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 static struct vm_struct *__get_vm_area_node(unsigned long size, unsigned long flags,
 164                                             unsigned long start, unsigned long end,
 165                                             int node, gfp_t gfp_mask)
 166 {
 167         struct vm_struct **p, *tmp, *area;
 168         unsigned long align = 1;
 169         unsigned long addr;
 170
 171         BUG_ON(in_interrupt());
 172         if (flags & VM_IOREMAP) {
 173                 int bit = fls(size);
 174
 175                 if (bit > IOREMAP_MAX_ORDER)
 176                         bit = IOREMAP_MAX_ORDER;
 177                 else if (bit < PAGE_SHIFT)
 178                         bit = PAGE_SHIFT;
 179
 180                 align = 1ul << bit;
 181         }
 182         addr = ALIGN(start, align);
 183         size = PAGE_ALIGN(size);
 184
 185         area = kmalloc_node(sizeof(*area), gfp_mask, node);
 186         if (unlikely(!area))
 187                 return NULL;
 188
 189         if (unlikely(!size)) {
 190                 kfree (area);
 191                 return NULL;
 192         }
 193
 194         /*
 195          * We always allocate a guard page.
 196          */
 197         size += PAGE_SIZE;
 198
 199         write_lock(&vmlist_lock);
 200         for (p = &vmlist; (tmp = *p) != NULL ;p = &tmp->next) {
 201                 if ((unsigned long)tmp->addr < addr) {
 202                         if((unsigned long)tmp->addr + tmp->size >= addr)
 203                                 addr = ALIGN(tmp->size +
 204                                              (unsigned long)tmp->addr, align);
 205                         continue;
 206                 }
 207                 if ((size + addr) < addr)
 208                         goto out;
 209                 if (size + addr <= (unsigned long)tmp->addr)
 210                         goto found;
 211                 addr = ALIGN(tmp->size + (unsigned long)tmp->addr, align);
 212                 if (addr > end - size)
 213                         goto out;
 214         }
 215
 216 found:
 217         area->next = *p;
 218         *p = area;
 219
 220         area->flags = flags;
 221         area->addr = (void *)addr;
 222         area->size = size;
 223         area->pages = NULL;
 224         area->nr_pages = 0;
 225         area->phys_addr = 0;
 226         write_unlock(&vmlist_lock);
 227
 228         return area;
 229
 230 out:
 231         write_unlock(&vmlist_lock);
 232         kfree(area);
 233         if (printk_ratelimit())
 234                 printk(KERN_WARNING "allocation failed: out of vmalloc space - use vmalloc=<size> to increase size.\n");
 235         return NULL;
 236 }
 237
 238 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
 239                                 unsigned long start, unsigned long end)
 240 {
 241         return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL);
 242 }
 243
 244 /**
 245  *      get_vm_area  -  reserve a contingous kernel virtual area
 246  *      @size:          size of the area
 247  *      @flags:         %VM_IOREMAP for I/O mappings or VM_ALLOC
 248  *
 249  *      Search an area of @size in the kernel virtual mapping area,
 250  *      and reserved it for out purposes.  Returns the area descriptor
 251  *      on success or %NULL on failure.
 252  */
 253 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
 254 {
 255         return __get_vm_area(size, flags, VMALLOC_START, VMALLOC_END);
 256 }
 257
 258 struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags,
 259                                    int node, gfp_t gfp_mask)
 260 {
 261         return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node,
 262                                   gfp_mask);
 263 }
 264
 265 /* Caller must hold vmlist_lock */
 266 static struct vm_struct *__find_vm_area(void *addr)
 267 {
 268         struct vm_struct *tmp;
 269
 270         for (tmp = vmlist; tmp != NULL; tmp = tmp->next) {
 271                  if (tmp->addr == addr)
 272                         break;
 273         }
 274
 275         return tmp;
 276 }
 277
 278 /* Caller must hold vmlist_lock */
 279 static struct vm_struct *__remove_vm_area(void *addr)
 280 {
 281         struct vm_struct **p, *tmp;
 282
 283         for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) {
 284                  if (tmp->addr == addr)
 285                          goto found;
 286         }
 287         return NULL;
 288
 289 found:
 290         unmap_vm_area(tmp);
 291         *p = tmp->next;
 292
 293         /*
 294          * Remove the guard page.
 295          */
 296         tmp->size -= PAGE_SIZE;
 297         return tmp;
 298 }
 299
 300 /**
 301  *      remove_vm_area  -  find and remove a contingous kernel virtual area
 302  *      @addr:          base address
 303  *
 304  *      Search for the kernel VM area starting at @addr, and remove it.
 305  *      This function returns the found VM area, but using it is NOT safe
 306  *      on SMP machines, except for its size or flags.
 307  */
 308 struct vm_struct *remove_vm_area(void *addr)
 309 {
 310         struct vm_struct *v;
 311         write_lock(&vmlist_lock);
 312         v = __remove_vm_area(addr);
 313         write_unlock(&vmlist_lock);
 314         return v;
 315 }
 316
 317 void __vunmap(void *addr, int deallocate_pages)
 318 {
 319         struct vm_struct *area;
 320
 321         if (!addr)
 322                 return;
 323
 324         if ((PAGE_SIZE-1) & (unsigned long)addr) {
 325                 printk(KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
 326                 WARN_ON(1);
 327                 return;
 328         }
 329
 330         area = remove_vm_area(addr);
 331         if (unlikely(!area)) {
 332                 printk(KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
 333                                 addr);
 334                 WARN_ON(1);
 335                 return;
 336         }
 337
 338         debug_check_no_locks_freed(addr, area->size);
 339
 340         if (deallocate_pages) {
 341                 int i;
 342
 343                 for (i = 0; i < area->nr_pages; i++) {
 344                         BUG_ON(!area->pages[i]);
 345                         __free_page(area->pages[i]);
 346                 }
 347
 348                 if (area->flags & VM_VPAGES)
 349                         vfree(area->pages);
 350                 else
 351                         kfree(area->pages);
 352         }
 353
 354         kfree(area);
 355         return;
 356 }
 357
 358 /**
 359  *      vfree  -  release memory allocated by vmalloc()
 360  *      @addr:          memory base address
 361  *
 362  *      Free the virtually contiguous memory area starting at @addr, as
 363  *      obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
 364  *      NULL, no operation is performed.
 365  *
 366  *      Must not be called in interrupt context.
 367  */
 368 void vfree(void *addr)
 369 {
 370         BUG_ON(in_interrupt());
 371         __vunmap(addr, 1);
 372 }
 373 EXPORT_SYMBOL(vfree);
 374
 375 /**
 376  *      vunmap  -  release virtual mapping obtained by vmap()
 377  *      @addr:          memory base address
 378  *
 379  *      Free the virtually contiguous memory area starting at @addr,
 380  *      which was created from the page array passed to vmap().
 381  *
 382  *      Must not be called in interrupt context.
 383  */
 384 void vunmap(void *addr)
 385 {
 386         BUG_ON(in_interrupt());
 387         __vunmap(addr, 0);
 388 }
 389 EXPORT_SYMBOL(vunmap);
 390
 391 /**
 392  *      vmap  -  map an array of pages into virtually contiguous space
 393  *      @pages:         array of page pointers
 394  *      @count:         number of pages to map
 395  *      @flags:         vm_area->flags
 396  *      @prot:          page protection for the mapping
 397  *
 398  *      Maps @count pages from @pages into contiguous kernel virtual
 399  *      space.
 400  */
 401 void *vmap(struct page **pages, unsigned int count,
 402                 unsigned long flags, pgprot_t prot)
 403 {
 404         struct vm_struct *area;
 405
 406         if (count > num_physpages)
 407                 return NULL;
 408
 409         area = get_vm_area((count << PAGE_SHIFT), flags);
 410         if (!area)
 411                 return NULL;
 412         if (map_vm_area(area, prot, &pages)) {
 413                 vunmap(area->addr);
 414                 return NULL;
 415         }
 416
 417         return area->addr;
 418 }
 419 EXPORT_SYMBOL(vmap);
 420
 421 void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
 422                                 pgprot_t prot, int node)
 423 {
 424         struct page **pages;
 425         unsigned int nr_pages, array_size, i;
 426
 427         nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
 428         array_size = (nr_pages * sizeof(struct page *));
 429
 430         area->nr_pages = nr_pages;
 431         /* Please note that the recursion is strictly bounded. */
 432         if (array_size > PAGE_SIZE) {
 433                 pages = __vmalloc_node(array_size, gfp_mask, PAGE_KERNEL, node);
 434                 area->flags |= VM_VPAGES;
 435         } else {
 436                 pages = kmalloc_node(array_size,
 437                                 (gfp_mask & ~(__GFP_HIGHMEM | __GFP_ZERO)),
 438                                 node);
 439         }
 440         area->pages = pages;
 441         if (!area->pages) {
 442                 remove_vm_area(area->addr);
 443                 kfree(area);
 444                 return NULL;
 445         }
 446         memset(area->pages, 0, array_size);
 447
 448         for (i = 0; i < area->nr_pages; i++) {
 449                 if (node < 0)
 450                         area->pages[i] = alloc_page(gfp_mask);
 451                 else
 452                         area->pages[i] = alloc_pages_node(node, gfp_mask, 0);
 453                 if (unlikely(!area->pages[i])) {
 454                         /* Successfully allocated i pages, free them in __vunmap() */
 455                         area->nr_pages = i;
 456                         goto fail;
 457                 }
 458         }
 459
 460         if (map_vm_area(area, prot, &pages))
 461                 goto fail;
 462         return area->addr;
 463
 464 fail:
 465         vfree(area->addr);
 466         return NULL;
 467 }
 468
 469 void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
 470 {
 471         return __vmalloc_area_node(area, gfp_mask, prot, -1);
 472 }
 473
 474 /**
 475  *      __vmalloc_node  -  allocate virtually contiguous memory
 476  *      @size:          allocation size
 477  *      @gfp_mask:      flags for the page level allocator
 478  *      @prot:          protection mask for the allocated pages
 479  *      @node:          node to use for allocation or -1
 480  *
 481  *      Allocate enough pages to cover @size from the page level
 482  *      allocator with @gfp_mask flags.  Map them into contiguous
 483  *      kernel virtual space, using a pagetable protection of @prot.
 484  */
 485 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
 486                             int node)
 487 {
 488         struct vm_struct *area;
 489
 490         size = PAGE_ALIGN(size);
 491         if (!size || (size >> PAGE_SHIFT) > num_physpages)
 492                 return NULL;
 493
 494         area = get_vm_area_node(size, VM_ALLOC, node, gfp_mask);
 495         if (!area)
 496                 return NULL;
 497
 498         return __vmalloc_area_node(area, gfp_mask, prot, node);
 499 }
 500
 501 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
 502 {
 503         return __vmalloc_node(size, gfp_mask, prot, -1);
 504 }
 505 EXPORT_SYMBOL(__vmalloc);
 506
 507 /**
 508  *      vmalloc  -  allocate virtually contiguous memory
 509  *      @size:          allocation size
 510  *      Allocate enough pages to cover @size from the page level
 511  *      allocator and map them into contiguous kernel virtual space.
 512  *
 513  *      For tight control over page level allocator and protection flags
 514  *      use __vmalloc() instead.
 515  */
 516 void *vmalloc(unsigned long size)
 517 {
 518         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
 519 }
 520 EXPORT_SYMBOL(vmalloc);
 521
 522 /**
 523  * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
 524  * @size: allocation size
 525  *
 526  * The resulting memory area is zeroed so it can be mapped to userspace
 527  * without leaking data.
 528  */
 529 void *vmalloc_user(unsigned long size)
 530 {
 531         struct vm_struct *area;
 532         void *ret;
 533
 534         ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL);
 535         write_lock(&vmlist_lock);
 536         area = __find_vm_area(ret);
 537         area->flags |= VM_USERMAP;
 538         write_unlock(&vmlist_lock);
 539
 540         return ret;
 541 }
 542 EXPORT_SYMBOL(vmalloc_user);
 543
 544 /**
 545  *      vmalloc_node  -  allocate memory on a specific node
 546  *      @size:          allocation size
 547  *      @node:          numa node
 548  *
 549  *      Allocate enough pages to cover @size from the page level
 550  *      allocator and map them into contiguous kernel virtual space.
 551  *
 552  *      For tight control over page level allocator and protection flags
 553  *      use __vmalloc() instead.
 554  */
 555 void *vmalloc_node(unsigned long size, int node)
 556 {
 557         return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, node);
 558 }
 559 EXPORT_SYMBOL(vmalloc_node);
 560
 561 #ifndef PAGE_KERNEL_EXEC
 562 # define PAGE_KERNEL_EXEC PAGE_KERNEL
 563 #endif
 564
 565 /**
 566  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
 567  *      @size:          allocation size
 568  *
 569  *      Kernel-internal function to allocate enough pages to cover @size
 570  *      the page level allocator and map them into contiguous and
 571  *      executable kernel virtual space.
 572  *
 573  *      For tight control over page level allocator and protection flags
 574  *      use __vmalloc() instead.
 575  */
 576
 577 void *vmalloc_exec(unsigned long size)
 578 {
 579         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
 580 }
 581
 582 /**
 583  *      vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
 584  *      @size:          allocation size
 585  *
 586  *      Allocate enough 32bit PA addressable pages to cover @size from the
 587  *      page level allocator and map them into contiguous kernel virtual space.
 588  */
 589 void *vmalloc_32(unsigned long size)
 590 {
 591         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
 592 }
 593 EXPORT_SYMBOL(vmalloc_32);
 594
 595 /**
 596  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
 597  *      @size:          allocation size
 598  *
 599  * The resulting memory area is 32bit addressable and zeroed so it can be
 600  * mapped to userspace without leaking data.
 601  */
 602 void *vmalloc_32_user(unsigned long size)
 603 {
 604         struct vm_struct *area;
 605         void *ret;
 606
 607         ret = __vmalloc(size, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
 608         write_lock(&vmlist_lock);
 609         area = __find_vm_area(ret);
 610         area->flags |= VM_USERMAP;
 611         write_unlock(&vmlist_lock);
 612
 613         return ret;
 614 }
 615 EXPORT_SYMBOL(vmalloc_32_user);
 616
 617 long vread(char *buf, char *addr, unsigned long count)
 618 {
 619         struct vm_struct *tmp;
 620         char *vaddr, *buf_start = buf;
 621         unsigned long n;
 622
 623         /* Don't allow overflow */
 624         if ((unsigned long) addr + count < count)
 625                 count = -(unsigned long) addr;
 626
 627         read_lock(&vmlist_lock);
 628         for (tmp = vmlist; tmp; tmp = tmp->next) {
 629                 vaddr = (char *) tmp->addr;
 630                 if (addr >= vaddr + tmp->size - PAGE_SIZE)
 631                         continue;
 632                 while (addr < vaddr) {
 633                         if (count == 0)
 634                                 goto finished;
 635                         *buf = '\0';
 636                         buf++;
 637                         addr++;
 638                         count--;
 639                 }
 640                 n = vaddr + tmp->size - PAGE_SIZE - addr;
 641                 do {
 642                         if (count == 0)
 643                                 goto finished;
 644                         *buf = *addr;
 645                         buf++;
 646                         addr++;
 647                         count--;
 648                 } while (--n > 0);
 649         }
 650 finished:
 651         read_unlock(&vmlist_lock);
 652         return buf - buf_start;
 653 }
 654
 655 long vwrite(char *buf, char *addr, unsigned long count)
 656 {
 657         struct vm_struct *tmp;
 658         char *vaddr, *buf_start = buf;
 659         unsigned long n;
 660
 661         /* Don't allow overflow */
 662         if ((unsigned long) addr + count < count)
 663                 count = -(unsigned long) addr;
 664
 665         read_lock(&vmlist_lock);
 666         for (tmp = vmlist; tmp; tmp = tmp->next) {
 667                 vaddr = (char *) tmp->addr;
 668                 if (addr >= vaddr + tmp->size - PAGE_SIZE)
 669                         continue;
 670                 while (addr < vaddr) {
 671                         if (count == 0)
 672                                 goto finished;
 673                         buf++;
 674                         addr++;
 675                         count--;
 676                 }
 677                 n = vaddr + tmp->size - PAGE_SIZE - addr;
 678                 do {
 679                         if (count == 0)
 680                                 goto finished;
 681                         *addr = *buf;
 682                         buf++;
 683                         addr++;
 684                         count--;
 685                 } while (--n > 0);
 686         }
 687 finished:
 688         read_unlock(&vmlist_lock);
 689         return buf - buf_start;
 690 }
 691
 692 /**
 693  *      remap_vmalloc_range  -  map vmalloc pages to userspace
 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