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 <asm/mmu_context.h>
  37 #include "internal.h"
  38
  39 static inline __attribute__((format(printf, 1, 2)))
  40 void no_printk(const char *fmt, ...)
  41 {
  42 }
  43
  44 #if 0
  45 #define kenter(FMT, ...) \
  46         printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
  47 #define kleave(FMT, ...) \
  48         printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
  49 #define kdebug(FMT, ...) \
  50         printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__)
  51 #else
  52 #define kenter(FMT, ...) \
  53         no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
  54 #define kleave(FMT, ...) \
  55         no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
  56 #define kdebug(FMT, ...) \
  57         no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__)
  58 #endif
  59
  60 void *high_memory;
  61 struct page *mem_map;
  62 unsigned long max_mapnr;
  63 unsigned long num_physpages;
  64 unsigned long highest_memmap_pfn;
  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 atomic_long_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 const struct vm_operations_struct generic_file_vm_ops = {
  83 };
  84
  85 /*
  86  * Return the total memory allocated for this pointer, not
  87  * just what the caller asked for.
  88  *
  89  * Doesn't have to be accurate, i.e. may have races.
  90  */
  91 unsigned int kobjsize(const void *objp)
  92 {
  93         struct page *page;
  94
  95         /*
  96          * If the object we have should not have ksize performed on it,
  97          * return size of 0
  98          */
  99         if (!objp || !virt_addr_valid(objp))
 100                 return 0;
 101
 102         page = virt_to_head_page(objp);
 103
 104         /*
 105          * If the allocator sets PageSlab, we know the pointer came from
 106          * kmalloc().
 107          */
 108         if (PageSlab(page))
 109                 return ksize(objp);
 110
 111         /*
 112          * If it's not a compound page, see if we have a matching VMA
 113          * region. This test is intentionally done in reverse order,
 114          * so if there's no VMA, we still fall through and hand back
 115          * PAGE_SIZE for 0-order pages.
 116          */
 117         if (!PageCompound(page)) {
 118                 struct vm_area_struct *vma;
 119
 120                 vma = find_vma(current->mm, (unsigned long)objp);
 121                 if (vma)
 122                         return vma->vm_end - vma->vm_start;
 123         }
 124
 125         /*
 126          * The ksize() function is only guaranteed to work for pointers
 127          * returned by kmalloc(). So handle arbitrary pointers here.
 128          */
 129         return PAGE_SIZE << compound_order(page);
 130 }
 131
 132 int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
 133                      unsigned long start, int nr_pages, unsigned int foll_flags,
 134                      struct page **pages, struct vm_area_struct **vmas)
 135 {
 136         struct vm_area_struct *vma;
 137         unsigned long vm_flags;
 138         int i;
 139
 140         /* calculate required read or write permissions.
 141          * If FOLL_FORCE is set, we only require the "MAY" flags.
 142          */
 143         vm_flags  = (foll_flags & FOLL_WRITE) ?
 144                         (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
 145         vm_flags &= (foll_flags & FOLL_FORCE) ?
 146                         (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
 147
 148         for (i = 0; i < nr_pages; i++) {
 149                 vma = find_vma(mm, start);
 150                 if (!vma)
 151                         goto finish_or_fault;
 152
 153                 /* protect what we can, including chardevs */
 154                 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
 155                     !(vm_flags & vma->vm_flags))
 156                         goto finish_or_fault;
 157
 158                 if (pages) {
 159                         pages[i] = virt_to_page(start);
 160                         if (pages[i])
 161                                 page_cache_get(pages[i]);
 162                 }
 163                 if (vmas)
 164                         vmas[i] = vma;
 165                 start = (start + PAGE_SIZE) & PAGE_MASK;
 166         }
 167
 168         return i;
 169
 170 finish_or_fault:
 171         return i ? : -EFAULT;
 172 }
 173
 174 /*
 175  * get a list of pages in an address range belonging to the specified process
 176  * and indicate the VMA that covers each page
 177  * - this is potentially dodgy as we may end incrementing the page count of a
 178  *   slab page or a secondary page from a compound page
 179  * - don't permit access to VMAs that don't support it, such as I/O mappings
 180  */
 181 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
 182         unsigned long start, int nr_pages, int write, int force,
 183         struct page **pages, struct vm_area_struct **vmas)
 184 {
 185         int flags = 0;
 186
 187         if (write)
 188                 flags |= FOLL_WRITE;
 189         if (force)
 190                 flags |= FOLL_FORCE;
 191
 192         return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas);
 193 }
 194 EXPORT_SYMBOL(get_user_pages);
 195
 196 /**
 197  * follow_pfn - look up PFN at a user virtual address
 198  * @vma: memory mapping
 199  * @address: user virtual address
 200  * @pfn: location to store found PFN
 201  *
 202  * Only IO mappings and raw PFN mappings are allowed.
 203  *
 204  * Returns zero and the pfn at @pfn on success, -ve otherwise.
 205  */
 206 int follow_pfn(struct vm_area_struct *vma, unsigned long address,
 207         unsigned long *pfn)
 208 {
 209         if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
 210                 return -EINVAL;
 211
 212         *pfn = address >> PAGE_SHIFT;
 213         return 0;
 214 }
 215 EXPORT_SYMBOL(follow_pfn);
 216
 217 DEFINE_RWLOCK(vmlist_lock);
 218 struct vm_struct *vmlist;
 219
 220 void vfree(const void *addr)
 221 {
 222         kfree(addr);
 223 }
 224 EXPORT_SYMBOL(vfree);
 225
 226 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
 227 {
 228         /*
 229          *  You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
 230          * returns only a logical address.
 231          */
 232         return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
 233 }
 234 EXPORT_SYMBOL(__vmalloc);
 235
 236 void *vmalloc_user(unsigned long size)
 237 {
 238         void *ret;
 239
 240         ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
 241                         PAGE_KERNEL);
 242         if (ret) {
 243                 struct vm_area_struct *vma;
 244
 245                 down_write(&current->mm->mmap_sem);
 246                 vma = find_vma(current->mm, (unsigned long)ret);
 247                 if (vma)
 248                         vma->vm_flags |= VM_USERMAP;
 249                 up_write(&current->mm->mmap_sem);
 250         }
 251
 252         return ret;
 253 }
 254 EXPORT_SYMBOL(vmalloc_user);
 255
 256 struct page *vmalloc_to_page(const void *addr)
 257 {
 258         return virt_to_page(addr);
 259 }
 260 EXPORT_SYMBOL(vmalloc_to_page);
 261
 262 unsigned long vmalloc_to_pfn(const void *addr)
 263 {
 264         return page_to_pfn(virt_to_page(addr));
 265 }
 266 EXPORT_SYMBOL(vmalloc_to_pfn);
 267
 268 long vread(char *buf, char *addr, unsigned long count)
 269 {
 270         memcpy(buf, addr, count);
 271         return count;
 272 }
 273
 274 long vwrite(char *buf, char *addr, unsigned long count)
 275 {
 276         /* Don't allow overflow */
 277         if ((unsigned long) addr + count < count)
 278                 count = -(unsigned long) addr;
 279
 280         memcpy(addr, buf, count);
 281         return(count);
 282 }
 283
 284 /*
 285  *      vmalloc  -  allocate virtually continguos memory
 286  *
 287  *      @size:          allocation size
 288  *
 289  *      Allocate enough pages to cover @size from the page level
 290  *      allocator and map them into continguos kernel virtual space.
 291  *
 292  *      For tight control over page level allocator and protection flags
 293  *      use __vmalloc() instead.
 294  */
 295 void *vmalloc(unsigned long size)
 296 {
 297        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
 298 }
 299 EXPORT_SYMBOL(vmalloc);
 300
 301 void *vmalloc_node(unsigned long size, int node)
 302 {
 303         return vmalloc(size);
 304 }
 305 EXPORT_SYMBOL(vmalloc_node);
 306
 307 #ifndef PAGE_KERNEL_EXEC
 308 # define PAGE_KERNEL_EXEC PAGE_KERNEL
 309 #endif
 310
 311 /**
 312  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
 313  *      @size:          allocation size
 314  *
 315  *      Kernel-internal function to allocate enough pages to cover @size
 316  *      the page level allocator and map them into contiguous and
 317  *      executable kernel virtual space.
 318  *
 319  *      For tight control over page level allocator and protection flags
 320  *      use __vmalloc() instead.
 321  */
 322
 323 void *vmalloc_exec(unsigned long size)
 324 {
 325         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
 326 }
 327
 328 /**
 329  * vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
 330  *      @size:          allocation size
 331  *
 332  *      Allocate enough 32bit PA addressable pages to cover @size from the
 333  *      page level allocator and map them into continguos kernel virtual space.
 334  */
 335 void *vmalloc_32(unsigned long size)
 336 {
 337         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
 338 }
 339 EXPORT_SYMBOL(vmalloc_32);
 340
 341 /**
 342  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
 343  *      @size:          allocation size
 344  *
 345  * The resulting memory area is 32bit addressable and zeroed so it can be
 346  * mapped to userspace without leaking data.
 347  *
 348  * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
 349  * remap_vmalloc_range() are permissible.
 350  */
 351 void *vmalloc_32_user(unsigned long size)
 352 {
 353         /*
 354          * We'll have to sort out the ZONE_DMA bits for 64-bit,
 355          * but for now this can simply use vmalloc_user() directly.
 356          */
 357         return vmalloc_user(size);
 358 }
 359 EXPORT_SYMBOL(vmalloc_32_user);
 360
 361 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
 362 {
 363         BUG();
 364         return NULL;
 365 }
 366 EXPORT_SYMBOL(vmap);
 367
 368 void vunmap(const void *addr)
 369 {
 370         BUG();
 371 }
 372 EXPORT_SYMBOL(vunmap);
 373
 374 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
 375 {
 376         BUG();
 377         return NULL;
 378 }
 379 EXPORT_SYMBOL(vm_map_ram);
 380
 381 void vm_unmap_ram(const void *mem, unsigned int count)
 382 {
 383         BUG();
 384 }
 385 EXPORT_SYMBOL(vm_unmap_ram);
 386
 387 void vm_unmap_aliases(void)
 388 {
 389 }
 390 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
 391
 392 /*
 393  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
 394  * have one.
 395  */
 396 void  __attribute__((weak)) vmalloc_sync_all(void)
 397 {
 398 }
 399
 400 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
 401                    struct page *page)
 402 {
 403         return -EINVAL;
 404 }
 405 EXPORT_SYMBOL(vm_insert_page);
 406
 407 /*
 408  *  sys_brk() for the most part doesn't need the global kernel
 409  *  lock, except when an application is doing something nasty
 410  *  like trying to un-brk an area that has already been mapped
 411  *  to a regular file.  in this case, the unmapping will need
 412  *  to invoke file system routines that need the global lock.
 413  */
 414 SYSCALL_DEFINE1(brk, unsigned long, brk)
 415 {
 416         struct mm_struct *mm = current->mm;
 417
 418         if (brk < mm->start_brk || brk > mm->context.end_brk)
 419                 return mm->brk;
 420
 421         if (mm->brk == brk)
 422                 return mm->brk;
 423
 424         /*
 425          * Always allow shrinking brk
 426          */
 427         if (brk <= mm->brk) {
 428                 mm->brk = brk;
 429                 return brk;
 430         }
 431
 432         /*
 433          * Ok, looks good - let it rip.
 434          */
 435         flush_icache_range(mm->brk, brk);
 436         return mm->brk = brk;
 437 }
 438
 439 /*
 440  * initialise the VMA and region record slabs
 441  */
 442 void __init mmap_init(void)
 443 {
 444         int ret;
 445
 446         ret = percpu_counter_init(&vm_committed_as, 0);
 447         VM_BUG_ON(ret);
 448         vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
 449 }
 450
 451 /*
 452  * validate the region tree
 453  * - the caller must hold the region lock
 454  */
 455 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
 456 static noinline void validate_nommu_regions(void)
 457 {
 458         struct vm_region *region, *last;
 459         struct rb_node *p, *lastp;
 460
 461         lastp = rb_first(&nommu_region_tree);
 462         if (!lastp)
 463                 return;
 464
 465         last = rb_entry(lastp, struct vm_region, vm_rb);
 466         BUG_ON(unlikely(last->vm_end <= last->vm_start));
 467         BUG_ON(unlikely(last->vm_top < last->vm_end));
 468
 469         while ((p = rb_next(lastp))) {
 470                 region = rb_entry(p, struct vm_region, vm_rb);
 471                 last = rb_entry(lastp, struct vm_region, vm_rb);
 472
 473                 BUG_ON(unlikely(region->vm_end <= region->vm_start));
 474                 BUG_ON(unlikely(region->vm_top < region->vm_end));
 475                 BUG_ON(unlikely(region->vm_start < last->vm_top));
 476
 477                 lastp = p;
 478         }
 479 }
 480 #else
 481 static void validate_nommu_regions(void)
 482 {
 483 }
 484 #endif
 485
 486 /*
 487  * add a region into the global tree
 488  */
 489 static void add_nommu_region(struct vm_region *region)
 490 {
 491         struct vm_region *pregion;
 492         struct rb_node **p, *parent;
 493
 494         validate_nommu_regions();
 495
 496         parent = NULL;
 497         p = &nommu_region_tree.rb_node;
 498         while (*p) {
 499                 parent = *p;
 500                 pregion = rb_entry(parent, struct vm_region, vm_rb);
 501                 if (region->vm_start < pregion->vm_start)
 502                         p = &(*p)->rb_left;
 503                 else if (region->vm_start > pregion->vm_start)
 504                         p = &(*p)->rb_right;
 505                 else if (pregion == region)
 506                         return;
 507                 else
 508                         BUG();
 509         }
 510
 511         rb_link_node(&region->vm_rb, parent, p);
 512         rb_insert_color(&region->vm_rb, &nommu_region_tree);
 513
 514         validate_nommu_regions();
 515 }
 516
 517 /*
 518  * delete a region from the global tree
 519  */
 520 static void delete_nommu_region(struct vm_region *region)
 521 {
 522         BUG_ON(!nommu_region_tree.rb_node);
 523
 524         validate_nommu_regions();
 525         rb_erase(&region->vm_rb, &nommu_region_tree);
 526         validate_nommu_regions();
 527 }
 528
 529 /*
 530  * free a contiguous series of pages
 531  */
 532 static void free_page_series(unsigned long from, unsigned long to)
 533 {
 534         for (; from < to; from += PAGE_SIZE) {
 535                 struct page *page = virt_to_page(from);
 536
 537                 kdebug("- free %lx", from);
 538                 atomic_long_dec(&mmap_pages_allocated);
 539                 if (page_count(page) != 1)
 540                         kdebug("free page %p: refcount not one: %d",
 541                                page, page_count(page));
 542                 put_page(page);
 543         }
 544 }
 545
 546 /*
 547  * release a reference to a region
 548  * - the caller must hold the region semaphore for writing, which this releases
 549  * - the region may not have been added to the tree yet, in which case vm_top
 550  *   will equal vm_start
 551  */
 552 static void __put_nommu_region(struct vm_region *region)
 553         __releases(nommu_region_sem)
 554 {
 555         kenter("%p{%d}", region, region->vm_usage);
 556
 557         BUG_ON(!nommu_region_tree.rb_node);
 558
 559         if (--region->vm_usage == 0) {
 560                 if (region->vm_top > region->vm_start)
 561                         delete_nommu_region(region);
 562                 up_write(&nommu_region_sem);
 563
 564                 if (region->vm_file)
 565                         fput(region->vm_file);
 566
 567                 /* IO memory and memory shared directly out of the pagecache
 568                  * from ramfs/tmpfs mustn't be released here */
 569                 if (region->vm_flags & VM_MAPPED_COPY) {
 570                         kdebug("free series");
 571                         free_page_series(region->vm_start, region->vm_top);
 572                 }
 573                 kmem_cache_free(vm_region_jar, region);
 574         } else {
 575                 up_write(&nommu_region_sem);
 576         }
 577 }
 578
 579 /*
 580  * release a reference to a region
 581  */
 582 static void put_nommu_region(struct vm_region *region)
 583 {
 584         down_write(&nommu_region_sem);
 585         __put_nommu_region(region);
 586 }
 587
 588 /*
 589  * update protection on a vma
 590  */
 591 static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
 592 {
 593 #ifdef CONFIG_MPU
 594         struct mm_struct *mm = vma->vm_mm;
 595         long start = vma->vm_start & PAGE_MASK;
 596         while (start < vma->vm_end) {
 597                 protect_page(mm, start, flags);
 598                 start += PAGE_SIZE;
 599         }
 600         update_protections(mm);
 601 #endif
 602 }
 603
 604 /*
 605  * add a VMA into a process's mm_struct in the appropriate place in the list
 606  * and tree and add to the address space's page tree also if not an anonymous
 607  * page
 608  * - should be called with mm->mmap_sem held writelocked
 609  */
 610 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
 611 {
 612         struct vm_area_struct *pvma, **pp;
 613         struct address_space *mapping;
 614         struct rb_node **p, *parent;
 615
 616         kenter(",%p", vma);
 617
 618         BUG_ON(!vma->vm_region);
 619
 620         mm->map_count++;
 621         vma->vm_mm = mm;
 622
 623         protect_vma(vma, vma->vm_flags);
 624
 625         /* add the VMA to the mapping */
 626         if (vma->vm_file) {
 627                 mapping = vma->vm_file->f_mapping;
 628
 629                 flush_dcache_mmap_lock(mapping);
 630                 vma_prio_tree_insert(vma, &mapping->i_mmap);
 631                 flush_dcache_mmap_unlock(mapping);
 632         }
 633
 634         /* add the VMA to the tree */
 635         parent = NULL;
 636         p = &mm->mm_rb.rb_node;
 637         while (*p) {
 638                 parent = *p;
 639                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
 640
 641                 /* sort by: start addr, end addr, VMA struct addr in that order
 642                  * (the latter is necessary as we may get identical VMAs) */
 643                 if (vma->vm_start < pvma->vm_start)
 644                         p = &(*p)->rb_left;
 645                 else if (vma->vm_start > pvma->vm_start)
 646                         p = &(*p)->rb_right;
 647                 else if (vma->vm_end < pvma->vm_end)
 648                         p = &(*p)->rb_left;
 649                 else if (vma->vm_end > pvma->vm_end)
 650                         p = &(*p)->rb_right;
 651                 else if (vma < pvma)
 652                         p = &(*p)->rb_left;
 653                 else if (vma > pvma)
 654                         p = &(*p)->rb_right;
 655                 else
 656                         BUG();
 657         }
 658
 659         rb_link_node(&vma->vm_rb, parent, p);
 660         rb_insert_color(&vma->vm_rb, &mm->mm_rb);
 661
 662         /* add VMA to the VMA list also */
 663         for (pp = &mm->mmap; (pvma = *pp); pp = &(*pp)->vm_next) {
 664                 if (pvma->vm_start > vma->vm_start)
 665                         break;
 666                 if (pvma->vm_start < vma->vm_start)
 667                         continue;
 668                 if (pvma->vm_end < vma->vm_end)
 669                         break;
 670         }
 671
 672         vma->vm_next = *pp;
 673         *pp = vma;
 674 }
 675
 676 /*
 677  * delete a VMA from its owning mm_struct and address space
 678  */
 679 static void delete_vma_from_mm(struct vm_area_struct *vma)
 680 {
 681         struct vm_area_struct **pp;
 682         struct address_space *mapping;
 683         struct mm_struct *mm = vma->vm_mm;
 684
 685         kenter("%p", vma);
 686
 687         protect_vma(vma, 0);
 688
 689         mm->map_count--;
 690         if (mm->mmap_cache == vma)
 691                 mm->mmap_cache = NULL;
 692
 693         /* remove the VMA from the mapping */
 694         if (vma->vm_file) {
 695                 mapping = vma->vm_file->f_mapping;
 696
 697                 flush_dcache_mmap_lock(mapping);
 698                 vma_prio_tree_remove(vma, &mapping->i_mmap);
 699                 flush_dcache_mmap_unlock(mapping);
 700         }
 701
 702         /* remove from the MM's tree and list */
 703         rb_erase(&vma->vm_rb, &mm->mm_rb);
 704         for (pp = &mm->mmap; *pp; pp = &(*pp)->vm_next) {
 705                 if (*pp == vma) {
 706                         *pp = vma->vm_next;
 707                         break;
 708                 }
 709         }
 710
 711         vma->vm_mm = NULL;
 712 }
 713
 714 /*
 715  * destroy a VMA record
 716  */
 717 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
 718 {
 719         kenter("%p", vma);
 720         if (vma->vm_ops && vma->vm_ops->close)
 721                 vma->vm_ops->close(vma);
 722         if (vma->vm_file) {
 723                 fput(vma->vm_file);
 724                 if (vma->vm_flags & VM_EXECUTABLE)
 725                         removed_exe_file_vma(mm);
 726         }
 727         put_nommu_region(vma->vm_region);
 728         kmem_cache_free(vm_area_cachep, vma);
 729 }
 730
 731 /*
 732  * look up the first VMA in which addr resides, NULL if none
 733  * - should be called with mm->mmap_sem at least held readlocked
 734  */
 735 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
 736 {
 737         struct vm_area_struct *vma;
 738         struct rb_node *n = mm->mm_rb.rb_node;
 739
 740         /* check the cache first */
 741         vma = mm->mmap_cache;
 742         if (vma && vma->vm_start <= addr && vma->vm_end > addr)
 743                 return vma;
 744
 745         /* trawl the tree (there may be multiple mappings in which addr
 746          * resides) */
 747         for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
 748                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
 749                 if (vma->vm_start > addr)
 750                         return NULL;
 751                 if (vma->vm_end > addr) {
 752                         mm->mmap_cache = vma;
 753                         return vma;
 754                 }
 755         }
 756
 757         return NULL;
 758 }
 759 EXPORT_SYMBOL(find_vma);
 760
 761 /*
 762  * find a VMA
 763  * - we don't extend stack VMAs under NOMMU conditions
 764  */
 765 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
 766 {
 767         return find_vma(mm, addr);
 768 }
 769
 770 /*
 771  * expand a stack to a given address
 772  * - not supported under NOMMU conditions
 773  */
 774 int expand_stack(struct vm_area_struct *vma, unsigned long address)
 775 {
 776         return -ENOMEM;
 777 }
 778
 779 /*
 780  * look up the first VMA exactly that exactly matches addr
 781  * - should be called with mm->mmap_sem at least held readlocked
 782  */
 783 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
 784                                              unsigned long addr,
 785                                              unsigned long len)
 786 {
 787         struct vm_area_struct *vma;
 788         struct rb_node *n = mm->mm_rb.rb_node;
 789         unsigned long end = addr + len;
 790
 791         /* check the cache first */
 792         vma = mm->mmap_cache;
 793         if (vma && vma->vm_start == addr && vma->vm_end == end)
 794                 return vma;
 795
 796         /* trawl the tree (there may be multiple mappings in which addr
 797          * resides) */
 798         for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
 799                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
 800                 if (vma->vm_start < addr)
 801                         continue;
 802                 if (vma->vm_start > addr)
 803                         return NULL;
 804                 if (vma->vm_end == end) {
 805                         mm->mmap_cache = vma;
 806                         return vma;
 807                 }
 808         }
 809
 810         return NULL;
 811 }
 812
 813 /*
 814  * determine whether a mapping should be permitted and, if so, what sort of
 815  * mapping we're capable of supporting
 816  */
 817 static int validate_mmap_request(struct file *file,
 818                                  unsigned long addr,
 819                                  unsigned long len,
 820                                  unsigned long prot,
 821                                  unsigned long flags,
 822                                  unsigned long pgoff,
 823                                  unsigned long *_capabilities)
 824 {
 825         unsigned long capabilities, rlen;
 826         unsigned long reqprot = prot;
 827         int ret;
 828
 829         /* do the simple checks first */
 830         if (flags & MAP_FIXED) {
 831                 printk(KERN_DEBUG
 832                        "%d: Can't do fixed-address/overlay mmap of RAM\n",
 833                        current->pid);
 834                 return -EINVAL;
 835         }
 836
 837         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
 838             (flags & MAP_TYPE) != MAP_SHARED)
 839                 return -EINVAL;
 840
 841         if (!len)
 842                 return -EINVAL;
 843
 844         /* Careful about overflows.. */
 845         rlen = PAGE_ALIGN(len);
 846         if (!rlen || rlen > TASK_SIZE)
 847                 return -ENOMEM;
 848
 849         /* offset overflow? */
 850         if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
 851                 return -EOVERFLOW;
 852
 853         if (file) {
 854                 /* validate file mapping requests */
 855                 struct address_space *mapping;
 856
 857                 /* files must support mmap */
 858                 if (!file->f_op || !file->f_op->mmap)
 859                         return -ENODEV;
 860
 861                 /* work out if what we've got could possibly be shared
 862                  * - we support chardevs that provide their own "memory"
 863                  * - we support files/blockdevs that are memory backed
 864                  */
 865                 mapping = file->f_mapping;
 866                 if (!mapping)
 867                         mapping = file->f_path.dentry->d_inode->i_mapping;
 868
 869                 capabilities = 0;
 870                 if (mapping && mapping->backing_dev_info)
 871                         capabilities = mapping->backing_dev_info->capabilities;
 872
 873                 if (!capabilities) {
 874                         /* no explicit capabilities set, so assume some
 875                          * defaults */
 876                         switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
 877                         case S_IFREG:
 878                         case S_IFBLK:
 879                                 capabilities = BDI_CAP_MAP_COPY;
 880                                 break;
 881
 882                         case S_IFCHR:
 883                                 capabilities =
 884                                         BDI_CAP_MAP_DIRECT |
 885                                         BDI_CAP_READ_MAP |
 886                                         BDI_CAP_WRITE_MAP;
 887                                 break;
 888
 889                         default:
 890                                 return -EINVAL;
 891                         }
 892                 }
 893
 894                 /* eliminate any capabilities that we can't support on this
 895                  * device */
 896                 if (!file->f_op->get_unmapped_area)
 897                         capabilities &= ~BDI_CAP_MAP_DIRECT;
 898                 if (!file->f_op->read)
 899                         capabilities &= ~BDI_CAP_MAP_COPY;
 900
 901                 /* The file shall have been opened with read permission. */
 902                 if (!(file->f_mode & FMODE_READ))
 903                         return -EACCES;
 904
 905                 if (flags & MAP_SHARED) {
 906                         /* do checks for writing, appending and locking */
 907                         if ((prot & PROT_WRITE) &&
 908                             !(file->f_mode & FMODE_WRITE))
 909                                 return -EACCES;
 910
 911                         if (IS_APPEND(file->f_path.dentry->d_inode) &&
 912                             (file->f_mode & FMODE_WRITE))
 913                                 return -EACCES;
 914
 915                         if (locks_verify_locked(file->f_path.dentry->d_inode))
 916                                 return -EAGAIN;
 917
 918                         if (!(capabilities & BDI_CAP_MAP_DIRECT))
 919                                 return -ENODEV;
 920
 921                         /* we mustn't privatise shared mappings */
 922                         capabilities &= ~BDI_CAP_MAP_COPY;
 923                 }
 924                 else {
 925                         /* we're going to read the file into private memory we
 926                          * allocate */
 927                         if (!(capabilities & BDI_CAP_MAP_COPY))
 928                                 return -ENODEV;
 929
 930                         /* we don't permit a private writable mapping to be
 931                          * shared with the backing device */
 932                         if (prot & PROT_WRITE)
 933                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
 934                 }
 935
 936                 if (capabilities & BDI_CAP_MAP_DIRECT) {
 937                         if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
 938                             ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
 939                             ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
 940                             ) {
 941                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
 942                                 if (flags & MAP_SHARED) {
 943                                         printk(KERN_WARNING
 944                                                "MAP_SHARED not completely supported on !MMU\n");
 945                                         return -EINVAL;
 946                                 }
 947                         }
 948                 }
 949
 950                 /* handle executable mappings and implied executable
 951                  * mappings */
 952                 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
 953                         if (prot & PROT_EXEC)
 954                                 return -EPERM;
 955                 }
 956                 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
 957                         /* handle implication of PROT_EXEC by PROT_READ */
 958                         if (current->personality & READ_IMPLIES_EXEC) {
 959                                 if (capabilities & BDI_CAP_EXEC_MAP)
 960                                         prot |= PROT_EXEC;
 961                         }
 962                 }
 963                 else if ((prot & PROT_READ) &&
 964                          (prot & PROT_EXEC) &&
 965                          !(capabilities & BDI_CAP_EXEC_MAP)
 966                          ) {
 967                         /* backing file is not executable, try to copy */
 968                         capabilities &= ~BDI_CAP_MAP_DIRECT;
 969                 }
 970         }
 971         else {
 972                 /* anonymous mappings are always memory backed and can be
 973                  * privately mapped
 974                  */
 975                 capabilities = BDI_CAP_MAP_COPY;
 976
 977                 /* handle PROT_EXEC implication by PROT_READ */
 978                 if ((prot & PROT_READ) &&
 979                     (current->personality & READ_IMPLIES_EXEC))
 980                         prot |= PROT_EXEC;
 981         }
 982
 983         /* allow the security API to have its say */
 984         ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
 985         if (ret < 0)
 986                 return ret;
 987
 988         /* looks okay */
 989         *_capabilities = capabilities;
 990         return 0;
 991 }
 992
 993 /*
 994  * we've determined that we can make the mapping, now translate what we
 995  * now know into VMA flags
 996  */
 997 static unsigned long determine_vm_flags(struct file *file,
 998                                         unsigned long prot,
 999                                         unsigned long flags,
1000                                         unsigned long capabilities)
1001 {
1002         unsigned long vm_flags;
1003
1004         vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
1005         /* vm_flags |= mm->def_flags; */
1006
1007         if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
1008                 /* attempt to share read-only copies of mapped file chunks */
1009                 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1010                 if (file && !(prot & PROT_WRITE))
1011                         vm_flags |= VM_MAYSHARE;
1012         } else {
1013                 /* overlay a shareable mapping on the backing device or inode
1014                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1015                  * romfs/cramfs */
1016                 vm_flags |= VM_MAYSHARE | (capabilities & BDI_CAP_VMFLAGS);
1017                 if (flags & MAP_SHARED)
1018                         vm_flags |= VM_SHARED;
1019         }
1020
1021         /* refuse to let anyone share private mappings with this process if
1022          * it's being traced - otherwise breakpoints set in it may interfere
1023          * with another untraced process
1024          */
1025         if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current))
1026                 vm_flags &= ~VM_MAYSHARE;
1027
1028         return vm_flags;
1029 }
1030
1031 /*
1032  * set up a shared mapping on a file (the driver or filesystem provides and
1033  * pins the storage)
1034  */
1035 static int do_mmap_shared_file(struct vm_area_struct *vma)
1036 {
1037         int ret;
1038
1039         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1040         if (ret == 0) {
1041                 vma->vm_region->vm_top = vma->vm_region->vm_end;
1042                 return 0;
1043         }
1044         if (ret != -ENOSYS)
1045                 return ret;
1046
1047         /* getting -ENOSYS indicates that direct mmap isn't possible (as
1048          * opposed to tried but failed) so we can only give a suitable error as
1049          * it's not possible to make a private copy if MAP_SHARED was given */
1050         return -ENODEV;
1051 }
1052
1053 /*
1054  * set up a private mapping or an anonymous shared mapping
1055  */
1056 static int do_mmap_private(struct vm_area_struct *vma,
1057                            struct vm_region *region,
1058                            unsigned long len,
1059                            unsigned long capabilities)
1060 {
1061         struct page *pages;
1062         unsigned long total, point, n, rlen;
1063         void *base;
1064         int ret, order;
1065
1066         /* invoke the file's mapping function so that it can keep track of
1067          * shared mappings on devices or memory
1068          * - VM_MAYSHARE will be set if it may attempt to share
1069          */
1070         if (capabilities & BDI_CAP_MAP_DIRECT) {
1071                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1072                 if (ret == 0) {
1073                         /* shouldn't return success if we're not sharing */
1074                         BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1075                         vma->vm_region->vm_top = vma->vm_region->vm_end;
1076                         return 0;
1077                 }
1078                 if (ret != -ENOSYS)
1079                         return ret;
1080
1081                 /* getting an ENOSYS error indicates that direct mmap isn't
1082                  * possible (as opposed to tried but failed) so we'll try to
1083                  * make a private copy of the data and map that instead */
1084         }
1085
1086         rlen = PAGE_ALIGN(len);
1087
1088         /* allocate some memory to hold the mapping
1089          * - note that this may not return a page-aligned address if the object
1090          *   we're allocating is smaller than a page
1091          */
1092         order = get_order(rlen);
1093         kdebug("alloc order %d for %lx", order, len);
1094
1095         pages = alloc_pages(GFP_KERNEL, order);
1096         if (!pages)
1097                 goto enomem;
1098
1099         total = 1 << order;
1100         atomic_long_add(total, &mmap_pages_allocated);
1101
1102         point = rlen >> PAGE_SHIFT;
1103
1104         /* we allocated a power-of-2 sized page set, so we may want to trim off
1105          * the excess */
1106         if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
1107                 while (total > point) {
1108                         order = ilog2(total - point);
1109                         n = 1 << order;
1110                         kdebug("shave %lu/%lu @%lu", n, total - point, total);
1111                         atomic_long_sub(n, &mmap_pages_allocated);
1112                         total -= n;
1113                         set_page_refcounted(pages + total);
1114                         __free_pages(pages + total, order);
1115                 }
1116         }
1117
1118         for (point = 1; point < total; point++)
1119                 set_page_refcounted(&pages[point]);
1120
1121         base = page_address(pages);
1122         region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1123         region->vm_start = (unsigned long) base;
1124         region->vm_end   = region->vm_start + rlen;
1125         region->vm_top   = region->vm_start + (total << PAGE_SHIFT);
1126
1127         vma->vm_start = region->vm_start;
1128         vma->vm_end   = region->vm_start + len;
1129
1130         if (vma->vm_file) {
1131                 /* read the contents of a file into the copy */
1132                 mm_segment_t old_fs;
1133                 loff_t fpos;
1134
1135                 fpos = vma->vm_pgoff;
1136                 fpos <<= PAGE_SHIFT;
1137
1138                 old_fs = get_fs();
1139                 set_fs(KERNEL_DS);
1140                 ret = vma->vm_file->f_op->read(vma->vm_file, base, rlen, &fpos);
1141                 set_fs(old_fs);
1142
1143                 if (ret < 0)
1144                         goto error_free;
1145
1146                 /* clear the last little bit */
1147                 if (ret < rlen)
1148                         memset(base + ret, 0, rlen - ret);
1149
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         /* decide whether we should attempt the mapping, and if so what sort of
1187          * mapping */
1188         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1189                                     &capabilities);
1190         if (ret < 0) {
1191                 kleave(" = %d [val]", ret);
1192                 return ret;
1193         }
1194
1195         /* we ignore the address hint */
1196         addr = 0;
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         region->vm_usage = 1;
1212         region->vm_flags = vm_flags;
1213         region->vm_pgoff = pgoff;
1214
1215         INIT_LIST_HEAD(&vma->anon_vma_chain);
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                         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                                         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 (capabilities & BDI_CAP_MAP_DIRECT) {
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          * - the region is filled in if BDI_CAP_MAP_DIRECT is still set
1338          */
1339         if (file && vma->vm_flags & VM_SHARED)
1340                 ret = do_mmap_shared_file(vma);
1341         else
1342                 ret = do_mmap_private(vma, region, len, capabilities);
1343         if (ret < 0)
1344                 goto error_just_free;
1345         add_nommu_region(region);
1346
1347         /* clear anonymous mappings that don't ask for uninitialized data */
1348         if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1349                 memset((void *)region->vm_start, 0,
1350                        region->vm_end - region->vm_start);
1351
1352         /* okay... we have a mapping; now we have to register it */
1353         result = vma->vm_start;
1354
1355         current->mm->total_vm += len >> PAGE_SHIFT;
1356
1357 share:
1358         add_vma_to_mm(current->mm, vma);
1359
1360         /* we flush the region from the icache only when the first executable
1361          * mapping of it is made  */
1362         if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1363                 flush_icache_range(region->vm_start, region->vm_end);
1364                 region->vm_icache_flushed = true;
1365         }
1366
1367         up_write(&nommu_region_sem);
1368
1369         kleave(" = %lx", result);
1370         return result;
1371
1372 error_just_free:
1373         up_write(&nommu_region_sem);
1374 error:
1375         if (region->vm_file)
1376                 fput(region->vm_file);
1377         kmem_cache_free(vm_region_jar, region);
1378         if (vma->vm_file)
1379                 fput(vma->vm_file);
1380         if (vma->vm_flags & VM_EXECUTABLE)
1381                 removed_exe_file_vma(vma->vm_mm);
1382         kmem_cache_free(vm_area_cachep, vma);
1383         kleave(" = %d", ret);
1384         return ret;
1385
1386 sharing_violation:
1387         up_write(&nommu_region_sem);
1388         printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1389         ret = -EINVAL;
1390         goto error;
1391
1392 error_getting_vma:
1393         kmem_cache_free(vm_region_jar, region);
1394         printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1395                " from process %d failed\n",
1396                len, current->pid);
1397         show_free_areas();
1398         return -ENOMEM;
1399
1400 error_getting_region:
1401         printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1402                " from process %d failed\n",
1403                len, current->pid);
1404         show_free_areas();
1405         return -ENOMEM;
1406 }
1407 EXPORT_SYMBOL(do_mmap_pgoff);
1408
1409 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1410                 unsigned long, prot, unsigned long, flags,
1411                 unsigned long, fd, unsigned long, pgoff)
1412 {
1413         struct file *file = NULL;
1414         unsigned long retval = -EBADF;
1415
1416         if (!(flags & MAP_ANONYMOUS)) {
1417                 file = fget(fd);
1418                 if (!file)
1419                         goto out;
1420         }
1421
1422         flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1423
1424         down_write(&current->mm->mmap_sem);
1425         retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1426         up_write(&current->mm->mmap_sem);
1427
1428         if (file)
1429                 fput(file);
1430 out:
1431         return retval;
1432 }
1433
1434 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1435 struct mmap_arg_struct {
1436         unsigned long addr;
1437         unsigned long len;
1438         unsigned long prot;
1439         unsigned long flags;
1440         unsigned long fd;
1441         unsigned long offset;
1442 };
1443
1444 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1445 {
1446         struct mmap_arg_struct a;
1447
1448         if (copy_from_user(&a, arg, sizeof(a)))
1449                 return -EFAULT;
1450         if (a.offset & ~PAGE_MASK)
1451                 return -EINVAL;
1452
1453         return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1454                               a.offset >> PAGE_SHIFT);
1455 }
1456 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1457
1458 /*
1459  * split a vma into two pieces at address 'addr', a new vma is allocated either
1460  * for the first part or the tail.
1461  */
1462 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1463               unsigned long addr, int new_below)
1464 {
1465         struct vm_area_struct *new;
1466         struct vm_region *region;
1467         unsigned long npages;
1468
1469         kenter("");
1470
1471         /* we're only permitted to split anonymous regions (these should have
1472          * only a single usage on the region) */
1473         if (vma->vm_file)
1474                 return -ENOMEM;
1475
1476         if (mm->map_count >= sysctl_max_map_count)
1477                 return -ENOMEM;
1478
1479         region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1480         if (!region)
1481                 return -ENOMEM;
1482
1483         new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1484         if (!new) {
1485                 kmem_cache_free(vm_region_jar, region);
1486                 return -ENOMEM;
1487         }
1488
1489         /* most fields are the same, copy all, and then fixup */
1490         *new = *vma;
1491         *region = *vma->vm_region;
1492         new->vm_region = region;
1493
1494         npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1495
1496         if (new_below) {
1497                 region->vm_top = region->vm_end = new->vm_end = addr;
1498         } else {
1499                 region->vm_start = new->vm_start = addr;
1500                 region->vm_pgoff = new->vm_pgoff += npages;
1501         }
1502
1503         if (new->vm_ops && new->vm_ops->open)
1504                 new->vm_ops->open(new);
1505
1506         delete_vma_from_mm(vma);
1507         down_write(&nommu_region_sem);
1508         delete_nommu_region(vma->vm_region);
1509         if (new_below) {
1510                 vma->vm_region->vm_start = vma->vm_start = addr;
1511                 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1512         } else {
1513                 vma->vm_region->vm_end = vma->vm_end = addr;
1514                 vma->vm_region->vm_top = addr;
1515         }
1516         add_nommu_region(vma->vm_region);
1517         add_nommu_region(new->vm_region);
1518         up_write(&nommu_region_sem);
1519         add_vma_to_mm(mm, vma);
1520         add_vma_to_mm(mm, new);
1521         return 0;
1522 }
1523
1524 /*
1525  * shrink a VMA by removing the specified chunk from either the beginning or
1526  * the end
1527  */
1528 static int shrink_vma(struct mm_struct *mm,
1529                       struct vm_area_struct *vma,
1530                       unsigned long from, unsigned long to)
1531 {
1532         struct vm_region *region;
1533
1534         kenter("");
1535
1536         /* adjust the VMA's pointers, which may reposition it in the MM's tree
1537          * and list */
1538         delete_vma_from_mm(vma);
1539         if (from > vma->vm_start)
1540                 vma->vm_end = from;
1541         else
1542                 vma->vm_start = to;
1543         add_vma_to_mm(mm, vma);
1544
1545         /* cut the backing region down to size */
1546         region = vma->vm_region;
1547         BUG_ON(region->vm_usage != 1);
1548
1549         down_write(&nommu_region_sem);
1550         delete_nommu_region(region);
1551         if (from > region->vm_start) {
1552                 to = region->vm_top;
1553                 region->vm_top = region->vm_end = from;
1554         } else {
1555                 region->vm_start = to;
1556         }
1557         add_nommu_region(region);
1558         up_write(&nommu_region_sem);
1559
1560         free_page_series(from, to);
1561         return 0;
1562 }
1563
1564 /*
1565  * release a mapping
1566  * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1567  *   VMA, though it need not cover the whole VMA
1568  */
1569 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1570 {
1571         struct vm_area_struct *vma;
1572         struct rb_node *rb;
1573         unsigned long end = start + len;
1574         int ret;
1575
1576         kenter(",%lx,%zx", start, len);
1577
1578         if (len == 0)
1579                 return -EINVAL;
1580
1581         /* find the first potentially overlapping VMA */
1582         vma = find_vma(mm, start);
1583         if (!vma) {
1584                 static int limit = 0;
1585                 if (limit < 5) {
1586                         printk(KERN_WARNING
1587                                "munmap of memory not mmapped by process %d"
1588                                " (%s): 0x%lx-0x%lx\n",
1589                                current->pid, current->comm,
1590                                start, start + len - 1);
1591                         limit++;
1592                 }
1593                 return -EINVAL;
1594         }
1595
1596         /* we're allowed to split an anonymous VMA but not a file-backed one */
1597         if (vma->vm_file) {
1598                 do {
1599                         if (start > vma->vm_start) {
1600                                 kleave(" = -EINVAL [miss]");
1601                                 return -EINVAL;
1602                         }
1603                         if (end == vma->vm_end)
1604                                 goto erase_whole_vma;
1605                         rb = rb_next(&vma->vm_rb);
1606                         vma = rb_entry(rb, struct vm_area_struct, vm_rb);
1607                 } while (rb);
1608                 kleave(" = -EINVAL [split file]");
1609                 return -EINVAL;
1610         } else {
1611                 /* the chunk must be a subset of the VMA found */
1612                 if (start == vma->vm_start && end == vma->vm_end)
1613                         goto erase_whole_vma;
1614                 if (start < vma->vm_start || end > vma->vm_end) {
1615                         kleave(" = -EINVAL [superset]");
1616                         return -EINVAL;
1617                 }
1618                 if (start & ~PAGE_MASK) {
1619                         kleave(" = -EINVAL [unaligned start]");
1620                         return -EINVAL;
1621                 }
1622                 if (end != vma->vm_end && end & ~PAGE_MASK) {
1623                         kleave(" = -EINVAL [unaligned split]");
1624                         return -EINVAL;
1625                 }
1626                 if (start != vma->vm_start && end != vma->vm_end) {
1627                         ret = split_vma(mm, vma, start, 1);
1628                         if (ret < 0) {
1629                                 kleave(" = %d [split]", ret);
1630                                 return ret;
1631                         }
1632                 }
1633                 return shrink_vma(mm, vma, start, end);
1634         }
1635
1636 erase_whole_vma:
1637         delete_vma_from_mm(vma);
1638         delete_vma(mm, vma);
1639         kleave(" = 0");
1640         return 0;
1641 }
1642 EXPORT_SYMBOL(do_munmap);
1643
1644 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1645 {
1646         int ret;
1647         struct mm_struct *mm = current->mm;
1648
1649         down_write(&mm->mmap_sem);
1650         ret = do_munmap(mm, addr, len);
1651         up_write(&mm->mmap_sem);
1652         return ret;
1653 }
1654
1655 /*
1656  * release all the mappings made in a process's VM space
1657  */
1658 void exit_mmap(struct mm_struct *mm)
1659 {
1660         struct vm_area_struct *vma;
1661
1662         if (!mm)
1663                 return;
1664
1665         kenter("");
1666
1667         mm->total_vm = 0;
1668
1669         while ((vma = mm->mmap)) {
1670                 mm->mmap = vma->vm_next;
1671                 delete_vma_from_mm(vma);
1672                 delete_vma(mm, vma);
1673         }
1674
1675         kleave("");
1676 }
1677
1678 unsigned long do_brk(unsigned long addr, unsigned long len)
1679 {
1680         return -ENOMEM;
1681 }
1682
1683 /*
1684  * expand (or shrink) an existing mapping, potentially moving it at the same
1685  * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1686  *
1687  * under NOMMU conditions, we only permit changing a mapping's size, and only
1688  * as long as it stays within the region allocated by do_mmap_private() and the
1689  * block is not shareable
1690  *
1691  * MREMAP_FIXED is not supported under NOMMU conditions
1692  */
1693 unsigned long do_mremap(unsigned long addr,
1694                         unsigned long old_len, unsigned long new_len,
1695                         unsigned long flags, unsigned long new_addr)
1696 {
1697         struct vm_area_struct *vma;
1698
1699         /* insanity checks first */
1700         if (old_len == 0 || new_len == 0)
1701                 return (unsigned long) -EINVAL;
1702
1703         if (addr & ~PAGE_MASK)
1704                 return -EINVAL;
1705
1706         if (flags & MREMAP_FIXED && new_addr != addr)
1707                 return (unsigned long) -EINVAL;
1708
1709         vma = find_vma_exact(current->mm, addr, old_len);
1710         if (!vma)
1711                 return (unsigned long) -EINVAL;
1712
1713         if (vma->vm_end != vma->vm_start + old_len)
1714                 return (unsigned long) -EFAULT;
1715
1716         if (vma->vm_flags & VM_MAYSHARE)
1717                 return (unsigned long) -EPERM;
1718
1719         if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1720                 return (unsigned long) -ENOMEM;
1721
1722         /* all checks complete - do it */
1723         vma->vm_end = vma->vm_start + new_len;
1724         return vma->vm_start;
1725 }
1726 EXPORT_SYMBOL(do_mremap);
1727
1728 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1729                 unsigned long, new_len, unsigned long, flags,
1730                 unsigned long, new_addr)
1731 {
1732         unsigned long ret;
1733
1734         down_write(&current->mm->mmap_sem);
1735         ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1736         up_write(&current->mm->mmap_sem);
1737         return ret;
1738 }
1739
1740 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1741                         unsigned int foll_flags)
1742 {
1743         return NULL;
1744 }
1745
1746 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1747                 unsigned long to, unsigned long size, pgprot_t prot)
1748 {
1749         vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1750         return 0;
1751 }
1752 EXPORT_SYMBOL(remap_pfn_range);
1753
1754 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1755                         unsigned long pgoff)
1756 {
1757         unsigned int size = vma->vm_end - vma->vm_start;
1758
1759         if (!(vma->vm_flags & VM_USERMAP))
1760                 return -EINVAL;
1761
1762         vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1763         vma->vm_end = vma->vm_start + size;
1764
1765         return 0;
1766 }
1767 EXPORT_SYMBOL(remap_vmalloc_range);
1768
1769 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1770 {
1771 }
1772
1773 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1774         unsigned long len, unsigned long pgoff, unsigned long flags)
1775 {
1776         return -ENOMEM;
1777 }
1778
1779 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1780 {
1781 }
1782
1783 void unmap_mapping_range(struct address_space *mapping,
1784                          loff_t const holebegin, loff_t const holelen,
1785                          int even_cows)
1786 {
1787 }
1788 EXPORT_SYMBOL(unmap_mapping_range);
1789
1790 /*
1791  * Check that a process has enough memory to allocate a new virtual
1792  * mapping. 0 means there is enough memory for the allocation to
1793  * succeed and -ENOMEM implies there is not.
1794  *
1795  * We currently support three overcommit policies, which are set via the
1796  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1797  *
1798  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1799  * Additional code 2002 Jul 20 by Robert Love.
1800  *
1801  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1802  *
1803  * Note this is a helper function intended to be used by LSMs which
1804  * wish to use this logic.
1805  */
1806 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1807 {
1808         unsigned long free, allowed;
1809
1810         vm_acct_memory(pages);
1811
1812         /*
1813          * Sometimes we want to use more memory than we have
1814          */
1815         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1816                 return 0;
1817
1818         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1819                 unsigned long n;
1820
1821                 free = global_page_state(NR_FILE_PAGES);
1822                 free += nr_swap_pages;
1823
1824                 /*
1825                  * Any slabs which are created with the
1826                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1827                  * which are reclaimable, under pressure.  The dentry
1828                  * cache and most inode caches should fall into this
1829                  */
1830                 free += global_page_state(NR_SLAB_RECLAIMABLE);
1831
1832                 /*
1833                  * Leave the last 3% for root
1834                  */
1835                 if (!cap_sys_admin)
1836                         free -= free / 32;
1837
1838                 if (free > pages)
1839                         return 0;
1840
1841                 /*
1842                  * nr_free_pages() is very expensive on large systems,
1843                  * only call if we're about to fail.
1844                  */
1845                 n = nr_free_pages();
1846
1847                 /*
1848                  * Leave reserved pages. The pages are not for anonymous pages.
1849                  */
1850                 if (n <= totalreserve_pages)
1851                         goto error;
1852                 else
1853                         n -= totalreserve_pages;
1854
1855                 /*
1856                  * Leave the last 3% for root
1857                  */
1858                 if (!cap_sys_admin)
1859                         n -= n / 32;
1860                 free += n;
1861
1862                 if (free > pages)
1863                         return 0;
1864
1865                 goto error;
1866         }
1867
1868         allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1869         /*
1870          * Leave the last 3% for root
1871          */
1872         if (!cap_sys_admin)
1873                 allowed -= allowed / 32;
1874         allowed += total_swap_pages;
1875
1876         /* Don't let a single process grow too big:
1877            leave 3% of the size of this process for other processes */
1878         if (mm)
1879                 allowed -= mm->total_vm / 32;
1880
1881         if (percpu_counter_read_positive(&vm_committed_as) < allowed)
1882                 return 0;
1883
1884 error:
1885         vm_unacct_memory(pages);
1886
1887         return -ENOMEM;
1888 }
1889
1890 int in_gate_area_no_task(unsigned long addr)
1891 {
1892         return 0;
1893 }
1894
1895 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1896 {
1897         BUG();
1898         return 0;
1899 }
1900 EXPORT_SYMBOL(filemap_fault);
1901
1902 /*
1903  * Access another process' address space.
1904  * - source/target buffer must be kernel space
1905  */
1906 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1907 {
1908         struct vm_area_struct *vma;
1909         struct mm_struct *mm;
1910
1911         if (addr + len < addr)
1912                 return 0;
1913
1914         mm = get_task_mm(tsk);
1915         if (!mm)
1916                 return 0;
1917
1918         down_read(&mm->mmap_sem);
1919
1920         /* the access must start within one of the target process's mappings */
1921         vma = find_vma(mm, addr);
1922         if (vma) {
1923                 /* don't overrun this mapping */
1924                 if (addr + len >= vma->vm_end)
1925                         len = vma->vm_end - addr;
1926
1927                 /* only read or write mappings where it is permitted */
1928                 if (write && vma->vm_flags & VM_MAYWRITE)
1929                         copy_to_user_page(vma, NULL, addr,
1930                                          (void *) addr, buf, len);
1931                 else if (!write && vma->vm_flags & VM_MAYREAD)
1932                         copy_from_user_page(vma, NULL, addr,
1933                                             buf, (void *) addr, len);
1934                 else
1935                         len = 0;
1936         } else {
1937                 len = 0;
1938         }
1939
1940         up_read(&mm->mmap_sem);
1941         mmput(mm);
1942         return len;
1943 }
1944
1945 /**
1946  * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1947  * @inode: The inode to check
1948  * @size: The current filesize of the inode
1949  * @newsize: The proposed filesize of the inode
1950  *
1951  * Check the shared mappings on an inode on behalf of a shrinking truncate to
1952  * make sure that that any outstanding VMAs aren't broken and then shrink the
1953  * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
1954  * automatically grant mappings that are too large.
1955  */
1956 int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
1957                                 size_t newsize)
1958 {
1959         struct vm_area_struct *vma;
1960         struct prio_tree_iter iter;
1961         struct vm_region *region;
1962         pgoff_t low, high;
1963         size_t r_size, r_top;
1964
1965         low = newsize >> PAGE_SHIFT;
1966         high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1967
1968         down_write(&nommu_region_sem);
1969
1970         /* search for VMAs that fall within the dead zone */
1971         vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
1972                               low, high) {
1973                 /* found one - only interested if it's shared out of the page
1974                  * cache */
1975                 if (vma->vm_flags & VM_SHARED) {
1976                         up_write(&nommu_region_sem);
1977                         return -ETXTBSY; /* not quite true, but near enough */
1978                 }
1979         }
1980
1981         /* reduce any regions that overlap the dead zone - if in existence,
1982          * these will be pointed to by VMAs that don't overlap the dead zone
1983          *
1984          * we don't check for any regions that start beyond the EOF as there
1985          * shouldn't be any
1986          */
1987         vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
1988                               0, ULONG_MAX) {
1989                 if (!(vma->vm_flags & VM_SHARED))
1990                         continue;
1991
1992                 region = vma->vm_region;
1993                 r_size = region->vm_top - region->vm_start;
1994                 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
1995
1996                 if (r_top > newsize) {
1997                         region->vm_top -= r_top - newsize;
1998                         if (region->vm_end > region->vm_top)
1999                                 region->vm_end = region->vm_top;
2000                 }
2001         }
2002
2003         up_write(&nommu_region_sem);
2004         return 0;
2005 }