mmap.c 7.0 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250
  1. // SPDX-License-Identifier: GPL-2.0-or-later
  2. /*
  3. * Flexible mmap layout support
  4. *
  5. * Based on code by Ingo Molnar and Andi Kleen, copyrighted
  6. * as follows:
  7. *
  8. * Copyright 2003-2009 Red Hat Inc.
  9. * All Rights Reserved.
  10. * Copyright 2005 Andi Kleen, SUSE Labs.
  11. * Copyright 2007 Jiri Kosina, SUSE Labs.
  12. */
  13. #include <linux/personality.h>
  14. #include <linux/mm.h>
  15. #include <linux/random.h>
  16. #include <linux/limits.h>
  17. #include <linux/sched/signal.h>
  18. #include <linux/sched/mm.h>
  19. #include <linux/compat.h>
  20. #include <linux/elf-randomize.h>
  21. #include <asm/elf.h>
  22. #include <asm/io.h>
  23. #include "physaddr.h"
  24. struct va_alignment __read_mostly va_align = {
  25. .flags = -1,
  26. };
  27. unsigned long task_size_32bit(void)
  28. {
  29. return IA32_PAGE_OFFSET;
  30. }
  31. unsigned long task_size_64bit(int full_addr_space)
  32. {
  33. return full_addr_space ? TASK_SIZE_MAX : DEFAULT_MAP_WINDOW;
  34. }
  35. static unsigned long stack_maxrandom_size(unsigned long task_size)
  36. {
  37. unsigned long max = 0;
  38. if (current->flags & PF_RANDOMIZE) {
  39. max = (-1UL) & __STACK_RND_MASK(task_size == task_size_32bit());
  40. max <<= PAGE_SHIFT;
  41. }
  42. return max;
  43. }
  44. #ifdef CONFIG_COMPAT
  45. # define mmap32_rnd_bits mmap_rnd_compat_bits
  46. # define mmap64_rnd_bits mmap_rnd_bits
  47. #else
  48. # define mmap32_rnd_bits mmap_rnd_bits
  49. # define mmap64_rnd_bits mmap_rnd_bits
  50. #endif
  51. #define SIZE_128M (128 * 1024 * 1024UL)
  52. static int mmap_is_legacy(void)
  53. {
  54. if (current->personality & ADDR_COMPAT_LAYOUT)
  55. return 1;
  56. return sysctl_legacy_va_layout;
  57. }
  58. static unsigned long arch_rnd(unsigned int rndbits)
  59. {
  60. if (!(current->flags & PF_RANDOMIZE))
  61. return 0;
  62. return (get_random_long() & ((1UL << rndbits) - 1)) << PAGE_SHIFT;
  63. }
  64. unsigned long arch_mmap_rnd(void)
  65. {
  66. return arch_rnd(mmap_is_ia32() ? mmap32_rnd_bits : mmap64_rnd_bits);
  67. }
  68. static unsigned long mmap_base(unsigned long rnd, unsigned long task_size,
  69. struct rlimit *rlim_stack)
  70. {
  71. unsigned long gap = rlim_stack->rlim_cur;
  72. unsigned long pad = stack_maxrandom_size(task_size) + stack_guard_gap;
  73. unsigned long gap_min, gap_max;
  74. /* Values close to RLIM_INFINITY can overflow. */
  75. if (gap + pad > gap)
  76. gap += pad;
  77. /*
  78. * Top of mmap area (just below the process stack).
  79. * Leave an at least ~128 MB hole with possible stack randomization.
  80. */
  81. gap_min = SIZE_128M;
  82. gap_max = (task_size / 6) * 5;
  83. if (gap < gap_min)
  84. gap = gap_min;
  85. else if (gap > gap_max)
  86. gap = gap_max;
  87. return PAGE_ALIGN(task_size - gap - rnd);
  88. }
  89. static unsigned long mmap_legacy_base(unsigned long rnd,
  90. unsigned long task_size)
  91. {
  92. return __TASK_UNMAPPED_BASE(task_size) + rnd;
  93. }
  94. /*
  95. * This function, called very early during the creation of a new
  96. * process VM image, sets up which VM layout function to use:
  97. */
  98. static void arch_pick_mmap_base(unsigned long *base, unsigned long *legacy_base,
  99. unsigned long random_factor, unsigned long task_size,
  100. struct rlimit *rlim_stack)
  101. {
  102. *legacy_base = mmap_legacy_base(random_factor, task_size);
  103. if (mmap_is_legacy())
  104. *base = *legacy_base;
  105. else
  106. *base = mmap_base(random_factor, task_size, rlim_stack);
  107. }
  108. void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
  109. {
  110. if (mmap_is_legacy())
  111. mm->get_unmapped_area = arch_get_unmapped_area;
  112. else
  113. mm->get_unmapped_area = arch_get_unmapped_area_topdown;
  114. arch_pick_mmap_base(&mm->mmap_base, &mm->mmap_legacy_base,
  115. arch_rnd(mmap64_rnd_bits), task_size_64bit(0),
  116. rlim_stack);
  117. #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
  118. /*
  119. * The mmap syscall mapping base decision depends solely on the
  120. * syscall type (64-bit or compat). This applies for 64bit
  121. * applications and 32bit applications. The 64bit syscall uses
  122. * mmap_base, the compat syscall uses mmap_compat_base.
  123. */
  124. arch_pick_mmap_base(&mm->mmap_compat_base, &mm->mmap_compat_legacy_base,
  125. arch_rnd(mmap32_rnd_bits), task_size_32bit(),
  126. rlim_stack);
  127. #endif
  128. }
  129. unsigned long get_mmap_base(int is_legacy)
  130. {
  131. struct mm_struct *mm = current->mm;
  132. #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
  133. if (in_32bit_syscall()) {
  134. return is_legacy ? mm->mmap_compat_legacy_base
  135. : mm->mmap_compat_base;
  136. }
  137. #endif
  138. return is_legacy ? mm->mmap_legacy_base : mm->mmap_base;
  139. }
  140. const char *arch_vma_name(struct vm_area_struct *vma)
  141. {
  142. return NULL;
  143. }
  144. /**
  145. * mmap_address_hint_valid - Validate the address hint of mmap
  146. * @addr: Address hint
  147. * @len: Mapping length
  148. *
  149. * Check whether @addr and @addr + @len result in a valid mapping.
  150. *
  151. * On 32bit this only checks whether @addr + @len is <= TASK_SIZE.
  152. *
  153. * On 64bit with 5-level page tables another sanity check is required
  154. * because mappings requested by mmap(@addr, 0) which cross the 47-bit
  155. * virtual address boundary can cause the following theoretical issue:
  156. *
  157. * An application calls mmap(addr, 0), i.e. without MAP_FIXED, where @addr
  158. * is below the border of the 47-bit address space and @addr + @len is
  159. * above the border.
  160. *
  161. * With 4-level paging this request succeeds, but the resulting mapping
  162. * address will always be within the 47-bit virtual address space, because
  163. * the hint address does not result in a valid mapping and is
  164. * ignored. Hence applications which are not prepared to handle virtual
  165. * addresses above 47-bit work correctly.
  166. *
  167. * With 5-level paging this request would be granted and result in a
  168. * mapping which crosses the border of the 47-bit virtual address
  169. * space. If the application cannot handle addresses above 47-bit this
  170. * will lead to misbehaviour and hard to diagnose failures.
  171. *
  172. * Therefore ignore address hints which would result in a mapping crossing
  173. * the 47-bit virtual address boundary.
  174. *
  175. * Note, that in the same scenario with MAP_FIXED the behaviour is
  176. * different. The request with @addr < 47-bit and @addr + @len > 47-bit
  177. * fails on a 4-level paging machine but succeeds on a 5-level paging
  178. * machine. It is reasonable to expect that an application does not rely on
  179. * the failure of such a fixed mapping request, so the restriction is not
  180. * applied.
  181. */
  182. bool mmap_address_hint_valid(unsigned long addr, unsigned long len)
  183. {
  184. if (TASK_SIZE - len < addr)
  185. return false;
  186. return (addr > DEFAULT_MAP_WINDOW) == (addr + len > DEFAULT_MAP_WINDOW);
  187. }
  188. /* Can we access it for direct reading/writing? Must be RAM: */
  189. int valid_phys_addr_range(phys_addr_t addr, size_t count)
  190. {
  191. return addr + count - 1 <= __pa(high_memory - 1);
  192. }
  193. /* Can we access it through mmap? Must be a valid physical address: */
  194. int valid_mmap_phys_addr_range(unsigned long pfn, size_t count)
  195. {
  196. phys_addr_t addr = (phys_addr_t)pfn << PAGE_SHIFT;
  197. return phys_addr_valid(addr + count - 1);
  198. }
  199. /*
  200. * Only allow root to set high MMIO mappings to PROT_NONE.
  201. * This prevents an unpriv. user to set them to PROT_NONE and invert
  202. * them, then pointing to valid memory for L1TF speculation.
  203. *
  204. * Note: for locked down kernels may want to disable the root override.
  205. */
  206. bool pfn_modify_allowed(unsigned long pfn, pgprot_t prot)
  207. {
  208. if (!boot_cpu_has_bug(X86_BUG_L1TF))
  209. return true;
  210. if (!__pte_needs_invert(pgprot_val(prot)))
  211. return true;
  212. /* If it's real memory always allow */
  213. if (pfn_valid(pfn))
  214. return true;
  215. if (pfn >= l1tf_pfn_limit() && !capable(CAP_SYS_ADMIN))
  216. return false;
  217. return true;
  218. }