mm/nommu.c

   1 /*
   2  *  linux/mm/nommu.c
   3  *
   4  *  Replacement code for mm functions to support CPU's that don't
   5  *  have any form of memory management unit (thus no virtual memory).
   6  *
   7  *  See Documentation/nommu-mmap.txt
   8  *
   9  *  Copyright (c) 2004-2008 David Howells <dhowells@redhat.com>
  10  *  Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
  11  *  Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
  12  *  Copyright (c) 2002      Greg Ungerer <gerg@snapgear.com>
  13  *  Copyright (c) 2007-2009 Paul Mundt <lethal@linux-sh.org>
  14  */
  15
  16 #include <linux/module.h>
  17 #include <linux/mm.h>
  18 #include <linux/mman.h>
  19 #include <linux/swap.h>
  20 #include <linux/file.h>
  21 #include <linux/highmem.h>
  22 #include <linux/pagemap.h>
  23 #include <linux/slab.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/tracehook.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/backing-dev.h>
  28 #include <linux/mount.h>
  29 #include <linux/personality.h>
  30 #include <linux/security.h>
  31 #include <linux/syscalls.h>
  32
  33 #include <asm/uaccess.h>
  34 #include <asm/tlb.h>
  35 #include <asm/tlbflush.h>
  36 #include "internal.h"
  37
  38 static inline __attribute__((format(printf, 1, 2)))
  39 void no_printk(const char *fmt, ...)
  40 {
  41 }
  42
  43 #if 0
  44 #define kenter(FMT, ...) \
  45         printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
  46 #define kleave(FMT, ...) \
  47         printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
  48 #define kdebug(FMT, ...) \
  49         printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__)
  50 #else
  51 #define kenter(FMT, ...) \
  52         no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
  53 #define kleave(FMT, ...) \
  54         no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
  55 #define kdebug(FMT, ...) \
  56         no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__)
  57 #endif
  58
  59 #include "internal.h"
  60
  61 void *high_memory;
  62 struct page *mem_map;
  63 unsigned long max_mapnr;
  64 unsigned long num_physpages;
  65 atomic_long_t vm_committed_space = ATOMIC_LONG_INIT(0);
  66 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
  67 int sysctl_overcommit_ratio = 50; /* default is 50% */
  68 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
  69 int sysctl_nr_trim_pages = 1; /* page trimming behaviour */
  70 int heap_stack_gap = 0;
  71
  72 atomic_t mmap_pages_allocated;
  73
  74 EXPORT_SYMBOL(mem_map);
  75 EXPORT_SYMBOL(num_physpages);
  76
  77 /* list of mapped, potentially shareable regions */
  78 static struct kmem_cache *vm_region_jar;
  79 struct rb_root nommu_region_tree = RB_ROOT;
  80 DECLARE_RWSEM(nommu_region_sem);
  81
  82 struct vm_operations_struct generic_file_vm_ops = {
  83 };
  84
  85 /*
  86  * Handle all mappings that got truncated by a "truncate()"
  87  * system call.
  88  *
  89  * NOTE! We have to be ready to update the memory sharing
  90  * between the file and the memory map for a potential last
  91  * incomplete page.  Ugly, but necessary.
  92  */
  93 int vmtruncate(struct inode *inode, loff_t offset)
  94 {
  95         struct address_space *mapping = inode->i_mapping;
  96         unsigned long limit;
  97
  98         if (inode->i_size < offset)
  99                 goto do_expand;
 100         i_size_write(inode, offset);
 101
 102         truncate_inode_pages(mapping, offset);
 103         goto out_truncate;
 104
 105 do_expand:
 106         limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
 107         if (limit != RLIM_INFINITY && offset > limit)
 108                 goto out_sig;
 109         if (offset > inode->i_sb->s_maxbytes)
 110                 goto out;
 111         i_size_write(inode, offset);
 112
 113 out_truncate:
 114         if (inode->i_op->truncate)
 115                 inode->i_op->truncate(inode);
 116         return 0;
 117 out_sig:
 118         send_sig(SIGXFSZ, current, 0);
 119 out:
 120         return -EFBIG;
 121 }
 122
 123 EXPORT_SYMBOL(vmtruncate);
 124
 125 /*
 126  * Return the total memory allocated for this pointer, not
 127  * just what the caller asked for.
 128  *
 129  * Doesn't have to be accurate, i.e. may have races.
 130  */
 131 unsigned int kobjsize(const void *objp)
 132 {
 133         struct page *page;
 134
 135         /*
 136          * If the object we have should not have ksize performed on it,
 137          * return size of 0
 138          */
 139         if (!objp || !virt_addr_valid(objp))
 140                 return 0;
 141
 142         page = virt_to_head_page(objp);
 143
 144         /*
 145          * If the allocator sets PageSlab, we know the pointer came from
 146          * kmalloc().
 147          */
 148         if (PageSlab(page))
 149                 return ksize(objp);
 150
 151         /*
 152          * If it's not a compound page, see if we have a matching VMA
 153          * region. This test is intentionally done in reverse order,
 154          * so if there's no VMA, we still fall through and hand back
 155          * PAGE_SIZE for 0-order pages.
 156          */
 157         if (!PageCompound(page)) {
 158                 struct vm_area_struct *vma;
 159
 160                 vma = find_vma(current->mm, (unsigned long)objp);
 161                 if (vma)
 162                         return vma->vm_end - vma->vm_start;
 163         }
 164
 165         /*
 166          * The ksize() function is only guaranteed to work for pointers
 167          * returned by kmalloc(). So handle arbitrary pointers here.
 168          */
 169         return PAGE_SIZE << compound_order(page);
 170 }
 171
 172 int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
 173                      unsigned long start, int len, int flags,
 174                 struct page **pages, struct vm_area_struct **vmas)
 175 {
 176         struct vm_area_struct *vma;
 177         unsigned long vm_flags;
 178         int i;
 179         int write = !!(flags & GUP_FLAGS_WRITE);
 180         int force = !!(flags & GUP_FLAGS_FORCE);
 181         int ignore = !!(flags & GUP_FLAGS_IGNORE_VMA_PERMISSIONS);
 182
 183         /* calculate required read or write permissions.
 184          * - if 'force' is set, we only require the "MAY" flags.
 185          */
 186         vm_flags  = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
 187         vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
 188
 189         for (i = 0; i < len; i++) {
 190                 vma = find_vma(mm, start);
 191                 if (!vma)
 192                         goto finish_or_fault;
 193
 194                 /* protect what we can, including chardevs */
 195                 if (vma->vm_flags & (VM_IO | VM_PFNMAP) ||
 196                     (!ignore && !(vm_flags & vma->vm_flags)))
 197                         goto finish_or_fault;
 198
 199                 if (pages) {
 200                         pages[i] = virt_to_page(start);
 201                         if (pages[i])
 202                                 page_cache_get(pages[i]);
 203                 }
 204                 if (vmas)
 205                         vmas[i] = vma;
 206                 start += PAGE_SIZE;
 207         }
 208
 209         return i;
 210
 211 finish_or_fault:
 212         return i ? : -EFAULT;
 213 }
 214
 215
 216 /*
 217  * get a list of pages in an address range belonging to the specified process
 218  * and indicate the VMA that covers each page
 219  * - this is potentially dodgy as we may end incrementing the page count of a
 220  *   slab page or a secondary page from a compound page
 221  * - don't permit access to VMAs that don't support it, such as I/O mappings
 222  */
 223 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
 224         unsigned long start, int len, int write, int force,
 225         struct page **pages, struct vm_area_struct **vmas)
 226 {
 227         int flags = 0;
 228
 229         if (write)
 230                 flags |= GUP_FLAGS_WRITE;
 231         if (force)
 232                 flags |= GUP_FLAGS_FORCE;
 233
 234         return __get_user_pages(tsk, mm,
 235                                 start, len, flags,
 236                                 pages, vmas);
 237 }
 238 EXPORT_SYMBOL(get_user_pages);
 239
 240 DEFINE_RWLOCK(vmlist_lock);
 241 struct vm_struct *vmlist;
 242
 243 void vfree(const void *addr)
 244 {
 245         kfree(addr);
 246 }
 247 EXPORT_SYMBOL(vfree);
 248
 249 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
 250 {
 251         /*
 252          *  You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
 253          * returns only a logical address.
 254          */
 255         return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
 256 }
 257 EXPORT_SYMBOL(__vmalloc);
 258
 259 void *vmalloc_user(unsigned long size)
 260 {
 261         void *ret;
 262
 263         ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
 264                         PAGE_KERNEL);
 265         if (ret) {
 266                 struct vm_area_struct *vma;
 267
 268                 down_write(&current->mm->mmap_sem);
 269                 vma = find_vma(current->mm, (unsigned long)ret);
 270                 if (vma)
 271                         vma->vm_flags |= VM_USERMAP;
 272                 up_write(&current->mm->mmap_sem);
 273         }
 274
 275         return ret;
 276 }
 277 EXPORT_SYMBOL(vmalloc_user);
 278
 279 struct page *vmalloc_to_page(const void *addr)
 280 {
 281         return virt_to_page(addr);
 282 }
 283 EXPORT_SYMBOL(vmalloc_to_page);
 284
 285 unsigned long vmalloc_to_pfn(const void *addr)
 286 {
 287         return page_to_pfn(virt_to_page(addr));
 288 }
 289 EXPORT_SYMBOL(vmalloc_to_pfn);
 290
 291 long vread(char *buf, char *addr, unsigned long count)
 292 {
 293         memcpy(buf, addr, count);
 294         return count;
 295 }
 296
 297 long vwrite(char *buf, char *addr, unsigned long count)
 298 {
 299         /* Don't allow overflow */
 300         if ((unsigned long) addr + count < count)
 301                 count = -(unsigned long) addr;
 302
 303         memcpy(addr, buf, count);
 304         return(count);
 305 }
 306
 307 /*
 308  *      vmalloc  -  allocate virtually continguos memory
 309  *
 310  *      @size:          allocation size
 311  *
 312  *      Allocate enough pages to cover @size from the page level
 313  *      allocator and map them into continguos kernel virtual space.
 314  *
 315  *      For tight control over page level allocator and protection flags
 316  *      use __vmalloc() instead.
 317  */
 318 void *vmalloc(unsigned long size)
 319 {
 320        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
 321 }
 322 EXPORT_SYMBOL(vmalloc);
 323
 324 void *vmalloc_node(unsigned long size, int node)
 325 {
 326         return vmalloc(size);
 327 }
 328 EXPORT_SYMBOL(vmalloc_node);
 329
 330 #ifndef PAGE_KERNEL_EXEC
 331 # define PAGE_KERNEL_EXEC PAGE_KERNEL
 332 #endif
 333
 334 /**
 335  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
 336  *      @size:          allocation size
 337  *
 338  *      Kernel-internal function to allocate enough pages to cover @size
 339  *      the page level allocator and map them into contiguous and
 340  *      executable kernel virtual space.
 341  *
 342  *      For tight control over page level allocator and protection flags
 343  *      use __vmalloc() instead.
 344  */
 345
 346 void *vmalloc_exec(unsigned long size)
 347 {
 348         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
 349 }
 350
 351 /**
 352  * vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
 353  *      @size:          allocation size
 354  *
 355  *      Allocate enough 32bit PA addressable pages to cover @size from the
 356  *      page level allocator and map them into continguos kernel virtual space.
 357  */
 358 void *vmalloc_32(unsigned long size)
 359 {
 360         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
 361 }
 362 EXPORT_SYMBOL(vmalloc_32);
 363
 364 /**
 365  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
 366  *      @size:          allocation size
 367  *
 368  * The resulting memory area is 32bit addressable and zeroed so it can be
 369  * mapped to userspace without leaking data.
 370  *
 371  * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
 372  * remap_vmalloc_range() are permissible.
 373  */
 374 void *vmalloc_32_user(unsigned long size)
 375 {
 376         /*
 377          * We'll have to sort out the ZONE_DMA bits for 64-bit,
 378          * but for now this can simply use vmalloc_user() directly.
 379          */
 380         return vmalloc_user(size);
 381 }
 382 EXPORT_SYMBOL(vmalloc_32_user);
 383
 384 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
 385 {
 386         BUG();
 387         return NULL;
 388 }
 389 EXPORT_SYMBOL(vmap);
 390
 391 void vunmap(const void *addr)
 392 {
 393         BUG();
 394 }
 395 EXPORT_SYMBOL(vunmap);
 396
 397 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
 398 {
 399         BUG();
 400         return NULL;
 401 }
 402 EXPORT_SYMBOL(vm_map_ram);
 403
 404 void vm_unmap_ram(const void *mem, unsigned int count)
 405 {
 406         BUG();
 407 }
 408 EXPORT_SYMBOL(vm_unmap_ram);
 409
 410 void vm_unmap_aliases(void)
 411 {
 412 }
 413 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
 414
 415 /*
 416  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
 417  * have one.
 418  */
 419 void  __attribute__((weak)) vmalloc_sync_all(void)
 420 {
 421 }
 422
 423 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
 424                    struct page *page)
 425 {
 426         return -EINVAL;
 427 }
 428 EXPORT_SYMBOL(vm_insert_page);
 429
 430 /*
 431  *  sys_brk() for the most part doesn't need the global kernel
 432  *  lock, except when an application is doing something nasty
 433  *  like trying to un-brk an area that has already been mapped
 434  *  to a regular file.  in this case, the unmapping will need
 435  *  to invoke file system routines that need the global lock.
 436  */
 437 SYSCALL_DEFINE1(brk, unsigned long, brk)
 438 {
 439         struct mm_struct *mm = current->mm;
 440
 441         if (brk < mm->start_brk || brk > mm->context.end_brk)
 442                 return mm->brk;
 443
 444         if (mm->brk == brk)
 445                 return mm->brk;
 446
 447         /*
 448          * Always allow shrinking brk
 449          */
 450         if (brk <= mm->brk) {
 451                 mm->brk = brk;
 452                 return brk;
 453         }
 454
 455         /*
 456          * Ok, looks good - let it rip.
 457          */
 458         return mm->brk = brk;
 459 }
 460
 461 /*
 462  * initialise the VMA and region record slabs
 463  */
 464 void __init mmap_init(void)
 465 {
 466         vm_region_jar = kmem_cache_create("vm_region_jar",
 467                                           sizeof(struct vm_region), 0,
 468                                           SLAB_PANIC, NULL);
 469         vm_area_cachep = kmem_cache_create("vm_area_struct",
 470                                            sizeof(struct vm_area_struct), 0,
 471                                            SLAB_PANIC, NULL);
 472 }
 473
 474 /*
 475  * validate the region tree
 476  * - the caller must hold the region lock
 477  */
 478 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
 479 static noinline void validate_nommu_regions(void)
 480 {
 481         struct vm_region *region, *last;
 482         struct rb_node *p, *lastp;
 483
 484         lastp = rb_first(&nommu_region_tree);
 485         if (!lastp)
 486                 return;
 487
 488         last = rb_entry(lastp, struct vm_region, vm_rb);
 489         if (unlikely(last->vm_end <= last->vm_start))
 490                 BUG();
 491         if (unlikely(last->vm_top < last->vm_end))
 492                 BUG();
 493
 494         while ((p = rb_next(lastp))) {
 495                 region = rb_entry(p, struct vm_region, vm_rb);
 496                 last = rb_entry(lastp, struct vm_region, vm_rb);
 497
 498                 if (unlikely(region->vm_end <= region->vm_start))
 499                         BUG();
 500                 if (unlikely(region->vm_top < region->vm_end))
 501                         BUG();
 502                 if (unlikely(region->vm_start < last->vm_top))
 503                         BUG();
 504
 505                 lastp = p;
 506         }
 507 }
 508 #else
 509 #define validate_nommu_regions() do {} while(0)
 510 #endif
 511
 512 /*
 513  * add a region into the global tree
 514  */
 515 static void add_nommu_region(struct vm_region *region)
 516 {
 517         struct vm_region *pregion;
 518         struct rb_node **p, *parent;
 519
 520         validate_nommu_regions();
 521
 522         BUG_ON(region->vm_start & ~PAGE_MASK);
 523
 524         parent = NULL;
 525         p = &nommu_region_tree.rb_node;
 526         while (*p) {
 527                 parent = *p;
 528                 pregion = rb_entry(parent, struct vm_region, vm_rb);
 529                 if (region->vm_start < pregion->vm_start)
 530                         p = &(*p)->rb_left;
 531                 else if (region->vm_start > pregion->vm_start)
 532                         p = &(*p)->rb_right;
 533                 else if (pregion == region)
 534                         return;
 535                 else
 536                         BUG();
 537         }
 538
 539         rb_link_node(&region->vm_rb, parent, p);
 540         rb_insert_color(&region->vm_rb, &nommu_region_tree);
 541
 542         validate_nommu_regions();
 543 }
 544
 545 /*
 546  * delete a region from the global tree
 547  */
 548 static void delete_nommu_region(struct vm_region *region)
 549 {
 550         BUG_ON(!nommu_region_tree.rb_node);
 551
 552         validate_nommu_regions();
 553         rb_erase(&region->vm_rb, &nommu_region_tree);
 554         validate_nommu_regions();
 555 }
 556
 557 /*
 558  * free a contiguous series of pages
 559  */
 560 static void free_page_series(unsigned long from, unsigned long to)
 561 {
 562         for (; from < to; from += PAGE_SIZE) {
 563                 struct page *page = virt_to_page(from);
 564
 565                 kdebug("- free %lx", from);
 566                 atomic_dec(&mmap_pages_allocated);
 567                 if (page_count(page) != 1)
 568                         kdebug("free page %p [%d]", page, page_count(page));
 569                 put_page(page);
 570         }
 571 }
 572
 573 /*
 574  * release a reference to a region
 575  * - the caller must hold the region semaphore, which this releases
 576  * - the region may not have been added to the tree yet, in which case vm_top
 577  *   will equal vm_start
 578  */
 579 static void __put_nommu_region(struct vm_region *region)
 580         __releases(nommu_region_sem)
 581 {
 582         kenter("%p{%d}", region, atomic_read(&region->vm_usage));
 583
 584         BUG_ON(!nommu_region_tree.rb_node);
 585
 586         if (atomic_dec_and_test(&region->vm_usage)) {
 587                 if (region->vm_top > region->vm_start)
 588                         delete_nommu_region(region);
 589                 up_write(&nommu_region_sem);
 590
 591                 if (region->vm_file)
 592                         fput(region->vm_file);
 593
 594                 /* IO memory and memory shared directly out of the pagecache
 595                  * from ramfs/tmpfs mustn't be released here */
 596                 if (region->vm_flags & VM_MAPPED_COPY) {
 597                         kdebug("free series");
 598                         free_page_series(region->vm_start, region->vm_top);
 599                 }
 600                 kmem_cache_free(vm_region_jar, region);
 601         } else {
 602                 up_write(&nommu_region_sem);
 603         }
 604 }
 605
 606 /*
 607  * release a reference to a region
 608  */
 609 static void put_nommu_region(struct vm_region *region)
 610 {
 611         down_write(&nommu_region_sem);
 612         __put_nommu_region(region);
 613 }
 614
 615 /*
 616  * add a VMA into a process's mm_struct in the appropriate place in the list
 617  * and tree and add to the address space's page tree also if not an anonymous
 618  * page
 619  * - should be called with mm->mmap_sem held writelocked
 620  */
 621 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
 622 {
 623         struct vm_area_struct *pvma, **pp;
 624         struct address_space *mapping;
 625         struct rb_node **p, *parent;
 626
 627         kenter(",%p", vma);
 628
 629         BUG_ON(!vma->vm_region);
 630
 631         mm->map_count++;
 632         vma->vm_mm = mm;
 633
 634         /* add the VMA to the mapping */
 635         if (vma->vm_file) {
 636                 mapping = vma->vm_file->f_mapping;
 637
 638                 flush_dcache_mmap_lock(mapping);
 639                 vma_prio_tree_insert(vma, &mapping->i_mmap);
 640                 flush_dcache_mmap_unlock(mapping);
 641         }
 642
 643         /* add the VMA to the tree */
 644         parent = NULL;
 645         p = &mm->mm_rb.rb_node;
 646         while (*p) {
 647                 parent = *p;
 648                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
 649
 650                 /* sort by: start addr, end addr, VMA struct addr in that order
 651                  * (the latter is necessary as we may get identical VMAs) */
 652                 if (vma->vm_start < pvma->vm_start)
 653                         p = &(*p)->rb_left;
 654                 else if (vma->vm_start > pvma->vm_start)
 655                         p = &(*p)->rb_right;
 656                 else if (vma->vm_end < pvma->vm_end)
 657                         p = &(*p)->rb_left;
 658                 else if (vma->vm_end > pvma->vm_end)
 659                         p = &(*p)->rb_right;
 660                 else if (vma < pvma)
 661                         p = &(*p)->rb_left;
 662                 else if (vma > pvma)
 663                         p = &(*p)->rb_right;
 664                 else
 665                         BUG();
 666         }
 667
 668         rb_link_node(&vma->vm_rb, parent, p);
 669         rb_insert_color(&vma->vm_rb, &mm->mm_rb);
 670
 671         /* add VMA to the VMA list also */
 672         for (pp = &mm->mmap; (pvma = *pp); pp = &(*pp)->vm_next) {
 673                 if (pvma->vm_start > vma->vm_start)
 674                         break;
 675                 if (pvma->vm_start < vma->vm_start)
 676                         continue;
 677                 if (pvma->vm_end < vma->vm_end)
 678                         break;
 679         }
 680
 681         vma->vm_next = *pp;
 682         *pp = vma;
 683 }
 684
 685 /*
 686  * delete a VMA from its owning mm_struct and address space
 687  */
 688 static void delete_vma_from_mm(struct vm_area_struct *vma)
 689 {
 690         struct vm_area_struct **pp;
 691         struct address_space *mapping;
 692         struct mm_struct *mm = vma->vm_mm;
 693
 694         kenter("%p", vma);
 695
 696         mm->map_count--;
 697         if (mm->mmap_cache == vma)
 698                 mm->mmap_cache = NULL;
 699
 700         /* remove the VMA from the mapping */
 701         if (vma->vm_file) {
 702                 mapping = vma->vm_file->f_mapping;
 703
 704                 flush_dcache_mmap_lock(mapping);
 705                 vma_prio_tree_remove(vma, &mapping->i_mmap);
 706                 flush_dcache_mmap_unlock(mapping);
 707         }
 708
 709         /* remove from the MM's tree and list */
 710         rb_erase(&vma->vm_rb, &mm->mm_rb);
 711         for (pp = &mm->mmap; *pp; pp = &(*pp)->vm_next) {
 712                 if (*pp == vma) {
 713                         *pp = vma->vm_next;
 714                         break;
 715                 }
 716         }
 717
 718         vma->vm_mm = NULL;
 719 }
 720
 721 /*
 722  * destroy a VMA record
 723  */
 724 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
 725 {
 726         kenter("%p", vma);
 727         if (vma->vm_ops && vma->vm_ops->close)
 728                 vma->vm_ops->close(vma);
 729         if (vma->vm_file) {
 730                 fput(vma->vm_file);
 731                 if (vma->vm_flags & VM_EXECUTABLE)
 732                         removed_exe_file_vma(mm);
 733         }
 734         put_nommu_region(vma->vm_region);
 735         kmem_cache_free(vm_area_cachep, vma);
 736 }
 737
 738 /*
 739  * look up the first VMA in which addr resides, NULL if none
 740  * - should be called with mm->mmap_sem at least held readlocked
 741  */
 742 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
 743 {
 744         struct vm_area_struct *vma;
 745         struct rb_node *n = mm->mm_rb.rb_node;
 746
 747         /* check the cache first */
 748         vma = mm->mmap_cache;
 749         if (vma && vma->vm_start <= addr && vma->vm_end > addr)
 750                 return vma;
 751
 752         /* trawl the tree (there may be multiple mappings in which addr
 753          * resides) */
 754         for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
 755                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
 756                 if (vma->vm_start > addr)
 757                         return NULL;
 758                 if (vma->vm_end > addr) {
 759                         mm->mmap_cache = vma;
 760                         return vma;
 761                 }
 762         }
 763
 764         return NULL;
 765 }
 766 EXPORT_SYMBOL(find_vma);
 767
 768 /*
 769  * find a VMA
 770  * - we don't extend stack VMAs under NOMMU conditions
 771  */
 772 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
 773 {
 774         return find_vma(mm, addr);
 775 }
 776
 777 /*
 778  * expand a stack to a given address
 779  * - not supported under NOMMU conditions
 780  */
 781 int expand_stack(struct vm_area_struct *vma, unsigned long address)
 782 {
 783         return -ENOMEM;
 784 }
 785
 786 /*
 787  * look up the first VMA exactly that exactly matches addr
 788  * - should be called with mm->mmap_sem at least held readlocked
 789  */
 790 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
 791                                              unsigned long addr,
 792                                              unsigned long len)
 793 {
 794         struct vm_area_struct *vma;
 795         struct rb_node *n = mm->mm_rb.rb_node;
 796         unsigned long end = addr + len;
 797
 798         /* check the cache first */
 799         vma = mm->mmap_cache;
 800         if (vma && vma->vm_start == addr && vma->vm_end == end)
 801                 return vma;
 802
 803         /* trawl the tree (there may be multiple mappings in which addr
 804          * resides) */
 805         for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
 806                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
 807                 if (vma->vm_start < addr)
 808                         continue;
 809                 if (vma->vm_start > addr)
 810                         return NULL;
 811                 if (vma->vm_end == end) {
 812                         mm->mmap_cache = vma;
 813                         return vma;
 814                 }
 815         }
 816
 817         return NULL;
 818 }
 819
 820 /*
 821  * determine whether a mapping should be permitted and, if so, what sort of
 822  * mapping we're capable of supporting
 823  */
 824 static int validate_mmap_request(struct file *file,
 825                                  unsigned long addr,
 826                                  unsigned long len,
 827                                  unsigned long prot,
 828                                  unsigned long flags,
 829                                  unsigned long pgoff,
 830                                  unsigned long *_capabilities)
 831 {
 832         unsigned long capabilities, rlen;
 833         unsigned long reqprot = prot;
 834         int ret;
 835
 836         /* do the simple checks first */
 837         if (flags & MAP_FIXED || addr) {
 838                 printk(KERN_DEBUG
 839                        "%d: Can't do fixed-address/overlay mmap of RAM\n",
 840                        current->pid);
 841                 return -EINVAL;
 842         }
 843
 844         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
 845             (flags & MAP_TYPE) != MAP_SHARED)
 846                 return -EINVAL;
 847
 848         if (!len)
 849                 return -EINVAL;
 850
 851         /* Careful about overflows.. */
 852         rlen = PAGE_ALIGN(len);
 853         if (!rlen || rlen > TASK_SIZE)
 854                 return -ENOMEM;
 855
 856         /* offset overflow? */
 857         if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
 858                 return -EOVERFLOW;
 859
 860         if (file) {
 861                 /* validate file mapping requests */
 862                 struct address_space *mapping;
 863
 864                 /* files must support mmap */
 865                 if (!file->f_op || !file->f_op->mmap)
 866                         return -ENODEV;
 867
 868                 /* work out if what we've got could possibly be shared
 869                  * - we support chardevs that provide their own "memory"
 870                  * - we support files/blockdevs that are memory backed
 871                  */
 872                 mapping = file->f_mapping;
 873                 if (!mapping)
 874                         mapping = file->f_path.dentry->d_inode->i_mapping;
 875
 876                 capabilities = 0;
 877                 if (mapping && mapping->backing_dev_info)
 878                         capabilities = mapping->backing_dev_info->capabilities;
 879
 880                 if (!capabilities) {
 881                         /* no explicit capabilities set, so assume some
 882                          * defaults */
 883                         switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
 884                         case S_IFREG:
 885                         case S_IFBLK:
 886                                 capabilities = BDI_CAP_MAP_COPY;
 887                                 break;
 888
 889                         case S_IFCHR:
 890                                 capabilities =
 891                                         BDI_CAP_MAP_DIRECT |
 892                                         BDI_CAP_READ_MAP |
 893                                         BDI_CAP_WRITE_MAP;
 894                                 break;
 895
 896                         default:
 897                                 return -EINVAL;
 898                         }
 899                 }
 900
 901                 /* eliminate any capabilities that we can't support on this
 902                  * device */
 903                 if (!file->f_op->get_unmapped_area)
 904                         capabilities &= ~BDI_CAP_MAP_DIRECT;
 905                 if (!file->f_op->read)
 906                         capabilities &= ~BDI_CAP_MAP_COPY;
 907
 908                 if (flags & MAP_SHARED) {
 909                         /* do checks for writing, appending and locking */
 910                         if ((prot & PROT_WRITE) &&
 911                             !(file->f_mode & FMODE_WRITE))
 912                                 return -EACCES;
 913
 914                         if (IS_APPEND(file->f_path.dentry->d_inode) &&
 915                             (file->f_mode & FMODE_WRITE))
 916                                 return -EACCES;
 917
 918                         if (locks_verify_locked(file->f_path.dentry->d_inode))
 919                                 return -EAGAIN;
 920
 921                         if (!(capabilities & BDI_CAP_MAP_DIRECT))
 922                                 return -ENODEV;
 923
 924                         if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
 925                             ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
 926                             ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
 927                             ) {
 928                                 printk("MAP_SHARED not completely supported on !MMU\n");
 929                                 return -EINVAL;
 930                         }
 931
 932                         /* we mustn't privatise shared mappings */
 933                         capabilities &= ~BDI_CAP_MAP_COPY;
 934                 }
 935                 else {
 936                         /* we're going to read the file into private memory we
 937                          * allocate */
 938                         if (!(capabilities & BDI_CAP_MAP_COPY))
 939                                 return -ENODEV;
 940
 941                         /* we don't permit a private writable mapping to be
 942                          * shared with the backing device */
 943                         if (prot & PROT_WRITE)
 944                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
 945                 }
 946
 947                 /* handle executable mappings and implied executable
 948                  * mappings */
 949                 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
 950                         if (prot & PROT_EXEC)
 951                                 return -EPERM;
 952                 }
 953                 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
 954                         /* handle implication of PROT_EXEC by PROT_READ */
 955                         if (current->personality & READ_IMPLIES_EXEC) {
 956                                 if (capabilities & BDI_CAP_EXEC_MAP)
 957                                         prot |= PROT_EXEC;
 958                         }
 959                 }
 960                 else if ((prot & PROT_READ) &&
 961                          (prot & PROT_EXEC) &&
 962                          !(capabilities & BDI_CAP_EXEC_MAP)
 963                          ) {
 964                         /* backing file is not executable, try to copy */
 965                         capabilities &= ~BDI_CAP_MAP_DIRECT;
 966                 }
 967         }
 968         else {
 969                 /* anonymous mappings are always memory backed and can be
 970                  * privately mapped
 971                  */
 972                 capabilities = BDI_CAP_MAP_COPY;
 973
 974                 /* handle PROT_EXEC implication by PROT_READ */
 975                 if ((prot & PROT_READ) &&
 976                     (current->personality & READ_IMPLIES_EXEC))
 977                         prot |= PROT_EXEC;
 978         }
 979
 980         /* allow the security API to have its say */
 981         ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
 982         if (ret < 0)
 983                 return ret;
 984
 985         /* looks okay */
 986         *_capabilities = capabilities;
 987         return 0;
 988 }
 989
 990 /*
 991  * we've determined that we can make the mapping, now translate what we
 992  * now know into VMA flags
 993  */
 994 static unsigned long determine_vm_flags(struct file *file,
 995                                         unsigned long prot,
 996                                         unsigned long flags,
 997                                         unsigned long capabilities)
 998 {
 999         unsigned long vm_flags;
1000
1001         vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
1002         vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1003         /* vm_flags |= mm->def_flags; */
1004
1005         if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
1006                 /* attempt to share read-only copies of mapped file chunks */
1007                 if (file && !(prot & PROT_WRITE))
1008                         vm_flags |= VM_MAYSHARE;
1009         }
1010         else {
1011                 /* overlay a shareable mapping on the backing device or inode
1012                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1013                  * romfs/cramfs */
1014                 if (flags & MAP_SHARED)
1015                         vm_flags |= VM_MAYSHARE | VM_SHARED;
1016                 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
1017                         vm_flags |= VM_MAYSHARE;
1018         }
1019
1020         /* refuse to let anyone share private mappings with this process if
1021          * it's being traced - otherwise breakpoints set in it may interfere
1022          * with another untraced process
1023          */
1024         if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current))
1025                 vm_flags &= ~VM_MAYSHARE;
1026
1027         return vm_flags;
1028 }
1029
1030 /*
1031  * set up a shared mapping on a file (the driver or filesystem provides and
1032  * pins the storage)
1033  */
1034 static int do_mmap_shared_file(struct vm_area_struct *vma)
1035 {
1036         int ret;
1037
1038         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1039         if (ret == 0) {
1040                 vma->vm_region->vm_top = vma->vm_region->vm_end;
1041                 return ret;
1042         }
1043         if (ret != -ENOSYS)
1044                 return ret;
1045
1046         /* getting an ENOSYS error indicates that direct mmap isn't
1047          * possible (as opposed to tried but failed) so we'll fall
1048          * through to making a private copy of the data and mapping
1049          * that if we can */
1050         return -ENODEV;
1051 }
1052
1053 /*
1054  * set up a private mapping or an anonymous shared mapping
1055  */
1056 static int do_mmap_private(struct vm_area_struct *vma,
1057                            struct vm_region *region,
1058                            unsigned long len)
1059 {
1060         struct page *pages;
1061         unsigned long total, point, n, rlen;
1062         void *base;
1063         int ret, order;
1064
1065         /* invoke the file's mapping function so that it can keep track of
1066          * shared mappings on devices or memory
1067          * - VM_MAYSHARE will be set if it may attempt to share
1068          */
1069         if (vma->vm_file) {
1070                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1071                 if (ret == 0) {
1072                         /* shouldn't return success if we're not sharing */
1073                         BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1074                         vma->vm_region->vm_top = vma->vm_region->vm_end;
1075                         return ret;
1076                 }
1077                 if (ret != -ENOSYS)
1078                         return ret;
1079
1080                 /* getting an ENOSYS error indicates that direct mmap isn't
1081                  * possible (as opposed to tried but failed) so we'll try to
1082                  * make a private copy of the data and map that instead */
1083         }
1084
1085         rlen = PAGE_ALIGN(len);
1086
1087         /* allocate some memory to hold the mapping
1088          * - note that this may not return a page-aligned address if the object
1089          *   we're allocating is smaller than a page
1090          */
1091         order = get_order(rlen);
1092         kdebug("alloc order %d for %lx", order, len);
1093
1094         pages = alloc_pages(GFP_KERNEL, order);
1095         if (!pages)
1096                 goto enomem;
1097
1098         total = 1 << order;
1099         atomic_add(total, &mmap_pages_allocated);
1100
1101         point = rlen >> PAGE_SHIFT;
1102
1103         /* we allocated a power-of-2 sized page set, so we may want to trim off
1104          * the excess */
1105         if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
1106                 while (total > point) {
1107                         order = ilog2(total - point);
1108                         n = 1 << order;
1109                         kdebug("shave %lu/%lu @%lu", n, total - point, total);
1110                         atomic_sub(n, &mmap_pages_allocated);
1111                         total -= n;
1112                         set_page_refcounted(pages + total);
1113                         __free_pages(pages + total, order);
1114                 }
1115         }
1116
1117         for (point = 1; point < total; point++)
1118                 set_page_refcounted(&pages[point]);
1119
1120         base = page_address(pages);
1121         region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1122         region->vm_start = (unsigned long) base;
1123         region->vm_end   = region->vm_start + rlen;
1124         region->vm_top   = region->vm_start + (total << PAGE_SHIFT);
1125
1126         vma->vm_start = region->vm_start;
1127         vma->vm_end   = region->vm_start + len;
1128
1129         if (vma->vm_file) {
1130                 /* read the contents of a file into the copy */
1131                 mm_segment_t old_fs;
1132                 loff_t fpos;
1133
1134                 fpos = vma->vm_pgoff;
1135                 fpos <<= PAGE_SHIFT;
1136
1137                 old_fs = get_fs();
1138                 set_fs(KERNEL_DS);
1139                 ret = vma->vm_file->f_op->read(vma->vm_file, base, rlen, &fpos);
1140                 set_fs(old_fs);
1141
1142                 if (ret < 0)
1143                         goto error_free;
1144
1145                 /* clear the last little bit */
1146                 if (ret < rlen)
1147                         memset(base + ret, 0, rlen - ret);
1148
1149         } else {
1150                 /* if it's an anonymous mapping, then just clear it */
1151                 memset(base, 0, rlen);
1152         }
1153
1154         return 0;
1155
1156 error_free:
1157         free_page_series(region->vm_start, region->vm_end);
1158         region->vm_start = vma->vm_start = 0;
1159         region->vm_end   = vma->vm_end = 0;
1160         region->vm_top   = 0;
1161         return ret;
1162
1163 enomem:
1164         printk("Allocation of length %lu from process %d failed\n",
1165                len, current->pid);
1166         show_free_areas();
1167         return -ENOMEM;
1168 }
1169
1170 /*
1171  * handle mapping creation for uClinux
1172  */
1173 unsigned long do_mmap_pgoff(struct file *file,
1174                             unsigned long addr,
1175                             unsigned long len,
1176                             unsigned long prot,
1177                             unsigned long flags,
1178                             unsigned long pgoff)
1179 {
1180         struct vm_area_struct *vma;
1181         struct vm_region *region;
1182         struct rb_node *rb;
1183         unsigned long capabilities, vm_flags, result;
1184         int ret;
1185
1186         kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1187
1188         if (!(flags & MAP_FIXED))
1189                 addr = round_hint_to_min(addr);
1190
1191         /* decide whether we should attempt the mapping, and if so what sort of
1192          * mapping */
1193         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1194                                     &capabilities);
1195         if (ret < 0) {
1196                 kleave(" = %d [val]", ret);
1197                 return ret;
1198         }
1199
1200         /* we've determined that we can make the mapping, now translate what we
1201          * now know into VMA flags */
1202         vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1203
1204         /* we're going to need to record the mapping */
1205         region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1206         if (!region)
1207                 goto error_getting_region;
1208
1209         vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1210         if (!vma)
1211                 goto error_getting_vma;
1212
1213         atomic_set(&region->vm_usage, 1);
1214         region->vm_flags = vm_flags;
1215         region->vm_pgoff = pgoff;
1216
1217         INIT_LIST_HEAD(&vma->anon_vma_node);
1218         vma->vm_flags = vm_flags;
1219         vma->vm_pgoff = pgoff;
1220
1221         if (file) {
1222                 region->vm_file = file;
1223                 get_file(file);
1224                 vma->vm_file = file;
1225                 get_file(file);
1226                 if (vm_flags & VM_EXECUTABLE) {
1227                         added_exe_file_vma(current->mm);
1228                         vma->vm_mm = current->mm;
1229                 }
1230         }
1231
1232         down_write(&nommu_region_sem);
1233
1234         /* if we want to share, we need to check for regions created by other
1235          * mmap() calls that overlap with our proposed mapping
1236          * - we can only share with a superset match on most regular files
1237          * - shared mappings on character devices and memory backed files are
1238          *   permitted to overlap inexactly as far as we are concerned for in
1239          *   these cases, sharing is handled in the driver or filesystem rather
1240          *   than here
1241          */
1242         if (vm_flags & VM_MAYSHARE) {
1243                 struct vm_region *pregion;
1244                 unsigned long pglen, rpglen, pgend, rpgend, start;
1245
1246                 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1247                 pgend = pgoff + pglen;
1248
1249                 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1250                         pregion = rb_entry(rb, struct vm_region, vm_rb);
1251
1252                         if (!(pregion->vm_flags & VM_MAYSHARE))
1253                                 continue;
1254
1255                         /* search for overlapping mappings on the same file */
1256                         if (pregion->vm_file->f_path.dentry->d_inode !=
1257                             file->f_path.dentry->d_inode)
1258                                 continue;
1259
1260                         if (pregion->vm_pgoff >= pgend)
1261                                 continue;
1262
1263                         rpglen = pregion->vm_end - pregion->vm_start;
1264                         rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1265                         rpgend = pregion->vm_pgoff + rpglen;
1266                         if (pgoff >= rpgend)
1267                                 continue;
1268
1269                         /* handle inexactly overlapping matches between
1270                          * mappings */
1271                         if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1272                             !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1273                                 /* new mapping is not a subset of the region */
1274                                 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1275                                         goto sharing_violation;
1276                                 continue;
1277                         }
1278
1279                         /* we've found a region we can share */
1280                         atomic_inc(&pregion->vm_usage);
1281                         vma->vm_region = pregion;
1282                         start = pregion->vm_start;
1283                         start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1284                         vma->vm_start = start;
1285                         vma->vm_end = start + len;
1286
1287                         if (pregion->vm_flags & VM_MAPPED_COPY) {
1288                                 kdebug("share copy");
1289                                 vma->vm_flags |= VM_MAPPED_COPY;
1290                         } else {
1291                                 kdebug("share mmap");
1292                                 ret = do_mmap_shared_file(vma);
1293                                 if (ret < 0) {
1294                                         vma->vm_region = NULL;
1295                                         vma->vm_start = 0;
1296                                         vma->vm_end = 0;
1297                                         atomic_dec(&pregion->vm_usage);
1298                                         pregion = NULL;
1299                                         goto error_just_free;
1300                                 }
1301                         }
1302                         fput(region->vm_file);
1303                         kmem_cache_free(vm_region_jar, region);
1304                         region = pregion;
1305                         result = start;
1306                         goto share;
1307                 }
1308
1309                 /* obtain the address at which to make a shared mapping
1310                  * - this is the hook for quasi-memory character devices to
1311                  *   tell us the location of a shared mapping
1312                  */
1313                 if (file && file->f_op->get_unmapped_area) {
1314                         addr = file->f_op->get_unmapped_area(file, addr, len,
1315                                                              pgoff, flags);
1316                         if (IS_ERR((void *) addr)) {
1317                                 ret = addr;
1318                                 if (ret != (unsigned long) -ENOSYS)
1319                                         goto error_just_free;
1320
1321                                 /* the driver refused to tell us where to site
1322                                  * the mapping so we'll have to attempt to copy
1323                                  * it */
1324                                 ret = (unsigned long) -ENODEV;
1325                                 if (!(capabilities & BDI_CAP_MAP_COPY))
1326                                         goto error_just_free;
1327
1328                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
1329                         } else {
1330                                 vma->vm_start = region->vm_start = addr;
1331                                 vma->vm_end = region->vm_end = addr + len;
1332                         }
1333                 }
1334         }
1335
1336         vma->vm_region = region;
1337
1338         /* set up the mapping */
1339         if (file && vma->vm_flags & VM_SHARED)
1340                 ret = do_mmap_shared_file(vma);
1341         else
1342                 ret = do_mmap_private(vma, region, len);
1343         if (ret < 0)
1344                 goto error_put_region;
1345
1346         add_nommu_region(region);
1347
1348         /* okay... we have a mapping; now we have to register it */
1349         result = vma->vm_start;
1350
1351         current->mm->total_vm += len >> PAGE_SHIFT;
1352
1353 share:
1354         add_vma_to_mm(current->mm, vma);
1355
1356         up_write(&nommu_region_sem);
1357
1358         if (prot & PROT_EXEC)
1359                 flush_icache_range(result, result + len);
1360
1361         kleave(" = %lx", result);
1362         return result;
1363
1364 error_put_region:
1365         __put_nommu_region(region);
1366         if (vma) {
1367                 if (vma->vm_file) {
1368                         fput(vma->vm_file);
1369                         if (vma->vm_flags & VM_EXECUTABLE)
1370                                 removed_exe_file_vma(vma->vm_mm);
1371                 }
1372                 kmem_cache_free(vm_area_cachep, vma);
1373         }
1374         kleave(" = %d [pr]", ret);
1375         return ret;
1376
1377 error_just_free:
1378         up_write(&nommu_region_sem);
1379 error:
1380         fput(region->vm_file);
1381         kmem_cache_free(vm_region_jar, region);
1382         fput(vma->vm_file);
1383         if (vma->vm_flags & VM_EXECUTABLE)
1384                 removed_exe_file_vma(vma->vm_mm);
1385         kmem_cache_free(vm_area_cachep, vma);
1386         kleave(" = %d", ret);
1387         return ret;
1388
1389 sharing_violation:
1390         up_write(&nommu_region_sem);
1391         printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1392         ret = -EINVAL;
1393         goto error;
1394
1395 error_getting_vma:
1396         kmem_cache_free(vm_region_jar, region);
1397         printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1398                " from process %d failed\n",
1399                len, current->pid);
1400         show_free_areas();
1401         return -ENOMEM;
1402
1403 error_getting_region:
1404         printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1405                " from process %d failed\n",
1406                len, current->pid);
1407         show_free_areas();
1408         return -ENOMEM;
1409 }
1410 EXPORT_SYMBOL(do_mmap_pgoff);
1411
1412 /*
1413  * split a vma into two pieces at address 'addr', a new vma is allocated either
1414  * for the first part or the tail.
1415  */
1416 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1417               unsigned long addr, int new_below)
1418 {
1419         struct vm_area_struct *new;
1420         struct vm_region *region;
1421         unsigned long npages;
1422
1423         kenter("");
1424
1425         /* we're only permitted to split anonymous regions that have a single
1426          * owner */
1427         if (vma->vm_file ||
1428             atomic_read(&vma->vm_region->vm_usage) != 1)
1429                 return -ENOMEM;
1430
1431         if (mm->map_count >= sysctl_max_map_count)
1432                 return -ENOMEM;
1433
1434         region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1435         if (!region)
1436                 return -ENOMEM;
1437
1438         new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1439         if (!new) {
1440                 kmem_cache_free(vm_region_jar, region);
1441                 return -ENOMEM;
1442         }
1443
1444         /* most fields are the same, copy all, and then fixup */
1445         *new = *vma;
1446         *region = *vma->vm_region;
1447         new->vm_region = region;
1448
1449         npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1450
1451         if (new_below) {
1452                 region->vm_top = region->vm_end = new->vm_end = addr;
1453         } else {
1454                 region->vm_start = new->vm_start = addr;
1455                 region->vm_pgoff = new->vm_pgoff += npages;
1456         }
1457
1458         if (new->vm_ops && new->vm_ops->open)
1459                 new->vm_ops->open(new);
1460
1461         delete_vma_from_mm(vma);
1462         down_write(&nommu_region_sem);
1463         delete_nommu_region(vma->vm_region);
1464         if (new_below) {
1465                 vma->vm_region->vm_start = vma->vm_start = addr;
1466                 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1467         } else {
1468                 vma->vm_region->vm_end = vma->vm_end = addr;
1469                 vma->vm_region->vm_top = addr;
1470         }
1471         add_nommu_region(vma->vm_region);
1472         add_nommu_region(new->vm_region);
1473         up_write(&nommu_region_sem);
1474         add_vma_to_mm(mm, vma);
1475         add_vma_to_mm(mm, new);
1476         return 0;
1477 }
1478
1479 /*
1480  * shrink a VMA by removing the specified chunk from either the beginning or
1481  * the end
1482  */
1483 static int shrink_vma(struct mm_struct *mm,
1484                       struct vm_area_struct *vma,
1485                       unsigned long from, unsigned long to)
1486 {
1487         struct vm_region *region;
1488
1489         kenter("");
1490
1491         /* adjust the VMA's pointers, which may reposition it in the MM's tree
1492          * and list */
1493         delete_vma_from_mm(vma);
1494         if (from > vma->vm_start)
1495                 vma->vm_end = from;
1496         else
1497                 vma->vm_start = to;
1498         add_vma_to_mm(mm, vma);
1499
1500         /* cut the backing region down to size */
1501         region = vma->vm_region;
1502         BUG_ON(atomic_read(&region->vm_usage) != 1);
1503
1504         down_write(&nommu_region_sem);
1505         delete_nommu_region(region);
1506         if (from > region->vm_start) {
1507                 to = region->vm_top;
1508                 region->vm_top = region->vm_end = from;
1509         } else {
1510                 region->vm_start = to;
1511         }
1512         add_nommu_region(region);
1513         up_write(&nommu_region_sem);
1514
1515         free_page_series(from, to);
1516         return 0;
1517 }
1518
1519 /*
1520  * release a mapping
1521  * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1522  *   VMA, though it need not cover the whole VMA
1523  */
1524 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1525 {
1526         struct vm_area_struct *vma;
1527         struct rb_node *rb;
1528         unsigned long end = start + len;
1529         int ret;
1530
1531         kenter(",%lx,%zx", start, len);
1532
1533         if (len == 0)
1534                 return -EINVAL;
1535
1536         /* find the first potentially overlapping VMA */
1537         vma = find_vma(mm, start);
1538         if (!vma) {
1539                 printk(KERN_WARNING
1540                        "munmap of memory not mmapped by process %d (%s):"
1541                        " 0x%lx-0x%lx\n",
1542                        current->pid, current->comm, start, start + len - 1);
1543                 return -EINVAL;
1544         }
1545
1546         /* we're allowed to split an anonymous VMA but not a file-backed one */
1547         if (vma->vm_file) {
1548                 do {
1549                         if (start > vma->vm_start) {
1550                                 kleave(" = -EINVAL [miss]");
1551                                 return -EINVAL;
1552                         }
1553                         if (end == vma->vm_end)
1554                                 goto erase_whole_vma;
1555                         rb = rb_next(&vma->vm_rb);
1556                         vma = rb_entry(rb, struct vm_area_struct, vm_rb);
1557                 } while (rb);
1558                 kleave(" = -EINVAL [split file]");
1559                 return -EINVAL;
1560         } else {
1561                 /* the chunk must be a subset of the VMA found */
1562                 if (start == vma->vm_start && end == vma->vm_end)
1563                         goto erase_whole_vma;
1564                 if (start < vma->vm_start || end > vma->vm_end) {
1565                         kleave(" = -EINVAL [superset]");
1566                         return -EINVAL;
1567                 }
1568                 if (start & ~PAGE_MASK) {
1569                         kleave(" = -EINVAL [unaligned start]");
1570                         return -EINVAL;
1571                 }
1572                 if (end != vma->vm_end && end & ~PAGE_MASK) {
1573                         kleave(" = -EINVAL [unaligned split]");
1574                         return -EINVAL;
1575                 }
1576                 if (start != vma->vm_start && end != vma->vm_end) {
1577                         ret = split_vma(mm, vma, start, 1);
1578                         if (ret < 0) {
1579                                 kleave(" = %d [split]", ret);
1580                                 return ret;
1581                         }
1582                 }
1583                 return shrink_vma(mm, vma, start, end);
1584         }
1585
1586 erase_whole_vma:
1587         delete_vma_from_mm(vma);
1588         delete_vma(mm, vma);
1589         kleave(" = 0");
1590         return 0;
1591 }
1592 EXPORT_SYMBOL(do_munmap);
1593
1594 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1595 {
1596         int ret;
1597         struct mm_struct *mm = current->mm;
1598
1599         down_write(&mm->mmap_sem);
1600         ret = do_munmap(mm, addr, len);
1601         up_write(&mm->mmap_sem);
1602         return ret;
1603 }
1604
1605 /*
1606  * release all the mappings made in a process's VM space
1607  */
1608 void exit_mmap(struct mm_struct *mm)
1609 {
1610         struct vm_area_struct *vma;
1611
1612         if (!mm)
1613                 return;
1614
1615         kenter("");
1616
1617         mm->total_vm = 0;
1618
1619         while ((vma = mm->mmap)) {
1620                 mm->mmap = vma->vm_next;
1621                 delete_vma_from_mm(vma);
1622                 delete_vma(mm, vma);
1623         }
1624
1625         kleave("");
1626 }
1627
1628 unsigned long do_brk(unsigned long addr, unsigned long len)
1629 {
1630         return -ENOMEM;
1631 }
1632
1633 /*
1634  * expand (or shrink) an existing mapping, potentially moving it at the same
1635  * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1636  *
1637  * under NOMMU conditions, we only permit changing a mapping's size, and only
1638  * as long as it stays within the region allocated by do_mmap_private() and the
1639  * block is not shareable
1640  *
1641  * MREMAP_FIXED is not supported under NOMMU conditions
1642  */
1643 unsigned long do_mremap(unsigned long addr,
1644                         unsigned long old_len, unsigned long new_len,
1645                         unsigned long flags, unsigned long new_addr)
1646 {
1647         struct vm_area_struct *vma;
1648
1649         /* insanity checks first */
1650         if (old_len == 0 || new_len == 0)
1651                 return (unsigned long) -EINVAL;
1652
1653         if (addr & ~PAGE_MASK)
1654                 return -EINVAL;
1655
1656         if (flags & MREMAP_FIXED && new_addr != addr)
1657                 return (unsigned long) -EINVAL;
1658
1659         vma = find_vma_exact(current->mm, addr, old_len);
1660         if (!vma)
1661                 return (unsigned long) -EINVAL;
1662
1663         if (vma->vm_end != vma->vm_start + old_len)
1664                 return (unsigned long) -EFAULT;
1665
1666         if (vma->vm_flags & VM_MAYSHARE)
1667                 return (unsigned long) -EPERM;
1668
1669         if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1670                 return (unsigned long) -ENOMEM;
1671
1672         /* all checks complete - do it */
1673         vma->vm_end = vma->vm_start + new_len;
1674         return vma->vm_start;
1675 }
1676 EXPORT_SYMBOL(do_mremap);
1677
1678 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1679                 unsigned long, new_len, unsigned long, flags,
1680                 unsigned long, new_addr)
1681 {
1682         unsigned long ret;
1683
1684         down_write(&current->mm->mmap_sem);
1685         ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1686         up_write(&current->mm->mmap_sem);
1687         return ret;
1688 }
1689
1690 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1691                         unsigned int foll_flags)
1692 {
1693         return NULL;
1694 }
1695
1696 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1697                 unsigned long to, unsigned long size, pgprot_t prot)
1698 {
1699         vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1700         return 0;
1701 }
1702 EXPORT_SYMBOL(remap_pfn_range);
1703
1704 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1705                         unsigned long pgoff)
1706 {
1707         unsigned int size = vma->vm_end - vma->vm_start;
1708
1709         if (!(vma->vm_flags & VM_USERMAP))
1710                 return -EINVAL;
1711
1712         vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1713         vma->vm_end = vma->vm_start + size;
1714
1715         return 0;
1716 }
1717 EXPORT_SYMBOL(remap_vmalloc_range);
1718
1719 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1720 {
1721 }
1722
1723 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1724         unsigned long len, unsigned long pgoff, unsigned long flags)
1725 {
1726         return -ENOMEM;
1727 }
1728
1729 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1730 {
1731 }
1732
1733 void unmap_mapping_range(struct address_space *mapping,
1734                          loff_t const holebegin, loff_t const holelen,
1735                          int even_cows)
1736 {
1737 }
1738 EXPORT_SYMBOL(unmap_mapping_range);
1739
1740 /*
1741  * ask for an unmapped area at which to create a mapping on a file
1742  */
1743 unsigned long get_unmapped_area(struct file *file, unsigned long addr,
1744                                 unsigned long len, unsigned long pgoff,
1745                                 unsigned long flags)
1746 {
1747         unsigned long (*get_area)(struct file *, unsigned long, unsigned long,
1748                                   unsigned long, unsigned long);
1749
1750         get_area = current->mm->get_unmapped_area;
1751         if (file && file->f_op && file->f_op->get_unmapped_area)
1752                 get_area = file->f_op->get_unmapped_area;
1753
1754         if (!get_area)
1755                 return -ENOSYS;
1756
1757         return get_area(file, addr, len, pgoff, flags);
1758 }
1759 EXPORT_SYMBOL(get_unmapped_area);
1760
1761 /*
1762  * Check that a process has enough memory to allocate a new virtual
1763  * mapping. 0 means there is enough memory for the allocation to
1764  * succeed and -ENOMEM implies there is not.
1765  *
1766  * We currently support three overcommit policies, which are set via the
1767  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1768  *
1769  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1770  * Additional code 2002 Jul 20 by Robert Love.
1771  *
1772  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1773  *
1774  * Note this is a helper function intended to be used by LSMs which
1775  * wish to use this logic.
1776  */
1777 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1778 {
1779         unsigned long free, allowed;
1780
1781         vm_acct_memory(pages);
1782
1783         /*
1784          * Sometimes we want to use more memory than we have
1785          */
1786         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1787                 return 0;
1788
1789         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1790                 unsigned long n;
1791
1792                 free = global_page_state(NR_FILE_PAGES);
1793                 free += nr_swap_pages;
1794
1795                 /*
1796                  * Any slabs which are created with the
1797                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1798                  * which are reclaimable, under pressure.  The dentry
1799                  * cache and most inode caches should fall into this
1800                  */
1801                 free += global_page_state(NR_SLAB_RECLAIMABLE);
1802
1803                 /*
1804                  * Leave the last 3% for root
1805                  */
1806                 if (!cap_sys_admin)
1807                         free -= free / 32;
1808
1809                 if (free > pages)
1810                         return 0;
1811
1812                 /*
1813                  * nr_free_pages() is very expensive on large systems,
1814                  * only call if we're about to fail.
1815                  */
1816                 n = nr_free_pages();
1817
1818                 /*
1819                  * Leave reserved pages. The pages are not for anonymous pages.
1820                  */
1821                 if (n <= totalreserve_pages)
1822                         goto error;
1823                 else
1824                         n -= totalreserve_pages;
1825
1826                 /*
1827                  * Leave the last 3% for root
1828                  */
1829                 if (!cap_sys_admin)
1830                         n -= n / 32;
1831                 free += n;
1832
1833                 if (free > pages)
1834                         return 0;
1835
1836                 goto error;
1837         }
1838
1839         allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1840         /*
1841          * Leave the last 3% for root
1842          */
1843         if (!cap_sys_admin)
1844                 allowed -= allowed / 32;
1845         allowed += total_swap_pages;
1846
1847         /* Don't let a single process grow too big:
1848            leave 3% of the size of this process for other processes */
1849         if (mm)
1850                 allowed -= mm->total_vm / 32;
1851
1852         /*
1853          * cast `allowed' as a signed long because vm_committed_space
1854          * sometimes has a negative value
1855          */
1856         if (atomic_long_read(&vm_committed_space) < (long)allowed)
1857                 return 0;
1858 error:
1859         vm_unacct_memory(pages);
1860
1861         return -ENOMEM;
1862 }
1863
1864 int in_gate_area_no_task(unsigned long addr)
1865 {
1866         return 0;
1867 }
1868
1869 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1870 {
1871         BUG();
1872         return 0;
1873 }
1874 EXPORT_SYMBOL(filemap_fault);
1875
1876 /*
1877  * Access another process' address space.
1878  * - source/target buffer must be kernel space
1879  */
1880 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1881 {
1882         struct vm_area_struct *vma;
1883         struct mm_struct *mm;
1884
1885         if (addr + len < addr)
1886                 return 0;
1887
1888         mm = get_task_mm(tsk);
1889         if (!mm)
1890                 return 0;
1891
1892         down_read(&mm->mmap_sem);
1893
1894         /* the access must start within one of the target process's mappings */
1895         vma = find_vma(mm, addr);
1896         if (vma) {
1897                 /* don't overrun this mapping */
1898                 if (addr + len >= vma->vm_end)
1899                         len = vma->vm_end - addr;
1900
1901                 /* only read or write mappings where it is permitted */
1902                 if (write && vma->vm_flags & VM_MAYWRITE)
1903                         len -= copy_to_user((void *) addr, buf, len);
1904                 else if (!write && vma->vm_flags & VM_MAYREAD)
1905                         len -= copy_from_user(buf, (void *) addr, len);
1906                 else
1907                         len = 0;
1908         } else {
1909                 len = 0;
1910         }
1911
1912         up_read(&mm->mmap_sem);
1913         mmput(mm);
1914         return len;
1915 }