322                 else if (vma->vm_start > pvma->vm_start) {
 323                         p = &(*p)->rb_right;
 324                 }
 325                 else {
 326                         /* mappings are at the same address - this can only
 327                          * happen for shared-mem chardevs and shared file
 328                          * mappings backed by ramfs/tmpfs */
 329                         BUG_ON(!(pvma->vm_flags & VM_SHARED));
 330
 331                         if (vma < pvma)
 332                                 p = &(*p)->rb_left;
 333                         else if (vma > pvma)
 334                                 p = &(*p)->rb_right;
 335                         else
 336                                 BUG();
 337                 }
 338         }
 339
 340         rb_link_node(&vma->vm_rb, parent, p);
 341         rb_insert_color(&vma->vm_rb, &nommu_vma_tree);
 342 }
 343
 344 static void delete_nommu_vma(struct vm_area_struct *vma)
 345 {
 346         struct address_space *mapping;
 347
 348         /* remove the VMA from the mapping */
 349         if (vma->vm_file) {
 350                 mapping = vma->vm_file->f_mapping;
 351
 352                 flush_dcache_mmap_lock(mapping);
 353                 vma_prio_tree_remove(vma, &mapping->i_mmap);
 354                 flush_dcache_mmap_unlock(mapping);
 355         }
 356
 357         /* remove from the master list */
 358         rb_erase(&vma->vm_rb, &nommu_vma_tree);
 359 }
 360
 361 /*
 362  * determine whether a mapping should be permitted and, if so, what sort of
 363  * mapping we're capable of supporting
 364  */
 365 static int validate_mmap_request(struct file *file,
 366                                  unsigned long addr,
 367                                  unsigned long len,
 368                                  unsigned long prot,
 369                                  unsigned long flags,
 370                                  unsigned long pgoff,
 371                                  unsigned long *_capabilities)
 372 {
 373         unsigned long capabilities;
 374         unsigned long reqprot = prot;
 375         int ret;
 376
 377         /* do the simple checks first */
 378         if (flags & MAP_FIXED || addr) {
 379                 printk(KERN_DEBUG
 380                        "%d: Can't do fixed-address/overlay mmap of RAM\n",
 381                        current->pid);
 382                 return -EINVAL;
 383         }
 384
 385         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
 386             (flags & MAP_TYPE) != MAP_SHARED)
 387                 return -EINVAL;
 388
 389         if (PAGE_ALIGN(len) == 0)
 390                 return addr;
 391
 392         if (len > TASK_SIZE)
 393                 return -EINVAL;
 394
 395         /* offset overflow? */
 396         if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
 397                 return -EINVAL;
 398
 399         if (file) {
 400                 /* validate file mapping requests */
 401                 struct address_space *mapping;
 402
 403                 /* files must support mmap */
 404                 if (!file->f_op || !file->f_op->mmap)
 405                         return -ENODEV;
 406
 407                 /* work out if what we've got could possibly be shared
 408                  * - we support chardevs that provide their own "memory"
 409                  * - we support files/blockdevs that are memory backed
 410                  */
 411                 mapping = file->f_mapping;
 412                 if (!mapping)
 413                         mapping = file->f_dentry->d_inode->i_mapping;
 414
 415                 capabilities = 0;
 416                 if (mapping && mapping->backing_dev_info)
 417                         capabilities = mapping->backing_dev_info->capabilities;
 418
 419                 if (!capabilities) {
 420                         /* no explicit capabilities set, so assume some
 421                          * defaults */
 422                         switch (file->f_dentry->d_inode->i_mode & S_IFMT) {
 423                         case S_IFREG:
 424                         case S_IFBLK:
 425                                 capabilities = BDI_CAP_MAP_COPY;
 426                                 break;
 427
 428                         case S_IFCHR:
 429                                 capabilities =
 430                                         BDI_CAP_MAP_DIRECT |
 431                                         BDI_CAP_READ_MAP |
 432                                         BDI_CAP_WRITE_MAP;
 433                                 break;
 434
 435                         default:
 436                                 return -EINVAL;
 437                         }
 438                 }
 439
 440                 /* eliminate any capabilities that we can't support on this
 441                  * device */
 442                 if (!file->f_op->get_unmapped_area)
 443                         capabilities &= ~BDI_CAP_MAP_DIRECT;
 444                 if (!file->f_op->read)
 445                         capabilities &= ~BDI_CAP_MAP_COPY;
 446
 447                 if (flags & MAP_SHARED) {
 448                         /* do checks for writing, appending and locking */
 449                         if ((prot & PROT_WRITE) &&
 450                             !(file->f_mode & FMODE_WRITE))
 451                                 return -EACCES;
 452
 453                         if (IS_APPEND(file->f_dentry->d_inode) &&
 454                             (file->f_mode & FMODE_WRITE))
 455                                 return -EACCES;
 456
 457                         if (locks_verify_locked(file->f_dentry->d_inode))
 458                                 return -EAGAIN;
 459
 460                         if (!(capabilities & BDI_CAP_MAP_DIRECT))
 461                                 return -ENODEV;
 462
 463                         if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
 464                             ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
 465                             ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
 466                             ) {
 467                                 printk("MAP_SHARED not completely supported on !MMU\n");
 468                                 return -EINVAL;
 469                         }
 470
 471                         /* we mustn't privatise shared mappings */
 472                         capabilities &= ~BDI_CAP_MAP_COPY;
 473                 }
 474                 else {
 475                         /* we're going to read the file into private memory we
 476                          * allocate */
 477                         if (!(capabilities & BDI_CAP_MAP_COPY))
 478                                 return -ENODEV;
 479
 480                         /* we don't permit a private writable mapping to be
 481                          * shared with the backing device */
 482                         if (prot & PROT_WRITE)
 483                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
 484                 }
 485
 486                 /* handle executable mappings and implied executable
 487                  * mappings */
 488                 if (file->f_vfsmnt->mnt_flags & MNT_NOEXEC) {
 489                         if (prot & PROT_EXEC)
 490                                 return -EPERM;
 491                 }
 492                 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
 493                         /* handle implication of PROT_EXEC by PROT_READ */
 494                         if (current->personality & READ_IMPLIES_EXEC) {
 495                                 if (capabilities & BDI_CAP_EXEC_MAP)
 496                                         prot |= PROT_EXEC;
 497                         }
 498                 }
 499                 else if ((prot & PROT_READ) &&
 500                          (prot & PROT_EXEC) &&
 501                          !(capabilities & BDI_CAP_EXEC_MAP)
 502                          ) {
 503                         /* backing file is not executable, try to copy */
 504                         capabilities &= ~BDI_CAP_MAP_DIRECT;
 505                 }
 506         }
 507         else {
 508                 /* anonymous mappings are always memory backed and can be
 509                  * privately mapped
 510                  */
 511                 capabilities = BDI_CAP_MAP_COPY;
 512
 513                 /* handle PROT_EXEC implication by PROT_READ */
 514                 if ((prot & PROT_READ) &&
 515                     (current->personality & READ_IMPLIES_EXEC))
 516                         prot |= PROT_EXEC;
 517         }
 518
 519         /* allow the security API to have its say */
 520         ret = security_file_mmap(file, reqprot, prot, flags);
 521         if (ret < 0)
 522                 return ret;
 523
 524         /* looks okay */
 525         *_capabilities = capabilities;
 526         return 0;
 527 }
 528
 529 /*
 530  * we've determined that we can make the mapping, now translate what we
 531  * now know into VMA flags
 532  */
 533 static unsigned long determine_vm_flags(struct file *file,
 534                                         unsigned long prot,
 535                                         unsigned long flags,
 536                                         unsigned long capabilities)
 537 {
 538         unsigned long vm_flags;
 539
 540         vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
 541         vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 542         /* vm_flags |= mm->def_flags; */
 543
 544         if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
 545                 /* attempt to share read-only copies of mapped file chunks */
 546                 if (file && !(prot & PROT_WRITE))
 547                         vm_flags |= VM_MAYSHARE;
 548         }
 549         else {
 550                 /* overlay a shareable mapping on the backing device or inode
 551                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
 552                  * romfs/cramfs */
 553                 if (flags & MAP_SHARED)
 554                         vm_flags |= VM_MAYSHARE | VM_SHARED;
 555                 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
 556                         vm_flags |= VM_MAYSHARE;
 557         }
 558
 559         /* refuse to let anyone share private mappings with this process if
 560          * it's being traced - otherwise breakpoints set in it may interfere
 561          * with another untraced process
 562          */
 563         if ((flags & MAP_PRIVATE) && (current->ptrace & PT_PTRACED))
 564                 vm_flags &= ~VM_MAYSHARE;
 565
 566         return vm_flags;
 567 }
 568
 569 /*
 570  * set up a shared mapping on a file
 571  */
 572 static int do_mmap_shared_file(struct vm_area_struct *vma, unsigned long len)
 573 {
 574         int ret;
 575
 576         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
 577         if (ret != -ENOSYS)
 578                 return ret;
 579
 580         /* getting an ENOSYS error indicates that direct mmap isn't
 581          * possible (as opposed to tried but failed) so we'll fall
 582          * through to making a private copy of the data and mapping
 583          * that if we can */
 584         return -ENODEV;
 585 }
 586
 587 /*
 588  * set up a private mapping or an anonymous shared mapping
 589  */
 590 static int do_mmap_private(struct vm_area_struct *vma, unsigned long len)
 591 {
 592         void *base;
 593         int ret;
 594
 595         /* invoke the file's mapping function so that it can keep track of
 596          * shared mappings on devices or memory
 597          * - VM_MAYSHARE will be set if it may attempt to share
 598          */
 599         if (vma->vm_file) {
 600                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
 601                 if (ret != -ENOSYS) {
 602                         /* shouldn't return success if we're not sharing */
 603                         BUG_ON(ret == 0 && !(vma->vm_flags & VM_MAYSHARE));
 604                         return ret; /* success or a real error */
 605                 }
 606
 607                 /* getting an ENOSYS error indicates that direct mmap isn't
 608                  * possible (as opposed to tried but failed) so we'll try to
 609                  * make a private copy of the data and map that instead */
 610         }
 611
 612         /* allocate some memory to hold the mapping
 613          * - note that this may not return a page-aligned address if the object
 614          *   we're allocating is smaller than a page
 615          */
 616         base = kmalloc(len, GFP_KERNEL);
 617         if (!base)
 618                 goto enomem;
 619
 620         vma->vm_start = (unsigned long) base;
 621         vma->vm_end = vma->vm_start + len;
 622         vma->vm_flags |= VM_MAPPED_COPY;
 623
 624 #ifdef WARN_ON_SLACK
 625         if (len + WARN_ON_SLACK <= kobjsize(result))
 626                 printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n",
 627                        len, current->pid, kobjsize(result) - len);
 628 #endif
 629
 630         if (vma->vm_file) {
 631                 /* read the contents of a file into the copy */
 632                 mm_segment_t old_fs;
 633                 loff_t fpos;
 634
 635                 fpos = vma->vm_pgoff;
 636                 fpos <<= PAGE_SHIFT;
 637
 638                 old_fs = get_fs();
 639                 set_fs(KERNEL_DS);
 640                 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
 641                 set_fs(old_fs);
 642
 643                 if (ret < 0)
 644                         goto error_free;
 645
 646                 /* clear the last little bit */
 647                 if (ret < len)
 648                         memset(base + ret, 0, len - ret);
 649
 650         } else {
 651                 /* if it's an anonymous mapping, then just clear it */
 652                 memset(base, 0, len);
 653         }
 654
 655         return 0;
 656
 657 error_free:
 658         kfree(base);
 659         vma->vm_start = 0;
 660         return ret;
 661
 662 enomem:
 663         printk("Allocation of length %lu from process %d failed\n",
 664                len, current->pid);
 665         show_free_areas();
 666         return -ENOMEM;
 667 }
 668
 669 /*
 670  * handle mapping creation for uClinux
 671  */
 672 unsigned long do_mmap_pgoff(struct file *file,
 673                             unsigned long addr,
 674                             unsigned long len,
 675                             unsigned long prot,
 676                             unsigned long flags,
 677                             unsigned long pgoff)
 678 {
 679         struct vm_list_struct *vml = NULL;
 680         struct vm_area_struct *vma = NULL;
 681         struct rb_node *rb;
 682         unsigned long capabilities, vm_flags;
 683         void *result;
 684         int ret;
 685
 686         /* decide whether we should attempt the mapping, and if so what sort of
 687          * mapping */
 688         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
 689                                     &capabilities);
 690         if (ret < 0)
 691                 return ret;
 692
 693         /* we've determined that we can make the mapping, now translate what we
 694          * now know into VMA flags */
 695         vm_flags = determine_vm_flags(file, prot, flags, capabilities);
 696
 697         /* we're going to need to record the mapping if it works */
 698         vml = kmalloc(sizeof(struct vm_list_struct), GFP_KERNEL);
 699         if (!vml)
 700                 goto error_getting_vml;
 701         memset(vml, 0, sizeof(*vml));
 702
 703         down_write(&nommu_vma_sem);
 704
 705         /* if we want to share, we need to check for VMAs created by other
 706          * mmap() calls that overlap with our proposed mapping
 707          * - we can only share with an exact match on most regular files
 708          * - shared mappings on character devices and memory backed files are
 709          *   permitted to overlap inexactly as far as we are concerned for in
 710          *   these cases, sharing is handled in the driver or filesystem rather
 711          *   than here
 712          */
 713         if (vm_flags & VM_MAYSHARE) {
 714                 unsigned long pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 715                 unsigned long vmpglen;
 716
 717                 for (rb = rb_first(&nommu_vma_tree); rb; rb = rb_next(rb)) {
 718                         vma = rb_entry(rb, struct vm_area_struct, vm_rb);
 719
 720                         if (!(vma->vm_flags & VM_MAYSHARE))
 721                                 continue;
 722
 723                         /* search for overlapping mappings on the same file */
 724                         if (vma->vm_file->f_dentry->d_inode != file->f_dentry->d_inode)
 725                                 continue;
 726
 727                         if (vma->vm_pgoff >= pgoff + pglen)
 728                                 continue;
 729
 730                         vmpglen = vma->vm_end - vma->vm_start + PAGE_SIZE - 1;
 731                         vmpglen >>= PAGE_SHIFT;
 732                         if (pgoff >= vma->vm_pgoff + vmpglen)
 733                                 continue;
 734
 735                         /* handle inexactly overlapping matches between mappings */
 736                         if (vma->vm_pgoff != pgoff || vmpglen != pglen) {
 737                                 if (!(capabilities & BDI_CAP_MAP_DIRECT))
 738                                         goto sharing_violation;
 739                                 continue;
 740                         }
 741
 742                         /* we've found a VMA we can share */
 743                         atomic_inc(&vma->vm_usage);
 744
 745                         vml->vma = vma;
 746                         result = (void *) vma->vm_start;
 747                         goto shared;
 748                 }
 749
 750                 vma = NULL;
 751
 752                 /* obtain the address at which to make a shared mapping
 753                  * - this is the hook for quasi-memory character devices to
 754                  *   tell us the location of a shared mapping
 755                  */
 756                 if (file && file->f_op->get_unmapped_area) {
 757                         addr = file->f_op->get_unmapped_area(file, addr, len,
 758                                                              pgoff, flags);
 759                         if (IS_ERR((void *) addr)) {
 760                                 ret = addr;
 761                                 if (ret != (unsigned long) -ENOSYS)
 762                                         goto error;
 763
 764                                 /* the driver refused to tell us where to site
 765                                  * the mapping so we'll have to attempt to copy
 766                                  * it */
 767                                 ret = (unsigned long) -ENODEV;
 768                                 if (!(capabilities & BDI_CAP_MAP_COPY))
 769                                         goto error;
 770
 771                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
 772                         }
 773                 }
 774         }
 775
 776         /* we're going to need a VMA struct as well */
 777         vma = kmalloc(sizeof(struct vm_area_struct), GFP_KERNEL);
 778         if (!vma)
 779                 goto error_getting_vma;
 780
 781         memset(vma, 0, sizeof(*vma));
 782         INIT_LIST_HEAD(&vma->anon_vma_node);
 783         atomic_set(&vma->vm_usage, 1);
 784         if (file)
 785                 get_file(file);
 786         vma->vm_file    = file;
 787         vma->vm_flags   = vm_flags;
 788         vma->vm_start   = addr;
 789         vma->vm_end     = addr + len;
 790         vma->vm_pgoff   = pgoff;
 791
 792         vml->vma = vma;
 793
 794         /* set up the mapping */
 795         if (file && vma->vm_flags & VM_SHARED)
 796                 ret = do_mmap_shared_file(vma, len);
 797         else
 798                 ret = do_mmap_private(vma, len);
 799         if (ret < 0)
 800                 goto error;
 801
 802         /* okay... we have a mapping; now we have to register it */
 803         result = (void *) vma->vm_start;
 804
 805         if (vma->vm_flags & VM_MAPPED_COPY) {
 806                 realalloc += kobjsize(result);
 807                 askedalloc += len;
 808         }
 809
 810         realalloc += kobjsize(vma);
 811         askedalloc += sizeof(*vma);
 812
 813         current->mm->total_vm += len >> PAGE_SHIFT;
 814
 815         add_nommu_vma(vma);
 816
 817  shared:
 818         realalloc += kobjsize(vml);
 819         askedalloc += sizeof(*vml);
 820
 821         vml->next = current->mm->context.vmlist;
 822         current->mm->context.vmlist = vml;
 823
 824         up_write(&nommu_vma_sem);
 825
 826         if (prot & PROT_EXEC)
 827                 flush_icache_range((unsigned long) result,
 828                                    (unsigned long) result + len);
 829
 830 #ifdef DEBUG
 831         printk("do_mmap:\n");
 832         show_process_blocks();
 833 #endif
 834
 835         return (unsigned long) result;
 836
 837  error:
 838         up_write(&nommu_vma_sem);
 839         kfree(vml);
 840         if (vma) {
 841                 fput(vma->vm_file);
 842                 kfree(vma);
 843         }
 844         return ret;
 845
 846  sharing_violation:
 847         up_write(&nommu_vma_sem);
 848         printk("Attempt to share mismatched mappings\n");
 849         kfree(vml);
 850         return -EINVAL;
 851
 852  error_getting_vma:
 853         up_write(&nommu_vma_sem);
 854         kfree(vml);
 855         printk("Allocation of vml for %lu byte allocation from process %d failed\n",
 856                len, current->pid);
 857         show_free_areas();
 858         return -ENOMEM;
 859
 860  error_getting_vml:
 861         printk("Allocation of vml for %lu byte allocation from process %d failed\n",
 862                len, current->pid);
 863         show_free_areas();
 864         return -ENOMEM;
 865 }
 866
 867 /*
 868  * handle mapping disposal for uClinux
 869  */
 870 static void put_vma(struct vm_area_struct *vma)
 871 {
 872         if (vma) {
 873                 down_write(&nommu_vma_sem);
 874
 875                 if (atomic_dec_and_test(&vma->vm_usage)) {
 876                         delete_nommu_vma(vma);
 877
 878                         if (vma->vm_ops && vma->vm_ops->close)
 879                                 vma->vm_ops->close(vma);
 880
 881                         /* IO memory and memory shared directly out of the pagecache from
 882                          * ramfs/tmpfs mustn't be released here */
 883                         if (vma->vm_flags & VM_MAPPED_COPY) {
 884                                 realalloc -= kobjsize((void *) vma->vm_start);
 885                                 askedalloc -= vma->vm_end - vma->vm_start;
 886                                 kfree((void *) vma->vm_start);
 887                         }
 888
 889                         realalloc -= kobjsize(vma);
 890                         askedalloc -= sizeof(*vma);
 891
 892                         if (vma->vm_file)
 893                                 fput(vma->vm_file);
 894                         kfree(vma);
 895                 }
 896
 897                 up_write(&nommu_vma_sem);
 898         }
 899 }
 900
 901 int do_munmap(struct mm_struct *mm, unsigned long addr, size_t len)
 902 {
 903         struct vm_list_struct *vml, **parent;
 904         unsigned long end = addr + len;
 905
 906 #ifdef DEBUG
 907         printk("do_munmap:\n");
 908 #endif
 909
 910         for (parent = &mm->context.vmlist; *parent; parent = &(*parent)->next)
 911                 if ((*parent)->vma->vm_start == addr &&
 912                     (*parent)->vma->vm_end == end)
 913                         goto found;
 914
 915         printk("munmap of non-mmaped memory by process %d (%s): %p\n",
 916                current->pid, current->comm, (void *) addr);
 917         return -EINVAL;
 918
 919  found:
 920         vml = *parent;
 921
 922         put_vma(vml->vma);
 923
 924         *parent = vml->next;
 925         realalloc -= kobjsize(vml);
 926         askedalloc -= sizeof(*vml);
 927         kfree(vml);
 928         mm->total_vm -= len >> PAGE_SHIFT;
 929
 930 #ifdef DEBUG
 931         show_process_blocks();
 932 #endif
 933
 934         return 0;
 935 }
 936
 937 /* Release all mmaps. */
 938 void exit_mmap(struct mm_struct * mm)
 939 {
 940         struct vm_list_struct *tmp;
 941
 942         if (mm) {
 943 #ifdef DEBUG
 944                 printk("Exit_mmap:\n");
 945 #endif
 946
 947                 mm->total_vm = 0;
 948
 949                 while ((tmp = mm->context.vmlist)) {
 950                         mm->context.vmlist = tmp->next;
 951                         put_vma(tmp->vma);
 952
 953                         realalloc -= kobjsize(tmp);
 954                         askedalloc -= sizeof(*tmp);
 955                         kfree(tmp);
 956                 }
 957
 958 #ifdef DEBUG
 959                 show_process_blocks();
 960 #endif
 961         }
 962 }
 963
 964 asmlinkage long sys_munmap(unsigned long addr, size_t len)
 965 {
 966         int ret;
 967         struct mm_struct *mm = current->mm;
 968
 969         down_write(&mm->mmap_sem);
 970         ret = do_munmap(mm, addr, len);
 971         up_write(&mm->mmap_sem);
 972         return ret;
 973 }
 974
 975 unsigned long do_brk(unsigned long addr, unsigned long len)
 976 {
 977         return -ENOMEM;
 978 }
 979
 980 /*
 981  * Expand (or shrink) an existing mapping, potentially moving it at the
 982  * same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
 983  *
 984  * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise
 985  * This option implies MREMAP_MAYMOVE.
 986  *
 987  * on uClinux, we only permit changing a mapping's size, and only as long as it stays within the
 988  * hole allocated by the kmalloc() call in do_mmap_pgoff() and the block is not shareable
 989  */
 990 unsigned long do_mremap(unsigned long addr,
 991                         unsigned long old_len, unsigned long new_len,
 992                         unsigned long flags, unsigned long new_addr)
 993 {
 994         struct vm_list_struct *vml = NULL;
 995
 996         /* insanity checks first */
 997         if (new_len == 0)
 998                 return (unsigned long) -EINVAL;
 999
1000         if (flags & MREMAP_FIXED && new_addr != addr)
1001                 return (unsigned long) -EINVAL;
1002
1003         for (vml = current->mm->context.vmlist; vml; vml = vml->next)
1004                 if (vml->vma->vm_start == addr)
1005                         goto found;
1006
1007         return (unsigned long) -EINVAL;
1008
1009  found:
1010         if (vml->vma->vm_end != vml->vma->vm_start + old_len)
1011                 return (unsigned long) -EFAULT;
1012
1013         if (vml->vma->vm_flags & VM_MAYSHARE)
1014                 return (unsigned long) -EPERM;
1015
1016         if (new_len > kobjsize((void *) addr))
1017                 return (unsigned long) -ENOMEM;
1018
1019         /* all checks complete - do it */
1020         vml->vma->vm_end = vml->vma->vm_start + new_len;
1021
1022         askedalloc -= old_len;
1023         askedalloc += new_len;
1024
1025         return vml->vma->vm_start;
1026 }
1027
1028 /*
1029  * Look up the first VMA which satisfies  addr < vm_end,  NULL if none
1030  */
1031 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
1032 {
1033         struct vm_list_struct *vml;
1034
1035         for (vml = mm->context.vmlist; vml; vml = vml->next)
1036                 if (addr >= vml->vma->vm_start && addr < vml->vma->vm_end)
1037                         return vml->vma;
1038
1039         return NULL;
1040 }
1041
1042 EXPORT_SYMBOL(find_vma);
1043
1044 struct page * follow_page(struct mm_struct *mm, unsigned long addr, int write)
1045 {
1046         return NULL;
1047 }
1048
1049 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
1050 {
1051         return NULL;
1052 }
1053
1054 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1055                 unsigned long to, unsigned long size, pgprot_t prot)
1056 {
1057         return -EPERM;
1058 }
1059
1060 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1061 {
1062 }
1063
1064 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1065         unsigned long len, unsigned long pgoff, unsigned long flags)
1066 {
1067         return -ENOMEM;
1068 }
1069
1070 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1071 {
1072 }
1073
1074 void update_mem_hiwater(struct task_struct *tsk)
1075 {
1076         unsigned long rss = get_mm_counter(tsk->mm, rss);
1077
1078         if (likely(tsk->mm)) {
1079                 if (tsk->mm->hiwater_rss < rss)
1080                         tsk->mm->hiwater_rss = rss;
1081                 if (tsk->mm->hiwater_vm < tsk->mm->total_vm)
1082                         tsk->mm->hiwater_vm = tsk->mm->total_vm;
1083         }
1084 }
1085
1086 void unmap_mapping_range(struct address_space *mapping,
1087                          loff_t const holebegin, loff_t const holelen,
1088                          int even_cows)
1089 {
1090 }
1091
1092 /*
1093  * Check that a process has enough memory to allocate a new virtual
1094  * mapping. 0 means there is enough memory for the allocation to
1095  * succeed and -ENOMEM implies there is not.
1096  *
1097  * We currently support three overcommit policies, which are set via the
1098  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1099  *
1100  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1101  * Additional code 2002 Jul 20 by Robert Love.
1102  *
1103  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1104  *
1105  * Note this is a helper function intended to be used by LSMs which
1106  * wish to use this logic.
1107  */
1108 int __vm_enough_memory(long pages, int cap_sys_admin)
1109 {
1110         unsigned long free, allowed;
1111
1112         vm_acct_memory(pages);
1113
1114         /*
1115          * Sometimes we want to use more memory than we have
1116          */
1117         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1118                 return 0;
1119
1120         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1121                 unsigned long n;
1122
1123                 free = get_page_cache_size();
1124                 free += nr_swap_pages;
1125
1126                 /*
1127                  * Any slabs which are created with the
1128                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1129                  * which are reclaimable, under pressure.  The dentry
1130                  * cache and most inode caches should fall into this
1131                  */
1132                 free += atomic_read(&slab_reclaim_pages);
1133
1134                 /*
1135                  * Leave the last 3% for root
1136                  */
1137                 if (!cap_sys_admin)
1138                         free -= free / 32;
1139
1140                 if (free > pages)
1141                         return 0;
1142
1143                 /*
1144                  * nr_free_pages() is very expensive on large systems,
1145                  * only call if we're about to fail.
1146                  */
1147                 n = nr_free_pages();
1148                 if (!cap_sys_admin)
1149                         n -= n / 32;
1150                 free += n;
1151
1152                 if (free > pages)
1153                         return 0;
1154                 vm_unacct_memory(pages);
1155                 return -ENOMEM;
1156         }
1157
1158         allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1159         /*
1160          * Leave the last 3% for root
1161          */
1162         if (!cap_sys_admin)
1163                 allowed -= allowed / 32;
1164         allowed += total_swap_pages;
1165
1166         /* Don't let a single process grow too big:
1167            leave 3% of the size of this process for other processes */
1168         allowed -= current->mm->total_vm / 32;
1169
1170         /*
1171          * cast `allowed' as a signed long because vm_committed_space
1172          * sometimes has a negative value
1173          */
1174         if (atomic_read(&vm_committed_space) < (long)allowed)
1175                 return 0;
1176
1177         vm_unacct_memory(pages);
1178
1179         return -ENOMEM;
1180 }
1181
1182 int in_gate_area_no_task(unsigned long addr)
1183 {
1184         return 0;
1185 }