xref: /linux/mm/usercopy.c (revision b687034b1a4d85333ced0fe07f67b17276cccdc8)
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 
13 #include <linux/mm.h>
14 #include <linux/highmem.h>
15 #include <linux/kstrtox.h>
16 #include <linux/slab.h>
17 #include <linux/sched.h>
18 #include <linux/sched/task.h>
19 #include <linux/sched/task_stack.h>
20 #include <linux/ucopysize.h>
21 #include <linux/vmalloc.h>
22 #include <linux/atomic.h>
23 #include <linux/jump_label.h>
24 #include <asm/sections.h>
25 #include "slab.h"
26 
27 /*
28  * Checks if a given pointer and length is contained by the current
29  * stack frame (if possible).
30  *
31  * Returns:
32  *	NOT_STACK: not at all on the stack
33  *	GOOD_FRAME: fully within a valid stack frame
34  *	GOOD_STACK: within the current stack (when can't frame-check exactly)
35  *	BAD_STACK: error condition (invalid stack position or bad stack frame)
36  */
check_stack_object(const void * obj,unsigned long len)37 static noinline int check_stack_object(const void *obj, unsigned long len)
38 {
39 	const void * const stack = task_stack_page(current);
40 	const void * const stackend = stack + THREAD_SIZE;
41 	int ret;
42 
43 	/* Object is not on the stack at all. */
44 	if (obj + len <= stack || stackend <= obj)
45 		return NOT_STACK;
46 
47 	/*
48 	 * Reject: object partially overlaps the stack (passing the
49 	 * check above means at least one end is within the stack,
50 	 * so if this check fails, the other end is outside the stack).
51 	 */
52 	if (obj < stack || stackend < obj + len)
53 		return BAD_STACK;
54 
55 	/* Check if object is safely within a valid frame. */
56 	ret = arch_within_stack_frames(stack, stackend, obj, len);
57 	if (ret)
58 		return ret;
59 
60 	/* Finally, check stack depth if possible. */
61 #ifdef CONFIG_ARCH_HAS_CURRENT_STACK_POINTER
62 	if (IS_ENABLED(CONFIG_STACK_GROWSUP)) {
63 		if ((void *)current_stack_pointer < obj + len)
64 			return BAD_STACK;
65 	} else {
66 		if (obj < (void *)current_stack_pointer)
67 			return BAD_STACK;
68 	}
69 #endif
70 
71 	return GOOD_STACK;
72 }
73 
74 /*
75  * If these functions are reached, then CONFIG_HARDENED_USERCOPY has found
76  * an unexpected state during a copy_from_user() or copy_to_user() call.
77  * There are several checks being performed on the buffer by the
78  * __check_object_size() function. Normal stack buffer usage should never
79  * trip the checks, and kernel text addressing will always trip the check.
80  * For cache objects, it is checking that only the whitelisted range of
81  * bytes for a given cache is being accessed (via the cache's usersize and
82  * useroffset fields). To adjust a cache whitelist, use the usercopy-aware
83  * kmem_cache_create_usercopy() function to create the cache (and
84  * carefully audit the whitelist range).
85  */
usercopy_abort(const char * name,const char * detail,bool to_user,unsigned long offset,unsigned long len)86 void __noreturn usercopy_abort(const char *name, const char *detail,
87 			       bool to_user, unsigned long offset,
88 			       unsigned long len)
89 {
90 	pr_emerg("Kernel memory %s attempt detected %s %s%s%s%s (offset %lu, size %lu)!\n",
91 		 to_user ? "exposure" : "overwrite",
92 		 to_user ? "from" : "to",
93 		 name ? : "unknown?!",
94 		 detail ? " '" : "", detail ? : "", detail ? "'" : "",
95 		 offset, len);
96 
97 	/*
98 	 * For greater effect, it would be nice to do do_group_exit(),
99 	 * but BUG() actually hooks all the lock-breaking and per-arch
100 	 * Oops code, so that is used here instead.
101 	 */
102 	BUG();
103 }
104 
105 /* Returns true if any portion of [ptr,ptr+n) over laps with [low,high). */
overlaps(const unsigned long ptr,unsigned long n,unsigned long low,unsigned long high)106 static bool overlaps(const unsigned long ptr, unsigned long n,
107 		     unsigned long low, unsigned long high)
108 {
109 	const unsigned long check_low = ptr;
110 	unsigned long check_high = check_low + n;
111 
112 	/* Does not overlap if entirely above or entirely below. */
113 	if (check_low >= high || check_high <= low)
114 		return false;
115 
116 	return true;
117 }
118 
119 /* Is this address range in the kernel text area? */
check_kernel_text_object(const unsigned long ptr,unsigned long n,bool to_user)120 static inline void check_kernel_text_object(const unsigned long ptr,
121 					    unsigned long n, bool to_user)
122 {
123 	unsigned long textlow = (unsigned long)_stext;
124 	unsigned long texthigh = (unsigned long)_etext;
125 	unsigned long textlow_linear, texthigh_linear;
126 
127 	if (overlaps(ptr, n, textlow, texthigh))
128 		usercopy_abort("kernel text", NULL, to_user, ptr - textlow, n);
129 
130 	/*
131 	 * Some architectures have virtual memory mappings with a secondary
132 	 * mapping of the kernel text, i.e. there is more than one virtual
133 	 * kernel address that points to the kernel image. It is usually
134 	 * when there is a separate linear physical memory mapping, in that
135 	 * __pa() is not just the reverse of __va(). This can be detected
136 	 * and checked:
137 	 */
138 	textlow_linear = (unsigned long)lm_alias(textlow);
139 	/* No different mapping: we're done. */
140 	if (textlow_linear == textlow)
141 		return;
142 
143 	/* Check the secondary mapping... */
144 	texthigh_linear = (unsigned long)lm_alias(texthigh);
145 	if (overlaps(ptr, n, textlow_linear, texthigh_linear))
146 		usercopy_abort("linear kernel text", NULL, to_user,
147 			       ptr - textlow_linear, n);
148 }
149 
check_bogus_address(const unsigned long ptr,unsigned long n,bool to_user)150 static inline void check_bogus_address(const unsigned long ptr, unsigned long n,
151 				       bool to_user)
152 {
153 	/* Reject if object wraps past end of memory. */
154 	if (ptr + (n - 1) < ptr)
155 		usercopy_abort("wrapped address", NULL, to_user, 0, ptr + n);
156 
157 	/* Reject if NULL or ZERO-allocation. */
158 	if (ZERO_OR_NULL_PTR(ptr))
159 		usercopy_abort("null address", NULL, to_user, ptr, n);
160 }
161 
check_heap_object(const void * ptr,unsigned long n,bool to_user)162 static inline void check_heap_object(const void *ptr, unsigned long n,
163 				     bool to_user)
164 {
165 	unsigned long addr = (unsigned long)ptr;
166 	unsigned long offset;
167 	struct page *page;
168 	struct slab *slab;
169 
170 	if (is_kmap_addr(ptr)) {
171 		offset = offset_in_page(ptr);
172 		if (n > PAGE_SIZE - offset)
173 			usercopy_abort("kmap", NULL, to_user, offset, n);
174 		return;
175 	}
176 
177 	if (is_vmalloc_addr(ptr) && !pagefault_disabled()) {
178 		struct vmap_area *area = find_vmap_area(addr);
179 
180 		if (!area)
181 			usercopy_abort("vmalloc", "no area", to_user, 0, n);
182 
183 		if (n > area->va_end - addr) {
184 			offset = addr - area->va_start;
185 			usercopy_abort("vmalloc", NULL, to_user, offset, n);
186 		}
187 		return;
188 	}
189 
190 	if (!virt_addr_valid(ptr))
191 		return;
192 
193 	page = virt_to_page(ptr);
194 	slab = page_slab(page);
195 	if (slab) {
196 		/* Check slab allocator for flags and size. */
197 		__check_heap_object(ptr, n, slab, to_user);
198 	} else if (PageCompound(page)) {
199 		page = compound_head(page);
200 		offset = ptr - page_address(page);
201 		if (n > page_size(page) - offset)
202 			usercopy_abort("page alloc", NULL, to_user, offset, n);
203 	}
204 
205 	/*
206 	 * We cannot check non-compound pages.  They might be part of
207 	 * a large allocation, in which case crossing a page boundary
208 	 * is fine.
209 	 */
210 }
211 
212 DEFINE_STATIC_KEY_MAYBE_RO(CONFIG_HARDENED_USERCOPY_DEFAULT_ON,
213 			   validate_usercopy_range);
214 EXPORT_SYMBOL(validate_usercopy_range);
215 
216 /*
217  * Validates that the given object is:
218  * - not bogus address
219  * - fully contained by stack (or stack frame, when available)
220  * - fully within SLAB object (or object whitelist area, when available)
221  * - not in kernel text
222  */
__check_object_size(const void * ptr,unsigned long n,bool to_user)223 void __check_object_size(const void *ptr, unsigned long n, bool to_user)
224 {
225 	/* Skip all tests if size is zero. */
226 	if (!n)
227 		return;
228 
229 	/* Check for invalid addresses. */
230 	check_bogus_address((const unsigned long)ptr, n, to_user);
231 
232 	/* Check for bad stack object. */
233 	switch (check_stack_object(ptr, n)) {
234 	case NOT_STACK:
235 		/* Object is not touching the current process stack. */
236 		break;
237 	case GOOD_FRAME:
238 	case GOOD_STACK:
239 		/*
240 		 * Object is either in the correct frame (when it
241 		 * is possible to check) or just generally on the
242 		 * process stack (when frame checking not available).
243 		 */
244 		return;
245 	default:
246 		usercopy_abort("process stack", NULL, to_user,
247 #ifdef CONFIG_ARCH_HAS_CURRENT_STACK_POINTER
248 			IS_ENABLED(CONFIG_STACK_GROWSUP) ?
249 				ptr - (void *)current_stack_pointer :
250 				(void *)current_stack_pointer - ptr,
251 #else
252 			0,
253 #endif
254 			n);
255 	}
256 
257 	/* Check for bad heap object. */
258 	check_heap_object(ptr, n, to_user);
259 
260 	/* Check for object in kernel to avoid text exposure. */
261 	check_kernel_text_object((const unsigned long)ptr, n, to_user);
262 }
263 EXPORT_SYMBOL(__check_object_size);
264 
265 static bool enable_checks __initdata =
266 		IS_ENABLED(CONFIG_HARDENED_USERCOPY_DEFAULT_ON);
267 
parse_hardened_usercopy(char * str)268 static int __init parse_hardened_usercopy(char *str)
269 {
270 	if (kstrtobool(str, &enable_checks))
271 		pr_warn("Invalid option string for hardened_usercopy: '%s'\n",
272 			str);
273 	return 1;
274 }
275 
276 __setup("hardened_usercopy=", parse_hardened_usercopy);
277 
set_hardened_usercopy(void)278 static int __init set_hardened_usercopy(void)
279 {
280 	if (enable_checks)
281 		static_branch_enable(&validate_usercopy_range);
282 	else
283 		static_branch_disable(&validate_usercopy_range);
284 	return 1;
285 }
286 
287 late_initcall(set_hardened_usercopy);
288