Sasha Smundak | 0fc590b | 2020-10-07 08:11:59 -0700 | [diff] [blame] | 1 | //===- FuzzedDataProvider.h - Utility header for fuzz targets ---*- C++ -* ===// |
| 2 | // |
| 3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 | // See https://llvm.org/LICENSE.txt for license information. |
| 5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 6 | // |
| 7 | //===----------------------------------------------------------------------===// |
| 8 | // A single header library providing an utility class to break up an array of |
| 9 | // bytes. Whenever run on the same input, provides the same output, as long as |
| 10 | // its methods are called in the same order, with the same arguments. |
| 11 | //===----------------------------------------------------------------------===// |
| 12 | |
| 13 | #ifndef LLVM_FUZZER_FUZZED_DATA_PROVIDER_H_ |
| 14 | #define LLVM_FUZZER_FUZZED_DATA_PROVIDER_H_ |
| 15 | |
| 16 | #include <algorithm> |
Pirama Arumuga Nainar | 986b880 | 2021-06-03 16:00:34 -0700 | [diff] [blame] | 17 | #include <array> |
Sasha Smundak | 0fc590b | 2020-10-07 08:11:59 -0700 | [diff] [blame] | 18 | #include <climits> |
| 19 | #include <cstddef> |
| 20 | #include <cstdint> |
| 21 | #include <cstring> |
| 22 | #include <initializer_list> |
Pirama Arumuga Nainar | 7e1f839 | 2021-08-16 17:30:48 -0700 | [diff] [blame] | 23 | #include <limits> |
Sasha Smundak | 0fc590b | 2020-10-07 08:11:59 -0700 | [diff] [blame] | 24 | #include <string> |
| 25 | #include <type_traits> |
| 26 | #include <utility> |
| 27 | #include <vector> |
| 28 | |
| 29 | // In addition to the comments below, the API is also briefly documented at |
| 30 | // https://github.com/google/fuzzing/blob/master/docs/split-inputs.md#fuzzed-data-provider |
| 31 | class FuzzedDataProvider { |
| 32 | public: |
| 33 | // |data| is an array of length |size| that the FuzzedDataProvider wraps to |
| 34 | // provide more granular access. |data| must outlive the FuzzedDataProvider. |
| 35 | FuzzedDataProvider(const uint8_t *data, size_t size) |
| 36 | : data_ptr_(data), remaining_bytes_(size) {} |
| 37 | ~FuzzedDataProvider() = default; |
| 38 | |
| 39 | // See the implementation below (after the class definition) for more verbose |
| 40 | // comments for each of the methods. |
| 41 | |
| 42 | // Methods returning std::vector of bytes. These are the most popular choice |
| 43 | // when splitting fuzzing input into pieces, as every piece is put into a |
| 44 | // separate buffer (i.e. ASan would catch any under-/overflow) and the memory |
| 45 | // will be released automatically. |
| 46 | template <typename T> std::vector<T> ConsumeBytes(size_t num_bytes); |
| 47 | template <typename T> |
| 48 | std::vector<T> ConsumeBytesWithTerminator(size_t num_bytes, T terminator = 0); |
| 49 | template <typename T> std::vector<T> ConsumeRemainingBytes(); |
| 50 | |
| 51 | // Methods returning strings. Use only when you need a std::string or a null |
| 52 | // terminated C-string. Otherwise, prefer the methods returning std::vector. |
| 53 | std::string ConsumeBytesAsString(size_t num_bytes); |
| 54 | std::string ConsumeRandomLengthString(size_t max_length); |
| 55 | std::string ConsumeRandomLengthString(); |
| 56 | std::string ConsumeRemainingBytesAsString(); |
| 57 | |
| 58 | // Methods returning integer values. |
| 59 | template <typename T> T ConsumeIntegral(); |
| 60 | template <typename T> T ConsumeIntegralInRange(T min, T max); |
| 61 | |
| 62 | // Methods returning floating point values. |
| 63 | template <typename T> T ConsumeFloatingPoint(); |
| 64 | template <typename T> T ConsumeFloatingPointInRange(T min, T max); |
| 65 | |
| 66 | // 0 <= return value <= 1. |
| 67 | template <typename T> T ConsumeProbability(); |
| 68 | |
| 69 | bool ConsumeBool(); |
| 70 | |
| 71 | // Returns a value chosen from the given enum. |
| 72 | template <typename T> T ConsumeEnum(); |
| 73 | |
| 74 | // Returns a value from the given array. |
| 75 | template <typename T, size_t size> T PickValueInArray(const T (&array)[size]); |
Pirama Arumuga Nainar | 986b880 | 2021-06-03 16:00:34 -0700 | [diff] [blame] | 76 | template <typename T, size_t size> |
| 77 | T PickValueInArray(const std::array<T, size> &array); |
Sasha Smundak | 0fc590b | 2020-10-07 08:11:59 -0700 | [diff] [blame] | 78 | template <typename T> T PickValueInArray(std::initializer_list<const T> list); |
| 79 | |
| 80 | // Writes data to the given destination and returns number of bytes written. |
| 81 | size_t ConsumeData(void *destination, size_t num_bytes); |
| 82 | |
| 83 | // Reports the remaining bytes available for fuzzed input. |
| 84 | size_t remaining_bytes() { return remaining_bytes_; } |
| 85 | |
| 86 | private: |
| 87 | FuzzedDataProvider(const FuzzedDataProvider &) = delete; |
| 88 | FuzzedDataProvider &operator=(const FuzzedDataProvider &) = delete; |
| 89 | |
| 90 | void CopyAndAdvance(void *destination, size_t num_bytes); |
| 91 | |
| 92 | void Advance(size_t num_bytes); |
| 93 | |
| 94 | template <typename T> |
| 95 | std::vector<T> ConsumeBytes(size_t size, size_t num_bytes); |
| 96 | |
| 97 | template <typename TS, typename TU> TS ConvertUnsignedToSigned(TU value); |
| 98 | |
| 99 | const uint8_t *data_ptr_; |
| 100 | size_t remaining_bytes_; |
| 101 | }; |
| 102 | |
| 103 | // Returns a std::vector containing |num_bytes| of input data. If fewer than |
| 104 | // |num_bytes| of data remain, returns a shorter std::vector containing all |
| 105 | // of the data that's left. Can be used with any byte sized type, such as |
| 106 | // char, unsigned char, uint8_t, etc. |
| 107 | template <typename T> |
| 108 | std::vector<T> FuzzedDataProvider::ConsumeBytes(size_t num_bytes) { |
| 109 | num_bytes = std::min(num_bytes, remaining_bytes_); |
| 110 | return ConsumeBytes<T>(num_bytes, num_bytes); |
| 111 | } |
| 112 | |
| 113 | // Similar to |ConsumeBytes|, but also appends the terminator value at the end |
| 114 | // of the resulting vector. Useful, when a mutable null-terminated C-string is |
| 115 | // needed, for example. But that is a rare case. Better avoid it, if possible, |
| 116 | // and prefer using |ConsumeBytes| or |ConsumeBytesAsString| methods. |
| 117 | template <typename T> |
| 118 | std::vector<T> FuzzedDataProvider::ConsumeBytesWithTerminator(size_t num_bytes, |
| 119 | T terminator) { |
| 120 | num_bytes = std::min(num_bytes, remaining_bytes_); |
| 121 | std::vector<T> result = ConsumeBytes<T>(num_bytes + 1, num_bytes); |
| 122 | result.back() = terminator; |
| 123 | return result; |
| 124 | } |
| 125 | |
| 126 | // Returns a std::vector containing all remaining bytes of the input data. |
| 127 | template <typename T> |
| 128 | std::vector<T> FuzzedDataProvider::ConsumeRemainingBytes() { |
| 129 | return ConsumeBytes<T>(remaining_bytes_); |
| 130 | } |
| 131 | |
| 132 | // Returns a std::string containing |num_bytes| of input data. Using this and |
| 133 | // |.c_str()| on the resulting string is the best way to get an immutable |
| 134 | // null-terminated C string. If fewer than |num_bytes| of data remain, returns |
| 135 | // a shorter std::string containing all of the data that's left. |
| 136 | inline std::string FuzzedDataProvider::ConsumeBytesAsString(size_t num_bytes) { |
| 137 | static_assert(sizeof(std::string::value_type) == sizeof(uint8_t), |
| 138 | "ConsumeBytesAsString cannot convert the data to a string."); |
| 139 | |
| 140 | num_bytes = std::min(num_bytes, remaining_bytes_); |
| 141 | std::string result( |
| 142 | reinterpret_cast<const std::string::value_type *>(data_ptr_), num_bytes); |
| 143 | Advance(num_bytes); |
| 144 | return result; |
| 145 | } |
| 146 | |
| 147 | // Returns a std::string of length from 0 to |max_length|. When it runs out of |
| 148 | // input data, returns what remains of the input. Designed to be more stable |
| 149 | // with respect to a fuzzer inserting characters than just picking a random |
| 150 | // length and then consuming that many bytes with |ConsumeBytes|. |
| 151 | inline std::string |
| 152 | FuzzedDataProvider::ConsumeRandomLengthString(size_t max_length) { |
| 153 | // Reads bytes from the start of |data_ptr_|. Maps "\\" to "\", and maps "\" |
| 154 | // followed by anything else to the end of the string. As a result of this |
| 155 | // logic, a fuzzer can insert characters into the string, and the string |
| 156 | // will be lengthened to include those new characters, resulting in a more |
| 157 | // stable fuzzer than picking the length of a string independently from |
| 158 | // picking its contents. |
| 159 | std::string result; |
| 160 | |
| 161 | // Reserve the anticipated capaticity to prevent several reallocations. |
| 162 | result.reserve(std::min(max_length, remaining_bytes_)); |
| 163 | for (size_t i = 0; i < max_length && remaining_bytes_ != 0; ++i) { |
| 164 | char next = ConvertUnsignedToSigned<char>(data_ptr_[0]); |
| 165 | Advance(1); |
| 166 | if (next == '\\' && remaining_bytes_ != 0) { |
| 167 | next = ConvertUnsignedToSigned<char>(data_ptr_[0]); |
| 168 | Advance(1); |
| 169 | if (next != '\\') |
| 170 | break; |
| 171 | } |
| 172 | result += next; |
| 173 | } |
| 174 | |
| 175 | result.shrink_to_fit(); |
| 176 | return result; |
| 177 | } |
| 178 | |
| 179 | // Returns a std::string of length from 0 to |remaining_bytes_|. |
| 180 | inline std::string FuzzedDataProvider::ConsumeRandomLengthString() { |
| 181 | return ConsumeRandomLengthString(remaining_bytes_); |
| 182 | } |
| 183 | |
| 184 | // Returns a std::string containing all remaining bytes of the input data. |
| 185 | // Prefer using |ConsumeRemainingBytes| unless you actually need a std::string |
| 186 | // object. |
| 187 | inline std::string FuzzedDataProvider::ConsumeRemainingBytesAsString() { |
| 188 | return ConsumeBytesAsString(remaining_bytes_); |
| 189 | } |
| 190 | |
| 191 | // Returns a number in the range [Type's min, Type's max]. The value might |
| 192 | // not be uniformly distributed in the given range. If there's no input data |
| 193 | // left, always returns |min|. |
| 194 | template <typename T> T FuzzedDataProvider::ConsumeIntegral() { |
| 195 | return ConsumeIntegralInRange(std::numeric_limits<T>::min(), |
| 196 | std::numeric_limits<T>::max()); |
| 197 | } |
| 198 | |
| 199 | // Returns a number in the range [min, max] by consuming bytes from the |
| 200 | // input data. The value might not be uniformly distributed in the given |
| 201 | // range. If there's no input data left, always returns |min|. |min| must |
| 202 | // be less than or equal to |max|. |
| 203 | template <typename T> |
| 204 | T FuzzedDataProvider::ConsumeIntegralInRange(T min, T max) { |
| 205 | static_assert(std::is_integral<T>::value, "An integral type is required."); |
| 206 | static_assert(sizeof(T) <= sizeof(uint64_t), "Unsupported integral type."); |
| 207 | |
| 208 | if (min > max) |
| 209 | abort(); |
| 210 | |
| 211 | // Use the biggest type possible to hold the range and the result. |
| 212 | uint64_t range = static_cast<uint64_t>(max) - min; |
| 213 | uint64_t result = 0; |
| 214 | size_t offset = 0; |
| 215 | |
| 216 | while (offset < sizeof(T) * CHAR_BIT && (range >> offset) > 0 && |
| 217 | remaining_bytes_ != 0) { |
| 218 | // Pull bytes off the end of the seed data. Experimentally, this seems to |
| 219 | // allow the fuzzer to more easily explore the input space. This makes |
| 220 | // sense, since it works by modifying inputs that caused new code to run, |
| 221 | // and this data is often used to encode length of data read by |
| 222 | // |ConsumeBytes|. Separating out read lengths makes it easier modify the |
| 223 | // contents of the data that is actually read. |
| 224 | --remaining_bytes_; |
| 225 | result = (result << CHAR_BIT) | data_ptr_[remaining_bytes_]; |
| 226 | offset += CHAR_BIT; |
| 227 | } |
| 228 | |
| 229 | // Avoid division by 0, in case |range + 1| results in overflow. |
| 230 | if (range != std::numeric_limits<decltype(range)>::max()) |
| 231 | result = result % (range + 1); |
| 232 | |
| 233 | return static_cast<T>(min + result); |
| 234 | } |
| 235 | |
| 236 | // Returns a floating point value in the range [Type's lowest, Type's max] by |
| 237 | // consuming bytes from the input data. If there's no input data left, always |
| 238 | // returns approximately 0. |
| 239 | template <typename T> T FuzzedDataProvider::ConsumeFloatingPoint() { |
| 240 | return ConsumeFloatingPointInRange<T>(std::numeric_limits<T>::lowest(), |
| 241 | std::numeric_limits<T>::max()); |
| 242 | } |
| 243 | |
| 244 | // Returns a floating point value in the given range by consuming bytes from |
| 245 | // the input data. If there's no input data left, returns |min|. Note that |
| 246 | // |min| must be less than or equal to |max|. |
| 247 | template <typename T> |
| 248 | T FuzzedDataProvider::ConsumeFloatingPointInRange(T min, T max) { |
| 249 | if (min > max) |
| 250 | abort(); |
| 251 | |
| 252 | T range = .0; |
| 253 | T result = min; |
| 254 | constexpr T zero(.0); |
| 255 | if (max > zero && min < zero && max > min + std::numeric_limits<T>::max()) { |
| 256 | // The diff |max - min| would overflow the given floating point type. Use |
| 257 | // the half of the diff as the range and consume a bool to decide whether |
| 258 | // the result is in the first of the second part of the diff. |
| 259 | range = (max / 2.0) - (min / 2.0); |
| 260 | if (ConsumeBool()) { |
| 261 | result += range; |
| 262 | } |
| 263 | } else { |
| 264 | range = max - min; |
| 265 | } |
| 266 | |
| 267 | return result + range * ConsumeProbability<T>(); |
| 268 | } |
| 269 | |
| 270 | // Returns a floating point number in the range [0.0, 1.0]. If there's no |
| 271 | // input data left, always returns 0. |
| 272 | template <typename T> T FuzzedDataProvider::ConsumeProbability() { |
| 273 | static_assert(std::is_floating_point<T>::value, |
| 274 | "A floating point type is required."); |
| 275 | |
| 276 | // Use different integral types for different floating point types in order |
| 277 | // to provide better density of the resulting values. |
| 278 | using IntegralType = |
| 279 | typename std::conditional<(sizeof(T) <= sizeof(uint32_t)), uint32_t, |
| 280 | uint64_t>::type; |
| 281 | |
| 282 | T result = static_cast<T>(ConsumeIntegral<IntegralType>()); |
| 283 | result /= static_cast<T>(std::numeric_limits<IntegralType>::max()); |
| 284 | return result; |
| 285 | } |
| 286 | |
| 287 | // Reads one byte and returns a bool, or false when no data remains. |
| 288 | inline bool FuzzedDataProvider::ConsumeBool() { |
| 289 | return 1 & ConsumeIntegral<uint8_t>(); |
| 290 | } |
| 291 | |
| 292 | // Returns an enum value. The enum must start at 0 and be contiguous. It must |
| 293 | // also contain |kMaxValue| aliased to its largest (inclusive) value. Such as: |
| 294 | // enum class Foo { SomeValue, OtherValue, kMaxValue = OtherValue }; |
| 295 | template <typename T> T FuzzedDataProvider::ConsumeEnum() { |
| 296 | static_assert(std::is_enum<T>::value, "|T| must be an enum type."); |
| 297 | return static_cast<T>( |
| 298 | ConsumeIntegralInRange<uint32_t>(0, static_cast<uint32_t>(T::kMaxValue))); |
| 299 | } |
| 300 | |
| 301 | // Returns a copy of the value selected from the given fixed-size |array|. |
| 302 | template <typename T, size_t size> |
| 303 | T FuzzedDataProvider::PickValueInArray(const T (&array)[size]) { |
| 304 | static_assert(size > 0, "The array must be non empty."); |
| 305 | return array[ConsumeIntegralInRange<size_t>(0, size - 1)]; |
| 306 | } |
| 307 | |
Pirama Arumuga Nainar | 986b880 | 2021-06-03 16:00:34 -0700 | [diff] [blame] | 308 | template <typename T, size_t size> |
| 309 | T FuzzedDataProvider::PickValueInArray(const std::array<T, size> &array) { |
| 310 | static_assert(size > 0, "The array must be non empty."); |
| 311 | return array[ConsumeIntegralInRange<size_t>(0, size - 1)]; |
| 312 | } |
| 313 | |
Sasha Smundak | 0fc590b | 2020-10-07 08:11:59 -0700 | [diff] [blame] | 314 | template <typename T> |
| 315 | T FuzzedDataProvider::PickValueInArray(std::initializer_list<const T> list) { |
| 316 | // TODO(Dor1s): switch to static_assert once C++14 is allowed. |
| 317 | if (!list.size()) |
| 318 | abort(); |
| 319 | |
| 320 | return *(list.begin() + ConsumeIntegralInRange<size_t>(0, list.size() - 1)); |
| 321 | } |
| 322 | |
| 323 | // Writes |num_bytes| of input data to the given destination pointer. If there |
| 324 | // is not enough data left, writes all remaining bytes. Return value is the |
| 325 | // number of bytes written. |
| 326 | // In general, it's better to avoid using this function, but it may be useful |
| 327 | // in cases when it's necessary to fill a certain buffer or object with |
| 328 | // fuzzing data. |
| 329 | inline size_t FuzzedDataProvider::ConsumeData(void *destination, |
| 330 | size_t num_bytes) { |
| 331 | num_bytes = std::min(num_bytes, remaining_bytes_); |
| 332 | CopyAndAdvance(destination, num_bytes); |
| 333 | return num_bytes; |
| 334 | } |
| 335 | |
| 336 | // Private methods. |
| 337 | inline void FuzzedDataProvider::CopyAndAdvance(void *destination, |
| 338 | size_t num_bytes) { |
| 339 | std::memcpy(destination, data_ptr_, num_bytes); |
| 340 | Advance(num_bytes); |
| 341 | } |
| 342 | |
| 343 | inline void FuzzedDataProvider::Advance(size_t num_bytes) { |
| 344 | if (num_bytes > remaining_bytes_) |
| 345 | abort(); |
| 346 | |
| 347 | data_ptr_ += num_bytes; |
| 348 | remaining_bytes_ -= num_bytes; |
| 349 | } |
| 350 | |
| 351 | template <typename T> |
| 352 | std::vector<T> FuzzedDataProvider::ConsumeBytes(size_t size, size_t num_bytes) { |
| 353 | static_assert(sizeof(T) == sizeof(uint8_t), "Incompatible data type."); |
| 354 | |
| 355 | // The point of using the size-based constructor below is to increase the |
| 356 | // odds of having a vector object with capacity being equal to the length. |
| 357 | // That part is always implementation specific, but at least both libc++ and |
| 358 | // libstdc++ allocate the requested number of bytes in that constructor, |
| 359 | // which seems to be a natural choice for other implementations as well. |
| 360 | // To increase the odds even more, we also call |shrink_to_fit| below. |
| 361 | std::vector<T> result(size); |
| 362 | if (size == 0) { |
| 363 | if (num_bytes != 0) |
| 364 | abort(); |
| 365 | return result; |
| 366 | } |
| 367 | |
| 368 | CopyAndAdvance(result.data(), num_bytes); |
| 369 | |
| 370 | // Even though |shrink_to_fit| is also implementation specific, we expect it |
| 371 | // to provide an additional assurance in case vector's constructor allocated |
| 372 | // a buffer which is larger than the actual amount of data we put inside it. |
| 373 | result.shrink_to_fit(); |
| 374 | return result; |
| 375 | } |
| 376 | |
| 377 | template <typename TS, typename TU> |
| 378 | TS FuzzedDataProvider::ConvertUnsignedToSigned(TU value) { |
| 379 | static_assert(sizeof(TS) == sizeof(TU), "Incompatible data types."); |
| 380 | static_assert(!std::numeric_limits<TU>::is_signed, |
| 381 | "Source type must be unsigned."); |
| 382 | |
| 383 | // TODO(Dor1s): change to `if constexpr` once C++17 becomes mainstream. |
| 384 | if (std::numeric_limits<TS>::is_modulo) |
| 385 | return static_cast<TS>(value); |
| 386 | |
| 387 | // Avoid using implementation-defined unsigned to signed conversions. |
| 388 | // To learn more, see https://stackoverflow.com/questions/13150449. |
| 389 | if (value <= std::numeric_limits<TS>::max()) { |
| 390 | return static_cast<TS>(value); |
| 391 | } else { |
| 392 | constexpr auto TS_min = std::numeric_limits<TS>::min(); |
Pirama Arumuga Nainar | 7e1f839 | 2021-08-16 17:30:48 -0700 | [diff] [blame] | 393 | return TS_min + static_cast<TS>(value - TS_min); |
Sasha Smundak | 0fc590b | 2020-10-07 08:11:59 -0700 | [diff] [blame] | 394 | } |
| 395 | } |
| 396 | |
| 397 | #endif // LLVM_FUZZER_FUZZED_DATA_PROVIDER_H_ |