• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /* SPDX-License-Identifier: GPL-2.0 OR MIT */
2 #ifndef __LINUX_OVERFLOW_H
3 #define __LINUX_OVERFLOW_H
4 
5 #include <linux/compiler.h>
6 #include <linux/limits.h>
7 
8 /*
9  * In the fallback code below, we need to compute the minimum and
10  * maximum values representable in a given type. These macros may also
11  * be useful elsewhere, so we provide them outside the
12  * COMPILER_HAS_GENERIC_BUILTIN_OVERFLOW block.
13  *
14  * It would seem more obvious to do something like
15  *
16  * #define type_min(T) (T)(is_signed_type(T) ? (T)1 << (8*sizeof(T)-1) : 0)
17  * #define type_max(T) (T)(is_signed_type(T) ? ((T)1 << (8*sizeof(T)-1)) - 1 : ~(T)0)
18  *
19  * Unfortunately, the middle expressions, strictly speaking, have
20  * undefined behaviour, and at least some versions of gcc warn about
21  * the type_max expression (but not if -fsanitize=undefined is in
22  * effect; in that case, the warning is deferred to runtime...).
23  *
24  * The slightly excessive casting in type_min is to make sure the
25  * macros also produce sensible values for the exotic type _Bool. [The
26  * overflow checkers only almost work for _Bool, but that's
27  * a-feature-not-a-bug, since people shouldn't be doing arithmetic on
28  * _Bools. Besides, the gcc builtins don't allow _Bool* as third
29  * argument.]
30  *
31  * Idea stolen from
32  * https://mail-index.netbsd.org/tech-misc/2007/02/05/0000.html -
33  * credit to Christian Biere.
34  */
35 #define is_signed_type(type)       (((type)(-1)) < (type)1)
36 #define __type_half_max(type) ((type)1 << (8*sizeof(type) - 1 - is_signed_type(type)))
37 #define type_max(T) ((T)((__type_half_max(T) - 1) + __type_half_max(T)))
38 #define type_min(T) ((T)((T)-type_max(T)-(T)1))
39 
40 /*
41  * Avoids triggering -Wtype-limits compilation warning,
42  * while using unsigned data types to check a < 0.
43  */
44 #define is_non_negative(a) ((a) > 0 || (a) == 0)
45 #define is_negative(a) (!(is_non_negative(a)))
46 
47 /*
48  * Allows for effectively applying __must_check to a macro so we can have
49  * both the type-agnostic benefits of the macros while also being able to
50  * enforce that the return value is, in fact, checked.
51  */
__must_check_overflow(bool overflow)52 static inline bool __must_check __must_check_overflow(bool overflow)
53 {
54 	return unlikely(overflow);
55 }
56 
57 #ifdef COMPILER_HAS_GENERIC_BUILTIN_OVERFLOW
58 /*
59  * For simplicity and code hygiene, the fallback code below insists on
60  * a, b and *d having the same type (similar to the min() and max()
61  * macros), whereas gcc's type-generic overflow checkers accept
62  * different types. Hence we don't just make check_add_overflow an
63  * alias for __builtin_add_overflow, but add type checks similar to
64  * below.
65  */
66 #define check_add_overflow(a, b, d) __must_check_overflow(({	\
67 	typeof(a) __a = (a);			\
68 	typeof(b) __b = (b);			\
69 	typeof(d) __d = (d);			\
70 	(void) (&__a == &__b);			\
71 	(void) (&__a == __d);			\
72 	__builtin_add_overflow(__a, __b, __d);	\
73 }))
74 
75 #define check_sub_overflow(a, b, d) __must_check_overflow(({	\
76 	typeof(a) __a = (a);			\
77 	typeof(b) __b = (b);			\
78 	typeof(d) __d = (d);			\
79 	(void) (&__a == &__b);			\
80 	(void) (&__a == __d);			\
81 	__builtin_sub_overflow(__a, __b, __d);	\
82 }))
83 
84 #define check_mul_overflow(a, b, d) __must_check_overflow(({	\
85 	typeof(a) __a = (a);			\
86 	typeof(b) __b = (b);			\
87 	typeof(d) __d = (d);			\
88 	(void) (&__a == &__b);			\
89 	(void) (&__a == __d);			\
90 	__builtin_mul_overflow(__a, __b, __d);	\
91 }))
92 
93 #else
94 
95 
96 /* Checking for unsigned overflow is relatively easy without causing UB. */
97 #define __unsigned_add_overflow(a, b, d) ({	\
98 	typeof(a) __a = (a);			\
99 	typeof(b) __b = (b);			\
100 	typeof(d) __d = (d);			\
101 	(void) (&__a == &__b);			\
102 	(void) (&__a == __d);			\
103 	*__d = __a + __b;			\
104 	*__d < __a;				\
105 })
106 #define __unsigned_sub_overflow(a, b, d) ({	\
107 	typeof(a) __a = (a);			\
108 	typeof(b) __b = (b);			\
109 	typeof(d) __d = (d);			\
110 	(void) (&__a == &__b);			\
111 	(void) (&__a == __d);			\
112 	*__d = __a - __b;			\
113 	__a < __b;				\
114 })
115 /*
116  * If one of a or b is a compile-time constant, this avoids a division.
117  */
118 #define __unsigned_mul_overflow(a, b, d) ({		\
119 	typeof(a) __a = (a);				\
120 	typeof(b) __b = (b);				\
121 	typeof(d) __d = (d);				\
122 	(void) (&__a == &__b);				\
123 	(void) (&__a == __d);				\
124 	*__d = __a * __b;				\
125 	__builtin_constant_p(__b) ?			\
126 	  __b > 0 && __a > type_max(typeof(__a)) / __b : \
127 	  __a > 0 && __b > type_max(typeof(__b)) / __a;	 \
128 })
129 
130 /*
131  * For signed types, detecting overflow is much harder, especially if
132  * we want to avoid UB. But the interface of these macros is such that
133  * we must provide a result in *d, and in fact we must produce the
134  * result promised by gcc's builtins, which is simply the possibly
135  * wrapped-around value. Fortunately, we can just formally do the
136  * operations in the widest relevant unsigned type (u64) and then
137  * truncate the result - gcc is smart enough to generate the same code
138  * with and without the (u64) casts.
139  */
140 
141 /*
142  * Adding two signed integers can overflow only if they have the same
143  * sign, and overflow has happened iff the result has the opposite
144  * sign.
145  */
146 #define __signed_add_overflow(a, b, d) ({	\
147 	typeof(a) __a = (a);			\
148 	typeof(b) __b = (b);			\
149 	typeof(d) __d = (d);			\
150 	(void) (&__a == &__b);			\
151 	(void) (&__a == __d);			\
152 	*__d = (u64)__a + (u64)__b;		\
153 	(((~(__a ^ __b)) & (*__d ^ __a))	\
154 		& type_min(typeof(__a))) != 0;	\
155 })
156 
157 /*
158  * Subtraction is similar, except that overflow can now happen only
159  * when the signs are opposite. In this case, overflow has happened if
160  * the result has the opposite sign of a.
161  */
162 #define __signed_sub_overflow(a, b, d) ({	\
163 	typeof(a) __a = (a);			\
164 	typeof(b) __b = (b);			\
165 	typeof(d) __d = (d);			\
166 	(void) (&__a == &__b);			\
167 	(void) (&__a == __d);			\
168 	*__d = (u64)__a - (u64)__b;		\
169 	((((__a ^ __b)) & (*__d ^ __a))		\
170 		& type_min(typeof(__a))) != 0;	\
171 })
172 
173 /*
174  * Signed multiplication is rather hard. gcc always follows C99, so
175  * division is truncated towards 0. This means that we can write the
176  * overflow check like this:
177  *
178  * (a > 0 && (b > MAX/a || b < MIN/a)) ||
179  * (a < -1 && (b > MIN/a || b < MAX/a) ||
180  * (a == -1 && b == MIN)
181  *
182  * The redundant casts of -1 are to silence an annoying -Wtype-limits
183  * (included in -Wextra) warning: When the type is u8 or u16, the
184  * __b_c_e in check_mul_overflow obviously selects
185  * __unsigned_mul_overflow, but unfortunately gcc still parses this
186  * code and warns about the limited range of __b.
187  */
188 
189 #define __signed_mul_overflow(a, b, d) ({				\
190 	typeof(a) __a = (a);						\
191 	typeof(b) __b = (b);						\
192 	typeof(d) __d = (d);						\
193 	typeof(a) __tmax = type_max(typeof(a));				\
194 	typeof(a) __tmin = type_min(typeof(a));				\
195 	(void) (&__a == &__b);						\
196 	(void) (&__a == __d);						\
197 	*__d = (u64)__a * (u64)__b;					\
198 	(__b > 0   && (__a > __tmax/__b || __a < __tmin/__b)) ||	\
199 	(__b < (typeof(__b))-1  && (__a > __tmin/__b || __a < __tmax/__b)) || \
200 	(__b == (typeof(__b))-1 && __a == __tmin);			\
201 })
202 
203 
204 #define check_add_overflow(a, b, d)	__must_check_overflow(		\
205 	__builtin_choose_expr(is_signed_type(typeof(a)),		\
206 			__signed_add_overflow(a, b, d),			\
207 			__unsigned_add_overflow(a, b, d)))
208 
209 #define check_sub_overflow(a, b, d)	__must_check_overflow(		\
210 	__builtin_choose_expr(is_signed_type(typeof(a)),		\
211 			__signed_sub_overflow(a, b, d),			\
212 			__unsigned_sub_overflow(a, b, d)))
213 
214 #define check_mul_overflow(a, b, d)	__must_check_overflow(		\
215 	__builtin_choose_expr(is_signed_type(typeof(a)),		\
216 			__signed_mul_overflow(a, b, d),			\
217 			__unsigned_mul_overflow(a, b, d)))
218 
219 #endif /* COMPILER_HAS_GENERIC_BUILTIN_OVERFLOW */
220 
221 /** check_shl_overflow() - Calculate a left-shifted value and check overflow
222  *
223  * @a: Value to be shifted
224  * @s: How many bits left to shift
225  * @d: Pointer to where to store the result
226  *
227  * Computes *@d = (@a << @s)
228  *
229  * Returns true if '*d' cannot hold the result or when 'a << s' doesn't
230  * make sense. Example conditions:
231  * - 'a << s' causes bits to be lost when stored in *d.
232  * - 's' is garbage (e.g. negative) or so large that the result of
233  *   'a << s' is guaranteed to be 0.
234  * - 'a' is negative.
235  * - 'a << s' sets the sign bit, if any, in '*d'.
236  *
237  * '*d' will hold the results of the attempted shift, but is not
238  * considered "safe for use" if false is returned.
239  */
240 #define check_shl_overflow(a, s, d) __must_check_overflow(({		\
241 	typeof(a) _a = a;						\
242 	typeof(s) _s = s;						\
243 	typeof(d) _d = d;						\
244 	u64 _a_full = _a;						\
245 	unsigned int _to_shift =					\
246 		is_non_negative(_s) && _s < 8 * sizeof(*d) ? _s : 0;	\
247 	*_d = (_a_full << _to_shift);					\
248 	(_to_shift != _s || is_negative(*_d) || is_negative(_a) ||	\
249 	(*_d >> _to_shift) != _a);					\
250 }))
251 
252 /**
253  * array_size() - Calculate size of 2-dimensional array.
254  *
255  * @a: dimension one
256  * @b: dimension two
257  *
258  * Calculates size of 2-dimensional array: @a * @b.
259  *
260  * Returns: number of bytes needed to represent the array or SIZE_MAX on
261  * overflow.
262  */
array_size(size_t a,size_t b)263 static inline __must_check size_t array_size(size_t a, size_t b)
264 {
265 	size_t bytes;
266 
267 	if (check_mul_overflow(a, b, &bytes))
268 		return SIZE_MAX;
269 
270 	return bytes;
271 }
272 
273 /**
274  * array3_size() - Calculate size of 3-dimensional array.
275  *
276  * @a: dimension one
277  * @b: dimension two
278  * @c: dimension three
279  *
280  * Calculates size of 3-dimensional array: @a * @b * @c.
281  *
282  * Returns: number of bytes needed to represent the array or SIZE_MAX on
283  * overflow.
284  */
array3_size(size_t a,size_t b,size_t c)285 static inline __must_check size_t array3_size(size_t a, size_t b, size_t c)
286 {
287 	size_t bytes;
288 
289 	if (check_mul_overflow(a, b, &bytes))
290 		return SIZE_MAX;
291 	if (check_mul_overflow(bytes, c, &bytes))
292 		return SIZE_MAX;
293 
294 	return bytes;
295 }
296 
297 /*
298  * Compute a*b+c, returning SIZE_MAX on overflow. Internal helper for
299  * struct_size() below.
300  */
__ab_c_size(size_t a,size_t b,size_t c)301 static inline __must_check size_t __ab_c_size(size_t a, size_t b, size_t c)
302 {
303 	size_t bytes;
304 
305 	if (check_mul_overflow(a, b, &bytes))
306 		return SIZE_MAX;
307 	if (check_add_overflow(bytes, c, &bytes))
308 		return SIZE_MAX;
309 
310 	return bytes;
311 }
312 
313 /**
314  * struct_size() - Calculate size of structure with trailing array.
315  * @p: Pointer to the structure.
316  * @member: Name of the array member.
317  * @count: Number of elements in the array.
318  *
319  * Calculates size of memory needed for structure @p followed by an
320  * array of @count number of @member elements.
321  *
322  * Return: number of bytes needed or SIZE_MAX on overflow.
323  */
324 #define struct_size(p, member, count)					\
325 	__ab_c_size(count,						\
326 		    sizeof(*(p)->member) + __must_be_array((p)->member),\
327 		    sizeof(*(p)))
328 
329 /**
330  * flex_array_size() - Calculate size of a flexible array member
331  *                     within an enclosing structure.
332  *
333  * @p: Pointer to the structure.
334  * @member: Name of the flexible array member.
335  * @count: Number of elements in the array.
336  *
337  * Calculates size of a flexible array of @count number of @member
338  * elements, at the end of structure @p.
339  *
340  * Return: number of bytes needed or SIZE_MAX on overflow.
341  */
342 #define flex_array_size(p, member, count)				\
343 	array_size(count,						\
344 		    sizeof(*(p)->member) + __must_be_array((p)->member))
345 
346 #endif /* __LINUX_OVERFLOW_H */
347