git.oblomov.eu Git - 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         static struct vm_area_struct dummy_vma;
 133
 134         for (i = 0; i < len; i++) {
 135                 if (pages) {
 136                         pages[i] = virt_to_page(start);
 137                         if (pages[i])
 138                                 page_cache_get(pages[i]);
 139                 }
 140                 if (vmas)
 141                         vmas[i] = &dummy_vma;
 142                 start += PAGE_SIZE;
 143         }
 144         return(i);
 145 }
 146
 147 EXPORT_SYMBOL(get_user_pages);
 148
 149 DEFINE_RWLOCK(vmlist_lock);
 150 struct vm_struct *vmlist;
 151
 152 void vfree(void *addr)
 153 {
 154         kfree(addr);
 155 }
 156
 157 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
 158 {
 159         /*
 160          * kmalloc doesn't like __GFP_HIGHMEM for some reason
 161          */
 162         return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
 163 }
 164
 165 struct page * vmalloc_to_page(void *addr)
 166 {
 167         return virt_to_page(addr);
 168 }
 169
 170 unsigned long vmalloc_to_pfn(void *addr)
 171 {
 172         return page_to_pfn(virt_to_page(addr));
 173 }
 174
 175
 176 long vread(char *buf, char *addr, unsigned long count)
 177 {
 178         memcpy(buf, addr, count);
 179         return count;
 180 }
 181
 182 long vwrite(char *buf, char *addr, unsigned long count)
 183 {
 184         /* Don't allow overflow */
 185         if ((unsigned long) addr + count < count)
 186                 count = -(unsigned long) addr;
 187
 188         memcpy(addr, buf, count);
 189         return(count);
 190 }
 191
 192 /*
 193  *      vmalloc  -  allocate virtually continguos memory
 194  *
 195  *      @size:          allocation size
 196  *
 197  *      Allocate enough pages to cover @size from the page level
 198  *      allocator and map them into continguos kernel virtual space.
 199  *
 200  *      For tight cotrol over page level allocator and protection flags
 201  *      use __vmalloc() instead.
 202  */
 203 void *vmalloc(unsigned long size)
 204 {
 205        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
 206 }
 207 EXPORT_SYMBOL(vmalloc);
 208
 209 void *vmalloc_node(unsigned long size, int node)
 210 {
 211         return vmalloc(size);
 212 }
 213 EXPORT_SYMBOL(vmalloc_node);
 214
 215 /*
 216  *      vmalloc_32  -  allocate virtually continguos memory (32bit addressable)
 217  *
 218  *      @size:          allocation size
 219  *
 220  *      Allocate enough 32bit PA addressable pages to cover @size from the
 221  *      page level allocator and map them into continguos kernel virtual space.
 222  */
 223 void *vmalloc_32(unsigned long size)
 224 {
 225         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
 226 }
 227
 228 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
 229 {
 230         BUG();
 231         return NULL;
 232 }
 233
 234 void vunmap(void *addr)
 235 {
 236         BUG();
 237 }
 238
 239 /*
 240  *  sys_brk() for the most part doesn't need the global kernel
 241  *  lock, except when an application is doing something nasty
 242  *  like trying to un-brk an area that has already been mapped
 243  *  to a regular file.  in this case, the unmapping will need
 244  *  to invoke file system routines that need the global lock.
 245  */
 246 asmlinkage unsigned long sys_brk(unsigned long brk)
 247 {
 248         struct mm_struct *mm = current->mm;
 249
 250         if (brk < mm->start_brk || brk > mm->context.end_brk)
 251                 return mm->brk;
 252
 253         if (mm->brk == brk)
 254                 return mm->brk;
 255
 256         /*
 257          * Always allow shrinking brk
 258          */
 259         if (brk <= mm->brk) {
 260                 mm->brk = brk;
 261                 return brk;
 262         }
 263
 264         /*
 265          * Ok, looks good - let it rip.
 266          */
 267         return mm->brk = brk;
 268 }
 269
 270 #ifdef DEBUG
 271 static void show_process_blocks(void)
 272 {
 273         struct vm_list_struct *vml;
 274
 275         printk("Process blocks %d:", current->pid);
 276
 277         for (vml = &current->mm->context.vmlist; vml; vml = vml->next) {
 278                 printk(" %p: %p", vml, vml->vma);
 279                 if (vml->vma)
 280                         printk(" (%d @%lx #%d)",
 281                                kobjsize((void *) vml->vma->vm_start),
 282                                vml->vma->vm_start,
 283                                atomic_read(&vml->vma->vm_usage));
 284                 printk(vml->next ? " ->" : ".\n");
 285         }
 286 }
 287 #endif /* DEBUG */
 288
 289 static inline struct vm_area_struct *find_nommu_vma(unsigned long start)
 290 {
 291         struct vm_area_struct *vma;
 292         struct rb_node *n = nommu_vma_tree.rb_node;
 293
 294         while (n) {
 295                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
 296
 297                 if (start < vma->vm_start)
 298                         n = n->rb_left;
 299                 else if (start > vma->vm_start)
 300                         n = n->rb_right;
 301                 else
 302                         return vma;
 303         }
 304
 305         return NULL;
 306 }
 307
 308 static void add_nommu_vma(struct vm_area_struct *vma)
 309 {
 310         struct vm_area_struct *pvma;
 311         struct address_space *mapping;
 312         struct rb_node **p = &nommu_vma_tree.rb_node;
 313         struct rb_node *parent = NULL;
 314
 315         /* add the VMA to the mapping */
 316         if (vma->vm_file) {
 317                 mapping = vma->vm_file->f_mapping;
 318
 319                 flush_dcache_mmap_lock(mapping);
 320                 vma_prio_tree_insert(vma, &mapping->i_mmap);
 321                 flush_dcache_mmap_unlock(mapping);
 322         }
 323
 324         /* add the VMA to the master list */
 325         while (*p) {
 326                 parent = *p;
 327                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
 328
 329                 if (vma->vm_start < pvma->vm_start) {
 330                         p = &(*p)->rb_left;
 331                 }
 332                 else if (vma->vm_start > pvma->vm_start) {
 333                         p = &(*p)->rb_right;
 334                 }
 335                 else {
 336                         /* mappings are at the same address - this can only
 337                          * happen for shared-mem chardevs and shared file
 338                          * mappings backed by ramfs/tmpfs */
 339                         BUG_ON(!(pvma->vm_flags & VM_SHARED));
 340
 341                         if (vma < pvma)
 342                                 p = &(*p)->rb_left;
 343                         else if (vma > pvma)
 344                                 p = &(*p)->rb_right;
 345                         else
 346                                 BUG();
 347                 }
 348         }
 349
 350         rb_link_node(&vma->vm_rb, parent, p);
 351         rb_insert_color(&vma->vm_rb, &nommu_vma_tree);
 352 }
 353
 354 static void delete_nommu_vma(struct vm_area_struct *vma)
 355 {
 356         struct address_space *mapping;
 357
 358         /* remove the VMA from the mapping */
 359         if (vma->vm_file) {
 360                 mapping = vma->vm_file->f_mapping;
 361
 362                 flush_dcache_mmap_lock(mapping);
 363                 vma_prio_tree_remove(vma, &mapping->i_mmap);
 364                 flush_dcache_mmap_unlock(mapping);
 365         }
 366
 367         /* remove from the master list */
 368         rb_erase(&vma->vm_rb, &nommu_vma_tree);
 369 }
 370
 371 /*
 372  * determine whether a mapping should be permitted and, if so, what sort of
 373  * mapping we're capable of supporting
 374  */
 375 static int validate_mmap_request(struct file *file,
 376                                  unsigned long addr,
 377                                  unsigned long len,
 378                                  unsigned long prot,
 379                                  unsigned long flags,
 380                                  unsigned long pgoff,
 381                                  unsigned long *_capabilities)
 382 {
 383         unsigned long capabilities;
 384         unsigned long reqprot = prot;
 385         int ret;
 386
 387         /* do the simple checks first */
 388         if (flags & MAP_FIXED || addr) {
 389                 printk(KERN_DEBUG
 390                        "%d: Can't do fixed-address/overlay mmap of RAM\n",
 391                        current->pid);
 392                 return -EINVAL;
 393         }
 394
 395         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
 396             (flags & MAP_TYPE) != MAP_SHARED)
 397                 return -EINVAL;
 398
 399         if (PAGE_ALIGN(len) == 0)
 400                 return addr;
 401
 402         if (len > TASK_SIZE)
 403                 return -EINVAL;
 404
 405         /* offset overflow? */
 406         if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
 407                 return -EINVAL;
 408
 409         if (file) {
 410                 /* validate file mapping requests */
 411                 struct address_space *mapping;
 412
 413                 /* files must support mmap */
 414                 if (!file->f_op || !file->f_op->mmap)
 415                         return -ENODEV;
 416
 417                 /* work out if what we've got could possibly be shared
 418                  * - we support chardevs that provide their own "memory"
 419                  * - we support files/blockdevs that are memory backed
 420                  */
 421                 mapping = file->f_mapping;
 422                 if (!mapping)
 423                         mapping = file->f_dentry->d_inode->i_mapping;
 424
 425                 capabilities = 0;
 426                 if (mapping && mapping->backing_dev_info)
 427                         capabilities = mapping->backing_dev_info->capabilities;
 428
 429                 if (!capabilities) {
 430                         /* no explicit capabilities set, so assume some
 431                          * defaults */
 432                         switch (file->f_dentry->d_inode->i_mode & S_IFMT) {
 433                         case S_IFREG:
 434                         case S_IFBLK:
 435                                 capabilities = BDI_CAP_MAP_COPY;
 436                                 break;
 437
 438                         case S_IFCHR:
 439                                 capabilities =
 440                                         BDI_CAP_MAP_DIRECT |
 441                                         BDI_CAP_READ_MAP |
 442                                         BDI_CAP_WRITE_MAP;
 443                                 break;
 444
 445                         default:
 446                                 return -EINVAL;
 447                         }
 448                 }
 449
 450                 /* eliminate any capabilities that we can't support on this
 451                  * device */
 452                 if (!file->f_op->get_unmapped_area)
 453                         capabilities &= ~BDI_CAP_MAP_DIRECT;
 454                 if (!file->f_op->read)
 455                         capabilities &= ~BDI_CAP_MAP_COPY;
 456
 457                 if (flags & MAP_SHARED) {
 458                         /* do checks for writing, appending and locking */
 459                         if ((prot & PROT_WRITE) &&
 460                             !(file->f_mode & FMODE_WRITE))
 461                                 return -EACCES;
 462
 463                         if (IS_APPEND(file->f_dentry->d_inode) &&
 464                             (file->f_mode & FMODE_WRITE))
 465                                 return -EACCES;
 466
 467                         if (locks_verify_locked(file->f_dentry->d_inode))
 468                                 return -EAGAIN;
 469
 470                         if (!(capabilities & BDI_CAP_MAP_DIRECT))
 471                                 return -ENODEV;
 472
 473                         if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
 474                             ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
 475                             ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
 476                             ) {
 477                                 printk("MAP_SHARED not completely supported on !MMU\n");
 478                                 return -EINVAL;
 479                         }
 480
 481                         /* we mustn't privatise shared mappings */
 482                         capabilities &= ~BDI_CAP_MAP_COPY;
 483                 }
 484                 else {
 485                         /* we're going to read the file into private memory we
 486                          * allocate */
 487                         if (!(capabilities & BDI_CAP_MAP_COPY))
 488                                 return -ENODEV;
 489
 490                         /* we don't permit a private writable mapping to be
 491                          * shared with the backing device */
 492                         if (prot & PROT_WRITE)
 493                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
 494                 }
 495
 496                 /* handle executable mappings and implied executable
 497                  * mappings */
 498                 if (file->f_vfsmnt->mnt_flags & MNT_NOEXEC) {
 499                         if (prot & PROT_EXEC)
 500                                 return -EPERM;
 501                 }
 502                 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
 503                         /* handle implication of PROT_EXEC by PROT_READ */
 504                         if (current->personality & READ_IMPLIES_EXEC) {
 505                                 if (capabilities & BDI_CAP_EXEC_MAP)
 506                                         prot |= PROT_EXEC;
 507                         }
 508                 }
 509                 else if ((prot & PROT_READ) &&
 510                          (prot & PROT_EXEC) &&
 511                          !(capabilities & BDI_CAP_EXEC_MAP)
 512                          ) {
 513                         /* backing file is not executable, try to copy */
 514                         capabilities &= ~BDI_CAP_MAP_DIRECT;
 515                 }
 516         }
 517         else {
 518                 /* anonymous mappings are always memory backed and can be
 519                  * privately mapped
 520                  */
 521                 capabilities = BDI_CAP_MAP_COPY;
 522
 523                 /* handle PROT_EXEC implication by PROT_READ */
 524                 if ((prot & PROT_READ) &&
 525                     (current->personality & READ_IMPLIES_EXEC))
 526                         prot |= PROT_EXEC;
 527         }
 528
 529         /* allow the security API to have its say */
 530         ret = security_file_mmap(file, reqprot, prot, flags);
 531         if (ret < 0)
 532                 return ret;
 533
 534         /* looks okay */
 535         *_capabilities = capabilities;
 536         return 0;
 537 }
 538
 539 /*
 540  * we've determined that we can make the mapping, now translate what we
 541  * now know into VMA flags
 542  */
 543 static unsigned long determine_vm_flags(struct file *file,
 544                                         unsigned long prot,
 545                                         unsigned long flags,
 546                                         unsigned long capabilities)
 547 {
 548         unsigned long vm_flags;
 549
 550         vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
 551         vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 552         /* vm_flags |= mm->def_flags; */
 553
 554         if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
 555                 /* attempt to share read-only copies of mapped file chunks */
 556                 if (file && !(prot & PROT_WRITE))
 557                         vm_flags |= VM_MAYSHARE;
 558         }
 559         else {
 560                 /* overlay a shareable mapping on the backing device or inode
 561                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
 562                  * romfs/cramfs */
 563                 if (flags & MAP_SHARED)
 564                         vm_flags |= VM_MAYSHARE | VM_SHARED;
 565                 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
 566                         vm_flags |= VM_MAYSHARE;
 567         }
 568
 569         /* refuse to let anyone share private mappings with this process if
 570          * it's being traced - otherwise breakpoints set in it may interfere
 571          * with another untraced process
 572          */
 573         if ((flags & MAP_PRIVATE) && (current->ptrace & PT_PTRACED))
 574                 vm_flags &= ~VM_MAYSHARE;
 575
 576         return vm_flags;
 577 }
 578
 579 /*
 580  * set up a shared mapping on a file
 581  */
 582 static int do_mmap_shared_file(struct vm_area_struct *vma, unsigned long len)
 583 {
 584         int ret;
 585
 586         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
 587         if (ret != -ENOSYS)
 588                 return ret;
 589
 590         /* getting an ENOSYS error indicates that direct mmap isn't
 591          * possible (as opposed to tried but failed) so we'll fall
 592          * through to making a private copy of the data and mapping
 593          * that if we can */
 594         return -ENODEV;
 595 }
 596
 597 /*
 598  * set up a private mapping or an anonymous shared mapping
 599  */
 600 static int do_mmap_private(struct vm_area_struct *vma, unsigned long len)
 601 {
 602         void *base;
 603         int ret;
 604
 605         /* invoke the file's mapping function so that it can keep track of
 606          * shared mappings on devices or memory
 607          * - VM_MAYSHARE will be set if it may attempt to share
 608          */
 609         if (vma->vm_file) {
 610                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
 611                 if (ret != -ENOSYS) {
 612                         /* shouldn't return success if we're not sharing */
 613                         BUG_ON(ret == 0 && !(vma->vm_flags & VM_MAYSHARE));
 614                         return ret; /* success or a real error */
 615                 }
 616
 617                 /* getting an ENOSYS error indicates that direct mmap isn't
 618                  * possible (as opposed to tried but failed) so we'll try to
 619                  * make a private copy of the data and map that instead */
 620         }
 621
 622         /* allocate some memory to hold the mapping
 623          * - note that this may not return a page-aligned address if the object
 624          *   we're allocating is smaller than a page
 625          */
 626         base = kmalloc(len, GFP_KERNEL|__GFP_COMP);
 627         if (!base)
 628                 goto enomem;
 629
 630         vma->vm_start = (unsigned long) base;
 631         vma->vm_end = vma->vm_start + len;
 632         vma->vm_flags |= VM_MAPPED_COPY;
 633
 634 #ifdef WARN_ON_SLACK
 635         if (len + WARN_ON_SLACK <= kobjsize(result))
 636                 printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n",
 637                        len, current->pid, kobjsize(result) - len);
 638 #endif
 639
 640         if (vma->vm_file) {
 641                 /* read the contents of a file into the copy */
 642                 mm_segment_t old_fs;
 643                 loff_t fpos;
 644
 645                 fpos = vma->vm_pgoff;
 646                 fpos <<= PAGE_SHIFT;
 647
 648                 old_fs = get_fs();
 649                 set_fs(KERNEL_DS);
 650                 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
 651                 set_fs(old_fs);
 652
 653                 if (ret < 0)
 654                         goto error_free;
 655
 656                 /* clear the last little bit */
 657                 if (ret < len)
 658                         memset(base + ret, 0, len - ret);
 659
 660         } else {
 661                 /* if it's an anonymous mapping, then just clear it */
 662                 memset(base, 0, len);
 663         }
 664
 665         return 0;
 666
 667 error_free:
 668         kfree(base);
 669         vma->vm_start = 0;
 670         return ret;
 671
 672 enomem:
 673         printk("Allocation of length %lu from process %d failed\n",
 674                len, current->pid);
 675         show_free_areas();
 676         return -ENOMEM;
 677 }
 678
 679 /*
 680  * handle mapping creation for uClinux
 681  */
 682 unsigned long do_mmap_pgoff(struct file *file,
 683                             unsigned long addr,
 684                             unsigned long len,
 685                             unsigned long prot,
 686                             unsigned long flags,
 687                             unsigned long pgoff)
 688 {
 689         struct vm_list_struct *vml = NULL;
 690         struct vm_area_struct *vma = NULL;
 691         struct rb_node *rb;
 692         unsigned long capabilities, vm_flags;
 693         void *result;
 694         int ret;
 695
 696         /* decide whether we should attempt the mapping, and if so what sort of
 697          * mapping */
 698         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
 699                                     &capabilities);
 700         if (ret < 0)
 701                 return ret;
 702
 703         /* we've determined that we can make the mapping, now translate what we
 704          * now know into VMA flags */
 705         vm_flags = determine_vm_flags(file, prot, flags, capabilities);
 706
 707         /* we're going to need to record the mapping if it works */
 708         vml = kmalloc(sizeof(struct vm_list_struct), GFP_KERNEL);
 709         if (!vml)
 710                 goto error_getting_vml;
 711         memset(vml, 0, sizeof(*vml));
 712
 713         down_write(&nommu_vma_sem);
 714
 715         /* if we want to share, we need to check for VMAs created by other
 716          * mmap() calls that overlap with our proposed mapping
 717          * - we can only share with an exact match on most regular files
 718          * - shared mappings on character devices and memory backed files are
 719          *   permitted to overlap inexactly as far as we are concerned for in
 720          *   these cases, sharing is handled in the driver or filesystem rather
 721          *   than here
 722          */
 723         if (vm_flags & VM_MAYSHARE) {
 724                 unsigned long pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 725                 unsigned long vmpglen;
 726
 727                 for (rb = rb_first(&nommu_vma_tree); rb; rb = rb_next(rb)) {
 728                         vma = rb_entry(rb, struct vm_area_struct, vm_rb);
 729
 730                         if (!(vma->vm_flags & VM_MAYSHARE))
 731                                 continue;
 732
 733                         /* search for overlapping mappings on the same file */
 734                         if (vma->vm_file->f_dentry->d_inode != file->f_dentry->d_inode)
 735                                 continue;
 736
 737                         if (vma->vm_pgoff >= pgoff + pglen)
 738                                 continue;
 739
 740                         vmpglen = vma->vm_end - vma->vm_start + PAGE_SIZE - 1;
 741                         vmpglen >>= PAGE_SHIFT;
 742                         if (pgoff >= vma->vm_pgoff + vmpglen)
 743                                 continue;
 744
 745                         /* handle inexactly overlapping matches between mappings */
 746                         if (vma->vm_pgoff != pgoff || vmpglen != pglen) {
 747                                 if (!(capabilities & BDI_CAP_MAP_DIRECT))
 748                                         goto sharing_violation;
 749                                 continue;
 750                         }
 751
 752                         /* we've found a VMA we can share */
 753                         atomic_inc(&vma->vm_usage);
 754
 755                         vml->vma = vma;
 756                         result = (void *) vma->vm_start;
 757                         goto shared;
 758                 }
 759
 760                 vma = NULL;
 761
 762                 /* obtain the address at which to make a shared mapping
 763                  * - this is the hook for quasi-memory character devices to
 764                  *   tell us the location of a shared mapping
 765                  */
 766                 if (file && file->f_op->get_unmapped_area) {
 767                         addr = file->f_op->get_unmapped_area(file, addr, len,
 768                                                              pgoff, flags);
 769                         if (IS_ERR((void *) addr)) {
 770                                 ret = addr;
 771                                 if (ret != (unsigned long) -ENOSYS)
 772                                         goto error;
 773
 774                                 /* the driver refused to tell us where to site
 775                                  * the mapping so we'll have to attempt to copy
 776                                  * it */
 777                                 ret = (unsigned long) -ENODEV;
 778                                 if (!(capabilities & BDI_CAP_MAP_COPY))
 779                                         goto error;
 780
 781                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
 782                         }
 783                 }
 784         }
 785
 786         /* we're going to need a VMA struct as well */
 787         vma = kmalloc(sizeof(struct vm_area_struct), GFP_KERNEL);
 788         if (!vma)
 789                 goto error_getting_vma;
 790
 791         memset(vma, 0, sizeof(*vma));
 792         INIT_LIST_HEAD(&vma->anon_vma_node);
 793         atomic_set(&vma->vm_usage, 1);
 794         if (file)
 795                 get_file(file);
 796         vma->vm_file    = file;
 797         vma->vm_flags   = vm_flags;
 798         vma->vm_start   = addr;
 799         vma->vm_end     = addr + len;
 800         vma->vm_pgoff   = pgoff;
 801
 802         vml->vma = vma;
 803
 804         /* set up the mapping */
 805         if (file && vma->vm_flags & VM_SHARED)
 806                 ret = do_mmap_shared_file(vma, len);
 807         else
 808                 ret = do_mmap_private(vma, len);
 809         if (ret < 0)
 810                 goto error;
 811
 812         /* okay... we have a mapping; now we have to register it */
 813         result = (void *) vma->vm_start;
 814
 815         if (vma->vm_flags & VM_MAPPED_COPY) {
 816                 realalloc += kobjsize(result);
 817                 askedalloc += len;
 818         }
 819
 820         realalloc += kobjsize(vma);
 821         askedalloc += sizeof(*vma);
 822
 823         current->mm->total_vm += len >> PAGE_SHIFT;
 824
 825         add_nommu_vma(vma);
 826
 827  shared:
 828         realalloc += kobjsize(vml);
 829         askedalloc += sizeof(*vml);
 830
 831         vml->next = current->mm->context.vmlist;
 832         current->mm->context.vmlist = vml;
 833
 834         up_write(&nommu_vma_sem);
 835
 836         if (prot & PROT_EXEC)
 837                 flush_icache_range((unsigned long) result,
 838                                    (unsigned long) result + len);
 839
 840 #ifdef DEBUG
 841         printk("do_mmap:\n");
 842         show_process_blocks();
 843 #endif
 844
 845         return (unsigned long) result;
 846
 847  error:
 848         up_write(&nommu_vma_sem);
 849         kfree(vml);
 850         if (vma) {
 851                 fput(vma->vm_file);
 852                 kfree(vma);
 853         }
 854         return ret;
 855
 856  sharing_violation:
 857         up_write(&nommu_vma_sem);
 858         printk("Attempt to share mismatched mappings\n");
 859         kfree(vml);
 860         return -EINVAL;
 861
 862  error_getting_vma:
 863         up_write(&nommu_vma_sem);
 864         kfree(vml);
 865         printk("Allocation of vma for %lu byte allocation from process %d failed\n",
 866                len, current->pid);
 867         show_free_areas();
 868         return -ENOMEM;
 869
 870  error_getting_vml:
 871         printk("Allocation of vml for %lu byte allocation from process %d failed\n",
 872                len, current->pid);
 873         show_free_areas();
 874         return -ENOMEM;
 875 }
 876
 877 /*
 878  * handle mapping disposal for uClinux
 879  */
 880 static void put_vma(struct vm_area_struct *vma)
 881 {
 882         if (vma) {
 883                 down_write(&nommu_vma_sem);
 884
 885                 if (atomic_dec_and_test(&vma->vm_usage)) {
 886                         delete_nommu_vma(vma);
 887
 888                         if (vma->vm_ops && vma->vm_ops->close)
 889                                 vma->vm_ops->close(vma);
 890
 891                         /* IO memory and memory shared directly out of the pagecache from
 892                          * ramfs/tmpfs mustn't be released here */
 893                         if (vma->vm_flags & VM_MAPPED_COPY) {
 894                                 realalloc -= kobjsize((void *) vma->vm_start);
 895                                 askedalloc -= vma->vm_end - vma->vm_start;
 896                                 kfree((void *) vma->vm_start);
 897                         }
 898
 899                         realalloc -= kobjsize(vma);
 900                         askedalloc -= sizeof(*vma);
 901
 902                         if (vma->vm_file)
 903                                 fput(vma->vm_file);
 904                         kfree(vma);
 905                 }
 906
 907                 up_write(&nommu_vma_sem);
 908         }
 909 }
 910
 911 int do_munmap(struct mm_struct *mm, unsigned long addr, size_t len)
 912 {
 913         struct vm_list_struct *vml, **parent;
 914         unsigned long end = addr + len;
 915
 916 #ifdef DEBUG
 917         printk("do_munmap:\n");
 918 #endif
 919
 920         for (parent = &mm->context.vmlist; *parent; parent = &(*parent)->next)
 921                 if ((*parent)->vma->vm_start == addr &&
 922                     ((len == 0) || ((*parent)->vma->vm_end == end)))
 923                         goto found;
 924
 925         printk("munmap of non-mmaped memory by process %d (%s): %p\n",
 926                current->pid, current->comm, (void *) addr);
 927         return -EINVAL;
 928
 929  found:
 930         vml = *parent;
 931
 932         put_vma(vml->vma);
 933
 934         *parent = vml->next;
 935         realalloc -= kobjsize(vml);
 936         askedalloc -= sizeof(*vml);
 937         kfree(vml);
 938
 939         update_hiwater_vm(mm);
 940         mm->total_vm -= len >> PAGE_SHIFT;
 941
 942 #ifdef DEBUG
 943         show_process_blocks();
 944 #endif
 945
 946         return 0;
 947 }
 948
 949 /* Release all mmaps. */
 950 void exit_mmap(struct mm_struct * mm)
 951 {
 952         struct vm_list_struct *tmp;
 953
 954         if (mm) {
 955 #ifdef DEBUG
 956                 printk("Exit_mmap:\n");
 957 #endif
 958
 959                 mm->total_vm = 0;
 960
 961                 while ((tmp = mm->context.vmlist)) {
 962                         mm->context.vmlist = tmp->next;
 963                         put_vma(tmp->vma);
 964
 965                         realalloc -= kobjsize(tmp);
 966                         askedalloc -= sizeof(*tmp);
 967                         kfree(tmp);
 968                 }
 969
 970 #ifdef DEBUG
 971                 show_process_blocks();
 972 #endif
 973         }
 974 }
 975
 976 asmlinkage long sys_munmap(unsigned long addr, size_t len)
 977 {
 978         int ret;
 979         struct mm_struct *mm = current->mm;
 980
 981         down_write(&mm->mmap_sem);
 982         ret = do_munmap(mm, addr, len);
 983         up_write(&mm->mmap_sem);
 984         return ret;
 985 }
 986
 987 unsigned long do_brk(unsigned long addr, unsigned long len)
 988 {
 989         return -ENOMEM;
 990 }
 991
 992 /*
 993  * Expand (or shrink) an existing mapping, potentially moving it at the
 994  * same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
 995  *
 996  * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise
 997  * This option implies MREMAP_MAYMOVE.
 998  *
 999  * on uClinux, we only permit changing a mapping's size, and only as long as it stays within the
1000  * hole allocated by the kmalloc() call in do_mmap_pgoff() and the block is not shareable
1001  */
1002 unsigned long do_mremap(unsigned long addr,
1003                         unsigned long old_len, unsigned long new_len,
1004                         unsigned long flags, unsigned long new_addr)
1005 {
1006         struct vm_list_struct *vml = NULL;
1007
1008         /* insanity checks first */
1009         if (new_len == 0)
1010                 return (unsigned long) -EINVAL;
1011
1012         if (flags & MREMAP_FIXED && new_addr != addr)
1013                 return (unsigned long) -EINVAL;
1014
1015         for (vml = current->mm->context.vmlist; vml; vml = vml->next)
1016                 if (vml->vma->vm_start == addr)
1017                         goto found;
1018
1019         return (unsigned long) -EINVAL;
1020
1021  found:
1022         if (vml->vma->vm_end != vml->vma->vm_start + old_len)
1023                 return (unsigned long) -EFAULT;
1024
1025         if (vml->vma->vm_flags & VM_MAYSHARE)
1026                 return (unsigned long) -EPERM;
1027
1028         if (new_len > kobjsize((void *) addr))
1029                 return (unsigned long) -ENOMEM;
1030
1031         /* all checks complete - do it */
1032         vml->vma->vm_end = vml->vma->vm_start + new_len;
1033
1034         askedalloc -= old_len;
1035         askedalloc += new_len;
1036
1037         return vml->vma->vm_start;
1038 }
1039
1040 /*
1041  * Look up the first VMA which satisfies  addr < vm_end,  NULL if none
1042  */
1043 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
1044 {
1045         struct vm_list_struct *vml;
1046
1047         for (vml = mm->context.vmlist; vml; vml = vml->next)
1048                 if (addr >= vml->vma->vm_start && addr < vml->vma->vm_end)
1049                         return vml->vma;
1050
1051         return NULL;
1052 }
1053
1054 EXPORT_SYMBOL(find_vma);
1055
1056 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1057                         unsigned int foll_flags)
1058 {
1059         return NULL;
1060 }
1061
1062 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
1063 {
1064         return NULL;
1065 }
1066
1067 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1068                 unsigned long to, unsigned long size, pgprot_t prot)
1069 {
1070         vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1071         return 0;
1072 }
1073
1074 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1075 {
1076 }
1077
1078 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1079         unsigned long len, unsigned long pgoff, unsigned long flags)
1080 {
1081         return -ENOMEM;
1082 }
1083
1084 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1085 {
1086 }
1087
1088 void unmap_mapping_range(struct address_space *mapping,
1089                          loff_t const holebegin, loff_t const holelen,
1090                          int even_cows)
1091 {
1092 }
1093
1094 /*
1095  * Check that a process has enough memory to allocate a new virtual
1096  * mapping. 0 means there is enough memory for the allocation to
1097  * succeed and -ENOMEM implies there is not.
1098  *
1099  * We currently support three overcommit policies, which are set via the
1100  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1101  *
1102  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1103  * Additional code 2002 Jul 20 by Robert Love.
1104  *
1105  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1106  *
1107  * Note this is a helper function intended to be used by LSMs which
1108  * wish to use this logic.
1109  */
1110 int __vm_enough_memory(long pages, int cap_sys_admin)
1111 {
1112         unsigned long free, allowed;
1113
1114         vm_acct_memory(pages);
1115
1116         /*
1117          * Sometimes we want to use more memory than we have
1118          */
1119         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1120                 return 0;
1121
1122         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1123                 unsigned long n;
1124
1125                 free = get_page_cache_size();
1126                 free += nr_swap_pages;
1127
1128                 /*
1129                  * Any slabs which are created with the
1130                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1131                  * which are reclaimable, under pressure.  The dentry
1132                  * cache and most inode caches should fall into this
1133                  */
1134                 free += atomic_read(&slab_reclaim_pages);
1135
1136                 /*
1137                  * Leave the last 3% for root
1138                  */
1139                 if (!cap_sys_admin)
1140                         free -= free / 32;
1141
1142                 if (free > pages)
1143                         return 0;
1144
1145                 /*
1146                  * nr_free_pages() is very expensive on large systems,
1147                  * only call if we're about to fail.
1148                  */
1149                 n = nr_free_pages();
1150
1151                 /*
1152                  * Leave reserved pages. The pages are not for anonymous pages.
1153                  */
1154                 if (n <= totalreserve_pages)
1155                         goto error;
1156                 else
1157                         n -= totalreserve_pages;
1158
1159                 /*
1160                  * Leave the last 3% for root
1161                  */
1162                 if (!cap_sys_admin)
1163                         n -= n / 32;
1164                 free += n;
1165
1166                 if (free > pages)
1167                         return 0;
1168
1169                 goto error;
1170         }
1171
1172         allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1173         /*
1174          * Leave the last 3% for root
1175          */
1176         if (!cap_sys_admin)
1177                 allowed -= allowed / 32;
1178         allowed += total_swap_pages;
1179
1180         /* Don't let a single process grow too big:
1181            leave 3% of the size of this process for other processes */
1182         allowed -= current->mm->total_vm / 32;
1183
1184         /*
1185          * cast `allowed' as a signed long because vm_committed_space
1186          * sometimes has a negative value
1187          */
1188         if (atomic_read(&vm_committed_space) < (long)allowed)
1189                 return 0;
1190 error:
1191         vm_unacct_memory(pages);
1192
1193         return -ENOMEM;
1194 }
1195
1196 int in_gate_area_no_task(unsigned long addr)
1197 {
1198         return 0;
1199 }
1200
1201 struct page *filemap_nopage(struct vm_area_struct *area,
1202                         unsigned long address, int *type)
1203 {
1204         BUG();
1205         return NULL;
1206 }