usercopy.c 8.1 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277
  1. // SPDX-License-Identifier: GPL-2.0-only
  2. /*
  3. * This implements the various checks for CONFIG_HARDENED_USERCOPY*,
  4. * which are designed to protect kernel memory from needless exposure
  5. * and overwrite under many unintended conditions. This code is based
  6. * on PAX_USERCOPY, which is:
  7. *
  8. * Copyright (C) 2001-2016 PaX Team, Bradley Spengler, Open Source
  9. * Security Inc.
  10. */
  11. #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12. #include <linux/mm.h>
  13. #include <linux/highmem.h>
  14. #include <linux/kstrtox.h>
  15. #include <linux/slab.h>
  16. #include <linux/sched.h>
  17. #include <linux/sched/task.h>
  18. #include <linux/sched/task_stack.h>
  19. #include <linux/thread_info.h>
  20. #include <linux/vmalloc.h>
  21. #include <linux/atomic.h>
  22. #include <linux/jump_label.h>
  23. #include <asm/sections.h>
  24. #include "slab.h"
  25. /*
  26. * Checks if a given pointer and length is contained by the current
  27. * stack frame (if possible).
  28. *
  29. * Returns:
  30. * NOT_STACK: not at all on the stack
  31. * GOOD_FRAME: fully within a valid stack frame
  32. * GOOD_STACK: within the current stack (when can't frame-check exactly)
  33. * BAD_STACK: error condition (invalid stack position or bad stack frame)
  34. */
  35. static noinline int check_stack_object(const void *obj, unsigned long len)
  36. {
  37. const void * const stack = task_stack_page(current);
  38. const void * const stackend = stack + THREAD_SIZE;
  39. int ret;
  40. /* Object is not on the stack at all. */
  41. if (obj + len <= stack || stackend <= obj)
  42. return NOT_STACK;
  43. /*
  44. * Reject: object partially overlaps the stack (passing the
  45. * check above means at least one end is within the stack,
  46. * so if this check fails, the other end is outside the stack).
  47. */
  48. if (obj < stack || stackend < obj + len)
  49. return BAD_STACK;
  50. /* Check if object is safely within a valid frame. */
  51. ret = arch_within_stack_frames(stack, stackend, obj, len);
  52. if (ret)
  53. return ret;
  54. /* Finally, check stack depth if possible. */
  55. #ifdef CONFIG_ARCH_HAS_CURRENT_STACK_POINTER
  56. if (IS_ENABLED(CONFIG_STACK_GROWSUP)) {
  57. if ((void *)current_stack_pointer < obj + len)
  58. return BAD_STACK;
  59. } else {
  60. if (obj < (void *)current_stack_pointer)
  61. return BAD_STACK;
  62. }
  63. #endif
  64. return GOOD_STACK;
  65. }
  66. /*
  67. * If these functions are reached, then CONFIG_HARDENED_USERCOPY has found
  68. * an unexpected state during a copy_from_user() or copy_to_user() call.
  69. * There are several checks being performed on the buffer by the
  70. * __check_object_size() function. Normal stack buffer usage should never
  71. * trip the checks, and kernel text addressing will always trip the check.
  72. * For cache objects, it is checking that only the whitelisted range of
  73. * bytes for a given cache is being accessed (via the cache's usersize and
  74. * useroffset fields). To adjust a cache whitelist, use the usercopy-aware
  75. * kmem_cache_create_usercopy() function to create the cache (and
  76. * carefully audit the whitelist range).
  77. */
  78. void __noreturn usercopy_abort(const char *name, const char *detail,
  79. bool to_user, unsigned long offset,
  80. unsigned long len)
  81. {
  82. pr_emerg("Kernel memory %s attempt detected %s %s%s%s%s (offset %lu, size %lu)!\n",
  83. to_user ? "exposure" : "overwrite",
  84. to_user ? "from" : "to",
  85. name ? : "unknown?!",
  86. detail ? " '" : "", detail ? : "", detail ? "'" : "",
  87. offset, len);
  88. /*
  89. * For greater effect, it would be nice to do do_group_exit(),
  90. * but BUG() actually hooks all the lock-breaking and per-arch
  91. * Oops code, so that is used here instead.
  92. */
  93. BUG();
  94. }
  95. /* Returns true if any portion of [ptr,ptr+n) over laps with [low,high). */
  96. static bool overlaps(const unsigned long ptr, unsigned long n,
  97. unsigned long low, unsigned long high)
  98. {
  99. const unsigned long check_low = ptr;
  100. unsigned long check_high = check_low + n;
  101. /* Does not overlap if entirely above or entirely below. */
  102. if (check_low >= high || check_high <= low)
  103. return false;
  104. return true;
  105. }
  106. /* Is this address range in the kernel text area? */
  107. static inline void check_kernel_text_object(const unsigned long ptr,
  108. unsigned long n, bool to_user)
  109. {
  110. unsigned long textlow = (unsigned long)_stext;
  111. unsigned long texthigh = (unsigned long)_etext;
  112. unsigned long textlow_linear, texthigh_linear;
  113. if (overlaps(ptr, n, textlow, texthigh))
  114. usercopy_abort("kernel text", NULL, to_user, ptr - textlow, n);
  115. /*
  116. * Some architectures have virtual memory mappings with a secondary
  117. * mapping of the kernel text, i.e. there is more than one virtual
  118. * kernel address that points to the kernel image. It is usually
  119. * when there is a separate linear physical memory mapping, in that
  120. * __pa() is not just the reverse of __va(). This can be detected
  121. * and checked:
  122. */
  123. textlow_linear = (unsigned long)lm_alias(textlow);
  124. /* No different mapping: we're done. */
  125. if (textlow_linear == textlow)
  126. return;
  127. /* Check the secondary mapping... */
  128. texthigh_linear = (unsigned long)lm_alias(texthigh);
  129. if (overlaps(ptr, n, textlow_linear, texthigh_linear))
  130. usercopy_abort("linear kernel text", NULL, to_user,
  131. ptr - textlow_linear, n);
  132. }
  133. static inline void check_bogus_address(const unsigned long ptr, unsigned long n,
  134. bool to_user)
  135. {
  136. /* Reject if object wraps past end of memory. */
  137. if (ptr + (n - 1) < ptr)
  138. usercopy_abort("wrapped address", NULL, to_user, 0, ptr + n);
  139. /* Reject if NULL or ZERO-allocation. */
  140. if (ZERO_OR_NULL_PTR(ptr))
  141. usercopy_abort("null address", NULL, to_user, ptr, n);
  142. }
  143. static inline void check_heap_object(const void *ptr, unsigned long n,
  144. bool to_user)
  145. {
  146. unsigned long addr = (unsigned long)ptr;
  147. unsigned long offset;
  148. struct folio *folio;
  149. if (is_kmap_addr(ptr)) {
  150. offset = offset_in_page(ptr);
  151. if (n > PAGE_SIZE - offset)
  152. usercopy_abort("kmap", NULL, to_user, offset, n);
  153. return;
  154. }
  155. if (is_vmalloc_addr(ptr) && !pagefault_disabled()) {
  156. struct vmap_area *area = find_vmap_area(addr);
  157. if (!area)
  158. usercopy_abort("vmalloc", "no area", to_user, 0, n);
  159. if (n > area->va_end - addr) {
  160. offset = addr - area->va_start;
  161. usercopy_abort("vmalloc", NULL, to_user, offset, n);
  162. }
  163. return;
  164. }
  165. if (!virt_addr_valid(ptr))
  166. return;
  167. folio = virt_to_folio(ptr);
  168. if (folio_test_slab(folio)) {
  169. /* Check slab allocator for flags and size. */
  170. __check_heap_object(ptr, n, folio_slab(folio), to_user);
  171. } else if (folio_test_large(folio)) {
  172. offset = ptr - folio_address(folio);
  173. if (n > folio_size(folio) - offset)
  174. usercopy_abort("page alloc", NULL, to_user, offset, n);
  175. }
  176. }
  177. static DEFINE_STATIC_KEY_FALSE_RO(bypass_usercopy_checks);
  178. /*
  179. * Validates that the given object is:
  180. * - not bogus address
  181. * - fully contained by stack (or stack frame, when available)
  182. * - fully within SLAB object (or object whitelist area, when available)
  183. * - not in kernel text
  184. */
  185. void __check_object_size(const void *ptr, unsigned long n, bool to_user)
  186. {
  187. if (static_branch_unlikely(&bypass_usercopy_checks))
  188. return;
  189. /* Skip all tests if size is zero. */
  190. if (!n)
  191. return;
  192. /* Check for invalid addresses. */
  193. check_bogus_address((const unsigned long)ptr, n, to_user);
  194. /* Check for bad stack object. */
  195. switch (check_stack_object(ptr, n)) {
  196. case NOT_STACK:
  197. /* Object is not touching the current process stack. */
  198. break;
  199. case GOOD_FRAME:
  200. case GOOD_STACK:
  201. /*
  202. * Object is either in the correct frame (when it
  203. * is possible to check) or just generally on the
  204. * process stack (when frame checking not available).
  205. */
  206. return;
  207. default:
  208. usercopy_abort("process stack", NULL, to_user,
  209. #ifdef CONFIG_ARCH_HAS_CURRENT_STACK_POINTER
  210. IS_ENABLED(CONFIG_STACK_GROWSUP) ?
  211. ptr - (void *)current_stack_pointer :
  212. (void *)current_stack_pointer - ptr,
  213. #else
  214. 0,
  215. #endif
  216. n);
  217. }
  218. /* Check for bad heap object. */
  219. check_heap_object(ptr, n, to_user);
  220. /* Check for object in kernel to avoid text exposure. */
  221. check_kernel_text_object((const unsigned long)ptr, n, to_user);
  222. }
  223. EXPORT_SYMBOL(__check_object_size);
  224. static bool enable_checks __initdata = true;
  225. static int __init parse_hardened_usercopy(char *str)
  226. {
  227. if (kstrtobool(str, &enable_checks))
  228. pr_warn("Invalid option string for hardened_usercopy: '%s'\n",
  229. str);
  230. return 1;
  231. }
  232. __setup("hardened_usercopy=", parse_hardened_usercopy);
  233. static int __init set_hardened_usercopy(void)
  234. {
  235. if (enable_checks == false)
  236. static_branch_enable(&bypass_usercopy_checks);
  237. return 1;
  238. }
  239. late_initcall(set_hardened_usercopy);