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/export.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, pte_t **ptes)
 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                mutex_lock(&mapping->i_mmap_mutex);
 700                flush_dcache_mmap_lock(mapping);
 701                vma_prio_tree_insert(vma, &mapping->i_mmap);
 702                flush_dcache_mmap_unlock(mapping);
 703                mutex_unlock(&mapping->i_mmap_mutex);
 704        }
 705
 706        /* add the VMA to the tree */
 707        parent = rb_prev = NULL;
 708        p = &mm->mm_rb.rb_node;
 709        while (*p) {
 710                parent = *p;
 711                pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
 712
 713                /* sort by: start addr, end addr, VMA struct addr in that order
 714                 * (the latter is necessary as we may get identical VMAs) */
 715                if (vma->vm_start < pvma->vm_start)
 716                        p = &(*p)->rb_left;
 717                else if (vma->vm_start > pvma->vm_start) {
 718                        rb_prev = parent;
 719                        p = &(*p)->rb_right;
 720                } else if (vma->vm_end < pvma->vm_end)
 721                        p = &(*p)->rb_left;
 722                else if (vma->vm_end > pvma->vm_end) {
 723                        rb_prev = parent;
 724                        p = &(*p)->rb_right;
 725                } else if (vma < pvma)
 726                        p = &(*p)->rb_left;
 727                else if (vma > pvma) {
 728                        rb_prev = parent;
 729                        p = &(*p)->rb_right;
 730                } else
 731                        BUG();
 732        }
 733
 734        rb_link_node(&vma->vm_rb, parent, p);
 735        rb_insert_color(&vma->vm_rb, &mm->mm_rb);
 736
 737        /* add VMA to the VMA list also */
 738        prev = NULL;
 739        if (rb_prev)
 740                prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
 741
 742        __vma_link_list(mm, vma, prev, parent);
 743}
 744
 745/*
 746 * delete a VMA from its owning mm_struct and address space
 747 */
 748static void delete_vma_from_mm(struct vm_area_struct *vma)
 749{
 750        struct address_space *mapping;
 751        struct mm_struct *mm = vma->vm_mm;
 752
 753        kenter("%p", vma);
 754
 755        protect_vma(vma, 0);
 756
 757        mm->map_count--;
 758        if (mm->mmap_cache == vma)
 759                mm->mmap_cache = NULL;
 760
 761        /* remove the VMA from the mapping */
 762        if (vma->vm_file) {
 763                mapping = vma->vm_file->f_mapping;
 764
 765                mutex_lock(&mapping->i_mmap_mutex);
 766                flush_dcache_mmap_lock(mapping);
 767                vma_prio_tree_remove(vma, &mapping->i_mmap);
 768                flush_dcache_mmap_unlock(mapping);
 769                mutex_unlock(&mapping->i_mmap_mutex);
 770        }
 771
 772        /* remove from the MM's tree and list */
 773        rb_erase(&vma->vm_rb, &mm->mm_rb);
 774
 775        if (vma->vm_prev)
 776                vma->vm_prev->vm_next = vma->vm_next;
 777        else
 778                mm->mmap = vma->vm_next;
 779
 780        if (vma->vm_next)
 781                vma->vm_next->vm_prev = vma->vm_prev;
 782}
 783
 784/*
 785 * destroy a VMA record
 786 */
 787static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
 788{
 789        kenter("%p", vma);
 790        if (vma->vm_ops && vma->vm_ops->close)
 791                vma->vm_ops->close(vma);
 792        if (vma->vm_file) {
 793                fput(vma->vm_file);
 794                if (vma->vm_flags & VM_EXECUTABLE)
 795                        removed_exe_file_vma(mm);
 796        }
 797        put_nommu_region(vma->vm_region);
 798        kmem_cache_free(vm_area_cachep, vma);
 799}
 800
 801/*
 802 * look up the first VMA in which addr resides, NULL if none
 803 * - should be called with mm->mmap_sem at least held readlocked
 804 */
 805struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
 806{
 807        struct vm_area_struct *vma;
 808
 809        /* check the cache first */
 810        vma = mm->mmap_cache;
 811        if (vma && vma->vm_start <= addr && vma->vm_end > addr)
 812                return vma;
 813
 814        /* trawl the list (there may be multiple mappings in which addr
 815         * resides) */
 816        for (vma = mm->mmap; vma; vma = vma->vm_next) {
 817                if (vma->vm_start > addr)
 818                        return NULL;
 819                if (vma->vm_end > addr) {
 820                        mm->mmap_cache = vma;
 821                        return vma;
 822                }
 823        }
 824
 825        return NULL;
 826}
 827EXPORT_SYMBOL(find_vma);
 828
 829/*
 830 * find a VMA
 831 * - we don't extend stack VMAs under NOMMU conditions
 832 */
 833struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
 834{
 835        return find_vma(mm, addr);
 836}
 837
 838/*
 839 * expand a stack to a given address
 840 * - not supported under NOMMU conditions
 841 */
 842int expand_stack(struct vm_area_struct *vma, unsigned long address)
 843{
 844        return -ENOMEM;
 845}
 846
 847/*
 848 * look up the first VMA exactly that exactly matches addr
 849 * - should be called with mm->mmap_sem at least held readlocked
 850 */
 851static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
 852                                             unsigned long addr,
 853                                             unsigned long len)
 854{
 855        struct vm_area_struct *vma;
 856        unsigned long end = addr + len;
 857
 858        /* check the cache first */
 859        vma = mm->mmap_cache;
 860        if (vma && vma->vm_start == addr && vma->vm_end == end)
 861                return vma;
 862
 863        /* trawl the list (there may be multiple mappings in which addr
 864         * resides) */
 865        for (vma = mm->mmap; vma; vma = vma->vm_next) {
 866                if (vma->vm_start < addr)
 867                        continue;
 868                if (vma->vm_start > addr)
 869                        return NULL;
 870                if (vma->vm_end == end) {
 871                        mm->mmap_cache = vma;
 872                        return vma;
 873                }
 874        }
 875
 876        return NULL;
 877}
 878
 879/*
 880 * determine whether a mapping should be permitted and, if so, what sort of
 881 * mapping we're capable of supporting
 882 */
 883static int validate_mmap_request(struct file *file,
 884                                 unsigned long addr,
 885                                 unsigned long len,
 886                                 unsigned long prot,
 887                                 unsigned long flags,
 888                                 unsigned long pgoff,
 889                                 unsigned long *_capabilities)
 890{
 891        unsigned long capabilities, rlen;
 892        int ret;
 893
 894        /* do the simple checks first */
 895        if (flags & MAP_FIXED) {
 896                printk(KERN_DEBUG
 897                       "%d: Can't do fixed-address/overlay mmap of RAM\n",
 898                       current->pid);
 899                return -EINVAL;
 900        }
 901
 902        if ((flags & MAP_TYPE) != MAP_PRIVATE &&
 903            (flags & MAP_TYPE) != MAP_SHARED)
 904                return -EINVAL;
 905
 906        if (!len)
 907                return -EINVAL;
 908
 909        /* Careful about overflows.. */
 910        rlen = PAGE_ALIGN(len);
 911        if (!rlen || rlen > TASK_SIZE)
 912                return -ENOMEM;
 913
 914        /* offset overflow? */
 915        if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
 916                return -EOVERFLOW;
 917
 918        if (file) {
 919                /* validate file mapping requests */
 920                struct address_space *mapping;
 921
 922                /* files must support mmap */
 923                if (!file->f_op || !file->f_op->mmap)
 924                        return -ENODEV;
 925
 926                /* work out if what we've got could possibly be shared
 927                 * - we support chardevs that provide their own "memory"
 928                 * - we support files/blockdevs that are memory backed
 929                 */
 930                mapping = file->f_mapping;
 931                if (!mapping)
 932                        mapping = file->f_path.dentry->d_inode->i_mapping;
 933
 934                capabilities = 0;
 935                if (mapping && mapping->backing_dev_info)
 936                        capabilities = mapping->backing_dev_info->capabilities;
 937
 938                if (!capabilities) {
 939                        /* no explicit capabilities set, so assume some
 940                         * defaults */
 941                        switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
 942                        case S_IFREG:
 943                        case S_IFBLK:
 944                                capabilities = BDI_CAP_MAP_COPY;
 945                                break;
 946
 947                        case S_IFCHR:
 948                                capabilities =
 949                                        BDI_CAP_MAP_DIRECT |
 950                                        BDI_CAP_READ_MAP |
 951                                        BDI_CAP_WRITE_MAP;
 952                                break;
 953
 954                        default:
 955                                return -EINVAL;
 956                        }
 957                }
 958
 959                /* eliminate any capabilities that we can't support on this
 960                 * device */
 961                if (!file->f_op->get_unmapped_area)
 962                        capabilities &= ~BDI_CAP_MAP_DIRECT;
 963                if (!file->f_op->read)
 964                        capabilities &= ~BDI_CAP_MAP_COPY;
 965
 966                /* The file shall have been opened with read permission. */
 967                if (!(file->f_mode & FMODE_READ))
 968                        return -EACCES;
 969
 970                if (flags & MAP_SHARED) {
 971                        /* do checks for writing, appending and locking */
 972                        if ((prot & PROT_WRITE) &&
 973                            !(file->f_mode & FMODE_WRITE))
 974                                return -EACCES;
 975
 976                        if (IS_APPEND(file->f_path.dentry->d_inode) &&
 977                            (file->f_mode & FMODE_WRITE))
 978                                return -EACCES;
 979
 980                        if (locks_verify_locked(file->f_path.dentry->d_inode))
 981                                return -EAGAIN;
 982
 983                        if (!(capabilities & BDI_CAP_MAP_DIRECT))
 984                                return -ENODEV;
 985
 986                        /* we mustn't privatise shared mappings */
 987                        capabilities &= ~BDI_CAP_MAP_COPY;
 988                }
 989                else {
 990                        /* we're going to read the file into private memory we
 991                         * allocate */
 992                        if (!(capabilities & BDI_CAP_MAP_COPY))
 993                                return -ENODEV;
 994
 995                        /* we don't permit a private writable mapping to be
 996                         * shared with the backing device */
 997                        if (prot & PROT_WRITE)
 998                                capabilities &= ~BDI_CAP_MAP_DIRECT;
 999                }
1000
1001                if (capabilities & BDI_CAP_MAP_DIRECT) {
1002                        if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
1003                            ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
1004                            ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
1005                            ) {
1006                                capabilities &= ~BDI_CAP_MAP_DIRECT;
1007                                if (flags & MAP_SHARED) {
1008                                        printk(KERN_WARNING
1009                                               "MAP_SHARED not completely supported on !MMU\n");
1010                                        return -EINVAL;
1011                                }
1012                        }
1013                }
1014
1015                /* handle executable mappings and implied executable
1016                 * mappings */
1017                if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1018                        if (prot & PROT_EXEC)
1019                                return -EPERM;
1020                }
1021                else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
1022                        /* handle implication of PROT_EXEC by PROT_READ */
1023                        if (current->personality & READ_IMPLIES_EXEC) {
1024                                if (capabilities & BDI_CAP_EXEC_MAP)
1025                                        prot |= PROT_EXEC;
1026                        }
1027                }
1028                else if ((prot & PROT_READ) &&
1029                         (prot & PROT_EXEC) &&
1030                         !(capabilities & BDI_CAP_EXEC_MAP)
1031                         ) {
1032                        /* backing file is not executable, try to copy */
1033                        capabilities &= ~BDI_CAP_MAP_DIRECT;
1034                }
1035        }
1036        else {
1037                /* anonymous mappings are always memory backed and can be
1038                 * privately mapped
1039                 */
1040                capabilities = BDI_CAP_MAP_COPY;
1041
1042                /* handle PROT_EXEC implication by PROT_READ */
1043                if ((prot & PROT_READ) &&
1044                    (current->personality & READ_IMPLIES_EXEC))
1045                        prot |= PROT_EXEC;
1046        }
1047
1048        /* allow the security API to have its say */
1049        ret = security_mmap_addr(addr);
1050        if (ret < 0)
1051                return ret;
1052
1053        /* looks okay */
1054        *_capabilities = capabilities;
1055        return 0;
1056}
1057
1058/*
1059 * we've determined that we can make the mapping, now translate what we
1060 * now know into VMA flags
1061 */
1062static unsigned long determine_vm_flags(struct file *file,
1063                                        unsigned long prot,
1064                                        unsigned long flags,
1065                                        unsigned long capabilities)
1066{
1067        unsigned long vm_flags;
1068
1069        vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
1070        /* vm_flags |= mm->def_flags; */
1071
1072        if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
1073                /* attempt to share read-only copies of mapped file chunks */
1074                vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1075                if (file && !(prot & PROT_WRITE))
1076                        vm_flags |= VM_MAYSHARE;
1077        } else {
1078                /* overlay a shareable mapping on the backing device or inode
1079                 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1080                 * romfs/cramfs */
1081                vm_flags |= VM_MAYSHARE | (capabilities & BDI_CAP_VMFLAGS);
1082                if (flags & MAP_SHARED)
1083                        vm_flags |= VM_SHARED;
1084        }
1085
1086        /* refuse to let anyone share private mappings with this process if
1087         * it's being traced - otherwise breakpoints set in it may interfere
1088         * with another untraced process
1089         */
1090        if ((flags & MAP_PRIVATE) && current->ptrace)
1091                vm_flags &= ~VM_MAYSHARE;
1092
1093        return vm_flags;
1094}
1095
1096/*
1097 * set up a shared mapping on a file (the driver or filesystem provides and
1098 * pins the storage)
1099 */
1100static int do_mmap_shared_file(struct vm_area_struct *vma)
1101{
1102        int ret;
1103
1104        ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1105        if (ret == 0) {
1106                vma->vm_region->vm_top = vma->vm_region->vm_end;
1107                return 0;
1108        }
1109        if (ret != -ENOSYS)
1110                return ret;
1111
1112        /* getting -ENOSYS indicates that direct mmap isn't possible (as
1113         * opposed to tried but failed) so we can only give a suitable error as
1114         * it's not possible to make a private copy if MAP_SHARED was given */
1115        return -ENODEV;
1116}
1117
1118/*
1119 * set up a private mapping or an anonymous shared mapping
1120 */
1121static int do_mmap_private(struct vm_area_struct *vma,
1122                           struct vm_region *region,
1123                           unsigned long len,
1124                           unsigned long capabilities)
1125{
1126        struct page *pages;
1127        unsigned long total, point, n;
1128        void *base;
1129        int ret, order;
1130
1131        /* invoke the file's mapping function so that it can keep track of
1132         * shared mappings on devices or memory
1133         * - VM_MAYSHARE will be set if it may attempt to share
1134         */
1135        if (capabilities & BDI_CAP_MAP_DIRECT) {
1136                ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1137                if (ret == 0) {
1138                        /* shouldn't return success if we're not sharing */
1139                        BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1140                        vma->vm_region->vm_top = vma->vm_region->vm_end;
1141                        return 0;
1142                }
1143                if (ret != -ENOSYS)
1144                        return ret;
1145
1146                /* getting an ENOSYS error indicates that direct mmap isn't
1147                 * possible (as opposed to tried but failed) so we'll try to
1148                 * make a private copy of the data and map that instead */
1149        }
1150
1151
1152        /* allocate some memory to hold the mapping
1153         * - note that this may not return a page-aligned address if the object
1154         *   we're allocating is smaller than a page
1155         */
1156        order = get_order(len);
1157        kdebug("alloc order %d for %lx", order, len);
1158
1159        pages = alloc_pages(GFP_KERNEL, order);
1160        if (!pages)
1161                goto enomem;
1162
1163        total = 1 << order;
1164        atomic_long_add(total, &mmap_pages_allocated);
1165
1166        point = len >> PAGE_SHIFT;
1167
1168        /* we allocated a power-of-2 sized page set, so we may want to trim off
1169         * the excess */
1170        if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
1171                while (total > point) {
1172                        order = ilog2(total - point);
1173                        n = 1 << order;
1174                        kdebug("shave %lu/%lu @%lu", n, total - point, total);
1175                        atomic_long_sub(n, &mmap_pages_allocated);
1176                        total -= n;
1177                        set_page_refcounted(pages + total);
1178                        __free_pages(pages + total, order);
1179                }
1180        }
1181
1182        for (point = 1; point < total; point++)
1183                set_page_refcounted(&pages[point]);
1184
1185        base = page_address(pages);
1186        region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1187        region->vm_start = (unsigned long) base;
1188        region->vm_end   = region->vm_start + len;
1189        region->vm_top   = region->vm_start + (total << PAGE_SHIFT);
1190
1191        vma->vm_start = region->vm_start;
1192        vma->vm_end   = region->vm_start + len;
1193
1194        if (vma->vm_file) {
1195                /* read the contents of a file into the copy */
1196                mm_segment_t old_fs;
1197                loff_t fpos;
1198
1199                fpos = vma->vm_pgoff;
1200                fpos <<= PAGE_SHIFT;
1201
1202                old_fs = get_fs();
1203                set_fs(KERNEL_DS);
1204                ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
1205                set_fs(old_fs);
1206
1207                if (ret < 0)
1208                        goto error_free;
1209
1210                /* clear the last little bit */
1211                if (ret < len)
1212                        memset(base + ret, 0, len - ret);
1213
1214        }
1215
1216        return 0;
1217
1218error_free:
1219        free_page_series(region->vm_start, region->vm_top);
1220        region->vm_start = vma->vm_start = 0;
1221        region->vm_end   = vma->vm_end = 0;
1222        region->vm_top   = 0;
1223        return ret;
1224
1225enomem:
1226        printk("Allocation of length %lu from process %d (%s) failed\n",
1227               len, current->pid, current->comm);
1228        show_free_areas(0);
1229        return -ENOMEM;
1230}
1231
1232/*
1233 * handle mapping creation for uClinux
1234 */
1235unsigned long do_mmap_pgoff(struct file *file,
1236                            unsigned long addr,
1237                            unsigned long len,
1238                            unsigned long prot,
1239                            unsigned long flags,
1240                            unsigned long pgoff)
1241{
1242        struct vm_area_struct *vma;
1243        struct vm_region *region;
1244        struct rb_node *rb;
1245        unsigned long capabilities, vm_flags, result;
1246        int ret;
1247
1248        kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1249
1250        /* decide whether we should attempt the mapping, and if so what sort of
1251         * mapping */
1252        ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1253                                    &capabilities);
1254        if (ret < 0) {
1255                kleave(" = %d [val]", ret);
1256                return ret;
1257        }
1258
1259        /* we ignore the address hint */
1260        addr = 0;
1261        len = PAGE_ALIGN(len);
1262
1263        /* we've determined that we can make the mapping, now translate what we
1264         * now know into VMA flags */
1265        vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1266
1267        /* we're going to need to record the mapping */
1268        region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1269        if (!region)
1270                goto error_getting_region;
1271
1272        vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1273        if (!vma)
1274                goto error_getting_vma;
1275
1276        region->vm_usage = 1;
1277        region->vm_flags = vm_flags;
1278        region->vm_pgoff = pgoff;
1279
1280        INIT_LIST_HEAD(&vma->anon_vma_chain);
1281        vma->vm_flags = vm_flags;
1282        vma->vm_pgoff = pgoff;
1283
1284        if (file) {
1285                region->vm_file = file;
1286                get_file(file);
1287                vma->vm_file = file;
1288                get_file(file);
1289                if (vm_flags & VM_EXECUTABLE) {
1290                        added_exe_file_vma(current->mm);
1291                        vma->vm_mm = current->mm;
1292                }
1293        }
1294
1295        down_write(&nommu_region_sem);
1296
1297        /* if we want to share, we need to check for regions created by other
1298         * mmap() calls that overlap with our proposed mapping
1299         * - we can only share with a superset match on most regular files
1300         * - shared mappings on character devices and memory backed files are
1301         *   permitted to overlap inexactly as far as we are concerned for in
1302         *   these cases, sharing is handled in the driver or filesystem rather
1303         *   than here
1304         */
1305        if (vm_flags & VM_MAYSHARE) {
1306                struct vm_region *pregion;
1307                unsigned long pglen, rpglen, pgend, rpgend, start;
1308
1309                pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1310                pgend = pgoff + pglen;
1311
1312                for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1313                        pregion = rb_entry(rb, struct vm_region, vm_rb);
1314
1315                        if (!(pregion->vm_flags & VM_MAYSHARE))
1316                                continue;
1317
1318                        /* search for overlapping mappings on the same file */
1319                        if (pregion->vm_file->f_path.dentry->d_inode !=
1320                            file->f_path.dentry->d_inode)
1321                                continue;
1322
1323                        if (pregion->vm_pgoff >= pgend)
1324                                continue;
1325
1326                        rpglen = pregion->vm_end - pregion->vm_start;
1327                        rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1328                        rpgend = pregion->vm_pgoff + rpglen;
1329                        if (pgoff >= rpgend)
1330                                continue;
1331
1332                        /* handle inexactly overlapping matches between
1333                         * mappings */
1334                        if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1335                            !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1336                                /* new mapping is not a subset of the region */
1337                                if (!(capabilities & BDI_CAP_MAP_DIRECT))
1338                                        goto sharing_violation;
1339                                continue;
1340                        }
1341
1342                        /* we've found a region we can share */
1343                        pregion->vm_usage++;
1344                        vma->vm_region = pregion;
1345                        start = pregion->vm_start;
1346                        start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1347                        vma->vm_start = start;
1348                        vma->vm_end = start + len;
1349
1350                        if (pregion->vm_flags & VM_MAPPED_COPY) {
1351                                kdebug("share copy");
1352                                vma->vm_flags |= VM_MAPPED_COPY;
1353                        } else {
1354                                kdebug("share mmap");
1355                                ret = do_mmap_shared_file(vma);
1356                                if (ret < 0) {
1357                                        vma->vm_region = NULL;
1358                                        vma->vm_start = 0;
1359                                        vma->vm_end = 0;
1360                                        pregion->vm_usage--;
1361                                        pregion = NULL;
1362                                        goto error_just_free;
1363                                }
1364                        }
1365                        fput(region->vm_file);
1366                        kmem_cache_free(vm_region_jar, region);
1367                        region = pregion;
1368                        result = start;
1369                        goto share;
1370                }
1371
1372                /* obtain the address at which to make a shared mapping
1373                 * - this is the hook for quasi-memory character devices to
1374                 *   tell us the location of a shared mapping
1375                 */
1376                if (capabilities & BDI_CAP_MAP_DIRECT) {
1377                        addr = file->f_op->get_unmapped_area(file, addr, len,
1378                                                             pgoff, flags);
1379                        if (IS_ERR_VALUE(addr)) {
1380                                ret = addr;
1381                                if (ret != -ENOSYS)
1382                                        goto error_just_free;
1383
1384                                /* the driver refused to tell us where to site
1385                                 * the mapping so we'll have to attempt to copy
1386                                 * it */
1387                                ret = -ENODEV;
1388                                if (!(capabilities & BDI_CAP_MAP_COPY))
1389                                        goto error_just_free;
1390
1391                                capabilities &= ~BDI_CAP_MAP_DIRECT;
1392                        } else {
1393                                vma->vm_start = region->vm_start = addr;
1394                                vma->vm_end = region->vm_end = addr + len;
1395                        }
1396                }
1397        }
1398
1399        vma->vm_region = region;
1400
1401        /* set up the mapping
1402         * - the region is filled in if BDI_CAP_MAP_DIRECT is still set
1403         */
1404        if (file && vma->vm_flags & VM_SHARED)
1405                ret = do_mmap_shared_file(vma);
1406        else
1407                ret = do_mmap_private(vma, region, len, capabilities);
1408        if (ret < 0)
1409                goto error_just_free;
1410        add_nommu_region(region);
1411
1412        /* clear anonymous mappings that don't ask for uninitialized data */
1413        if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1414                memset((void *)region->vm_start, 0,
1415                       region->vm_end - region->vm_start);
1416
1417        /* okay... we have a mapping; now we have to register it */
1418        result = vma->vm_start;
1419
1420        current->mm->total_vm += len >> PAGE_SHIFT;
1421
1422share:
1423        add_vma_to_mm(current->mm, vma);
1424
1425        /* we flush the region from the icache only when the first executable
1426         * mapping of it is made  */
1427        if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1428                flush_icache_range(region->vm_start, region->vm_end);
1429                region->vm_icache_flushed = true;
1430        }
1431
1432        up_write(&nommu_region_sem);
1433
1434        kleave(" = %lx", result);
1435        return result;
1436
1437error_just_free:
1438        up_write(&nommu_region_sem);
1439error:
1440        if (region->vm_file)
1441                fput(region->vm_file);
1442        kmem_cache_free(vm_region_jar, region);
1443        if (vma->vm_file)
1444                fput(vma->vm_file);
1445        if (vma->vm_flags & VM_EXECUTABLE)
1446                removed_exe_file_vma(vma->vm_mm);
1447        kmem_cache_free(vm_area_cachep, vma);
1448        kleave(" = %d", ret);
1449        return ret;
1450
1451sharing_violation:
1452        up_write(&nommu_region_sem);
1453        printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1454        ret = -EINVAL;
1455        goto error;
1456
1457error_getting_vma:
1458        kmem_cache_free(vm_region_jar, region);
1459        printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1460               " from process %d failed\n",
1461               len, current->pid);
1462        show_free_areas(0);
1463        return -ENOMEM;
1464
1465error_getting_region:
1466        printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1467               " from process %d failed\n",
1468               len, current->pid);
1469        show_free_areas(0);
1470        return -ENOMEM;
1471}
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        retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1490
1491        if (file)
1492                fput(file);
1493out:
1494        return retval;
1495}
1496
1497#ifdef __ARCH_WANT_SYS_OLD_MMAP
1498struct mmap_arg_struct {
1499        unsigned long addr;
1500        unsigned long len;
1501        unsigned long prot;
1502        unsigned long flags;
1503        unsigned long fd;
1504        unsigned long offset;
1505};
1506
1507SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1508{
1509        struct mmap_arg_struct a;
1510
1511        if (copy_from_user(&a, arg, sizeof(a)))
1512                return -EFAULT;
1513        if (a.offset & ~PAGE_MASK)
1514                return -EINVAL;
1515
1516        return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1517                              a.offset >> PAGE_SHIFT);
1518}
1519#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1520
1521/*
1522 * split a vma into two pieces at address 'addr', a new vma is allocated either
1523 * for the first part or the tail.
1524 */
1525int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1526              unsigned long addr, int new_below)
1527{
1528        struct vm_area_struct *new;
1529        struct vm_region *region;
1530        unsigned long npages;
1531
1532        kenter("");
1533
1534        /* we're only permitted to split anonymous regions (these should have
1535         * only a single usage on the region) */
1536        if (vma->vm_file)
1537                return -ENOMEM;
1538
1539        if (mm->map_count >= sysctl_max_map_count)
1540                return -ENOMEM;
1541
1542        region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1543        if (!region)
1544                return -ENOMEM;
1545
1546        new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1547        if (!new) {
1548                kmem_cache_free(vm_region_jar, region);
1549                return -ENOMEM;
1550        }
1551
1552        /* most fields are the same, copy all, and then fixup */
1553        *new = *vma;
1554        *region = *vma->vm_region;
1555        new->vm_region = region;
1556
1557        npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1558
1559        if (new_below) {
1560                region->vm_top = region->vm_end = new->vm_end = addr;
1561        } else {
1562                region->vm_start = new->vm_start = addr;
1563                region->vm_pgoff = new->vm_pgoff += npages;
1564        }
1565
1566        if (new->vm_ops && new->vm_ops->open)
1567                new->vm_ops->open(new);
1568
1569        delete_vma_from_mm(vma);
1570        down_write(&nommu_region_sem);
1571        delete_nommu_region(vma->vm_region);
1572        if (new_below) {
1573                vma->vm_region->vm_start = vma->vm_start = addr;
1574                vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1575        } else {
1576                vma->vm_region->vm_end = vma->vm_end = addr;
1577                vma->vm_region->vm_top = addr;
1578        }
1579        add_nommu_region(vma->vm_region);
1580        add_nommu_region(new->vm_region);
1581        up_write(&nommu_region_sem);
1582        add_vma_to_mm(mm, vma);
1583        add_vma_to_mm(mm, new);
1584        return 0;
1585}
1586
1587/*
1588 * shrink a VMA by removing the specified chunk from either the beginning or
1589 * the end
1590 */
1591static int shrink_vma(struct mm_struct *mm,
1592                      struct vm_area_struct *vma,
1593                      unsigned long from, unsigned long to)
1594{
1595        struct vm_region *region;
1596
1597        kenter("");
1598
1599        /* adjust the VMA's pointers, which may reposition it in the MM's tree
1600         * and list */
1601        delete_vma_from_mm(vma);
1602        if (from > vma->vm_start)
1603                vma->vm_end = from;
1604        else
1605                vma->vm_start = to;
1606        add_vma_to_mm(mm, vma);
1607
1608        /* cut the backing region down to size */
1609        region = vma->vm_region;
1610        BUG_ON(region->vm_usage != 1);
1611
1612        down_write(&nommu_region_sem);
1613        delete_nommu_region(region);
1614        if (from > region->vm_start) {
1615                to = region->vm_top;
1616                region->vm_top = region->vm_end = from;
1617        } else {
1618                region->vm_start = to;
1619        }
1620        add_nommu_region(region);
1621        up_write(&nommu_region_sem);
1622
1623        free_page_series(from, to);
1624        return 0;
1625}
1626
1627/*
1628 * release a mapping
1629 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1630 *   VMA, though it need not cover the whole VMA
1631 */
1632int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1633{
1634        struct vm_area_struct *vma;
1635        unsigned long end;
1636        int ret;
1637
1638        kenter(",%lx,%zx", start, len);
1639
1640        len = PAGE_ALIGN(len);
1641        if (len == 0)
1642                return -EINVAL;
1643
1644        end = start + len;
1645
1646        /* find the first potentially overlapping VMA */
1647        vma = find_vma(mm, start);
1648        if (!vma) {
1649                static int limit = 0;
1650                if (limit < 5) {
1651                        printk(KERN_WARNING
1652                               "munmap of memory not mmapped by process %d"
1653                               " (%s): 0x%lx-0x%lx\n",
1654                               current->pid, current->comm,
1655                               start, start + len - 1);
1656                        limit++;
1657                }
1658                return -EINVAL;
1659        }
1660
1661        /* we're allowed to split an anonymous VMA but not a file-backed one */
1662        if (vma->vm_file) {
1663                do {
1664                        if (start > vma->vm_start) {
1665                                kleave(" = -EINVAL [miss]");
1666                                return -EINVAL;
1667                        }
1668                        if (end == vma->vm_end)
1669                                goto erase_whole_vma;
1670                        vma = vma->vm_next;
1671                } while (vma);
1672                kleave(" = -EINVAL [split file]");
1673                return -EINVAL;
1674        } else {
1675                /* the chunk must be a subset of the VMA found */
1676                if (start == vma->vm_start && end == vma->vm_end)
1677                        goto erase_whole_vma;
1678                if (start < vma->vm_start || end > vma->vm_end) {
1679                        kleave(" = -EINVAL [superset]");
1680                        return -EINVAL;
1681                }
1682                if (start & ~PAGE_MASK) {
1683                        kleave(" = -EINVAL [unaligned start]");
1684                        return -EINVAL;
1685                }
1686                if (end != vma->vm_end && end & ~PAGE_MASK) {
1687                        kleave(" = -EINVAL [unaligned split]");
1688                        return -EINVAL;
1689                }
1690                if (start != vma->vm_start && end != vma->vm_end) {
1691                        ret = split_vma(mm, vma, start, 1);
1692                        if (ret < 0) {
1693                                kleave(" = %d [split]", ret);
1694                                return ret;
1695                        }
1696                }
1697                return shrink_vma(mm, vma, start, end);
1698        }
1699
1700erase_whole_vma:
1701        delete_vma_from_mm(vma);
1702        delete_vma(mm, vma);
1703        kleave(" = 0");
1704        return 0;
1705}
1706EXPORT_SYMBOL(do_munmap);
1707
1708int vm_munmap(unsigned long addr, size_t len)
1709{
1710        struct mm_struct *mm = current->mm;
1711        int ret;
1712
1713        down_write(&mm->mmap_sem);
1714        ret = do_munmap(mm, addr, len);
1715        up_write(&mm->mmap_sem);
1716        return ret;
1717}
1718EXPORT_SYMBOL(vm_munmap);
1719
1720SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1721{
1722        return vm_munmap(addr, len);
1723}
1724
1725/*
1726 * release all the mappings made in a process's VM space
1727 */
1728void exit_mmap(struct mm_struct *mm)
1729{
1730        struct vm_area_struct *vma;
1731
1732        if (!mm)
1733                return;
1734
1735        kenter("");
1736
1737        mm->total_vm = 0;
1738
1739        while ((vma = mm->mmap)) {
1740                mm->mmap = vma->vm_next;
1741                delete_vma_from_mm(vma);
1742                delete_vma(mm, vma);
1743                cond_resched();
1744        }
1745
1746        kleave("");
1747}
1748
1749unsigned long vm_brk(unsigned long addr, unsigned long len)
1750{
1751        return -ENOMEM;
1752}
1753
1754/*
1755 * expand (or shrink) an existing mapping, potentially moving it at the same
1756 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1757 *
1758 * under NOMMU conditions, we only permit changing a mapping's size, and only
1759 * as long as it stays within the region allocated by do_mmap_private() and the
1760 * block is not shareable
1761 *
1762 * MREMAP_FIXED is not supported under NOMMU conditions
1763 */
1764unsigned long do_mremap(unsigned long addr,
1765                        unsigned long old_len, unsigned long new_len,
1766                        unsigned long flags, unsigned long new_addr)
1767{
1768        struct vm_area_struct *vma;
1769
1770        /* insanity checks first */
1771        old_len = PAGE_ALIGN(old_len);
1772        new_len = PAGE_ALIGN(new_len);
1773        if (old_len == 0 || new_len == 0)
1774                return (unsigned long) -EINVAL;
1775
1776        if (addr & ~PAGE_MASK)
1777                return -EINVAL;
1778
1779        if (flags & MREMAP_FIXED && new_addr != addr)
1780                return (unsigned long) -EINVAL;
1781
1782        vma = find_vma_exact(current->mm, addr, old_len);
1783        if (!vma)
1784                return (unsigned long) -EINVAL;
1785
1786        if (vma->vm_end != vma->vm_start + old_len)
1787                return (unsigned long) -EFAULT;
1788
1789        if (vma->vm_flags & VM_MAYSHARE)
1790                return (unsigned long) -EPERM;
1791
1792        if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1793                return (unsigned long) -ENOMEM;
1794
1795        /* all checks complete - do it */
1796        vma->vm_end = vma->vm_start + new_len;
1797        return vma->vm_start;
1798}
1799EXPORT_SYMBOL(do_mremap);
1800
1801SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1802                unsigned long, new_len, unsigned long, flags,
1803                unsigned long, new_addr)
1804{
1805        unsigned long ret;
1806
1807        down_write(&current->mm->mmap_sem);
1808        ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1809        up_write(&current->mm->mmap_sem);
1810        return ret;
1811}
1812
1813struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1814                        unsigned int foll_flags)
1815{
1816        return NULL;
1817}
1818
1819int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1820                unsigned long pfn, unsigned long size, pgprot_t prot)
1821{
1822        if (addr != (pfn << PAGE_SHIFT))
1823                return -EINVAL;
1824
1825        vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
1826        return 0;
1827}
1828EXPORT_SYMBOL(remap_pfn_range);
1829
1830int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1831                        unsigned long pgoff)
1832{
1833        unsigned int size = vma->vm_end - vma->vm_start;
1834
1835        if (!(vma->vm_flags & VM_USERMAP))
1836                return -EINVAL;
1837
1838        vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1839        vma->vm_end = vma->vm_start + size;
1840
1841        return 0;
1842}
1843EXPORT_SYMBOL(remap_vmalloc_range);
1844
1845unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1846        unsigned long len, unsigned long pgoff, unsigned long flags)
1847{
1848        return -ENOMEM;
1849}
1850
1851void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1852{
1853}
1854
1855void unmap_mapping_range(struct address_space *mapping,
1856                         loff_t const holebegin, loff_t const holelen,
1857                         int even_cows)
1858{
1859}
1860EXPORT_SYMBOL(unmap_mapping_range);
1861
1862/*
1863 * Check that a process has enough memory to allocate a new virtual
1864 * mapping. 0 means there is enough memory for the allocation to
1865 * succeed and -ENOMEM implies there is not.
1866 *
1867 * We currently support three overcommit policies, which are set via the
1868 * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1869 *
1870 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1871 * Additional code 2002 Jul 20 by Robert Love.
1872 *
1873 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1874 *
1875 * Note this is a helper function intended to be used by LSMs which
1876 * wish to use this logic.
1877 */
1878int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1879{
1880        unsigned long free, allowed;
1881
1882        vm_acct_memory(pages);
1883
1884        /*
1885         * Sometimes we want to use more memory than we have
1886         */
1887        if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1888                return 0;
1889
1890        if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1891                free = global_page_state(NR_FREE_PAGES);
1892                free += global_page_state(NR_FILE_PAGES);
1893
1894                /*
1895                 * shmem pages shouldn't be counted as free in this
1896                 * case, they can't be purged, only swapped out, and
1897                 * that won't affect the overall amount of available
1898                 * memory in the system.
1899                 */
1900                free -= global_page_state(NR_SHMEM);
1901
1902                free += nr_swap_pages;
1903
1904                /*
1905                 * Any slabs which are created with the
1906                 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1907                 * which are reclaimable, under pressure.  The dentry
1908                 * cache and most inode caches should fall into this
1909                 */
1910                free += global_page_state(NR_SLAB_RECLAIMABLE);
1911
1912                /*
1913                 * Leave reserved pages. The pages are not for anonymous pages.
1914                 */
1915                if (free <= totalreserve_pages)
1916                        goto error;
1917                else
1918                        free -= totalreserve_pages;
1919
1920                /*
1921                 * Leave the last 3% for root
1922                 */
1923                if (!cap_sys_admin)
1924                        free -= free / 32;
1925
1926                if (free > pages)
1927                        return 0;
1928
1929                goto error;
1930        }
1931
1932        allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1933        /*
1934         * Leave the last 3% for root
1935         */
1936        if (!cap_sys_admin)
1937                allowed -= allowed / 32;
1938        allowed += total_swap_pages;
1939
1940        /* Don't let a single process grow too big:
1941           leave 3% of the size of this process for other processes */
1942        if (mm)
1943                allowed -= mm->total_vm / 32;
1944
1945        if (percpu_counter_read_positive(&vm_committed_as) < allowed)
1946                return 0;
1947
1948error:
1949        vm_unacct_memory(pages);
1950
1951        return -ENOMEM;
1952}
1953
1954int in_gate_area_no_mm(unsigned long addr)
1955{
1956        return 0;
1957}
1958
1959int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1960{
1961        BUG();
1962        return 0;
1963}
1964EXPORT_SYMBOL(filemap_fault);
1965
1966static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
1967                unsigned long addr, void *buf, int len, int write)
1968{
1969        struct vm_area_struct *vma;
1970
1971        down_read(&mm->mmap_sem);
1972
1973        /* the access must start within one of the target process's mappings */
1974        vma = find_vma(mm, addr);
1975        if (vma) {
1976                /* don't overrun this mapping */
1977                if (addr + len >= vma->vm_end)
1978                        len = vma->vm_end - addr;
1979
1980                /* only read or write mappings where it is permitted */
1981                if (write && vma->vm_flags & VM_MAYWRITE)
1982                        copy_to_user_page(vma, NULL, addr,
1983                                         (void *) addr, buf, len);
1984                else if (!write && vma->vm_flags & VM_MAYREAD)
1985                        copy_from_user_page(vma, NULL, addr,
1986                                            buf, (void *) addr, len);
1987                else
1988                        len = 0;
1989        } else {
1990                len = 0;
1991        }
1992
1993        up_read(&mm->mmap_sem);
1994
1995        return len;
1996}
1997
1998/**
1999 * @access_remote_vm - access another process' address space
2000 * @mm:         the mm_struct of the target address space
2001 * @addr:       start address to access
2002 * @buf:        source or destination buffer
2003 * @len:        number of bytes to transfer
2004 * @write:      whether the access is a write
2005 *
2006 * The caller must hold a reference on @mm.
2007 */
2008int access_remote_vm(struct mm_struct *mm, unsigned long addr,
2009                void *buf, int len, int write)
2010{
2011        return __access_remote_vm(NULL, mm, addr, buf, len, write);
2012}
2013
2014/*
2015 * Access another process' address space.
2016 * - source/target buffer must be kernel space
2017 */
2018int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2019{
2020        struct mm_struct *mm;
2021
2022        if (addr + len < addr)
2023                return 0;
2024
2025        mm = get_task_mm(tsk);
2026        if (!mm)
2027                return 0;
2028
2029        len = __access_remote_vm(tsk, mm, addr, buf, len, write);
2030
2031        mmput(mm);
2032        return len;
2033}
2034
2035/**
2036 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
2037 * @inode: The inode to check
2038 * @size: The current filesize of the inode
2039 * @newsize: The proposed filesize of the inode
2040 *
2041 * Check the shared mappings on an inode on behalf of a shrinking truncate to
2042 * make sure that that any outstanding VMAs aren't broken and then shrink the
2043 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
2044 * automatically grant mappings that are too large.
2045 */
2046int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
2047                                size_t newsize)
2048{
2049        struct vm_area_struct *vma;
2050        struct prio_tree_iter iter;
2051        struct vm_region *region;
2052        pgoff_t low, high;
2053        size_t r_size, r_top;
2054
2055        low = newsize >> PAGE_SHIFT;
2056        high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2057
2058        down_write(&nommu_region_sem);
2059        mutex_lock(&inode->i_mapping->i_mmap_mutex);
2060
2061        /* search for VMAs that fall within the dead zone */
2062        vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
2063                              low, high) {
2064                /* found one - only interested if it's shared out of the page
2065                 * cache */
2066                if (vma->vm_flags & VM_SHARED) {
2067                        mutex_unlock(&inode->i_mapping->i_mmap_mutex);
2068                        up_write(&nommu_region_sem);
2069                        return -ETXTBSY; /* not quite true, but near enough */
2070                }
2071        }
2072
2073        /* reduce any regions that overlap the dead zone - if in existence,
2074         * these will be pointed to by VMAs that don't overlap the dead zone
2075         *
2076         * we don't check for any regions that start beyond the EOF as there
2077         * shouldn't be any
2078         */
2079        vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
2080                              0, ULONG_MAX) {
2081                if (!(vma->vm_flags & VM_SHARED))
2082                        continue;
2083
2084                region = vma->vm_region;
2085                r_size = region->vm_top - region->vm_start;
2086                r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
2087
2088                if (r_top > newsize) {
2089                        region->vm_top -= r_top - newsize;
2090                        if (region->vm_end > region->vm_top)
2091                                region->vm_end = region->vm_top;
2092                }
2093        }
2094
2095        mutex_unlock(&inode->i_mapping->i_mmap_mutex);
2096        up_write(&nommu_region_sem);
2097        return 0;
2098}
2099
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.