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