linux/mm/nommu.c
<<
>>
Prefs
   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) 2007-2010 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/blkdev.h>
  26#include <linux/backing-dev.h>
  27#include <linux/mount.h>
  28#include <linux/personality.h>
  29#include <linux/security.h>
  30#include <linux/syscalls.h>
  31#include <linux/audit.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
  55void *high_memory;
  56struct page *mem_map;
  57unsigned long max_mapnr;
  58unsigned long num_physpages;
  59unsigned long highest_memmap_pfn;
  60struct percpu_counter vm_committed_as;
  61int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
  62int sysctl_overcommit_ratio = 50; /* default is 50% */
  63int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
  64int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
  65int heap_stack_gap = 0;
  66
  67atomic_long_t mmap_pages_allocated;
  68
  69EXPORT_SYMBOL(mem_map);
  70EXPORT_SYMBOL(num_physpages);
  71
  72/* list of mapped, potentially shareable regions */
  73static struct kmem_cache *vm_region_jar;
  74struct rb_root nommu_region_tree = RB_ROOT;
  75DECLARE_RWSEM(nommu_region_sem);
  76
  77const 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 */
  86unsigned 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
 127int __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                     int *retry)
 131{
 132        struct vm_area_struct *vma;
 133        unsigned long vm_flags;
 134        int i;
 135
 136        /* calculate required read or write permissions.
 137         * If FOLL_FORCE is set, we only require the "MAY" flags.
 138         */
 139        vm_flags  = (foll_flags & FOLL_WRITE) ?
 140                        (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
 141        vm_flags &= (foll_flags & FOLL_FORCE) ?
 142                        (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
 143
 144        for (i = 0; i < nr_pages; i++) {
 145                vma = find_vma(mm, start);
 146                if (!vma)
 147                        goto finish_or_fault;
 148
 149                /* protect what we can, including chardevs */
 150                if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
 151                    !(vm_flags & vma->vm_flags))
 152                        goto finish_or_fault;
 153
 154                if (pages) {
 155                        pages[i] = virt_to_page(start);
 156                        if (pages[i])
 157                                page_cache_get(pages[i]);
 158                }
 159                if (vmas)
 160                        vmas[i] = vma;
 161                start = (start + PAGE_SIZE) & PAGE_MASK;
 162        }
 163
 164        return i;
 165
 166finish_or_fault:
 167        return i ? : -EFAULT;
 168}
 169
 170/*
 171 * get a list of pages in an address range belonging to the specified process
 172 * and indicate the VMA that covers each page
 173 * - this is potentially dodgy as we may end incrementing the page count of a
 174 *   slab page or a secondary page from a compound page
 175 * - don't permit access to VMAs that don't support it, such as I/O mappings
 176 */
 177int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
 178        unsigned long start, int nr_pages, int write, int force,
 179        struct page **pages, struct vm_area_struct **vmas)
 180{
 181        int flags = 0;
 182
 183        if (write)
 184                flags |= FOLL_WRITE;
 185        if (force)
 186                flags |= FOLL_FORCE;
 187
 188        return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas,
 189                                NULL);
 190}
 191EXPORT_SYMBOL(get_user_pages);
 192
 193/**
 194 * follow_pfn - look up PFN at a user virtual address
 195 * @vma: memory mapping
 196 * @address: user virtual address
 197 * @pfn: location to store found PFN
 198 *
 199 * Only IO mappings and raw PFN mappings are allowed.
 200 *
 201 * Returns zero and the pfn at @pfn on success, -ve otherwise.
 202 */
 203int follow_pfn(struct vm_area_struct *vma, unsigned long address,
 204        unsigned long *pfn)
 205{
 206        if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
 207                return -EINVAL;
 208
 209        *pfn = address >> PAGE_SHIFT;
 210        return 0;
 211}
 212EXPORT_SYMBOL(follow_pfn);
 213
 214DEFINE_RWLOCK(vmlist_lock);
 215struct vm_struct *vmlist;
 216
 217void vfree(const void *addr)
 218{
 219        kfree(addr);
 220}
 221EXPORT_SYMBOL(vfree);
 222
 223void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
 224{
 225        /*
 226         *  You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
 227         * returns only a logical address.
 228         */
 229        return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
 230}
 231EXPORT_SYMBOL(__vmalloc);
 232
 233void *vmalloc_user(unsigned long size)
 234{
 235        void *ret;
 236
 237        ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
 238                        PAGE_KERNEL);
 239        if (ret) {
 240                struct vm_area_struct *vma;
 241
 242                down_write(&current->mm->mmap_sem);
 243                vma = find_vma(current->mm, (unsigned long)ret);
 244                if (vma)
 245                        vma->vm_flags |= VM_USERMAP;
 246                up_write(&current->mm->mmap_sem);
 247        }
 248
 249        return ret;
 250}
 251EXPORT_SYMBOL(vmalloc_user);
 252
 253struct page *vmalloc_to_page(const void *addr)
 254{
 255        return virt_to_page(addr);
 256}
 257EXPORT_SYMBOL(vmalloc_to_page);
 258
 259unsigned long vmalloc_to_pfn(const void *addr)
 260{
 261        return page_to_pfn(virt_to_page(addr));
 262}
 263EXPORT_SYMBOL(vmalloc_to_pfn);
 264
 265long vread(char *buf, char *addr, unsigned long count)
 266{
 267        memcpy(buf, addr, count);
 268        return count;
 269}
 270
 271long vwrite(char *buf, char *addr, unsigned long count)
 272{
 273        /* Don't allow overflow */
 274        if ((unsigned long) addr + count < count)
 275                count = -(unsigned long) addr;
 276
 277        memcpy(addr, buf, count);
 278        return(count);
 279}
 280
 281/*
 282 *      vmalloc  -  allocate virtually continguos memory
 283 *
 284 *      @size:          allocation size
 285 *
 286 *      Allocate enough pages to cover @size from the page level
 287 *      allocator and map them into continguos kernel virtual space.
 288 *
 289 *      For tight control over page level allocator and protection flags
 290 *      use __vmalloc() instead.
 291 */
 292void *vmalloc(unsigned long size)
 293{
 294       return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
 295}
 296EXPORT_SYMBOL(vmalloc);
 297
 298/*
 299 *      vzalloc - allocate virtually continguos memory with zero fill
 300 *
 301 *      @size:          allocation size
 302 *
 303 *      Allocate enough pages to cover @size from the page level
 304 *      allocator and map them into continguos kernel virtual space.
 305 *      The memory allocated is set to zero.
 306 *
 307 *      For tight control over page level allocator and protection flags
 308 *      use __vmalloc() instead.
 309 */
 310void *vzalloc(unsigned long size)
 311{
 312        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
 313                        PAGE_KERNEL);
 314}
 315EXPORT_SYMBOL(vzalloc);
 316
 317/**
 318 * vmalloc_node - allocate memory on a specific node
 319 * @size:       allocation size
 320 * @node:       numa node
 321 *
 322 * Allocate enough pages to cover @size from the page level
 323 * allocator and map them into contiguous kernel virtual space.
 324 *
 325 * For tight control over page level allocator and protection flags
 326 * use __vmalloc() instead.
 327 */
 328void *vmalloc_node(unsigned long size, int node)
 329{
 330        return vmalloc(size);
 331}
 332EXPORT_SYMBOL(vmalloc_node);
 333
 334/**
 335 * vzalloc_node - allocate memory on a specific node with zero fill
 336 * @size:       allocation size
 337 * @node:       numa node
 338 *
 339 * Allocate enough pages to cover @size from the page level
 340 * allocator and map them into contiguous kernel virtual space.
 341 * The memory allocated is set to zero.
 342 *
 343 * For tight control over page level allocator and protection flags
 344 * use __vmalloc() instead.
 345 */
 346void *vzalloc_node(unsigned long size, int node)
 347{
 348        return vzalloc(size);
 349}
 350EXPORT_SYMBOL(vzalloc_node);
 351
 352#ifndef PAGE_KERNEL_EXEC
 353# define PAGE_KERNEL_EXEC PAGE_KERNEL
 354#endif
 355
 356/**
 357 *      vmalloc_exec  -  allocate virtually contiguous, executable memory
 358 *      @size:          allocation size
 359 *
 360 *      Kernel-internal function to allocate enough pages to cover @size
 361 *      the page level allocator and map them into contiguous and
 362 *      executable kernel virtual space.
 363 *
 364 *      For tight control over page level allocator and protection flags
 365 *      use __vmalloc() instead.
 366 */
 367
 368void *vmalloc_exec(unsigned long size)
 369{
 370        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
 371}
 372
 373/**
 374 * vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
 375 *      @size:          allocation size
 376 *
 377 *      Allocate enough 32bit PA addressable pages to cover @size from the
 378 *      page level allocator and map them into continguos kernel virtual space.
 379 */
 380void *vmalloc_32(unsigned long size)
 381{
 382        return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
 383}
 384EXPORT_SYMBOL(vmalloc_32);
 385
 386/**
 387 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
 388 *      @size:          allocation size
 389 *
 390 * The resulting memory area is 32bit addressable and zeroed so it can be
 391 * mapped to userspace without leaking data.
 392 *
 393 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
 394 * remap_vmalloc_range() are permissible.
 395 */
 396void *vmalloc_32_user(unsigned long size)
 397{
 398        /*
 399         * We'll have to sort out the ZONE_DMA bits for 64-bit,
 400         * but for now this can simply use vmalloc_user() directly.
 401         */
 402        return vmalloc_user(size);
 403}
 404EXPORT_SYMBOL(vmalloc_32_user);
 405
 406void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
 407{
 408        BUG();
 409        return NULL;
 410}
 411EXPORT_SYMBOL(vmap);
 412
 413void vunmap(const void *addr)
 414{
 415        BUG();
 416}
 417EXPORT_SYMBOL(vunmap);
 418
 419void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
 420{
 421        BUG();
 422        return NULL;
 423}
 424EXPORT_SYMBOL(vm_map_ram);
 425
 426void vm_unmap_ram(const void *mem, unsigned int count)
 427{
 428        BUG();
 429}
 430EXPORT_SYMBOL(vm_unmap_ram);
 431
 432void vm_unmap_aliases(void)
 433{
 434}
 435EXPORT_SYMBOL_GPL(vm_unmap_aliases);
 436
 437/*
 438 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
 439 * have one.
 440 */
 441void  __attribute__((weak)) vmalloc_sync_all(void)
 442{
 443}
 444
 445/**
 446 *      alloc_vm_area - allocate a range of kernel address space
 447 *      @size:          size of the area
 448 *
 449 *      Returns:        NULL on failure, vm_struct on success
 450 *
 451 *      This function reserves a range of kernel address space, and
 452 *      allocates pagetables to map that range.  No actual mappings
 453 *      are created.  If the kernel address space is not shared
 454 *      between processes, it syncs the pagetable across all
 455 *      processes.
 456 */
 457struct vm_struct *alloc_vm_area(size_t size)
 458{
 459        BUG();
 460        return NULL;
 461}
 462EXPORT_SYMBOL_GPL(alloc_vm_area);
 463
 464void free_vm_area(struct vm_struct *area)
 465{
 466        BUG();
 467}
 468EXPORT_SYMBOL_GPL(free_vm_area);
 469
 470int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
 471                   struct page *page)
 472{
 473        return -EINVAL;
 474}
 475EXPORT_SYMBOL(vm_insert_page);
 476
 477/*
 478 *  sys_brk() for the most part doesn't need the global kernel
 479 *  lock, except when an application is doing something nasty
 480 *  like trying to un-brk an area that has already been mapped
 481 *  to a regular file.  in this case, the unmapping will need
 482 *  to invoke file system routines that need the global lock.
 483 */
 484SYSCALL_DEFINE1(brk, unsigned long, brk)
 485{
 486        struct mm_struct *mm = current->mm;
 487
 488        if (brk < mm->start_brk || brk > mm->context.end_brk)
 489                return mm->brk;
 490
 491        if (mm->brk == brk)
 492                return mm->brk;
 493
 494        /*
 495         * Always allow shrinking brk
 496         */
 497        if (brk <= mm->brk) {
 498                mm->brk = brk;
 499                return brk;
 500        }
 501
 502        /*
 503         * Ok, looks good - let it rip.
 504         */
 505        flush_icache_range(mm->brk, brk);
 506        return mm->brk = brk;
 507}
 508
 509/*
 510 * initialise the VMA and region record slabs
 511 */
 512void __init mmap_init(void)
 513{
 514        int ret;
 515
 516        ret = percpu_counter_init(&vm_committed_as, 0);
 517        VM_BUG_ON(ret);
 518        vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
 519}
 520
 521/*
 522 * validate the region tree
 523 * - the caller must hold the region lock
 524 */
 525#ifdef CONFIG_DEBUG_NOMMU_REGIONS
 526static noinline void validate_nommu_regions(void)
 527{
 528        struct vm_region *region, *last;
 529        struct rb_node *p, *lastp;
 530
 531        lastp = rb_first(&nommu_region_tree);
 532        if (!lastp)
 533                return;
 534
 535        last = rb_entry(lastp, struct vm_region, vm_rb);
 536        BUG_ON(unlikely(last->vm_end <= last->vm_start));
 537        BUG_ON(unlikely(last->vm_top < last->vm_end));
 538
 539        while ((p = rb_next(lastp))) {
 540                region = rb_entry(p, struct vm_region, vm_rb);
 541                last = rb_entry(lastp, struct vm_region, vm_rb);
 542
 543                BUG_ON(unlikely(region->vm_end <= region->vm_start));
 544                BUG_ON(unlikely(region->vm_top < region->vm_end));
 545                BUG_ON(unlikely(region->vm_start < last->vm_top));
 546
 547                lastp = p;
 548        }
 549}
 550#else
 551static void validate_nommu_regions(void)
 552{
 553}
 554#endif
 555
 556/*
 557 * add a region into the global tree
 558 */
 559static void add_nommu_region(struct vm_region *region)
 560{
 561        struct vm_region *pregion;
 562        struct rb_node **p, *parent;
 563
 564        validate_nommu_regions();
 565
 566        parent = NULL;
 567        p = &nommu_region_tree.rb_node;
 568        while (*p) {
 569                parent = *p;
 570                pregion = rb_entry(parent, struct vm_region, vm_rb);
 571                if (region->vm_start < pregion->vm_start)
 572                        p = &(*p)->rb_left;
 573                else if (region->vm_start > pregion->vm_start)
 574                        p = &(*p)->rb_right;
 575                else if (pregion == region)
 576                        return;
 577                else
 578                        BUG();
 579        }
 580
 581        rb_link_node(&region->vm_rb, parent, p);
 582        rb_insert_color(&region->vm_rb, &nommu_region_tree);
 583
 584        validate_nommu_regions();
 585}
 586
 587/*
 588 * delete a region from the global tree
 589 */
 590static void delete_nommu_region(struct vm_region *region)
 591{
 592        BUG_ON(!nommu_region_tree.rb_node);
 593
 594        validate_nommu_regions();
 595        rb_erase(&region->vm_rb, &nommu_region_tree);
 596        validate_nommu_regions();
 597}
 598
 599/*
 600 * free a contiguous series of pages
 601 */
 602static void free_page_series(unsigned long from, unsigned long to)
 603{
 604        for (; from < to; from += PAGE_SIZE) {
 605                struct page *page = virt_to_page(from);
 606
 607                kdebug("- free %lx", from);
 608                atomic_long_dec(&mmap_pages_allocated);
 609                if (page_count(page) != 1)
 610                        kdebug("free page %p: refcount not one: %d",
 611                               page, page_count(page));
 612                put_page(page);
 613        }
 614}
 615
 616/*
 617 * release a reference to a region
 618 * - the caller must hold the region semaphore for writing, which this releases
 619 * - the region may not have been added to the tree yet, in which case vm_top
 620 *   will equal vm_start
 621 */
 622static void __put_nommu_region(struct vm_region *region)
 623        __releases(nommu_region_sem)
 624{
 625        kenter("%p{%d}", region, region->vm_usage);
 626
 627        BUG_ON(!nommu_region_tree.rb_node);
 628
 629        if (--region->vm_usage == 0) {
 630                if (region->vm_top > region->vm_start)
 631                        delete_nommu_region(region);
 632                up_write(&nommu_region_sem);
 633
 634                if (region->vm_file)
 635                        fput(region->vm_file);
 636
 637                /* IO memory and memory shared directly out of the pagecache
 638                 * from ramfs/tmpfs mustn't be released here */
 639                if (region->vm_flags & VM_MAPPED_COPY) {
 640                        kdebug("free series");
 641                        free_page_series(region->vm_start, region->vm_top);
 642                }
 643                kmem_cache_free(vm_region_jar, region);
 644        } else {
 645                up_write(&nommu_region_sem);
 646        }
 647}
 648
 649/*
 650 * release a reference to a region
 651 */
 652static void put_nommu_region(struct vm_region *region)
 653{
 654        down_write(&nommu_region_sem);
 655        __put_nommu_region(region);
 656}
 657
 658/*
 659 * update protection on a vma
 660 */
 661static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
 662{
 663#ifdef CONFIG_MPU
 664        struct mm_struct *mm = vma->vm_mm;
 665        long start = vma->vm_start & PAGE_MASK;
 666        while (start < vma->vm_end) {
 667                protect_page(mm, start, flags);
 668                start += PAGE_SIZE;
 669        }
 670        update_protections(mm);
 671#endif
 672}
 673
 674/*
 675 * add a VMA into a process's mm_struct in the appropriate place in the list
 676 * and tree and add to the address space's page tree also if not an anonymous
 677 * page
 678 * - should be called with mm->mmap_sem held writelocked
 679 */
 680static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
 681{
 682        struct vm_area_struct *pvma, *prev;
 683        struct address_space *mapping;
 684        struct rb_node **p, *parent, *rb_prev;
 685
 686        kenter(",%p", vma);
 687
 688        BUG_ON(!vma->vm_region);
 689
 690        mm->map_count++;
 691        vma->vm_mm = mm;
 692
 693        protect_vma(vma, vma->vm_flags);
 694
 695        /* add the VMA to the mapping */
 696        if (vma->vm_file) {
 697                mapping = vma->vm_file->f_mapping;
 698
 699                flush_dcache_mmap_lock(mapping);
 700                vma_prio_tree_insert(vma, &mapping->i_mmap);
 701                flush_dcache_mmap_unlock(mapping);
 702        }
 703
 704        /* add the VMA to the tree */
 705        parent = rb_prev = NULL;
 706        p = &mm->mm_rb.rb_node;
 707        while (*p) {
 708                parent = *p;
 709                pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
 710
 711                /* sort by: start addr, end addr, VMA struct addr in that order
 712                 * (the latter is necessary as we may get identical VMAs) */
 713                if (vma->vm_start < pvma->vm_start)
 714                        p = &(*p)->rb_left;
 715                else if (vma->vm_start > pvma->vm_start) {
 716                        rb_prev = parent;
 717                        p = &(*p)->rb_right;
 718                } else if (vma->vm_end < pvma->vm_end)
 719                        p = &(*p)->rb_left;
 720                else if (vma->vm_end > pvma->vm_end) {
 721                        rb_prev = parent;
 722                        p = &(*p)->rb_right;
 723                } else if (vma < pvma)
 724                        p = &(*p)->rb_left;
 725                else if (vma > pvma) {
 726                        rb_prev = parent;
 727                        p = &(*p)->rb_right;
 728                } else
 729                        BUG();
 730        }
 731
 732        rb_link_node(&vma->vm_rb, parent, p);
 733        rb_insert_color(&vma->vm_rb, &mm->mm_rb);
 734
 735        /* add VMA to the VMA list also */
 736        prev = NULL;
 737        if (rb_prev)
 738                prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
 739
 740        __vma_link_list(mm, vma, prev, parent);
 741}
 742
 743/*
 744 * delete a VMA from its owning mm_struct and address space
 745 */
 746static void delete_vma_from_mm(struct vm_area_struct *vma)
 747{
 748        struct address_space *mapping;
 749        struct mm_struct *mm = vma->vm_mm;
 750
 751        kenter("%p", vma);
 752
 753        protect_vma(vma, 0);
 754
 755        mm->map_count--;
 756        if (mm->mmap_cache == vma)
 757                mm->mmap_cache = NULL;
 758
 759        /* remove the VMA from the mapping */
 760        if (vma->vm_file) {
 761                mapping = vma->vm_file->f_mapping;
 762
 763                flush_dcache_mmap_lock(mapping);
 764                vma_prio_tree_remove(vma, &mapping->i_mmap);
 765                flush_dcache_mmap_unlock(mapping);
 766        }
 767
 768        /* remove from the MM's tree and list */
 769        rb_erase(&vma->vm_rb, &mm->mm_rb);
 770
 771        if (vma->vm_prev)
 772                vma->vm_prev->vm_next = vma->vm_next;
 773        else
 774                mm->mmap = vma->vm_next;
 775
 776        if (vma->vm_next)
 777                vma->vm_next->vm_prev = vma->vm_prev;
 778
 779        vma->vm_mm = NULL;
 780}
 781
 782/*
 783 * destroy a VMA record
 784 */
 785static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
 786{
 787        kenter("%p", vma);
 788        if (vma->vm_ops && vma->vm_ops->close)
 789                vma->vm_ops->close(vma);
 790        if (vma->vm_file) {
 791                fput(vma->vm_file);
 792                if (vma->vm_flags & VM_EXECUTABLE)
 793                        removed_exe_file_vma(mm);
 794        }
 795        put_nommu_region(vma->vm_region);
 796        kmem_cache_free(vm_area_cachep, vma);
 797}
 798
 799/*
 800 * look up the first VMA in which addr resides, NULL if none
 801 * - should be called with mm->mmap_sem at least held readlocked
 802 */
 803struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
 804{
 805        struct vm_area_struct *vma;
 806
 807        /* check the cache first */
 808        vma = mm->mmap_cache;
 809        if (vma && vma->vm_start <= addr && vma->vm_end > addr)
 810                return vma;
 811
 812        /* trawl the list (there may be multiple mappings in which addr
 813         * resides) */
 814        for (vma = mm->mmap; vma; vma = vma->vm_next) {
 815                if (vma->vm_start > addr)
 816                        return NULL;
 817                if (vma->vm_end > addr) {
 818                        mm->mmap_cache = vma;
 819                        return vma;
 820                }
 821        }
 822
 823        return NULL;
 824}
 825EXPORT_SYMBOL(find_vma);
 826
 827/*
 828 * find a VMA
 829 * - we don't extend stack VMAs under NOMMU conditions
 830 */
 831struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
 832{
 833        return find_vma(mm, addr);
 834}
 835
 836/*
 837 * expand a stack to a given address
 838 * - not supported under NOMMU conditions
 839 */
 840int expand_stack(struct vm_area_struct *vma, unsigned long address)
 841{
 842        return -ENOMEM;
 843}
 844
 845/*
 846 * look up the first VMA exactly that exactly matches addr
 847 * - should be called with mm->mmap_sem at least held readlocked
 848 */
 849static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
 850                                             unsigned long addr,
 851                                             unsigned long len)
 852{
 853        struct vm_area_struct *vma;
 854        unsigned long end = addr + len;
 855
 856        /* check the cache first */
 857        vma = mm->mmap_cache;
 858        if (vma && vma->vm_start == addr && vma->vm_end == end)
 859                return vma;
 860
 861        /* trawl the list (there may be multiple mappings in which addr
 862         * resides) */
 863        for (vma = mm->mmap; vma; vma = vma->vm_next) {
 864                if (vma->vm_start < addr)
 865                        continue;
 866                if (vma->vm_start > addr)
 867                        return NULL;
 868                if (vma->vm_end == end) {
 869                        mm->mmap_cache = vma;
 870                        return vma;
 871                }
 872        }
 873
 874        return NULL;
 875}
 876
 877/*
 878 * determine whether a mapping should be permitted and, if so, what sort of
 879 * mapping we're capable of supporting
 880 */
 881static int validate_mmap_request(struct file *file,
 882                                 unsigned long addr,
 883                                 unsigned long len,
 884                                 unsigned long prot,
 885                                 unsigned long flags,
 886                                 unsigned long pgoff,
 887                                 unsigned long *_capabilities)
 888{
 889        unsigned long capabilities, rlen;
 890        unsigned long reqprot = prot;
 891        int ret;
 892
 893        /* do the simple checks first */
 894        if (flags & MAP_FIXED) {
 895                printk(KERN_DEBUG
 896                       "%d: Can't do fixed-address/overlay mmap of RAM\n",
 897                       current->pid);
 898                return -EINVAL;
 899        }
 900
 901        if ((flags & MAP_TYPE) != MAP_PRIVATE &&
 902            (flags & MAP_TYPE) != MAP_SHARED)
 903                return -EINVAL;
 904
 905        if (!len)
 906                return -EINVAL;
 907
 908        /* Careful about overflows.. */
 909        rlen = PAGE_ALIGN(len);
 910        if (!rlen || rlen > TASK_SIZE)
 911                return -ENOMEM;
 912
 913        /* offset overflow? */
 914        if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
 915                return -EOVERFLOW;
 916
 917        if (file) {
 918                /* validate file mapping requests */
 919                struct address_space *mapping;
 920
 921                /* files must support mmap */
 922                if (!file->f_op || !file->f_op->mmap)
 923                        return -ENODEV;
 924
 925                /* work out if what we've got could possibly be shared
 926                 * - we support chardevs that provide their own "memory"
 927                 * - we support files/blockdevs that are memory backed
 928                 */
 929                mapping = file->f_mapping;
 930                if (!mapping)
 931                        mapping = file->f_path.dentry->d_inode->i_mapping;
 932
 933                capabilities = 0;
 934                if (mapping && mapping->backing_dev_info)
 935                        capabilities = mapping->backing_dev_info->capabilities;
 936
 937                if (!capabilities) {
 938                        /* no explicit capabilities set, so assume some
 939                         * defaults */
 940                        switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
 941                        case S_IFREG:
 942                        case S_IFBLK:
 943                                capabilities = BDI_CAP_MAP_COPY;
 944                                break;
 945
 946                        case S_IFCHR:
 947                                capabilities =
 948                                        BDI_CAP_MAP_DIRECT |
 949                                        BDI_CAP_READ_MAP |
 950                                        BDI_CAP_WRITE_MAP;
 951                                break;
 952
 953                        default:
 954                                return -EINVAL;
 955                        }
 956                }
 957
 958                /* eliminate any capabilities that we can't support on this
 959                 * device */
 960                if (!file->f_op->get_unmapped_area)
 961                        capabilities &= ~BDI_CAP_MAP_DIRECT;
 962                if (!file->f_op->read)
 963                        capabilities &= ~BDI_CAP_MAP_COPY;
 964
 965                /* The file shall have been opened with read permission. */
 966                if (!(file->f_mode & FMODE_READ))
 967                        return -EACCES;
 968
 969                if (flags & MAP_SHARED) {
 970                        /* do checks for writing, appending and locking */
 971                        if ((prot & PROT_WRITE) &&
 972                            !(file->f_mode & FMODE_WRITE))
 973                                return -EACCES;
 974
 975                        if (IS_APPEND(file->f_path.dentry->d_inode) &&
 976                            (file->f_mode & FMODE_WRITE))
 977                                return -EACCES;
 978
 979                        if (locks_verify_locked(file->f_path.dentry->d_inode))
 980                                return -EAGAIN;
 981
 982                        if (!(capabilities & BDI_CAP_MAP_DIRECT))
 983                                return -ENODEV;
 984
 985                        /* we mustn't privatise shared mappings */
 986                        capabilities &= ~BDI_CAP_MAP_COPY;
 987                }
 988                else {
 989                        /* we're going to read the file into private memory we
 990                         * allocate */
 991                        if (!(capabilities & BDI_CAP_MAP_COPY))
 992                                return -ENODEV;
 993
 994                        /* we don't permit a private writable mapping to be
 995                         * shared with the backing device */
 996                        if (prot & PROT_WRITE)
 997                                capabilities &= ~BDI_CAP_MAP_DIRECT;
 998                }
 999
1000                if (capabilities & BDI_CAP_MAP_DIRECT) {
1001                        if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
1002                            ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
1003                            ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
1004                            ) {
1005                                capabilities &= ~BDI_CAP_MAP_DIRECT;
1006                                if (flags & MAP_SHARED) {
1007                                        printk(KERN_WARNING
1008                                               "MAP_SHARED not completely supported on !MMU\n");
1009                                        return -EINVAL;
1010                                }
1011                        }
1012                }
1013
1014                /* handle executable mappings and implied executable
1015                 * mappings */
1016                if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1017                        if (prot & PROT_EXEC)
1018                                return -EPERM;
1019                }
1020                else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
1021                        /* handle implication of PROT_EXEC by PROT_READ */
1022                        if (current->personality & READ_IMPLIES_EXEC) {
1023                                if (capabilities & BDI_CAP_EXEC_MAP)
1024                                        prot |= PROT_EXEC;
1025                        }
1026                }
1027                else if ((prot & PROT_READ) &&
1028                         (prot & PROT_EXEC) &&
1029                         !(capabilities & BDI_CAP_EXEC_MAP)
1030                         ) {
1031                        /* backing file is not executable, try to copy */
1032                        capabilities &= ~BDI_CAP_MAP_DIRECT;
1033                }
1034        }
1035        else {
1036                /* anonymous mappings are always memory backed and can be
1037                 * privately mapped
1038                 */
1039                capabilities = BDI_CAP_MAP_COPY;
1040
1041                /* handle PROT_EXEC implication by PROT_READ */
1042                if ((prot & PROT_READ) &&
1043                    (current->personality & READ_IMPLIES_EXEC))
1044                        prot |= PROT_EXEC;
1045        }
1046
1047        /* allow the security API to have its say */
1048        ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
1049        if (ret < 0)
1050                return ret;
1051
1052        /* looks okay */
1053        *_capabilities = capabilities;
1054        return 0;
1055}
1056
1057/*
1058 * we've determined that we can make the mapping, now translate what we
1059 * now know into VMA flags
1060 */
1061static unsigned long determine_vm_flags(struct file *file,
1062                                        unsigned long prot,
1063                                        unsigned long flags,
1064                                        unsigned long capabilities)
1065{
1066        unsigned long vm_flags;
1067
1068        vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
1069        /* vm_flags |= mm->def_flags; */
1070
1071        if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
1072                /* attempt to share read-only copies of mapped file chunks */
1073                vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1074                if (file && !(prot & PROT_WRITE))
1075                        vm_flags |= VM_MAYSHARE;
1076        } else {
1077                /* overlay a shareable mapping on the backing device or inode
1078                 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1079                 * romfs/cramfs */
1080                vm_flags |= VM_MAYSHARE | (capabilities & BDI_CAP_VMFLAGS);
1081                if (flags & MAP_SHARED)
1082                        vm_flags |= VM_SHARED;
1083        }
1084
1085        /* refuse to let anyone share private mappings with this process if
1086         * it's being traced - otherwise breakpoints set in it may interfere
1087         * with another untraced process
1088         */
1089        if ((flags & MAP_PRIVATE) && current->ptrace)
1090                vm_flags &= ~VM_MAYSHARE;
1091
1092        return vm_flags;
1093}
1094
1095/*
1096 * set up a shared mapping on a file (the driver or filesystem provides and
1097 * pins the storage)
1098 */
1099static int do_mmap_shared_file(struct vm_area_struct *vma)
1100{
1101        int ret;
1102
1103        ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1104        if (ret == 0) {
1105                vma->vm_region->vm_top = vma->vm_region->vm_end;
1106                return 0;
1107        }
1108        if (ret != -ENOSYS)
1109                return ret;
1110
1111        /* getting -ENOSYS indicates that direct mmap isn't possible (as
1112         * opposed to tried but failed) so we can only give a suitable error as
1113         * it's not possible to make a private copy if MAP_SHARED was given */
1114        return -ENODEV;
1115}
1116
1117/*
1118 * set up a private mapping or an anonymous shared mapping
1119 */
1120static int do_mmap_private(struct vm_area_struct *vma,
1121                           struct vm_region *region,
1122                           unsigned long len,
1123                           unsigned long capabilities)
1124{
1125        struct page *pages;
1126        unsigned long total, point, n;
1127        void *base;
1128        int ret, order;
1129
1130        /* invoke the file's mapping function so that it can keep track of
1131         * shared mappings on devices or memory
1132         * - VM_MAYSHARE will be set if it may attempt to share
1133         */
1134        if (capabilities & BDI_CAP_MAP_DIRECT) {
1135                ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1136                if (ret == 0) {
1137                        /* shouldn't return success if we're not sharing */
1138                        BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1139                        vma->vm_region->vm_top = vma->vm_region->vm_end;
1140                        return 0;
1141                }
1142                if (ret != -ENOSYS)
1143                        return ret;
1144
1145                /* getting an ENOSYS error indicates that direct mmap isn't
1146                 * possible (as opposed to tried but failed) so we'll try to
1147                 * make a private copy of the data and map that instead */
1148        }
1149
1150
1151        /* allocate some memory to hold the mapping
1152         * - note that this may not return a page-aligned address if the object
1153         *   we're allocating is smaller than a page
1154         */
1155        order = get_order(len);
1156        kdebug("alloc order %d for %lx", order, len);
1157
1158        pages = alloc_pages(GFP_KERNEL, order);
1159        if (!pages)
1160                goto enomem;
1161
1162        total = 1 << order;
1163        atomic_long_add(total, &mmap_pages_allocated);
1164
1165        point = len >> PAGE_SHIFT;
1166
1167        /* we allocated a power-of-2 sized page set, so we may want to trim off
1168         * the excess */
1169        if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
1170                while (total > point) {
1171                        order = ilog2(total - point);
1172                        n = 1 << order;
1173                        kdebug("shave %lu/%lu @%lu", n, total - point, total);
1174                        atomic_long_sub(n, &mmap_pages_allocated);
1175                        total -= n;
1176                        set_page_refcounted(pages + total);
1177                        __free_pages(pages + total, order);
1178                }
1179        }
1180
1181        for (point = 1; point < total; point++)
1182                set_page_refcounted(&pages[point]);
1183
1184        base = page_address(pages);
1185        region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1186        region->vm_start = (unsigned long) base;
1187        region->vm_end   = region->vm_start + len;
1188        region->vm_top   = region->vm_start + (total << PAGE_SHIFT);
1189
1190        vma->vm_start = region->vm_start;
1191        vma->vm_end   = region->vm_start + len;
1192
1193        if (vma->vm_file) {
1194                /* read the contents of a file into the copy */
1195                mm_segment_t old_fs;
1196                loff_t fpos;
1197
1198                fpos = vma->vm_pgoff;
1199                fpos <<= PAGE_SHIFT;
1200
1201                old_fs = get_fs();
1202                set_fs(KERNEL_DS);
1203                ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
1204                set_fs(old_fs);
1205
1206                if (ret < 0)
1207                        goto error_free;
1208
1209                /* clear the last little bit */
1210                if (ret < len)
1211                        memset(base + ret, 0, len - ret);
1212
1213        }
1214
1215        return 0;
1216
1217error_free:
1218        free_page_series(region->vm_start, region->vm_top);
1219        region->vm_start = vma->vm_start = 0;
1220        region->vm_end   = vma->vm_end = 0;
1221        region->vm_top   = 0;
1222        return ret;
1223
1224enomem:
1225        printk("Allocation of length %lu from process %d (%s) failed\n",
1226               len, current->pid, current->comm);
1227        show_free_areas(0);
1228        return -ENOMEM;
1229}
1230
1231/*
1232 * handle mapping creation for uClinux
1233 */
1234unsigned long do_mmap_pgoff(struct file *file,
1235                            unsigned long addr,
1236                            unsigned long len,
1237                            unsigned long prot,
1238                            unsigned long flags,
1239                            unsigned long pgoff)
1240{
1241        struct vm_area_struct *vma;
1242        struct vm_region *region;
1243        struct rb_node *rb;
1244        unsigned long capabilities, vm_flags, result;
1245        int ret;
1246
1247        kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1248
1249        /* decide whether we should attempt the mapping, and if so what sort of
1250         * mapping */
1251        ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1252                                    &capabilities);
1253        if (ret < 0) {
1254                kleave(" = %d [val]", ret);
1255                return ret;
1256        }
1257
1258        /* we ignore the address hint */
1259        addr = 0;
1260        len = PAGE_ALIGN(len);
1261
1262        /* we've determined that we can make the mapping, now translate what we
1263         * now know into VMA flags */
1264        vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1265
1266        /* we're going to need to record the mapping */
1267        region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1268        if (!region)
1269                goto error_getting_region;
1270
1271        vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1272        if (!vma)
1273                goto error_getting_vma;
1274
1275        region->vm_usage = 1;
1276        region->vm_flags = vm_flags;
1277        region->vm_pgoff = pgoff;
1278
1279        INIT_LIST_HEAD(&vma->anon_vma_chain);
1280        vma->vm_flags = vm_flags;
1281        vma->vm_pgoff = pgoff;
1282
1283        if (file) {
1284                region->vm_file = file;
1285                get_file(file);
1286                vma->vm_file = file;
1287                get_file(file);
1288                if (vm_flags & VM_EXECUTABLE) {
1289                        added_exe_file_vma(current->mm);
1290                        vma->vm_mm = current->mm;
1291                }
1292        }
1293
1294        down_write(&nommu_region_sem);
1295
1296        /* if we want to share, we need to check for regions created by other
1297         * mmap() calls that overlap with our proposed mapping
1298         * - we can only share with a superset match on most regular files
1299         * - shared mappings on character devices and memory backed files are
1300         *   permitted to overlap inexactly as far as we are concerned for in
1301         *   these cases, sharing is handled in the driver or filesystem rather
1302         *   than here
1303         */
1304        if (vm_flags & VM_MAYSHARE) {
1305                struct vm_region *pregion;
1306                unsigned long pglen, rpglen, pgend, rpgend, start;
1307
1308                pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1309                pgend = pgoff + pglen;
1310
1311                for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1312                        pregion = rb_entry(rb, struct vm_region, vm_rb);
1313
1314                        if (!(pregion->vm_flags & VM_MAYSHARE))
1315                                continue;
1316
1317                        /* search for overlapping mappings on the same file */
1318                        if (pregion->vm_file->f_path.dentry->d_inode !=
1319                            file->f_path.dentry->d_inode)
1320                                continue;
1321
1322                        if (pregion->vm_pgoff >= pgend)
1323                                continue;
1324
1325                        rpglen = pregion->vm_end - pregion->vm_start;
1326                        rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1327                        rpgend = pregion->vm_pgoff + rpglen;
1328                        if (pgoff >= rpgend)
1329                                continue;
1330
1331                        /* handle inexactly overlapping matches between
1332                         * mappings */
1333                        if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1334                            !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1335                                /* new mapping is not a subset of the region */
1336                                if (!(capabilities & BDI_CAP_MAP_DIRECT))
1337                                        goto sharing_violation;
1338                                continue;
1339                        }
1340
1341                        /* we've found a region we can share */
1342                        pregion->vm_usage++;
1343                        vma->vm_region = pregion;
1344                        start = pregion->vm_start;
1345                        start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1346                        vma->vm_start = start;
1347                        vma->vm_end = start + len;
1348
1349                        if (pregion->vm_flags & VM_MAPPED_COPY) {
1350                                kdebug("share copy");
1351                                vma->vm_flags |= VM_MAPPED_COPY;
1352                        } else {
1353                                kdebug("share mmap");
1354                                ret = do_mmap_shared_file(vma);
1355                                if (ret < 0) {
1356                                        vma->vm_region = NULL;
1357                                        vma->vm_start = 0;
1358                                        vma->vm_end = 0;
1359                                        pregion->vm_usage--;
1360                                        pregion = NULL;
1361                                        goto error_just_free;
1362                                }
1363                        }
1364                        fput(region->vm_file);
1365                        kmem_cache_free(vm_region_jar, region);
1366                        region = pregion;
1367                        result = start;
1368                        goto share;
1369                }
1370
1371                /* obtain the address at which to make a shared mapping
1372                 * - this is the hook for quasi-memory character devices to
1373                 *   tell us the location of a shared mapping
1374                 */
1375                if (capabilities & BDI_CAP_MAP_DIRECT) {
1376                        addr = file->f_op->get_unmapped_area(file, addr, len,
1377                                                             pgoff, flags);
1378                        if (IS_ERR_VALUE(addr)) {
1379                                ret = addr;
1380                                if (ret != -ENOSYS)
1381                                        goto error_just_free;
1382
1383                                /* the driver refused to tell us where to site
1384                                 * the mapping so we'll have to attempt to copy
1385                                 * it */
1386                                ret = -ENODEV;
1387                                if (!(capabilities & BDI_CAP_MAP_COPY))
1388                                        goto error_just_free;
1389
1390                                capabilities &= ~BDI_CAP_MAP_DIRECT;
1391                        } else {
1392                                vma->vm_start = region->vm_start = addr;
1393                                vma->vm_end = region->vm_end = addr + len;
1394                        }
1395                }
1396        }
1397
1398        vma->vm_region = region;
1399
1400        /* set up the mapping
1401         * - the region is filled in if BDI_CAP_MAP_DIRECT is still set
1402         */
1403        if (file && vma->vm_flags & VM_SHARED)
1404                ret = do_mmap_shared_file(vma);
1405        else
1406                ret = do_mmap_private(vma, region, len, capabilities);
1407        if (ret < 0)
1408                goto error_just_free;
1409        add_nommu_region(region);
1410
1411        /* clear anonymous mappings that don't ask for uninitialized data */
1412        if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1413                memset((void *)region->vm_start, 0,
1414                       region->vm_end - region->vm_start);
1415
1416        /* okay... we have a mapping; now we have to register it */
1417        result = vma->vm_start;
1418
1419        current->mm->total_vm += len >> PAGE_SHIFT;
1420
1421share:
1422        add_vma_to_mm(current->mm, vma);
1423
1424        /* we flush the region from the icache only when the first executable
1425         * mapping of it is made  */
1426        if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1427                flush_icache_range(region->vm_start, region->vm_end);
1428                region->vm_icache_flushed = true;
1429        }
1430
1431        up_write(&nommu_region_sem);
1432
1433        kleave(" = %lx", result);
1434        return result;
1435
1436error_just_free:
1437        up_write(&nommu_region_sem);
1438error:
1439        if (region->vm_file)
1440                fput(region->vm_file);
1441        kmem_cache_free(vm_region_jar, region);
1442        if (vma->vm_file)
1443                fput(vma->vm_file);
1444        if (vma->vm_flags & VM_EXECUTABLE)
1445                removed_exe_file_vma(vma->vm_mm);
1446        kmem_cache_free(vm_area_cachep, vma);
1447        kleave(" = %d", ret);
1448        return ret;
1449
1450sharing_violation:
1451        up_write(&nommu_region_sem);
1452        printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1453        ret = -EINVAL;
1454        goto error;
1455
1456error_getting_vma:
1457        kmem_cache_free(vm_region_jar, region);
1458        printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1459               " from process %d failed\n",
1460               len, current->pid);
1461        show_free_areas(0);
1462        return -ENOMEM;
1463
1464error_getting_region:
1465        printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1466               " from process %d failed\n",
1467               len, current->pid);
1468        show_free_areas(0);
1469        return -ENOMEM;
1470}
1471EXPORT_SYMBOL(do_mmap_pgoff);
1472
1473SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1474                unsigned long, prot, unsigned long, flags,
1475                unsigned long, fd, unsigned long, pgoff)
1476{
1477        struct file *file = NULL;
1478        unsigned long retval = -EBADF;
1479
1480        audit_mmap_fd(fd, flags);
1481        if (!(flags & MAP_ANONYMOUS)) {
1482                file = fget(fd);
1483                if (!file)
1484                        goto out;
1485        }
1486
1487        flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1488
1489        down_write(&current->mm->mmap_sem);
1490        retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1491        up_write(&current->mm->mmap_sem);
1492
1493        if (file)
1494                fput(file);
1495out:
1496        return retval;
1497}
1498
1499#ifdef __ARCH_WANT_SYS_OLD_MMAP
1500struct mmap_arg_struct {
1501        unsigned long addr;
1502        unsigned long len;
1503        unsigned long prot;
1504        unsigned long flags;
1505        unsigned long fd;
1506        unsigned long offset;
1507};
1508
1509SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1510{
1511        struct mmap_arg_struct a;
1512
1513        if (copy_from_user(&a, arg, sizeof(a)))
1514                return -EFAULT;
1515        if (a.offset & ~PAGE_MASK)
1516                return -EINVAL;
1517
1518        return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1519                              a.offset >> PAGE_SHIFT);
1520}
1521#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1522
1523/*
1524 * split a vma into two pieces at address 'addr', a new vma is allocated either
1525 * for the first part or the tail.
1526 */
1527int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1528              unsigned long addr, int new_below)
1529{
1530        struct vm_area_struct *new;
1531        struct vm_region *region;
1532        unsigned long npages;
1533
1534        kenter("");
1535
1536        /* we're only permitted to split anonymous regions (these should have
1537         * only a single usage on the region) */
1538        if (vma->vm_file)
1539                return -ENOMEM;
1540
1541        if (mm->map_count >= sysctl_max_map_count)
1542                return -ENOMEM;
1543
1544        region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1545        if (!region)
1546                return -ENOMEM;
1547
1548        new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1549        if (!new) {
1550                kmem_cache_free(vm_region_jar, region);
1551                return -ENOMEM;
1552        }
1553
1554        /* most fields are the same, copy all, and then fixup */
1555        *new = *vma;
1556        *region = *vma->vm_region;
1557        new->vm_region = region;
1558
1559        npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1560
1561        if (new_below) {
1562                region->vm_top = region->vm_end = new->vm_end = addr;
1563        } else {
1564                region->vm_start = new->vm_start = addr;
1565                region->vm_pgoff = new->vm_pgoff += npages;
1566        }
1567
1568        if (new->vm_ops && new->vm_ops->open)
1569                new->vm_ops->open(new);
1570
1571        delete_vma_from_mm(vma);
1572        down_write(&nommu_region_sem);
1573        delete_nommu_region(vma->vm_region);
1574        if (new_below) {
1575                vma->vm_region->vm_start = vma->vm_start = addr;
1576                vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1577        } else {
1578                vma->vm_region->vm_end = vma->vm_end = addr;
1579                vma->vm_region->vm_top = addr;
1580        }
1581        add_nommu_region(vma->vm_region);
1582        add_nommu_region(new->vm_region);
1583        up_write(&nommu_region_sem);
1584        add_vma_to_mm(mm, vma);
1585        add_vma_to_mm(mm, new);
1586        return 0;
1587}
1588
1589/*
1590 * shrink a VMA by removing the specified chunk from either the beginning or
1591 * the end
1592 */
1593static int shrink_vma(struct mm_struct *mm,
1594                      struct vm_area_struct *vma,
1595                      unsigned long from, unsigned long to)
1596{
1597        struct vm_region *region;
1598
1599        kenter("");
1600
1601        /* adjust the VMA's pointers, which may reposition it in the MM's tree
1602         * and list */
1603        delete_vma_from_mm(vma);
1604        if (from > vma->vm_start)
1605                vma->vm_end = from;
1606        else
1607                vma->vm_start = to;
1608        add_vma_to_mm(mm, vma);
1609
1610        /* cut the backing region down to size */
1611        region = vma->vm_region;
1612        BUG_ON(region->vm_usage != 1);
1613
1614        down_write(&nommu_region_sem);
1615        delete_nommu_region(region);
1616        if (from > region->vm_start) {
1617                to = region->vm_top;
1618                region->vm_top = region->vm_end = from;
1619        } else {
1620                region->vm_start = to;
1621        }
1622        add_nommu_region(region);
1623        up_write(&nommu_region_sem);
1624
1625        free_page_series(from, to);
1626        return 0;
1627}
1628
1629/*
1630 * release a mapping
1631 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1632 *   VMA, though it need not cover the whole VMA
1633 */
1634int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1635{
1636        struct vm_area_struct *vma;
1637        unsigned long end;
1638        int ret;
1639
1640        kenter(",%lx,%zx", start, len);
1641
1642        len = PAGE_ALIGN(len);
1643        if (len == 0)
1644                return -EINVAL;
1645
1646        end = start + len;
1647
1648        /* find the first potentially overlapping VMA */
1649        vma = find_vma(mm, start);
1650        if (!vma) {
1651                static int limit = 0;
1652                if (limit < 5) {
1653                        printk(KERN_WARNING
1654                               "munmap of memory not mmapped by process %d"
1655                               " (%s): 0x%lx-0x%lx\n",
1656                               current->pid, current->comm,
1657                               start, start + len - 1);
1658                        limit++;
1659                }
1660                return -EINVAL;
1661        }
1662
1663        /* we're allowed to split an anonymous VMA but not a file-backed one */
1664        if (vma->vm_file) {
1665                do {
1666                        if (start > vma->vm_start) {
1667                                kleave(" = -EINVAL [miss]");
1668                                return -EINVAL;
1669                        }
1670                        if (end == vma->vm_end)
1671                                goto erase_whole_vma;
1672                        vma = vma->vm_next;
1673                } while (vma);
1674                kleave(" = -EINVAL [split file]");
1675                return -EINVAL;
1676        } else {
1677                /* the chunk must be a subset of the VMA found */
1678                if (start == vma->vm_start && end == vma->vm_end)
1679                        goto erase_whole_vma;
1680                if (start < vma->vm_start || end > vma->vm_end) {
1681                        kleave(" = -EINVAL [superset]");
1682                        return -EINVAL;
1683                }
1684                if (start & ~PAGE_MASK) {
1685                        kleave(" = -EINVAL [unaligned start]");
1686                        return -EINVAL;
1687                }
1688                if (end != vma->vm_end && end & ~PAGE_MASK) {
1689                        kleave(" = -EINVAL [unaligned split]");
1690                        return -EINVAL;
1691                }
1692                if (start != vma->vm_start && end != vma->vm_end) {
1693                        ret = split_vma(mm, vma, start, 1);
1694                        if (ret < 0) {
1695                                kleave(" = %d [split]", ret);
1696                                return ret;
1697                        }
1698                }
1699                return shrink_vma(mm, vma, start, end);
1700        }
1701
1702erase_whole_vma:
1703        delete_vma_from_mm(vma);
1704        delete_vma(mm, vma);
1705        kleave(" = 0");
1706        return 0;
1707}
1708EXPORT_SYMBOL(do_munmap);
1709
1710SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1711{
1712        int ret;
1713        struct mm_struct *mm = current->mm;
1714
1715        down_write(&mm->mmap_sem);
1716        ret = do_munmap(mm, addr, len);
1717        up_write(&mm->mmap_sem);
1718        return ret;
1719}
1720
1721/*
1722 * release all the mappings made in a process's VM space
1723 */
1724void exit_mmap(struct mm_struct *mm)
1725{
1726        struct vm_area_struct *vma;
1727
1728        if (!mm)
1729                return;
1730
1731        kenter("");
1732
1733        mm->total_vm = 0;
1734
1735        while ((vma = mm->mmap)) {
1736                mm->mmap = vma->vm_next;
1737                delete_vma_from_mm(vma);
1738                delete_vma(mm, vma);
1739                cond_resched();
1740        }
1741
1742        kleave("");
1743}
1744
1745unsigned long do_brk(unsigned long addr, unsigned long len)
1746{
1747        return -ENOMEM;
1748}
1749
1750/*
1751 * expand (or shrink) an existing mapping, potentially moving it at the same
1752 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1753 *
1754 * under NOMMU conditions, we only permit changing a mapping's size, and only
1755 * as long as it stays within the region allocated by do_mmap_private() and the
1756 * block is not shareable
1757 *
1758 * MREMAP_FIXED is not supported under NOMMU conditions
1759 */
1760unsigned long do_mremap(unsigned long addr,
1761                        unsigned long old_len, unsigned long new_len,
1762                        unsigned long flags, unsigned long new_addr)
1763{
1764        struct vm_area_struct *vma;
1765
1766        /* insanity checks first */
1767        old_len = PAGE_ALIGN(old_len);
1768        new_len = PAGE_ALIGN(new_len);
1769        if (old_len == 0 || new_len == 0)
1770                return (unsigned long) -EINVAL;
1771
1772        if (addr & ~PAGE_MASK)
1773                return -EINVAL;
1774
1775        if (flags & MREMAP_FIXED && new_addr != addr)
1776                return (unsigned long) -EINVAL;
1777
1778        vma = find_vma_exact(current->mm, addr, old_len);
1779        if (!vma)
1780                return (unsigned long) -EINVAL;
1781
1782        if (vma->vm_end != vma->vm_start + old_len)
1783                return (unsigned long) -EFAULT;
1784
1785        if (vma->vm_flags & VM_MAYSHARE)
1786                return (unsigned long) -EPERM;
1787
1788        if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1789                return (unsigned long) -ENOMEM;
1790
1791        /* all checks complete - do it */
1792        vma->vm_end = vma->vm_start + new_len;
1793        return vma->vm_start;
1794}
1795EXPORT_SYMBOL(do_mremap);
1796
1797SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1798                unsigned long, new_len, unsigned long, flags,
1799                unsigned long, new_addr)
1800{
1801        unsigned long ret;
1802
1803        down_write(&current->mm->mmap_sem);
1804        ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1805        up_write(&current->mm->mmap_sem);
1806        return ret;
1807}
1808
1809struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1810                        unsigned int foll_flags)
1811{
1812        return NULL;
1813}
1814
1815int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1816                unsigned long pfn, unsigned long size, pgprot_t prot)
1817{
1818        if (addr != (pfn << PAGE_SHIFT))
1819                return -EINVAL;
1820
1821        vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
1822        return 0;
1823}
1824EXPORT_SYMBOL(remap_pfn_range);
1825
1826int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1827                        unsigned long pgoff)
1828{
1829        unsigned int size = vma->vm_end - vma->vm_start;
1830
1831        if (!(vma->vm_flags & VM_USERMAP))
1832                return -EINVAL;
1833
1834        vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1835        vma->vm_end = vma->vm_start + size;
1836
1837        return 0;
1838}
1839EXPORT_SYMBOL(remap_vmalloc_range);
1840
1841unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1842        unsigned long len, unsigned long pgoff, unsigned long flags)
1843{
1844        return -ENOMEM;
1845}
1846
1847void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1848{
1849}
1850
1851void unmap_mapping_range(struct address_space *mapping,
1852                         loff_t const holebegin, loff_t const holelen,
1853                         int even_cows)
1854{
1855}
1856EXPORT_SYMBOL(unmap_mapping_range);
1857
1858/*
1859 * Check that a process has enough memory to allocate a new virtual
1860 * mapping. 0 means there is enough memory for the allocation to
1861 * succeed and -ENOMEM implies there is not.
1862 *
1863 * We currently support three overcommit policies, which are set via the
1864 * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1865 *
1866 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1867 * Additional code 2002 Jul 20 by Robert Love.
1868 *
1869 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1870 *
1871 * Note this is a helper function intended to be used by LSMs which
1872 * wish to use this logic.
1873 */
1874int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1875{
1876        unsigned long free, allowed;
1877
1878        vm_acct_memory(pages);
1879
1880        /*
1881         * Sometimes we want to use more memory than we have
1882         */
1883        if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1884                return 0;
1885
1886        if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1887                free = global_page_state(NR_FREE_PAGES);
1888                free += global_page_state(NR_FILE_PAGES);
1889
1890                /*
1891                 * shmem pages shouldn't be counted as free in this
1892                 * case, they can't be purged, only swapped out, and
1893                 * that won't affect the overall amount of available
1894                 * memory in the system.
1895                 */
1896                free -= global_page_state(NR_SHMEM);
1897
1898                free += nr_swap_pages;
1899
1900                /*
1901                 * Any slabs which are created with the
1902                 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1903                 * which are reclaimable, under pressure.  The dentry
1904                 * cache and most inode caches should fall into this
1905                 */
1906                free += global_page_state(NR_SLAB_RECLAIMABLE);
1907
1908                /*
1909                 * Leave reserved pages. The pages are not for anonymous pages.
1910                 */
1911                if (free <= totalreserve_pages)
1912                        goto error;
1913                else
1914                        free -= totalreserve_pages;
1915
1916                /*
1917                 * Leave the last 3% for root
1918                 */
1919                if (!cap_sys_admin)
1920                        free -= free / 32;
1921
1922                if (free > pages)
1923                        return 0;
1924
1925                goto error;
1926        }
1927
1928        allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1929        /*
1930         * Leave the last 3% for root
1931         */
1932        if (!cap_sys_admin)
1933                allowed -= allowed / 32;
1934        allowed += total_swap_pages;
1935
1936        /* Don't let a single process grow too big:
1937           leave 3% of the size of this process for other processes */
1938        if (mm)
1939                allowed -= mm->total_vm / 32;
1940
1941        if (percpu_counter_read_positive(&vm_committed_as) < allowed)
1942                return 0;
1943
1944error:
1945        vm_unacct_memory(pages);
1946
1947        return -ENOMEM;
1948}
1949
1950int in_gate_area_no_mm(unsigned long addr)
1951{
1952        return 0;
1953}
1954
1955int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1956{
1957        BUG();
1958        return 0;
1959}
1960EXPORT_SYMBOL(filemap_fault);
1961
1962static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
1963                unsigned long addr, void *buf, int len, int write)
1964{
1965        struct vm_area_struct *vma;
1966
1967        down_read(&mm->mmap_sem);
1968
1969        /* the access must start within one of the target process's mappings */
1970        vma = find_vma(mm, addr);
1971        if (vma) {
1972                /* don't overrun this mapping */
1973                if (addr + len >= vma->vm_end)
1974                        len = vma->vm_end - addr;
1975
1976                /* only read or write mappings where it is permitted */
1977                if (write && vma->vm_flags & VM_MAYWRITE)
1978                        copy_to_user_page(vma, NULL, addr,
1979                                         (void *) addr, buf, len);
1980                else if (!write && vma->vm_flags & VM_MAYREAD)
1981                        copy_from_user_page(vma, NULL, addr,
1982                                            buf, (void *) addr, len);
1983                else
1984                        len = 0;
1985        } else {
1986                len = 0;
1987        }
1988
1989        up_read(&mm->mmap_sem);
1990
1991        return len;
1992}
1993
1994/**
1995 * @access_remote_vm - access another process' address space
1996 * @mm:         the mm_struct of the target address space
1997 * @addr:       start address to access
1998 * @buf:        source or destination buffer
1999 * @len:        number of bytes to transfer
2000 * @write:      whether the access is a write
2001 *
2002 * The caller must hold a reference on @mm.
2003 */
2004int access_remote_vm(struct mm_struct *mm, unsigned long addr,
2005                void *buf, int len, int write)
2006{
2007        return __access_remote_vm(NULL, mm, addr, buf, len, write);
2008}
2009
2010/*
2011 * Access another process' address space.
2012 * - source/target buffer must be kernel space
2013 */
2014int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2015{
2016        struct mm_struct *mm;
2017
2018        if (addr + len < addr)
2019                return 0;
2020
2021        mm = get_task_mm(tsk);
2022        if (!mm)
2023                return 0;
2024
2025        len = __access_remote_vm(tsk, mm, addr, buf, len, write);
2026
2027        mmput(mm);
2028        return len;
2029}
2030
2031/**
2032 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
2033 * @inode: The inode to check
2034 * @size: The current filesize of the inode
2035 * @newsize: The proposed filesize of the inode
2036 *
2037 * Check the shared mappings on an inode on behalf of a shrinking truncate to
2038 * make sure that that any outstanding VMAs aren't broken and then shrink the
2039 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
2040 * automatically grant mappings that are too large.
2041 */
2042int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
2043                                size_t newsize)
2044{
2045        struct vm_area_struct *vma;
2046        struct prio_tree_iter iter;
2047        struct vm_region *region;
2048        pgoff_t low, high;
2049        size_t r_size, r_top;
2050
2051        low = newsize >> PAGE_SHIFT;
2052        high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2053
2054        down_write(&nommu_region_sem);
2055
2056        /* search for VMAs that fall within the dead zone */
2057        vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
2058                              low, high) {
2059                /* found one - only interested if it's shared out of the page
2060                 * cache */
2061                if (vma->vm_flags & VM_SHARED) {
2062                        up_write(&nommu_region_sem);
2063                        return -ETXTBSY; /* not quite true, but near enough */
2064                }
2065        }
2066
2067        /* reduce any regions that overlap the dead zone - if in existence,
2068         * these will be pointed to by VMAs that don't overlap the dead zone
2069         *
2070         * we don't check for any regions that start beyond the EOF as there
2071         * shouldn't be any
2072         */
2073        vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
2074                              0, ULONG_MAX) {
2075                if (!(vma->vm_flags & VM_SHARED))
2076                        continue;
2077
2078                region = vma->vm_region;
2079                r_size = region->vm_top - region->vm_start;
2080                r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
2081
2082                if (r_top > newsize) {
2083                        region->vm_top -= r_top - newsize;
2084                        if (region->vm_end > region->vm_top)
2085                                region->vm_end = region->vm_top;
2086                }
2087        }
2088
2089        up_write(&nommu_region_sem);
2090        return 0;
2091}
2092
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.