Brian Silverman | 9c614bc | 2016-02-15 20:20:02 -0500 | [diff] [blame] | 1 | // Protocol Buffers - Google's data interchange format |
| 2 | // Copyright 2008 Google Inc. All rights reserved. |
| 3 | // https://developers.google.com/protocol-buffers/ |
| 4 | // |
| 5 | // Redistribution and use in source and binary forms, with or without |
| 6 | // modification, are permitted provided that the following conditions are |
| 7 | // met: |
| 8 | // |
| 9 | // * Redistributions of source code must retain the above copyright |
| 10 | // notice, this list of conditions and the following disclaimer. |
| 11 | // * Redistributions in binary form must reproduce the above |
| 12 | // copyright notice, this list of conditions and the following disclaimer |
| 13 | // in the documentation and/or other materials provided with the |
| 14 | // distribution. |
| 15 | // * Neither the name of Google Inc. nor the names of its |
| 16 | // contributors may be used to endorse or promote products derived from |
| 17 | // this software without specific prior written permission. |
| 18 | // |
| 19 | // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
| 20 | // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
| 21 | // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR |
| 22 | // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT |
| 23 | // OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, |
| 24 | // SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT |
| 25 | // LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, |
| 26 | // DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY |
| 27 | // THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
| 28 | // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
| 29 | // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
| 30 | |
| 31 | // Author: kenton@google.com (Kenton Varda) |
| 32 | // Based on original Protocol Buffers design by |
| 33 | // Sanjay Ghemawat, Jeff Dean, and others. |
| 34 | // |
| 35 | // This implementation is heavily optimized to make reads and writes |
| 36 | // of small values (especially varints) as fast as possible. In |
| 37 | // particular, we optimize for the common case that a read or a write |
| 38 | // will not cross the end of the buffer, since we can avoid a lot |
| 39 | // of branching in this case. |
| 40 | |
| 41 | #include <google/protobuf/io/coded_stream_inl.h> |
| 42 | #include <algorithm> |
| 43 | #include <utility> |
| 44 | #include <limits.h> |
| 45 | #include <google/protobuf/io/zero_copy_stream.h> |
| 46 | #include <google/protobuf/arena.h> |
| 47 | #include <google/protobuf/stubs/logging.h> |
| 48 | #include <google/protobuf/stubs/common.h> |
| 49 | #include <google/protobuf/stubs/stl_util.h> |
| 50 | |
| 51 | |
| 52 | namespace google { |
| 53 | namespace protobuf { |
| 54 | namespace io { |
| 55 | |
| 56 | namespace { |
| 57 | |
| 58 | static const int kMaxVarintBytes = 10; |
| 59 | static const int kMaxVarint32Bytes = 5; |
| 60 | |
| 61 | |
| 62 | inline bool NextNonEmpty(ZeroCopyInputStream* input, |
| 63 | const void** data, int* size) { |
| 64 | bool success; |
| 65 | do { |
| 66 | success = input->Next(data, size); |
| 67 | } while (success && *size == 0); |
| 68 | return success; |
| 69 | } |
| 70 | |
| 71 | } // namespace |
| 72 | |
| 73 | // CodedInputStream ================================================== |
| 74 | |
| 75 | CodedInputStream::~CodedInputStream() { |
| 76 | if (input_ != NULL) { |
| 77 | BackUpInputToCurrentPosition(); |
| 78 | } |
Brian Silverman | 9c614bc | 2016-02-15 20:20:02 -0500 | [diff] [blame] | 79 | } |
| 80 | |
| 81 | // Static. |
| 82 | int CodedInputStream::default_recursion_limit_ = 100; |
| 83 | |
| 84 | |
| 85 | void CodedOutputStream::EnableAliasing(bool enabled) { |
| 86 | aliasing_enabled_ = enabled && output_->AllowsAliasing(); |
| 87 | } |
| 88 | |
| 89 | void CodedInputStream::BackUpInputToCurrentPosition() { |
| 90 | int backup_bytes = BufferSize() + buffer_size_after_limit_ + overflow_bytes_; |
| 91 | if (backup_bytes > 0) { |
| 92 | input_->BackUp(backup_bytes); |
| 93 | |
| 94 | // total_bytes_read_ doesn't include overflow_bytes_. |
| 95 | total_bytes_read_ -= BufferSize() + buffer_size_after_limit_; |
| 96 | buffer_end_ = buffer_; |
| 97 | buffer_size_after_limit_ = 0; |
| 98 | overflow_bytes_ = 0; |
| 99 | } |
| 100 | } |
| 101 | |
| 102 | inline void CodedInputStream::RecomputeBufferLimits() { |
| 103 | buffer_end_ += buffer_size_after_limit_; |
Austin Schuh | 40c1652 | 2018-10-28 20:27:54 -0700 | [diff] [blame^] | 104 | int closest_limit = std::min(current_limit_, total_bytes_limit_); |
Brian Silverman | 9c614bc | 2016-02-15 20:20:02 -0500 | [diff] [blame] | 105 | if (closest_limit < total_bytes_read_) { |
| 106 | // The limit position is in the current buffer. We must adjust |
| 107 | // the buffer size accordingly. |
| 108 | buffer_size_after_limit_ = total_bytes_read_ - closest_limit; |
| 109 | buffer_end_ -= buffer_size_after_limit_; |
| 110 | } else { |
| 111 | buffer_size_after_limit_ = 0; |
| 112 | } |
| 113 | } |
| 114 | |
| 115 | CodedInputStream::Limit CodedInputStream::PushLimit(int byte_limit) { |
| 116 | // Current position relative to the beginning of the stream. |
| 117 | int current_position = CurrentPosition(); |
| 118 | |
| 119 | Limit old_limit = current_limit_; |
| 120 | |
| 121 | // security: byte_limit is possibly evil, so check for negative values |
Austin Schuh | 40c1652 | 2018-10-28 20:27:54 -0700 | [diff] [blame^] | 122 | // and overflow. Also check that the new requested limit is before the |
| 123 | // previous limit; otherwise we continue to enforce the previous limit. |
| 124 | if (GOOGLE_PREDICT_TRUE(byte_limit >= 0 && |
| 125 | byte_limit <= INT_MAX - current_position && |
| 126 | byte_limit < current_limit_ - current_position)) { |
Brian Silverman | 9c614bc | 2016-02-15 20:20:02 -0500 | [diff] [blame] | 127 | current_limit_ = current_position + byte_limit; |
Austin Schuh | 40c1652 | 2018-10-28 20:27:54 -0700 | [diff] [blame^] | 128 | RecomputeBufferLimits(); |
Brian Silverman | 9c614bc | 2016-02-15 20:20:02 -0500 | [diff] [blame] | 129 | } |
| 130 | |
Brian Silverman | 9c614bc | 2016-02-15 20:20:02 -0500 | [diff] [blame] | 131 | return old_limit; |
| 132 | } |
| 133 | |
| 134 | void CodedInputStream::PopLimit(Limit limit) { |
| 135 | // The limit passed in is actually the *old* limit, which we returned from |
| 136 | // PushLimit(). |
| 137 | current_limit_ = limit; |
| 138 | RecomputeBufferLimits(); |
| 139 | |
| 140 | // We may no longer be at a legitimate message end. ReadTag() needs to be |
| 141 | // called again to find out. |
| 142 | legitimate_message_end_ = false; |
| 143 | } |
| 144 | |
| 145 | std::pair<CodedInputStream::Limit, int> |
| 146 | CodedInputStream::IncrementRecursionDepthAndPushLimit(int byte_limit) { |
| 147 | return std::make_pair(PushLimit(byte_limit), --recursion_budget_); |
| 148 | } |
| 149 | |
| 150 | CodedInputStream::Limit CodedInputStream::ReadLengthAndPushLimit() { |
| 151 | uint32 length; |
| 152 | return PushLimit(ReadVarint32(&length) ? length : 0); |
| 153 | } |
| 154 | |
| 155 | bool CodedInputStream::DecrementRecursionDepthAndPopLimit(Limit limit) { |
| 156 | bool result = ConsumedEntireMessage(); |
| 157 | PopLimit(limit); |
| 158 | GOOGLE_DCHECK_LT(recursion_budget_, recursion_limit_); |
| 159 | ++recursion_budget_; |
| 160 | return result; |
| 161 | } |
| 162 | |
| 163 | bool CodedInputStream::CheckEntireMessageConsumedAndPopLimit(Limit limit) { |
| 164 | bool result = ConsumedEntireMessage(); |
| 165 | PopLimit(limit); |
| 166 | return result; |
| 167 | } |
| 168 | |
| 169 | int CodedInputStream::BytesUntilLimit() const { |
| 170 | if (current_limit_ == INT_MAX) return -1; |
| 171 | int current_position = CurrentPosition(); |
| 172 | |
| 173 | return current_limit_ - current_position; |
| 174 | } |
| 175 | |
Austin Schuh | 40c1652 | 2018-10-28 20:27:54 -0700 | [diff] [blame^] | 176 | void CodedInputStream::SetTotalBytesLimit(int total_bytes_limit) { |
Brian Silverman | 9c614bc | 2016-02-15 20:20:02 -0500 | [diff] [blame] | 177 | // Make sure the limit isn't already past, since this could confuse other |
| 178 | // code. |
| 179 | int current_position = CurrentPosition(); |
Austin Schuh | 40c1652 | 2018-10-28 20:27:54 -0700 | [diff] [blame^] | 180 | total_bytes_limit_ = std::max(current_position, total_bytes_limit); |
Brian Silverman | 9c614bc | 2016-02-15 20:20:02 -0500 | [diff] [blame] | 181 | RecomputeBufferLimits(); |
| 182 | } |
| 183 | |
| 184 | int CodedInputStream::BytesUntilTotalBytesLimit() const { |
| 185 | if (total_bytes_limit_ == INT_MAX) return -1; |
| 186 | return total_bytes_limit_ - CurrentPosition(); |
| 187 | } |
| 188 | |
| 189 | void CodedInputStream::PrintTotalBytesLimitError() { |
| 190 | GOOGLE_LOG(ERROR) << "A protocol message was rejected because it was too " |
| 191 | "big (more than " << total_bytes_limit_ |
| 192 | << " bytes). To increase the limit (or to disable these " |
| 193 | "warnings), see CodedInputStream::SetTotalBytesLimit() " |
| 194 | "in google/protobuf/io/coded_stream.h."; |
| 195 | } |
| 196 | |
Austin Schuh | 40c1652 | 2018-10-28 20:27:54 -0700 | [diff] [blame^] | 197 | bool CodedInputStream::SkipFallback(int count, int original_buffer_size) { |
Brian Silverman | 9c614bc | 2016-02-15 20:20:02 -0500 | [diff] [blame] | 198 | if (buffer_size_after_limit_ > 0) { |
| 199 | // We hit a limit inside this buffer. Advance to the limit and fail. |
| 200 | Advance(original_buffer_size); |
| 201 | return false; |
| 202 | } |
| 203 | |
| 204 | count -= original_buffer_size; |
| 205 | buffer_ = NULL; |
| 206 | buffer_end_ = buffer_; |
| 207 | |
| 208 | // Make sure this skip doesn't try to skip past the current limit. |
Austin Schuh | 40c1652 | 2018-10-28 20:27:54 -0700 | [diff] [blame^] | 209 | int closest_limit = std::min(current_limit_, total_bytes_limit_); |
Brian Silverman | 9c614bc | 2016-02-15 20:20:02 -0500 | [diff] [blame] | 210 | int bytes_until_limit = closest_limit - total_bytes_read_; |
| 211 | if (bytes_until_limit < count) { |
| 212 | // We hit the limit. Skip up to it then fail. |
| 213 | if (bytes_until_limit > 0) { |
| 214 | total_bytes_read_ = closest_limit; |
| 215 | input_->Skip(bytes_until_limit); |
| 216 | } |
| 217 | return false; |
| 218 | } |
| 219 | |
Austin Schuh | 40c1652 | 2018-10-28 20:27:54 -0700 | [diff] [blame^] | 220 | if (!input_->Skip(count)) { |
| 221 | total_bytes_read_ = input_->ByteCount(); |
| 222 | return false; |
| 223 | } |
Brian Silverman | 9c614bc | 2016-02-15 20:20:02 -0500 | [diff] [blame] | 224 | total_bytes_read_ += count; |
Austin Schuh | 40c1652 | 2018-10-28 20:27:54 -0700 | [diff] [blame^] | 225 | return true; |
Brian Silverman | 9c614bc | 2016-02-15 20:20:02 -0500 | [diff] [blame] | 226 | } |
| 227 | |
| 228 | bool CodedInputStream::GetDirectBufferPointer(const void** data, int* size) { |
| 229 | if (BufferSize() == 0 && !Refresh()) return false; |
| 230 | |
| 231 | *data = buffer_; |
| 232 | *size = BufferSize(); |
| 233 | return true; |
| 234 | } |
| 235 | |
| 236 | bool CodedInputStream::ReadRaw(void* buffer, int size) { |
| 237 | return InternalReadRawInline(buffer, size); |
| 238 | } |
| 239 | |
| 240 | bool CodedInputStream::ReadString(string* buffer, int size) { |
| 241 | if (size < 0) return false; // security: size is often user-supplied |
| 242 | return InternalReadStringInline(buffer, size); |
| 243 | } |
| 244 | |
| 245 | bool CodedInputStream::ReadStringFallback(string* buffer, int size) { |
| 246 | if (!buffer->empty()) { |
| 247 | buffer->clear(); |
| 248 | } |
| 249 | |
Austin Schuh | 40c1652 | 2018-10-28 20:27:54 -0700 | [diff] [blame^] | 250 | int closest_limit = std::min(current_limit_, total_bytes_limit_); |
Brian Silverman | 9c614bc | 2016-02-15 20:20:02 -0500 | [diff] [blame] | 251 | if (closest_limit != INT_MAX) { |
| 252 | int bytes_to_limit = closest_limit - CurrentPosition(); |
| 253 | if (bytes_to_limit > 0 && size > 0 && size <= bytes_to_limit) { |
| 254 | buffer->reserve(size); |
| 255 | } |
| 256 | } |
| 257 | |
| 258 | int current_buffer_size; |
| 259 | while ((current_buffer_size = BufferSize()) < size) { |
| 260 | // Some STL implementations "helpfully" crash on buffer->append(NULL, 0). |
| 261 | if (current_buffer_size != 0) { |
| 262 | // Note: string1.append(string2) is O(string2.size()) (as opposed to |
| 263 | // O(string1.size() + string2.size()), which would be bad). |
| 264 | buffer->append(reinterpret_cast<const char*>(buffer_), |
| 265 | current_buffer_size); |
| 266 | } |
| 267 | size -= current_buffer_size; |
| 268 | Advance(current_buffer_size); |
| 269 | if (!Refresh()) return false; |
| 270 | } |
| 271 | |
| 272 | buffer->append(reinterpret_cast<const char*>(buffer_), size); |
| 273 | Advance(size); |
| 274 | |
| 275 | return true; |
| 276 | } |
| 277 | |
| 278 | |
| 279 | bool CodedInputStream::ReadLittleEndian32Fallback(uint32* value) { |
| 280 | uint8 bytes[sizeof(*value)]; |
| 281 | |
| 282 | const uint8* ptr; |
| 283 | if (BufferSize() >= sizeof(*value)) { |
| 284 | // Fast path: Enough bytes in the buffer to read directly. |
| 285 | ptr = buffer_; |
| 286 | Advance(sizeof(*value)); |
| 287 | } else { |
| 288 | // Slow path: Had to read past the end of the buffer. |
| 289 | if (!ReadRaw(bytes, sizeof(*value))) return false; |
| 290 | ptr = bytes; |
| 291 | } |
| 292 | ReadLittleEndian32FromArray(ptr, value); |
| 293 | return true; |
| 294 | } |
| 295 | |
| 296 | bool CodedInputStream::ReadLittleEndian64Fallback(uint64* value) { |
| 297 | uint8 bytes[sizeof(*value)]; |
| 298 | |
| 299 | const uint8* ptr; |
| 300 | if (BufferSize() >= sizeof(*value)) { |
| 301 | // Fast path: Enough bytes in the buffer to read directly. |
| 302 | ptr = buffer_; |
| 303 | Advance(sizeof(*value)); |
| 304 | } else { |
| 305 | // Slow path: Had to read past the end of the buffer. |
| 306 | if (!ReadRaw(bytes, sizeof(*value))) return false; |
| 307 | ptr = bytes; |
| 308 | } |
| 309 | ReadLittleEndian64FromArray(ptr, value); |
| 310 | return true; |
| 311 | } |
| 312 | |
| 313 | namespace { |
| 314 | |
| 315 | // Read a varint from the given buffer, write it to *value, and return a pair. |
| 316 | // The first part of the pair is true iff the read was successful. The second |
| 317 | // part is buffer + (number of bytes read). This function is always inlined, |
| 318 | // so returning a pair is costless. |
Austin Schuh | 40c1652 | 2018-10-28 20:27:54 -0700 | [diff] [blame^] | 319 | GOOGLE_PROTOBUF_ATTRIBUTE_ALWAYS_INLINE |
| 320 | ::std::pair<bool, const uint8*> ReadVarint32FromArray( |
Brian Silverman | 9c614bc | 2016-02-15 20:20:02 -0500 | [diff] [blame] | 321 | uint32 first_byte, const uint8* buffer, |
| 322 | uint32* value); |
| 323 | inline ::std::pair<bool, const uint8*> ReadVarint32FromArray( |
| 324 | uint32 first_byte, const uint8* buffer, uint32* value) { |
| 325 | // Fast path: We have enough bytes left in the buffer to guarantee that |
| 326 | // this read won't cross the end, so we can skip the checks. |
| 327 | GOOGLE_DCHECK_EQ(*buffer, first_byte); |
| 328 | GOOGLE_DCHECK_EQ(first_byte & 0x80, 0x80) << first_byte; |
| 329 | const uint8* ptr = buffer; |
| 330 | uint32 b; |
| 331 | uint32 result = first_byte - 0x80; |
| 332 | ++ptr; // We just processed the first byte. Move on to the second. |
| 333 | b = *(ptr++); result += b << 7; if (!(b & 0x80)) goto done; |
| 334 | result -= 0x80 << 7; |
| 335 | b = *(ptr++); result += b << 14; if (!(b & 0x80)) goto done; |
| 336 | result -= 0x80 << 14; |
| 337 | b = *(ptr++); result += b << 21; if (!(b & 0x80)) goto done; |
| 338 | result -= 0x80 << 21; |
| 339 | b = *(ptr++); result += b << 28; if (!(b & 0x80)) goto done; |
| 340 | // "result -= 0x80 << 28" is irrevelant. |
| 341 | |
| 342 | // If the input is larger than 32 bits, we still need to read it all |
| 343 | // and discard the high-order bits. |
| 344 | for (int i = 0; i < kMaxVarintBytes - kMaxVarint32Bytes; i++) { |
| 345 | b = *(ptr++); if (!(b & 0x80)) goto done; |
| 346 | } |
| 347 | |
| 348 | // We have overrun the maximum size of a varint (10 bytes). Assume |
| 349 | // the data is corrupt. |
| 350 | return std::make_pair(false, ptr); |
| 351 | |
| 352 | done: |
| 353 | *value = result; |
| 354 | return std::make_pair(true, ptr); |
| 355 | } |
| 356 | |
Austin Schuh | 40c1652 | 2018-10-28 20:27:54 -0700 | [diff] [blame^] | 357 | GOOGLE_PROTOBUF_ATTRIBUTE_ALWAYS_INLINE::std::pair<bool, const uint8*> |
| 358 | ReadVarint64FromArray(const uint8* buffer, uint64* value); |
| 359 | inline ::std::pair<bool, const uint8*> ReadVarint64FromArray( |
| 360 | const uint8* buffer, uint64* value) { |
| 361 | const uint8* ptr = buffer; |
| 362 | uint32 b; |
| 363 | |
| 364 | // Splitting into 32-bit pieces gives better performance on 32-bit |
| 365 | // processors. |
| 366 | uint32 part0 = 0, part1 = 0, part2 = 0; |
| 367 | |
| 368 | b = *(ptr++); part0 = b ; if (!(b & 0x80)) goto done; |
| 369 | part0 -= 0x80; |
| 370 | b = *(ptr++); part0 += b << 7; if (!(b & 0x80)) goto done; |
| 371 | part0 -= 0x80 << 7; |
| 372 | b = *(ptr++); part0 += b << 14; if (!(b & 0x80)) goto done; |
| 373 | part0 -= 0x80 << 14; |
| 374 | b = *(ptr++); part0 += b << 21; if (!(b & 0x80)) goto done; |
| 375 | part0 -= 0x80 << 21; |
| 376 | b = *(ptr++); part1 = b ; if (!(b & 0x80)) goto done; |
| 377 | part1 -= 0x80; |
| 378 | b = *(ptr++); part1 += b << 7; if (!(b & 0x80)) goto done; |
| 379 | part1 -= 0x80 << 7; |
| 380 | b = *(ptr++); part1 += b << 14; if (!(b & 0x80)) goto done; |
| 381 | part1 -= 0x80 << 14; |
| 382 | b = *(ptr++); part1 += b << 21; if (!(b & 0x80)) goto done; |
| 383 | part1 -= 0x80 << 21; |
| 384 | b = *(ptr++); part2 = b ; if (!(b & 0x80)) goto done; |
| 385 | part2 -= 0x80; |
| 386 | b = *(ptr++); part2 += b << 7; if (!(b & 0x80)) goto done; |
| 387 | // "part2 -= 0x80 << 7" is irrelevant because (0x80 << 7) << 56 is 0. |
| 388 | |
| 389 | // We have overrun the maximum size of a varint (10 bytes). Assume |
| 390 | // the data is corrupt. |
| 391 | return std::make_pair(false, ptr); |
| 392 | |
| 393 | done: |
| 394 | *value = (static_cast<uint64>(part0)) | |
| 395 | (static_cast<uint64>(part1) << 28) | |
| 396 | (static_cast<uint64>(part2) << 56); |
| 397 | return std::make_pair(true, ptr); |
| 398 | } |
| 399 | |
Brian Silverman | 9c614bc | 2016-02-15 20:20:02 -0500 | [diff] [blame] | 400 | } // namespace |
| 401 | |
| 402 | bool CodedInputStream::ReadVarint32Slow(uint32* value) { |
| 403 | // Directly invoke ReadVarint64Fallback, since we already tried to optimize |
| 404 | // for one-byte varints. |
| 405 | std::pair<uint64, bool> p = ReadVarint64Fallback(); |
| 406 | *value = static_cast<uint32>(p.first); |
| 407 | return p.second; |
| 408 | } |
| 409 | |
| 410 | int64 CodedInputStream::ReadVarint32Fallback(uint32 first_byte_or_zero) { |
| 411 | if (BufferSize() >= kMaxVarintBytes || |
| 412 | // Optimization: We're also safe if the buffer is non-empty and it ends |
| 413 | // with a byte that would terminate a varint. |
| 414 | (buffer_end_ > buffer_ && !(buffer_end_[-1] & 0x80))) { |
| 415 | GOOGLE_DCHECK_NE(first_byte_or_zero, 0) |
| 416 | << "Caller should provide us with *buffer_ when buffer is non-empty"; |
| 417 | uint32 temp; |
| 418 | ::std::pair<bool, const uint8*> p = |
| 419 | ReadVarint32FromArray(first_byte_or_zero, buffer_, &temp); |
| 420 | if (!p.first) return -1; |
| 421 | buffer_ = p.second; |
| 422 | return temp; |
| 423 | } else { |
| 424 | // Really slow case: we will incur the cost of an extra function call here, |
| 425 | // but moving this out of line reduces the size of this function, which |
| 426 | // improves the common case. In micro benchmarks, this is worth about 10-15% |
| 427 | uint32 temp; |
| 428 | return ReadVarint32Slow(&temp) ? static_cast<int64>(temp) : -1; |
| 429 | } |
| 430 | } |
| 431 | |
Austin Schuh | 40c1652 | 2018-10-28 20:27:54 -0700 | [diff] [blame^] | 432 | int CodedInputStream::ReadVarintSizeAsIntSlow() { |
| 433 | // Directly invoke ReadVarint64Fallback, since we already tried to optimize |
| 434 | // for one-byte varints. |
| 435 | std::pair<uint64, bool> p = ReadVarint64Fallback(); |
| 436 | if (!p.second || p.first > static_cast<uint64>(INT_MAX)) return -1; |
| 437 | return p.first; |
| 438 | } |
| 439 | |
| 440 | int CodedInputStream::ReadVarintSizeAsIntFallback() { |
| 441 | if (BufferSize() >= kMaxVarintBytes || |
| 442 | // Optimization: We're also safe if the buffer is non-empty and it ends |
| 443 | // with a byte that would terminate a varint. |
| 444 | (buffer_end_ > buffer_ && !(buffer_end_[-1] & 0x80))) { |
| 445 | uint64 temp; |
| 446 | ::std::pair<bool, const uint8*> p = ReadVarint64FromArray(buffer_, &temp); |
| 447 | if (!p.first || temp > static_cast<uint64>(INT_MAX)) return -1; |
| 448 | buffer_ = p.second; |
| 449 | return temp; |
| 450 | } else { |
| 451 | // Really slow case: we will incur the cost of an extra function call here, |
| 452 | // but moving this out of line reduces the size of this function, which |
| 453 | // improves the common case. In micro benchmarks, this is worth about 10-15% |
| 454 | return ReadVarintSizeAsIntSlow(); |
| 455 | } |
| 456 | } |
| 457 | |
Brian Silverman | 9c614bc | 2016-02-15 20:20:02 -0500 | [diff] [blame] | 458 | uint32 CodedInputStream::ReadTagSlow() { |
| 459 | if (buffer_ == buffer_end_) { |
| 460 | // Call refresh. |
| 461 | if (!Refresh()) { |
| 462 | // Refresh failed. Make sure that it failed due to EOF, not because |
| 463 | // we hit total_bytes_limit_, which, unlike normal limits, is not a |
| 464 | // valid place to end a message. |
| 465 | int current_position = total_bytes_read_ - buffer_size_after_limit_; |
| 466 | if (current_position >= total_bytes_limit_) { |
| 467 | // Hit total_bytes_limit_. But if we also hit the normal limit, |
| 468 | // we're still OK. |
| 469 | legitimate_message_end_ = current_limit_ == total_bytes_limit_; |
| 470 | } else { |
| 471 | legitimate_message_end_ = true; |
| 472 | } |
| 473 | return 0; |
| 474 | } |
| 475 | } |
| 476 | |
| 477 | // For the slow path, just do a 64-bit read. Try to optimize for one-byte tags |
| 478 | // again, since we have now refreshed the buffer. |
| 479 | uint64 result = 0; |
| 480 | if (!ReadVarint64(&result)) return 0; |
| 481 | return static_cast<uint32>(result); |
| 482 | } |
| 483 | |
| 484 | uint32 CodedInputStream::ReadTagFallback(uint32 first_byte_or_zero) { |
| 485 | const int buf_size = BufferSize(); |
| 486 | if (buf_size >= kMaxVarintBytes || |
| 487 | // Optimization: We're also safe if the buffer is non-empty and it ends |
| 488 | // with a byte that would terminate a varint. |
| 489 | (buf_size > 0 && !(buffer_end_[-1] & 0x80))) { |
| 490 | GOOGLE_DCHECK_EQ(first_byte_or_zero, buffer_[0]); |
| 491 | if (first_byte_or_zero == 0) { |
| 492 | ++buffer_; |
| 493 | return 0; |
| 494 | } |
| 495 | uint32 tag; |
| 496 | ::std::pair<bool, const uint8*> p = |
| 497 | ReadVarint32FromArray(first_byte_or_zero, buffer_, &tag); |
| 498 | if (!p.first) { |
| 499 | return 0; |
| 500 | } |
| 501 | buffer_ = p.second; |
| 502 | return tag; |
| 503 | } else { |
| 504 | // We are commonly at a limit when attempting to read tags. Try to quickly |
| 505 | // detect this case without making another function call. |
| 506 | if ((buf_size == 0) && |
| 507 | ((buffer_size_after_limit_ > 0) || |
| 508 | (total_bytes_read_ == current_limit_)) && |
| 509 | // Make sure that the limit we hit is not total_bytes_limit_, since |
| 510 | // in that case we still need to call Refresh() so that it prints an |
| 511 | // error. |
| 512 | total_bytes_read_ - buffer_size_after_limit_ < total_bytes_limit_) { |
| 513 | // We hit a byte limit. |
| 514 | legitimate_message_end_ = true; |
| 515 | return 0; |
| 516 | } |
| 517 | return ReadTagSlow(); |
| 518 | } |
| 519 | } |
| 520 | |
| 521 | bool CodedInputStream::ReadVarint64Slow(uint64* value) { |
| 522 | // Slow path: This read might cross the end of the buffer, so we |
| 523 | // need to check and refresh the buffer if and when it does. |
| 524 | |
| 525 | uint64 result = 0; |
| 526 | int count = 0; |
| 527 | uint32 b; |
| 528 | |
| 529 | do { |
Austin Schuh | 40c1652 | 2018-10-28 20:27:54 -0700 | [diff] [blame^] | 530 | if (count == kMaxVarintBytes) { |
| 531 | *value = 0; |
| 532 | return false; |
| 533 | } |
Brian Silverman | 9c614bc | 2016-02-15 20:20:02 -0500 | [diff] [blame] | 534 | while (buffer_ == buffer_end_) { |
Austin Schuh | 40c1652 | 2018-10-28 20:27:54 -0700 | [diff] [blame^] | 535 | if (!Refresh()) { |
| 536 | *value = 0; |
| 537 | return false; |
| 538 | } |
Brian Silverman | 9c614bc | 2016-02-15 20:20:02 -0500 | [diff] [blame] | 539 | } |
| 540 | b = *buffer_; |
| 541 | result |= static_cast<uint64>(b & 0x7F) << (7 * count); |
| 542 | Advance(1); |
| 543 | ++count; |
| 544 | } while (b & 0x80); |
| 545 | |
| 546 | *value = result; |
| 547 | return true; |
| 548 | } |
| 549 | |
| 550 | std::pair<uint64, bool> CodedInputStream::ReadVarint64Fallback() { |
| 551 | if (BufferSize() >= kMaxVarintBytes || |
| 552 | // Optimization: We're also safe if the buffer is non-empty and it ends |
| 553 | // with a byte that would terminate a varint. |
| 554 | (buffer_end_ > buffer_ && !(buffer_end_[-1] & 0x80))) { |
Austin Schuh | 40c1652 | 2018-10-28 20:27:54 -0700 | [diff] [blame^] | 555 | uint64 temp; |
| 556 | ::std::pair<bool, const uint8*> p = ReadVarint64FromArray(buffer_, &temp); |
| 557 | if (!p.first) { |
| 558 | return std::make_pair(0, false); |
| 559 | } |
| 560 | buffer_ = p.second; |
| 561 | return std::make_pair(temp, true); |
Brian Silverman | 9c614bc | 2016-02-15 20:20:02 -0500 | [diff] [blame] | 562 | } else { |
| 563 | uint64 temp; |
| 564 | bool success = ReadVarint64Slow(&temp); |
| 565 | return std::make_pair(temp, success); |
| 566 | } |
| 567 | } |
| 568 | |
| 569 | bool CodedInputStream::Refresh() { |
| 570 | GOOGLE_DCHECK_EQ(0, BufferSize()); |
| 571 | |
| 572 | if (buffer_size_after_limit_ > 0 || overflow_bytes_ > 0 || |
| 573 | total_bytes_read_ == current_limit_) { |
| 574 | // We've hit a limit. Stop. |
| 575 | int current_position = total_bytes_read_ - buffer_size_after_limit_; |
| 576 | |
| 577 | if (current_position >= total_bytes_limit_ && |
| 578 | total_bytes_limit_ != current_limit_) { |
| 579 | // Hit total_bytes_limit_. |
| 580 | PrintTotalBytesLimitError(); |
| 581 | } |
| 582 | |
| 583 | return false; |
| 584 | } |
| 585 | |
Brian Silverman | 9c614bc | 2016-02-15 20:20:02 -0500 | [diff] [blame] | 586 | const void* void_buffer; |
| 587 | int buffer_size; |
| 588 | if (NextNonEmpty(input_, &void_buffer, &buffer_size)) { |
| 589 | buffer_ = reinterpret_cast<const uint8*>(void_buffer); |
| 590 | buffer_end_ = buffer_ + buffer_size; |
| 591 | GOOGLE_CHECK_GE(buffer_size, 0); |
| 592 | |
| 593 | if (total_bytes_read_ <= INT_MAX - buffer_size) { |
| 594 | total_bytes_read_ += buffer_size; |
| 595 | } else { |
| 596 | // Overflow. Reset buffer_end_ to not include the bytes beyond INT_MAX. |
| 597 | // We can't get that far anyway, because total_bytes_limit_ is guaranteed |
| 598 | // to be less than it. We need to keep track of the number of bytes |
| 599 | // we discarded, though, so that we can call input_->BackUp() to back |
| 600 | // up over them on destruction. |
| 601 | |
| 602 | // The following line is equivalent to: |
| 603 | // overflow_bytes_ = total_bytes_read_ + buffer_size - INT_MAX; |
| 604 | // except that it avoids overflows. Signed integer overflow has |
| 605 | // undefined results according to the C standard. |
| 606 | overflow_bytes_ = total_bytes_read_ - (INT_MAX - buffer_size); |
| 607 | buffer_end_ -= overflow_bytes_; |
| 608 | total_bytes_read_ = INT_MAX; |
| 609 | } |
| 610 | |
| 611 | RecomputeBufferLimits(); |
| 612 | return true; |
| 613 | } else { |
| 614 | buffer_ = NULL; |
| 615 | buffer_end_ = NULL; |
| 616 | return false; |
| 617 | } |
| 618 | } |
| 619 | |
| 620 | // CodedOutputStream ================================================= |
| 621 | |
Austin Schuh | 40c1652 | 2018-10-28 20:27:54 -0700 | [diff] [blame^] | 622 | std::atomic<bool> CodedOutputStream::default_serialization_deterministic_{ |
| 623 | false}; |
| 624 | |
Brian Silverman | 9c614bc | 2016-02-15 20:20:02 -0500 | [diff] [blame] | 625 | CodedOutputStream::CodedOutputStream(ZeroCopyOutputStream* output) |
Austin Schuh | 40c1652 | 2018-10-28 20:27:54 -0700 | [diff] [blame^] | 626 | : CodedOutputStream(output, true) {} |
Brian Silverman | 9c614bc | 2016-02-15 20:20:02 -0500 | [diff] [blame] | 627 | |
| 628 | CodedOutputStream::CodedOutputStream(ZeroCopyOutputStream* output, |
| 629 | bool do_eager_refresh) |
| 630 | : output_(output), |
| 631 | buffer_(NULL), |
| 632 | buffer_size_(0), |
| 633 | total_bytes_(0), |
| 634 | had_error_(false), |
Austin Schuh | 40c1652 | 2018-10-28 20:27:54 -0700 | [diff] [blame^] | 635 | aliasing_enabled_(false), |
| 636 | is_serialization_deterministic_(IsDefaultSerializationDeterministic()) { |
Brian Silverman | 9c614bc | 2016-02-15 20:20:02 -0500 | [diff] [blame] | 637 | if (do_eager_refresh) { |
| 638 | // Eagerly Refresh() so buffer space is immediately available. |
| 639 | Refresh(); |
| 640 | // The Refresh() may have failed. If the client doesn't write any data, |
| 641 | // though, don't consider this an error. If the client does write data, then |
| 642 | // another Refresh() will be attempted and it will set the error once again. |
| 643 | had_error_ = false; |
| 644 | } |
| 645 | } |
| 646 | |
| 647 | CodedOutputStream::~CodedOutputStream() { |
| 648 | Trim(); |
| 649 | } |
| 650 | |
| 651 | void CodedOutputStream::Trim() { |
| 652 | if (buffer_size_ > 0) { |
| 653 | output_->BackUp(buffer_size_); |
| 654 | total_bytes_ -= buffer_size_; |
| 655 | buffer_size_ = 0; |
| 656 | buffer_ = NULL; |
| 657 | } |
| 658 | } |
| 659 | |
| 660 | bool CodedOutputStream::Skip(int count) { |
| 661 | if (count < 0) return false; |
| 662 | |
| 663 | while (count > buffer_size_) { |
| 664 | count -= buffer_size_; |
| 665 | if (!Refresh()) return false; |
| 666 | } |
| 667 | |
| 668 | Advance(count); |
| 669 | return true; |
| 670 | } |
| 671 | |
| 672 | bool CodedOutputStream::GetDirectBufferPointer(void** data, int* size) { |
| 673 | if (buffer_size_ == 0 && !Refresh()) return false; |
| 674 | |
| 675 | *data = buffer_; |
| 676 | *size = buffer_size_; |
| 677 | return true; |
| 678 | } |
| 679 | |
| 680 | void CodedOutputStream::WriteRaw(const void* data, int size) { |
| 681 | while (buffer_size_ < size) { |
| 682 | memcpy(buffer_, data, buffer_size_); |
| 683 | size -= buffer_size_; |
| 684 | data = reinterpret_cast<const uint8*>(data) + buffer_size_; |
| 685 | if (!Refresh()) return; |
| 686 | } |
| 687 | |
| 688 | memcpy(buffer_, data, size); |
| 689 | Advance(size); |
| 690 | } |
| 691 | |
| 692 | uint8* CodedOutputStream::WriteRawToArray( |
| 693 | const void* data, int size, uint8* target) { |
| 694 | memcpy(target, data, size); |
| 695 | return target + size; |
| 696 | } |
| 697 | |
| 698 | |
| 699 | void CodedOutputStream::WriteAliasedRaw(const void* data, int size) { |
| 700 | if (size < buffer_size_ |
| 701 | ) { |
| 702 | WriteRaw(data, size); |
| 703 | } else { |
| 704 | Trim(); |
| 705 | |
| 706 | total_bytes_ += size; |
| 707 | had_error_ |= !output_->WriteAliasedRaw(data, size); |
| 708 | } |
| 709 | } |
| 710 | |
| 711 | void CodedOutputStream::WriteLittleEndian32(uint32 value) { |
| 712 | uint8 bytes[sizeof(value)]; |
| 713 | |
| 714 | bool use_fast = buffer_size_ >= sizeof(value); |
| 715 | uint8* ptr = use_fast ? buffer_ : bytes; |
| 716 | |
| 717 | WriteLittleEndian32ToArray(value, ptr); |
| 718 | |
| 719 | if (use_fast) { |
| 720 | Advance(sizeof(value)); |
| 721 | } else { |
| 722 | WriteRaw(bytes, sizeof(value)); |
| 723 | } |
| 724 | } |
| 725 | |
| 726 | void CodedOutputStream::WriteLittleEndian64(uint64 value) { |
| 727 | uint8 bytes[sizeof(value)]; |
| 728 | |
| 729 | bool use_fast = buffer_size_ >= sizeof(value); |
| 730 | uint8* ptr = use_fast ? buffer_ : bytes; |
| 731 | |
| 732 | WriteLittleEndian64ToArray(value, ptr); |
| 733 | |
| 734 | if (use_fast) { |
| 735 | Advance(sizeof(value)); |
| 736 | } else { |
| 737 | WriteRaw(bytes, sizeof(value)); |
| 738 | } |
| 739 | } |
| 740 | |
| 741 | void CodedOutputStream::WriteVarint32SlowPath(uint32 value) { |
| 742 | uint8 bytes[kMaxVarint32Bytes]; |
| 743 | uint8* target = &bytes[0]; |
| 744 | uint8* end = WriteVarint32ToArray(value, target); |
| 745 | int size = end - target; |
| 746 | WriteRaw(bytes, size); |
| 747 | } |
| 748 | |
Austin Schuh | 40c1652 | 2018-10-28 20:27:54 -0700 | [diff] [blame^] | 749 | void CodedOutputStream::WriteVarint64SlowPath(uint64 value) { |
| 750 | uint8 bytes[kMaxVarintBytes]; |
| 751 | uint8* target = &bytes[0]; |
| 752 | uint8* end = WriteVarint64ToArray(value, target); |
| 753 | int size = end - target; |
| 754 | WriteRaw(bytes, size); |
Brian Silverman | 9c614bc | 2016-02-15 20:20:02 -0500 | [diff] [blame] | 755 | } |
| 756 | |
| 757 | bool CodedOutputStream::Refresh() { |
| 758 | void* void_buffer; |
| 759 | if (output_->Next(&void_buffer, &buffer_size_)) { |
| 760 | buffer_ = reinterpret_cast<uint8*>(void_buffer); |
| 761 | total_bytes_ += buffer_size_; |
| 762 | return true; |
| 763 | } else { |
| 764 | buffer_ = NULL; |
| 765 | buffer_size_ = 0; |
| 766 | had_error_ = true; |
| 767 | return false; |
| 768 | } |
| 769 | } |
| 770 | |
Brian Silverman | 9c614bc | 2016-02-15 20:20:02 -0500 | [diff] [blame] | 771 | uint8* CodedOutputStream::WriteStringWithSizeToArray(const string& str, |
| 772 | uint8* target) { |
| 773 | GOOGLE_DCHECK_LE(str.size(), kuint32max); |
| 774 | target = WriteVarint32ToArray(str.size(), target); |
| 775 | return WriteStringToArray(str, target); |
| 776 | } |
| 777 | |
| 778 | } // namespace io |
| 779 | } // namespace protobuf |
| 780 | } // namespace google |