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/debugobjects.h>
  19 #include <linux/vmalloc.h>
  20 #include <linux/kallsyms.h>
  21
  22 #include <asm/uaccess.h>
  23 #include <asm/tlbflush.h>
  24
  25
  26 DEFINE_RWLOCK(vmlist_lock);
  27 struct vm_struct *vmlist;
  28
  29 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
  30                             int node, void *caller);
  31
  32 static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
  33 {
  34         pte_t *pte;
  35
  36         pte = pte_offset_kernel(pmd, addr);
  37         do {
  38                 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
  39                 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
  40         } while (pte++, addr += PAGE_SIZE, addr != end);
  41 }
  42
  43 static inline void vunmap_pmd_range(pud_t *pud, unsigned long addr,
  44                                                 unsigned long end)
  45 {
  46         pmd_t *pmd;
  47         unsigned long next;
  48
  49         pmd = pmd_offset(pud, addr);
  50         do {
  51                 next = pmd_addr_end(addr, end);
  52                 if (pmd_none_or_clear_bad(pmd))
  53                         continue;
  54                 vunmap_pte_range(pmd, addr, next);
  55         } while (pmd++, addr = next, addr != end);
  56 }
  57
  58 static inline void vunmap_pud_range(pgd_t *pgd, unsigned long addr,
  59                                                 unsigned long end)
  60 {
  61         pud_t *pud;
  62         unsigned long next;
  63
  64         pud = pud_offset(pgd, addr);
  65         do {
  66                 next = pud_addr_end(addr, end);
  67                 if (pud_none_or_clear_bad(pud))
  68                         continue;
  69                 vunmap_pmd_range(pud, addr, next);
  70         } while (pud++, addr = next, addr != end);
  71 }
  72
  73 void unmap_kernel_range(unsigned long addr, unsigned long size)
  74 {
  75         pgd_t *pgd;
  76         unsigned long next;
  77         unsigned long start = addr;
  78         unsigned long end = addr + size;
  79
  80         BUG_ON(addr >= end);
  81         pgd = pgd_offset_k(addr);
  82         flush_cache_vunmap(addr, end);
  83         do {
  84                 next = pgd_addr_end(addr, end);
  85                 if (pgd_none_or_clear_bad(pgd))
  86                         continue;
  87                 vunmap_pud_range(pgd, addr, next);
  88         } while (pgd++, addr = next, addr != end);
  89         flush_tlb_kernel_range(start, end);
  90 }
  91
  92 static void unmap_vm_area(struct vm_struct *area)
  93 {
  94         unmap_kernel_range((unsigned long)area->addr, area->size);
  95 }
  96
  97 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
  98                         unsigned long end, pgprot_t prot, struct page ***pages)
  99 {
 100         pte_t *pte;
 101
 102         pte = pte_alloc_kernel(pmd, addr);
 103         if (!pte)
 104                 return -ENOMEM;
 105         do {
 106                 struct page *page = **pages;
 107                 WARN_ON(!pte_none(*pte));
 108                 if (!page)
 109                         return -ENOMEM;
 110                 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
 111                 (*pages)++;
 112         } while (pte++, addr += PAGE_SIZE, addr != end);
 113         return 0;
 114 }
 115
 116 static inline int vmap_pmd_range(pud_t *pud, unsigned long addr,
 117                         unsigned long end, pgprot_t prot, struct page ***pages)
 118 {
 119         pmd_t *pmd;
 120         unsigned long next;
 121
 122         pmd = pmd_alloc(&init_mm, pud, addr);
 123         if (!pmd)
 124                 return -ENOMEM;
 125         do {
 126                 next = pmd_addr_end(addr, end);
 127                 if (vmap_pte_range(pmd, addr, next, prot, pages))
 128                         return -ENOMEM;
 129         } while (pmd++, addr = next, addr != end);
 130         return 0;
 131 }
 132
 133 static inline int vmap_pud_range(pgd_t *pgd, unsigned long addr,
 134                         unsigned long end, pgprot_t prot, struct page ***pages)
 135 {
 136         pud_t *pud;
 137         unsigned long next;
 138
 139         pud = pud_alloc(&init_mm, pgd, addr);
 140         if (!pud)
 141                 return -ENOMEM;
 142         do {
 143                 next = pud_addr_end(addr, end);
 144                 if (vmap_pmd_range(pud, addr, next, prot, pages))
 145                         return -ENOMEM;
 146         } while (pud++, addr = next, addr != end);
 147         return 0;
 148 }
 149
 150 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
 151 {
 152         pgd_t *pgd;
 153         unsigned long next;
 154         unsigned long addr = (unsigned long) area->addr;
 155         unsigned long end = addr + area->size - PAGE_SIZE;
 156         int err;
 157
 158         BUG_ON(addr >= end);
 159         pgd = pgd_offset_k(addr);
 160         do {
 161                 next = pgd_addr_end(addr, end);
 162                 err = vmap_pud_range(pgd, addr, next, prot, pages);
 163                 if (err)
 164                         break;
 165         } while (pgd++, addr = next, addr != end);
 166         flush_cache_vmap((unsigned long) area->addr, end);
 167         return err;
 168 }
 169 EXPORT_SYMBOL_GPL(map_vm_area);
 170
 171 /*
 172  * Map a vmalloc()-space virtual address to the physical page.
 173  */
 174 struct page *vmalloc_to_page(const void *vmalloc_addr)
 175 {
 176         unsigned long addr = (unsigned long) vmalloc_addr;
 177         struct page *page = NULL;
 178         pgd_t *pgd = pgd_offset_k(addr);
 179         pud_t *pud;
 180         pmd_t *pmd;
 181         pte_t *ptep, pte;
 182
 183         if (!pgd_none(*pgd)) {
 184                 pud = pud_offset(pgd, addr);
 185                 if (!pud_none(*pud)) {
 186                         pmd = pmd_offset(pud, addr);
 187                         if (!pmd_none(*pmd)) {
 188                                 ptep = pte_offset_map(pmd, addr);
 189                                 pte = *ptep;
 190                                 if (pte_present(pte))
 191                                         page = pte_page(pte);
 192                                 pte_unmap(ptep);
 193                         }
 194                 }
 195         }
 196         return page;
 197 }
 198 EXPORT_SYMBOL(vmalloc_to_page);
 199
 200 /*
 201  * Map a vmalloc()-space virtual address to the physical page frame number.
 202  */
 203 unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
 204 {
 205         return page_to_pfn(vmalloc_to_page(vmalloc_addr));
 206 }
 207 EXPORT_SYMBOL(vmalloc_to_pfn);
 208
 209 static struct vm_struct *
 210 __get_vm_area_node(unsigned long size, unsigned long flags, unsigned long start,
 211                 unsigned long end, int node, gfp_t gfp_mask, void *caller)
 212 {
 213         struct vm_struct **p, *tmp, *area;
 214         unsigned long align = 1;
 215         unsigned long addr;
 216
 217         BUG_ON(in_interrupt());
 218         if (flags & VM_IOREMAP) {
 219                 int bit = fls(size);
 220
 221                 if (bit > IOREMAP_MAX_ORDER)
 222                         bit = IOREMAP_MAX_ORDER;
 223                 else if (bit < PAGE_SHIFT)
 224                         bit = PAGE_SHIFT;
 225
 226                 align = 1ul << bit;
 227         }
 228         addr = ALIGN(start, align);
 229         size = PAGE_ALIGN(size);
 230         if (unlikely(!size))
 231                 return NULL;
 232
 233         area = kmalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
 234
 235         if (unlikely(!area))
 236                 return NULL;
 237
 238         /*
 239          * We always allocate a guard page.
 240          */
 241         size += PAGE_SIZE;
 242
 243         write_lock(&vmlist_lock);
 244         for (p = &vmlist; (tmp = *p) != NULL ;p = &tmp->next) {
 245                 if ((unsigned long)tmp->addr < addr) {
 246                         if((unsigned long)tmp->addr + tmp->size >= addr)
 247                                 addr = ALIGN(tmp->size +
 248                                              (unsigned long)tmp->addr, align);
 249                         continue;
 250                 }
 251                 if ((size + addr) < addr)
 252                         goto out;
 253                 if (size + addr <= (unsigned long)tmp->addr)
 254                         goto found;
 255                 addr = ALIGN(tmp->size + (unsigned long)tmp->addr, align);
 256                 if (addr > end - size)
 257                         goto out;
 258         }
 259         if ((size + addr) < addr)
 260                 goto out;
 261         if (addr > end - size)
 262                 goto out;
 263
 264 found:
 265         area->next = *p;
 266         *p = area;
 267
 268         area->flags = flags;
 269         area->addr = (void *)addr;
 270         area->size = size;
 271         area->pages = NULL;
 272         area->nr_pages = 0;
 273         area->phys_addr = 0;
 274         area->caller = caller;
 275         write_unlock(&vmlist_lock);
 276
 277         return area;
 278
 279 out:
 280         write_unlock(&vmlist_lock);
 281         kfree(area);
 282         if (printk_ratelimit())
 283                 printk(KERN_WARNING "allocation failed: out of vmalloc space - use vmalloc=<size> to increase size.\n");
 284         return NULL;
 285 }
 286
 287 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
 288                                 unsigned long start, unsigned long end)
 289 {
 290         return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL,
 291                                                 __builtin_return_address(0));
 292 }
 293 EXPORT_SYMBOL_GPL(__get_vm_area);
 294
 295 /**
 296  *      get_vm_area  -  reserve a contiguous kernel virtual area
 297  *      @size:          size of the area
 298  *      @flags:         %VM_IOREMAP for I/O mappings or VM_ALLOC
 299  *
 300  *      Search an area of @size in the kernel virtual mapping area,
 301  *      and reserved it for out purposes.  Returns the area descriptor
 302  *      on success or %NULL on failure.
 303  */
 304 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
 305 {
 306         return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END,
 307                                 -1, GFP_KERNEL, __builtin_return_address(0));
 308 }
 309
 310 struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
 311                                 void *caller)
 312 {
 313         return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END,
 314                                                 -1, GFP_KERNEL, caller);
 315 }
 316
 317 struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags,
 318                                    int node, gfp_t gfp_mask)
 319 {
 320         return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node,
 321                                   gfp_mask, __builtin_return_address(0));
 322 }
 323
 324 /* Caller must hold vmlist_lock */
 325 static struct vm_struct *__find_vm_area(const void *addr)
 326 {
 327         struct vm_struct *tmp;
 328
 329         for (tmp = vmlist; tmp != NULL; tmp = tmp->next) {
 330                  if (tmp->addr == addr)
 331                         break;
 332         }
 333
 334         return tmp;
 335 }
 336
 337 /* Caller must hold vmlist_lock */
 338 static struct vm_struct *__remove_vm_area(const void *addr)
 339 {
 340         struct vm_struct **p, *tmp;
 341
 342         for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) {
 343                  if (tmp->addr == addr)
 344                          goto found;
 345         }
 346         return NULL;
 347
 348 found:
 349         unmap_vm_area(tmp);
 350         *p = tmp->next;
 351
 352         /*
 353          * Remove the guard page.
 354          */
 355         tmp->size -= PAGE_SIZE;
 356         return tmp;
 357 }
 358
 359 /**
 360  *      remove_vm_area  -  find and remove a continuous kernel virtual area
 361  *      @addr:          base address
 362  *
 363  *      Search for the kernel VM area starting at @addr, and remove it.
 364  *      This function returns the found VM area, but using it is NOT safe
 365  *      on SMP machines, except for its size or flags.
 366  */
 367 struct vm_struct *remove_vm_area(const void *addr)
 368 {
 369         struct vm_struct *v;
 370         write_lock(&vmlist_lock);
 371         v = __remove_vm_area(addr);
 372         write_unlock(&vmlist_lock);
 373         return v;
 374 }
 375
 376 static void __vunmap(const void *addr, int deallocate_pages)
 377 {
 378         struct vm_struct *area;
 379
 380         if (!addr)
 381                 return;
 382
 383         if ((PAGE_SIZE-1) & (unsigned long)addr) {
 384                 printk(KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
 385                 WARN_ON(1);
 386                 return;
 387         }
 388
 389         area = remove_vm_area(addr);
 390         if (unlikely(!area)) {
 391                 printk(KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
 392                                 addr);
 393                 WARN_ON(1);
 394                 return;
 395         }
 396
 397         debug_check_no_locks_freed(addr, area->size);
 398         debug_check_no_obj_freed(addr, area->size);
 399
 400         if (deallocate_pages) {
 401                 int i;
 402
 403                 for (i = 0; i < area->nr_pages; i++) {
 404                         struct page *page = area->pages[i];
 405
 406                         BUG_ON(!page);
 407                         __free_page(page);
 408                 }
 409
 410                 if (area->flags & VM_VPAGES)
 411                         vfree(area->pages);
 412                 else
 413                         kfree(area->pages);
 414         }
 415
 416         kfree(area);
 417         return;
 418 }
 419
 420 /**
 421  *      vfree  -  release memory allocated by vmalloc()
 422  *      @addr:          memory base address
 423  *
 424  *      Free the virtually continuous memory area starting at @addr, as
 425  *      obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
 426  *      NULL, no operation is performed.
 427  *
 428  *      Must not be called in interrupt context.
 429  */
 430 void vfree(const void *addr)
 431 {
 432         BUG_ON(in_interrupt());
 433         __vunmap(addr, 1);
 434 }
 435 EXPORT_SYMBOL(vfree);
 436
 437 /**
 438  *      vunmap  -  release virtual mapping obtained by vmap()
 439  *      @addr:          memory base address
 440  *
 441  *      Free the virtually contiguous memory area starting at @addr,
 442  *      which was created from the page array passed to vmap().
 443  *
 444  *      Must not be called in interrupt context.
 445  */
 446 void vunmap(const void *addr)
 447 {
 448         BUG_ON(in_interrupt());
 449         __vunmap(addr, 0);
 450 }
 451 EXPORT_SYMBOL(vunmap);
 452
 453 /**
 454  *      vmap  -  map an array of pages into virtually contiguous space
 455  *      @pages:         array of page pointers
 456  *      @count:         number of pages to map
 457  *      @flags:         vm_area->flags
 458  *      @prot:          page protection for the mapping
 459  *
 460  *      Maps @count pages from @pages into contiguous kernel virtual
 461  *      space.
 462  */
 463 void *vmap(struct page **pages, unsigned int count,
 464                 unsigned long flags, pgprot_t prot)
 465 {
 466         struct vm_struct *area;
 467
 468         if (count > num_physpages)
 469                 return NULL;
 470
 471         area = get_vm_area_caller((count << PAGE_SHIFT), flags,
 472                                         __builtin_return_address(0));
 473         if (!area)
 474                 return NULL;
 475
 476         if (map_vm_area(area, prot, &pages)) {
 477                 vunmap(area->addr);
 478                 return NULL;
 479         }
 480
 481         return area->addr;
 482 }
 483 EXPORT_SYMBOL(vmap);
 484
 485 static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
 486                                  pgprot_t prot, int node, void *caller)
 487 {
 488         struct page **pages;
 489         unsigned int nr_pages, array_size, i;
 490
 491         nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
 492         array_size = (nr_pages * sizeof(struct page *));
 493
 494         area->nr_pages = nr_pages;
 495         /* Please note that the recursion is strictly bounded. */
 496         if (array_size > PAGE_SIZE) {
 497                 pages = __vmalloc_node(array_size, gfp_mask | __GFP_ZERO,
 498                                 PAGE_KERNEL, node, caller);
 499                 area->flags |= VM_VPAGES;
 500         } else {
 501                 pages = kmalloc_node(array_size,
 502                                 (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO,
 503                                 node);
 504         }
 505         area->pages = pages;
 506         area->caller = caller;
 507         if (!area->pages) {
 508                 remove_vm_area(area->addr);
 509                 kfree(area);
 510                 return NULL;
 511         }
 512
 513         for (i = 0; i < area->nr_pages; i++) {
 514                 struct page *page;
 515
 516                 if (node < 0)
 517                         page = alloc_page(gfp_mask);
 518                 else
 519                         page = alloc_pages_node(node, gfp_mask, 0);
 520
 521                 if (unlikely(!page)) {
 522                         /* Successfully allocated i pages, free them in __vunmap() */
 523                         area->nr_pages = i;
 524                         goto fail;
 525                 }
 526                 area->pages[i] = page;
 527         }
 528
 529         if (map_vm_area(area, prot, &pages))
 530                 goto fail;
 531         return area->addr;
 532
 533 fail:
 534         vfree(area->addr);
 535         return NULL;
 536 }
 537
 538 void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
 539 {
 540         return __vmalloc_area_node(area, gfp_mask, prot, -1,
 541                                         __builtin_return_address(0));
 542 }
 543
 544 /**
 545  *      __vmalloc_node  -  allocate virtually contiguous memory
 546  *      @size:          allocation size
 547  *      @gfp_mask:      flags for the page level allocator
 548  *      @prot:          protection mask for the allocated pages
 549  *      @node:          node to use for allocation or -1
 550  *      @caller:        caller's return address
 551  *
 552  *      Allocate enough pages to cover @size from the page level
 553  *      allocator with @gfp_mask flags.  Map them into contiguous
 554  *      kernel virtual space, using a pagetable protection of @prot.
 555  */
 556 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
 557                                                 int node, void *caller)
 558 {
 559         struct vm_struct *area;
 560
 561         size = PAGE_ALIGN(size);
 562         if (!size || (size >> PAGE_SHIFT) > num_physpages)
 563                 return NULL;
 564
 565         area = __get_vm_area_node(size, VM_ALLOC, VMALLOC_START, VMALLOC_END,
 566                                                 node, gfp_mask, caller);
 567
 568         if (!area)
 569                 return NULL;
 570
 571         return __vmalloc_area_node(area, gfp_mask, prot, node, caller);
 572 }
 573
 574 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
 575 {
 576         return __vmalloc_node(size, gfp_mask, prot, -1,
 577                                 __builtin_return_address(0));
 578 }
 579 EXPORT_SYMBOL(__vmalloc);
 580
 581 /**
 582  *      vmalloc  -  allocate virtually contiguous memory
 583  *      @size:          allocation size
 584  *      Allocate enough pages to cover @size from the page level
 585  *      allocator and map them into contiguous kernel virtual space.
 586  *
 587  *      For tight control over page level allocator and protection flags
 588  *      use __vmalloc() instead.
 589  */
 590 void *vmalloc(unsigned long size)
 591 {
 592         return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
 593                                         -1, __builtin_return_address(0));
 594 }
 595 EXPORT_SYMBOL(vmalloc);
 596
 597 /**
 598  * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
 599  * @size: allocation size
 600  *
 601  * The resulting memory area is zeroed so it can be mapped to userspace
 602  * without leaking data.
 603  */
 604 void *vmalloc_user(unsigned long size)
 605 {
 606         struct vm_struct *area;
 607         void *ret;
 608
 609         ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL);
 610         if (ret) {
 611                 write_lock(&vmlist_lock);
 612                 area = __find_vm_area(ret);
 613                 area->flags |= VM_USERMAP;
 614                 write_unlock(&vmlist_lock);
 615         }
 616         return ret;
 617 }
 618 EXPORT_SYMBOL(vmalloc_user);
 619
 620 /**
 621  *      vmalloc_node  -  allocate memory on a specific node
 622  *      @size:          allocation size
 623  *      @node:          numa node
 624  *
 625  *      Allocate enough pages to cover @size from the page level
 626  *      allocator and map them into contiguous kernel virtual space.
 627  *
 628  *      For tight control over page level allocator and protection flags
 629  *      use __vmalloc() instead.
 630  */
 631 void *vmalloc_node(unsigned long size, int node)
 632 {
 633         return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
 634                                         node, __builtin_return_address(0));
 635 }
 636 EXPORT_SYMBOL(vmalloc_node);
 637
 638 #ifndef PAGE_KERNEL_EXEC
 639 # define PAGE_KERNEL_EXEC PAGE_KERNEL
 640 #endif
 641
 642 /**
 643  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
 644  *      @size:          allocation size
 645  *
 646  *      Kernel-internal function to allocate enough pages to cover @size
 647  *      the page level allocator and map them into contiguous and
 648  *      executable kernel virtual space.
 649  *
 650  *      For tight control over page level allocator and protection flags
 651  *      use __vmalloc() instead.
 652  */
 653
 654 void *vmalloc_exec(unsigned long size)
 655 {
 656         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
 657 }
 658
 659 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
 660 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
 661 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
 662 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
 663 #else
 664 #define GFP_VMALLOC32 GFP_KERNEL
 665 #endif
 666
 667 /**
 668  *      vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
 669  *      @size:          allocation size
 670  *
 671  *      Allocate enough 32bit PA addressable pages to cover @size from the
 672  *      page level allocator and map them into contiguous kernel virtual space.
 673  */
 674 void *vmalloc_32(unsigned long size)
 675 {
 676         return __vmalloc(size, GFP_VMALLOC32, PAGE_KERNEL);
 677 }
 678 EXPORT_SYMBOL(vmalloc_32);
 679
 680 /**
 681  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
 682  *      @size:          allocation size
 683  *
 684  * The resulting memory area is 32bit addressable and zeroed so it can be
 685  * mapped to userspace without leaking data.
 686  */
 687 void *vmalloc_32_user(unsigned long size)
 688 {
 689         struct vm_struct *area;
 690         void *ret;
 691
 692         ret = __vmalloc(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL);
 693         if (ret) {
 694                 write_lock(&vmlist_lock);
 695                 area = __find_vm_area(ret);
 696                 area->flags |= VM_USERMAP;
 697                 write_unlock(&vmlist_lock);
 698         }
 699         return ret;
 700 }
 701 EXPORT_SYMBOL(vmalloc_32_user);
 702
 703 long vread(char *buf, char *addr, unsigned long count)
 704 {
 705         struct vm_struct *tmp;
 706         char *vaddr, *buf_start = buf;
 707         unsigned long n;
 708
 709         /* Don't allow overflow */
 710         if ((unsigned long) addr + count < count)
 711                 count = -(unsigned long) addr;
 712
 713         read_lock(&vmlist_lock);
 714         for (tmp = vmlist; tmp; tmp = tmp->next) {
 715                 vaddr = (char *) tmp->addr;
 716                 if (addr >= vaddr + tmp->size - PAGE_SIZE)
 717                         continue;
 718                 while (addr < vaddr) {
 719                         if (count == 0)
 720                                 goto finished;
 721                         *buf = '\0';
 722                         buf++;
 723                         addr++;
 724                         count--;
 725                 }
 726                 n = vaddr + tmp->size - PAGE_SIZE - addr;
 727                 do {
 728                         if (count == 0)
 729                                 goto finished;
 730                         *buf = *addr;
 731                         buf++;
 732                         addr++;
 733                         count--;
 734                 } while (--n > 0);
 735         }
 736 finished:
 737         read_unlock(&vmlist_lock);
 738         return buf - buf_start;
 739 }
 740
 741 long vwrite(char *buf, char *addr, unsigned long count)
 742 {
 743         struct vm_struct *tmp;
 744         char *vaddr, *buf_start = buf;
 745         unsigned long n;
 746
 747         /* Don't allow overflow */
 748         if ((unsigned long) addr + count < count)
 749                 count = -(unsigned long) addr;
 750
 751         read_lock(&vmlist_lock);
 752         for (tmp = vmlist; tmp; tmp = tmp->next) {
 753                 vaddr = (char *) tmp->addr;
 754                 if (addr >= vaddr + tmp->size - PAGE_SIZE)
 755                         continue;
 756                 while (addr < vaddr) {
 757                         if (count == 0)
 758                                 goto finished;
 759                         buf++;
 760                         addr++;
 761                         count--;
 762                 }
 763                 n = vaddr + tmp->size - PAGE_SIZE - addr;
 764                 do {
 765                         if (count == 0)
 766                                 goto finished;
 767                         *addr = *buf;
 768                         buf++;
 769                         addr++;
 770                         count--;
 771                 } while (--n > 0);
 772         }
 773 finished:
 774         read_unlock(&vmlist_lock);
 775         return buf - buf_start;
 776 }
 777
 778 /**
 779  *      remap_vmalloc_range  -  map vmalloc pages to userspace
 780  *      @vma:           vma to cover (map full range of vma)
 781  *      @addr:          vmalloc memory
 782  *      @pgoff:         number of pages into addr before first page to map
 783  *
 784  *      Returns:        0 for success, -Exxx on failure
 785  *
 786  *      This function checks that addr is a valid vmalloc'ed area, and
 787  *      that it is big enough to cover the vma. Will return failure if
 788  *      that criteria isn't met.
 789  *
 790  *      Similar to remap_pfn_range() (see mm/memory.c)
 791  */
 792 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
 793                                                 unsigned long pgoff)
 794 {
 795         struct vm_struct *area;
 796         unsigned long uaddr = vma->vm_start;
 797         unsigned long usize = vma->vm_end - vma->vm_start;
 798         int ret;
 799
 800         if ((PAGE_SIZE-1) & (unsigned long)addr)
 801                 return -EINVAL;
 802
 803         read_lock(&vmlist_lock);
 804         area = __find_vm_area(addr);
 805         if (!area)
 806                 goto out_einval_locked;
 807
 808         if (!(area->flags & VM_USERMAP))
 809                 goto out_einval_locked;
 810
 811         if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
 812                 goto out_einval_locked;
 813         read_unlock(&vmlist_lock);
 814
 815         addr += pgoff << PAGE_SHIFT;
 816         do {
 817                 struct page *page = vmalloc_to_page(addr);
 818                 ret = vm_insert_page(vma, uaddr, page);
 819                 if (ret)
 820                         return ret;
 821
 822                 uaddr += PAGE_SIZE;
 823                 addr += PAGE_SIZE;
 824                 usize -= PAGE_SIZE;
 825         } while (usize > 0);
 826
 827         /* Prevent "things" like memory migration? VM_flags need a cleanup... */
 828         vma->vm_flags |= VM_RESERVED;
 829
 830         return ret;
 831
 832 out_einval_locked:
 833         read_unlock(&vmlist_lock);
 834         return -EINVAL;
 835 }
 836 EXPORT_SYMBOL(remap_vmalloc_range);
 837
 838 /*
 839  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
 840  * have one.
 841  */
 842 void  __attribute__((weak)) vmalloc_sync_all(void)
 843 {
 844 }
 845
 846
 847 static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
 848 {
 849         /* apply_to_page_range() does all the hard work. */
 850         return 0;
 851 }
 852
 853 /**
 854  *      alloc_vm_area - allocate a range of kernel address space
 855  *      @size:          size of the area
 856  *
 857  *      Returns:        NULL on failure, vm_struct on success
 858  *
 859  *      This function reserves a range of kernel address space, and
 860  *      allocates pagetables to map that range.  No actual mappings
 861  *      are created.  If the kernel address space is not shared
 862  *      between processes, it syncs the pagetable across all
 863  *      processes.
 864  */
 865 struct vm_struct *alloc_vm_area(size_t size)
 866 {
 867         struct vm_struct *area;
 868
 869         area = get_vm_area_caller(size, VM_IOREMAP,
 870                                 __builtin_return_address(0));
 871         if (area == NULL)
 872                 return NULL;
 873
 874         /*
 875          * This ensures that page tables are constructed for this region
 876          * of kernel virtual address space and mapped into init_mm.
 877          */
 878         if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
 879                                 area->size, f, NULL)) {
 880                 free_vm_area(area);
 881                 return NULL;
 882         }
 883
 884         /* Make sure the pagetables are constructed in process kernel
 885            mappings */
 886         vmalloc_sync_all();
 887
 888         return area;
 889 }
 890 EXPORT_SYMBOL_GPL(alloc_vm_area);
 891
 892 void free_vm_area(struct vm_struct *area)
 893 {
 894         struct vm_struct *ret;
 895         ret = remove_vm_area(area->addr);
 896         BUG_ON(ret != area);
 897         kfree(area);
 898 }
 899 EXPORT_SYMBOL_GPL(free_vm_area);
 900
 901
 902 #ifdef CONFIG_PROC_FS
 903 static void *s_start(struct seq_file *m, loff_t *pos)
 904 {
 905         loff_t n = *pos;
 906         struct vm_struct *v;
 907
 908         read_lock(&vmlist_lock);
 909         v = vmlist;
 910         while (n > 0 && v) {
 911                 n--;
 912                 v = v->next;
 913         }
 914         if (!n)
 915                 return v;
 916
 917         return NULL;
 918
 919 }
 920
 921 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
 922 {
 923         struct vm_struct *v = p;
 924
 925         ++*pos;
 926         return v->next;
 927 }
 928
 929 static void s_stop(struct seq_file *m, void *p)
 930 {
 931         read_unlock(&vmlist_lock);
 932 }
 933
 934 static void show_numa_info(struct seq_file *m, struct vm_struct *v)
 935 {
 936         if (NUMA_BUILD) {
 937                 unsigned int nr, *counters = m->private;
 938
 939                 if (!counters)
 940                         return;
 941
 942                 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
 943
 944                 for (nr = 0; nr < v->nr_pages; nr++)
 945                         counters[page_to_nid(v->pages[nr])]++;
 946
 947                 for_each_node_state(nr, N_HIGH_MEMORY)
 948                         if (counters[nr])
 949                                 seq_printf(m, " N%u=%u", nr, counters[nr]);
 950         }
 951 }
 952
 953 static int s_show(struct seq_file *m, void *p)
 954 {
 955         struct vm_struct *v = p;
 956
 957         seq_printf(m, "0x%p-0x%p %7ld",
 958                 v->addr, v->addr + v->size, v->size);
 959
 960         if (v->caller) {
 961                 char buff[2 * KSYM_NAME_LEN];
 962
 963                 seq_putc(m, ' ');
 964                 sprint_symbol(buff, (unsigned long)v->caller);
 965                 seq_puts(m, buff);
 966         }
 967
 968         if (v->nr_pages)
 969                 seq_printf(m, " pages=%d", v->nr_pages);
 970
 971         if (v->phys_addr)
 972                 seq_printf(m, " phys=%lx", v->phys_addr);
 973
 974         if (v->flags & VM_IOREMAP)
 975                 seq_printf(m, " ioremap");
 976
 977         if (v->flags & VM_ALLOC)
 978                 seq_printf(m, " vmalloc");
 979
 980         if (v->flags & VM_MAP)
 981                 seq_printf(m, " vmap");
 982
 983         if (v->flags & VM_USERMAP)
 984                 seq_printf(m, " user");
 985
 986         if (v->flags & VM_VPAGES)
 987                 seq_printf(m, " vpages");
 988
 989         show_numa_info(m, v);
 990         seq_putc(m, '\n');
 991         return 0;
 992 }
 993
 994 const struct seq_operations vmalloc_op = {
 995         .start = s_start,
 996         .next = s_next,
 997         .stop = s_stop,
 998         .show = s_show,
 999 };
1000 #endif
1001