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(num_physpages);
  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 /*
  57  * Handle all mappings that got truncated by a "truncate()"
  58  * system call.
  59  *
  60  * NOTE! We have to be ready to update the memory sharing
  61  * between the file and the memory map for a potential last
  62  * incomplete page.  Ugly, but necessary.
  63  */
  64 int vmtruncate(struct inode *inode, loff_t offset)
  65 {
  66         struct address_space *mapping = inode->i_mapping;
  67         unsigned long limit;
  68
  69         if (inode->i_size < offset)
  70                 goto do_expand;
  71         i_size_write(inode, offset);
  72
  73         truncate_inode_pages(mapping, offset);
  74         goto out_truncate;
  75
  76 do_expand:
  77         limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
  78         if (limit != RLIM_INFINITY && offset > limit)
  79                 goto out_sig;
  80         if (offset > inode->i_sb->s_maxbytes)
  81                 goto out;
  82         i_size_write(inode, offset);
  83
  84 out_truncate:
  85         if (inode->i_op && inode->i_op->truncate)
  86                 inode->i_op->truncate(inode);
  87         return 0;
  88 out_sig:
  89         send_sig(SIGXFSZ, current, 0);
  90 out:
  91         return -EFBIG;
  92 }
  93
  94 EXPORT_SYMBOL(vmtruncate);
  95
  96 /*
  97  * Return the total memory allocated for this pointer, not
  98  * just what the caller asked for.
  99  *
 100  * Doesn't have to be accurate, i.e. may have races.
 101  */
 102 unsigned int kobjsize(const void *objp)
 103 {
 104         struct page *page;
 105
 106         if (!objp || !((page = virt_to_page(objp))))
 107                 return 0;
 108
 109         if (PageSlab(page))
 110                 return ksize(objp);
 111
 112         BUG_ON(page->index < 0);
 113         BUG_ON(page->index >= MAX_ORDER);
 114
 115         return (PAGE_SIZE << page->index);
 116 }
 117
 118 /*
 119  * get a list of pages in an address range belonging to the specified process
 120  * and indicate the VMA that covers each page
 121  * - this is potentially dodgy as we may end incrementing the page count of a
 122  *   slab page or a secondary page from a compound page
 123  * - don't permit access to VMAs that don't support it, such as I/O mappings
 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         struct vm_area_struct *vma;
 130         unsigned long vm_flags;
 131         int i;
 132
 133         /* calculate required read or write permissions.
 134          * - if 'force' is set, we only require the "MAY" flags.
 135          */
 136         vm_flags  = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
 137         vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
 138
 139         for (i = 0; i < len; i++) {
 140                 vma = find_vma(mm, start);
 141                 if (!vma)
 142                         goto finish_or_fault;
 143
 144                 /* protect what we can, including chardevs */
 145                 if (vma->vm_flags & (VM_IO | VM_PFNMAP) ||
 146                     !(vm_flags & vma->vm_flags))
 147                         goto finish_or_fault;
 148
 149                 if (pages) {
 150                         pages[i] = virt_to_page(start);
 151                         if (pages[i])
 152                                 page_cache_get(pages[i]);
 153                 }
 154                 if (vmas)
 155                         vmas[i] = vma;
 156                 start += PAGE_SIZE;
 157         }
 158
 159         return i;
 160
 161 finish_or_fault:
 162         return i ? : -EFAULT;
 163 }
 164 EXPORT_SYMBOL(get_user_pages);
 165
 166 DEFINE_RWLOCK(vmlist_lock);
 167 struct vm_struct *vmlist;
 168
 169 void vfree(void *addr)
 170 {
 171         kfree(addr);
 172 }
 173 EXPORT_SYMBOL(vfree);
 174
 175 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
 176 {
 177         /*
 178          *  You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
 179          * returns only a logical address.
 180          */
 181         return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
 182 }
 183 EXPORT_SYMBOL(__vmalloc);
 184
 185 struct page * vmalloc_to_page(void *addr)
 186 {
 187         return virt_to_page(addr);
 188 }
 189 EXPORT_SYMBOL(vmalloc_to_page);
 190
 191 unsigned long vmalloc_to_pfn(void *addr)
 192 {
 193         return page_to_pfn(virt_to_page(addr));
 194 }
 195 EXPORT_SYMBOL(vmalloc_to_pfn);
 196
 197 long vread(char *buf, char *addr, unsigned long count)
 198 {
 199         memcpy(buf, addr, count);
 200         return count;
 201 }
 202
 203 long vwrite(char *buf, char *addr, unsigned long count)
 204 {
 205         /* Don't allow overflow */
 206         if ((unsigned long) addr + count < count)
 207                 count = -(unsigned long) addr;
 208
 209         memcpy(addr, buf, count);
 210         return(count);
 211 }
 212
 213 /*
 214  *      vmalloc  -  allocate virtually continguos memory
 215  *
 216  *      @size:          allocation size
 217  *
 218  *      Allocate enough pages to cover @size from the page level
 219  *      allocator and map them into continguos kernel virtual space.
 220  *
 221  *      For tight control over page level allocator and protection flags
 222  *      use __vmalloc() instead.
 223  */
 224 void *vmalloc(unsigned long size)
 225 {
 226        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
 227 }
 228 EXPORT_SYMBOL(vmalloc);
 229
 230 void *vmalloc_node(unsigned long size, int node)
 231 {
 232         return vmalloc(size);
 233 }
 234 EXPORT_SYMBOL(vmalloc_node);
 235
 236 /**
 237  * vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
 238  *      @size:          allocation size
 239  *
 240  *      Allocate enough 32bit PA addressable pages to cover @size from the
 241  *      page level allocator and map them into continguos kernel virtual space.
 242  */
 243 void *vmalloc_32(unsigned long size)
 244 {
 245         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
 246 }
 247 EXPORT_SYMBOL(vmalloc_32);
 248
 249 /**
 250  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
 251  *      @size:          allocation size
 252  *
 253  * The resulting memory area is 32bit addressable and zeroed so it can be
 254  * mapped to userspace without leaking data.
 255  */
 256 void *vmalloc_32_user(unsigned long size)
 257 {
 258         return __vmalloc(size, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
 259 }
 260 EXPORT_SYMBOL(vmalloc_32_user);
 261
 262 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
 263 {
 264         BUG();
 265         return NULL;
 266 }
 267 EXPORT_SYMBOL(vmap);
 268
 269 void vunmap(void *addr)
 270 {
 271         BUG();
 272 }
 273 EXPORT_SYMBOL(vunmap);
 274
 275 /*
 276  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
 277  * have one.
 278  */
 279 void  __attribute__((weak)) vmalloc_sync_all(void)
 280 {
 281 }
 282
 283 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
 284                    struct page *page)
 285 {
 286         return -EINVAL;
 287 }
 288 EXPORT_SYMBOL(vm_insert_page);
 289
 290 /*
 291  *  sys_brk() for the most part doesn't need the global kernel
 292  *  lock, except when an application is doing something nasty
 293  *  like trying to un-brk an area that has already been mapped
 294  *  to a regular file.  in this case, the unmapping will need
 295  *  to invoke file system routines that need the global lock.
 296  */
 297 asmlinkage unsigned long sys_brk(unsigned long brk)
 298 {
 299         struct mm_struct *mm = current->mm;
 300
 301         if (brk < mm->start_brk || brk > mm->context.end_brk)
 302                 return mm->brk;
 303
 304         if (mm->brk == brk)
 305                 return mm->brk;
 306
 307         /*
 308          * Always allow shrinking brk
 309          */
 310         if (brk <= mm->brk) {
 311                 mm->brk = brk;
 312                 return brk;
 313         }
 314
 315         /*
 316          * Ok, looks good - let it rip.
 317          */
 318         return mm->brk = brk;
 319 }
 320
 321 #ifdef DEBUG
 322 static void show_process_blocks(void)
 323 {
 324         struct vm_list_struct *vml;
 325
 326         printk("Process blocks %d:", current->pid);
 327
 328         for (vml = &current->mm->context.vmlist; vml; vml = vml->next) {
 329                 printk(" %p: %p", vml, vml->vma);
 330                 if (vml->vma)
 331                         printk(" (%d @%lx #%d)",
 332                                kobjsize((void *) vml->vma->vm_start),
 333                                vml->vma->vm_start,
 334                                atomic_read(&vml->vma->vm_usage));
 335                 printk(vml->next ? " ->" : ".\n");
 336         }
 337 }
 338 #endif /* DEBUG */
 339
 340 /*
 341  * add a VMA into a process's mm_struct in the appropriate place in the list
 342  * - should be called with mm->mmap_sem held writelocked
 343  */
 344 static void add_vma_to_mm(struct mm_struct *mm, struct vm_list_struct *vml)
 345 {
 346         struct vm_list_struct **ppv;
 347
 348         for (ppv = &current->mm->context.vmlist; *ppv; ppv = &(*ppv)->next)
 349                 if ((*ppv)->vma->vm_start > vml->vma->vm_start)
 350                         break;
 351
 352         vml->next = *ppv;
 353         *ppv = vml;
 354 }
 355
 356 /*
 357  * look up the first VMA in which addr resides, NULL if none
 358  * - should be called with mm->mmap_sem at least held readlocked
 359  */
 360 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
 361 {
 362         struct vm_list_struct *loop, *vml;
 363
 364         /* search the vm_start ordered list */
 365         vml = NULL;
 366         for (loop = mm->context.vmlist; loop; loop = loop->next) {
 367                 if (loop->vma->vm_start > addr)
 368                         break;
 369                 vml = loop;
 370         }
 371
 372         if (vml && vml->vma->vm_end > addr)
 373                 return vml->vma;
 374
 375         return NULL;
 376 }
 377 EXPORT_SYMBOL(find_vma);
 378
 379 /*
 380  * find a VMA
 381  * - we don't extend stack VMAs under NOMMU conditions
 382  */
 383 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
 384 {
 385         return find_vma(mm, addr);
 386 }
 387
 388 int expand_stack(struct vm_area_struct *vma, unsigned long address)
 389 {
 390         return -ENOMEM;
 391 }
 392
 393 /*
 394  * look up the first VMA exactly that exactly matches addr
 395  * - should be called with mm->mmap_sem at least held readlocked
 396  */
 397 static inline struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
 398                                                     unsigned long addr)
 399 {
 400         struct vm_list_struct *vml;
 401
 402         /* search the vm_start ordered list */
 403         for (vml = mm->context.vmlist; vml; vml = vml->next) {
 404                 if (vml->vma->vm_start == addr)
 405                         return vml->vma;
 406                 if (vml->vma->vm_start > addr)
 407                         break;
 408         }
 409
 410         return NULL;
 411 }
 412
 413 /*
 414  * find a VMA in the global tree
 415  */
 416 static inline struct vm_area_struct *find_nommu_vma(unsigned long start)
 417 {
 418         struct vm_area_struct *vma;
 419         struct rb_node *n = nommu_vma_tree.rb_node;
 420
 421         while (n) {
 422                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
 423
 424                 if (start < vma->vm_start)
 425                         n = n->rb_left;
 426                 else if (start > vma->vm_start)
 427                         n = n->rb_right;
 428                 else
 429                         return vma;
 430         }
 431
 432         return NULL;
 433 }
 434
 435 /*
 436  * add a VMA in the global tree
 437  */
 438 static void add_nommu_vma(struct vm_area_struct *vma)
 439 {
 440         struct vm_area_struct *pvma;
 441         struct address_space *mapping;
 442         struct rb_node **p = &nommu_vma_tree.rb_node;
 443         struct rb_node *parent = NULL;
 444
 445         /* add the VMA to the mapping */
 446         if (vma->vm_file) {
 447                 mapping = vma->vm_file->f_mapping;
 448
 449                 flush_dcache_mmap_lock(mapping);
 450                 vma_prio_tree_insert(vma, &mapping->i_mmap);
 451                 flush_dcache_mmap_unlock(mapping);
 452         }
 453
 454         /* add the VMA to the master list */
 455         while (*p) {
 456                 parent = *p;
 457                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
 458
 459                 if (vma->vm_start < pvma->vm_start) {
 460                         p = &(*p)->rb_left;
 461                 }
 462                 else if (vma->vm_start > pvma->vm_start) {
 463                         p = &(*p)->rb_right;
 464                 }
 465                 else {
 466                         /* mappings are at the same address - this can only
 467                          * happen for shared-mem chardevs and shared file
 468                          * mappings backed by ramfs/tmpfs */
 469                         BUG_ON(!(pvma->vm_flags & VM_SHARED));
 470
 471                         if (vma < pvma)
 472                                 p = &(*p)->rb_left;
 473                         else if (vma > pvma)
 474                                 p = &(*p)->rb_right;
 475                         else
 476                                 BUG();
 477                 }
 478         }
 479
 480         rb_link_node(&vma->vm_rb, parent, p);
 481         rb_insert_color(&vma->vm_rb, &nommu_vma_tree);
 482 }
 483
 484 /*
 485  * delete a VMA from the global list
 486  */
 487 static void delete_nommu_vma(struct vm_area_struct *vma)
 488 {
 489         struct address_space *mapping;
 490
 491         /* remove the VMA from the mapping */
 492         if (vma->vm_file) {
 493                 mapping = vma->vm_file->f_mapping;
 494
 495                 flush_dcache_mmap_lock(mapping);
 496                 vma_prio_tree_remove(vma, &mapping->i_mmap);
 497                 flush_dcache_mmap_unlock(mapping);
 498         }
 499
 500         /* remove from the master list */
 501         rb_erase(&vma->vm_rb, &nommu_vma_tree);
 502 }
 503
 504 /*
 505  * determine whether a mapping should be permitted and, if so, what sort of
 506  * mapping we're capable of supporting
 507  */
 508 static int validate_mmap_request(struct file *file,
 509                                  unsigned long addr,
 510                                  unsigned long len,
 511                                  unsigned long prot,
 512                                  unsigned long flags,
 513                                  unsigned long pgoff,
 514                                  unsigned long *_capabilities)
 515 {
 516         unsigned long capabilities;
 517         unsigned long reqprot = prot;
 518         int ret;
 519
 520         /* do the simple checks first */
 521         if (flags & MAP_FIXED || addr) {
 522                 printk(KERN_DEBUG
 523                        "%d: Can't do fixed-address/overlay mmap of RAM\n",
 524                        current->pid);
 525                 return -EINVAL;
 526         }
 527
 528         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
 529             (flags & MAP_TYPE) != MAP_SHARED)
 530                 return -EINVAL;
 531
 532         if (!len)
 533                 return -EINVAL;
 534
 535         /* Careful about overflows.. */
 536         len = PAGE_ALIGN(len);
 537         if (!len || len > TASK_SIZE)
 538                 return -ENOMEM;
 539
 540         /* offset overflow? */
 541         if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
 542                 return -EOVERFLOW;
 543
 544         if (file) {
 545                 /* validate file mapping requests */
 546                 struct address_space *mapping;
 547
 548                 /* files must support mmap */
 549                 if (!file->f_op || !file->f_op->mmap)
 550                         return -ENODEV;
 551
 552                 /* work out if what we've got could possibly be shared
 553                  * - we support chardevs that provide their own "memory"
 554                  * - we support files/blockdevs that are memory backed
 555                  */
 556                 mapping = file->f_mapping;
 557                 if (!mapping)
 558                         mapping = file->f_path.dentry->d_inode->i_mapping;
 559
 560                 capabilities = 0;
 561                 if (mapping && mapping->backing_dev_info)
 562                         capabilities = mapping->backing_dev_info->capabilities;
 563
 564                 if (!capabilities) {
 565                         /* no explicit capabilities set, so assume some
 566                          * defaults */
 567                         switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
 568                         case S_IFREG:
 569                         case S_IFBLK:
 570                                 capabilities = BDI_CAP_MAP_COPY;
 571                                 break;
 572
 573                         case S_IFCHR:
 574                                 capabilities =
 575                                         BDI_CAP_MAP_DIRECT |
 576                                         BDI_CAP_READ_MAP |
 577                                         BDI_CAP_WRITE_MAP;
 578                                 break;
 579
 580                         default:
 581                                 return -EINVAL;
 582                         }
 583                 }
 584
 585                 /* eliminate any capabilities that we can't support on this
 586                  * device */
 587                 if (!file->f_op->get_unmapped_area)
 588                         capabilities &= ~BDI_CAP_MAP_DIRECT;
 589                 if (!file->f_op->read)
 590                         capabilities &= ~BDI_CAP_MAP_COPY;
 591
 592                 if (flags & MAP_SHARED) {
 593                         /* do checks for writing, appending and locking */
 594                         if ((prot & PROT_WRITE) &&
 595                             !(file->f_mode & FMODE_WRITE))
 596                                 return -EACCES;
 597
 598                         if (IS_APPEND(file->f_path.dentry->d_inode) &&
 599                             (file->f_mode & FMODE_WRITE))
 600                                 return -EACCES;
 601
 602                         if (locks_verify_locked(file->f_path.dentry->d_inode))
 603                                 return -EAGAIN;
 604
 605                         if (!(capabilities & BDI_CAP_MAP_DIRECT))
 606                                 return -ENODEV;
 607
 608                         if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
 609                             ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
 610                             ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
 611                             ) {
 612                                 printk("MAP_SHARED not completely supported on !MMU\n");
 613                                 return -EINVAL;
 614                         }
 615
 616                         /* we mustn't privatise shared mappings */
 617                         capabilities &= ~BDI_CAP_MAP_COPY;
 618                 }
 619                 else {
 620                         /* we're going to read the file into private memory we
 621                          * allocate */
 622                         if (!(capabilities & BDI_CAP_MAP_COPY))
 623                                 return -ENODEV;
 624
 625                         /* we don't permit a private writable mapping to be
 626                          * shared with the backing device */
 627                         if (prot & PROT_WRITE)
 628                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
 629                 }
 630
 631                 /* handle executable mappings and implied executable
 632                  * mappings */
 633                 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
 634                         if (prot & PROT_EXEC)
 635                                 return -EPERM;
 636                 }
 637                 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
 638                         /* handle implication of PROT_EXEC by PROT_READ */
 639                         if (current->personality & READ_IMPLIES_EXEC) {
 640                                 if (capabilities & BDI_CAP_EXEC_MAP)
 641                                         prot |= PROT_EXEC;
 642                         }
 643                 }
 644                 else if ((prot & PROT_READ) &&
 645                          (prot & PROT_EXEC) &&
 646                          !(capabilities & BDI_CAP_EXEC_MAP)
 647                          ) {
 648                         /* backing file is not executable, try to copy */
 649                         capabilities &= ~BDI_CAP_MAP_DIRECT;
 650                 }
 651         }
 652         else {
 653                 /* anonymous mappings are always memory backed and can be
 654                  * privately mapped
 655                  */
 656                 capabilities = BDI_CAP_MAP_COPY;
 657
 658                 /* handle PROT_EXEC implication by PROT_READ */
 659                 if ((prot & PROT_READ) &&
 660                     (current->personality & READ_IMPLIES_EXEC))
 661                         prot |= PROT_EXEC;
 662         }
 663
 664         /* allow the security API to have its say */
 665         ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
 666         if (ret < 0)
 667                 return ret;
 668
 669         /* looks okay */
 670         *_capabilities = capabilities;
 671         return 0;
 672 }
 673
 674 /*
 675  * we've determined that we can make the mapping, now translate what we
 676  * now know into VMA flags
 677  */
 678 static unsigned long determine_vm_flags(struct file *file,
 679                                         unsigned long prot,
 680                                         unsigned long flags,
 681                                         unsigned long capabilities)
 682 {
 683         unsigned long vm_flags;
 684
 685         vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
 686         vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 687         /* vm_flags |= mm->def_flags; */
 688
 689         if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
 690                 /* attempt to share read-only copies of mapped file chunks */
 691                 if (file && !(prot & PROT_WRITE))
 692                         vm_flags |= VM_MAYSHARE;
 693         }
 694         else {
 695                 /* overlay a shareable mapping on the backing device or inode
 696                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
 697                  * romfs/cramfs */
 698                 if (flags & MAP_SHARED)
 699                         vm_flags |= VM_MAYSHARE | VM_SHARED;
 700                 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
 701                         vm_flags |= VM_MAYSHARE;
 702         }
 703
 704         /* refuse to let anyone share private mappings with this process if
 705          * it's being traced - otherwise breakpoints set in it may interfere
 706          * with another untraced process
 707          */
 708         if ((flags & MAP_PRIVATE) && (current->ptrace & PT_PTRACED))
 709                 vm_flags &= ~VM_MAYSHARE;
 710
 711         return vm_flags;
 712 }
 713
 714 /*
 715  * set up a shared mapping on a file
 716  */
 717 static int do_mmap_shared_file(struct vm_area_struct *vma, unsigned long len)
 718 {
 719         int ret;
 720
 721         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
 722         if (ret != -ENOSYS)
 723                 return ret;
 724
 725         /* getting an ENOSYS error indicates that direct mmap isn't
 726          * possible (as opposed to tried but failed) so we'll fall
 727          * through to making a private copy of the data and mapping
 728          * that if we can */
 729         return -ENODEV;
 730 }
 731
 732 /*
 733  * set up a private mapping or an anonymous shared mapping
 734  */
 735 static int do_mmap_private(struct vm_area_struct *vma, unsigned long len)
 736 {
 737         void *base;
 738         int ret;
 739
 740         /* invoke the file's mapping function so that it can keep track of
 741          * shared mappings on devices or memory
 742          * - VM_MAYSHARE will be set if it may attempt to share
 743          */
 744         if (vma->vm_file) {
 745                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
 746                 if (ret != -ENOSYS) {
 747                         /* shouldn't return success if we're not sharing */
 748                         BUG_ON(ret == 0 && !(vma->vm_flags & VM_MAYSHARE));
 749                         return ret; /* success or a real error */
 750                 }
 751
 752                 /* getting an ENOSYS error indicates that direct mmap isn't
 753                  * possible (as opposed to tried but failed) so we'll try to
 754                  * make a private copy of the data and map that instead */
 755         }
 756
 757         /* allocate some memory to hold the mapping
 758          * - note that this may not return a page-aligned address if the object
 759          *   we're allocating is smaller than a page
 760          */
 761         base = kmalloc(len, GFP_KERNEL|__GFP_COMP);
 762         if (!base)
 763                 goto enomem;
 764
 765         vma->vm_start = (unsigned long) base;
 766         vma->vm_end = vma->vm_start + len;
 767         vma->vm_flags |= VM_MAPPED_COPY;
 768
 769 #ifdef WARN_ON_SLACK
 770         if (len + WARN_ON_SLACK <= kobjsize(result))
 771                 printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n",
 772                        len, current->pid, kobjsize(result) - len);
 773 #endif
 774
 775         if (vma->vm_file) {
 776                 /* read the contents of a file into the copy */
 777                 mm_segment_t old_fs;
 778                 loff_t fpos;
 779
 780                 fpos = vma->vm_pgoff;
 781                 fpos <<= PAGE_SHIFT;
 782
 783                 old_fs = get_fs();
 784                 set_fs(KERNEL_DS);
 785                 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
 786                 set_fs(old_fs);
 787
 788                 if (ret < 0)
 789                         goto error_free;
 790
 791                 /* clear the last little bit */
 792                 if (ret < len)
 793                         memset(base + ret, 0, len - ret);
 794
 795         } else {
 796                 /* if it's an anonymous mapping, then just clear it */
 797                 memset(base, 0, len);
 798         }
 799
 800         return 0;
 801
 802 error_free:
 803         kfree(base);
 804         vma->vm_start = 0;
 805         return ret;
 806
 807 enomem:
 808         printk("Allocation of length %lu from process %d failed\n",
 809                len, current->pid);
 810         show_free_areas();
 811         return -ENOMEM;
 812 }
 813
 814 /*
 815  * handle mapping creation for uClinux
 816  */
 817 unsigned long do_mmap_pgoff(struct file *file,
 818                             unsigned long addr,
 819                             unsigned long len,
 820                             unsigned long prot,
 821                             unsigned long flags,
 822                             unsigned long pgoff)
 823 {
 824         struct vm_list_struct *vml = NULL;
 825         struct vm_area_struct *vma = NULL;
 826         struct rb_node *rb;
 827         unsigned long capabilities, vm_flags;
 828         void *result;
 829         int ret;
 830
 831         /* decide whether we should attempt the mapping, and if so what sort of
 832          * mapping */
 833         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
 834                                     &capabilities);
 835         if (ret < 0)
 836                 return ret;
 837
 838         /* we've determined that we can make the mapping, now translate what we
 839          * now know into VMA flags */
 840         vm_flags = determine_vm_flags(file, prot, flags, capabilities);
 841
 842         /* we're going to need to record the mapping if it works */
 843         vml = kzalloc(sizeof(struct vm_list_struct), GFP_KERNEL);
 844         if (!vml)
 845                 goto error_getting_vml;
 846
 847         down_write(&nommu_vma_sem);
 848
 849         /* if we want to share, we need to check for VMAs created by other
 850          * mmap() calls that overlap with our proposed mapping
 851          * - we can only share with an exact match on most regular files
 852          * - shared mappings on character devices and memory backed files are
 853          *   permitted to overlap inexactly as far as we are concerned for in
 854          *   these cases, sharing is handled in the driver or filesystem rather
 855          *   than here
 856          */
 857         if (vm_flags & VM_MAYSHARE) {
 858                 unsigned long pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 859                 unsigned long vmpglen;
 860
 861                 /* suppress VMA sharing for shared regions */
 862                 if (vm_flags & VM_SHARED &&
 863                     capabilities & BDI_CAP_MAP_DIRECT)
 864                         goto dont_share_VMAs;
 865
 866                 for (rb = rb_first(&nommu_vma_tree); rb; rb = rb_next(rb)) {
 867                         vma = rb_entry(rb, struct vm_area_struct, vm_rb);
 868
 869                         if (!(vma->vm_flags & VM_MAYSHARE))
 870                                 continue;
 871
 872                         /* search for overlapping mappings on the same file */
 873                         if (vma->vm_file->f_path.dentry->d_inode != file->f_path.dentry->d_inode)
 874                                 continue;
 875
 876                         if (vma->vm_pgoff >= pgoff + pglen)
 877                                 continue;
 878
 879                         vmpglen = vma->vm_end - vma->vm_start + PAGE_SIZE - 1;
 880                         vmpglen >>= PAGE_SHIFT;
 881                         if (pgoff >= vma->vm_pgoff + vmpglen)
 882                                 continue;
 883
 884                         /* handle inexactly overlapping matches between mappings */
 885                         if (vma->vm_pgoff != pgoff || vmpglen != pglen) {
 886                                 if (!(capabilities & BDI_CAP_MAP_DIRECT))
 887                                         goto sharing_violation;
 888                                 continue;
 889                         }
 890
 891                         /* we've found a VMA we can share */
 892                         atomic_inc(&vma->vm_usage);
 893
 894                         vml->vma = vma;
 895                         result = (void *) vma->vm_start;
 896                         goto shared;
 897                 }
 898
 899         dont_share_VMAs:
 900                 vma = NULL;
 901
 902                 /* obtain the address at which to make a shared mapping
 903                  * - this is the hook for quasi-memory character devices to
 904                  *   tell us the location of a shared mapping
 905                  */
 906                 if (file && file->f_op->get_unmapped_area) {
 907                         addr = file->f_op->get_unmapped_area(file, addr, len,
 908                                                              pgoff, flags);
 909                         if (IS_ERR((void *) addr)) {
 910                                 ret = addr;
 911                                 if (ret != (unsigned long) -ENOSYS)
 912                                         goto error;
 913
 914                                 /* the driver refused to tell us where to site
 915                                  * the mapping so we'll have to attempt to copy
 916                                  * it */
 917                                 ret = (unsigned long) -ENODEV;
 918                                 if (!(capabilities & BDI_CAP_MAP_COPY))
 919                                         goto error;
 920
 921                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
 922                         }
 923                 }
 924         }
 925
 926         /* we're going to need a VMA struct as well */
 927         vma = kzalloc(sizeof(struct vm_area_struct), GFP_KERNEL);
 928         if (!vma)
 929                 goto error_getting_vma;
 930
 931         INIT_LIST_HEAD(&vma->anon_vma_node);
 932         atomic_set(&vma->vm_usage, 1);
 933         if (file)
 934                 get_file(file);
 935         vma->vm_file    = file;
 936         vma->vm_flags   = vm_flags;
 937         vma->vm_start   = addr;
 938         vma->vm_end     = addr + len;
 939         vma->vm_pgoff   = pgoff;
 940
 941         vml->vma = vma;
 942
 943         /* set up the mapping */
 944         if (file && vma->vm_flags & VM_SHARED)
 945                 ret = do_mmap_shared_file(vma, len);
 946         else
 947                 ret = do_mmap_private(vma, len);
 948         if (ret < 0)
 949                 goto error;
 950
 951         /* okay... we have a mapping; now we have to register it */
 952         result = (void *) vma->vm_start;
 953
 954         if (vma->vm_flags & VM_MAPPED_COPY) {
 955                 realalloc += kobjsize(result);
 956                 askedalloc += len;
 957         }
 958
 959         realalloc += kobjsize(vma);
 960         askedalloc += sizeof(*vma);
 961
 962         current->mm->total_vm += len >> PAGE_SHIFT;
 963
 964         add_nommu_vma(vma);
 965
 966  shared:
 967         realalloc += kobjsize(vml);
 968         askedalloc += sizeof(*vml);
 969
 970         add_vma_to_mm(current->mm, vml);
 971
 972         up_write(&nommu_vma_sem);
 973
 974         if (prot & PROT_EXEC)
 975                 flush_icache_range((unsigned long) result,
 976                                    (unsigned long) result + len);
 977
 978 #ifdef DEBUG
 979         printk("do_mmap:\n");
 980         show_process_blocks();
 981 #endif
 982
 983         return (unsigned long) result;
 984
 985  error:
 986         up_write(&nommu_vma_sem);
 987         kfree(vml);
 988         if (vma) {
 989                 if (vma->vm_file)
 990                         fput(vma->vm_file);
 991                 kfree(vma);
 992         }
 993         return ret;
 994
 995  sharing_violation:
 996         up_write(&nommu_vma_sem);
 997         printk("Attempt to share mismatched mappings\n");
 998         kfree(vml);
 999         return -EINVAL;
1000
1001  error_getting_vma:
1002         up_write(&nommu_vma_sem);
1003         kfree(vml);
1004         printk("Allocation of vma for %lu byte allocation from process %d failed\n",
1005                len, current->pid);
1006         show_free_areas();
1007         return -ENOMEM;
1008
1009  error_getting_vml:
1010         printk("Allocation of vml for %lu byte allocation from process %d failed\n",
1011                len, current->pid);
1012         show_free_areas();
1013         return -ENOMEM;
1014 }
1015 EXPORT_SYMBOL(do_mmap_pgoff);
1016
1017 /*
1018  * handle mapping disposal for uClinux
1019  */
1020 static void put_vma(struct vm_area_struct *vma)
1021 {
1022         if (vma) {
1023                 down_write(&nommu_vma_sem);
1024
1025                 if (atomic_dec_and_test(&vma->vm_usage)) {
1026                         delete_nommu_vma(vma);
1027
1028                         if (vma->vm_ops && vma->vm_ops->close)
1029                                 vma->vm_ops->close(vma);
1030
1031                         /* IO memory and memory shared directly out of the pagecache from
1032                          * ramfs/tmpfs mustn't be released here */
1033                         if (vma->vm_flags & VM_MAPPED_COPY) {
1034                                 realalloc -= kobjsize((void *) vma->vm_start);
1035                                 askedalloc -= vma->vm_end - vma->vm_start;
1036                                 kfree((void *) vma->vm_start);
1037                         }
1038
1039                         realalloc -= kobjsize(vma);
1040                         askedalloc -= sizeof(*vma);
1041
1042                         if (vma->vm_file)
1043                                 fput(vma->vm_file);
1044                         kfree(vma);
1045                 }
1046
1047                 up_write(&nommu_vma_sem);
1048         }
1049 }
1050
1051 /*
1052  * release a mapping
1053  * - under NOMMU conditions the parameters must match exactly to the mapping to
1054  *   be removed
1055  */
1056 int do_munmap(struct mm_struct *mm, unsigned long addr, size_t len)
1057 {
1058         struct vm_list_struct *vml, **parent;
1059         unsigned long end = addr + len;
1060
1061 #ifdef DEBUG
1062         printk("do_munmap:\n");
1063 #endif
1064
1065         for (parent = &mm->context.vmlist; *parent; parent = &(*parent)->next) {
1066                 if ((*parent)->vma->vm_start > addr)
1067                         break;
1068                 if ((*parent)->vma->vm_start == addr &&
1069                     ((len == 0) || ((*parent)->vma->vm_end == end)))
1070                         goto found;
1071         }
1072
1073         printk("munmap of non-mmaped memory by process %d (%s): %p\n",
1074                current->pid, current->comm, (void *) addr);
1075         return -EINVAL;
1076
1077  found:
1078         vml = *parent;
1079
1080         put_vma(vml->vma);
1081
1082         *parent = vml->next;
1083         realalloc -= kobjsize(vml);
1084         askedalloc -= sizeof(*vml);
1085         kfree(vml);
1086
1087         update_hiwater_vm(mm);
1088         mm->total_vm -= len >> PAGE_SHIFT;
1089
1090 #ifdef DEBUG
1091         show_process_blocks();
1092 #endif
1093
1094         return 0;
1095 }
1096 EXPORT_SYMBOL(do_munmap);
1097
1098 asmlinkage long sys_munmap(unsigned long addr, size_t len)
1099 {
1100         int ret;
1101         struct mm_struct *mm = current->mm;
1102
1103         down_write(&mm->mmap_sem);
1104         ret = do_munmap(mm, addr, len);
1105         up_write(&mm->mmap_sem);
1106         return ret;
1107 }
1108
1109 /*
1110  * Release all mappings
1111  */
1112 void exit_mmap(struct mm_struct * mm)
1113 {
1114         struct vm_list_struct *tmp;
1115
1116         if (mm) {
1117 #ifdef DEBUG
1118                 printk("Exit_mmap:\n");
1119 #endif
1120
1121                 mm->total_vm = 0;
1122
1123                 while ((tmp = mm->context.vmlist)) {
1124                         mm->context.vmlist = tmp->next;
1125                         put_vma(tmp->vma);
1126
1127                         realalloc -= kobjsize(tmp);
1128                         askedalloc -= sizeof(*tmp);
1129                         kfree(tmp);
1130                 }
1131
1132 #ifdef DEBUG
1133                 show_process_blocks();
1134 #endif
1135         }
1136 }
1137
1138 unsigned long do_brk(unsigned long addr, unsigned long len)
1139 {
1140         return -ENOMEM;
1141 }
1142
1143 /*
1144  * expand (or shrink) an existing mapping, potentially moving it at the same
1145  * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1146  *
1147  * under NOMMU conditions, we only permit changing a mapping's size, and only
1148  * as long as it stays within the hole allocated by the kmalloc() call in
1149  * do_mmap_pgoff() and the block is not shareable
1150  *
1151  * MREMAP_FIXED is not supported under NOMMU conditions
1152  */
1153 unsigned long do_mremap(unsigned long addr,
1154                         unsigned long old_len, unsigned long new_len,
1155                         unsigned long flags, unsigned long new_addr)
1156 {
1157         struct vm_area_struct *vma;
1158
1159         /* insanity checks first */
1160         if (new_len == 0)
1161                 return (unsigned long) -EINVAL;
1162
1163         if (flags & MREMAP_FIXED && new_addr != addr)
1164                 return (unsigned long) -EINVAL;
1165
1166         vma = find_vma_exact(current->mm, addr);
1167         if (!vma)
1168                 return (unsigned long) -EINVAL;
1169
1170         if (vma->vm_end != vma->vm_start + old_len)
1171                 return (unsigned long) -EFAULT;
1172
1173         if (vma->vm_flags & VM_MAYSHARE)
1174                 return (unsigned long) -EPERM;
1175
1176         if (new_len > kobjsize((void *) addr))
1177                 return (unsigned long) -ENOMEM;
1178
1179         /* all checks complete - do it */
1180         vma->vm_end = vma->vm_start + new_len;
1181
1182         askedalloc -= old_len;
1183         askedalloc += new_len;
1184
1185         return vma->vm_start;
1186 }
1187 EXPORT_SYMBOL(do_mremap);
1188
1189 asmlinkage unsigned long sys_mremap(unsigned long addr,
1190         unsigned long old_len, unsigned long new_len,
1191         unsigned long flags, unsigned long new_addr)
1192 {
1193         unsigned long ret;
1194
1195         down_write(&current->mm->mmap_sem);
1196         ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1197         up_write(&current->mm->mmap_sem);
1198         return ret;
1199 }
1200
1201 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1202                         unsigned int foll_flags)
1203 {
1204         return NULL;
1205 }
1206
1207 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1208                 unsigned long to, unsigned long size, pgprot_t prot)
1209 {
1210         vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1211         return 0;
1212 }
1213 EXPORT_SYMBOL(remap_pfn_range);
1214
1215 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1216 {
1217 }
1218
1219 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1220         unsigned long len, unsigned long pgoff, unsigned long flags)
1221 {
1222         return -ENOMEM;
1223 }
1224
1225 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1226 {
1227 }
1228
1229 void unmap_mapping_range(struct address_space *mapping,
1230                          loff_t const holebegin, loff_t const holelen,
1231                          int even_cows)
1232 {
1233 }
1234 EXPORT_SYMBOL(unmap_mapping_range);
1235
1236 /*
1237  * ask for an unmapped area at which to create a mapping on a file
1238  */
1239 unsigned long get_unmapped_area(struct file *file, unsigned long addr,
1240                                 unsigned long len, unsigned long pgoff,
1241                                 unsigned long flags)
1242 {
1243         unsigned long (*get_area)(struct file *, unsigned long, unsigned long,
1244                                   unsigned long, unsigned long);
1245
1246         get_area = current->mm->get_unmapped_area;
1247         if (file && file->f_op && file->f_op->get_unmapped_area)
1248                 get_area = file->f_op->get_unmapped_area;
1249
1250         if (!get_area)
1251                 return -ENOSYS;
1252
1253         return get_area(file, addr, len, pgoff, flags);
1254 }
1255 EXPORT_SYMBOL(get_unmapped_area);
1256
1257 /*
1258  * Check that a process has enough memory to allocate a new virtual
1259  * mapping. 0 means there is enough memory for the allocation to
1260  * succeed and -ENOMEM implies there is not.
1261  *
1262  * We currently support three overcommit policies, which are set via the
1263  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1264  *
1265  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1266  * Additional code 2002 Jul 20 by Robert Love.
1267  *
1268  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1269  *
1270  * Note this is a helper function intended to be used by LSMs which
1271  * wish to use this logic.
1272  */
1273 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1274 {
1275         unsigned long free, allowed;
1276
1277         vm_acct_memory(pages);
1278
1279         /*
1280          * Sometimes we want to use more memory than we have
1281          */
1282         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1283                 return 0;
1284
1285         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1286                 unsigned long n;
1287
1288                 free = global_page_state(NR_FILE_PAGES);
1289                 free += nr_swap_pages;
1290
1291                 /*
1292                  * Any slabs which are created with the
1293                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1294                  * which are reclaimable, under pressure.  The dentry
1295                  * cache and most inode caches should fall into this
1296                  */
1297                 free += global_page_state(NR_SLAB_RECLAIMABLE);
1298
1299                 /*
1300                  * Leave the last 3% for root
1301                  */
1302                 if (!cap_sys_admin)
1303                         free -= free / 32;
1304
1305                 if (free > pages)
1306                         return 0;
1307
1308                 /*
1309                  * nr_free_pages() is very expensive on large systems,
1310                  * only call if we're about to fail.
1311                  */
1312                 n = nr_free_pages();
1313
1314                 /*
1315                  * Leave reserved pages. The pages are not for anonymous pages.
1316                  */
1317                 if (n <= totalreserve_pages)
1318                         goto error;
1319                 else
1320                         n -= totalreserve_pages;
1321
1322                 /*
1323                  * Leave the last 3% for root
1324                  */
1325                 if (!cap_sys_admin)
1326                         n -= n / 32;
1327                 free += n;
1328
1329                 if (free > pages)
1330                         return 0;
1331
1332                 goto error;
1333         }
1334
1335         allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1336         /*
1337          * Leave the last 3% for root
1338          */
1339         if (!cap_sys_admin)
1340                 allowed -= allowed / 32;
1341         allowed += total_swap_pages;
1342
1343         /* Don't let a single process grow too big:
1344            leave 3% of the size of this process for other processes */
1345         allowed -= current->mm->total_vm / 32;
1346
1347         /*
1348          * cast `allowed' as a signed long because vm_committed_space
1349          * sometimes has a negative value
1350          */
1351         if (atomic_read(&vm_committed_space) < (long)allowed)
1352                 return 0;
1353 error:
1354         vm_unacct_memory(pages);
1355
1356         return -ENOMEM;
1357 }
1358
1359 int in_gate_area_no_task(unsigned long addr)
1360 {
1361         return 0;
1362 }
1363
1364 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1365 {
1366         BUG();
1367         return 0;
1368 }
1369 EXPORT_SYMBOL(filemap_fault);
1370
1371 /*
1372  * Access another process' address space.
1373  * - source/target buffer must be kernel space
1374  */
1375 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1376 {
1377         struct vm_area_struct *vma;
1378         struct mm_struct *mm;
1379
1380         if (addr + len < addr)
1381                 return 0;
1382
1383         mm = get_task_mm(tsk);
1384         if (!mm)
1385                 return 0;
1386
1387         down_read(&mm->mmap_sem);
1388
1389         /* the access must start within one of the target process's mappings */
1390         vma = find_vma(mm, addr);
1391         if (vma) {
1392                 /* don't overrun this mapping */
1393                 if (addr + len >= vma->vm_end)
1394                         len = vma->vm_end - addr;
1395
1396                 /* only read or write mappings where it is permitted */
1397                 if (write && vma->vm_flags & VM_MAYWRITE)
1398                         len -= copy_to_user((void *) addr, buf, len);
1399                 else if (!write && vma->vm_flags & VM_MAYREAD)
1400                         len -= copy_from_user(buf, (void *) addr, len);
1401                 else
1402                         len = 0;
1403         } else {
1404                 len = 0;
1405         }
1406
1407         up_read(&mm->mmap_sem);
1408         mmput(mm);
1409         return len;
1410 }