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 vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
  28 {
  29         pte_t *pte;
  30
  31         pte = pte_offset_kernel(pmd, addr);
  32         do {
  33                 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
  34                 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
  35         } while (pte++, addr += PAGE_SIZE, addr != end);
  36 }
  37
  38 static inline void vunmap_pmd_range(pud_t *pud, unsigned long addr,
  39                                                 unsigned long end)
  40 {
  41         pmd_t *pmd;
  42         unsigned long next;
  43
  44         pmd = pmd_offset(pud, addr);
  45         do {
  46                 next = pmd_addr_end(addr, end);
  47                 if (pmd_none_or_clear_bad(pmd))
  48                         continue;
  49                 vunmap_pte_range(pmd, addr, next);
  50         } while (pmd++, addr = next, addr != end);
  51 }
  52
  53 static inline void vunmap_pud_range(pgd_t *pgd, unsigned long addr,
  54                                                 unsigned long end)
  55 {
  56         pud_t *pud;
  57         unsigned long next;
  58
  59         pud = pud_offset(pgd, addr);
  60         do {
  61                 next = pud_addr_end(addr, end);
  62                 if (pud_none_or_clear_bad(pud))
  63                         continue;
  64                 vunmap_pmd_range(pud, addr, next);
  65         } while (pud++, addr = next, addr != end);
  66 }
  67
  68 void unmap_vm_area(struct vm_struct *area)
  69 {
  70         pgd_t *pgd;
  71         unsigned long next;
  72         unsigned long addr = (unsigned long) area->addr;
  73         unsigned long end = addr + area->size;
  74
  75         BUG_ON(addr >= end);
  76         pgd = pgd_offset_k(addr);
  77         flush_cache_vunmap(addr, end);
  78         do {
  79                 next = pgd_addr_end(addr, end);
  80                 if (pgd_none_or_clear_bad(pgd))
  81                         continue;
  82                 vunmap_pud_range(pgd, addr, next);
  83         } while (pgd++, addr = next, addr != end);
  84         flush_tlb_kernel_range((unsigned long) area->addr, end);
  85 }
  86
  87 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
  88                         unsigned long end, pgprot_t prot, struct page ***pages)
  89 {
  90         pte_t *pte;
  91
  92         pte = pte_alloc_kernel(pmd, addr);
  93         if (!pte)
  94                 return -ENOMEM;
  95         do {
  96                 struct page *page = **pages;
  97                 WARN_ON(!pte_none(*pte));
  98                 if (!page)
  99                         return -ENOMEM;
 100                 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
 101                 (*pages)++;
 102         } while (pte++, addr += PAGE_SIZE, addr != end);
 103         return 0;
 104 }
 105
 106 static inline int vmap_pmd_range(pud_t *pud, unsigned long addr,
 107                         unsigned long end, pgprot_t prot, struct page ***pages)
 108 {
 109         pmd_t *pmd;
 110         unsigned long next;
 111
 112         pmd = pmd_alloc(&init_mm, pud, addr);
 113         if (!pmd)
 114                 return -ENOMEM;
 115         do {
 116                 next = pmd_addr_end(addr, end);
 117                 if (vmap_pte_range(pmd, addr, next, prot, pages))
 118                         return -ENOMEM;
 119         } while (pmd++, addr = next, addr != end);
 120         return 0;
 121 }
 122
 123 static inline int vmap_pud_range(pgd_t *pgd, unsigned long addr,
 124                         unsigned long end, pgprot_t prot, struct page ***pages)
 125 {
 126         pud_t *pud;
 127         unsigned long next;
 128
 129         pud = pud_alloc(&init_mm, pgd, addr);
 130         if (!pud)
 131                 return -ENOMEM;
 132         do {
 133                 next = pud_addr_end(addr, end);
 134                 if (vmap_pmd_range(pud, addr, next, prot, pages))
 135                         return -ENOMEM;
 136         } while (pud++, addr = next, addr != end);
 137         return 0;
 138 }
 139
 140 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
 141 {
 142         pgd_t *pgd;
 143         unsigned long next;
 144         unsigned long addr = (unsigned long) area->addr;
 145         unsigned long end = addr + area->size - PAGE_SIZE;
 146         int err;
 147
 148         BUG_ON(addr >= end);
 149         pgd = pgd_offset_k(addr);
 150         do {
 151                 next = pgd_addr_end(addr, end);
 152                 err = vmap_pud_range(pgd, addr, next, prot, pages);
 153                 if (err)
 154                         break;
 155         } while (pgd++, addr = next, addr != end);
 156         flush_cache_vmap((unsigned long) area->addr, end);
 157         return err;
 158 }
 159
 160 struct vm_struct *__get_vm_area_node(unsigned long size, unsigned long flags,
 161                                 unsigned long start, unsigned long end, int node)
 162 {
 163         struct vm_struct **p, *tmp, *area;
 164         unsigned long align = 1;
 165         unsigned long addr;
 166
 167         if (flags & VM_IOREMAP) {
 168                 int bit = fls(size);
 169
 170                 if (bit > IOREMAP_MAX_ORDER)
 171                         bit = IOREMAP_MAX_ORDER;
 172                 else if (bit < PAGE_SHIFT)
 173                         bit = PAGE_SHIFT;
 174
 175                 align = 1ul << bit;
 176         }
 177         addr = ALIGN(start, align);
 178         size = PAGE_ALIGN(size);
 179
 180         area = kmalloc_node(sizeof(*area), GFP_KERNEL, node);
 181         if (unlikely(!area))
 182                 return NULL;
 183
 184         if (unlikely(!size)) {
 185                 kfree (area);
 186                 return NULL;
 187         }
 188
 189         /*
 190          * We always allocate a guard page.
 191          */
 192         size += PAGE_SIZE;
 193
 194         write_lock(&vmlist_lock);
 195         for (p = &vmlist; (tmp = *p) != NULL ;p = &tmp->next) {
 196                 if ((unsigned long)tmp->addr < addr) {
 197                         if((unsigned long)tmp->addr + tmp->size >= addr)
 198                                 addr = ALIGN(tmp->size +
 199                                              (unsigned long)tmp->addr, align);
 200                         continue;
 201                 }
 202                 if ((size + addr) < addr)
 203                         goto out;
 204                 if (size + addr <= (unsigned long)tmp->addr)
 205                         goto found;
 206                 addr = ALIGN(tmp->size + (unsigned long)tmp->addr, align);
 207                 if (addr > end - size)
 208                         goto out;
 209         }
 210
 211 found:
 212         area->next = *p;
 213         *p = area;
 214
 215         area->flags = flags;
 216         area->addr = (void *)addr;
 217         area->size = size;
 218         area->pages = NULL;
 219         area->nr_pages = 0;
 220         area->phys_addr = 0;
 221         write_unlock(&vmlist_lock);
 222
 223         return area;
 224
 225 out:
 226         write_unlock(&vmlist_lock);
 227         kfree(area);
 228         if (printk_ratelimit())
 229                 printk(KERN_WARNING "allocation failed: out of vmalloc space - use vmalloc=<size> to increase size.\n");
 230         return NULL;
 231 }
 232
 233 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
 234                                 unsigned long start, unsigned long end)
 235 {
 236         return __get_vm_area_node(size, flags, start, end, -1);
 237 }
 238
 239 /**
 240  *      get_vm_area  -  reserve a contingous kernel virtual area
 241  *
 242  *      @size:          size of the area
 243  *      @flags:         %VM_IOREMAP for I/O mappings or VM_ALLOC
 244  *
 245  *      Search an area of @size in the kernel virtual mapping area,
 246  *      and reserved it for out purposes.  Returns the area descriptor
 247  *      on success or %NULL on failure.
 248  */
 249 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
 250 {
 251         return __get_vm_area(size, flags, VMALLOC_START, VMALLOC_END);
 252 }
 253
 254 struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags, int node)
 255 {
 256         return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node);
 257 }
 258
 259 /* Caller must hold vmlist_lock */
 260 static struct vm_struct *__find_vm_area(void *addr)
 261 {
 262         struct vm_struct *tmp;
 263
 264         for (tmp = vmlist; tmp != NULL; tmp = tmp->next) {
 265                  if (tmp->addr == addr)
 266                         break;
 267         }
 268
 269         return tmp;
 270 }
 271
 272 /* Caller must hold vmlist_lock */
 273 struct vm_struct *__remove_vm_area(void *addr)
 274 {
 275         struct vm_struct **p, *tmp;
 276
 277         for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) {
 278                  if (tmp->addr == addr)
 279                          goto found;
 280         }
 281         return NULL;
 282
 283 found:
 284         unmap_vm_area(tmp);
 285         *p = tmp->next;
 286
 287         /*
 288          * Remove the guard page.
 289          */
 290         tmp->size -= PAGE_SIZE;
 291         return tmp;
 292 }
 293
 294 /**
 295  *      remove_vm_area  -  find and remove a contingous kernel virtual area
 296  *
 297  *      @addr:          base address
 298  *
 299  *      Search for the kernel VM area starting at @addr, and remove it.
 300  *      This function returns the found VM area, but using it is NOT safe
 301  *      on SMP machines, except for its size or flags.
 302  */
 303 struct vm_struct *remove_vm_area(void *addr)
 304 {
 305         struct vm_struct *v;
 306         write_lock(&vmlist_lock);
 307         v = __remove_vm_area(addr);
 308         write_unlock(&vmlist_lock);
 309         return v;
 310 }
 311
 312 void __vunmap(void *addr, int deallocate_pages)
 313 {
 314         struct vm_struct *area;
 315
 316         if (!addr)
 317                 return;
 318
 319         if ((PAGE_SIZE-1) & (unsigned long)addr) {
 320                 printk(KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
 321                 WARN_ON(1);
 322                 return;
 323         }
 324
 325         area = remove_vm_area(addr);
 326         if (unlikely(!area)) {
 327                 printk(KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
 328                                 addr);
 329                 WARN_ON(1);
 330                 return;
 331         }
 332
 333         if (deallocate_pages) {
 334                 int i;
 335
 336                 for (i = 0; i < area->nr_pages; i++) {
 337                         BUG_ON(!area->pages[i]);
 338                         __free_page(area->pages[i]);
 339                 }
 340
 341                 if (area->nr_pages > PAGE_SIZE/sizeof(struct page *))
 342                         vfree(area->pages);
 343                 else
 344                         kfree(area->pages);
 345         }
 346
 347         kfree(area);
 348         return;
 349 }
 350
 351 /**
 352  *      vfree  -  release memory allocated by vmalloc()
 353  *
 354  *      @addr:          memory base address
 355  *
 356  *      Free the virtually contiguous memory area starting at @addr, as
 357  *      obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
 358  *      NULL, no operation is performed.
 359  *
 360  *      Must not be called in interrupt context.
 361  */
 362 void vfree(void *addr)
 363 {
 364         BUG_ON(in_interrupt());
 365         __vunmap(addr, 1);
 366 }
 367 EXPORT_SYMBOL(vfree);
 368
 369 /**
 370  *      vunmap  -  release virtual mapping obtained by vmap()
 371  *
 372  *      @addr:          memory base address
 373  *
 374  *      Free the virtually contiguous memory area starting at @addr,
 375  *      which was created from the page array passed to vmap().
 376  *
 377  *      Must not be called in interrupt context.
 378  */
 379 void vunmap(void *addr)
 380 {
 381         BUG_ON(in_interrupt());
 382         __vunmap(addr, 0);
 383 }
 384 EXPORT_SYMBOL(vunmap);
 385
 386 /**
 387  *      vmap  -  map an array of pages into virtually contiguous space
 388  *
 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         else
 431                 pages = kmalloc_node(array_size, (gfp_mask & ~__GFP_HIGHMEM), node);
 432         area->pages = pages;
 433         if (!area->pages) {
 434                 remove_vm_area(area->addr);
 435                 kfree(area);
 436                 return NULL;
 437         }
 438         memset(area->pages, 0, array_size);
 439
 440         for (i = 0; i < area->nr_pages; i++) {
 441                 if (node < 0)
 442                         area->pages[i] = alloc_page(gfp_mask);
 443                 else
 444                         area->pages[i] = alloc_pages_node(node, gfp_mask, 0);
 445                 if (unlikely(!area->pages[i])) {
 446                         /* Successfully allocated i pages, free them in __vunmap() */
 447                         area->nr_pages = i;
 448                         goto fail;
 449                 }
 450         }
 451
 452         if (map_vm_area(area, prot, &pages))
 453                 goto fail;
 454         return area->addr;
 455
 456 fail:
 457         vfree(area->addr);
 458         return NULL;
 459 }
 460
 461 void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
 462 {
 463         return __vmalloc_area_node(area, gfp_mask, prot, -1);
 464 }
 465
 466 /**
 467  *      __vmalloc_node  -  allocate virtually contiguous memory
 468  *
 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 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 EXPORT_SYMBOL(__vmalloc_node);
 494
 495 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
 496 {
 497         return __vmalloc_node(size, gfp_mask, prot, -1);
 498 }
 499 EXPORT_SYMBOL(__vmalloc);
 500
 501 /**
 502  *      vmalloc  -  allocate virtually contiguous memory
 503  *
 504  *      @size:          allocation size
 505  *
 506  *      Allocate enough pages to cover @size from the page level
 507  *      allocator and map them into contiguous kernel virtual space.
 508  *
 509  *      For tight cotrol over page level allocator and protection flags
 510  *      use __vmalloc() instead.
 511  */
 512 void *vmalloc(unsigned long size)
 513 {
 514         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
 515 }
 516 EXPORT_SYMBOL(vmalloc);
 517
 518 /**
 519  *      vmalloc_user  -  allocate virtually contiguous memory which has
 520  *                         been zeroed so it can be mapped to userspace without
 521  *                         leaking data.
 522  *
 523  *      @size:          allocation size
 524  */
 525 void *vmalloc_user(unsigned long size)
 526 {
 527         struct vm_struct *area;
 528         void *ret;
 529
 530         ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL);
 531         write_lock(&vmlist_lock);
 532         area = __find_vm_area(ret);
 533         area->flags |= VM_USERMAP;
 534         write_unlock(&vmlist_lock);
 535
 536         return ret;
 537 }
 538 EXPORT_SYMBOL(vmalloc_user);
 539
 540 /**
 541  *      vmalloc_node  -  allocate memory on a specific node
 542  *
 543  *      @size:          allocation size
 544  *      @node:          numa node
 545  *
 546  *      Allocate enough pages to cover @size from the page level
 547  *      allocator and map them into contiguous kernel virtual space.
 548  *
 549  *      For tight cotrol over page level allocator and protection flags
 550  *      use __vmalloc() instead.
 551  */
 552 void *vmalloc_node(unsigned long size, int node)
 553 {
 554         return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, node);
 555 }
 556 EXPORT_SYMBOL(vmalloc_node);
 557
 558 #ifndef PAGE_KERNEL_EXEC
 559 # define PAGE_KERNEL_EXEC PAGE_KERNEL
 560 #endif
 561
 562 /**
 563  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
 564  *
 565  *      @size:          allocation size
 566  *
 567  *      Kernel-internal function to allocate enough pages to cover @size
 568  *      the page level allocator and map them into contiguous and
 569  *      executable kernel virtual space.
 570  *
 571  *      For tight cotrol over page level allocator and protection flags
 572  *      use __vmalloc() instead.
 573  */
 574
 575 void *vmalloc_exec(unsigned long size)
 576 {
 577         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
 578 }
 579
 580 /**
 581  *      vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
 582  *
 583  *      @size:          allocation size
 584  *
 585  *      Allocate enough 32bit PA addressable pages to cover @size from the
 586  *      page level allocator and map them into contiguous kernel virtual space.
 587  */
 588 void *vmalloc_32(unsigned long size)
 589 {
 590         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
 591 }
 592 EXPORT_SYMBOL(vmalloc_32);
 593
 594 /**
 595  *      vmalloc_32_user  -  allocate virtually contiguous memory (32bit
 596  *                            addressable) which is zeroed so it can be
 597  *                            mapped to userspace without leaking data.
 598  *
 599  *      @size:          allocation size
 600  */
 601 void *vmalloc_32_user(unsigned long size)
 602 {
 603         struct vm_struct *area;
 604         void *ret;
 605
 606         ret = __vmalloc(size, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
 607         write_lock(&vmlist_lock);
 608         area = __find_vm_area(ret);
 609         area->flags |= VM_USERMAP;
 610         write_unlock(&vmlist_lock);
 611
 612         return ret;
 613 }
 614 EXPORT_SYMBOL(vmalloc_32_user);
 615
 616 long vread(char *buf, char *addr, unsigned long count)
 617 {
 618         struct vm_struct *tmp;
 619         char *vaddr, *buf_start = buf;
 620         unsigned long n;
 621
 622         /* Don't allow overflow */
 623         if ((unsigned long) addr + count < count)
 624                 count = -(unsigned long) addr;
 625
 626         read_lock(&vmlist_lock);
 627         for (tmp = vmlist; tmp; tmp = tmp->next) {
 628                 vaddr = (char *) tmp->addr;
 629                 if (addr >= vaddr + tmp->size - PAGE_SIZE)
 630                         continue;
 631                 while (addr < vaddr) {
 632                         if (count == 0)
 633                                 goto finished;
 634                         *buf = '\0';
 635                         buf++;
 636                         addr++;
 637                         count--;
 638                 }
 639                 n = vaddr + tmp->size - PAGE_SIZE - addr;
 640                 do {
 641                         if (count == 0)
 642                                 goto finished;
 643                         *buf = *addr;
 644                         buf++;
 645                         addr++;
 646                         count--;
 647                 } while (--n > 0);
 648         }
 649 finished:
 650         read_unlock(&vmlist_lock);
 651         return buf - buf_start;
 652 }
 653
 654 long vwrite(char *buf, char *addr, unsigned long count)
 655 {
 656         struct vm_struct *tmp;
 657         char *vaddr, *buf_start = buf;
 658         unsigned long n;
 659
 660         /* Don't allow overflow */
 661         if ((unsigned long) addr + count < count)
 662                 count = -(unsigned long) addr;
 663
 664         read_lock(&vmlist_lock);
 665         for (tmp = vmlist; tmp; tmp = tmp->next) {
 666                 vaddr = (char *) tmp->addr;
 667                 if (addr >= vaddr + tmp->size - PAGE_SIZE)
 668                         continue;
 669                 while (addr < vaddr) {
 670                         if (count == 0)
 671                                 goto finished;
 672                         buf++;
 673                         addr++;
 674                         count--;
 675                 }
 676                 n = vaddr + tmp->size - PAGE_SIZE - addr;
 677                 do {
 678                         if (count == 0)
 679                                 goto finished;
 680                         *addr = *buf;
 681                         buf++;
 682                         addr++;
 683                         count--;
 684                 } while (--n > 0);
 685         }
 686 finished:
 687         read_unlock(&vmlist_lock);
 688         return buf - buf_start;
 689 }
 690
 691 /**
 692  *      remap_vmalloc_range  -  map vmalloc pages to userspace
 693  *
 694  *      @vma:           vma to cover (map full range of vma)
 695  *      @addr:          vmalloc memory
 696  *      @pgoff:         number of pages into addr before first page to map
 697  *      @returns:       0 for success, -Exxx on failure
 698  *
 699  *      This function checks that addr is a valid vmalloc'ed area, and
 700  *      that it is big enough to cover the vma. Will return failure if
 701  *      that criteria isn't met.
 702  *
 703  *      Similar to remap_pfn_range (see mm/memory.c)
 704  */
 705 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
 706                                                 unsigned long pgoff)
 707 {
 708         struct vm_struct *area;
 709         unsigned long uaddr = vma->vm_start;
 710         unsigned long usize = vma->vm_end - vma->vm_start;
 711         int ret;
 712
 713         if ((PAGE_SIZE-1) & (unsigned long)addr)
 714                 return -EINVAL;
 715
 716         read_lock(&vmlist_lock);
 717         area = __find_vm_area(addr);
 718         if (!area)
 719                 goto out_einval_locked;
 720
 721         if (!(area->flags & VM_USERMAP))
 722                 goto out_einval_locked;
 723
 724         if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
 725                 goto out_einval_locked;
 726         read_unlock(&vmlist_lock);
 727
 728         addr += pgoff << PAGE_SHIFT;
 729         do {
 730                 struct page *page = vmalloc_to_page(addr);
 731                 ret = vm_insert_page(vma, uaddr, page);
 732                 if (ret)
 733                         return ret;
 734
 735                 uaddr += PAGE_SIZE;
 736                 addr += PAGE_SIZE;
 737                 usize -= PAGE_SIZE;
 738         } while (usize > 0);
 739
 740         /* Prevent "things" like memory migration? VM_flags need a cleanup... */
 741         vma->vm_flags |= VM_RESERVED;
 742
 743         return ret;
 744
 745 out_einval_locked:
 746         read_unlock(&vmlist_lock);
 747         return -EINVAL;
 748 }
 749 EXPORT_SYMBOL(remap_vmalloc_range);
 750