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) 2007-2009 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 struct percpu_counter vm_committed_as;
  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 = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
  70 int heap_stack_gap = 0;
  71
  72 /* amount of vm to protect from userspace access */
  73 unsigned long mmap_min_addr = CONFIG_DEFAULT_MMAP_MIN_ADDR;
  74
  75 atomic_long_t mmap_pages_allocated;
  76
  77 EXPORT_SYMBOL(mem_map);
  78 EXPORT_SYMBOL(num_physpages);
  79
  80 /* list of mapped, potentially shareable regions */
  81 static struct kmem_cache *vm_region_jar;
  82 struct rb_root nommu_region_tree = RB_ROOT;
  83 DECLARE_RWSEM(nommu_region_sem);
  84
  85 struct vm_operations_struct generic_file_vm_ops = {
  86 };
  87
  88 /*
  89  * Handle all mappings that got truncated by a "truncate()"
  90  * system call.
  91  *
  92  * NOTE! We have to be ready to update the memory sharing
  93  * between the file and the memory map for a potential last
  94  * incomplete page.  Ugly, but necessary.
  95  */
  96 int vmtruncate(struct inode *inode, loff_t offset)
  97 {
  98         struct address_space *mapping = inode->i_mapping;
  99         unsigned long limit;
 100
 101         if (inode->i_size < offset)
 102                 goto do_expand;
 103         i_size_write(inode, offset);
 104
 105         truncate_inode_pages(mapping, offset);
 106         goto out_truncate;
 107
 108 do_expand:
 109         limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
 110         if (limit != RLIM_INFINITY && offset > limit)
 111                 goto out_sig;
 112         if (offset > inode->i_sb->s_maxbytes)
 113                 goto out;
 114         i_size_write(inode, offset);
 115
 116 out_truncate:
 117         if (inode->i_op->truncate)
 118                 inode->i_op->truncate(inode);
 119         return 0;
 120 out_sig:
 121         send_sig(SIGXFSZ, current, 0);
 122 out:
 123         return -EFBIG;
 124 }
 125
 126 EXPORT_SYMBOL(vmtruncate);
 127
 128 /*
 129  * Return the total memory allocated for this pointer, not
 130  * just what the caller asked for.
 131  *
 132  * Doesn't have to be accurate, i.e. may have races.
 133  */
 134 unsigned int kobjsize(const void *objp)
 135 {
 136         struct page *page;
 137
 138         /*
 139          * If the object we have should not have ksize performed on it,
 140          * return size of 0
 141          */
 142         if (!objp || !virt_addr_valid(objp))
 143                 return 0;
 144
 145         page = virt_to_head_page(objp);
 146
 147         /*
 148          * If the allocator sets PageSlab, we know the pointer came from
 149          * kmalloc().
 150          */
 151         if (PageSlab(page))
 152                 return ksize(objp);
 153
 154         /*
 155          * If it's not a compound page, see if we have a matching VMA
 156          * region. This test is intentionally done in reverse order,
 157          * so if there's no VMA, we still fall through and hand back
 158          * PAGE_SIZE for 0-order pages.
 159          */
 160         if (!PageCompound(page)) {
 161                 struct vm_area_struct *vma;
 162
 163                 vma = find_vma(current->mm, (unsigned long)objp);
 164                 if (vma)
 165                         return vma->vm_end - vma->vm_start;
 166         }
 167
 168         /*
 169          * The ksize() function is only guaranteed to work for pointers
 170          * returned by kmalloc(). So handle arbitrary pointers here.
 171          */
 172         return PAGE_SIZE << compound_order(page);
 173 }
 174
 175 int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
 176                      unsigned long start, int len, int flags,
 177                 struct page **pages, struct vm_area_struct **vmas)
 178 {
 179         struct vm_area_struct *vma;
 180         unsigned long vm_flags;
 181         int i;
 182         int write = !!(flags & GUP_FLAGS_WRITE);
 183         int force = !!(flags & GUP_FLAGS_FORCE);
 184         int ignore = !!(flags & GUP_FLAGS_IGNORE_VMA_PERMISSIONS);
 185
 186         /* calculate required read or write permissions.
 187          * - if 'force' is set, we only require the "MAY" flags.
 188          */
 189         vm_flags  = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
 190         vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
 191
 192         for (i = 0; i < len; i++) {
 193                 vma = find_vma(mm, start);
 194                 if (!vma)
 195                         goto finish_or_fault;
 196
 197                 /* protect what we can, including chardevs */
 198                 if (vma->vm_flags & (VM_IO | VM_PFNMAP) ||
 199                     (!ignore && !(vm_flags & vma->vm_flags)))
 200                         goto finish_or_fault;
 201
 202                 if (pages) {
 203                         pages[i] = virt_to_page(start);
 204                         if (pages[i])
 205                                 page_cache_get(pages[i]);
 206                 }
 207                 if (vmas)
 208                         vmas[i] = vma;
 209                 start += PAGE_SIZE;
 210         }
 211
 212         return i;
 213
 214 finish_or_fault:
 215         return i ? : -EFAULT;
 216 }
 217
 218
 219 /*
 220  * get a list of pages in an address range belonging to the specified process
 221  * and indicate the VMA that covers each page
 222  * - this is potentially dodgy as we may end incrementing the page count of a
 223  *   slab page or a secondary page from a compound page
 224  * - don't permit access to VMAs that don't support it, such as I/O mappings
 225  */
 226 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
 227         unsigned long start, int len, int write, int force,
 228         struct page **pages, struct vm_area_struct **vmas)
 229 {
 230         int flags = 0;
 231
 232         if (write)
 233                 flags |= GUP_FLAGS_WRITE;
 234         if (force)
 235                 flags |= GUP_FLAGS_FORCE;
 236
 237         return __get_user_pages(tsk, mm,
 238                                 start, len, flags,
 239                                 pages, vmas);
 240 }
 241 EXPORT_SYMBOL(get_user_pages);
 242
 243 DEFINE_RWLOCK(vmlist_lock);
 244 struct vm_struct *vmlist;
 245
 246 void vfree(const void *addr)
 247 {
 248         kfree(addr);
 249 }
 250 EXPORT_SYMBOL(vfree);
 251
 252 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
 253 {
 254         /*
 255          *  You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
 256          * returns only a logical address.
 257          */
 258         return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
 259 }
 260 EXPORT_SYMBOL(__vmalloc);
 261
 262 void *vmalloc_user(unsigned long size)
 263 {
 264         void *ret;
 265
 266         ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
 267                         PAGE_KERNEL);
 268         if (ret) {
 269                 struct vm_area_struct *vma;
 270
 271                 down_write(&current->mm->mmap_sem);
 272                 vma = find_vma(current->mm, (unsigned long)ret);
 273                 if (vma)
 274                         vma->vm_flags |= VM_USERMAP;
 275                 up_write(&current->mm->mmap_sem);
 276         }
 277
 278         return ret;
 279 }
 280 EXPORT_SYMBOL(vmalloc_user);
 281
 282 struct page *vmalloc_to_page(const void *addr)
 283 {
 284         return virt_to_page(addr);
 285 }
 286 EXPORT_SYMBOL(vmalloc_to_page);
 287
 288 unsigned long vmalloc_to_pfn(const void *addr)
 289 {
 290         return page_to_pfn(virt_to_page(addr));
 291 }
 292 EXPORT_SYMBOL(vmalloc_to_pfn);
 293
 294 long vread(char *buf, char *addr, unsigned long count)
 295 {
 296         memcpy(buf, addr, count);
 297         return count;
 298 }
 299
 300 long vwrite(char *buf, char *addr, unsigned long count)
 301 {
 302         /* Don't allow overflow */
 303         if ((unsigned long) addr + count < count)
 304                 count = -(unsigned long) addr;
 305
 306         memcpy(addr, buf, count);
 307         return(count);
 308 }
 309
 310 /*
 311  *      vmalloc  -  allocate virtually continguos memory
 312  *
 313  *      @size:          allocation size
 314  *
 315  *      Allocate enough pages to cover @size from the page level
 316  *      allocator and map them into continguos kernel virtual space.
 317  *
 318  *      For tight control over page level allocator and protection flags
 319  *      use __vmalloc() instead.
 320  */
 321 void *vmalloc(unsigned long size)
 322 {
 323        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
 324 }
 325 EXPORT_SYMBOL(vmalloc);
 326
 327 void *vmalloc_node(unsigned long size, int node)
 328 {
 329         return vmalloc(size);
 330 }
 331 EXPORT_SYMBOL(vmalloc_node);
 332
 333 #ifndef PAGE_KERNEL_EXEC
 334 # define PAGE_KERNEL_EXEC PAGE_KERNEL
 335 #endif
 336
 337 /**
 338  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
 339  *      @size:          allocation size
 340  *
 341  *      Kernel-internal function to allocate enough pages to cover @size
 342  *      the page level allocator and map them into contiguous and
 343  *      executable kernel virtual space.
 344  *
 345  *      For tight control over page level allocator and protection flags
 346  *      use __vmalloc() instead.
 347  */
 348
 349 void *vmalloc_exec(unsigned long size)
 350 {
 351         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
 352 }
 353
 354 /**
 355  * vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
 356  *      @size:          allocation size
 357  *
 358  *      Allocate enough 32bit PA addressable pages to cover @size from the
 359  *      page level allocator and map them into continguos kernel virtual space.
 360  */
 361 void *vmalloc_32(unsigned long size)
 362 {
 363         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
 364 }
 365 EXPORT_SYMBOL(vmalloc_32);
 366
 367 /**
 368  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
 369  *      @size:          allocation size
 370  *
 371  * The resulting memory area is 32bit addressable and zeroed so it can be
 372  * mapped to userspace without leaking data.
 373  *
 374  * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
 375  * remap_vmalloc_range() are permissible.
 376  */
 377 void *vmalloc_32_user(unsigned long size)
 378 {
 379         /*
 380          * We'll have to sort out the ZONE_DMA bits for 64-bit,
 381          * but for now this can simply use vmalloc_user() directly.
 382          */
 383         return vmalloc_user(size);
 384 }
 385 EXPORT_SYMBOL(vmalloc_32_user);
 386
 387 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
 388 {
 389         BUG();
 390         return NULL;
 391 }
 392 EXPORT_SYMBOL(vmap);
 393
 394 void vunmap(const void *addr)
 395 {
 396         BUG();
 397 }
 398 EXPORT_SYMBOL(vunmap);
 399
 400 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
 401 {
 402         BUG();
 403         return NULL;
 404 }
 405 EXPORT_SYMBOL(vm_map_ram);
 406
 407 void vm_unmap_ram(const void *mem, unsigned int count)
 408 {
 409         BUG();
 410 }
 411 EXPORT_SYMBOL(vm_unmap_ram);
 412
 413 void vm_unmap_aliases(void)
 414 {
 415 }
 416 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
 417
 418 /*
 419  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
 420  * have one.
 421  */
 422 void  __attribute__((weak)) vmalloc_sync_all(void)
 423 {
 424 }
 425
 426 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
 427                    struct page *page)
 428 {
 429         return -EINVAL;
 430 }
 431 EXPORT_SYMBOL(vm_insert_page);
 432
 433 /*
 434  *  sys_brk() for the most part doesn't need the global kernel
 435  *  lock, except when an application is doing something nasty
 436  *  like trying to un-brk an area that has already been mapped
 437  *  to a regular file.  in this case, the unmapping will need
 438  *  to invoke file system routines that need the global lock.
 439  */
 440 SYSCALL_DEFINE1(brk, unsigned long, brk)
 441 {
 442         struct mm_struct *mm = current->mm;
 443
 444         if (brk < mm->start_brk || brk > mm->context.end_brk)
 445                 return mm->brk;
 446
 447         if (mm->brk == brk)
 448                 return mm->brk;
 449
 450         /*
 451          * Always allow shrinking brk
 452          */
 453         if (brk <= mm->brk) {
 454                 mm->brk = brk;
 455                 return brk;
 456         }
 457
 458         /*
 459          * Ok, looks good - let it rip.
 460          */
 461         return mm->brk = brk;
 462 }
 463
 464 /*
 465  * initialise the VMA and region record slabs
 466  */
 467 void __init mmap_init(void)
 468 {
 469         int ret;
 470
 471         ret = percpu_counter_init(&vm_committed_as, 0);
 472         VM_BUG_ON(ret);
 473         vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
 474 }
 475
 476 /*
 477  * validate the region tree
 478  * - the caller must hold the region lock
 479  */
 480 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
 481 static noinline void validate_nommu_regions(void)
 482 {
 483         struct vm_region *region, *last;
 484         struct rb_node *p, *lastp;
 485
 486         lastp = rb_first(&nommu_region_tree);
 487         if (!lastp)
 488                 return;
 489
 490         last = rb_entry(lastp, struct vm_region, vm_rb);
 491         BUG_ON(unlikely(last->vm_end <= last->vm_start));
 492         BUG_ON(unlikely(last->vm_top < last->vm_end));
 493
 494         while ((p = rb_next(lastp))) {
 495                 region = rb_entry(p, struct vm_region, vm_rb);
 496                 last = rb_entry(lastp, struct vm_region, vm_rb);
 497
 498                 BUG_ON(unlikely(region->vm_end <= region->vm_start));
 499                 BUG_ON(unlikely(region->vm_top < region->vm_end));
 500                 BUG_ON(unlikely(region->vm_start < last->vm_top));
 501
 502                 lastp = p;
 503         }
 504 }
 505 #else
 506 static void validate_nommu_regions(void)
 507 {
 508 }
 509 #endif
 510
 511 /*
 512  * add a region into the global tree
 513  */
 514 static void add_nommu_region(struct vm_region *region)
 515 {
 516         struct vm_region *pregion;
 517         struct rb_node **p, *parent;
 518
 519         validate_nommu_regions();
 520
 521         parent = NULL;
 522         p = &nommu_region_tree.rb_node;
 523         while (*p) {
 524                 parent = *p;
 525                 pregion = rb_entry(parent, struct vm_region, vm_rb);
 526                 if (region->vm_start < pregion->vm_start)
 527                         p = &(*p)->rb_left;
 528                 else if (region->vm_start > pregion->vm_start)
 529                         p = &(*p)->rb_right;
 530                 else if (pregion == region)
 531                         return;
 532                 else
 533                         BUG();
 534         }
 535
 536         rb_link_node(&region->vm_rb, parent, p);
 537         rb_insert_color(&region->vm_rb, &nommu_region_tree);
 538
 539         validate_nommu_regions();
 540 }
 541
 542 /*
 543  * delete a region from the global tree
 544  */
 545 static void delete_nommu_region(struct vm_region *region)
 546 {
 547         BUG_ON(!nommu_region_tree.rb_node);
 548
 549         validate_nommu_regions();
 550         rb_erase(&region->vm_rb, &nommu_region_tree);
 551         validate_nommu_regions();
 552 }
 553
 554 /*
 555  * free a contiguous series of pages
 556  */
 557 static void free_page_series(unsigned long from, unsigned long to)
 558 {
 559         for (; from < to; from += PAGE_SIZE) {
 560                 struct page *page = virt_to_page(from);
 561
 562                 kdebug("- free %lx", from);
 563                 atomic_long_dec(&mmap_pages_allocated);
 564                 if (page_count(page) != 1)
 565                         kdebug("free page %p: refcount not one: %d",
 566                                page, page_count(page));
 567                 put_page(page);
 568         }
 569 }
 570
 571 /*
 572  * release a reference to a region
 573  * - the caller must hold the region semaphore for writing, which this releases
 574  * - the region may not have been added to the tree yet, in which case vm_top
 575  *   will equal vm_start
 576  */
 577 static void __put_nommu_region(struct vm_region *region)
 578         __releases(nommu_region_sem)
 579 {
 580         kenter("%p{%d}", region, atomic_read(&region->vm_usage));
 581
 582         BUG_ON(!nommu_region_tree.rb_node);
 583
 584         if (atomic_dec_and_test(&region->vm_usage)) {
 585                 if (region->vm_top > region->vm_start)
 586                         delete_nommu_region(region);
 587                 up_write(&nommu_region_sem);
 588
 589                 if (region->vm_file)
 590                         fput(region->vm_file);
 591
 592                 /* IO memory and memory shared directly out of the pagecache
 593                  * from ramfs/tmpfs mustn't be released here */
 594                 if (region->vm_flags & VM_MAPPED_COPY) {
 595                         kdebug("free series");
 596                         free_page_series(region->vm_start, region->vm_top);
 597                 }
 598                 kmem_cache_free(vm_region_jar, region);
 599         } else {
 600                 up_write(&nommu_region_sem);
 601         }
 602 }
 603
 604 /*
 605  * release a reference to a region
 606  */
 607 static void put_nommu_region(struct vm_region *region)
 608 {
 609         down_write(&nommu_region_sem);
 610         __put_nommu_region(region);
 611 }
 612
 613 /*
 614  * add a VMA into a process's mm_struct in the appropriate place in the list
 615  * and tree and add to the address space's page tree also if not an anonymous
 616  * page
 617  * - should be called with mm->mmap_sem held writelocked
 618  */
 619 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
 620 {
 621         struct vm_area_struct *pvma, **pp;
 622         struct address_space *mapping;
 623         struct rb_node **p, *parent;
 624
 625         kenter(",%p", vma);
 626
 627         BUG_ON(!vma->vm_region);
 628
 629         mm->map_count++;
 630         vma->vm_mm = mm;
 631
 632         /* add the VMA to the mapping */
 633         if (vma->vm_file) {
 634                 mapping = vma->vm_file->f_mapping;
 635
 636                 flush_dcache_mmap_lock(mapping);
 637                 vma_prio_tree_insert(vma, &mapping->i_mmap);
 638                 flush_dcache_mmap_unlock(mapping);
 639         }
 640
 641         /* add the VMA to the tree */
 642         parent = NULL;
 643         p = &mm->mm_rb.rb_node;
 644         while (*p) {
 645                 parent = *p;
 646                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
 647
 648                 /* sort by: start addr, end addr, VMA struct addr in that order
 649                  * (the latter is necessary as we may get identical VMAs) */
 650                 if (vma->vm_start < pvma->vm_start)
 651                         p = &(*p)->rb_left;
 652                 else if (vma->vm_start > pvma->vm_start)
 653                         p = &(*p)->rb_right;
 654                 else if (vma->vm_end < pvma->vm_end)
 655                         p = &(*p)->rb_left;
 656                 else if (vma->vm_end > pvma->vm_end)
 657                         p = &(*p)->rb_right;
 658                 else if (vma < pvma)
 659                         p = &(*p)->rb_left;
 660                 else if (vma > pvma)
 661                         p = &(*p)->rb_right;
 662                 else
 663                         BUG();
 664         }
 665
 666         rb_link_node(&vma->vm_rb, parent, p);
 667         rb_insert_color(&vma->vm_rb, &mm->mm_rb);
 668
 669         /* add VMA to the VMA list also */
 670         for (pp = &mm->mmap; (pvma = *pp); pp = &(*pp)->vm_next) {
 671                 if (pvma->vm_start > vma->vm_start)
 672                         break;
 673                 if (pvma->vm_start < vma->vm_start)
 674                         continue;
 675                 if (pvma->vm_end < vma->vm_end)
 676                         break;
 677         }
 678
 679         vma->vm_next = *pp;
 680         *pp = vma;
 681 }
 682
 683 /*
 684  * delete a VMA from its owning mm_struct and address space
 685  */
 686 static void delete_vma_from_mm(struct vm_area_struct *vma)
 687 {
 688         struct vm_area_struct **pp;
 689         struct address_space *mapping;
 690         struct mm_struct *mm = vma->vm_mm;
 691
 692         kenter("%p", vma);
 693
 694         mm->map_count--;
 695         if (mm->mmap_cache == vma)
 696                 mm->mmap_cache = NULL;
 697
 698         /* remove the VMA from the mapping */
 699         if (vma->vm_file) {
 700                 mapping = vma->vm_file->f_mapping;
 701
 702                 flush_dcache_mmap_lock(mapping);
 703                 vma_prio_tree_remove(vma, &mapping->i_mmap);
 704                 flush_dcache_mmap_unlock(mapping);
 705         }
 706
 707         /* remove from the MM's tree and list */
 708         rb_erase(&vma->vm_rb, &mm->mm_rb);
 709         for (pp = &mm->mmap; *pp; pp = &(*pp)->vm_next) {
 710                 if (*pp == vma) {
 711                         *pp = vma->vm_next;
 712                         break;
 713                 }
 714         }
 715
 716         vma->vm_mm = NULL;
 717 }
 718
 719 /*
 720  * destroy a VMA record
 721  */
 722 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
 723 {
 724         kenter("%p", vma);
 725         if (vma->vm_ops && vma->vm_ops->close)
 726                 vma->vm_ops->close(vma);
 727         if (vma->vm_file) {
 728                 fput(vma->vm_file);
 729                 if (vma->vm_flags & VM_EXECUTABLE)
 730                         removed_exe_file_vma(mm);
 731         }
 732         put_nommu_region(vma->vm_region);
 733         kmem_cache_free(vm_area_cachep, vma);
 734 }
 735
 736 /*
 737  * look up the first VMA in which addr resides, NULL if none
 738  * - should be called with mm->mmap_sem at least held readlocked
 739  */
 740 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
 741 {
 742         struct vm_area_struct *vma;
 743         struct rb_node *n = mm->mm_rb.rb_node;
 744
 745         /* check the cache first */
 746         vma = mm->mmap_cache;
 747         if (vma && vma->vm_start <= addr && vma->vm_end > addr)
 748                 return vma;
 749
 750         /* trawl the tree (there may be multiple mappings in which addr
 751          * resides) */
 752         for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
 753                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
 754                 if (vma->vm_start > addr)
 755                         return NULL;
 756                 if (vma->vm_end > addr) {
 757                         mm->mmap_cache = vma;
 758                         return vma;
 759                 }
 760         }
 761
 762         return NULL;
 763 }
 764 EXPORT_SYMBOL(find_vma);
 765
 766 /*
 767  * find a VMA
 768  * - we don't extend stack VMAs under NOMMU conditions
 769  */
 770 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
 771 {
 772         return find_vma(mm, addr);
 773 }
 774
 775 /*
 776  * expand a stack to a given address
 777  * - not supported under NOMMU conditions
 778  */
 779 int expand_stack(struct vm_area_struct *vma, unsigned long address)
 780 {
 781         return -ENOMEM;
 782 }
 783
 784 /*
 785  * look up the first VMA exactly that exactly matches addr
 786  * - should be called with mm->mmap_sem at least held readlocked
 787  */
 788 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
 789                                              unsigned long addr,
 790                                              unsigned long len)
 791 {
 792         struct vm_area_struct *vma;
 793         struct rb_node *n = mm->mm_rb.rb_node;
 794         unsigned long end = addr + len;
 795
 796         /* check the cache first */
 797         vma = mm->mmap_cache;
 798         if (vma && vma->vm_start == addr && vma->vm_end == end)
 799                 return vma;
 800
 801         /* trawl the tree (there may be multiple mappings in which addr
 802          * resides) */
 803         for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
 804                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
 805                 if (vma->vm_start < addr)
 806                         continue;
 807                 if (vma->vm_start > addr)
 808                         return NULL;
 809                 if (vma->vm_end == end) {
 810                         mm->mmap_cache = vma;
 811                         return vma;
 812                 }
 813         }
 814
 815         return NULL;
 816 }
 817
 818 /*
 819  * determine whether a mapping should be permitted and, if so, what sort of
 820  * mapping we're capable of supporting
 821  */
 822 static int validate_mmap_request(struct file *file,
 823                                  unsigned long addr,
 824                                  unsigned long len,
 825                                  unsigned long prot,
 826                                  unsigned long flags,
 827                                  unsigned long pgoff,
 828                                  unsigned long *_capabilities)
 829 {
 830         unsigned long capabilities, rlen;
 831         unsigned long reqprot = prot;
 832         int ret;
 833
 834         /* do the simple checks first */
 835         if (flags & MAP_FIXED || addr) {
 836                 printk(KERN_DEBUG
 837                        "%d: Can't do fixed-address/overlay mmap of RAM\n",
 838                        current->pid);
 839                 return -EINVAL;
 840         }
 841
 842         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
 843             (flags & MAP_TYPE) != MAP_SHARED)
 844                 return -EINVAL;
 845
 846         if (!len)
 847                 return -EINVAL;
 848
 849         /* Careful about overflows.. */
 850         rlen = PAGE_ALIGN(len);
 851         if (!rlen || rlen > TASK_SIZE)
 852                 return -ENOMEM;
 853
 854         /* offset overflow? */
 855         if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
 856                 return -EOVERFLOW;
 857
 858         if (file) {
 859                 /* validate file mapping requests */
 860                 struct address_space *mapping;
 861
 862                 /* files must support mmap */
 863                 if (!file->f_op || !file->f_op->mmap)
 864                         return -ENODEV;
 865
 866                 /* work out if what we've got could possibly be shared
 867                  * - we support chardevs that provide their own "memory"
 868                  * - we support files/blockdevs that are memory backed
 869                  */
 870                 mapping = file->f_mapping;
 871                 if (!mapping)
 872                         mapping = file->f_path.dentry->d_inode->i_mapping;
 873
 874                 capabilities = 0;
 875                 if (mapping && mapping->backing_dev_info)
 876                         capabilities = mapping->backing_dev_info->capabilities;
 877
 878                 if (!capabilities) {
 879                         /* no explicit capabilities set, so assume some
 880                          * defaults */
 881                         switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
 882                         case S_IFREG:
 883                         case S_IFBLK:
 884                                 capabilities = BDI_CAP_MAP_COPY;
 885                                 break;
 886
 887                         case S_IFCHR:
 888                                 capabilities =
 889                                         BDI_CAP_MAP_DIRECT |
 890                                         BDI_CAP_READ_MAP |
 891                                         BDI_CAP_WRITE_MAP;
 892                                 break;
 893
 894                         default:
 895                                 return -EINVAL;
 896                         }
 897                 }
 898
 899                 /* eliminate any capabilities that we can't support on this
 900                  * device */
 901                 if (!file->f_op->get_unmapped_area)
 902                         capabilities &= ~BDI_CAP_MAP_DIRECT;
 903                 if (!file->f_op->read)
 904                         capabilities &= ~BDI_CAP_MAP_COPY;
 905
 906                 if (flags & MAP_SHARED) {
 907                         /* do checks for writing, appending and locking */
 908                         if ((prot & PROT_WRITE) &&
 909                             !(file->f_mode & FMODE_WRITE))
 910                                 return -EACCES;
 911
 912                         if (IS_APPEND(file->f_path.dentry->d_inode) &&
 913                             (file->f_mode & FMODE_WRITE))
 914                                 return -EACCES;
 915
 916                         if (locks_verify_locked(file->f_path.dentry->d_inode))
 917                                 return -EAGAIN;
 918
 919                         if (!(capabilities & BDI_CAP_MAP_DIRECT))
 920                                 return -ENODEV;
 921
 922                         if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
 923                             ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
 924                             ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
 925                             ) {
 926                                 printk("MAP_SHARED not completely supported on !MMU\n");
 927                                 return -EINVAL;
 928                         }
 929
 930                         /* we mustn't privatise shared mappings */
 931                         capabilities &= ~BDI_CAP_MAP_COPY;
 932                 }
 933                 else {
 934                         /* we're going to read the file into private memory we
 935                          * allocate */
 936                         if (!(capabilities & BDI_CAP_MAP_COPY))
 937                                 return -ENODEV;
 938
 939                         /* we don't permit a private writable mapping to be
 940                          * shared with the backing device */
 941                         if (prot & PROT_WRITE)
 942                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
 943                 }
 944
 945                 /* handle executable mappings and implied executable
 946                  * mappings */
 947                 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
 948                         if (prot & PROT_EXEC)
 949                                 return -EPERM;
 950                 }
 951                 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
 952                         /* handle implication of PROT_EXEC by PROT_READ */
 953                         if (current->personality & READ_IMPLIES_EXEC) {
 954                                 if (capabilities & BDI_CAP_EXEC_MAP)
 955                                         prot |= PROT_EXEC;
 956                         }
 957                 }
 958                 else if ((prot & PROT_READ) &&
 959                          (prot & PROT_EXEC) &&
 960                          !(capabilities & BDI_CAP_EXEC_MAP)
 961                          ) {
 962                         /* backing file is not executable, try to copy */
 963                         capabilities &= ~BDI_CAP_MAP_DIRECT;
 964                 }
 965         }
 966         else {
 967                 /* anonymous mappings are always memory backed and can be
 968                  * privately mapped
 969                  */
 970                 capabilities = BDI_CAP_MAP_COPY;
 971
 972                 /* handle PROT_EXEC implication by PROT_READ */
 973                 if ((prot & PROT_READ) &&
 974                     (current->personality & READ_IMPLIES_EXEC))
 975                         prot |= PROT_EXEC;
 976         }
 977
 978         /* allow the security API to have its say */
 979         ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
 980         if (ret < 0)
 981                 return ret;
 982
 983         /* looks okay */
 984         *_capabilities = capabilities;
 985         return 0;
 986 }
 987
 988 /*
 989  * we've determined that we can make the mapping, now translate what we
 990  * now know into VMA flags
 991  */
 992 static unsigned long determine_vm_flags(struct file *file,
 993                                         unsigned long prot,
 994                                         unsigned long flags,
 995                                         unsigned long capabilities)
 996 {
 997         unsigned long vm_flags;
 998
 999         vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
1000         vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1001         /* vm_flags |= mm->def_flags; */
1002
1003         if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
1004                 /* attempt to share read-only copies of mapped file chunks */
1005                 if (file && !(prot & PROT_WRITE))
1006                         vm_flags |= VM_MAYSHARE;
1007         }
1008         else {
1009                 /* overlay a shareable mapping on the backing device or inode
1010                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1011                  * romfs/cramfs */
1012                 if (flags & MAP_SHARED)
1013                         vm_flags |= VM_MAYSHARE | VM_SHARED;
1014                 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
1015                         vm_flags |= VM_MAYSHARE;
1016         }
1017
1018         /* refuse to let anyone share private mappings with this process if
1019          * it's being traced - otherwise breakpoints set in it may interfere
1020          * with another untraced process
1021          */
1022         if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current))
1023                 vm_flags &= ~VM_MAYSHARE;
1024
1025         return vm_flags;
1026 }
1027
1028 /*
1029  * set up a shared mapping on a file (the driver or filesystem provides and
1030  * pins the storage)
1031  */
1032 static int do_mmap_shared_file(struct vm_area_struct *vma)
1033 {
1034         int ret;
1035
1036         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1037         if (ret == 0) {
1038                 vma->vm_region->vm_top = vma->vm_region->vm_end;
1039                 return ret;
1040         }
1041         if (ret != -ENOSYS)
1042                 return ret;
1043
1044         /* getting an ENOSYS error indicates that direct mmap isn't
1045          * possible (as opposed to tried but failed) so we'll fall
1046          * through to making a private copy of the data and mapping
1047          * that if we can */
1048         return -ENODEV;
1049 }
1050
1051 /*
1052  * set up a private mapping or an anonymous shared mapping
1053  */
1054 static int do_mmap_private(struct vm_area_struct *vma,
1055                            struct vm_region *region,
1056                            unsigned long len)
1057 {
1058         struct page *pages;
1059         unsigned long total, point, n, rlen;
1060         void *base;
1061         int ret, order;
1062
1063         /* invoke the file's mapping function so that it can keep track of
1064          * shared mappings on devices or memory
1065          * - VM_MAYSHARE will be set if it may attempt to share
1066          */
1067         if (vma->vm_file) {
1068                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1069                 if (ret == 0) {
1070                         /* shouldn't return success if we're not sharing */
1071                         BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1072                         vma->vm_region->vm_top = vma->vm_region->vm_end;
1073                         return ret;
1074                 }
1075                 if (ret != -ENOSYS)
1076                         return ret;
1077
1078                 /* getting an ENOSYS error indicates that direct mmap isn't
1079                  * possible (as opposed to tried but failed) so we'll try to
1080                  * make a private copy of the data and map that instead */
1081         }
1082
1083         rlen = PAGE_ALIGN(len);
1084
1085         /* allocate some memory to hold the mapping
1086          * - note that this may not return a page-aligned address if the object
1087          *   we're allocating is smaller than a page
1088          */
1089         order = get_order(rlen);
1090         kdebug("alloc order %d for %lx", order, len);
1091
1092         pages = alloc_pages(GFP_KERNEL, order);
1093         if (!pages)
1094                 goto enomem;
1095
1096         total = 1 << order;
1097         atomic_long_add(total, &mmap_pages_allocated);
1098
1099         point = rlen >> PAGE_SHIFT;
1100
1101         /* we allocated a power-of-2 sized page set, so we may want to trim off
1102          * the excess */
1103         if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
1104                 while (total > point) {
1105                         order = ilog2(total - point);
1106                         n = 1 << order;
1107                         kdebug("shave %lu/%lu @%lu", n, total - point, total);
1108                         atomic_long_sub(n, &mmap_pages_allocated);
1109                         total -= n;
1110                         set_page_refcounted(pages + total);
1111                         __free_pages(pages + total, order);
1112                 }
1113         }
1114
1115         for (point = 1; point < total; point++)
1116                 set_page_refcounted(&pages[point]);
1117
1118         base = page_address(pages);
1119         region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1120         region->vm_start = (unsigned long) base;
1121         region->vm_end   = region->vm_start + rlen;
1122         region->vm_top   = region->vm_start + (total << PAGE_SHIFT);
1123
1124         vma->vm_start = region->vm_start;
1125         vma->vm_end   = region->vm_start + len;
1126
1127         if (vma->vm_file) {
1128                 /* read the contents of a file into the copy */
1129                 mm_segment_t old_fs;
1130                 loff_t fpos;
1131
1132                 fpos = vma->vm_pgoff;
1133                 fpos <<= PAGE_SHIFT;
1134
1135                 old_fs = get_fs();
1136                 set_fs(KERNEL_DS);
1137                 ret = vma->vm_file->f_op->read(vma->vm_file, base, rlen, &fpos);
1138                 set_fs(old_fs);
1139
1140                 if (ret < 0)
1141                         goto error_free;
1142
1143                 /* clear the last little bit */
1144                 if (ret < rlen)
1145                         memset(base + ret, 0, rlen - ret);
1146
1147         } else {
1148                 /* if it's an anonymous mapping, then just clear it */
1149                 memset(base, 0, rlen);
1150         }
1151
1152         return 0;
1153
1154 error_free:
1155         free_page_series(region->vm_start, region->vm_end);
1156         region->vm_start = vma->vm_start = 0;
1157         region->vm_end   = vma->vm_end = 0;
1158         region->vm_top   = 0;
1159         return ret;
1160
1161 enomem:
1162         printk("Allocation of length %lu from process %d (%s) failed\n",
1163                len, current->pid, current->comm);
1164         show_free_areas();
1165         return -ENOMEM;
1166 }
1167
1168 /*
1169  * handle mapping creation for uClinux
1170  */
1171 unsigned long do_mmap_pgoff(struct file *file,
1172                             unsigned long addr,
1173                             unsigned long len,
1174                             unsigned long prot,
1175                             unsigned long flags,
1176                             unsigned long pgoff)
1177 {
1178         struct vm_area_struct *vma;
1179         struct vm_region *region;
1180         struct rb_node *rb;
1181         unsigned long capabilities, vm_flags, result;
1182         int ret;
1183
1184         kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1185
1186         if (!(flags & MAP_FIXED))
1187                 addr = round_hint_to_min(addr);
1188
1189         /* decide whether we should attempt the mapping, and if so what sort of
1190          * mapping */
1191         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1192                                     &capabilities);
1193         if (ret < 0) {
1194                 kleave(" = %d [val]", ret);
1195                 return ret;
1196         }
1197
1198         /* we've determined that we can make the mapping, now translate what we
1199          * now know into VMA flags */
1200         vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1201
1202         /* we're going to need to record the mapping */
1203         region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1204         if (!region)
1205                 goto error_getting_region;
1206
1207         vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1208         if (!vma)
1209                 goto error_getting_vma;
1210
1211         atomic_set(&region->vm_usage, 1);
1212         region->vm_flags = vm_flags;
1213         region->vm_pgoff = pgoff;
1214
1215         INIT_LIST_HEAD(&vma->anon_vma_node);
1216         vma->vm_flags = vm_flags;
1217         vma->vm_pgoff = pgoff;
1218
1219         if (file) {
1220                 region->vm_file = file;
1221                 get_file(file);
1222                 vma->vm_file = file;
1223                 get_file(file);
1224                 if (vm_flags & VM_EXECUTABLE) {
1225                         added_exe_file_vma(current->mm);
1226                         vma->vm_mm = current->mm;
1227                 }
1228         }
1229
1230         down_write(&nommu_region_sem);
1231
1232         /* if we want to share, we need to check for regions created by other
1233          * mmap() calls that overlap with our proposed mapping
1234          * - we can only share with a superset match on most regular files
1235          * - shared mappings on character devices and memory backed files are
1236          *   permitted to overlap inexactly as far as we are concerned for in
1237          *   these cases, sharing is handled in the driver or filesystem rather
1238          *   than here
1239          */
1240         if (vm_flags & VM_MAYSHARE) {
1241                 struct vm_region *pregion;
1242                 unsigned long pglen, rpglen, pgend, rpgend, start;
1243
1244                 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1245                 pgend = pgoff + pglen;
1246
1247                 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1248                         pregion = rb_entry(rb, struct vm_region, vm_rb);
1249
1250                         if (!(pregion->vm_flags & VM_MAYSHARE))
1251                                 continue;
1252
1253                         /* search for overlapping mappings on the same file */
1254                         if (pregion->vm_file->f_path.dentry->d_inode !=
1255                             file->f_path.dentry->d_inode)
1256                                 continue;
1257
1258                         if (pregion->vm_pgoff >= pgend)
1259                                 continue;
1260
1261                         rpglen = pregion->vm_end - pregion->vm_start;
1262                         rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1263                         rpgend = pregion->vm_pgoff + rpglen;
1264                         if (pgoff >= rpgend)
1265                                 continue;
1266
1267                         /* handle inexactly overlapping matches between
1268                          * mappings */
1269                         if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1270                             !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1271                                 /* new mapping is not a subset of the region */
1272                                 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1273                                         goto sharing_violation;
1274                                 continue;
1275                         }
1276
1277                         /* we've found a region we can share */
1278                         atomic_inc(&pregion->vm_usage);
1279                         vma->vm_region = pregion;
1280                         start = pregion->vm_start;
1281                         start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1282                         vma->vm_start = start;
1283                         vma->vm_end = start + len;
1284
1285                         if (pregion->vm_flags & VM_MAPPED_COPY) {
1286                                 kdebug("share copy");
1287                                 vma->vm_flags |= VM_MAPPED_COPY;
1288                         } else {
1289                                 kdebug("share mmap");
1290                                 ret = do_mmap_shared_file(vma);
1291                                 if (ret < 0) {
1292                                         vma->vm_region = NULL;
1293                                         vma->vm_start = 0;
1294                                         vma->vm_end = 0;
1295                                         atomic_dec(&pregion->vm_usage);
1296                                         pregion = NULL;
1297                                         goto error_just_free;
1298                                 }
1299                         }
1300                         fput(region->vm_file);
1301                         kmem_cache_free(vm_region_jar, region);
1302                         region = pregion;
1303                         result = start;
1304                         goto share;
1305                 }
1306
1307                 /* obtain the address at which to make a shared mapping
1308                  * - this is the hook for quasi-memory character devices to
1309                  *   tell us the location of a shared mapping
1310                  */
1311                 if (file && file->f_op->get_unmapped_area) {
1312                         addr = file->f_op->get_unmapped_area(file, addr, len,
1313                                                              pgoff, flags);
1314                         if (IS_ERR((void *) addr)) {
1315                                 ret = addr;
1316                                 if (ret != (unsigned long) -ENOSYS)
1317                                         goto error_just_free;
1318
1319                                 /* the driver refused to tell us where to site
1320                                  * the mapping so we'll have to attempt to copy
1321                                  * it */
1322                                 ret = (unsigned long) -ENODEV;
1323                                 if (!(capabilities & BDI_CAP_MAP_COPY))
1324                                         goto error_just_free;
1325
1326                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
1327                         } else {
1328                                 vma->vm_start = region->vm_start = addr;
1329                                 vma->vm_end = region->vm_end = addr + len;
1330                         }
1331                 }
1332         }
1333
1334         vma->vm_region = region;
1335
1336         /* set up the mapping */
1337         if (file && vma->vm_flags & VM_SHARED)
1338                 ret = do_mmap_shared_file(vma);
1339         else
1340                 ret = do_mmap_private(vma, region, len);
1341         if (ret < 0)
1342                 goto error_put_region;
1343
1344         add_nommu_region(region);
1345
1346         /* okay... we have a mapping; now we have to register it */
1347         result = vma->vm_start;
1348
1349         current->mm->total_vm += len >> PAGE_SHIFT;
1350
1351 share:
1352         add_vma_to_mm(current->mm, vma);
1353
1354         up_write(&nommu_region_sem);
1355
1356         if (prot & PROT_EXEC)
1357                 flush_icache_range(result, result + len);
1358
1359         kleave(" = %lx", result);
1360         return result;
1361
1362 error_put_region:
1363         __put_nommu_region(region);
1364         if (vma) {
1365                 if (vma->vm_file) {
1366                         fput(vma->vm_file);
1367                         if (vma->vm_flags & VM_EXECUTABLE)
1368                                 removed_exe_file_vma(vma->vm_mm);
1369                 }
1370                 kmem_cache_free(vm_area_cachep, vma);
1371         }
1372         kleave(" = %d [pr]", ret);
1373         return ret;
1374
1375 error_just_free:
1376         up_write(&nommu_region_sem);
1377 error:
1378         fput(region->vm_file);
1379         kmem_cache_free(vm_region_jar, region);
1380         fput(vma->vm_file);
1381         if (vma->vm_flags & VM_EXECUTABLE)
1382                 removed_exe_file_vma(vma->vm_mm);
1383         kmem_cache_free(vm_area_cachep, vma);
1384         kleave(" = %d", ret);
1385         return ret;
1386
1387 sharing_violation:
1388         up_write(&nommu_region_sem);
1389         printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1390         ret = -EINVAL;
1391         goto error;
1392
1393 error_getting_vma:
1394         kmem_cache_free(vm_region_jar, region);
1395         printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1396                " from process %d failed\n",
1397                len, current->pid);
1398         show_free_areas();
1399         return -ENOMEM;
1400
1401 error_getting_region:
1402         printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1403                " from process %d failed\n",
1404                len, current->pid);
1405         show_free_areas();
1406         return -ENOMEM;
1407 }
1408 EXPORT_SYMBOL(do_mmap_pgoff);
1409
1410 /*
1411  * split a vma into two pieces at address 'addr', a new vma is allocated either
1412  * for the first part or the tail.
1413  */
1414 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1415               unsigned long addr, int new_below)
1416 {
1417         struct vm_area_struct *new;
1418         struct vm_region *region;
1419         unsigned long npages;
1420
1421         kenter("");
1422
1423         /* we're only permitted to split anonymous regions that have a single
1424          * owner */
1425         if (vma->vm_file ||
1426             atomic_read(&vma->vm_region->vm_usage) != 1)
1427                 return -ENOMEM;
1428
1429         if (mm->map_count >= sysctl_max_map_count)
1430                 return -ENOMEM;
1431
1432         region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1433         if (!region)
1434                 return -ENOMEM;
1435
1436         new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1437         if (!new) {
1438                 kmem_cache_free(vm_region_jar, region);
1439                 return -ENOMEM;
1440         }
1441
1442         /* most fields are the same, copy all, and then fixup */
1443         *new = *vma;
1444         *region = *vma->vm_region;
1445         new->vm_region = region;
1446
1447         npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1448
1449         if (new_below) {
1450                 region->vm_top = region->vm_end = new->vm_end = addr;
1451         } else {
1452                 region->vm_start = new->vm_start = addr;
1453                 region->vm_pgoff = new->vm_pgoff += npages;
1454         }
1455
1456         if (new->vm_ops && new->vm_ops->open)
1457                 new->vm_ops->open(new);
1458
1459         delete_vma_from_mm(vma);
1460         down_write(&nommu_region_sem);
1461         delete_nommu_region(vma->vm_region);
1462         if (new_below) {
1463                 vma->vm_region->vm_start = vma->vm_start = addr;
1464                 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1465         } else {
1466                 vma->vm_region->vm_end = vma->vm_end = addr;
1467                 vma->vm_region->vm_top = addr;
1468         }
1469         add_nommu_region(vma->vm_region);
1470         add_nommu_region(new->vm_region);
1471         up_write(&nommu_region_sem);
1472         add_vma_to_mm(mm, vma);
1473         add_vma_to_mm(mm, new);
1474         return 0;
1475 }
1476
1477 /*
1478  * shrink a VMA by removing the specified chunk from either the beginning or
1479  * the end
1480  */
1481 static int shrink_vma(struct mm_struct *mm,
1482                       struct vm_area_struct *vma,
1483                       unsigned long from, unsigned long to)
1484 {
1485         struct vm_region *region;
1486
1487         kenter("");
1488
1489         /* adjust the VMA's pointers, which may reposition it in the MM's tree
1490          * and list */
1491         delete_vma_from_mm(vma);
1492         if (from > vma->vm_start)
1493                 vma->vm_end = from;
1494         else
1495                 vma->vm_start = to;
1496         add_vma_to_mm(mm, vma);
1497
1498         /* cut the backing region down to size */
1499         region = vma->vm_region;
1500         BUG_ON(atomic_read(&region->vm_usage) != 1);
1501
1502         down_write(&nommu_region_sem);
1503         delete_nommu_region(region);
1504         if (from > region->vm_start) {
1505                 to = region->vm_top;
1506                 region->vm_top = region->vm_end = from;
1507         } else {
1508                 region->vm_start = to;
1509         }
1510         add_nommu_region(region);
1511         up_write(&nommu_region_sem);
1512
1513         free_page_series(from, to);
1514         return 0;
1515 }
1516
1517 /*
1518  * release a mapping
1519  * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1520  *   VMA, though it need not cover the whole VMA
1521  */
1522 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1523 {
1524         struct vm_area_struct *vma;
1525         struct rb_node *rb;
1526         unsigned long end = start + len;
1527         int ret;
1528
1529         kenter(",%lx,%zx", start, len);
1530
1531         if (len == 0)
1532                 return -EINVAL;
1533
1534         /* find the first potentially overlapping VMA */
1535         vma = find_vma(mm, start);
1536         if (!vma) {
1537                 static int limit = 0;
1538                 if (limit < 5) {
1539                         printk(KERN_WARNING
1540                                "munmap of memory not mmapped by process %d"
1541                                " (%s): 0x%lx-0x%lx\n",
1542                                current->pid, current->comm,
1543                                start, start + len - 1);
1544                         limit++;
1545                 }
1546                 return -EINVAL;
1547         }
1548
1549         /* we're allowed to split an anonymous VMA but not a file-backed one */
1550         if (vma->vm_file) {
1551                 do {
1552                         if (start > vma->vm_start) {
1553                                 kleave(" = -EINVAL [miss]");
1554                                 return -EINVAL;
1555                         }
1556                         if (end == vma->vm_end)
1557                                 goto erase_whole_vma;
1558                         rb = rb_next(&vma->vm_rb);
1559                         vma = rb_entry(rb, struct vm_area_struct, vm_rb);
1560                 } while (rb);
1561                 kleave(" = -EINVAL [split file]");
1562                 return -EINVAL;
1563         } else {
1564                 /* the chunk must be a subset of the VMA found */
1565                 if (start == vma->vm_start && end == vma->vm_end)
1566                         goto erase_whole_vma;
1567                 if (start < vma->vm_start || end > vma->vm_end) {
1568                         kleave(" = -EINVAL [superset]");
1569                         return -EINVAL;
1570                 }
1571                 if (start & ~PAGE_MASK) {
1572                         kleave(" = -EINVAL [unaligned start]");
1573                         return -EINVAL;
1574                 }
1575                 if (end != vma->vm_end && end & ~PAGE_MASK) {
1576                         kleave(" = -EINVAL [unaligned split]");
1577                         return -EINVAL;
1578                 }
1579                 if (start != vma->vm_start && end != vma->vm_end) {
1580                         ret = split_vma(mm, vma, start, 1);
1581                         if (ret < 0) {
1582                                 kleave(" = %d [split]", ret);
1583                                 return ret;
1584                         }
1585                 }
1586                 return shrink_vma(mm, vma, start, end);
1587         }
1588
1589 erase_whole_vma:
1590         delete_vma_from_mm(vma);
1591         delete_vma(mm, vma);
1592         kleave(" = 0");
1593         return 0;
1594 }
1595 EXPORT_SYMBOL(do_munmap);
1596
1597 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1598 {
1599         int ret;
1600         struct mm_struct *mm = current->mm;
1601
1602         down_write(&mm->mmap_sem);
1603         ret = do_munmap(mm, addr, len);
1604         up_write(&mm->mmap_sem);
1605         return ret;
1606 }
1607
1608 /*
1609  * release all the mappings made in a process's VM space
1610  */
1611 void exit_mmap(struct mm_struct *mm)
1612 {
1613         struct vm_area_struct *vma;
1614
1615         if (!mm)
1616                 return;
1617
1618         kenter("");
1619
1620         mm->total_vm = 0;
1621
1622         while ((vma = mm->mmap)) {
1623                 mm->mmap = vma->vm_next;
1624                 delete_vma_from_mm(vma);
1625                 delete_vma(mm, vma);
1626         }
1627
1628         kleave("");
1629 }
1630
1631 unsigned long do_brk(unsigned long addr, unsigned long len)
1632 {
1633         return -ENOMEM;
1634 }
1635
1636 /*
1637  * expand (or shrink) an existing mapping, potentially moving it at the same
1638  * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1639  *
1640  * under NOMMU conditions, we only permit changing a mapping's size, and only
1641  * as long as it stays within the region allocated by do_mmap_private() and the
1642  * block is not shareable
1643  *
1644  * MREMAP_FIXED is not supported under NOMMU conditions
1645  */
1646 unsigned long do_mremap(unsigned long addr,
1647                         unsigned long old_len, unsigned long new_len,
1648                         unsigned long flags, unsigned long new_addr)
1649 {
1650         struct vm_area_struct *vma;
1651
1652         /* insanity checks first */
1653         if (old_len == 0 || new_len == 0)
1654                 return (unsigned long) -EINVAL;
1655
1656         if (addr & ~PAGE_MASK)
1657                 return -EINVAL;
1658
1659         if (flags & MREMAP_FIXED && new_addr != addr)
1660                 return (unsigned long) -EINVAL;
1661
1662         vma = find_vma_exact(current->mm, addr, old_len);
1663         if (!vma)
1664                 return (unsigned long) -EINVAL;
1665
1666         if (vma->vm_end != vma->vm_start + old_len)
1667                 return (unsigned long) -EFAULT;
1668
1669         if (vma->vm_flags & VM_MAYSHARE)
1670                 return (unsigned long) -EPERM;
1671
1672         if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1673                 return (unsigned long) -ENOMEM;
1674
1675         /* all checks complete - do it */
1676         vma->vm_end = vma->vm_start + new_len;
1677         return vma->vm_start;
1678 }
1679 EXPORT_SYMBOL(do_mremap);
1680
1681 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1682                 unsigned long, new_len, unsigned long, flags,
1683                 unsigned long, new_addr)
1684 {
1685         unsigned long ret;
1686
1687         down_write(&current->mm->mmap_sem);
1688         ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1689         up_write(&current->mm->mmap_sem);
1690         return ret;
1691 }
1692
1693 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1694                         unsigned int foll_flags)
1695 {
1696         return NULL;
1697 }
1698
1699 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1700                 unsigned long to, unsigned long size, pgprot_t prot)
1701 {
1702         vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1703         return 0;
1704 }
1705 EXPORT_SYMBOL(remap_pfn_range);
1706
1707 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1708                         unsigned long pgoff)
1709 {
1710         unsigned int size = vma->vm_end - vma->vm_start;
1711
1712         if (!(vma->vm_flags & VM_USERMAP))
1713                 return -EINVAL;
1714
1715         vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1716         vma->vm_end = vma->vm_start + size;
1717
1718         return 0;
1719 }
1720 EXPORT_SYMBOL(remap_vmalloc_range);
1721
1722 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1723 {
1724 }
1725
1726 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1727         unsigned long len, unsigned long pgoff, unsigned long flags)
1728 {
1729         return -ENOMEM;
1730 }
1731
1732 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1733 {
1734 }
1735
1736 void unmap_mapping_range(struct address_space *mapping,
1737                          loff_t const holebegin, loff_t const holelen,
1738                          int even_cows)
1739 {
1740 }
1741 EXPORT_SYMBOL(unmap_mapping_range);
1742
1743 /*
1744  * ask for an unmapped area at which to create a mapping on a file
1745  */
1746 unsigned long get_unmapped_area(struct file *file, unsigned long addr,
1747                                 unsigned long len, unsigned long pgoff,
1748                                 unsigned long flags)
1749 {
1750         unsigned long (*get_area)(struct file *, unsigned long, unsigned long,
1751                                   unsigned long, unsigned long);
1752
1753         get_area = current->mm->get_unmapped_area;
1754         if (file && file->f_op && file->f_op->get_unmapped_area)
1755                 get_area = file->f_op->get_unmapped_area;
1756
1757         if (!get_area)
1758                 return -ENOSYS;
1759
1760         return get_area(file, addr, len, pgoff, flags);
1761 }
1762 EXPORT_SYMBOL(get_unmapped_area);
1763
1764 /*
1765  * Check that a process has enough memory to allocate a new virtual
1766  * mapping. 0 means there is enough memory for the allocation to
1767  * succeed and -ENOMEM implies there is not.
1768  *
1769  * We currently support three overcommit policies, which are set via the
1770  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1771  *
1772  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1773  * Additional code 2002 Jul 20 by Robert Love.
1774  *
1775  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1776  *
1777  * Note this is a helper function intended to be used by LSMs which
1778  * wish to use this logic.
1779  */
1780 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1781 {
1782         unsigned long free, allowed;
1783
1784         vm_acct_memory(pages);
1785
1786         /*
1787          * Sometimes we want to use more memory than we have
1788          */
1789         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1790                 return 0;
1791
1792         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1793                 unsigned long n;
1794
1795                 free = global_page_state(NR_FILE_PAGES);
1796                 free += nr_swap_pages;
1797
1798                 /*
1799                  * Any slabs which are created with the
1800                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1801                  * which are reclaimable, under pressure.  The dentry
1802                  * cache and most inode caches should fall into this
1803                  */
1804                 free += global_page_state(NR_SLAB_RECLAIMABLE);
1805
1806                 /*
1807                  * Leave the last 3% for root
1808                  */
1809                 if (!cap_sys_admin)
1810                         free -= free / 32;
1811
1812                 if (free > pages)
1813                         return 0;
1814
1815                 /*
1816                  * nr_free_pages() is very expensive on large systems,
1817                  * only call if we're about to fail.
1818                  */
1819                 n = nr_free_pages();
1820
1821                 /*
1822                  * Leave reserved pages. The pages are not for anonymous pages.
1823                  */
1824                 if (n <= totalreserve_pages)
1825                         goto error;
1826                 else
1827                         n -= totalreserve_pages;
1828
1829                 /*
1830                  * Leave the last 3% for root
1831                  */
1832                 if (!cap_sys_admin)
1833                         n -= n / 32;
1834                 free += n;
1835
1836                 if (free > pages)
1837                         return 0;
1838
1839                 goto error;
1840         }
1841
1842         allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1843         /*
1844          * Leave the last 3% for root
1845          */
1846         if (!cap_sys_admin)
1847                 allowed -= allowed / 32;
1848         allowed += total_swap_pages;
1849
1850         /* Don't let a single process grow too big:
1851            leave 3% of the size of this process for other processes */
1852         if (mm)
1853                 allowed -= mm->total_vm / 32;
1854
1855         if (percpu_counter_read_positive(&vm_committed_as) < allowed)
1856                 return 0;
1857
1858 error:
1859         vm_unacct_memory(pages);
1860
1861         return -ENOMEM;
1862 }
1863
1864 int in_gate_area_no_task(unsigned long addr)
1865 {
1866         return 0;
1867 }
1868
1869 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1870 {
1871         BUG();
1872         return 0;
1873 }
1874 EXPORT_SYMBOL(filemap_fault);
1875
1876 /*
1877  * Access another process' address space.
1878  * - source/target buffer must be kernel space
1879  */
1880 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1881 {
1882         struct vm_area_struct *vma;
1883         struct mm_struct *mm;
1884
1885         if (addr + len < addr)
1886                 return 0;
1887
1888         mm = get_task_mm(tsk);
1889         if (!mm)
1890                 return 0;
1891
1892         down_read(&mm->mmap_sem);
1893
1894         /* the access must start within one of the target process's mappings */
1895         vma = find_vma(mm, addr);
1896         if (vma) {
1897                 /* don't overrun this mapping */
1898                 if (addr + len >= vma->vm_end)
1899                         len = vma->vm_end - addr;
1900
1901                 /* only read or write mappings where it is permitted */
1902                 if (write && vma->vm_flags & VM_MAYWRITE)
1903                         len -= copy_to_user((void *) addr, buf, len);
1904                 else if (!write && vma->vm_flags & VM_MAYREAD)
1905                         len -= copy_from_user(buf, (void *) addr, len);
1906                 else
1907                         len = 0;
1908         } else {
1909                 len = 0;
1910         }
1911
1912         up_read(&mm->mmap_sem);
1913         mmput(mm);
1914         return len;
1915 }