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 #if 0
  40 #define kenter(FMT, ...) \
  41         printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
  42 #define kleave(FMT, ...) \
  43         printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
  44 #define kdebug(FMT, ...) \
  45         printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__)
  46 #else
  47 #define kenter(FMT, ...) \
  48         no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
  49 #define kleave(FMT, ...) \
  50         no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
  51 #define kdebug(FMT, ...) \
  52         no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__)
  53 #endif
  54
  55 void *high_memory;
  56 struct page *mem_map;
  57 unsigned long max_mapnr;
  58 unsigned long num_physpages;
  59 unsigned long highest_memmap_pfn;
  60 struct percpu_counter vm_committed_as;
  61 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
  62 int sysctl_overcommit_ratio = 50; /* default is 50% */
  63 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
  64 int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
  65 int heap_stack_gap = 0;
  66
  67 atomic_long_t mmap_pages_allocated;
  68
  69 EXPORT_SYMBOL(mem_map);
  70 EXPORT_SYMBOL(num_physpages);
  71
  72 /* list of mapped, potentially shareable regions */
  73 static struct kmem_cache *vm_region_jar;
  74 struct rb_root nommu_region_tree = RB_ROOT;
  75 DECLARE_RWSEM(nommu_region_sem);
  76
  77 const struct vm_operations_struct generic_file_vm_ops = {
  78 };
  79
  80 /*
  81  * Return the total memory allocated for this pointer, not
  82  * just what the caller asked for.
  83  *
  84  * Doesn't have to be accurate, i.e. may have races.
  85  */
  86 unsigned int kobjsize(const void *objp)
  87 {
  88         struct page *page;
  89
  90         /*
  91          * If the object we have should not have ksize performed on it,
  92          * return size of 0
  93          */
  94         if (!objp || !virt_addr_valid(objp))
  95                 return 0;
  96
  97         page = virt_to_head_page(objp);
  98
  99         /*
 100          * If the allocator sets PageSlab, we know the pointer came from
 101          * kmalloc().
 102          */
 103         if (PageSlab(page))
 104                 return ksize(objp);
 105
 106         /*
 107          * If it's not a compound page, see if we have a matching VMA
 108          * region. This test is intentionally done in reverse order,
 109          * so if there's no VMA, we still fall through and hand back
 110          * PAGE_SIZE for 0-order pages.
 111          */
 112         if (!PageCompound(page)) {
 113                 struct vm_area_struct *vma;
 114
 115                 vma = find_vma(current->mm, (unsigned long)objp);
 116                 if (vma)
 117                         return vma->vm_end - vma->vm_start;
 118         }
 119
 120         /*
 121          * The ksize() function is only guaranteed to work for pointers
 122          * returned by kmalloc(). So handle arbitrary pointers here.
 123          */
 124         return PAGE_SIZE << compound_order(page);
 125 }
 126
 127 int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
 128                      unsigned long start, int nr_pages, unsigned int foll_flags,
 129                      struct page **pages, struct vm_area_struct **vmas)
 130 {
 131         struct vm_area_struct *vma;
 132         unsigned long vm_flags;
 133         int i;
 134
 135         /* calculate required read or write permissions.
 136          * If FOLL_FORCE is set, we only require the "MAY" flags.
 137          */
 138         vm_flags  = (foll_flags & FOLL_WRITE) ?
 139                         (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
 140         vm_flags &= (foll_flags & FOLL_FORCE) ?
 141                         (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
 142
 143         for (i = 0; i < nr_pages; i++) {
 144                 vma = find_vma(mm, start);
 145                 if (!vma)
 146                         goto finish_or_fault;
 147
 148                 /* protect what we can, including chardevs */
 149                 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
 150                     !(vm_flags & vma->vm_flags))
 151                         goto finish_or_fault;
 152
 153                 if (pages) {
 154                         pages[i] = virt_to_page(start);
 155                         if (pages[i])
 156                                 page_cache_get(pages[i]);
 157                 }
 158                 if (vmas)
 159                         vmas[i] = vma;
 160                 start = (start + PAGE_SIZE) & PAGE_MASK;
 161         }
 162
 163         return i;
 164
 165 finish_or_fault:
 166         return i ? : -EFAULT;
 167 }
 168
 169 /*
 170  * get a list of pages in an address range belonging to the specified process
 171  * and indicate the VMA that covers each page
 172  * - this is potentially dodgy as we may end incrementing the page count of a
 173  *   slab page or a secondary page from a compound page
 174  * - don't permit access to VMAs that don't support it, such as I/O mappings
 175  */
 176 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
 177         unsigned long start, int nr_pages, int write, int force,
 178         struct page **pages, struct vm_area_struct **vmas)
 179 {
 180         int flags = 0;
 181
 182         if (write)
 183                 flags |= FOLL_WRITE;
 184         if (force)
 185                 flags |= FOLL_FORCE;
 186
 187         return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas);
 188 }
 189 EXPORT_SYMBOL(get_user_pages);
 190
 191 /**
 192  * follow_pfn - look up PFN at a user virtual address
 193  * @vma: memory mapping
 194  * @address: user virtual address
 195  * @pfn: location to store found PFN
 196  *
 197  * Only IO mappings and raw PFN mappings are allowed.
 198  *
 199  * Returns zero and the pfn at @pfn on success, -ve otherwise.
 200  */
 201 int follow_pfn(struct vm_area_struct *vma, unsigned long address,
 202         unsigned long *pfn)
 203 {
 204         if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
 205                 return -EINVAL;
 206
 207         *pfn = address >> PAGE_SHIFT;
 208         return 0;
 209 }
 210 EXPORT_SYMBOL(follow_pfn);
 211
 212 DEFINE_RWLOCK(vmlist_lock);
 213 struct vm_struct *vmlist;
 214
 215 void vfree(const void *addr)
 216 {
 217         kfree(addr);
 218 }
 219 EXPORT_SYMBOL(vfree);
 220
 221 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
 222 {
 223         /*
 224          *  You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
 225          * returns only a logical address.
 226          */
 227         return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
 228 }
 229 EXPORT_SYMBOL(__vmalloc);
 230
 231 void *vmalloc_user(unsigned long size)
 232 {
 233         void *ret;
 234
 235         ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
 236                         PAGE_KERNEL);
 237         if (ret) {
 238                 struct vm_area_struct *vma;
 239
 240                 down_write(&current->mm->mmap_sem);
 241                 vma = find_vma(current->mm, (unsigned long)ret);
 242                 if (vma)
 243                         vma->vm_flags |= VM_USERMAP;
 244                 up_write(&current->mm->mmap_sem);
 245         }
 246
 247         return ret;
 248 }
 249 EXPORT_SYMBOL(vmalloc_user);
 250
 251 struct page *vmalloc_to_page(const void *addr)
 252 {
 253         return virt_to_page(addr);
 254 }
 255 EXPORT_SYMBOL(vmalloc_to_page);
 256
 257 unsigned long vmalloc_to_pfn(const void *addr)
 258 {
 259         return page_to_pfn(virt_to_page(addr));
 260 }
 261 EXPORT_SYMBOL(vmalloc_to_pfn);
 262
 263 long vread(char *buf, char *addr, unsigned long count)
 264 {
 265         memcpy(buf, addr, count);
 266         return count;
 267 }
 268
 269 long vwrite(char *buf, char *addr, unsigned long count)
 270 {
 271         /* Don't allow overflow */
 272         if ((unsigned long) addr + count < count)
 273                 count = -(unsigned long) addr;
 274
 275         memcpy(addr, buf, count);
 276         return(count);
 277 }
 278
 279 /*
 280  *      vmalloc  -  allocate virtually continguos memory
 281  *
 282  *      @size:          allocation size
 283  *
 284  *      Allocate enough pages to cover @size from the page level
 285  *      allocator and map them into continguos kernel virtual space.
 286  *
 287  *      For tight control over page level allocator and protection flags
 288  *      use __vmalloc() instead.
 289  */
 290 void *vmalloc(unsigned long size)
 291 {
 292        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
 293 }
 294 EXPORT_SYMBOL(vmalloc);
 295
 296 void *vmalloc_node(unsigned long size, int node)
 297 {
 298         return vmalloc(size);
 299 }
 300 EXPORT_SYMBOL(vmalloc_node);
 301
 302 #ifndef PAGE_KERNEL_EXEC
 303 # define PAGE_KERNEL_EXEC PAGE_KERNEL
 304 #endif
 305
 306 /**
 307  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
 308  *      @size:          allocation size
 309  *
 310  *      Kernel-internal function to allocate enough pages to cover @size
 311  *      the page level allocator and map them into contiguous and
 312  *      executable kernel virtual space.
 313  *
 314  *      For tight control over page level allocator and protection flags
 315  *      use __vmalloc() instead.
 316  */
 317
 318 void *vmalloc_exec(unsigned long size)
 319 {
 320         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
 321 }
 322
 323 /**
 324  * vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
 325  *      @size:          allocation size
 326  *
 327  *      Allocate enough 32bit PA addressable pages to cover @size from the
 328  *      page level allocator and map them into continguos kernel virtual space.
 329  */
 330 void *vmalloc_32(unsigned long size)
 331 {
 332         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
 333 }
 334 EXPORT_SYMBOL(vmalloc_32);
 335
 336 /**
 337  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
 338  *      @size:          allocation size
 339  *
 340  * The resulting memory area is 32bit addressable and zeroed so it can be
 341  * mapped to userspace without leaking data.
 342  *
 343  * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
 344  * remap_vmalloc_range() are permissible.
 345  */
 346 void *vmalloc_32_user(unsigned long size)
 347 {
 348         /*
 349          * We'll have to sort out the ZONE_DMA bits for 64-bit,
 350          * but for now this can simply use vmalloc_user() directly.
 351          */
 352         return vmalloc_user(size);
 353 }
 354 EXPORT_SYMBOL(vmalloc_32_user);
 355
 356 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
 357 {
 358         BUG();
 359         return NULL;
 360 }
 361 EXPORT_SYMBOL(vmap);
 362
 363 void vunmap(const void *addr)
 364 {
 365         BUG();
 366 }
 367 EXPORT_SYMBOL(vunmap);
 368
 369 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
 370 {
 371         BUG();
 372         return NULL;
 373 }
 374 EXPORT_SYMBOL(vm_map_ram);
 375
 376 void vm_unmap_ram(const void *mem, unsigned int count)
 377 {
 378         BUG();
 379 }
 380 EXPORT_SYMBOL(vm_unmap_ram);
 381
 382 void vm_unmap_aliases(void)
 383 {
 384 }
 385 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
 386
 387 /*
 388  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
 389  * have one.
 390  */
 391 void  __attribute__((weak)) vmalloc_sync_all(void)
 392 {
 393 }
 394
 395 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
 396                    struct page *page)
 397 {
 398         return -EINVAL;
 399 }
 400 EXPORT_SYMBOL(vm_insert_page);
 401
 402 /*
 403  *  sys_brk() for the most part doesn't need the global kernel
 404  *  lock, except when an application is doing something nasty
 405  *  like trying to un-brk an area that has already been mapped
 406  *  to a regular file.  in this case, the unmapping will need
 407  *  to invoke file system routines that need the global lock.
 408  */
 409 SYSCALL_DEFINE1(brk, unsigned long, brk)
 410 {
 411         struct mm_struct *mm = current->mm;
 412
 413         if (brk < mm->start_brk || brk > mm->context.end_brk)
 414                 return mm->brk;
 415
 416         if (mm->brk == brk)
 417                 return mm->brk;
 418
 419         /*
 420          * Always allow shrinking brk
 421          */
 422         if (brk <= mm->brk) {
 423                 mm->brk = brk;
 424                 return brk;
 425         }
 426
 427         /*
 428          * Ok, looks good - let it rip.
 429          */
 430         flush_icache_range(mm->brk, brk);
 431         return mm->brk = brk;
 432 }
 433
 434 /*
 435  * initialise the VMA and region record slabs
 436  */
 437 void __init mmap_init(void)
 438 {
 439         int ret;
 440
 441         ret = percpu_counter_init(&vm_committed_as, 0);
 442         VM_BUG_ON(ret);
 443         vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
 444 }
 445
 446 /*
 447  * validate the region tree
 448  * - the caller must hold the region lock
 449  */
 450 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
 451 static noinline void validate_nommu_regions(void)
 452 {
 453         struct vm_region *region, *last;
 454         struct rb_node *p, *lastp;
 455
 456         lastp = rb_first(&nommu_region_tree);
 457         if (!lastp)
 458                 return;
 459
 460         last = rb_entry(lastp, struct vm_region, vm_rb);
 461         BUG_ON(unlikely(last->vm_end <= last->vm_start));
 462         BUG_ON(unlikely(last->vm_top < last->vm_end));
 463
 464         while ((p = rb_next(lastp))) {
 465                 region = rb_entry(p, struct vm_region, vm_rb);
 466                 last = rb_entry(lastp, struct vm_region, vm_rb);
 467
 468                 BUG_ON(unlikely(region->vm_end <= region->vm_start));
 469                 BUG_ON(unlikely(region->vm_top < region->vm_end));
 470                 BUG_ON(unlikely(region->vm_start < last->vm_top));
 471
 472                 lastp = p;
 473         }
 474 }
 475 #else
 476 static void validate_nommu_regions(void)
 477 {
 478 }
 479 #endif
 480
 481 /*
 482  * add a region into the global tree
 483  */
 484 static void add_nommu_region(struct vm_region *region)
 485 {
 486         struct vm_region *pregion;
 487         struct rb_node **p, *parent;
 488
 489         validate_nommu_regions();
 490
 491         parent = NULL;
 492         p = &nommu_region_tree.rb_node;
 493         while (*p) {
 494                 parent = *p;
 495                 pregion = rb_entry(parent, struct vm_region, vm_rb);
 496                 if (region->vm_start < pregion->vm_start)
 497                         p = &(*p)->rb_left;
 498                 else if (region->vm_start > pregion->vm_start)
 499                         p = &(*p)->rb_right;
 500                 else if (pregion == region)
 501                         return;
 502                 else
 503                         BUG();
 504         }
 505
 506         rb_link_node(&region->vm_rb, parent, p);
 507         rb_insert_color(&region->vm_rb, &nommu_region_tree);
 508
 509         validate_nommu_regions();
 510 }
 511
 512 /*
 513  * delete a region from the global tree
 514  */
 515 static void delete_nommu_region(struct vm_region *region)
 516 {
 517         BUG_ON(!nommu_region_tree.rb_node);
 518
 519         validate_nommu_regions();
 520         rb_erase(&region->vm_rb, &nommu_region_tree);
 521         validate_nommu_regions();
 522 }
 523
 524 /*
 525  * free a contiguous series of pages
 526  */
 527 static void free_page_series(unsigned long from, unsigned long to)
 528 {
 529         for (; from < to; from += PAGE_SIZE) {
 530                 struct page *page = virt_to_page(from);
 531
 532                 kdebug("- free %lx", from);
 533                 atomic_long_dec(&mmap_pages_allocated);
 534                 if (page_count(page) != 1)
 535                         kdebug("free page %p: refcount not one: %d",
 536                                page, page_count(page));
 537                 put_page(page);
 538         }
 539 }
 540
 541 /*
 542  * release a reference to a region
 543  * - the caller must hold the region semaphore for writing, which this releases
 544  * - the region may not have been added to the tree yet, in which case vm_top
 545  *   will equal vm_start
 546  */
 547 static void __put_nommu_region(struct vm_region *region)
 548         __releases(nommu_region_sem)
 549 {
 550         kenter("%p{%d}", region, region->vm_usage);
 551
 552         BUG_ON(!nommu_region_tree.rb_node);
 553
 554         if (--region->vm_usage == 0) {
 555                 if (region->vm_top > region->vm_start)
 556                         delete_nommu_region(region);
 557                 up_write(&nommu_region_sem);
 558
 559                 if (region->vm_file)
 560                         fput(region->vm_file);
 561
 562                 /* IO memory and memory shared directly out of the pagecache
 563                  * from ramfs/tmpfs mustn't be released here */
 564                 if (region->vm_flags & VM_MAPPED_COPY) {
 565                         kdebug("free series");
 566                         free_page_series(region->vm_start, region->vm_top);
 567                 }
 568                 kmem_cache_free(vm_region_jar, region);
 569         } else {
 570                 up_write(&nommu_region_sem);
 571         }
 572 }
 573
 574 /*
 575  * release a reference to a region
 576  */
 577 static void put_nommu_region(struct vm_region *region)
 578 {
 579         down_write(&nommu_region_sem);
 580         __put_nommu_region(region);
 581 }
 582
 583 /*
 584  * update protection on a vma
 585  */
 586 static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
 587 {
 588 #ifdef CONFIG_MPU
 589         struct mm_struct *mm = vma->vm_mm;
 590         long start = vma->vm_start & PAGE_MASK;
 591         while (start < vma->vm_end) {
 592                 protect_page(mm, start, flags);
 593                 start += PAGE_SIZE;
 594         }
 595         update_protections(mm);
 596 #endif
 597 }
 598
 599 /*
 600  * add a VMA into a process's mm_struct in the appropriate place in the list
 601  * and tree and add to the address space's page tree also if not an anonymous
 602  * page
 603  * - should be called with mm->mmap_sem held writelocked
 604  */
 605 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
 606 {
 607         struct vm_area_struct *pvma, **pp;
 608         struct address_space *mapping;
 609         struct rb_node **p, *parent;
 610
 611         kenter(",%p", vma);
 612
 613         BUG_ON(!vma->vm_region);
 614
 615         mm->map_count++;
 616         vma->vm_mm = mm;
 617
 618         protect_vma(vma, vma->vm_flags);
 619
 620         /* add the VMA to the mapping */
 621         if (vma->vm_file) {
 622                 mapping = vma->vm_file->f_mapping;
 623
 624                 flush_dcache_mmap_lock(mapping);
 625                 vma_prio_tree_insert(vma, &mapping->i_mmap);
 626                 flush_dcache_mmap_unlock(mapping);
 627         }
 628
 629         /* add the VMA to the tree */
 630         parent = NULL;
 631         p = &mm->mm_rb.rb_node;
 632         while (*p) {
 633                 parent = *p;
 634                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
 635
 636                 /* sort by: start addr, end addr, VMA struct addr in that order
 637                  * (the latter is necessary as we may get identical VMAs) */
 638                 if (vma->vm_start < pvma->vm_start)
 639                         p = &(*p)->rb_left;
 640                 else if (vma->vm_start > pvma->vm_start)
 641                         p = &(*p)->rb_right;
 642                 else if (vma->vm_end < pvma->vm_end)
 643                         p = &(*p)->rb_left;
 644                 else if (vma->vm_end > pvma->vm_end)
 645                         p = &(*p)->rb_right;
 646                 else if (vma < pvma)
 647                         p = &(*p)->rb_left;
 648                 else if (vma > pvma)
 649                         p = &(*p)->rb_right;
 650                 else
 651                         BUG();
 652         }
 653
 654         rb_link_node(&vma->vm_rb, parent, p);
 655         rb_insert_color(&vma->vm_rb, &mm->mm_rb);
 656
 657         /* add VMA to the VMA list also */
 658         for (pp = &mm->mmap; (pvma = *pp); pp = &(*pp)->vm_next) {
 659                 if (pvma->vm_start > vma->vm_start)
 660                         break;
 661                 if (pvma->vm_start < vma->vm_start)
 662                         continue;
 663                 if (pvma->vm_end < vma->vm_end)
 664                         break;
 665         }
 666
 667         vma->vm_next = *pp;
 668         *pp = vma;
 669 }
 670
 671 /*
 672  * delete a VMA from its owning mm_struct and address space
 673  */
 674 static void delete_vma_from_mm(struct vm_area_struct *vma)
 675 {
 676         struct vm_area_struct **pp;
 677         struct address_space *mapping;
 678         struct mm_struct *mm = vma->vm_mm;
 679
 680         kenter("%p", vma);
 681
 682         protect_vma(vma, 0);
 683
 684         mm->map_count--;
 685         if (mm->mmap_cache == vma)
 686                 mm->mmap_cache = NULL;
 687
 688         /* remove the VMA from the mapping */
 689         if (vma->vm_file) {
 690                 mapping = vma->vm_file->f_mapping;
 691
 692                 flush_dcache_mmap_lock(mapping);
 693                 vma_prio_tree_remove(vma, &mapping->i_mmap);
 694                 flush_dcache_mmap_unlock(mapping);
 695         }
 696
 697         /* remove from the MM's tree and list */
 698         rb_erase(&vma->vm_rb, &mm->mm_rb);
 699         for (pp = &mm->mmap; *pp; pp = &(*pp)->vm_next) {
 700                 if (*pp == vma) {
 701                         *pp = vma->vm_next;
 702                         break;
 703                 }
 704         }
 705
 706         vma->vm_mm = NULL;
 707 }
 708
 709 /*
 710  * destroy a VMA record
 711  */
 712 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
 713 {
 714         kenter("%p", vma);
 715         if (vma->vm_ops && vma->vm_ops->close)
 716                 vma->vm_ops->close(vma);
 717         if (vma->vm_file) {
 718                 fput(vma->vm_file);
 719                 if (vma->vm_flags & VM_EXECUTABLE)
 720                         removed_exe_file_vma(mm);
 721         }
 722         put_nommu_region(vma->vm_region);
 723         kmem_cache_free(vm_area_cachep, vma);
 724 }
 725
 726 /*
 727  * look up the first VMA in which addr resides, NULL if none
 728  * - should be called with mm->mmap_sem at least held readlocked
 729  */
 730 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
 731 {
 732         struct vm_area_struct *vma;
 733         struct rb_node *n = mm->mm_rb.rb_node;
 734
 735         /* check the cache first */
 736         vma = mm->mmap_cache;
 737         if (vma && vma->vm_start <= addr && vma->vm_end > addr)
 738                 return vma;
 739
 740         /* trawl the tree (there may be multiple mappings in which addr
 741          * resides) */
 742         for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
 743                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
 744                 if (vma->vm_start > addr)
 745                         return NULL;
 746                 if (vma->vm_end > addr) {
 747                         mm->mmap_cache = vma;
 748                         return vma;
 749                 }
 750         }
 751
 752         return NULL;
 753 }
 754 EXPORT_SYMBOL(find_vma);
 755
 756 /*
 757  * find a VMA
 758  * - we don't extend stack VMAs under NOMMU conditions
 759  */
 760 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
 761 {
 762         return find_vma(mm, addr);
 763 }
 764
 765 /*
 766  * expand a stack to a given address
 767  * - not supported under NOMMU conditions
 768  */
 769 int expand_stack(struct vm_area_struct *vma, unsigned long address)
 770 {
 771         return -ENOMEM;
 772 }
 773
 774 /*
 775  * look up the first VMA exactly that exactly matches addr
 776  * - should be called with mm->mmap_sem at least held readlocked
 777  */
 778 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
 779                                              unsigned long addr,
 780                                              unsigned long len)
 781 {
 782         struct vm_area_struct *vma;
 783         struct rb_node *n = mm->mm_rb.rb_node;
 784         unsigned long end = addr + len;
 785
 786         /* check the cache first */
 787         vma = mm->mmap_cache;
 788         if (vma && vma->vm_start == addr && vma->vm_end == end)
 789                 return vma;
 790
 791         /* trawl the tree (there may be multiple mappings in which addr
 792          * resides) */
 793         for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
 794                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
 795                 if (vma->vm_start < addr)
 796                         continue;
 797                 if (vma->vm_start > addr)
 798                         return NULL;
 799                 if (vma->vm_end == end) {
 800                         mm->mmap_cache = vma;
 801                         return vma;
 802                 }
 803         }
 804
 805         return NULL;
 806 }
 807
 808 /*
 809  * determine whether a mapping should be permitted and, if so, what sort of
 810  * mapping we're capable of supporting
 811  */
 812 static int validate_mmap_request(struct file *file,
 813                                  unsigned long addr,
 814                                  unsigned long len,
 815                                  unsigned long prot,
 816                                  unsigned long flags,
 817                                  unsigned long pgoff,
 818                                  unsigned long *_capabilities)
 819 {
 820         unsigned long capabilities, rlen;
 821         unsigned long reqprot = prot;
 822         int ret;
 823
 824         /* do the simple checks first */
 825         if (flags & MAP_FIXED) {
 826                 printk(KERN_DEBUG
 827                        "%d: Can't do fixed-address/overlay mmap of RAM\n",
 828                        current->pid);
 829                 return -EINVAL;
 830         }
 831
 832         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
 833             (flags & MAP_TYPE) != MAP_SHARED)
 834                 return -EINVAL;
 835
 836         if (!len)
 837                 return -EINVAL;
 838
 839         /* Careful about overflows.. */
 840         rlen = PAGE_ALIGN(len);
 841         if (!rlen || rlen > TASK_SIZE)
 842                 return -ENOMEM;
 843
 844         /* offset overflow? */
 845         if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
 846                 return -EOVERFLOW;
 847
 848         if (file) {
 849                 /* validate file mapping requests */
 850                 struct address_space *mapping;
 851
 852                 /* files must support mmap */
 853                 if (!file->f_op || !file->f_op->mmap)
 854                         return -ENODEV;
 855
 856                 /* work out if what we've got could possibly be shared
 857                  * - we support chardevs that provide their own "memory"
 858                  * - we support files/blockdevs that are memory backed
 859                  */
 860                 mapping = file->f_mapping;
 861                 if (!mapping)
 862                         mapping = file->f_path.dentry->d_inode->i_mapping;
 863
 864                 capabilities = 0;
 865                 if (mapping && mapping->backing_dev_info)
 866                         capabilities = mapping->backing_dev_info->capabilities;
 867
 868                 if (!capabilities) {
 869                         /* no explicit capabilities set, so assume some
 870                          * defaults */
 871                         switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
 872                         case S_IFREG:
 873                         case S_IFBLK:
 874                                 capabilities = BDI_CAP_MAP_COPY;
 875                                 break;
 876
 877                         case S_IFCHR:
 878                                 capabilities =
 879                                         BDI_CAP_MAP_DIRECT |
 880                                         BDI_CAP_READ_MAP |
 881                                         BDI_CAP_WRITE_MAP;
 882                                 break;
 883
 884                         default:
 885                                 return -EINVAL;
 886                         }
 887                 }
 888
 889                 /* eliminate any capabilities that we can't support on this
 890                  * device */
 891                 if (!file->f_op->get_unmapped_area)
 892                         capabilities &= ~BDI_CAP_MAP_DIRECT;
 893                 if (!file->f_op->read)
 894                         capabilities &= ~BDI_CAP_MAP_COPY;
 895
 896                 /* The file shall have been opened with read permission. */
 897                 if (!(file->f_mode & FMODE_READ))
 898                         return -EACCES;
 899
 900                 if (flags & MAP_SHARED) {
 901                         /* do checks for writing, appending and locking */
 902                         if ((prot & PROT_WRITE) &&
 903                             !(file->f_mode & FMODE_WRITE))
 904                                 return -EACCES;
 905
 906                         if (IS_APPEND(file->f_path.dentry->d_inode) &&
 907                             (file->f_mode & FMODE_WRITE))
 908                                 return -EACCES;
 909
 910                         if (locks_verify_locked(file->f_path.dentry->d_inode))
 911                                 return -EAGAIN;
 912
 913                         if (!(capabilities & BDI_CAP_MAP_DIRECT))
 914                                 return -ENODEV;
 915
 916                         /* we mustn't privatise shared mappings */
 917                         capabilities &= ~BDI_CAP_MAP_COPY;
 918                 }
 919                 else {
 920                         /* we're going to read the file into private memory we
 921                          * allocate */
 922                         if (!(capabilities & BDI_CAP_MAP_COPY))
 923                                 return -ENODEV;
 924
 925                         /* we don't permit a private writable mapping to be
 926                          * shared with the backing device */
 927                         if (prot & PROT_WRITE)
 928                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
 929                 }
 930
 931                 if (capabilities & BDI_CAP_MAP_DIRECT) {
 932                         if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
 933                             ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
 934                             ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
 935                             ) {
 936                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
 937                                 if (flags & MAP_SHARED) {
 938                                         printk(KERN_WARNING
 939                                                "MAP_SHARED not completely supported on !MMU\n");
 940                                         return -EINVAL;
 941                                 }
 942                         }
 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 |= mm->def_flags; */
1001
1002         if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
1003                 /* attempt to share read-only copies of mapped file chunks */
1004                 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1005                 if (file && !(prot & PROT_WRITE))
1006                         vm_flags |= VM_MAYSHARE;
1007         } else {
1008                 /* overlay a shareable mapping on the backing device or inode
1009                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1010                  * romfs/cramfs */
1011                 vm_flags |= VM_MAYSHARE | (capabilities & BDI_CAP_VMFLAGS);
1012                 if (flags & MAP_SHARED)
1013                         vm_flags |= VM_SHARED;
1014         }
1015
1016         /* refuse to let anyone share private mappings with this process if
1017          * it's being traced - otherwise breakpoints set in it may interfere
1018          * with another untraced process
1019          */
1020         if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current))
1021                 vm_flags &= ~VM_MAYSHARE;
1022
1023         return vm_flags;
1024 }
1025
1026 /*
1027  * set up a shared mapping on a file (the driver or filesystem provides and
1028  * pins the storage)
1029  */
1030 static int do_mmap_shared_file(struct vm_area_struct *vma)
1031 {
1032         int ret;
1033
1034         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1035         if (ret == 0) {
1036                 vma->vm_region->vm_top = vma->vm_region->vm_end;
1037                 return 0;
1038         }
1039         if (ret != -ENOSYS)
1040                 return ret;
1041
1042         /* getting -ENOSYS indicates that direct mmap isn't possible (as
1043          * opposed to tried but failed) so we can only give a suitable error as
1044          * it's not possible to make a private copy if MAP_SHARED was given */
1045         return -ENODEV;
1046 }
1047
1048 /*
1049  * set up a private mapping or an anonymous shared mapping
1050  */
1051 static int do_mmap_private(struct vm_area_struct *vma,
1052                            struct vm_region *region,
1053                            unsigned long len,
1054                            unsigned long capabilities)
1055 {
1056         struct page *pages;
1057         unsigned long total, point, n, rlen;
1058         void *base;
1059         int ret, order;
1060
1061         /* invoke the file's mapping function so that it can keep track of
1062          * shared mappings on devices or memory
1063          * - VM_MAYSHARE will be set if it may attempt to share
1064          */
1065         if (capabilities & BDI_CAP_MAP_DIRECT) {
1066                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1067                 if (ret == 0) {
1068                         /* shouldn't return success if we're not sharing */
1069                         BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1070                         vma->vm_region->vm_top = vma->vm_region->vm_end;
1071                         return 0;
1072                 }
1073                 if (ret != -ENOSYS)
1074                         return ret;
1075
1076                 /* getting an ENOSYS error indicates that direct mmap isn't
1077                  * possible (as opposed to tried but failed) so we'll try to
1078                  * make a private copy of the data and map that instead */
1079         }
1080
1081         rlen = PAGE_ALIGN(len);
1082
1083         /* allocate some memory to hold the mapping
1084          * - note that this may not return a page-aligned address if the object
1085          *   we're allocating is smaller than a page
1086          */
1087         order = get_order(rlen);
1088         kdebug("alloc order %d for %lx", order, len);
1089
1090         pages = alloc_pages(GFP_KERNEL, order);
1091         if (!pages)
1092                 goto enomem;
1093
1094         total = 1 << order;
1095         atomic_long_add(total, &mmap_pages_allocated);
1096
1097         point = rlen >> PAGE_SHIFT;
1098
1099         /* we allocated a power-of-2 sized page set, so we may want to trim off
1100          * the excess */
1101         if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
1102                 while (total > point) {
1103                         order = ilog2(total - point);
1104                         n = 1 << order;
1105                         kdebug("shave %lu/%lu @%lu", n, total - point, total);
1106                         atomic_long_sub(n, &mmap_pages_allocated);
1107                         total -= n;
1108                         set_page_refcounted(pages + total);
1109                         __free_pages(pages + total, order);
1110                 }
1111         }
1112
1113         for (point = 1; point < total; point++)
1114                 set_page_refcounted(&pages[point]);
1115
1116         base = page_address(pages);
1117         region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1118         region->vm_start = (unsigned long) base;
1119         region->vm_end   = region->vm_start + rlen;
1120         region->vm_top   = region->vm_start + (total << PAGE_SHIFT);
1121
1122         vma->vm_start = region->vm_start;
1123         vma->vm_end   = region->vm_start + len;
1124
1125         if (vma->vm_file) {
1126                 /* read the contents of a file into the copy */
1127                 mm_segment_t old_fs;
1128                 loff_t fpos;
1129
1130                 fpos = vma->vm_pgoff;
1131                 fpos <<= PAGE_SHIFT;
1132
1133                 old_fs = get_fs();
1134                 set_fs(KERNEL_DS);
1135                 ret = vma->vm_file->f_op->read(vma->vm_file, base, rlen, &fpos);
1136                 set_fs(old_fs);
1137
1138                 if (ret < 0)
1139                         goto error_free;
1140
1141                 /* clear the last little bit */
1142                 if (ret < rlen)
1143                         memset(base + ret, 0, rlen - ret);
1144
1145         }
1146
1147         return 0;
1148
1149 error_free:
1150         free_page_series(region->vm_start, region->vm_end);
1151         region->vm_start = vma->vm_start = 0;
1152         region->vm_end   = vma->vm_end = 0;
1153         region->vm_top   = 0;
1154         return ret;
1155
1156 enomem:
1157         printk("Allocation of length %lu from process %d (%s) failed\n",
1158                len, current->pid, current->comm);
1159         show_free_areas();
1160         return -ENOMEM;
1161 }
1162
1163 /*
1164  * handle mapping creation for uClinux
1165  */
1166 unsigned long do_mmap_pgoff(struct file *file,
1167                             unsigned long addr,
1168                             unsigned long len,
1169                             unsigned long prot,
1170                             unsigned long flags,
1171                             unsigned long pgoff)
1172 {
1173         struct vm_area_struct *vma;
1174         struct vm_region *region;
1175         struct rb_node *rb;
1176         unsigned long capabilities, vm_flags, result;
1177         int ret;
1178
1179         kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1180
1181         /* decide whether we should attempt the mapping, and if so what sort of
1182          * mapping */
1183         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1184                                     &capabilities);
1185         if (ret < 0) {
1186                 kleave(" = %d [val]", ret);
1187                 return ret;
1188         }
1189
1190         /* we ignore the address hint */
1191         addr = 0;
1192
1193         /* we've determined that we can make the mapping, now translate what we
1194          * now know into VMA flags */
1195         vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1196
1197         /* we're going to need to record the mapping */
1198         region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1199         if (!region)
1200                 goto error_getting_region;
1201
1202         vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1203         if (!vma)
1204                 goto error_getting_vma;
1205
1206         region->vm_usage = 1;
1207         region->vm_flags = vm_flags;
1208         region->vm_pgoff = pgoff;
1209
1210         INIT_LIST_HEAD(&vma->anon_vma_chain);
1211         vma->vm_flags = vm_flags;
1212         vma->vm_pgoff = pgoff;
1213
1214         if (file) {
1215                 region->vm_file = file;
1216                 get_file(file);
1217                 vma->vm_file = file;
1218                 get_file(file);
1219                 if (vm_flags & VM_EXECUTABLE) {
1220                         added_exe_file_vma(current->mm);
1221                         vma->vm_mm = current->mm;
1222                 }
1223         }
1224
1225         down_write(&nommu_region_sem);
1226
1227         /* if we want to share, we need to check for regions created by other
1228          * mmap() calls that overlap with our proposed mapping
1229          * - we can only share with a superset match on most regular files
1230          * - shared mappings on character devices and memory backed files are
1231          *   permitted to overlap inexactly as far as we are concerned for in
1232          *   these cases, sharing is handled in the driver or filesystem rather
1233          *   than here
1234          */
1235         if (vm_flags & VM_MAYSHARE) {
1236                 struct vm_region *pregion;
1237                 unsigned long pglen, rpglen, pgend, rpgend, start;
1238
1239                 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1240                 pgend = pgoff + pglen;
1241
1242                 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1243                         pregion = rb_entry(rb, struct vm_region, vm_rb);
1244
1245                         if (!(pregion->vm_flags & VM_MAYSHARE))
1246                                 continue;
1247
1248                         /* search for overlapping mappings on the same file */
1249                         if (pregion->vm_file->f_path.dentry->d_inode !=
1250                             file->f_path.dentry->d_inode)
1251                                 continue;
1252
1253                         if (pregion->vm_pgoff >= pgend)
1254                                 continue;
1255
1256                         rpglen = pregion->vm_end - pregion->vm_start;
1257                         rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1258                         rpgend = pregion->vm_pgoff + rpglen;
1259                         if (pgoff >= rpgend)
1260                                 continue;
1261
1262                         /* handle inexactly overlapping matches between
1263                          * mappings */
1264                         if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1265                             !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1266                                 /* new mapping is not a subset of the region */
1267                                 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1268                                         goto sharing_violation;
1269                                 continue;
1270                         }
1271
1272                         /* we've found a region we can share */
1273                         pregion->vm_usage++;
1274                         vma->vm_region = pregion;
1275                         start = pregion->vm_start;
1276                         start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1277                         vma->vm_start = start;
1278                         vma->vm_end = start + len;
1279
1280                         if (pregion->vm_flags & VM_MAPPED_COPY) {
1281                                 kdebug("share copy");
1282                                 vma->vm_flags |= VM_MAPPED_COPY;
1283                         } else {
1284                                 kdebug("share mmap");
1285                                 ret = do_mmap_shared_file(vma);
1286                                 if (ret < 0) {
1287                                         vma->vm_region = NULL;
1288                                         vma->vm_start = 0;
1289                                         vma->vm_end = 0;
1290                                         pregion->vm_usage--;
1291                                         pregion = NULL;
1292                                         goto error_just_free;
1293                                 }
1294                         }
1295                         fput(region->vm_file);
1296                         kmem_cache_free(vm_region_jar, region);
1297                         region = pregion;
1298                         result = start;
1299                         goto share;
1300                 }
1301
1302                 /* obtain the address at which to make a shared mapping
1303                  * - this is the hook for quasi-memory character devices to
1304                  *   tell us the location of a shared mapping
1305                  */
1306                 if (capabilities & BDI_CAP_MAP_DIRECT) {
1307                         addr = file->f_op->get_unmapped_area(file, addr, len,
1308                                                              pgoff, flags);
1309                         if (IS_ERR((void *) addr)) {
1310                                 ret = addr;
1311                                 if (ret != (unsigned long) -ENOSYS)
1312                                         goto error_just_free;
1313
1314                                 /* the driver refused to tell us where to site
1315                                  * the mapping so we'll have to attempt to copy
1316                                  * it */
1317                                 ret = (unsigned long) -ENODEV;
1318                                 if (!(capabilities & BDI_CAP_MAP_COPY))
1319                                         goto error_just_free;
1320
1321                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
1322                         } else {
1323                                 vma->vm_start = region->vm_start = addr;
1324                                 vma->vm_end = region->vm_end = addr + len;
1325                         }
1326                 }
1327         }
1328
1329         vma->vm_region = region;
1330
1331         /* set up the mapping
1332          * - the region is filled in if BDI_CAP_MAP_DIRECT is still set
1333          */
1334         if (file && vma->vm_flags & VM_SHARED)
1335                 ret = do_mmap_shared_file(vma);
1336         else
1337                 ret = do_mmap_private(vma, region, len, capabilities);
1338         if (ret < 0)
1339                 goto error_just_free;
1340         add_nommu_region(region);
1341
1342         /* clear anonymous mappings that don't ask for uninitialized data */
1343         if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1344                 memset((void *)region->vm_start, 0,
1345                        region->vm_end - region->vm_start);
1346
1347         /* okay... we have a mapping; now we have to register it */
1348         result = vma->vm_start;
1349
1350         current->mm->total_vm += len >> PAGE_SHIFT;
1351
1352 share:
1353         add_vma_to_mm(current->mm, vma);
1354
1355         /* we flush the region from the icache only when the first executable
1356          * mapping of it is made  */
1357         if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1358                 flush_icache_range(region->vm_start, region->vm_end);
1359                 region->vm_icache_flushed = true;
1360         }
1361
1362         up_write(&nommu_region_sem);
1363
1364         kleave(" = %lx", result);
1365         return result;
1366
1367 error_just_free:
1368         up_write(&nommu_region_sem);
1369 error:
1370         if (region->vm_file)
1371                 fput(region->vm_file);
1372         kmem_cache_free(vm_region_jar, region);
1373         if (vma->vm_file)
1374                 fput(vma->vm_file);
1375         if (vma->vm_flags & VM_EXECUTABLE)
1376                 removed_exe_file_vma(vma->vm_mm);
1377         kmem_cache_free(vm_area_cachep, vma);
1378         kleave(" = %d", ret);
1379         return ret;
1380
1381 sharing_violation:
1382         up_write(&nommu_region_sem);
1383         printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1384         ret = -EINVAL;
1385         goto error;
1386
1387 error_getting_vma:
1388         kmem_cache_free(vm_region_jar, region);
1389         printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1390                " from process %d failed\n",
1391                len, current->pid);
1392         show_free_areas();
1393         return -ENOMEM;
1394
1395 error_getting_region:
1396         printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1397                " from process %d failed\n",
1398                len, current->pid);
1399         show_free_areas();
1400         return -ENOMEM;
1401 }
1402 EXPORT_SYMBOL(do_mmap_pgoff);
1403
1404 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1405                 unsigned long, prot, unsigned long, flags,
1406                 unsigned long, fd, unsigned long, pgoff)
1407 {
1408         struct file *file = NULL;
1409         unsigned long retval = -EBADF;
1410
1411         if (!(flags & MAP_ANONYMOUS)) {
1412                 file = fget(fd);
1413                 if (!file)
1414                         goto out;
1415         }
1416
1417         flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1418
1419         down_write(&current->mm->mmap_sem);
1420         retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1421         up_write(&current->mm->mmap_sem);
1422
1423         if (file)
1424                 fput(file);
1425 out:
1426         return retval;
1427 }
1428
1429 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1430 struct mmap_arg_struct {
1431         unsigned long addr;
1432         unsigned long len;
1433         unsigned long prot;
1434         unsigned long flags;
1435         unsigned long fd;
1436         unsigned long offset;
1437 };
1438
1439 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1440 {
1441         struct mmap_arg_struct a;
1442
1443         if (copy_from_user(&a, arg, sizeof(a)))
1444                 return -EFAULT;
1445         if (a.offset & ~PAGE_MASK)
1446                 return -EINVAL;
1447
1448         return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1449                               a.offset >> PAGE_SHIFT);
1450 }
1451 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1452
1453 /*
1454  * split a vma into two pieces at address 'addr', a new vma is allocated either
1455  * for the first part or the tail.
1456  */
1457 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1458               unsigned long addr, int new_below)
1459 {
1460         struct vm_area_struct *new;
1461         struct vm_region *region;
1462         unsigned long npages;
1463
1464         kenter("");
1465
1466         /* we're only permitted to split anonymous regions (these should have
1467          * only a single usage on the region) */
1468         if (vma->vm_file)
1469                 return -ENOMEM;
1470
1471         if (mm->map_count >= sysctl_max_map_count)
1472                 return -ENOMEM;
1473
1474         region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1475         if (!region)
1476                 return -ENOMEM;
1477
1478         new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1479         if (!new) {
1480                 kmem_cache_free(vm_region_jar, region);
1481                 return -ENOMEM;
1482         }
1483
1484         /* most fields are the same, copy all, and then fixup */
1485         *new = *vma;
1486         *region = *vma->vm_region;
1487         new->vm_region = region;
1488
1489         npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1490
1491         if (new_below) {
1492                 region->vm_top = region->vm_end = new->vm_end = addr;
1493         } else {
1494                 region->vm_start = new->vm_start = addr;
1495                 region->vm_pgoff = new->vm_pgoff += npages;
1496         }
1497
1498         if (new->vm_ops && new->vm_ops->open)
1499                 new->vm_ops->open(new);
1500
1501         delete_vma_from_mm(vma);
1502         down_write(&nommu_region_sem);
1503         delete_nommu_region(vma->vm_region);
1504         if (new_below) {
1505                 vma->vm_region->vm_start = vma->vm_start = addr;
1506                 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1507         } else {
1508                 vma->vm_region->vm_end = vma->vm_end = addr;
1509                 vma->vm_region->vm_top = addr;
1510         }
1511         add_nommu_region(vma->vm_region);
1512         add_nommu_region(new->vm_region);
1513         up_write(&nommu_region_sem);
1514         add_vma_to_mm(mm, vma);
1515         add_vma_to_mm(mm, new);
1516         return 0;
1517 }
1518
1519 /*
1520  * shrink a VMA by removing the specified chunk from either the beginning or
1521  * the end
1522  */
1523 static int shrink_vma(struct mm_struct *mm,
1524                       struct vm_area_struct *vma,
1525                       unsigned long from, unsigned long to)
1526 {
1527         struct vm_region *region;
1528
1529         kenter("");
1530
1531         /* adjust the VMA's pointers, which may reposition it in the MM's tree
1532          * and list */
1533         delete_vma_from_mm(vma);
1534         if (from > vma->vm_start)
1535                 vma->vm_end = from;
1536         else
1537                 vma->vm_start = to;
1538         add_vma_to_mm(mm, vma);
1539
1540         /* cut the backing region down to size */
1541         region = vma->vm_region;
1542         BUG_ON(region->vm_usage != 1);
1543
1544         down_write(&nommu_region_sem);
1545         delete_nommu_region(region);
1546         if (from > region->vm_start) {
1547                 to = region->vm_top;
1548                 region->vm_top = region->vm_end = from;
1549         } else {
1550                 region->vm_start = to;
1551         }
1552         add_nommu_region(region);
1553         up_write(&nommu_region_sem);
1554
1555         free_page_series(from, to);
1556         return 0;
1557 }
1558
1559 /*
1560  * release a mapping
1561  * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1562  *   VMA, though it need not cover the whole VMA
1563  */
1564 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1565 {
1566         struct vm_area_struct *vma;
1567         struct rb_node *rb;
1568         unsigned long end = start + len;
1569         int ret;
1570
1571         kenter(",%lx,%zx", start, len);
1572
1573         if (len == 0)
1574                 return -EINVAL;
1575
1576         /* find the first potentially overlapping VMA */
1577         vma = find_vma(mm, start);
1578         if (!vma) {
1579                 static int limit = 0;
1580                 if (limit < 5) {
1581                         printk(KERN_WARNING
1582                                "munmap of memory not mmapped by process %d"
1583                                " (%s): 0x%lx-0x%lx\n",
1584                                current->pid, current->comm,
1585                                start, start + len - 1);
1586                         limit++;
1587                 }
1588                 return -EINVAL;
1589         }
1590
1591         /* we're allowed to split an anonymous VMA but not a file-backed one */
1592         if (vma->vm_file) {
1593                 do {
1594                         if (start > vma->vm_start) {
1595                                 kleave(" = -EINVAL [miss]");
1596                                 return -EINVAL;
1597                         }
1598                         if (end == vma->vm_end)
1599                                 goto erase_whole_vma;
1600                         rb = rb_next(&vma->vm_rb);
1601                         vma = rb_entry(rb, struct vm_area_struct, vm_rb);
1602                 } while (rb);
1603                 kleave(" = -EINVAL [split file]");
1604                 return -EINVAL;
1605         } else {
1606                 /* the chunk must be a subset of the VMA found */
1607                 if (start == vma->vm_start && end == vma->vm_end)
1608                         goto erase_whole_vma;
1609                 if (start < vma->vm_start || end > vma->vm_end) {
1610                         kleave(" = -EINVAL [superset]");
1611                         return -EINVAL;
1612                 }
1613                 if (start & ~PAGE_MASK) {
1614                         kleave(" = -EINVAL [unaligned start]");
1615                         return -EINVAL;
1616                 }
1617                 if (end != vma->vm_end && end & ~PAGE_MASK) {
1618                         kleave(" = -EINVAL [unaligned split]");
1619                         return -EINVAL;
1620                 }
1621                 if (start != vma->vm_start && end != vma->vm_end) {
1622                         ret = split_vma(mm, vma, start, 1);
1623                         if (ret < 0) {
1624                                 kleave(" = %d [split]", ret);
1625                                 return ret;
1626                         }
1627                 }
1628                 return shrink_vma(mm, vma, start, end);
1629         }
1630
1631 erase_whole_vma:
1632         delete_vma_from_mm(vma);
1633         delete_vma(mm, vma);
1634         kleave(" = 0");
1635         return 0;
1636 }
1637 EXPORT_SYMBOL(do_munmap);
1638
1639 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1640 {
1641         int ret;
1642         struct mm_struct *mm = current->mm;
1643
1644         down_write(&mm->mmap_sem);
1645         ret = do_munmap(mm, addr, len);
1646         up_write(&mm->mmap_sem);
1647         return ret;
1648 }
1649
1650 /*
1651  * release all the mappings made in a process's VM space
1652  */
1653 void exit_mmap(struct mm_struct *mm)
1654 {
1655         struct vm_area_struct *vma;
1656
1657         if (!mm)
1658                 return;
1659
1660         kenter("");
1661
1662         mm->total_vm = 0;
1663
1664         while ((vma = mm->mmap)) {
1665                 mm->mmap = vma->vm_next;
1666                 delete_vma_from_mm(vma);
1667                 delete_vma(mm, vma);
1668         }
1669
1670         kleave("");
1671 }
1672
1673 unsigned long do_brk(unsigned long addr, unsigned long len)
1674 {
1675         return -ENOMEM;
1676 }
1677
1678 /*
1679  * expand (or shrink) an existing mapping, potentially moving it at the same
1680  * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1681  *
1682  * under NOMMU conditions, we only permit changing a mapping's size, and only
1683  * as long as it stays within the region allocated by do_mmap_private() and the
1684  * block is not shareable
1685  *
1686  * MREMAP_FIXED is not supported under NOMMU conditions
1687  */
1688 unsigned long do_mremap(unsigned long addr,
1689                         unsigned long old_len, unsigned long new_len,
1690                         unsigned long flags, unsigned long new_addr)
1691 {
1692         struct vm_area_struct *vma;
1693
1694         /* insanity checks first */
1695         if (old_len == 0 || new_len == 0)
1696                 return (unsigned long) -EINVAL;
1697
1698         if (addr & ~PAGE_MASK)
1699                 return -EINVAL;
1700
1701         if (flags & MREMAP_FIXED && new_addr != addr)
1702                 return (unsigned long) -EINVAL;
1703
1704         vma = find_vma_exact(current->mm, addr, old_len);
1705         if (!vma)
1706                 return (unsigned long) -EINVAL;
1707
1708         if (vma->vm_end != vma->vm_start + old_len)
1709                 return (unsigned long) -EFAULT;
1710
1711         if (vma->vm_flags & VM_MAYSHARE)
1712                 return (unsigned long) -EPERM;
1713
1714         if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1715                 return (unsigned long) -ENOMEM;
1716
1717         /* all checks complete - do it */
1718         vma->vm_end = vma->vm_start + new_len;
1719         return vma->vm_start;
1720 }
1721 EXPORT_SYMBOL(do_mremap);
1722
1723 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1724                 unsigned long, new_len, unsigned long, flags,
1725                 unsigned long, new_addr)
1726 {
1727         unsigned long ret;
1728
1729         down_write(&current->mm->mmap_sem);
1730         ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1731         up_write(&current->mm->mmap_sem);
1732         return ret;
1733 }
1734
1735 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1736                         unsigned int foll_flags)
1737 {
1738         return NULL;
1739 }
1740
1741 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1742                 unsigned long to, unsigned long size, pgprot_t prot)
1743 {
1744         vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1745         return 0;
1746 }
1747 EXPORT_SYMBOL(remap_pfn_range);
1748
1749 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1750                         unsigned long pgoff)
1751 {
1752         unsigned int size = vma->vm_end - vma->vm_start;
1753
1754         if (!(vma->vm_flags & VM_USERMAP))
1755                 return -EINVAL;
1756
1757         vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1758         vma->vm_end = vma->vm_start + size;
1759
1760         return 0;
1761 }
1762 EXPORT_SYMBOL(remap_vmalloc_range);
1763
1764 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1765 {
1766 }
1767
1768 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1769         unsigned long len, unsigned long pgoff, unsigned long flags)
1770 {
1771         return -ENOMEM;
1772 }
1773
1774 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1775 {
1776 }
1777
1778 void unmap_mapping_range(struct address_space *mapping,
1779                          loff_t const holebegin, loff_t const holelen,
1780                          int even_cows)
1781 {
1782 }
1783 EXPORT_SYMBOL(unmap_mapping_range);
1784
1785 /*
1786  * Check that a process has enough memory to allocate a new virtual
1787  * mapping. 0 means there is enough memory for the allocation to
1788  * succeed and -ENOMEM implies there is not.
1789  *
1790  * We currently support three overcommit policies, which are set via the
1791  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1792  *
1793  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1794  * Additional code 2002 Jul 20 by Robert Love.
1795  *
1796  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1797  *
1798  * Note this is a helper function intended to be used by LSMs which
1799  * wish to use this logic.
1800  */
1801 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1802 {
1803         unsigned long free, allowed;
1804
1805         vm_acct_memory(pages);
1806
1807         /*
1808          * Sometimes we want to use more memory than we have
1809          */
1810         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1811                 return 0;
1812
1813         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1814                 unsigned long n;
1815
1816                 free = global_page_state(NR_FILE_PAGES);
1817                 free += nr_swap_pages;
1818
1819                 /*
1820                  * Any slabs which are created with the
1821                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1822                  * which are reclaimable, under pressure.  The dentry
1823                  * cache and most inode caches should fall into this
1824                  */
1825                 free += global_page_state(NR_SLAB_RECLAIMABLE);
1826
1827                 /*
1828                  * Leave the last 3% for root
1829                  */
1830                 if (!cap_sys_admin)
1831                         free -= free / 32;
1832
1833                 if (free > pages)
1834                         return 0;
1835
1836                 /*
1837                  * nr_free_pages() is very expensive on large systems,
1838                  * only call if we're about to fail.
1839                  */
1840                 n = nr_free_pages();
1841
1842                 /*
1843                  * Leave reserved pages. The pages are not for anonymous pages.
1844                  */
1845                 if (n <= totalreserve_pages)
1846                         goto error;
1847                 else
1848                         n -= totalreserve_pages;
1849
1850                 /*
1851                  * Leave the last 3% for root
1852                  */
1853                 if (!cap_sys_admin)
1854                         n -= n / 32;
1855                 free += n;
1856
1857                 if (free > pages)
1858                         return 0;
1859
1860                 goto error;
1861         }
1862
1863         allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1864         /*
1865          * Leave the last 3% for root
1866          */
1867         if (!cap_sys_admin)
1868                 allowed -= allowed / 32;
1869         allowed += total_swap_pages;
1870
1871         /* Don't let a single process grow too big:
1872            leave 3% of the size of this process for other processes */
1873         if (mm)
1874                 allowed -= mm->total_vm / 32;
1875
1876         if (percpu_counter_read_positive(&vm_committed_as) < allowed)
1877                 return 0;
1878
1879 error:
1880         vm_unacct_memory(pages);
1881
1882         return -ENOMEM;
1883 }
1884
1885 int in_gate_area_no_task(unsigned long addr)
1886 {
1887         return 0;
1888 }
1889
1890 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1891 {
1892         BUG();
1893         return 0;
1894 }
1895 EXPORT_SYMBOL(filemap_fault);
1896
1897 /*
1898  * Access another process' address space.
1899  * - source/target buffer must be kernel space
1900  */
1901 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1902 {
1903         struct vm_area_struct *vma;
1904         struct mm_struct *mm;
1905
1906         if (addr + len < addr)
1907                 return 0;
1908
1909         mm = get_task_mm(tsk);
1910         if (!mm)
1911                 return 0;
1912
1913         down_read(&mm->mmap_sem);
1914
1915         /* the access must start within one of the target process's mappings */
1916         vma = find_vma(mm, addr);
1917         if (vma) {
1918                 /* don't overrun this mapping */
1919                 if (addr + len >= vma->vm_end)
1920                         len = vma->vm_end - addr;
1921
1922                 /* only read or write mappings where it is permitted */
1923                 if (write && vma->vm_flags & VM_MAYWRITE)
1924                         copy_to_user_page(vma, NULL, addr,
1925                                          (void *) addr, buf, len);
1926                 else if (!write && vma->vm_flags & VM_MAYREAD)
1927                         copy_from_user_page(vma, NULL, addr,
1928                                             buf, (void *) addr, len);
1929                 else
1930                         len = 0;
1931         } else {
1932                 len = 0;
1933         }
1934
1935         up_read(&mm->mmap_sem);
1936         mmput(mm);
1937         return len;
1938 }
1939
1940 /**
1941  * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1942  * @inode: The inode to check
1943  * @size: The current filesize of the inode
1944  * @newsize: The proposed filesize of the inode
1945  *
1946  * Check the shared mappings on an inode on behalf of a shrinking truncate to
1947  * make sure that that any outstanding VMAs aren't broken and then shrink the
1948  * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
1949  * automatically grant mappings that are too large.
1950  */
1951 int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
1952                                 size_t newsize)
1953 {
1954         struct vm_area_struct *vma;
1955         struct prio_tree_iter iter;
1956         struct vm_region *region;
1957         pgoff_t low, high;
1958         size_t r_size, r_top;
1959
1960         low = newsize >> PAGE_SHIFT;
1961         high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1962
1963         down_write(&nommu_region_sem);
1964
1965         /* search for VMAs that fall within the dead zone */
1966         vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
1967                               low, high) {
1968                 /* found one - only interested if it's shared out of the page
1969                  * cache */
1970                 if (vma->vm_flags & VM_SHARED) {
1971                         up_write(&nommu_region_sem);
1972                         return -ETXTBSY; /* not quite true, but near enough */
1973                 }
1974         }
1975
1976         /* reduce any regions that overlap the dead zone - if in existence,
1977          * these will be pointed to by VMAs that don't overlap the dead zone
1978          *
1979          * we don't check for any regions that start beyond the EOF as there
1980          * shouldn't be any
1981          */
1982         vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
1983                               0, ULONG_MAX) {
1984                 if (!(vma->vm_flags & VM_SHARED))
1985                         continue;
1986
1987                 region = vma->vm_region;
1988                 r_size = region->vm_top - region->vm_start;
1989                 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
1990
1991                 if (r_top > newsize) {
1992                         region->vm_top -= r_top - newsize;
1993                         if (region->vm_end > region->vm_top)
1994                                 region->vm_end = region->vm_top;
1995                 }
1996         }
1997
1998         up_write(&nommu_region_sem);
1999         return 0;
2000 }