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 #include <linux/seq_file.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/kallsyms.h>
  20
  21 #include <asm/uaccess.h>
  22 #include <asm/tlbflush.h>
  23
  24
  25 DEFINE_RWLOCK(vmlist_lock);
  26 struct vm_struct *vmlist;
  27
  28 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
  29                             int node, void *caller);
  30
  31 static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
  32 {
  33         pte_t *pte;
  34
  35         pte = pte_offset_kernel(pmd, addr);
  36         do {
  37                 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
  38                 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
  39         } while (pte++, addr += PAGE_SIZE, addr != end);
  40 }
  41
  42 static inline void vunmap_pmd_range(pud_t *pud, unsigned long addr,
  43                                                 unsigned long end)
  44 {
  45         pmd_t *pmd;
  46         unsigned long next;
  47
  48         pmd = pmd_offset(pud, addr);
  49         do {
  50                 next = pmd_addr_end(addr, end);
  51                 if (pmd_none_or_clear_bad(pmd))
  52                         continue;
  53                 vunmap_pte_range(pmd, addr, next);
  54         } while (pmd++, addr = next, addr != end);
  55 }
  56
  57 static inline void vunmap_pud_range(pgd_t *pgd, unsigned long addr,
  58                                                 unsigned long end)
  59 {
  60         pud_t *pud;
  61         unsigned long next;
  62
  63         pud = pud_offset(pgd, addr);
  64         do {
  65                 next = pud_addr_end(addr, end);
  66                 if (pud_none_or_clear_bad(pud))
  67                         continue;
  68                 vunmap_pmd_range(pud, addr, next);
  69         } while (pud++, addr = next, addr != end);
  70 }
  71
  72 void unmap_kernel_range(unsigned long addr, unsigned long size)
  73 {
  74         pgd_t *pgd;
  75         unsigned long next;
  76         unsigned long start = addr;
  77         unsigned long end = addr + size;
  78
  79         BUG_ON(addr >= end);
  80         pgd = pgd_offset_k(addr);
  81         flush_cache_vunmap(addr, end);
  82         do {
  83                 next = pgd_addr_end(addr, end);
  84                 if (pgd_none_or_clear_bad(pgd))
  85                         continue;
  86                 vunmap_pud_range(pgd, addr, next);
  87         } while (pgd++, addr = next, addr != end);
  88         flush_tlb_kernel_range(start, end);
  89 }
  90
  91 static void unmap_vm_area(struct vm_struct *area)
  92 {
  93         unmap_kernel_range((unsigned long)area->addr, area->size);
  94 }
  95
  96 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
  97                         unsigned long end, pgprot_t prot, struct page ***pages)
  98 {
  99         pte_t *pte;
 100
 101         pte = pte_alloc_kernel(pmd, addr);
 102         if (!pte)
 103                 return -ENOMEM;
 104         do {
 105                 struct page *page = **pages;
 106                 WARN_ON(!pte_none(*pte));
 107                 if (!page)
 108                         return -ENOMEM;
 109                 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
 110                 (*pages)++;
 111         } while (pte++, addr += PAGE_SIZE, addr != end);
 112         return 0;
 113 }
 114
 115 static inline int vmap_pmd_range(pud_t *pud, unsigned long addr,
 116                         unsigned long end, pgprot_t prot, struct page ***pages)
 117 {
 118         pmd_t *pmd;
 119         unsigned long next;
 120
 121         pmd = pmd_alloc(&init_mm, pud, addr);
 122         if (!pmd)
 123                 return -ENOMEM;
 124         do {
 125                 next = pmd_addr_end(addr, end);
 126                 if (vmap_pte_range(pmd, addr, next, prot, pages))
 127                         return -ENOMEM;
 128         } while (pmd++, addr = next, addr != end);
 129         return 0;
 130 }
 131
 132 static inline int vmap_pud_range(pgd_t *pgd, unsigned long addr,
 133                         unsigned long end, pgprot_t prot, struct page ***pages)
 134 {
 135         pud_t *pud;
 136         unsigned long next;
 137
 138         pud = pud_alloc(&init_mm, pgd, addr);
 139         if (!pud)
 140                 return -ENOMEM;
 141         do {
 142                 next = pud_addr_end(addr, end);
 143                 if (vmap_pmd_range(pud, addr, next, prot, pages))
 144                         return -ENOMEM;
 145         } while (pud++, addr = next, addr != end);
 146         return 0;
 147 }
 148
 149 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
 150 {
 151         pgd_t *pgd;
 152         unsigned long next;
 153         unsigned long addr = (unsigned long) area->addr;
 154         unsigned long end = addr + area->size - PAGE_SIZE;
 155         int err;
 156
 157         BUG_ON(addr >= end);
 158         pgd = pgd_offset_k(addr);
 159         do {
 160                 next = pgd_addr_end(addr, end);
 161                 err = vmap_pud_range(pgd, addr, next, prot, pages);
 162                 if (err)
 163                         break;
 164         } while (pgd++, addr = next, addr != end);
 165         flush_cache_vmap((unsigned long) area->addr, end);
 166         return err;
 167 }
 168 EXPORT_SYMBOL_GPL(map_vm_area);
 169
 170 /*
 171  * Map a vmalloc()-space virtual address to the physical page.
 172  */
 173 struct page *vmalloc_to_page(const void *vmalloc_addr)
 174 {
 175         unsigned long addr = (unsigned long) vmalloc_addr;
 176         struct page *page = NULL;
 177         pgd_t *pgd = pgd_offset_k(addr);
 178         pud_t *pud;
 179         pmd_t *pmd;
 180         pte_t *ptep, pte;
 181
 182         if (!pgd_none(*pgd)) {
 183                 pud = pud_offset(pgd, addr);
 184                 if (!pud_none(*pud)) {
 185                         pmd = pmd_offset(pud, addr);
 186                         if (!pmd_none(*pmd)) {
 187                                 ptep = pte_offset_map(pmd, addr);
 188                                 pte = *ptep;
 189                                 if (pte_present(pte))
 190                                         page = pte_page(pte);
 191                                 pte_unmap(ptep);
 192                         }
 193                 }
 194         }
 195         return page;
 196 }
 197 EXPORT_SYMBOL(vmalloc_to_page);
 198
 199 /*
 200  * Map a vmalloc()-space virtual address to the physical page frame number.
 201  */
 202 unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
 203 {
 204         return page_to_pfn(vmalloc_to_page(vmalloc_addr));
 205 }
 206 EXPORT_SYMBOL(vmalloc_to_pfn);
 207
 208 static struct vm_struct *
 209 __get_vm_area_node(unsigned long size, unsigned long flags, unsigned long start,
 210                 unsigned long end, int node, gfp_t gfp_mask, void *caller)
 211 {
 212         struct vm_struct **p, *tmp, *area;
 213         unsigned long align = 1;
 214         unsigned long addr;
 215
 216         BUG_ON(in_interrupt());
 217         if (flags & VM_IOREMAP) {
 218                 int bit = fls(size);
 219
 220                 if (bit > IOREMAP_MAX_ORDER)
 221                         bit = IOREMAP_MAX_ORDER;
 222                 else if (bit < PAGE_SHIFT)
 223                         bit = PAGE_SHIFT;
 224
 225                 align = 1ul << bit;
 226         }
 227         addr = ALIGN(start, align);
 228         size = PAGE_ALIGN(size);
 229         if (unlikely(!size))
 230                 return NULL;
 231
 232         area = kmalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
 233
 234         if (unlikely(!area))
 235                 return NULL;
 236
 237         /*
 238          * We always allocate a guard page.
 239          */
 240         size += PAGE_SIZE;
 241
 242         write_lock(&vmlist_lock);
 243         for (p = &vmlist; (tmp = *p) != NULL ;p = &tmp->next) {
 244                 if ((unsigned long)tmp->addr < addr) {
 245                         if((unsigned long)tmp->addr + tmp->size >= addr)
 246                                 addr = ALIGN(tmp->size +
 247                                              (unsigned long)tmp->addr, align);
 248                         continue;
 249                 }
 250                 if ((size + addr) < addr)
 251                         goto out;
 252                 if (size + addr <= (unsigned long)tmp->addr)
 253                         goto found;
 254                 addr = ALIGN(tmp->size + (unsigned long)tmp->addr, align);
 255                 if (addr > end - size)
 256                         goto out;
 257         }
 258         if ((size + addr) < addr)
 259                 goto out;
 260         if (addr > end - size)
 261                 goto out;
 262
 263 found:
 264         area->next = *p;
 265         *p = area;
 266
 267         area->flags = flags;
 268         area->addr = (void *)addr;
 269         area->size = size;
 270         area->pages = NULL;
 271         area->nr_pages = 0;
 272         area->phys_addr = 0;
 273         area->caller = caller;
 274         write_unlock(&vmlist_lock);
 275
 276         return area;
 277
 278 out:
 279         write_unlock(&vmlist_lock);
 280         kfree(area);
 281         if (printk_ratelimit())
 282                 printk(KERN_WARNING "allocation failed: out of vmalloc space - use vmalloc=<size> to increase size.\n");
 283         return NULL;
 284 }
 285
 286 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
 287                                 unsigned long start, unsigned long end)
 288 {
 289         return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL,
 290                                                 __builtin_return_address(0));
 291 }
 292 EXPORT_SYMBOL_GPL(__get_vm_area);
 293
 294 /**
 295  *      get_vm_area  -  reserve a contiguous kernel virtual area
 296  *      @size:          size of the area
 297  *      @flags:         %VM_IOREMAP for I/O mappings or VM_ALLOC
 298  *
 299  *      Search an area of @size in the kernel virtual mapping area,
 300  *      and reserved it for out purposes.  Returns the area descriptor
 301  *      on success or %NULL on failure.
 302  */
 303 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
 304 {
 305         return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END,
 306                                 -1, GFP_KERNEL, __builtin_return_address(0));
 307 }
 308
 309 struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
 310                                 void *caller)
 311 {
 312         return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END,
 313                                                 -1, GFP_KERNEL, caller);
 314 }
 315
 316 struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags,
 317                                    int node, gfp_t gfp_mask)
 318 {
 319         return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node,
 320                                   gfp_mask, __builtin_return_address(0));
 321 }
 322
 323 /* Caller must hold vmlist_lock */
 324 static struct vm_struct *__find_vm_area(const void *addr)
 325 {
 326         struct vm_struct *tmp;
 327
 328         for (tmp = vmlist; tmp != NULL; tmp = tmp->next) {
 329                  if (tmp->addr == addr)
 330                         break;
 331         }
 332
 333         return tmp;
 334 }
 335
 336 /* Caller must hold vmlist_lock */
 337 static struct vm_struct *__remove_vm_area(const void *addr)
 338 {
 339         struct vm_struct **p, *tmp;
 340
 341         for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) {
 342                  if (tmp->addr == addr)
 343                          goto found;
 344         }
 345         return NULL;
 346
 347 found:
 348         unmap_vm_area(tmp);
 349         *p = tmp->next;
 350
 351         /*
 352          * Remove the guard page.
 353          */
 354         tmp->size -= PAGE_SIZE;
 355         return tmp;
 356 }
 357
 358 /**
 359  *      remove_vm_area  -  find and remove a continuous kernel virtual area
 360  *      @addr:          base address
 361  *
 362  *      Search for the kernel VM area starting at @addr, and remove it.
 363  *      This function returns the found VM area, but using it is NOT safe
 364  *      on SMP machines, except for its size or flags.
 365  */
 366 struct vm_struct *remove_vm_area(const void *addr)
 367 {
 368         struct vm_struct *v;
 369         write_lock(&vmlist_lock);
 370         v = __remove_vm_area(addr);
 371         write_unlock(&vmlist_lock);
 372         return v;
 373 }
 374
 375 static void __vunmap(const void *addr, int deallocate_pages)
 376 {
 377         struct vm_struct *area;
 378
 379         if (!addr)
 380                 return;
 381
 382         if ((PAGE_SIZE-1) & (unsigned long)addr) {
 383                 printk(KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
 384                 WARN_ON(1);
 385                 return;
 386         }
 387
 388         area = remove_vm_area(addr);
 389         if (unlikely(!area)) {
 390                 printk(KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
 391                                 addr);
 392                 WARN_ON(1);
 393                 return;
 394         }
 395
 396         debug_check_no_locks_freed(addr, area->size);
 397
 398         if (deallocate_pages) {
 399                 int i;
 400
 401                 for (i = 0; i < area->nr_pages; i++) {
 402                         struct page *page = area->pages[i];
 403
 404                         BUG_ON(!page);
 405                         __free_page(page);
 406                 }
 407
 408                 if (area->flags & VM_VPAGES)
 409                         vfree(area->pages);
 410                 else
 411                         kfree(area->pages);
 412         }
 413
 414         kfree(area);
 415         return;
 416 }
 417
 418 /**
 419  *      vfree  -  release memory allocated by vmalloc()
 420  *      @addr:          memory base address
 421  *
 422  *      Free the virtually continuous memory area starting at @addr, as
 423  *      obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
 424  *      NULL, no operation is performed.
 425  *
 426  *      Must not be called in interrupt context.
 427  */
 428 void vfree(const void *addr)
 429 {
 430         BUG_ON(in_interrupt());
 431         __vunmap(addr, 1);
 432 }
 433 EXPORT_SYMBOL(vfree);
 434
 435 /**
 436  *      vunmap  -  release virtual mapping obtained by vmap()
 437  *      @addr:          memory base address
 438  *
 439  *      Free the virtually contiguous memory area starting at @addr,
 440  *      which was created from the page array passed to vmap().
 441  *
 442  *      Must not be called in interrupt context.
 443  */
 444 void vunmap(const void *addr)
 445 {
 446         BUG_ON(in_interrupt());
 447         __vunmap(addr, 0);
 448 }
 449 EXPORT_SYMBOL(vunmap);
 450
 451 /**
 452  *      vmap  -  map an array of pages into virtually contiguous space
 453  *      @pages:         array of page pointers
 454  *      @count:         number of pages to map
 455  *      @flags:         vm_area->flags
 456  *      @prot:          page protection for the mapping
 457  *
 458  *      Maps @count pages from @pages into contiguous kernel virtual
 459  *      space.
 460  */
 461 void *vmap(struct page **pages, unsigned int count,
 462                 unsigned long flags, pgprot_t prot)
 463 {
 464         struct vm_struct *area;
 465
 466         if (count > num_physpages)
 467                 return NULL;
 468
 469         area = get_vm_area_caller((count << PAGE_SHIFT), flags,
 470                                         __builtin_return_address(0));
 471         if (!area)
 472                 return NULL;
 473
 474         if (map_vm_area(area, prot, &pages)) {
 475                 vunmap(area->addr);
 476                 return NULL;
 477         }
 478
 479         return area->addr;
 480 }
 481 EXPORT_SYMBOL(vmap);
 482
 483 static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
 484                                  pgprot_t prot, int node, void *caller)
 485 {
 486         struct page **pages;
 487         unsigned int nr_pages, array_size, i;
 488
 489         nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
 490         array_size = (nr_pages * sizeof(struct page *));
 491
 492         area->nr_pages = nr_pages;
 493         /* Please note that the recursion is strictly bounded. */
 494         if (array_size > PAGE_SIZE) {
 495                 pages = __vmalloc_node(array_size, gfp_mask | __GFP_ZERO,
 496                                 PAGE_KERNEL, node, caller);
 497                 area->flags |= VM_VPAGES;
 498         } else {
 499                 pages = kmalloc_node(array_size,
 500                                 (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO,
 501                                 node);
 502         }
 503         area->pages = pages;
 504         area->caller = caller;
 505         if (!area->pages) {
 506                 remove_vm_area(area->addr);
 507                 kfree(area);
 508                 return NULL;
 509         }
 510
 511         for (i = 0; i < area->nr_pages; i++) {
 512                 struct page *page;
 513
 514                 if (node < 0)
 515                         page = alloc_page(gfp_mask);
 516                 else
 517                         page = alloc_pages_node(node, gfp_mask, 0);
 518
 519                 if (unlikely(!page)) {
 520                         /* Successfully allocated i pages, free them in __vunmap() */
 521                         area->nr_pages = i;
 522                         goto fail;
 523                 }
 524                 area->pages[i] = page;
 525         }
 526
 527         if (map_vm_area(area, prot, &pages))
 528                 goto fail;
 529         return area->addr;
 530
 531 fail:
 532         vfree(area->addr);
 533         return NULL;
 534 }
 535
 536 void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
 537 {
 538         return __vmalloc_area_node(area, gfp_mask, prot, -1,
 539                                         __builtin_return_address(0));
 540 }
 541
 542 /**
 543  *      __vmalloc_node  -  allocate virtually contiguous memory
 544  *      @size:          allocation size
 545  *      @gfp_mask:      flags for the page level allocator
 546  *      @prot:          protection mask for the allocated pages
 547  *      @node:          node to use for allocation or -1
 548  *
 549  *      Allocate enough pages to cover @size from the page level
 550  *      allocator with @gfp_mask flags.  Map them into contiguous
 551  *      kernel virtual space, using a pagetable protection of @prot.
 552  */
 553 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
 554                                                 int node, void *caller)
 555 {
 556         struct vm_struct *area;
 557
 558         size = PAGE_ALIGN(size);
 559         if (!size || (size >> PAGE_SHIFT) > num_physpages)
 560                 return NULL;
 561
 562         area = __get_vm_area_node(size, VM_ALLOC, VMALLOC_START, VMALLOC_END,
 563                                                 node, gfp_mask, caller);
 564
 565         if (!area)
 566                 return NULL;
 567
 568         return __vmalloc_area_node(area, gfp_mask, prot, node, caller);
 569 }
 570
 571 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
 572 {
 573         return __vmalloc_node(size, gfp_mask, prot, -1,
 574                                 __builtin_return_address(0));
 575 }
 576 EXPORT_SYMBOL(__vmalloc);
 577
 578 /**
 579  *      vmalloc  -  allocate virtually contiguous memory
 580  *      @size:          allocation size
 581  *      Allocate enough pages to cover @size from the page level
 582  *      allocator and map them into contiguous kernel virtual space.
 583  *
 584  *      For tight control over page level allocator and protection flags
 585  *      use __vmalloc() instead.
 586  */
 587 void *vmalloc(unsigned long size)
 588 {
 589         return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
 590                                         -1, __builtin_return_address(0));
 591 }
 592 EXPORT_SYMBOL(vmalloc);
 593
 594 /**
 595  * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
 596  * @size: allocation size
 597  *
 598  * The resulting memory area is zeroed so it can be mapped to userspace
 599  * without leaking data.
 600  */
 601 void *vmalloc_user(unsigned long size)
 602 {
 603         struct vm_struct *area;
 604         void *ret;
 605
 606         ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL);
 607         if (ret) {
 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_user);
 616
 617 /**
 618  *      vmalloc_node  -  allocate memory on a specific node
 619  *      @size:          allocation size
 620  *      @node:          numa node
 621  *
 622  *      Allocate enough pages to cover @size from the page level
 623  *      allocator and map them into contiguous kernel virtual space.
 624  *
 625  *      For tight control over page level allocator and protection flags
 626  *      use __vmalloc() instead.
 627  */
 628 void *vmalloc_node(unsigned long size, int node)
 629 {
 630         return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
 631                                         node, __builtin_return_address(0));
 632 }
 633 EXPORT_SYMBOL(vmalloc_node);
 634
 635 #ifndef PAGE_KERNEL_EXEC
 636 # define PAGE_KERNEL_EXEC PAGE_KERNEL
 637 #endif
 638
 639 /**
 640  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
 641  *      @size:          allocation size
 642  *
 643  *      Kernel-internal function to allocate enough pages to cover @size
 644  *      the page level allocator and map them into contiguous and
 645  *      executable kernel virtual space.
 646  *
 647  *      For tight control over page level allocator and protection flags
 648  *      use __vmalloc() instead.
 649  */
 650
 651 void *vmalloc_exec(unsigned long size)
 652 {
 653         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
 654 }
 655
 656 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
 657 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
 658 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
 659 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
 660 #else
 661 #define GFP_VMALLOC32 GFP_KERNEL
 662 #endif
 663
 664 /**
 665  *      vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
 666  *      @size:          allocation size
 667  *
 668  *      Allocate enough 32bit PA addressable pages to cover @size from the
 669  *      page level allocator and map them into contiguous kernel virtual space.
 670  */
 671 void *vmalloc_32(unsigned long size)
 672 {
 673         return __vmalloc(size, GFP_VMALLOC32, PAGE_KERNEL);
 674 }
 675 EXPORT_SYMBOL(vmalloc_32);
 676
 677 /**
 678  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
 679  *      @size:          allocation size
 680  *
 681  * The resulting memory area is 32bit addressable and zeroed so it can be
 682  * mapped to userspace without leaking data.
 683  */
 684 void *vmalloc_32_user(unsigned long size)
 685 {
 686         struct vm_struct *area;
 687         void *ret;
 688
 689         ret = __vmalloc(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL);
 690         if (ret) {
 691                 write_lock(&vmlist_lock);
 692                 area = __find_vm_area(ret);
 693                 area->flags |= VM_USERMAP;
 694                 write_unlock(&vmlist_lock);
 695         }
 696         return ret;
 697 }
 698 EXPORT_SYMBOL(vmalloc_32_user);
 699
 700 long vread(char *buf, char *addr, unsigned long count)
 701 {
 702         struct vm_struct *tmp;
 703         char *vaddr, *buf_start = buf;
 704         unsigned long n;
 705
 706         /* Don't allow overflow */
 707         if ((unsigned long) addr + count < count)
 708                 count = -(unsigned long) addr;
 709
 710         read_lock(&vmlist_lock);
 711         for (tmp = vmlist; tmp; tmp = tmp->next) {
 712                 vaddr = (char *) tmp->addr;
 713                 if (addr >= vaddr + tmp->size - PAGE_SIZE)
 714                         continue;
 715                 while (addr < vaddr) {
 716                         if (count == 0)
 717                                 goto finished;
 718                         *buf = '\0';
 719                         buf++;
 720                         addr++;
 721                         count--;
 722                 }
 723                 n = vaddr + tmp->size - PAGE_SIZE - addr;
 724                 do {
 725                         if (count == 0)
 726                                 goto finished;
 727                         *buf = *addr;
 728                         buf++;
 729                         addr++;
 730                         count--;
 731                 } while (--n > 0);
 732         }
 733 finished:
 734         read_unlock(&vmlist_lock);
 735         return buf - buf_start;
 736 }
 737
 738 long vwrite(char *buf, char *addr, unsigned long count)
 739 {
 740         struct vm_struct *tmp;
 741         char *vaddr, *buf_start = buf;
 742         unsigned long n;
 743
 744         /* Don't allow overflow */
 745         if ((unsigned long) addr + count < count)
 746                 count = -(unsigned long) addr;
 747
 748         read_lock(&vmlist_lock);
 749         for (tmp = vmlist; tmp; tmp = tmp->next) {
 750                 vaddr = (char *) tmp->addr;
 751                 if (addr >= vaddr + tmp->size - PAGE_SIZE)
 752                         continue;
 753                 while (addr < vaddr) {
 754                         if (count == 0)
 755                                 goto finished;
 756                         buf++;
 757                         addr++;
 758                         count--;
 759                 }
 760                 n = vaddr + tmp->size - PAGE_SIZE - addr;
 761                 do {
 762                         if (count == 0)
 763                                 goto finished;
 764                         *addr = *buf;
 765                         buf++;
 766                         addr++;
 767                         count--;
 768                 } while (--n > 0);
 769         }
 770 finished:
 771         read_unlock(&vmlist_lock);
 772         return buf - buf_start;
 773 }
 774
 775 /**
 776  *      remap_vmalloc_range  -  map vmalloc pages to userspace
 777  *      @vma:           vma to cover (map full range of vma)
 778  *      @addr:          vmalloc memory
 779  *      @pgoff:         number of pages into addr before first page to map
 780  *
 781  *      Returns:        0 for success, -Exxx on failure
 782  *
 783  *      This function checks that addr is a valid vmalloc'ed area, and
 784  *      that it is big enough to cover the vma. Will return failure if
 785  *      that criteria isn't met.
 786  *
 787  *      Similar to remap_pfn_range() (see mm/memory.c)
 788  */
 789 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
 790                                                 unsigned long pgoff)
 791 {
 792         struct vm_struct *area;
 793         unsigned long uaddr = vma->vm_start;
 794         unsigned long usize = vma->vm_end - vma->vm_start;
 795         int ret;
 796
 797         if ((PAGE_SIZE-1) & (unsigned long)addr)
 798                 return -EINVAL;
 799
 800         read_lock(&vmlist_lock);
 801         area = __find_vm_area(addr);
 802         if (!area)
 803                 goto out_einval_locked;
 804
 805         if (!(area->flags & VM_USERMAP))
 806                 goto out_einval_locked;
 807
 808         if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
 809                 goto out_einval_locked;
 810         read_unlock(&vmlist_lock);
 811
 812         addr += pgoff << PAGE_SHIFT;
 813         do {
 814                 struct page *page = vmalloc_to_page(addr);
 815                 ret = vm_insert_page(vma, uaddr, page);
 816                 if (ret)
 817                         return ret;
 818
 819                 uaddr += PAGE_SIZE;
 820                 addr += PAGE_SIZE;
 821                 usize -= PAGE_SIZE;
 822         } while (usize > 0);
 823
 824         /* Prevent "things" like memory migration? VM_flags need a cleanup... */
 825         vma->vm_flags |= VM_RESERVED;
 826
 827         return ret;
 828
 829 out_einval_locked:
 830         read_unlock(&vmlist_lock);
 831         return -EINVAL;
 832 }
 833 EXPORT_SYMBOL(remap_vmalloc_range);
 834
 835 /*
 836  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
 837  * have one.
 838  */
 839 void  __attribute__((weak)) vmalloc_sync_all(void)
 840 {
 841 }
 842
 843
 844 static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
 845 {
 846         /* apply_to_page_range() does all the hard work. */
 847         return 0;
 848 }
 849
 850 /**
 851  *      alloc_vm_area - allocate a range of kernel address space
 852  *      @size:          size of the area
 853  *
 854  *      Returns:        NULL on failure, vm_struct on success
 855  *
 856  *      This function reserves a range of kernel address space, and
 857  *      allocates pagetables to map that range.  No actual mappings
 858  *      are created.  If the kernel address space is not shared
 859  *      between processes, it syncs the pagetable across all
 860  *      processes.
 861  */
 862 struct vm_struct *alloc_vm_area(size_t size)
 863 {
 864         struct vm_struct *area;
 865
 866         area = get_vm_area_caller(size, VM_IOREMAP,
 867                                 __builtin_return_address(0));
 868         if (area == NULL)
 869                 return NULL;
 870
 871         /*
 872          * This ensures that page tables are constructed for this region
 873          * of kernel virtual address space and mapped into init_mm.
 874          */
 875         if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
 876                                 area->size, f, NULL)) {
 877                 free_vm_area(area);
 878                 return NULL;
 879         }
 880
 881         /* Make sure the pagetables are constructed in process kernel
 882            mappings */
 883         vmalloc_sync_all();
 884
 885         return area;
 886 }
 887 EXPORT_SYMBOL_GPL(alloc_vm_area);
 888
 889 void free_vm_area(struct vm_struct *area)
 890 {
 891         struct vm_struct *ret;
 892         ret = remove_vm_area(area->addr);
 893         BUG_ON(ret != area);
 894         kfree(area);
 895 }
 896 EXPORT_SYMBOL_GPL(free_vm_area);
 897
 898
 899 #ifdef CONFIG_PROC_FS
 900 static void *s_start(struct seq_file *m, loff_t *pos)
 901 {
 902         loff_t n = *pos;
 903         struct vm_struct *v;
 904
 905         read_lock(&vmlist_lock);
 906         v = vmlist;
 907         while (n > 0 && v) {
 908                 n--;
 909                 v = v->next;
 910         }
 911         if (!n)
 912                 return v;
 913
 914         return NULL;
 915
 916 }
 917
 918 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
 919 {
 920         struct vm_struct *v = p;
 921
 922         ++*pos;
 923         return v->next;
 924 }
 925
 926 static void s_stop(struct seq_file *m, void *p)
 927 {
 928         read_unlock(&vmlist_lock);
 929 }
 930
 931 static int s_show(struct seq_file *m, void *p)
 932 {
 933         struct vm_struct *v = p;
 934
 935         seq_printf(m, "0x%p-0x%p %7ld",
 936                 v->addr, v->addr + v->size, v->size);
 937
 938         if (v->caller) {
 939                 char buff[2 * KSYM_NAME_LEN];
 940
 941                 seq_putc(m, ' ');
 942                 sprint_symbol(buff, (unsigned long)v->caller);
 943                 seq_puts(m, buff);
 944         }
 945
 946         if (v->nr_pages)
 947                 seq_printf(m, " pages=%d", v->nr_pages);
 948
 949         if (v->phys_addr)
 950                 seq_printf(m, " phys=%lx", v->phys_addr);
 951
 952         if (v->flags & VM_IOREMAP)
 953                 seq_printf(m, " ioremap");
 954
 955         if (v->flags & VM_ALLOC)
 956                 seq_printf(m, " vmalloc");
 957
 958         if (v->flags & VM_MAP)
 959                 seq_printf(m, " vmap");
 960
 961         if (v->flags & VM_USERMAP)
 962                 seq_printf(m, " user");
 963
 964         if (v->flags & VM_VPAGES)
 965                 seq_printf(m, " vpages");
 966
 967         seq_putc(m, '\n');
 968         return 0;
 969 }
 970
 971 const struct seq_operations vmalloc_op = {
 972         .start = s_start,
 973         .next = s_next,
 974         .stop = s_stop,
 975         .show = s_show,
 976 };
 977 #endif
 978