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 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 += PAGE_SIZE;
 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         return mm->brk = brk;
 436 }
 437
 438 /*
 439  * initialise the VMA and region record slabs
 440  */
 441 void __init mmap_init(void)
 442 {
 443         int ret;
 444
 445         ret = percpu_counter_init(&vm_committed_as, 0);
 446         VM_BUG_ON(ret);
 447         vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
 448 }
 449
 450 /*
 451  * validate the region tree
 452  * - the caller must hold the region lock
 453  */
 454 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
 455 static noinline void validate_nommu_regions(void)
 456 {
 457         struct vm_region *region, *last;
 458         struct rb_node *p, *lastp;
 459
 460         lastp = rb_first(&nommu_region_tree);
 461         if (!lastp)
 462                 return;
 463
 464         last = rb_entry(lastp, struct vm_region, vm_rb);
 465         BUG_ON(unlikely(last->vm_end <= last->vm_start));
 466         BUG_ON(unlikely(last->vm_top < last->vm_end));
 467
 468         while ((p = rb_next(lastp))) {
 469                 region = rb_entry(p, struct vm_region, vm_rb);
 470                 last = rb_entry(lastp, struct vm_region, vm_rb);
 471
 472                 BUG_ON(unlikely(region->vm_end <= region->vm_start));
 473                 BUG_ON(unlikely(region->vm_top < region->vm_end));
 474                 BUG_ON(unlikely(region->vm_start < last->vm_top));
 475
 476                 lastp = p;
 477         }
 478 }
 479 #else
 480 static void validate_nommu_regions(void)
 481 {
 482 }
 483 #endif
 484
 485 /*
 486  * add a region into the global tree
 487  */
 488 static void add_nommu_region(struct vm_region *region)
 489 {
 490         struct vm_region *pregion;
 491         struct rb_node **p, *parent;
 492
 493         validate_nommu_regions();
 494
 495         parent = NULL;
 496         p = &nommu_region_tree.rb_node;
 497         while (*p) {
 498                 parent = *p;
 499                 pregion = rb_entry(parent, struct vm_region, vm_rb);
 500                 if (region->vm_start < pregion->vm_start)
 501                         p = &(*p)->rb_left;
 502                 else if (region->vm_start > pregion->vm_start)
 503                         p = &(*p)->rb_right;
 504                 else if (pregion == region)
 505                         return;
 506                 else
 507                         BUG();
 508         }
 509
 510         rb_link_node(&region->vm_rb, parent, p);
 511         rb_insert_color(&region->vm_rb, &nommu_region_tree);
 512
 513         validate_nommu_regions();
 514 }
 515
 516 /*
 517  * delete a region from the global tree
 518  */
 519 static void delete_nommu_region(struct vm_region *region)
 520 {
 521         BUG_ON(!nommu_region_tree.rb_node);
 522
 523         validate_nommu_regions();
 524         rb_erase(&region->vm_rb, &nommu_region_tree);
 525         validate_nommu_regions();
 526 }
 527
 528 /*
 529  * free a contiguous series of pages
 530  */
 531 static void free_page_series(unsigned long from, unsigned long to)
 532 {
 533         for (; from < to; from += PAGE_SIZE) {
 534                 struct page *page = virt_to_page(from);
 535
 536                 kdebug("- free %lx", from);
 537                 atomic_long_dec(&mmap_pages_allocated);
 538                 if (page_count(page) != 1)
 539                         kdebug("free page %p: refcount not one: %d",
 540                                page, page_count(page));
 541                 put_page(page);
 542         }
 543 }
 544
 545 /*
 546  * release a reference to a region
 547  * - the caller must hold the region semaphore for writing, which this releases
 548  * - the region may not have been added to the tree yet, in which case vm_top
 549  *   will equal vm_start
 550  */
 551 static void __put_nommu_region(struct vm_region *region)
 552         __releases(nommu_region_sem)
 553 {
 554         kenter("%p{%d}", region, atomic_read(&region->vm_usage));
 555
 556         BUG_ON(!nommu_region_tree.rb_node);
 557
 558         if (atomic_dec_and_test(&region->vm_usage)) {
 559                 if (region->vm_top > region->vm_start)
 560                         delete_nommu_region(region);
 561                 up_write(&nommu_region_sem);
 562
 563                 if (region->vm_file)
 564                         fput(region->vm_file);
 565
 566                 /* IO memory and memory shared directly out of the pagecache
 567                  * from ramfs/tmpfs mustn't be released here */
 568                 if (region->vm_flags & VM_MAPPED_COPY) {
 569                         kdebug("free series");
 570                         free_page_series(region->vm_start, region->vm_top);
 571                 }
 572                 kmem_cache_free(vm_region_jar, region);
 573         } else {
 574                 up_write(&nommu_region_sem);
 575         }
 576 }
 577
 578 /*
 579  * release a reference to a region
 580  */
 581 static void put_nommu_region(struct vm_region *region)
 582 {
 583         down_write(&nommu_region_sem);
 584         __put_nommu_region(region);
 585 }
 586
 587 /*
 588  * update protection on a vma
 589  */
 590 static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
 591 {
 592 #ifdef CONFIG_MPU
 593         struct mm_struct *mm = vma->vm_mm;
 594         long start = vma->vm_start & PAGE_MASK;
 595         while (start < vma->vm_end) {
 596                 protect_page(mm, start, flags);
 597                 start += PAGE_SIZE;
 598         }
 599         update_protections(mm);
 600 #endif
 601 }
 602
 603 /*
 604  * add a VMA into a process's mm_struct in the appropriate place in the list
 605  * and tree and add to the address space's page tree also if not an anonymous
 606  * page
 607  * - should be called with mm->mmap_sem held writelocked
 608  */
 609 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
 610 {
 611         struct vm_area_struct *pvma, **pp;
 612         struct address_space *mapping;
 613         struct rb_node **p, *parent;
 614
 615         kenter(",%p", vma);
 616
 617         BUG_ON(!vma->vm_region);
 618
 619         mm->map_count++;
 620         vma->vm_mm = mm;
 621
 622         protect_vma(vma, vma->vm_flags);
 623
 624         /* add the VMA to the mapping */
 625         if (vma->vm_file) {
 626                 mapping = vma->vm_file->f_mapping;
 627
 628                 flush_dcache_mmap_lock(mapping);
 629                 vma_prio_tree_insert(vma, &mapping->i_mmap);
 630                 flush_dcache_mmap_unlock(mapping);
 631         }
 632
 633         /* add the VMA to the tree */
 634         parent = NULL;
 635         p = &mm->mm_rb.rb_node;
 636         while (*p) {
 637                 parent = *p;
 638                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
 639
 640                 /* sort by: start addr, end addr, VMA struct addr in that order
 641                  * (the latter is necessary as we may get identical VMAs) */
 642                 if (vma->vm_start < pvma->vm_start)
 643                         p = &(*p)->rb_left;
 644                 else if (vma->vm_start > pvma->vm_start)
 645                         p = &(*p)->rb_right;
 646                 else if (vma->vm_end < pvma->vm_end)
 647                         p = &(*p)->rb_left;
 648                 else if (vma->vm_end > pvma->vm_end)
 649                         p = &(*p)->rb_right;
 650                 else if (vma < pvma)
 651                         p = &(*p)->rb_left;
 652                 else if (vma > pvma)
 653                         p = &(*p)->rb_right;
 654                 else
 655                         BUG();
 656         }
 657
 658         rb_link_node(&vma->vm_rb, parent, p);
 659         rb_insert_color(&vma->vm_rb, &mm->mm_rb);
 660
 661         /* add VMA to the VMA list also */
 662         for (pp = &mm->mmap; (pvma = *pp); pp = &(*pp)->vm_next) {
 663                 if (pvma->vm_start > vma->vm_start)
 664                         break;
 665                 if (pvma->vm_start < vma->vm_start)
 666                         continue;
 667                 if (pvma->vm_end < vma->vm_end)
 668                         break;
 669         }
 670
 671         vma->vm_next = *pp;
 672         *pp = vma;
 673 }
 674
 675 /*
 676  * delete a VMA from its owning mm_struct and address space
 677  */
 678 static void delete_vma_from_mm(struct vm_area_struct *vma)
 679 {
 680         struct vm_area_struct **pp;
 681         struct address_space *mapping;
 682         struct mm_struct *mm = vma->vm_mm;
 683
 684         kenter("%p", vma);
 685
 686         protect_vma(vma, 0);
 687
 688         mm->map_count--;
 689         if (mm->mmap_cache == vma)
 690                 mm->mmap_cache = NULL;
 691
 692         /* remove the VMA from the mapping */
 693         if (vma->vm_file) {
 694                 mapping = vma->vm_file->f_mapping;
 695
 696                 flush_dcache_mmap_lock(mapping);
 697                 vma_prio_tree_remove(vma, &mapping->i_mmap);
 698                 flush_dcache_mmap_unlock(mapping);
 699         }
 700
 701         /* remove from the MM's tree and list */
 702         rb_erase(&vma->vm_rb, &mm->mm_rb);
 703         for (pp = &mm->mmap; *pp; pp = &(*pp)->vm_next) {
 704                 if (*pp == vma) {
 705                         *pp = vma->vm_next;
 706                         break;
 707                 }
 708         }
 709
 710         vma->vm_mm = NULL;
 711 }
 712
 713 /*
 714  * destroy a VMA record
 715  */
 716 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
 717 {
 718         kenter("%p", vma);
 719         if (vma->vm_ops && vma->vm_ops->close)
 720                 vma->vm_ops->close(vma);
 721         if (vma->vm_file) {
 722                 fput(vma->vm_file);
 723                 if (vma->vm_flags & VM_EXECUTABLE)
 724                         removed_exe_file_vma(mm);
 725         }
 726         put_nommu_region(vma->vm_region);
 727         kmem_cache_free(vm_area_cachep, vma);
 728 }
 729
 730 /*
 731  * look up the first VMA in which addr resides, NULL if none
 732  * - should be called with mm->mmap_sem at least held readlocked
 733  */
 734 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
 735 {
 736         struct vm_area_struct *vma;
 737         struct rb_node *n = mm->mm_rb.rb_node;
 738
 739         /* check the cache first */
 740         vma = mm->mmap_cache;
 741         if (vma && vma->vm_start <= addr && vma->vm_end > addr)
 742                 return vma;
 743
 744         /* trawl the tree (there may be multiple mappings in which addr
 745          * resides) */
 746         for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
 747                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
 748                 if (vma->vm_start > addr)
 749                         return NULL;
 750                 if (vma->vm_end > addr) {
 751                         mm->mmap_cache = vma;
 752                         return vma;
 753                 }
 754         }
 755
 756         return NULL;
 757 }
 758 EXPORT_SYMBOL(find_vma);
 759
 760 /*
 761  * find a VMA
 762  * - we don't extend stack VMAs under NOMMU conditions
 763  */
 764 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
 765 {
 766         return find_vma(mm, addr);
 767 }
 768
 769 /*
 770  * expand a stack to a given address
 771  * - not supported under NOMMU conditions
 772  */
 773 int expand_stack(struct vm_area_struct *vma, unsigned long address)
 774 {
 775         return -ENOMEM;
 776 }
 777
 778 /*
 779  * look up the first VMA exactly that exactly matches addr
 780  * - should be called with mm->mmap_sem at least held readlocked
 781  */
 782 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
 783                                              unsigned long addr,
 784                                              unsigned long len)
 785 {
 786         struct vm_area_struct *vma;
 787         struct rb_node *n = mm->mm_rb.rb_node;
 788         unsigned long end = addr + len;
 789
 790         /* check the cache first */
 791         vma = mm->mmap_cache;
 792         if (vma && vma->vm_start == addr && vma->vm_end == end)
 793                 return vma;
 794
 795         /* trawl the tree (there may be multiple mappings in which addr
 796          * resides) */
 797         for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
 798                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
 799                 if (vma->vm_start < addr)
 800                         continue;
 801                 if (vma->vm_start > addr)
 802                         return NULL;
 803                 if (vma->vm_end == end) {
 804                         mm->mmap_cache = vma;
 805                         return vma;
 806                 }
 807         }
 808
 809         return NULL;
 810 }
 811
 812 /*
 813  * determine whether a mapping should be permitted and, if so, what sort of
 814  * mapping we're capable of supporting
 815  */
 816 static int validate_mmap_request(struct file *file,
 817                                  unsigned long addr,
 818                                  unsigned long len,
 819                                  unsigned long prot,
 820                                  unsigned long flags,
 821                                  unsigned long pgoff,
 822                                  unsigned long *_capabilities)
 823 {
 824         unsigned long capabilities, rlen;
 825         unsigned long reqprot = prot;
 826         int ret;
 827
 828         /* do the simple checks first */
 829         if (flags & MAP_FIXED || addr) {
 830                 printk(KERN_DEBUG
 831                        "%d: Can't do fixed-address/overlay mmap of RAM\n",
 832                        current->pid);
 833                 return -EINVAL;
 834         }
 835
 836         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
 837             (flags & MAP_TYPE) != MAP_SHARED)
 838                 return -EINVAL;
 839
 840         if (!len)
 841                 return -EINVAL;
 842
 843         /* Careful about overflows.. */
 844         rlen = PAGE_ALIGN(len);
 845         if (!rlen || rlen > TASK_SIZE)
 846                 return -ENOMEM;
 847
 848         /* offset overflow? */
 849         if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
 850                 return -EOVERFLOW;
 851
 852         if (file) {
 853                 /* validate file mapping requests */
 854                 struct address_space *mapping;
 855
 856                 /* files must support mmap */
 857                 if (!file->f_op || !file->f_op->mmap)
 858                         return -ENODEV;
 859
 860                 /* work out if what we've got could possibly be shared
 861                  * - we support chardevs that provide their own "memory"
 862                  * - we support files/blockdevs that are memory backed
 863                  */
 864                 mapping = file->f_mapping;
 865                 if (!mapping)
 866                         mapping = file->f_path.dentry->d_inode->i_mapping;
 867
 868                 capabilities = 0;
 869                 if (mapping && mapping->backing_dev_info)
 870                         capabilities = mapping->backing_dev_info->capabilities;
 871
 872                 if (!capabilities) {
 873                         /* no explicit capabilities set, so assume some
 874                          * defaults */
 875                         switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
 876                         case S_IFREG:
 877                         case S_IFBLK:
 878                                 capabilities = BDI_CAP_MAP_COPY;
 879                                 break;
 880
 881                         case S_IFCHR:
 882                                 capabilities =
 883                                         BDI_CAP_MAP_DIRECT |
 884                                         BDI_CAP_READ_MAP |
 885                                         BDI_CAP_WRITE_MAP;
 886                                 break;
 887
 888                         default:
 889                                 return -EINVAL;
 890                         }
 891                 }
 892
 893                 /* eliminate any capabilities that we can't support on this
 894                  * device */
 895                 if (!file->f_op->get_unmapped_area)
 896                         capabilities &= ~BDI_CAP_MAP_DIRECT;
 897                 if (!file->f_op->read)
 898                         capabilities &= ~BDI_CAP_MAP_COPY;
 899
 900                 /* The file shall have been opened with read permission. */
 901                 if (!(file->f_mode & FMODE_READ))
 902                         return -EACCES;
 903
 904                 if (flags & MAP_SHARED) {
 905                         /* do checks for writing, appending and locking */
 906                         if ((prot & PROT_WRITE) &&
 907                             !(file->f_mode & FMODE_WRITE))
 908                                 return -EACCES;
 909
 910                         if (IS_APPEND(file->f_path.dentry->d_inode) &&
 911                             (file->f_mode & FMODE_WRITE))
 912                                 return -EACCES;
 913
 914                         if (locks_verify_locked(file->f_path.dentry->d_inode))
 915                                 return -EAGAIN;
 916
 917                         if (!(capabilities & BDI_CAP_MAP_DIRECT))
 918                                 return -ENODEV;
 919
 920                         if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
 921                             ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
 922                             ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
 923                             ) {
 924                                 printk("MAP_SHARED not completely supported on !MMU\n");
 925                                 return -EINVAL;
 926                         }
 927
 928                         /* we mustn't privatise shared mappings */
 929                         capabilities &= ~BDI_CAP_MAP_COPY;
 930                 }
 931                 else {
 932                         /* we're going to read the file into private memory we
 933                          * allocate */
 934                         if (!(capabilities & BDI_CAP_MAP_COPY))
 935                                 return -ENODEV;
 936
 937                         /* we don't permit a private writable mapping to be
 938                          * shared with the backing device */
 939                         if (prot & PROT_WRITE)
 940                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
 941                 }
 942
 943                 /* handle executable mappings and implied executable
 944                  * mappings */
 945                 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
 946                         if (prot & PROT_EXEC)
 947                                 return -EPERM;
 948                 }
 949                 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
 950                         /* handle implication of PROT_EXEC by PROT_READ */
 951                         if (current->personality & READ_IMPLIES_EXEC) {
 952                                 if (capabilities & BDI_CAP_EXEC_MAP)
 953                                         prot |= PROT_EXEC;
 954                         }
 955                 }
 956                 else if ((prot & PROT_READ) &&
 957                          (prot & PROT_EXEC) &&
 958                          !(capabilities & BDI_CAP_EXEC_MAP)
 959                          ) {
 960                         /* backing file is not executable, try to copy */
 961                         capabilities &= ~BDI_CAP_MAP_DIRECT;
 962                 }
 963         }
 964         else {
 965                 /* anonymous mappings are always memory backed and can be
 966                  * privately mapped
 967                  */
 968                 capabilities = BDI_CAP_MAP_COPY;
 969
 970                 /* handle PROT_EXEC implication by PROT_READ */
 971                 if ((prot & PROT_READ) &&
 972                     (current->personality & READ_IMPLIES_EXEC))
 973                         prot |= PROT_EXEC;
 974         }
 975
 976         /* allow the security API to have its say */
 977         ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
 978         if (ret < 0)
 979                 return ret;
 980
 981         /* looks okay */
 982         *_capabilities = capabilities;
 983         return 0;
 984 }
 985
 986 /*
 987  * we've determined that we can make the mapping, now translate what we
 988  * now know into VMA flags
 989  */
 990 static unsigned long determine_vm_flags(struct file *file,
 991                                         unsigned long prot,
 992                                         unsigned long flags,
 993                                         unsigned long capabilities)
 994 {
 995         unsigned long vm_flags;
 996
 997         vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
 998         vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 999         /* vm_flags |= mm->def_flags; */
1000
1001         if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
1002                 /* attempt to share read-only copies of mapped file chunks */
1003                 if (file && !(prot & PROT_WRITE))
1004                         vm_flags |= VM_MAYSHARE;
1005         }
1006         else {
1007                 /* overlay a shareable mapping on the backing device or inode
1008                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1009                  * romfs/cramfs */
1010                 if (flags & MAP_SHARED)
1011                         vm_flags |= VM_MAYSHARE | VM_SHARED;
1012                 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
1013                         vm_flags |= VM_MAYSHARE;
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 ret;
1038         }
1039         if (ret != -ENOSYS)
1040                 return ret;
1041
1042         /* getting an ENOSYS error indicates that direct mmap isn't
1043          * possible (as opposed to tried but failed) so we'll fall
1044          * through to making a private copy of the data and mapping
1045          * that if we can */
1046         return -ENODEV;
1047 }
1048
1049 /*
1050  * set up a private mapping or an anonymous shared mapping
1051  */
1052 static int do_mmap_private(struct vm_area_struct *vma,
1053                            struct vm_region *region,
1054                            unsigned long len)
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 (vma->vm_file) {
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 ret;
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         } else {
1146                 /* if it's an anonymous mapping, then just clear it */
1147                 memset(base, 0, rlen);
1148         }
1149
1150         return 0;
1151
1152 error_free:
1153         free_page_series(region->vm_start, region->vm_end);
1154         region->vm_start = vma->vm_start = 0;
1155         region->vm_end   = vma->vm_end = 0;
1156         region->vm_top   = 0;
1157         return ret;
1158
1159 enomem:
1160         printk("Allocation of length %lu from process %d (%s) failed\n",
1161                len, current->pid, current->comm);
1162         show_free_areas();
1163         return -ENOMEM;
1164 }
1165
1166 /*
1167  * handle mapping creation for uClinux
1168  */
1169 unsigned long do_mmap_pgoff(struct file *file,
1170                             unsigned long addr,
1171                             unsigned long len,
1172                             unsigned long prot,
1173                             unsigned long flags,
1174                             unsigned long pgoff)
1175 {
1176         struct vm_area_struct *vma;
1177         struct vm_region *region;
1178         struct rb_node *rb;
1179         unsigned long capabilities, vm_flags, result;
1180         int ret;
1181
1182         kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1183
1184         if (!(flags & MAP_FIXED))
1185                 addr = round_hint_to_min(addr);
1186
1187         /* decide whether we should attempt the mapping, and if so what sort of
1188          * mapping */
1189         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1190                                     &capabilities);
1191         if (ret < 0) {
1192                 kleave(" = %d [val]", ret);
1193                 return ret;
1194         }
1195
1196         /* we've determined that we can make the mapping, now translate what we
1197          * now know into VMA flags */
1198         vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1199
1200         /* we're going to need to record the mapping */
1201         region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1202         if (!region)
1203                 goto error_getting_region;
1204
1205         vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1206         if (!vma)
1207                 goto error_getting_vma;
1208
1209         atomic_set(&region->vm_usage, 1);
1210         region->vm_flags = vm_flags;
1211         region->vm_pgoff = pgoff;
1212
1213         INIT_LIST_HEAD(&vma->anon_vma_node);
1214         vma->vm_flags = vm_flags;
1215         vma->vm_pgoff = pgoff;
1216
1217         if (file) {
1218                 region->vm_file = file;
1219                 get_file(file);
1220                 vma->vm_file = file;
1221                 get_file(file);
1222                 if (vm_flags & VM_EXECUTABLE) {
1223                         added_exe_file_vma(current->mm);
1224                         vma->vm_mm = current->mm;
1225                 }
1226         }
1227
1228         down_write(&nommu_region_sem);
1229
1230         /* if we want to share, we need to check for regions created by other
1231          * mmap() calls that overlap with our proposed mapping
1232          * - we can only share with a superset match on most regular files
1233          * - shared mappings on character devices and memory backed files are
1234          *   permitted to overlap inexactly as far as we are concerned for in
1235          *   these cases, sharing is handled in the driver or filesystem rather
1236          *   than here
1237          */
1238         if (vm_flags & VM_MAYSHARE) {
1239                 struct vm_region *pregion;
1240                 unsigned long pglen, rpglen, pgend, rpgend, start;
1241
1242                 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1243                 pgend = pgoff + pglen;
1244
1245                 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1246                         pregion = rb_entry(rb, struct vm_region, vm_rb);
1247
1248                         if (!(pregion->vm_flags & VM_MAYSHARE))
1249                                 continue;
1250
1251                         /* search for overlapping mappings on the same file */
1252                         if (pregion->vm_file->f_path.dentry->d_inode !=
1253                             file->f_path.dentry->d_inode)
1254                                 continue;
1255
1256                         if (pregion->vm_pgoff >= pgend)
1257                                 continue;
1258
1259                         rpglen = pregion->vm_end - pregion->vm_start;
1260                         rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1261                         rpgend = pregion->vm_pgoff + rpglen;
1262                         if (pgoff >= rpgend)
1263                                 continue;
1264
1265                         /* handle inexactly overlapping matches between
1266                          * mappings */
1267                         if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1268                             !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1269                                 /* new mapping is not a subset of the region */
1270                                 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1271                                         goto sharing_violation;
1272                                 continue;
1273                         }
1274
1275                         /* we've found a region we can share */
1276                         atomic_inc(&pregion->vm_usage);
1277                         vma->vm_region = pregion;
1278                         start = pregion->vm_start;
1279                         start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1280                         vma->vm_start = start;
1281                         vma->vm_end = start + len;
1282
1283                         if (pregion->vm_flags & VM_MAPPED_COPY) {
1284                                 kdebug("share copy");
1285                                 vma->vm_flags |= VM_MAPPED_COPY;
1286                         } else {
1287                                 kdebug("share mmap");
1288                                 ret = do_mmap_shared_file(vma);
1289                                 if (ret < 0) {
1290                                         vma->vm_region = NULL;
1291                                         vma->vm_start = 0;
1292                                         vma->vm_end = 0;
1293                                         atomic_dec(&pregion->vm_usage);
1294                                         pregion = NULL;
1295                                         goto error_just_free;
1296                                 }
1297                         }
1298                         fput(region->vm_file);
1299                         kmem_cache_free(vm_region_jar, region);
1300                         region = pregion;
1301                         result = start;
1302                         goto share;
1303                 }
1304
1305                 /* obtain the address at which to make a shared mapping
1306                  * - this is the hook for quasi-memory character devices to
1307                  *   tell us the location of a shared mapping
1308                  */
1309                 if (file && file->f_op->get_unmapped_area) {
1310                         addr = file->f_op->get_unmapped_area(file, addr, len,
1311                                                              pgoff, flags);
1312                         if (IS_ERR((void *) addr)) {
1313                                 ret = addr;
1314                                 if (ret != (unsigned long) -ENOSYS)
1315                                         goto error_just_free;
1316
1317                                 /* the driver refused to tell us where to site
1318                                  * the mapping so we'll have to attempt to copy
1319                                  * it */
1320                                 ret = (unsigned long) -ENODEV;
1321                                 if (!(capabilities & BDI_CAP_MAP_COPY))
1322                                         goto error_just_free;
1323
1324                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
1325                         } else {
1326                                 vma->vm_start = region->vm_start = addr;
1327                                 vma->vm_end = region->vm_end = addr + len;
1328                         }
1329                 }
1330         }
1331
1332         vma->vm_region = region;
1333         add_nommu_region(region);
1334
1335         /* set up the mapping */
1336         if (file && vma->vm_flags & VM_SHARED)
1337                 ret = do_mmap_shared_file(vma);
1338         else
1339                 ret = do_mmap_private(vma, region, len);
1340         if (ret < 0)
1341                 goto error_put_region;
1342
1343         /* okay... we have a mapping; now we have to register it */
1344         result = vma->vm_start;
1345
1346         current->mm->total_vm += len >> PAGE_SHIFT;
1347
1348 share:
1349         add_vma_to_mm(current->mm, vma);
1350
1351         up_write(&nommu_region_sem);
1352
1353         if (prot & PROT_EXEC)
1354                 flush_icache_range(result, result + len);
1355
1356         kleave(" = %lx", result);
1357         return result;
1358
1359 error_put_region:
1360         __put_nommu_region(region);
1361         if (vma) {
1362                 if (vma->vm_file) {
1363                         fput(vma->vm_file);
1364                         if (vma->vm_flags & VM_EXECUTABLE)
1365                                 removed_exe_file_vma(vma->vm_mm);
1366                 }
1367                 kmem_cache_free(vm_area_cachep, vma);
1368         }
1369         kleave(" = %d [pr]", ret);
1370         return ret;
1371
1372 error_just_free:
1373         up_write(&nommu_region_sem);
1374 error:
1375         fput(region->vm_file);
1376         kmem_cache_free(vm_region_jar, region);
1377         fput(vma->vm_file);
1378         if (vma->vm_flags & VM_EXECUTABLE)
1379                 removed_exe_file_vma(vma->vm_mm);
1380         kmem_cache_free(vm_area_cachep, vma);
1381         kleave(" = %d", ret);
1382         return ret;
1383
1384 sharing_violation:
1385         up_write(&nommu_region_sem);
1386         printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1387         ret = -EINVAL;
1388         goto error;
1389
1390 error_getting_vma:
1391         kmem_cache_free(vm_region_jar, region);
1392         printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1393                " from process %d failed\n",
1394                len, current->pid);
1395         show_free_areas();
1396         return -ENOMEM;
1397
1398 error_getting_region:
1399         printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1400                " from process %d failed\n",
1401                len, current->pid);
1402         show_free_areas();
1403         return -ENOMEM;
1404 }
1405 EXPORT_SYMBOL(do_mmap_pgoff);
1406
1407 /*
1408  * split a vma into two pieces at address 'addr', a new vma is allocated either
1409  * for the first part or the tail.
1410  */
1411 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1412               unsigned long addr, int new_below)
1413 {
1414         struct vm_area_struct *new;
1415         struct vm_region *region;
1416         unsigned long npages;
1417
1418         kenter("");
1419
1420         /* we're only permitted to split anonymous regions that have a single
1421          * owner */
1422         if (vma->vm_file ||
1423             atomic_read(&vma->vm_region->vm_usage) != 1)
1424                 return -ENOMEM;
1425
1426         if (mm->map_count >= sysctl_max_map_count)
1427                 return -ENOMEM;
1428
1429         region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1430         if (!region)
1431                 return -ENOMEM;
1432
1433         new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1434         if (!new) {
1435                 kmem_cache_free(vm_region_jar, region);
1436                 return -ENOMEM;
1437         }
1438
1439         /* most fields are the same, copy all, and then fixup */
1440         *new = *vma;
1441         *region = *vma->vm_region;
1442         new->vm_region = region;
1443
1444         npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1445
1446         if (new_below) {
1447                 region->vm_top = region->vm_end = new->vm_end = addr;
1448         } else {
1449                 region->vm_start = new->vm_start = addr;
1450                 region->vm_pgoff = new->vm_pgoff += npages;
1451         }
1452
1453         if (new->vm_ops && new->vm_ops->open)
1454                 new->vm_ops->open(new);
1455
1456         delete_vma_from_mm(vma);
1457         down_write(&nommu_region_sem);
1458         delete_nommu_region(vma->vm_region);
1459         if (new_below) {
1460                 vma->vm_region->vm_start = vma->vm_start = addr;
1461                 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1462         } else {
1463                 vma->vm_region->vm_end = vma->vm_end = addr;
1464                 vma->vm_region->vm_top = addr;
1465         }
1466         add_nommu_region(vma->vm_region);
1467         add_nommu_region(new->vm_region);
1468         up_write(&nommu_region_sem);
1469         add_vma_to_mm(mm, vma);
1470         add_vma_to_mm(mm, new);
1471         return 0;
1472 }
1473
1474 /*
1475  * shrink a VMA by removing the specified chunk from either the beginning or
1476  * the end
1477  */
1478 static int shrink_vma(struct mm_struct *mm,
1479                       struct vm_area_struct *vma,
1480                       unsigned long from, unsigned long to)
1481 {
1482         struct vm_region *region;
1483
1484         kenter("");
1485
1486         /* adjust the VMA's pointers, which may reposition it in the MM's tree
1487          * and list */
1488         delete_vma_from_mm(vma);
1489         if (from > vma->vm_start)
1490                 vma->vm_end = from;
1491         else
1492                 vma->vm_start = to;
1493         add_vma_to_mm(mm, vma);
1494
1495         /* cut the backing region down to size */
1496         region = vma->vm_region;
1497         BUG_ON(atomic_read(&region->vm_usage) != 1);
1498
1499         down_write(&nommu_region_sem);
1500         delete_nommu_region(region);
1501         if (from > region->vm_start) {
1502                 to = region->vm_top;
1503                 region->vm_top = region->vm_end = from;
1504         } else {
1505                 region->vm_start = to;
1506         }
1507         add_nommu_region(region);
1508         up_write(&nommu_region_sem);
1509
1510         free_page_series(from, to);
1511         return 0;
1512 }
1513
1514 /*
1515  * release a mapping
1516  * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1517  *   VMA, though it need not cover the whole VMA
1518  */
1519 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1520 {
1521         struct vm_area_struct *vma;
1522         struct rb_node *rb;
1523         unsigned long end = start + len;
1524         int ret;
1525
1526         kenter(",%lx,%zx", start, len);
1527
1528         if (len == 0)
1529                 return -EINVAL;
1530
1531         /* find the first potentially overlapping VMA */
1532         vma = find_vma(mm, start);
1533         if (!vma) {
1534                 static int limit = 0;
1535                 if (limit < 5) {
1536                         printk(KERN_WARNING
1537                                "munmap of memory not mmapped by process %d"
1538                                " (%s): 0x%lx-0x%lx\n",
1539                                current->pid, current->comm,
1540                                start, start + len - 1);
1541                         limit++;
1542                 }
1543                 return -EINVAL;
1544         }
1545
1546         /* we're allowed to split an anonymous VMA but not a file-backed one */
1547         if (vma->vm_file) {
1548                 do {
1549                         if (start > vma->vm_start) {
1550                                 kleave(" = -EINVAL [miss]");
1551                                 return -EINVAL;
1552                         }
1553                         if (end == vma->vm_end)
1554                                 goto erase_whole_vma;
1555                         rb = rb_next(&vma->vm_rb);
1556                         vma = rb_entry(rb, struct vm_area_struct, vm_rb);
1557                 } while (rb);
1558                 kleave(" = -EINVAL [split file]");
1559                 return -EINVAL;
1560         } else {
1561                 /* the chunk must be a subset of the VMA found */
1562                 if (start == vma->vm_start && end == vma->vm_end)
1563                         goto erase_whole_vma;
1564                 if (start < vma->vm_start || end > vma->vm_end) {
1565                         kleave(" = -EINVAL [superset]");
1566                         return -EINVAL;
1567                 }
1568                 if (start & ~PAGE_MASK) {
1569                         kleave(" = -EINVAL [unaligned start]");
1570                         return -EINVAL;
1571                 }
1572                 if (end != vma->vm_end && end & ~PAGE_MASK) {
1573                         kleave(" = -EINVAL [unaligned split]");
1574                         return -EINVAL;
1575                 }
1576                 if (start != vma->vm_start && end != vma->vm_end) {
1577                         ret = split_vma(mm, vma, start, 1);
1578                         if (ret < 0) {
1579                                 kleave(" = %d [split]", ret);
1580                                 return ret;
1581                         }
1582                 }
1583                 return shrink_vma(mm, vma, start, end);
1584         }
1585
1586 erase_whole_vma:
1587         delete_vma_from_mm(vma);
1588         delete_vma(mm, vma);
1589         kleave(" = 0");
1590         return 0;
1591 }
1592 EXPORT_SYMBOL(do_munmap);
1593
1594 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1595 {
1596         int ret;
1597         struct mm_struct *mm = current->mm;
1598
1599         down_write(&mm->mmap_sem);
1600         ret = do_munmap(mm, addr, len);
1601         up_write(&mm->mmap_sem);
1602         return ret;
1603 }
1604
1605 /*
1606  * release all the mappings made in a process's VM space
1607  */
1608 void exit_mmap(struct mm_struct *mm)
1609 {
1610         struct vm_area_struct *vma;
1611
1612         if (!mm)
1613                 return;
1614
1615         kenter("");
1616
1617         mm->total_vm = 0;
1618
1619         while ((vma = mm->mmap)) {
1620                 mm->mmap = vma->vm_next;
1621                 delete_vma_from_mm(vma);
1622                 delete_vma(mm, vma);
1623         }
1624
1625         kleave("");
1626 }
1627
1628 unsigned long do_brk(unsigned long addr, unsigned long len)
1629 {
1630         return -ENOMEM;
1631 }
1632
1633 /*
1634  * expand (or shrink) an existing mapping, potentially moving it at the same
1635  * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1636  *
1637  * under NOMMU conditions, we only permit changing a mapping's size, and only
1638  * as long as it stays within the region allocated by do_mmap_private() and the
1639  * block is not shareable
1640  *
1641  * MREMAP_FIXED is not supported under NOMMU conditions
1642  */
1643 unsigned long do_mremap(unsigned long addr,
1644                         unsigned long old_len, unsigned long new_len,
1645                         unsigned long flags, unsigned long new_addr)
1646 {
1647         struct vm_area_struct *vma;
1648
1649         /* insanity checks first */
1650         if (old_len == 0 || new_len == 0)
1651                 return (unsigned long) -EINVAL;
1652
1653         if (addr & ~PAGE_MASK)
1654                 return -EINVAL;
1655
1656         if (flags & MREMAP_FIXED && new_addr != addr)
1657                 return (unsigned long) -EINVAL;
1658
1659         vma = find_vma_exact(current->mm, addr, old_len);
1660         if (!vma)
1661                 return (unsigned long) -EINVAL;
1662
1663         if (vma->vm_end != vma->vm_start + old_len)
1664                 return (unsigned long) -EFAULT;
1665
1666         if (vma->vm_flags & VM_MAYSHARE)
1667                 return (unsigned long) -EPERM;
1668
1669         if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1670                 return (unsigned long) -ENOMEM;
1671
1672         /* all checks complete - do it */
1673         vma->vm_end = vma->vm_start + new_len;
1674         return vma->vm_start;
1675 }
1676 EXPORT_SYMBOL(do_mremap);
1677
1678 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1679                 unsigned long, new_len, unsigned long, flags,
1680                 unsigned long, new_addr)
1681 {
1682         unsigned long ret;
1683
1684         down_write(&current->mm->mmap_sem);
1685         ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1686         up_write(&current->mm->mmap_sem);
1687         return ret;
1688 }
1689
1690 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1691                         unsigned int foll_flags)
1692 {
1693         return NULL;
1694 }
1695
1696 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1697                 unsigned long to, unsigned long size, pgprot_t prot)
1698 {
1699         vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1700         return 0;
1701 }
1702 EXPORT_SYMBOL(remap_pfn_range);
1703
1704 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1705                         unsigned long pgoff)
1706 {
1707         unsigned int size = vma->vm_end - vma->vm_start;
1708
1709         if (!(vma->vm_flags & VM_USERMAP))
1710                 return -EINVAL;
1711
1712         vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1713         vma->vm_end = vma->vm_start + size;
1714
1715         return 0;
1716 }
1717 EXPORT_SYMBOL(remap_vmalloc_range);
1718
1719 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1720 {
1721 }
1722
1723 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1724         unsigned long len, unsigned long pgoff, unsigned long flags)
1725 {
1726         return -ENOMEM;
1727 }
1728
1729 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1730 {
1731 }
1732
1733 void unmap_mapping_range(struct address_space *mapping,
1734                          loff_t const holebegin, loff_t const holelen,
1735                          int even_cows)
1736 {
1737 }
1738 EXPORT_SYMBOL(unmap_mapping_range);
1739
1740 /*
1741  * ask for an unmapped area at which to create a mapping on a file
1742  */
1743 unsigned long get_unmapped_area(struct file *file, unsigned long addr,
1744                                 unsigned long len, unsigned long pgoff,
1745                                 unsigned long flags)
1746 {
1747         unsigned long (*get_area)(struct file *, unsigned long, unsigned long,
1748                                   unsigned long, unsigned long);
1749
1750         get_area = current->mm->get_unmapped_area;
1751         if (file && file->f_op && file->f_op->get_unmapped_area)
1752                 get_area = file->f_op->get_unmapped_area;
1753
1754         if (!get_area)
1755                 return -ENOSYS;
1756
1757         return get_area(file, addr, len, pgoff, flags);
1758 }
1759 EXPORT_SYMBOL(get_unmapped_area);
1760
1761 /*
1762  * Check that a process has enough memory to allocate a new virtual
1763  * mapping. 0 means there is enough memory for the allocation to
1764  * succeed and -ENOMEM implies there is not.
1765  *
1766  * We currently support three overcommit policies, which are set via the
1767  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1768  *
1769  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1770  * Additional code 2002 Jul 20 by Robert Love.
1771  *
1772  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1773  *
1774  * Note this is a helper function intended to be used by LSMs which
1775  * wish to use this logic.
1776  */
1777 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1778 {
1779         unsigned long free, allowed;
1780
1781         vm_acct_memory(pages);
1782
1783         /*
1784          * Sometimes we want to use more memory than we have
1785          */
1786         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1787                 return 0;
1788
1789         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1790                 unsigned long n;
1791
1792                 free = global_page_state(NR_FILE_PAGES);
1793                 free += nr_swap_pages;
1794
1795                 /*
1796                  * Any slabs which are created with the
1797                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1798                  * which are reclaimable, under pressure.  The dentry
1799                  * cache and most inode caches should fall into this
1800                  */
1801                 free += global_page_state(NR_SLAB_RECLAIMABLE);
1802
1803                 /*
1804                  * Leave the last 3% for root
1805                  */
1806                 if (!cap_sys_admin)
1807                         free -= free / 32;
1808
1809                 if (free > pages)
1810                         return 0;
1811
1812                 /*
1813                  * nr_free_pages() is very expensive on large systems,
1814                  * only call if we're about to fail.
1815                  */
1816                 n = nr_free_pages();
1817
1818                 /*
1819                  * Leave reserved pages. The pages are not for anonymous pages.
1820                  */
1821                 if (n <= totalreserve_pages)
1822                         goto error;
1823                 else
1824                         n -= totalreserve_pages;
1825
1826                 /*
1827                  * Leave the last 3% for root
1828                  */
1829                 if (!cap_sys_admin)
1830                         n -= n / 32;
1831                 free += n;
1832
1833                 if (free > pages)
1834                         return 0;
1835
1836                 goto error;
1837         }
1838
1839         allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1840         /*
1841          * Leave the last 3% for root
1842          */
1843         if (!cap_sys_admin)
1844                 allowed -= allowed / 32;
1845         allowed += total_swap_pages;
1846
1847         /* Don't let a single process grow too big:
1848            leave 3% of the size of this process for other processes */
1849         if (mm)
1850                 allowed -= mm->total_vm / 32;
1851
1852         if (percpu_counter_read_positive(&vm_committed_as) < allowed)
1853                 return 0;
1854
1855 error:
1856         vm_unacct_memory(pages);
1857
1858         return -ENOMEM;
1859 }
1860
1861 int in_gate_area_no_task(unsigned long addr)
1862 {
1863         return 0;
1864 }
1865
1866 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1867 {
1868         BUG();
1869         return 0;
1870 }
1871 EXPORT_SYMBOL(filemap_fault);
1872
1873 /*
1874  * Access another process' address space.
1875  * - source/target buffer must be kernel space
1876  */
1877 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1878 {
1879         struct vm_area_struct *vma;
1880         struct mm_struct *mm;
1881
1882         if (addr + len < addr)
1883                 return 0;
1884
1885         mm = get_task_mm(tsk);
1886         if (!mm)
1887                 return 0;
1888
1889         down_read(&mm->mmap_sem);
1890
1891         /* the access must start within one of the target process's mappings */
1892         vma = find_vma(mm, addr);
1893         if (vma) {
1894                 /* don't overrun this mapping */
1895                 if (addr + len >= vma->vm_end)
1896                         len = vma->vm_end - addr;
1897
1898                 /* only read or write mappings where it is permitted */
1899                 if (write && vma->vm_flags & VM_MAYWRITE)
1900                         len -= copy_to_user((void *) addr, buf, len);
1901                 else if (!write && vma->vm_flags & VM_MAYREAD)
1902                         len -= copy_from_user(buf, (void *) addr, len);
1903                 else
1904                         len = 0;
1905         } else {
1906                 len = 0;
1907         }
1908
1909         up_read(&mm->mmap_sem);
1910         mmput(mm);
1911         return len;
1912 }