mm/vmalloc.c

   1 /*
   2  *  linux/mm/vmalloc.c
   3  *
   4  *  Copyright (C) 1993  Linus Torvalds
   5  *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
   6  *  SMP-safe vmalloc/vfree/ioremap, Tigran Aivazian <tigran@veritas.com>, May 2000
   7  *  Major rework to support vmap/vunmap, Christoph Hellwig, SGI, August 2002
   8  *  Numa awareness, Christoph Lameter, SGI, June 2005
   9  */
  10
  11 #include <linux/mm.h>
  12 #include <linux/module.h>
  13 #include <linux/highmem.h>
  14 #include <linux/slab.h>
  15 #include <linux/spinlock.h>
  16 #include <linux/interrupt.h>
  17
  18 #include <linux/vmalloc.h>
  19
  20 #include <asm/uaccess.h>
  21 #include <asm/tlbflush.h>
  22
  23
  24 DEFINE_RWLOCK(vmlist_lock);
  25 struct vm_struct *vmlist;
  26
  27 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
  28                             int node);
  29
  30 static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
  31 {
  32         pte_t *pte;
  33
  34         pte = pte_offset_kernel(pmd, addr);
  35         do {
  36                 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
  37                 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
  38         } while (pte++, addr += PAGE_SIZE, addr != end);
  39 }
  40
  41 static inline void vunmap_pmd_range(pud_t *pud, unsigned long addr,
  42                                                 unsigned long end)
  43 {
  44         pmd_t *pmd;
  45         unsigned long next;
  46
  47         pmd = pmd_offset(pud, addr);
  48         do {
  49                 next = pmd_addr_end(addr, end);
  50                 if (pmd_none_or_clear_bad(pmd))
  51                         continue;
  52                 vunmap_pte_range(pmd, addr, next);
  53         } while (pmd++, addr = next, addr != end);
  54 }
  55
  56 static inline void vunmap_pud_range(pgd_t *pgd, unsigned long addr,
  57                                                 unsigned long end)
  58 {
  59         pud_t *pud;
  60         unsigned long next;
  61
  62         pud = pud_offset(pgd, addr);
  63         do {
  64                 next = pud_addr_end(addr, end);
  65                 if (pud_none_or_clear_bad(pud))
  66                         continue;
  67                 vunmap_pmd_range(pud, addr, next);
  68         } while (pud++, addr = next, addr != end);
  69 }
  70
  71 void unmap_vm_area(struct vm_struct *area)
  72 {
  73         pgd_t *pgd;
  74         unsigned long next;
  75         unsigned long addr = (unsigned long) area->addr;
  76         unsigned long end = addr + area->size;
  77
  78         BUG_ON(addr >= end);
  79         pgd = pgd_offset_k(addr);
  80         flush_cache_vunmap(addr, end);
  81         do {
  82                 next = pgd_addr_end(addr, end);
  83                 if (pgd_none_or_clear_bad(pgd))
  84                         continue;
  85                 vunmap_pud_range(pgd, addr, next);
  86         } while (pgd++, addr = next, addr != end);
  87         flush_tlb_kernel_range((unsigned long) area->addr, end);
  88 }
  89
  90 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
  91                         unsigned long end, pgprot_t prot, struct page ***pages)
  92 {
  93         pte_t *pte;
  94
  95         pte = pte_alloc_kernel(pmd, addr);
  96         if (!pte)
  97                 return -ENOMEM;
  98         do {
  99                 struct page *page = **pages;
 100                 WARN_ON(!pte_none(*pte));
 101                 if (!page)
 102                         return -ENOMEM;
 103                 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
 104                 (*pages)++;
 105         } while (pte++, addr += PAGE_SIZE, addr != end);
 106         return 0;
 107 }
 108
 109 static inline int vmap_pmd_range(pud_t *pud, unsigned long addr,
 110                         unsigned long end, pgprot_t prot, struct page ***pages)
 111 {
 112         pmd_t *pmd;
 113         unsigned long next;
 114
 115         pmd = pmd_alloc(&init_mm, pud, addr);
 116         if (!pmd)
 117                 return -ENOMEM;
 118         do {
 119                 next = pmd_addr_end(addr, end);
 120                 if (vmap_pte_range(pmd, addr, next, prot, pages))
 121                         return -ENOMEM;
 122         } while (pmd++, addr = next, addr != end);
 123         return 0;
 124 }
 125
 126 static inline int vmap_pud_range(pgd_t *pgd, unsigned long addr,
 127                         unsigned long end, pgprot_t prot, struct page ***pages)
 128 {
 129         pud_t *pud;
 130         unsigned long next;
 131
 132         pud = pud_alloc(&init_mm, pgd, addr);
 133         if (!pud)
 134                 return -ENOMEM;
 135         do {
 136                 next = pud_addr_end(addr, end);
 137                 if (vmap_pmd_range(pud, addr, next, prot, pages))
 138                         return -ENOMEM;
 139         } while (pud++, addr = next, addr != end);
 140         return 0;
 141 }
 142
 143 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
 144 {
 145         pgd_t *pgd;
 146         unsigned long next;
 147         unsigned long addr = (unsigned long) area->addr;
 148         unsigned long end = addr + area->size - PAGE_SIZE;
 149         int err;
 150
 151         BUG_ON(addr >= end);
 152         pgd = pgd_offset_k(addr);
 153         do {
 154                 next = pgd_addr_end(addr, end);
 155                 err = vmap_pud_range(pgd, addr, next, prot, pages);
 156                 if (err)
 157                         break;
 158         } while (pgd++, addr = next, addr != end);
 159         flush_cache_vmap((unsigned long) area->addr, end);
 160         return err;
 161 }
 162
 163 struct vm_struct *__get_vm_area_node(unsigned long size, unsigned long flags,
 164                                 unsigned long start, unsigned long end, int node)
 165 {
 166         struct vm_struct **p, *tmp, *area;
 167         unsigned long align = 1;
 168         unsigned long addr;
 169
 170         if (flags & VM_IOREMAP) {
 171                 int bit = fls(size);
 172
 173                 if (bit > IOREMAP_MAX_ORDER)
 174                         bit = IOREMAP_MAX_ORDER;
 175                 else if (bit < PAGE_SHIFT)
 176                         bit = PAGE_SHIFT;
 177
 178                 align = 1ul << bit;
 179         }
 180         addr = ALIGN(start, align);
 181         size = PAGE_ALIGN(size);
 182
 183         area = kmalloc_node(sizeof(*area), GFP_KERNEL, node);
 184         if (unlikely(!area))
 185                 return NULL;
 186
 187         if (unlikely(!size)) {
 188                 kfree (area);
 189                 return NULL;
 190         }
 191
 192         /*
 193          * We always allocate a guard page.
 194          */
 195         size += PAGE_SIZE;
 196
 197         write_lock(&vmlist_lock);
 198         for (p = &vmlist; (tmp = *p) != NULL ;p = &tmp->next) {
 199                 if ((unsigned long)tmp->addr < addr) {
 200                         if((unsigned long)tmp->addr + tmp->size >= addr)
 201                                 addr = ALIGN(tmp->size +
 202                                              (unsigned long)tmp->addr, align);
 203                         continue;
 204                 }
 205                 if ((size + addr) < addr)
 206                         goto out;
 207                 if (size + addr <= (unsigned long)tmp->addr)
 208                         goto found;
 209                 addr = ALIGN(tmp->size + (unsigned long)tmp->addr, align);
 210                 if (addr > end - size)
 211                         goto out;
 212         }
 213
 214 found:
 215         area->next = *p;
 216         *p = area;
 217
 218         area->flags = flags;
 219         area->addr = (void *)addr;
 220         area->size = size;
 221         area->pages = NULL;
 222         area->nr_pages = 0;
 223         area->phys_addr = 0;
 224         write_unlock(&vmlist_lock);
 225
 226         return area;
 227
 228 out:
 229         write_unlock(&vmlist_lock);
 230         kfree(area);
 231         if (printk_ratelimit())
 232                 printk(KERN_WARNING "allocation failed: out of vmalloc space - use vmalloc=<size> to increase size.\n");
 233         return NULL;
 234 }
 235
 236 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
 237                                 unsigned long start, unsigned long end)
 238 {
 239         return __get_vm_area_node(size, flags, start, end, -1);
 240 }
 241
 242 /**
 243  *      get_vm_area  -  reserve a contingous kernel virtual area
 244  *      @size:          size of the area
 245  *      @flags:         %VM_IOREMAP for I/O mappings or VM_ALLOC
 246  *
 247  *      Search an area of @size in the kernel virtual mapping area,
 248  *      and reserved it for out purposes.  Returns the area descriptor
 249  *      on success or %NULL on failure.
 250  */
 251 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
 252 {
 253         return __get_vm_area(size, flags, VMALLOC_START, VMALLOC_END);
 254 }
 255
 256 struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags, int node)
 257 {
 258         return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node);
 259 }
 260
 261 /* Caller must hold vmlist_lock */
 262 static struct vm_struct *__find_vm_area(void *addr)
 263 {
 264         struct vm_struct *tmp;
 265
 266         for (tmp = vmlist; tmp != NULL; tmp = tmp->next) {
 267                  if (tmp->addr == addr)
 268                         break;
 269         }
 270
 271         return tmp;
 272 }
 273
 274 /* Caller must hold vmlist_lock */
 275 static struct vm_struct *__remove_vm_area(void *addr)
 276 {
 277         struct vm_struct **p, *tmp;
 278
 279         for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) {
 280                  if (tmp->addr == addr)
 281                          goto found;
 282         }
 283         return NULL;
 284
 285 found:
 286         unmap_vm_area(tmp);
 287         *p = tmp->next;
 288
 289         /*
 290          * Remove the guard page.
 291          */
 292         tmp->size -= PAGE_SIZE;
 293         return tmp;
 294 }
 295
 296 /**
 297  *      remove_vm_area  -  find and remove a contingous kernel virtual area
 298  *      @addr:          base address
 299  *
 300  *      Search for the kernel VM area starting at @addr, and remove it.
 301  *      This function returns the found VM area, but using it is NOT safe
 302  *      on SMP machines, except for its size or flags.
 303  */
 304 struct vm_struct *remove_vm_area(void *addr)
 305 {
 306         struct vm_struct *v;
 307         write_lock(&vmlist_lock);
 308         v = __remove_vm_area(addr);
 309         write_unlock(&vmlist_lock);
 310         return v;
 311 }
 312
 313 void __vunmap(void *addr, int deallocate_pages)
 314 {
 315         struct vm_struct *area;
 316
 317         if (!addr)
 318                 return;
 319
 320         if ((PAGE_SIZE-1) & (unsigned long)addr) {
 321                 printk(KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
 322                 WARN_ON(1);
 323                 return;
 324         }
 325
 326         area = remove_vm_area(addr);
 327         if (unlikely(!area)) {
 328                 printk(KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
 329                                 addr);
 330                 WARN_ON(1);
 331                 return;
 332         }
 333
 334         debug_check_no_locks_freed(addr, area->size);
 335
 336         if (deallocate_pages) {
 337                 int i;
 338
 339                 for (i = 0; i < area->nr_pages; i++) {
 340                         BUG_ON(!area->pages[i]);
 341                         __free_page(area->pages[i]);
 342                 }
 343
 344                 if (area->flags & VM_VPAGES)
 345                         vfree(area->pages);
 346                 else
 347                         kfree(area->pages);
 348         }
 349
 350         kfree(area);
 351         return;
 352 }
 353
 354 /**
 355  *      vfree  -  release memory allocated by vmalloc()
 356  *      @addr:          memory base address
 357  *
 358  *      Free the virtually contiguous memory area starting at @addr, as
 359  *      obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
 360  *      NULL, no operation is performed.
 361  *
 362  *      Must not be called in interrupt context.
 363  */
 364 void vfree(void *addr)
 365 {
 366         BUG_ON(in_interrupt());
 367         __vunmap(addr, 1);
 368 }
 369 EXPORT_SYMBOL(vfree);
 370
 371 /**
 372  *      vunmap  -  release virtual mapping obtained by vmap()
 373  *      @addr:          memory base address
 374  *
 375  *      Free the virtually contiguous memory area starting at @addr,
 376  *      which was created from the page array passed to vmap().
 377  *
 378  *      Must not be called in interrupt context.
 379  */
 380 void vunmap(void *addr)
 381 {
 382         BUG_ON(in_interrupt());
 383         __vunmap(addr, 0);
 384 }
 385 EXPORT_SYMBOL(vunmap);
 386
 387 /**
 388  *      vmap  -  map an array of pages into virtually contiguous space
 389  *      @pages:         array of page pointers
 390  *      @count:         number of pages to map
 391  *      @flags:         vm_area->flags
 392  *      @prot:          page protection for the mapping
 393  *
 394  *      Maps @count pages from @pages into contiguous kernel virtual
 395  *      space.
 396  */
 397 void *vmap(struct page **pages, unsigned int count,
 398                 unsigned long flags, pgprot_t prot)
 399 {
 400         struct vm_struct *area;
 401
 402         if (count > num_physpages)
 403                 return NULL;
 404
 405         area = get_vm_area((count << PAGE_SHIFT), flags);
 406         if (!area)
 407                 return NULL;
 408         if (map_vm_area(area, prot, &pages)) {
 409                 vunmap(area->addr);
 410                 return NULL;
 411         }
 412
 413         return area->addr;
 414 }
 415 EXPORT_SYMBOL(vmap);
 416
 417 void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
 418                                 pgprot_t prot, int node)
 419 {
 420         struct page **pages;
 421         unsigned int nr_pages, array_size, i;
 422
 423         nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
 424         array_size = (nr_pages * sizeof(struct page *));
 425
 426         area->nr_pages = nr_pages;
 427         /* Please note that the recursion is strictly bounded. */
 428         if (array_size > PAGE_SIZE) {
 429                 pages = __vmalloc_node(array_size, gfp_mask, PAGE_KERNEL, node);
 430                 area->flags |= VM_VPAGES;
 431         } else
 432                 pages = kmalloc_node(array_size, (gfp_mask & ~__GFP_HIGHMEM), node);
 433         area->pages = pages;
 434         if (!area->pages) {
 435                 remove_vm_area(area->addr);
 436                 kfree(area);
 437                 return NULL;
 438         }
 439         memset(area->pages, 0, array_size);
 440
 441         for (i = 0; i < area->nr_pages; i++) {
 442                 if (node < 0)
 443                         area->pages[i] = alloc_page(gfp_mask);
 444                 else
 445                         area->pages[i] = alloc_pages_node(node, gfp_mask, 0);
 446                 if (unlikely(!area->pages[i])) {
 447                         /* Successfully allocated i pages, free them in __vunmap() */
 448                         area->nr_pages = i;
 449                         goto fail;
 450                 }
 451         }
 452
 453         if (map_vm_area(area, prot, &pages))
 454                 goto fail;
 455         return area->addr;
 456
 457 fail:
 458         vfree(area->addr);
 459         return NULL;
 460 }
 461
 462 void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
 463 {
 464         return __vmalloc_area_node(area, gfp_mask, prot, -1);
 465 }
 466
 467 /**
 468  *      __vmalloc_node  -  allocate virtually contiguous memory
 469  *      @size:          allocation size
 470  *      @gfp_mask:      flags for the page level allocator
 471  *      @prot:          protection mask for the allocated pages
 472  *      @node:          node to use for allocation or -1
 473  *
 474  *      Allocate enough pages to cover @size from the page level
 475  *      allocator with @gfp_mask flags.  Map them into contiguous
 476  *      kernel virtual space, using a pagetable protection of @prot.
 477  */
 478 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
 479                             int node)
 480 {
 481         struct vm_struct *area;
 482
 483         size = PAGE_ALIGN(size);
 484         if (!size || (size >> PAGE_SHIFT) > num_physpages)
 485                 return NULL;
 486
 487         area = get_vm_area_node(size, VM_ALLOC, node);
 488         if (!area)
 489                 return NULL;
 490
 491         return __vmalloc_area_node(area, gfp_mask, prot, node);
 492 }
 493
 494 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
 495 {
 496         return __vmalloc_node(size, gfp_mask, prot, -1);
 497 }
 498 EXPORT_SYMBOL(__vmalloc);
 499
 500 /**
 501  *      vmalloc  -  allocate virtually contiguous memory
 502  *      @size:          allocation size
 503  *      Allocate enough pages to cover @size from the page level
 504  *      allocator and map them into contiguous kernel virtual space.
 505  *
 506  *      For tight cotrol over page level allocator and protection flags
 507  *      use __vmalloc() instead.
 508  */
 509 void *vmalloc(unsigned long size)
 510 {
 511         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
 512 }
 513 EXPORT_SYMBOL(vmalloc);
 514
 515 /**
 516  * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
 517  * @size: allocation size
 518  *
 519  * The resulting memory area is zeroed so it can be mapped to userspace
 520  * without leaking data.
 521  */
 522 void *vmalloc_user(unsigned long size)
 523 {
 524         struct vm_struct *area;
 525         void *ret;
 526
 527         ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL);
 528         write_lock(&vmlist_lock);
 529         area = __find_vm_area(ret);
 530         area->flags |= VM_USERMAP;
 531         write_unlock(&vmlist_lock);
 532
 533         return ret;
 534 }
 535 EXPORT_SYMBOL(vmalloc_user);
 536
 537 /**
 538  *      vmalloc_node  -  allocate memory on a specific node
 539  *      @size:          allocation size
 540  *      @node:          numa node
 541  *
 542  *      Allocate enough pages to cover @size from the page level
 543  *      allocator and map them into contiguous kernel virtual space.
 544  *
 545  *      For tight cotrol over page level allocator and protection flags
 546  *      use __vmalloc() instead.
 547  */
 548 void *vmalloc_node(unsigned long size, int node)
 549 {
 550         return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, node);
 551 }
 552 EXPORT_SYMBOL(vmalloc_node);
 553
 554 #ifndef PAGE_KERNEL_EXEC
 555 # define PAGE_KERNEL_EXEC PAGE_KERNEL
 556 #endif
 557
 558 /**
 559  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
 560  *      @size:          allocation size
 561  *
 562  *      Kernel-internal function to allocate enough pages to cover @size
 563  *      the page level allocator and map them into contiguous and
 564  *      executable kernel virtual space.
 565  *
 566  *      For tight cotrol over page level allocator and protection flags
 567  *      use __vmalloc() instead.
 568  */
 569
 570 void *vmalloc_exec(unsigned long size)
 571 {
 572         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
 573 }
 574
 575 /**
 576  *      vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
 577  *      @size:          allocation size
 578  *
 579  *      Allocate enough 32bit PA addressable pages to cover @size from the
 580  *      page level allocator and map them into contiguous kernel virtual space.
 581  */
 582 void *vmalloc_32(unsigned long size)
 583 {
 584         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
 585 }
 586 EXPORT_SYMBOL(vmalloc_32);
 587
 588 /**
 589  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
 590  *      @size:          allocation size
 591  *
 592  * The resulting memory area is 32bit addressable and zeroed so it can be
 593  * mapped to userspace without leaking data.
 594  */
 595 void *vmalloc_32_user(unsigned long size)
 596 {
 597         struct vm_struct *area;
 598         void *ret;
 599
 600         ret = __vmalloc(size, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
 601         write_lock(&vmlist_lock);
 602         area = __find_vm_area(ret);
 603         area->flags |= VM_USERMAP;
 604         write_unlock(&vmlist_lock);
 605
 606         return ret;
 607 }
 608 EXPORT_SYMBOL(vmalloc_32_user);
 609
 610 long vread(char *buf, char *addr, unsigned long count)
 611 {
 612         struct vm_struct *tmp;
 613         char *vaddr, *buf_start = buf;
 614         unsigned long n;
 615
 616         /* Don't allow overflow */
 617         if ((unsigned long) addr + count < count)
 618                 count = -(unsigned long) addr;
 619
 620         read_lock(&vmlist_lock);
 621         for (tmp = vmlist; tmp; tmp = tmp->next) {
 622                 vaddr = (char *) tmp->addr;
 623                 if (addr >= vaddr + tmp->size - PAGE_SIZE)
 624                         continue;
 625                 while (addr < vaddr) {
 626                         if (count == 0)
 627                                 goto finished;
 628                         *buf = '\0';
 629                         buf++;
 630                         addr++;
 631                         count--;
 632                 }
 633                 n = vaddr + tmp->size - PAGE_SIZE - addr;
 634                 do {
 635                         if (count == 0)
 636                                 goto finished;
 637                         *buf = *addr;
 638                         buf++;
 639                         addr++;
 640                         count--;
 641                 } while (--n > 0);
 642         }
 643 finished:
 644         read_unlock(&vmlist_lock);
 645         return buf - buf_start;
 646 }
 647
 648 long vwrite(char *buf, char *addr, unsigned long count)
 649 {
 650         struct vm_struct *tmp;
 651         char *vaddr, *buf_start = buf;
 652         unsigned long n;
 653
 654         /* Don't allow overflow */
 655         if ((unsigned long) addr + count < count)
 656                 count = -(unsigned long) addr;
 657
 658         read_lock(&vmlist_lock);
 659         for (tmp = vmlist; tmp; tmp = tmp->next) {
 660                 vaddr = (char *) tmp->addr;
 661                 if (addr >= vaddr + tmp->size - PAGE_SIZE)
 662                         continue;
 663                 while (addr < vaddr) {
 664                         if (count == 0)
 665                                 goto finished;
 666                         buf++;
 667                         addr++;
 668                         count--;
 669                 }
 670                 n = vaddr + tmp->size - PAGE_SIZE - addr;
 671                 do {
 672                         if (count == 0)
 673                                 goto finished;
 674                         *addr = *buf;
 675                         buf++;
 676                         addr++;
 677                         count--;
 678                 } while (--n > 0);
 679         }
 680 finished:
 681         read_unlock(&vmlist_lock);
 682         return buf - buf_start;
 683 }
 684
 685 /**
 686  *      remap_vmalloc_range  -  map vmalloc pages to userspace
 687  *      @vma:           vma to cover (map full range of vma)
 688  *      @addr:          vmalloc memory
 689  *      @pgoff:         number of pages into addr before first page to map
 690  *      @returns:       0 for success, -Exxx on failure
 691  *
 692  *      This function checks that addr is a valid vmalloc'ed area, and
 693  *      that it is big enough to cover the vma. Will return failure if
 694  *      that criteria isn't met.
 695  *
 696  *      Similar to remap_pfn_range (see mm/memory.c)
 697  */
 698 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
 699                                                 unsigned long pgoff)
 700 {
 701         struct vm_struct *area;
 702         unsigned long uaddr = vma->vm_start;
 703         unsigned long usize = vma->vm_end - vma->vm_start;
 704         int ret;
 705
 706         if ((PAGE_SIZE-1) & (unsigned long)addr)
 707                 return -EINVAL;
 708
 709         read_lock(&vmlist_lock);
 710         area = __find_vm_area(addr);
 711         if (!area)
 712                 goto out_einval_locked;
 713
 714         if (!(area->flags & VM_USERMAP))
 715                 goto out_einval_locked;
 716
 717         if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
 718                 goto out_einval_locked;
 719         read_unlock(&vmlist_lock);
 720
 721         addr += pgoff << PAGE_SHIFT;
 722         do {
 723                 struct page *page = vmalloc_to_page(addr);
 724                 ret = vm_insert_page(vma, uaddr, page);
 725                 if (ret)
 726                         return ret;
 727
 728                 uaddr += PAGE_SIZE;
 729                 addr += PAGE_SIZE;
 730                 usize -= PAGE_SIZE;
 731         } while (usize > 0);
 732
 733         /* Prevent "things" like memory migration? VM_flags need a cleanup... */
 734         vma->vm_flags |= VM_RESERVED;
 735
 736         return ret;
 737
 738 out_einval_locked:
 739         read_unlock(&vmlist_lock);
 740         return -EINVAL;
 741 }
 742 EXPORT_SYMBOL(remap_vmalloc_range);
 743