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) 2004-2008 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  *  Copyright (c) 2007-2008 Paul Mundt <lethal@linux-sh.org>
  14  */
  15
  16 #include <linux/module.h>
  17 #include <linux/mm.h>
  18 #include <linux/mman.h>
  19 #include <linux/swap.h>
  20 #include <linux/file.h>
  21 #include <linux/highmem.h>
  22 #include <linux/pagemap.h>
  23 #include <linux/slab.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/tracehook.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/backing-dev.h>
  28 #include <linux/mount.h>
  29 #include <linux/personality.h>
  30 #include <linux/security.h>
  31 #include <linux/syscalls.h>
  32
  33 #include <asm/uaccess.h>
  34 #include <asm/tlb.h>
  35 #include <asm/tlbflush.h>
  36 #include "internal.h"
  37
  38 static inline __attribute__((format(printf, 1, 2)))
  39 void no_printk(const char *fmt, ...)
  40 {
  41 }
  42
  43 #if 0
  44 #define kenter(FMT, ...) \
  45         printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
  46 #define kleave(FMT, ...) \
  47         printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
  48 #define kdebug(FMT, ...) \
  49         printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__)
  50 #else
  51 #define kenter(FMT, ...) \
  52         no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
  53 #define kleave(FMT, ...) \
  54         no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
  55 #define kdebug(FMT, ...) \
  56         no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__)
  57 #endif
  58
  59 #include "internal.h"
  60
  61 void *high_memory;
  62 struct page *mem_map;
  63 unsigned long max_mapnr;
  64 unsigned long num_physpages;
  65 atomic_long_t vm_committed_space = ATOMIC_LONG_INIT(0);
  66 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
  67 int sysctl_overcommit_ratio = 50; /* default is 50% */
  68 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
  69 int sysctl_nr_trim_pages = 1; /* page trimming behaviour */
  70 int heap_stack_gap = 0;
  71
  72 atomic_t mmap_pages_allocated;
  73
  74 EXPORT_SYMBOL(mem_map);
  75 EXPORT_SYMBOL(num_physpages);
  76
  77 /* list of mapped, potentially shareable regions */
  78 static struct kmem_cache *vm_region_jar;
  79 struct rb_root nommu_region_tree = RB_ROOT;
  80 DECLARE_RWSEM(nommu_region_sem);
  81
  82 struct vm_operations_struct generic_file_vm_ops = {
  83 };
  84
  85 /*
  86  * Handle all mappings that got truncated by a "truncate()"
  87  * system call.
  88  *
  89  * NOTE! We have to be ready to update the memory sharing
  90  * between the file and the memory map for a potential last
  91  * incomplete page.  Ugly, but necessary.
  92  */
  93 int vmtruncate(struct inode *inode, loff_t offset)
  94 {
  95         struct address_space *mapping = inode->i_mapping;
  96         unsigned long limit;
  97
  98         if (inode->i_size < offset)
  99                 goto do_expand;
 100         i_size_write(inode, offset);
 101
 102         truncate_inode_pages(mapping, offset);
 103         goto out_truncate;
 104
 105 do_expand:
 106         limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
 107         if (limit != RLIM_INFINITY && offset > limit)
 108                 goto out_sig;
 109         if (offset > inode->i_sb->s_maxbytes)
 110                 goto out;
 111         i_size_write(inode, offset);
 112
 113 out_truncate:
 114         if (inode->i_op->truncate)
 115                 inode->i_op->truncate(inode);
 116         return 0;
 117 out_sig:
 118         send_sig(SIGXFSZ, current, 0);
 119 out:
 120         return -EFBIG;
 121 }
 122
 123 EXPORT_SYMBOL(vmtruncate);
 124
 125 /*
 126  * Return the total memory allocated for this pointer, not
 127  * just what the caller asked for.
 128  *
 129  * Doesn't have to be accurate, i.e. may have races.
 130  */
 131 unsigned int kobjsize(const void *objp)
 132 {
 133         struct page *page;
 134
 135         /*
 136          * If the object we have should not have ksize performed on it,
 137          * return size of 0
 138          */
 139         if (!objp || !virt_addr_valid(objp))
 140                 return 0;
 141
 142         page = virt_to_head_page(objp);
 143
 144         /*
 145          * If the allocator sets PageSlab, we know the pointer came from
 146          * kmalloc().
 147          */
 148         if (PageSlab(page))
 149                 return ksize(objp);
 150
 151         /*
 152          * If it's not a compound page, see if we have a matching VMA
 153          * region. This test is intentionally done in reverse order,
 154          * so if there's no VMA, we still fall through and hand back
 155          * PAGE_SIZE for 0-order pages.
 156          */
 157         if (!PageCompound(page)) {
 158                 struct vm_area_struct *vma;
 159
 160                 vma = find_vma(current->mm, (unsigned long)objp);
 161                 if (vma)
 162                         return vma->vm_end - vma->vm_start;
 163         }
 164
 165         /*
 166          * The ksize() function is only guaranteed to work for pointers
 167          * returned by kmalloc(). So handle arbitrary pointers here.
 168          */
 169         return PAGE_SIZE << compound_order(page);
 170 }
 171
 172 int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
 173                      unsigned long start, int len, int flags,
 174                 struct page **pages, struct vm_area_struct **vmas)
 175 {
 176         struct vm_area_struct *vma;
 177         unsigned long vm_flags;
 178         int i;
 179         int write = !!(flags & GUP_FLAGS_WRITE);
 180         int force = !!(flags & GUP_FLAGS_FORCE);
 181         int ignore = !!(flags & GUP_FLAGS_IGNORE_VMA_PERMISSIONS);
 182
 183         /* calculate required read or write permissions.
 184          * - if 'force' is set, we only require the "MAY" flags.
 185          */
 186         vm_flags  = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
 187         vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
 188
 189         for (i = 0; i < len; i++) {
 190                 vma = find_vma(mm, start);
 191                 if (!vma)
 192                         goto finish_or_fault;
 193
 194                 /* protect what we can, including chardevs */
 195                 if (vma->vm_flags & (VM_IO | VM_PFNMAP) ||
 196                     (!ignore && !(vm_flags & vma->vm_flags)))
 197                         goto finish_or_fault;
 198
 199                 if (pages) {
 200                         pages[i] = virt_to_page(start);
 201                         if (pages[i])
 202                                 page_cache_get(pages[i]);
 203                 }
 204                 if (vmas)
 205                         vmas[i] = vma;
 206                 start += PAGE_SIZE;
 207         }
 208
 209         return i;
 210
 211 finish_or_fault:
 212         return i ? : -EFAULT;
 213 }
 214
 215
 216 /*
 217  * get a list of pages in an address range belonging to the specified process
 218  * and indicate the VMA that covers each page
 219  * - this is potentially dodgy as we may end incrementing the page count of a
 220  *   slab page or a secondary page from a compound page
 221  * - don't permit access to VMAs that don't support it, such as I/O mappings
 222  */
 223 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
 224         unsigned long start, int len, int write, int force,
 225         struct page **pages, struct vm_area_struct **vmas)
 226 {
 227         int flags = 0;
 228
 229         if (write)
 230                 flags |= GUP_FLAGS_WRITE;
 231         if (force)
 232                 flags |= GUP_FLAGS_FORCE;
 233
 234         return __get_user_pages(tsk, mm,
 235                                 start, len, flags,
 236                                 pages, vmas);
 237 }
 238 EXPORT_SYMBOL(get_user_pages);
 239
 240 DEFINE_RWLOCK(vmlist_lock);
 241 struct vm_struct *vmlist;
 242
 243 void vfree(const void *addr)
 244 {
 245         kfree(addr);
 246 }
 247 EXPORT_SYMBOL(vfree);
 248
 249 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
 250 {
 251         /*
 252          *  You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
 253          * returns only a logical address.
 254          */
 255         return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
 256 }
 257 EXPORT_SYMBOL(__vmalloc);
 258
 259 void *vmalloc_user(unsigned long size)
 260 {
 261         void *ret;
 262
 263         ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
 264                         PAGE_KERNEL);
 265         if (ret) {
 266                 struct vm_area_struct *vma;
 267
 268                 down_write(&current->mm->mmap_sem);
 269                 vma = find_vma(current->mm, (unsigned long)ret);
 270                 if (vma)
 271                         vma->vm_flags |= VM_USERMAP;
 272                 up_write(&current->mm->mmap_sem);
 273         }
 274
 275         return ret;
 276 }
 277 EXPORT_SYMBOL(vmalloc_user);
 278
 279 struct page *vmalloc_to_page(const void *addr)
 280 {
 281         return virt_to_page(addr);
 282 }
 283 EXPORT_SYMBOL(vmalloc_to_page);
 284
 285 unsigned long vmalloc_to_pfn(const void *addr)
 286 {
 287         return page_to_pfn(virt_to_page(addr));
 288 }
 289 EXPORT_SYMBOL(vmalloc_to_pfn);
 290
 291 long vread(char *buf, char *addr, unsigned long count)
 292 {
 293         memcpy(buf, addr, count);
 294         return count;
 295 }
 296
 297 long vwrite(char *buf, char *addr, unsigned long count)
 298 {
 299         /* Don't allow overflow */
 300         if ((unsigned long) addr + count < count)
 301                 count = -(unsigned long) addr;
 302
 303         memcpy(addr, buf, count);
 304         return(count);
 305 }
 306
 307 /*
 308  *      vmalloc  -  allocate virtually continguos memory
 309  *
 310  *      @size:          allocation size
 311  *
 312  *      Allocate enough pages to cover @size from the page level
 313  *      allocator and map them into continguos kernel virtual space.
 314  *
 315  *      For tight control over page level allocator and protection flags
 316  *      use __vmalloc() instead.
 317  */
 318 void *vmalloc(unsigned long size)
 319 {
 320        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
 321 }
 322 EXPORT_SYMBOL(vmalloc);
 323
 324 void *vmalloc_node(unsigned long size, int node)
 325 {
 326         return vmalloc(size);
 327 }
 328 EXPORT_SYMBOL(vmalloc_node);
 329
 330 #ifndef PAGE_KERNEL_EXEC
 331 # define PAGE_KERNEL_EXEC PAGE_KERNEL
 332 #endif
 333
 334 /**
 335  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
 336  *      @size:          allocation size
 337  *
 338  *      Kernel-internal function to allocate enough pages to cover @size
 339  *      the page level allocator and map them into contiguous and
 340  *      executable kernel virtual space.
 341  *
 342  *      For tight control over page level allocator and protection flags
 343  *      use __vmalloc() instead.
 344  */
 345
 346 void *vmalloc_exec(unsigned long size)
 347 {
 348         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
 349 }
 350
 351 /**
 352  * vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
 353  *      @size:          allocation size
 354  *
 355  *      Allocate enough 32bit PA addressable pages to cover @size from the
 356  *      page level allocator and map them into continguos kernel virtual space.
 357  */
 358 void *vmalloc_32(unsigned long size)
 359 {
 360         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
 361 }
 362 EXPORT_SYMBOL(vmalloc_32);
 363
 364 /**
 365  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
 366  *      @size:          allocation size
 367  *
 368  * The resulting memory area is 32bit addressable and zeroed so it can be
 369  * mapped to userspace without leaking data.
 370  *
 371  * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
 372  * remap_vmalloc_range() are permissible.
 373  */
 374 void *vmalloc_32_user(unsigned long size)
 375 {
 376         /*
 377          * We'll have to sort out the ZONE_DMA bits for 64-bit,
 378          * but for now this can simply use vmalloc_user() directly.
 379          */
 380         return vmalloc_user(size);
 381 }
 382 EXPORT_SYMBOL(vmalloc_32_user);
 383
 384 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
 385 {
 386         BUG();
 387         return NULL;
 388 }
 389 EXPORT_SYMBOL(vmap);
 390
 391 void vunmap(const void *addr)
 392 {
 393         BUG();
 394 }
 395 EXPORT_SYMBOL(vunmap);
 396
 397 /*
 398  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
 399  * have one.
 400  */
 401 void  __attribute__((weak)) vmalloc_sync_all(void)
 402 {
 403 }
 404
 405 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
 406                    struct page *page)
 407 {
 408         return -EINVAL;
 409 }
 410 EXPORT_SYMBOL(vm_insert_page);
 411
 412 /*
 413  *  sys_brk() for the most part doesn't need the global kernel
 414  *  lock, except when an application is doing something nasty
 415  *  like trying to un-brk an area that has already been mapped
 416  *  to a regular file.  in this case, the unmapping will need
 417  *  to invoke file system routines that need the global lock.
 418  */
 419 SYSCALL_DEFINE1(brk, unsigned long, brk)
 420 {
 421         struct mm_struct *mm = current->mm;
 422
 423         if (brk < mm->start_brk || brk > mm->context.end_brk)
 424                 return mm->brk;
 425
 426         if (mm->brk == brk)
 427                 return mm->brk;
 428
 429         /*
 430          * Always allow shrinking brk
 431          */
 432         if (brk <= mm->brk) {
 433                 mm->brk = brk;
 434                 return brk;
 435         }
 436
 437         /*
 438          * Ok, looks good - let it rip.
 439          */
 440         return mm->brk = brk;
 441 }
 442
 443 /*
 444  * initialise the VMA and region record slabs
 445  */
 446 void __init mmap_init(void)
 447 {
 448         vm_region_jar = kmem_cache_create("vm_region_jar",
 449                                           sizeof(struct vm_region), 0,
 450                                           SLAB_PANIC, NULL);
 451         vm_area_cachep = kmem_cache_create("vm_area_struct",
 452                                            sizeof(struct vm_area_struct), 0,
 453                                            SLAB_PANIC, NULL);
 454 }
 455
 456 /*
 457  * validate the region tree
 458  * - the caller must hold the region lock
 459  */
 460 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
 461 static noinline void validate_nommu_regions(void)
 462 {
 463         struct vm_region *region, *last;
 464         struct rb_node *p, *lastp;
 465
 466         lastp = rb_first(&nommu_region_tree);
 467         if (!lastp)
 468                 return;
 469
 470         last = rb_entry(lastp, struct vm_region, vm_rb);
 471         if (unlikely(last->vm_end <= last->vm_start))
 472                 BUG();
 473         if (unlikely(last->vm_top < last->vm_end))
 474                 BUG();
 475
 476         while ((p = rb_next(lastp))) {
 477                 region = rb_entry(p, struct vm_region, vm_rb);
 478                 last = rb_entry(lastp, struct vm_region, vm_rb);
 479
 480                 if (unlikely(region->vm_end <= region->vm_start))
 481                         BUG();
 482                 if (unlikely(region->vm_top < region->vm_end))
 483                         BUG();
 484                 if (unlikely(region->vm_start < last->vm_top))
 485                         BUG();
 486
 487                 lastp = p;
 488         }
 489 }
 490 #else
 491 #define validate_nommu_regions() do {} while(0)
 492 #endif
 493
 494 /*
 495  * add a region into the global tree
 496  */
 497 static void add_nommu_region(struct vm_region *region)
 498 {
 499         struct vm_region *pregion;
 500         struct rb_node **p, *parent;
 501
 502         validate_nommu_regions();
 503
 504         BUG_ON(region->vm_start & ~PAGE_MASK);
 505
 506         parent = NULL;
 507         p = &nommu_region_tree.rb_node;
 508         while (*p) {
 509                 parent = *p;
 510                 pregion = rb_entry(parent, struct vm_region, vm_rb);
 511                 if (region->vm_start < pregion->vm_start)
 512                         p = &(*p)->rb_left;
 513                 else if (region->vm_start > pregion->vm_start)
 514                         p = &(*p)->rb_right;
 515                 else if (pregion == region)
 516                         return;
 517                 else
 518                         BUG();
 519         }
 520
 521         rb_link_node(&region->vm_rb, parent, p);
 522         rb_insert_color(&region->vm_rb, &nommu_region_tree);
 523
 524         validate_nommu_regions();
 525 }
 526
 527 /*
 528  * delete a region from the global tree
 529  */
 530 static void delete_nommu_region(struct vm_region *region)
 531 {
 532         BUG_ON(!nommu_region_tree.rb_node);
 533
 534         validate_nommu_regions();
 535         rb_erase(&region->vm_rb, &nommu_region_tree);
 536         validate_nommu_regions();
 537 }
 538
 539 /*
 540  * free a contiguous series of pages
 541  */
 542 static void free_page_series(unsigned long from, unsigned long to)
 543 {
 544         for (; from < to; from += PAGE_SIZE) {
 545                 struct page *page = virt_to_page(from);
 546
 547                 kdebug("- free %lx", from);
 548                 atomic_dec(&mmap_pages_allocated);
 549                 if (page_count(page) != 1)
 550                         kdebug("free page %p [%d]", page, page_count(page));
 551                 put_page(page);
 552         }
 553 }
 554
 555 /*
 556  * release a reference to a region
 557  * - the caller must hold the region semaphore, which this releases
 558  * - the region may not have been added to the tree yet, in which case vm_top
 559  *   will equal vm_start
 560  */
 561 static void __put_nommu_region(struct vm_region *region)
 562         __releases(nommu_region_sem)
 563 {
 564         kenter("%p{%d}", region, atomic_read(&region->vm_usage));
 565
 566         BUG_ON(!nommu_region_tree.rb_node);
 567
 568         if (atomic_dec_and_test(&region->vm_usage)) {
 569                 if (region->vm_top > region->vm_start)
 570                         delete_nommu_region(region);
 571                 up_write(&nommu_region_sem);
 572
 573                 if (region->vm_file)
 574                         fput(region->vm_file);
 575
 576                 /* IO memory and memory shared directly out of the pagecache
 577                  * from ramfs/tmpfs mustn't be released here */
 578                 if (region->vm_flags & VM_MAPPED_COPY) {
 579                         kdebug("free series");
 580                         free_page_series(region->vm_start, region->vm_top);
 581                 }
 582                 kmem_cache_free(vm_region_jar, region);
 583         } else {
 584                 up_write(&nommu_region_sem);
 585         }
 586 }
 587
 588 /*
 589  * release a reference to a region
 590  */
 591 static void put_nommu_region(struct vm_region *region)
 592 {
 593         down_write(&nommu_region_sem);
 594         __put_nommu_region(region);
 595 }
 596
 597 /*
 598  * add a VMA into a process's mm_struct in the appropriate place in the list
 599  * and tree and add to the address space's page tree also if not an anonymous
 600  * page
 601  * - should be called with mm->mmap_sem held writelocked
 602  */
 603 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
 604 {
 605         struct vm_area_struct *pvma, **pp;
 606         struct address_space *mapping;
 607         struct rb_node **p, *parent;
 608
 609         kenter(",%p", vma);
 610
 611         BUG_ON(!vma->vm_region);
 612
 613         mm->map_count++;
 614         vma->vm_mm = mm;
 615
 616         /* add the VMA to the mapping */
 617         if (vma->vm_file) {
 618                 mapping = vma->vm_file->f_mapping;
 619
 620                 flush_dcache_mmap_lock(mapping);
 621                 vma_prio_tree_insert(vma, &mapping->i_mmap);
 622                 flush_dcache_mmap_unlock(mapping);
 623         }
 624
 625         /* add the VMA to the tree */
 626         parent = NULL;
 627         p = &mm->mm_rb.rb_node;
 628         while (*p) {
 629                 parent = *p;
 630                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
 631
 632                 /* sort by: start addr, end addr, VMA struct addr in that order
 633                  * (the latter is necessary as we may get identical VMAs) */
 634                 if (vma->vm_start < pvma->vm_start)
 635                         p = &(*p)->rb_left;
 636                 else if (vma->vm_start > pvma->vm_start)
 637                         p = &(*p)->rb_right;
 638                 else if (vma->vm_end < pvma->vm_end)
 639                         p = &(*p)->rb_left;
 640                 else if (vma->vm_end > pvma->vm_end)
 641                         p = &(*p)->rb_right;
 642                 else if (vma < pvma)
 643                         p = &(*p)->rb_left;
 644                 else if (vma > pvma)
 645                         p = &(*p)->rb_right;
 646                 else
 647                         BUG();
 648         }
 649
 650         rb_link_node(&vma->vm_rb, parent, p);
 651         rb_insert_color(&vma->vm_rb, &mm->mm_rb);
 652
 653         /* add VMA to the VMA list also */
 654         for (pp = &mm->mmap; (pvma = *pp); pp = &(*pp)->vm_next) {
 655                 if (pvma->vm_start > vma->vm_start)
 656                         break;
 657                 if (pvma->vm_start < vma->vm_start)
 658                         continue;
 659                 if (pvma->vm_end < vma->vm_end)
 660                         break;
 661         }
 662
 663         vma->vm_next = *pp;
 664         *pp = vma;
 665 }
 666
 667 /*
 668  * delete a VMA from its owning mm_struct and address space
 669  */
 670 static void delete_vma_from_mm(struct vm_area_struct *vma)
 671 {
 672         struct vm_area_struct **pp;
 673         struct address_space *mapping;
 674         struct mm_struct *mm = vma->vm_mm;
 675
 676         kenter("%p", vma);
 677
 678         mm->map_count--;
 679         if (mm->mmap_cache == vma)
 680                 mm->mmap_cache = NULL;
 681
 682         /* remove the VMA from the mapping */
 683         if (vma->vm_file) {
 684                 mapping = vma->vm_file->f_mapping;
 685
 686                 flush_dcache_mmap_lock(mapping);
 687                 vma_prio_tree_remove(vma, &mapping->i_mmap);
 688                 flush_dcache_mmap_unlock(mapping);
 689         }
 690
 691         /* remove from the MM's tree and list */
 692         rb_erase(&vma->vm_rb, &mm->mm_rb);
 693         for (pp = &mm->mmap; *pp; pp = &(*pp)->vm_next) {
 694                 if (*pp == vma) {
 695                         *pp = vma->vm_next;
 696                         break;
 697                 }
 698         }
 699
 700         vma->vm_mm = NULL;
 701 }
 702
 703 /*
 704  * destroy a VMA record
 705  */
 706 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
 707 {
 708         kenter("%p", vma);
 709         if (vma->vm_ops && vma->vm_ops->close)
 710                 vma->vm_ops->close(vma);
 711         if (vma->vm_file) {
 712                 fput(vma->vm_file);
 713                 if (vma->vm_flags & VM_EXECUTABLE)
 714                         removed_exe_file_vma(mm);
 715         }
 716         put_nommu_region(vma->vm_region);
 717         kmem_cache_free(vm_area_cachep, vma);
 718 }
 719
 720 /*
 721  * look up the first VMA in which addr resides, NULL if none
 722  * - should be called with mm->mmap_sem at least held readlocked
 723  */
 724 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
 725 {
 726         struct vm_area_struct *vma;
 727         struct rb_node *n = mm->mm_rb.rb_node;
 728
 729         /* check the cache first */
 730         vma = mm->mmap_cache;
 731         if (vma && vma->vm_start <= addr && vma->vm_end > addr)
 732                 return vma;
 733
 734         /* trawl the tree (there may be multiple mappings in which addr
 735          * resides) */
 736         for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
 737                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
 738                 if (vma->vm_start > addr)
 739                         return NULL;
 740                 if (vma->vm_end > addr) {
 741                         mm->mmap_cache = vma;
 742                         return vma;
 743                 }
 744         }
 745
 746         return NULL;
 747 }
 748 EXPORT_SYMBOL(find_vma);
 749
 750 /*
 751  * find a VMA
 752  * - we don't extend stack VMAs under NOMMU conditions
 753  */
 754 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
 755 {
 756         return find_vma(mm, addr);
 757 }
 758
 759 /*
 760  * expand a stack to a given address
 761  * - not supported under NOMMU conditions
 762  */
 763 int expand_stack(struct vm_area_struct *vma, unsigned long address)
 764 {
 765         return -ENOMEM;
 766 }
 767
 768 /*
 769  * look up the first VMA exactly that exactly matches addr
 770  * - should be called with mm->mmap_sem at least held readlocked
 771  */
 772 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
 773                                              unsigned long addr,
 774                                              unsigned long len)
 775 {
 776         struct vm_area_struct *vma;
 777         struct rb_node *n = mm->mm_rb.rb_node;
 778         unsigned long end = addr + len;
 779
 780         /* check the cache first */
 781         vma = mm->mmap_cache;
 782         if (vma && vma->vm_start == addr && vma->vm_end == end)
 783                 return vma;
 784
 785         /* trawl the tree (there may be multiple mappings in which addr
 786          * resides) */
 787         for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
 788                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
 789                 if (vma->vm_start < addr)
 790                         continue;
 791                 if (vma->vm_start > addr)
 792                         return NULL;
 793                 if (vma->vm_end == end) {
 794                         mm->mmap_cache = vma;
 795                         return vma;
 796                 }
 797         }
 798
 799         return NULL;
 800 }
 801
 802 /*
 803  * determine whether a mapping should be permitted and, if so, what sort of
 804  * mapping we're capable of supporting
 805  */
 806 static int validate_mmap_request(struct file *file,
 807                                  unsigned long addr,
 808                                  unsigned long len,
 809                                  unsigned long prot,
 810                                  unsigned long flags,
 811                                  unsigned long pgoff,
 812                                  unsigned long *_capabilities)
 813 {
 814         unsigned long capabilities, rlen;
 815         unsigned long reqprot = prot;
 816         int ret;
 817
 818         /* do the simple checks first */
 819         if (flags & MAP_FIXED || addr) {
 820                 printk(KERN_DEBUG
 821                        "%d: Can't do fixed-address/overlay mmap of RAM\n",
 822                        current->pid);
 823                 return -EINVAL;
 824         }
 825
 826         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
 827             (flags & MAP_TYPE) != MAP_SHARED)
 828                 return -EINVAL;
 829
 830         if (!len)
 831                 return -EINVAL;
 832
 833         /* Careful about overflows.. */
 834         rlen = PAGE_ALIGN(len);
 835         if (!rlen || rlen > TASK_SIZE)
 836                 return -ENOMEM;
 837
 838         /* offset overflow? */
 839         if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
 840                 return -EOVERFLOW;
 841
 842         if (file) {
 843                 /* validate file mapping requests */
 844                 struct address_space *mapping;
 845
 846                 /* files must support mmap */
 847                 if (!file->f_op || !file->f_op->mmap)
 848                         return -ENODEV;
 849
 850                 /* work out if what we've got could possibly be shared
 851                  * - we support chardevs that provide their own "memory"
 852                  * - we support files/blockdevs that are memory backed
 853                  */
 854                 mapping = file->f_mapping;
 855                 if (!mapping)
 856                         mapping = file->f_path.dentry->d_inode->i_mapping;
 857
 858                 capabilities = 0;
 859                 if (mapping && mapping->backing_dev_info)
 860                         capabilities = mapping->backing_dev_info->capabilities;
 861
 862                 if (!capabilities) {
 863                         /* no explicit capabilities set, so assume some
 864                          * defaults */
 865                         switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
 866                         case S_IFREG:
 867                         case S_IFBLK:
 868                                 capabilities = BDI_CAP_MAP_COPY;
 869                                 break;
 870
 871                         case S_IFCHR:
 872                                 capabilities =
 873                                         BDI_CAP_MAP_DIRECT |
 874                                         BDI_CAP_READ_MAP |
 875                                         BDI_CAP_WRITE_MAP;
 876                                 break;
 877
 878                         default:
 879                                 return -EINVAL;
 880                         }
 881                 }
 882
 883                 /* eliminate any capabilities that we can't support on this
 884                  * device */
 885                 if (!file->f_op->get_unmapped_area)
 886                         capabilities &= ~BDI_CAP_MAP_DIRECT;
 887                 if (!file->f_op->read)
 888                         capabilities &= ~BDI_CAP_MAP_COPY;
 889
 890                 if (flags & MAP_SHARED) {
 891                         /* do checks for writing, appending and locking */
 892                         if ((prot & PROT_WRITE) &&
 893                             !(file->f_mode & FMODE_WRITE))
 894                                 return -EACCES;
 895
 896                         if (IS_APPEND(file->f_path.dentry->d_inode) &&
 897                             (file->f_mode & FMODE_WRITE))
 898                                 return -EACCES;
 899
 900                         if (locks_verify_locked(file->f_path.dentry->d_inode))
 901                                 return -EAGAIN;
 902
 903                         if (!(capabilities & BDI_CAP_MAP_DIRECT))
 904                                 return -ENODEV;
 905
 906                         if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
 907                             ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
 908                             ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
 909                             ) {
 910                                 printk("MAP_SHARED not completely supported on !MMU\n");
 911                                 return -EINVAL;
 912                         }
 913
 914                         /* we mustn't privatise shared mappings */
 915                         capabilities &= ~BDI_CAP_MAP_COPY;
 916                 }
 917                 else {
 918                         /* we're going to read the file into private memory we
 919                          * allocate */
 920                         if (!(capabilities & BDI_CAP_MAP_COPY))
 921                                 return -ENODEV;
 922
 923                         /* we don't permit a private writable mapping to be
 924                          * shared with the backing device */
 925                         if (prot & PROT_WRITE)
 926                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
 927                 }
 928
 929                 /* handle executable mappings and implied executable
 930                  * mappings */
 931                 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
 932                         if (prot & PROT_EXEC)
 933                                 return -EPERM;
 934                 }
 935                 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
 936                         /* handle implication of PROT_EXEC by PROT_READ */
 937                         if (current->personality & READ_IMPLIES_EXEC) {
 938                                 if (capabilities & BDI_CAP_EXEC_MAP)
 939                                         prot |= PROT_EXEC;
 940                         }
 941                 }
 942                 else if ((prot & PROT_READ) &&
 943                          (prot & PROT_EXEC) &&
 944                          !(capabilities & BDI_CAP_EXEC_MAP)
 945                          ) {
 946                         /* backing file is not executable, try to copy */
 947                         capabilities &= ~BDI_CAP_MAP_DIRECT;
 948                 }
 949         }
 950         else {
 951                 /* anonymous mappings are always memory backed and can be
 952                  * privately mapped
 953                  */
 954                 capabilities = BDI_CAP_MAP_COPY;
 955
 956                 /* handle PROT_EXEC implication by PROT_READ */
 957                 if ((prot & PROT_READ) &&
 958                     (current->personality & READ_IMPLIES_EXEC))
 959                         prot |= PROT_EXEC;
 960         }
 961
 962         /* allow the security API to have its say */
 963         ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
 964         if (ret < 0)
 965                 return ret;
 966
 967         /* looks okay */
 968         *_capabilities = capabilities;
 969         return 0;
 970 }
 971
 972 /*
 973  * we've determined that we can make the mapping, now translate what we
 974  * now know into VMA flags
 975  */
 976 static unsigned long determine_vm_flags(struct file *file,
 977                                         unsigned long prot,
 978                                         unsigned long flags,
 979                                         unsigned long capabilities)
 980 {
 981         unsigned long vm_flags;
 982
 983         vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
 984         vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 985         /* vm_flags |= mm->def_flags; */
 986
 987         if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
 988                 /* attempt to share read-only copies of mapped file chunks */
 989                 if (file && !(prot & PROT_WRITE))
 990                         vm_flags |= VM_MAYSHARE;
 991         }
 992         else {
 993                 /* overlay a shareable mapping on the backing device or inode
 994                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
 995                  * romfs/cramfs */
 996                 if (flags & MAP_SHARED)
 997                         vm_flags |= VM_MAYSHARE | VM_SHARED;
 998                 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
 999                         vm_flags |= VM_MAYSHARE;
1000         }
1001
1002         /* refuse to let anyone share private mappings with this process if
1003          * it's being traced - otherwise breakpoints set in it may interfere
1004          * with another untraced process
1005          */
1006         if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current))
1007                 vm_flags &= ~VM_MAYSHARE;
1008
1009         return vm_flags;
1010 }
1011
1012 /*
1013  * set up a shared mapping on a file (the driver or filesystem provides and
1014  * pins the storage)
1015  */
1016 static int do_mmap_shared_file(struct vm_area_struct *vma)
1017 {
1018         int ret;
1019
1020         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1021         if (ret == 0) {
1022                 vma->vm_region->vm_top = vma->vm_region->vm_end;
1023                 return ret;
1024         }
1025         if (ret != -ENOSYS)
1026                 return ret;
1027
1028         /* getting an ENOSYS error indicates that direct mmap isn't
1029          * possible (as opposed to tried but failed) so we'll fall
1030          * through to making a private copy of the data and mapping
1031          * that if we can */
1032         return -ENODEV;
1033 }
1034
1035 /*
1036  * set up a private mapping or an anonymous shared mapping
1037  */
1038 static int do_mmap_private(struct vm_area_struct *vma,
1039                            struct vm_region *region,
1040                            unsigned long len)
1041 {
1042         struct page *pages;
1043         unsigned long total, point, n, rlen;
1044         void *base;
1045         int ret, order;
1046
1047         /* invoke the file's mapping function so that it can keep track of
1048          * shared mappings on devices or memory
1049          * - VM_MAYSHARE will be set if it may attempt to share
1050          */
1051         if (vma->vm_file) {
1052                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1053                 if (ret == 0) {
1054                         /* shouldn't return success if we're not sharing */
1055                         BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1056                         vma->vm_region->vm_top = vma->vm_region->vm_end;
1057                         return ret;
1058                 }
1059                 if (ret != -ENOSYS)
1060                         return ret;
1061
1062                 /* getting an ENOSYS error indicates that direct mmap isn't
1063                  * possible (as opposed to tried but failed) so we'll try to
1064                  * make a private copy of the data and map that instead */
1065         }
1066
1067         rlen = PAGE_ALIGN(len);
1068
1069         /* allocate some memory to hold the mapping
1070          * - note that this may not return a page-aligned address if the object
1071          *   we're allocating is smaller than a page
1072          */
1073         order = get_order(rlen);
1074         kdebug("alloc order %d for %lx", order, len);
1075
1076         pages = alloc_pages(GFP_KERNEL, order);
1077         if (!pages)
1078                 goto enomem;
1079
1080         total = 1 << order;
1081         atomic_add(total, &mmap_pages_allocated);
1082
1083         point = rlen >> PAGE_SHIFT;
1084
1085         /* we allocated a power-of-2 sized page set, so we may want to trim off
1086          * the excess */
1087         if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
1088                 while (total > point) {
1089                         order = ilog2(total - point);
1090                         n = 1 << order;
1091                         kdebug("shave %lu/%lu @%lu", n, total - point, total);
1092                         atomic_sub(n, &mmap_pages_allocated);
1093                         total -= n;
1094                         set_page_refcounted(pages + total);
1095                         __free_pages(pages + total, order);
1096                 }
1097         }
1098
1099         for (point = 1; point < total; point++)
1100                 set_page_refcounted(&pages[point]);
1101
1102         base = page_address(pages);
1103         region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1104         region->vm_start = (unsigned long) base;
1105         region->vm_end   = region->vm_start + rlen;
1106         region->vm_top   = region->vm_start + (total << PAGE_SHIFT);
1107
1108         vma->vm_start = region->vm_start;
1109         vma->vm_end   = region->vm_start + len;
1110
1111         if (vma->vm_file) {
1112                 /* read the contents of a file into the copy */
1113                 mm_segment_t old_fs;
1114                 loff_t fpos;
1115
1116                 fpos = vma->vm_pgoff;
1117                 fpos <<= PAGE_SHIFT;
1118
1119                 old_fs = get_fs();
1120                 set_fs(KERNEL_DS);
1121                 ret = vma->vm_file->f_op->read(vma->vm_file, base, rlen, &fpos);
1122                 set_fs(old_fs);
1123
1124                 if (ret < 0)
1125                         goto error_free;
1126
1127                 /* clear the last little bit */
1128                 if (ret < rlen)
1129                         memset(base + ret, 0, rlen - ret);
1130
1131         } else {
1132                 /* if it's an anonymous mapping, then just clear it */
1133                 memset(base, 0, rlen);
1134         }
1135
1136         return 0;
1137
1138 error_free:
1139         free_page_series(region->vm_start, region->vm_end);
1140         region->vm_start = vma->vm_start = 0;
1141         region->vm_end   = vma->vm_end = 0;
1142         region->vm_top   = 0;
1143         return ret;
1144
1145 enomem:
1146         printk("Allocation of length %lu from process %d failed\n",
1147                len, current->pid);
1148         show_free_areas();
1149         return -ENOMEM;
1150 }
1151
1152 /*
1153  * handle mapping creation for uClinux
1154  */
1155 unsigned long do_mmap_pgoff(struct file *file,
1156                             unsigned long addr,
1157                             unsigned long len,
1158                             unsigned long prot,
1159                             unsigned long flags,
1160                             unsigned long pgoff)
1161 {
1162         struct vm_area_struct *vma;
1163         struct vm_region *region;
1164         struct rb_node *rb;
1165         unsigned long capabilities, vm_flags, result;
1166         int ret;
1167
1168         kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1169
1170         if (!(flags & MAP_FIXED))
1171                 addr = round_hint_to_min(addr);
1172
1173         /* decide whether we should attempt the mapping, and if so what sort of
1174          * mapping */
1175         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1176                                     &capabilities);
1177         if (ret < 0) {
1178                 kleave(" = %d [val]", ret);
1179                 return ret;
1180         }
1181
1182         /* we've determined that we can make the mapping, now translate what we
1183          * now know into VMA flags */
1184         vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1185
1186         /* we're going to need to record the mapping */
1187         region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1188         if (!region)
1189                 goto error_getting_region;
1190
1191         vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1192         if (!vma)
1193                 goto error_getting_vma;
1194
1195         atomic_set(&region->vm_usage, 1);
1196         region->vm_flags = vm_flags;
1197         region->vm_pgoff = pgoff;
1198
1199         INIT_LIST_HEAD(&vma->anon_vma_node);
1200         vma->vm_flags = vm_flags;
1201         vma->vm_pgoff = pgoff;
1202
1203         if (file) {
1204                 region->vm_file = file;
1205                 get_file(file);
1206                 vma->vm_file = file;
1207                 get_file(file);
1208                 if (vm_flags & VM_EXECUTABLE) {
1209                         added_exe_file_vma(current->mm);
1210                         vma->vm_mm = current->mm;
1211                 }
1212         }
1213
1214         down_write(&nommu_region_sem);
1215
1216         /* if we want to share, we need to check for regions created by other
1217          * mmap() calls that overlap with our proposed mapping
1218          * - we can only share with a superset match on most regular files
1219          * - shared mappings on character devices and memory backed files are
1220          *   permitted to overlap inexactly as far as we are concerned for in
1221          *   these cases, sharing is handled in the driver or filesystem rather
1222          *   than here
1223          */
1224         if (vm_flags & VM_MAYSHARE) {
1225                 struct vm_region *pregion;
1226                 unsigned long pglen, rpglen, pgend, rpgend, start;
1227
1228                 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1229                 pgend = pgoff + pglen;
1230
1231                 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1232                         pregion = rb_entry(rb, struct vm_region, vm_rb);
1233
1234                         if (!(pregion->vm_flags & VM_MAYSHARE))
1235                                 continue;
1236
1237                         /* search for overlapping mappings on the same file */
1238                         if (pregion->vm_file->f_path.dentry->d_inode !=
1239                             file->f_path.dentry->d_inode)
1240                                 continue;
1241
1242                         if (pregion->vm_pgoff >= pgend)
1243                                 continue;
1244
1245                         rpglen = pregion->vm_end - pregion->vm_start;
1246                         rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1247                         rpgend = pregion->vm_pgoff + rpglen;
1248                         if (pgoff >= rpgend)
1249                                 continue;
1250
1251                         /* handle inexactly overlapping matches between
1252                          * mappings */
1253                         if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1254                             !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1255                                 /* new mapping is not a subset of the region */
1256                                 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1257                                         goto sharing_violation;
1258                                 continue;
1259                         }
1260
1261                         /* we've found a region we can share */
1262                         atomic_inc(&pregion->vm_usage);
1263                         vma->vm_region = pregion;
1264                         start = pregion->vm_start;
1265                         start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1266                         vma->vm_start = start;
1267                         vma->vm_end = start + len;
1268
1269                         if (pregion->vm_flags & VM_MAPPED_COPY) {
1270                                 kdebug("share copy");
1271                                 vma->vm_flags |= VM_MAPPED_COPY;
1272                         } else {
1273                                 kdebug("share mmap");
1274                                 ret = do_mmap_shared_file(vma);
1275                                 if (ret < 0) {
1276                                         vma->vm_region = NULL;
1277                                         vma->vm_start = 0;
1278                                         vma->vm_end = 0;
1279                                         atomic_dec(&pregion->vm_usage);
1280                                         pregion = NULL;
1281                                         goto error_just_free;
1282                                 }
1283                         }
1284                         fput(region->vm_file);
1285                         kmem_cache_free(vm_region_jar, region);
1286                         region = pregion;
1287                         result = start;
1288                         goto share;
1289                 }
1290
1291                 /* obtain the address at which to make a shared mapping
1292                  * - this is the hook for quasi-memory character devices to
1293                  *   tell us the location of a shared mapping
1294                  */
1295                 if (file && file->f_op->get_unmapped_area) {
1296                         addr = file->f_op->get_unmapped_area(file, addr, len,
1297                                                              pgoff, flags);
1298                         if (IS_ERR((void *) addr)) {
1299                                 ret = addr;
1300                                 if (ret != (unsigned long) -ENOSYS)
1301                                         goto error_just_free;
1302
1303                                 /* the driver refused to tell us where to site
1304                                  * the mapping so we'll have to attempt to copy
1305                                  * it */
1306                                 ret = (unsigned long) -ENODEV;
1307                                 if (!(capabilities & BDI_CAP_MAP_COPY))
1308                                         goto error_just_free;
1309
1310                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
1311                         } else {
1312                                 vma->vm_start = region->vm_start = addr;
1313                                 vma->vm_end = region->vm_end = addr + len;
1314                         }
1315                 }
1316         }
1317
1318         vma->vm_region = region;
1319
1320         /* set up the mapping */
1321         if (file && vma->vm_flags & VM_SHARED)
1322                 ret = do_mmap_shared_file(vma);
1323         else
1324                 ret = do_mmap_private(vma, region, len);
1325         if (ret < 0)
1326                 goto error_put_region;
1327
1328         add_nommu_region(region);
1329
1330         /* okay... we have a mapping; now we have to register it */
1331         result = vma->vm_start;
1332
1333         current->mm->total_vm += len >> PAGE_SHIFT;
1334
1335 share:
1336         add_vma_to_mm(current->mm, vma);
1337
1338         up_write(&nommu_region_sem);
1339
1340         if (prot & PROT_EXEC)
1341                 flush_icache_range(result, result + len);
1342
1343         kleave(" = %lx", result);
1344         return result;
1345
1346 error_put_region:
1347         __put_nommu_region(region);
1348         if (vma) {
1349                 if (vma->vm_file) {
1350                         fput(vma->vm_file);
1351                         if (vma->vm_flags & VM_EXECUTABLE)
1352                                 removed_exe_file_vma(vma->vm_mm);
1353                 }
1354                 kmem_cache_free(vm_area_cachep, vma);
1355         }
1356         kleave(" = %d [pr]", ret);
1357         return ret;
1358
1359 error_just_free:
1360         up_write(&nommu_region_sem);
1361 error:
1362         fput(region->vm_file);
1363         kmem_cache_free(vm_region_jar, region);
1364         fput(vma->vm_file);
1365         if (vma->vm_flags & VM_EXECUTABLE)
1366                 removed_exe_file_vma(vma->vm_mm);
1367         kmem_cache_free(vm_area_cachep, vma);
1368         kleave(" = %d", ret);
1369         return ret;
1370
1371 sharing_violation:
1372         up_write(&nommu_region_sem);
1373         printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1374         ret = -EINVAL;
1375         goto error;
1376
1377 error_getting_vma:
1378         kmem_cache_free(vm_region_jar, region);
1379         printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1380                " from process %d failed\n",
1381                len, current->pid);
1382         show_free_areas();
1383         return -ENOMEM;
1384
1385 error_getting_region:
1386         printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1387                " from process %d failed\n",
1388                len, current->pid);
1389         show_free_areas();
1390         return -ENOMEM;
1391 }
1392 EXPORT_SYMBOL(do_mmap_pgoff);
1393
1394 /*
1395  * split a vma into two pieces at address 'addr', a new vma is allocated either
1396  * for the first part or the tail.
1397  */
1398 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1399               unsigned long addr, int new_below)
1400 {
1401         struct vm_area_struct *new;
1402         struct vm_region *region;
1403         unsigned long npages;
1404
1405         kenter("");
1406
1407         /* we're only permitted to split anonymous regions that have a single
1408          * owner */
1409         if (vma->vm_file ||
1410             atomic_read(&vma->vm_region->vm_usage) != 1)
1411                 return -ENOMEM;
1412
1413         if (mm->map_count >= sysctl_max_map_count)
1414                 return -ENOMEM;
1415
1416         region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1417         if (!region)
1418                 return -ENOMEM;
1419
1420         new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1421         if (!new) {
1422                 kmem_cache_free(vm_region_jar, region);
1423                 return -ENOMEM;
1424         }
1425
1426         /* most fields are the same, copy all, and then fixup */
1427         *new = *vma;
1428         *region = *vma->vm_region;
1429         new->vm_region = region;
1430
1431         npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1432
1433         if (new_below) {
1434                 region->vm_top = region->vm_end = new->vm_end = addr;
1435         } else {
1436                 region->vm_start = new->vm_start = addr;
1437                 region->vm_pgoff = new->vm_pgoff += npages;
1438         }
1439
1440         if (new->vm_ops && new->vm_ops->open)
1441                 new->vm_ops->open(new);
1442
1443         delete_vma_from_mm(vma);
1444         down_write(&nommu_region_sem);
1445         delete_nommu_region(vma->vm_region);
1446         if (new_below) {
1447                 vma->vm_region->vm_start = vma->vm_start = addr;
1448                 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1449         } else {
1450                 vma->vm_region->vm_end = vma->vm_end = addr;
1451                 vma->vm_region->vm_top = addr;
1452         }
1453         add_nommu_region(vma->vm_region);
1454         add_nommu_region(new->vm_region);
1455         up_write(&nommu_region_sem);
1456         add_vma_to_mm(mm, vma);
1457         add_vma_to_mm(mm, new);
1458         return 0;
1459 }
1460
1461 /*
1462  * shrink a VMA by removing the specified chunk from either the beginning or
1463  * the end
1464  */
1465 static int shrink_vma(struct mm_struct *mm,
1466                       struct vm_area_struct *vma,
1467                       unsigned long from, unsigned long to)
1468 {
1469         struct vm_region *region;
1470
1471         kenter("");
1472
1473         /* adjust the VMA's pointers, which may reposition it in the MM's tree
1474          * and list */
1475         delete_vma_from_mm(vma);
1476         if (from > vma->vm_start)
1477                 vma->vm_end = from;
1478         else
1479                 vma->vm_start = to;
1480         add_vma_to_mm(mm, vma);
1481
1482         /* cut the backing region down to size */
1483         region = vma->vm_region;
1484         BUG_ON(atomic_read(&region->vm_usage) != 1);
1485
1486         down_write(&nommu_region_sem);
1487         delete_nommu_region(region);
1488         if (from > region->vm_start) {
1489                 to = region->vm_top;
1490                 region->vm_top = region->vm_end = from;
1491         } else {
1492                 region->vm_start = to;
1493         }
1494         add_nommu_region(region);
1495         up_write(&nommu_region_sem);
1496
1497         free_page_series(from, to);
1498         return 0;
1499 }
1500
1501 /*
1502  * release a mapping
1503  * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1504  *   VMA, though it need not cover the whole VMA
1505  */
1506 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1507 {
1508         struct vm_area_struct *vma;
1509         struct rb_node *rb;
1510         unsigned long end = start + len;
1511         int ret;
1512
1513         kenter(",%lx,%zx", start, len);
1514
1515         if (len == 0)
1516                 return -EINVAL;
1517
1518         /* find the first potentially overlapping VMA */
1519         vma = find_vma(mm, start);
1520         if (!vma) {
1521                 printk(KERN_WARNING
1522                        "munmap of memory not mmapped by process %d (%s):"
1523                        " 0x%lx-0x%lx\n",
1524                        current->pid, current->comm, start, start + len - 1);
1525                 return -EINVAL;
1526         }
1527
1528         /* we're allowed to split an anonymous VMA but not a file-backed one */
1529         if (vma->vm_file) {
1530                 do {
1531                         if (start > vma->vm_start) {
1532                                 kleave(" = -EINVAL [miss]");
1533                                 return -EINVAL;
1534                         }
1535                         if (end == vma->vm_end)
1536                                 goto erase_whole_vma;
1537                         rb = rb_next(&vma->vm_rb);
1538                         vma = rb_entry(rb, struct vm_area_struct, vm_rb);
1539                 } while (rb);
1540                 kleave(" = -EINVAL [split file]");
1541                 return -EINVAL;
1542         } else {
1543                 /* the chunk must be a subset of the VMA found */
1544                 if (start == vma->vm_start && end == vma->vm_end)
1545                         goto erase_whole_vma;
1546                 if (start < vma->vm_start || end > vma->vm_end) {
1547                         kleave(" = -EINVAL [superset]");
1548                         return -EINVAL;
1549                 }
1550                 if (start & ~PAGE_MASK) {
1551                         kleave(" = -EINVAL [unaligned start]");
1552                         return -EINVAL;
1553                 }
1554                 if (end != vma->vm_end && end & ~PAGE_MASK) {
1555                         kleave(" = -EINVAL [unaligned split]");
1556                         return -EINVAL;
1557                 }
1558                 if (start != vma->vm_start && end != vma->vm_end) {
1559                         ret = split_vma(mm, vma, start, 1);
1560                         if (ret < 0) {
1561                                 kleave(" = %d [split]", ret);
1562                                 return ret;
1563                         }
1564                 }
1565                 return shrink_vma(mm, vma, start, end);
1566         }
1567
1568 erase_whole_vma:
1569         delete_vma_from_mm(vma);
1570         delete_vma(mm, vma);
1571         kleave(" = 0");
1572         return 0;
1573 }
1574 EXPORT_SYMBOL(do_munmap);
1575
1576 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1577 {
1578         int ret;
1579         struct mm_struct *mm = current->mm;
1580
1581         down_write(&mm->mmap_sem);
1582         ret = do_munmap(mm, addr, len);
1583         up_write(&mm->mmap_sem);
1584         return ret;
1585 }
1586
1587 /*
1588  * release all the mappings made in a process's VM space
1589  */
1590 void exit_mmap(struct mm_struct *mm)
1591 {
1592         struct vm_area_struct *vma;
1593
1594         if (!mm)
1595                 return;
1596
1597         kenter("");
1598
1599         mm->total_vm = 0;
1600
1601         while ((vma = mm->mmap)) {
1602                 mm->mmap = vma->vm_next;
1603                 delete_vma_from_mm(vma);
1604                 delete_vma(mm, vma);
1605         }
1606
1607         kleave("");
1608 }
1609
1610 unsigned long do_brk(unsigned long addr, unsigned long len)
1611 {
1612         return -ENOMEM;
1613 }
1614
1615 /*
1616  * expand (or shrink) an existing mapping, potentially moving it at the same
1617  * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1618  *
1619  * under NOMMU conditions, we only permit changing a mapping's size, and only
1620  * as long as it stays within the region allocated by do_mmap_private() and the
1621  * block is not shareable
1622  *
1623  * MREMAP_FIXED is not supported under NOMMU conditions
1624  */
1625 unsigned long do_mremap(unsigned long addr,
1626                         unsigned long old_len, unsigned long new_len,
1627                         unsigned long flags, unsigned long new_addr)
1628 {
1629         struct vm_area_struct *vma;
1630
1631         /* insanity checks first */
1632         if (old_len == 0 || new_len == 0)
1633                 return (unsigned long) -EINVAL;
1634
1635         if (addr & ~PAGE_MASK)
1636                 return -EINVAL;
1637
1638         if (flags & MREMAP_FIXED && new_addr != addr)
1639                 return (unsigned long) -EINVAL;
1640
1641         vma = find_vma_exact(current->mm, addr, old_len);
1642         if (!vma)
1643                 return (unsigned long) -EINVAL;
1644
1645         if (vma->vm_end != vma->vm_start + old_len)
1646                 return (unsigned long) -EFAULT;
1647
1648         if (vma->vm_flags & VM_MAYSHARE)
1649                 return (unsigned long) -EPERM;
1650
1651         if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1652                 return (unsigned long) -ENOMEM;
1653
1654         /* all checks complete - do it */
1655         vma->vm_end = vma->vm_start + new_len;
1656         return vma->vm_start;
1657 }
1658 EXPORT_SYMBOL(do_mremap);
1659
1660 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1661                 unsigned long, new_len, unsigned long, flags,
1662                 unsigned long, new_addr)
1663 {
1664         unsigned long ret;
1665
1666         down_write(&current->mm->mmap_sem);
1667         ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1668         up_write(&current->mm->mmap_sem);
1669         return ret;
1670 }
1671
1672 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1673                         unsigned int foll_flags)
1674 {
1675         return NULL;
1676 }
1677
1678 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1679                 unsigned long to, unsigned long size, pgprot_t prot)
1680 {
1681         vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1682         return 0;
1683 }
1684 EXPORT_SYMBOL(remap_pfn_range);
1685
1686 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1687                         unsigned long pgoff)
1688 {
1689         unsigned int size = vma->vm_end - vma->vm_start;
1690
1691         if (!(vma->vm_flags & VM_USERMAP))
1692                 return -EINVAL;
1693
1694         vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1695         vma->vm_end = vma->vm_start + size;
1696
1697         return 0;
1698 }
1699 EXPORT_SYMBOL(remap_vmalloc_range);
1700
1701 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1702 {
1703 }
1704
1705 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1706         unsigned long len, unsigned long pgoff, unsigned long flags)
1707 {
1708         return -ENOMEM;
1709 }
1710
1711 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1712 {
1713 }
1714
1715 void unmap_mapping_range(struct address_space *mapping,
1716                          loff_t const holebegin, loff_t const holelen,
1717                          int even_cows)
1718 {
1719 }
1720 EXPORT_SYMBOL(unmap_mapping_range);
1721
1722 /*
1723  * ask for an unmapped area at which to create a mapping on a file
1724  */
1725 unsigned long get_unmapped_area(struct file *file, unsigned long addr,
1726                                 unsigned long len, unsigned long pgoff,
1727                                 unsigned long flags)
1728 {
1729         unsigned long (*get_area)(struct file *, unsigned long, unsigned long,
1730                                   unsigned long, unsigned long);
1731
1732         get_area = current->mm->get_unmapped_area;
1733         if (file && file->f_op && file->f_op->get_unmapped_area)
1734                 get_area = file->f_op->get_unmapped_area;
1735
1736         if (!get_area)
1737                 return -ENOSYS;
1738
1739         return get_area(file, addr, len, pgoff, flags);
1740 }
1741 EXPORT_SYMBOL(get_unmapped_area);
1742
1743 /*
1744  * Check that a process has enough memory to allocate a new virtual
1745  * mapping. 0 means there is enough memory for the allocation to
1746  * succeed and -ENOMEM implies there is not.
1747  *
1748  * We currently support three overcommit policies, which are set via the
1749  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1750  *
1751  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1752  * Additional code 2002 Jul 20 by Robert Love.
1753  *
1754  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1755  *
1756  * Note this is a helper function intended to be used by LSMs which
1757  * wish to use this logic.
1758  */
1759 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1760 {
1761         unsigned long free, allowed;
1762
1763         vm_acct_memory(pages);
1764
1765         /*
1766          * Sometimes we want to use more memory than we have
1767          */
1768         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1769                 return 0;
1770
1771         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1772                 unsigned long n;
1773
1774                 free = global_page_state(NR_FILE_PAGES);
1775                 free += nr_swap_pages;
1776
1777                 /*
1778                  * Any slabs which are created with the
1779                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1780                  * which are reclaimable, under pressure.  The dentry
1781                  * cache and most inode caches should fall into this
1782                  */
1783                 free += global_page_state(NR_SLAB_RECLAIMABLE);
1784
1785                 /*
1786                  * Leave the last 3% for root
1787                  */
1788                 if (!cap_sys_admin)
1789                         free -= free / 32;
1790
1791                 if (free > pages)
1792                         return 0;
1793
1794                 /*
1795                  * nr_free_pages() is very expensive on large systems,
1796                  * only call if we're about to fail.
1797                  */
1798                 n = nr_free_pages();
1799
1800                 /*
1801                  * Leave reserved pages. The pages are not for anonymous pages.
1802                  */
1803                 if (n <= totalreserve_pages)
1804                         goto error;
1805                 else
1806                         n -= totalreserve_pages;
1807
1808                 /*
1809                  * Leave the last 3% for root
1810                  */
1811                 if (!cap_sys_admin)
1812                         n -= n / 32;
1813                 free += n;
1814
1815                 if (free > pages)
1816                         return 0;
1817
1818                 goto error;
1819         }
1820
1821         allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1822         /*
1823          * Leave the last 3% for root
1824          */
1825         if (!cap_sys_admin)
1826                 allowed -= allowed / 32;
1827         allowed += total_swap_pages;
1828
1829         /* Don't let a single process grow too big:
1830            leave 3% of the size of this process for other processes */
1831         if (mm)
1832                 allowed -= mm->total_vm / 32;
1833
1834         /*
1835          * cast `allowed' as a signed long because vm_committed_space
1836          * sometimes has a negative value
1837          */
1838         if (atomic_long_read(&vm_committed_space) < (long)allowed)
1839                 return 0;
1840 error:
1841         vm_unacct_memory(pages);
1842
1843         return -ENOMEM;
1844 }
1845
1846 int in_gate_area_no_task(unsigned long addr)
1847 {
1848         return 0;
1849 }
1850
1851 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1852 {
1853         BUG();
1854         return 0;
1855 }
1856 EXPORT_SYMBOL(filemap_fault);
1857
1858 /*
1859  * Access another process' address space.
1860  * - source/target buffer must be kernel space
1861  */
1862 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1863 {
1864         struct vm_area_struct *vma;
1865         struct mm_struct *mm;
1866
1867         if (addr + len < addr)
1868                 return 0;
1869
1870         mm = get_task_mm(tsk);
1871         if (!mm)
1872                 return 0;
1873
1874         down_read(&mm->mmap_sem);
1875
1876         /* the access must start within one of the target process's mappings */
1877         vma = find_vma(mm, addr);
1878         if (vma) {
1879                 /* don't overrun this mapping */
1880                 if (addr + len >= vma->vm_end)
1881                         len = vma->vm_end - addr;
1882
1883                 /* only read or write mappings where it is permitted */
1884                 if (write && vma->vm_flags & VM_MAYWRITE)
1885                         len -= copy_to_user((void *) addr, buf, len);
1886                 else if (!write && vma->vm_flags & VM_MAYREAD)
1887                         len -= copy_from_user(buf, (void *) addr, len);
1888                 else
1889                         len = 0;
1890         } else {
1891                 len = 0;
1892         }
1893
1894         up_read(&mm->mmap_sem);
1895         mmput(mm);
1896         return len;
1897 }