| 1 | // Copyright (c) 2013-2014 Sandstorm Development Group, Inc. and contributors |
| 2 | // Licensed under the MIT License: |
| 3 | // |
| 4 | // Permission is hereby granted, free of charge, to any person obtaining a copy |
| 5 | // of this software and associated documentation files (the "Software"), to deal |
| 6 | // in the Software without restriction, including without limitation the rights |
| 7 | // to use, copy, modify, merge, publish, distribute, sublicense, and/or sell |
| 8 | // copies of the Software, and to permit persons to whom the Software is |
| 9 | // furnished to do so, subject to the following conditions: |
| 10 | // |
| 11 | // The above copyright notice and this permission notice shall be included in |
| 12 | // all copies or substantial portions of the Software. |
| 13 | // |
| 14 | // THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| 15 | // IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| 16 | // FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE |
| 17 | // AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
| 18 | // LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
| 19 | // OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN |
| 20 | // THE SOFTWARE. |
| 21 | |
| 22 | #include "serialize.h" |
| 23 | #include "layout.h" |
| 24 | #include <kj/debug.h> |
| 25 | #include <exception> |
| 26 | |
| 27 | namespace capnp { |
| 28 | |
| 29 | UnalignedFlatArrayMessageReader::UnalignedFlatArrayMessageReader( |
| 30 | kj::ArrayPtr<const word> array, ReaderOptions options) |
| 31 | : MessageReader(options), end(array.end()) { |
| 32 | if (array.size() < 1) { |
| 33 | // Assume empty message. |
| 34 | return; |
| 35 | } |
| 36 | |
| 37 | const _::WireValue<uint32_t>* table = |
| 38 | reinterpret_cast<const _::WireValue<uint32_t>*>(array.begin()); |
| 39 | |
| 40 | uint segmentCount = table[0].get() + 1; |
| 41 | size_t offset = segmentCount / 2u + 1u; |
| 42 | |
| 43 | KJ_REQUIRE(array.size() >= offset, "Message ends prematurely in segment table." ) { |
| 44 | return; |
| 45 | } |
| 46 | |
| 47 | { |
| 48 | uint segmentSize = table[1].get(); |
| 49 | |
| 50 | KJ_REQUIRE(array.size() >= offset + segmentSize, |
| 51 | "Message ends prematurely in first segment." ) { |
| 52 | return; |
| 53 | } |
| 54 | |
| 55 | segment0 = array.slice(offset, offset + segmentSize); |
| 56 | offset += segmentSize; |
| 57 | } |
| 58 | |
| 59 | if (segmentCount > 1) { |
| 60 | moreSegments = kj::heapArray<kj::ArrayPtr<const word>>(segmentCount - 1); |
| 61 | |
| 62 | for (uint i = 1; i < segmentCount; i++) { |
| 63 | uint segmentSize = table[i + 1].get(); |
| 64 | |
| 65 | KJ_REQUIRE(array.size() >= offset + segmentSize, "Message ends prematurely." ) { |
| 66 | moreSegments = nullptr; |
| 67 | return; |
| 68 | } |
| 69 | |
| 70 | moreSegments[i - 1] = array.slice(offset, offset + segmentSize); |
| 71 | offset += segmentSize; |
| 72 | } |
| 73 | } |
| 74 | |
| 75 | end = array.begin() + offset; |
| 76 | } |
| 77 | |
| 78 | size_t expectedSizeInWordsFromPrefix(kj::ArrayPtr<const word> array) { |
| 79 | if (array.size() < 1) { |
| 80 | // All messages are at least one word. |
| 81 | return 1; |
| 82 | } |
| 83 | |
| 84 | const _::WireValue<uint32_t>* table = |
| 85 | reinterpret_cast<const _::WireValue<uint32_t>*>(array.begin()); |
| 86 | |
| 87 | uint segmentCount = table[0].get() + 1; |
| 88 | size_t offset = segmentCount / 2u + 1u; |
| 89 | |
| 90 | // If the array is too small to contain the full segment table, truncate segmentCount to just |
| 91 | // what is available. |
| 92 | segmentCount = kj::min(segmentCount, array.size() * 2 - 1u); |
| 93 | |
| 94 | size_t totalSize = offset; |
| 95 | for (uint i = 0; i < segmentCount; i++) { |
| 96 | totalSize += table[i + 1].get(); |
| 97 | } |
| 98 | return totalSize; |
| 99 | } |
| 100 | |
| 101 | kj::ArrayPtr<const word> UnalignedFlatArrayMessageReader::getSegment(uint id) { |
| 102 | if (id == 0) { |
| 103 | return segment0; |
| 104 | } else if (id <= moreSegments.size()) { |
| 105 | return moreSegments[id - 1]; |
| 106 | } else { |
| 107 | return nullptr; |
| 108 | } |
| 109 | } |
| 110 | |
| 111 | kj::ArrayPtr<const word> FlatArrayMessageReader::checkAlignment(kj::ArrayPtr<const word> array) { |
| 112 | KJ_REQUIRE((uintptr_t)array.begin() % sizeof(void*) == 0, |
| 113 | "Input to FlatArrayMessageReader is not aligned. If your architecture supports unaligned " |
| 114 | "access (e.g. x86/x64/modern ARM), you may use UnalignedFlatArrayMessageReader instead, " |
| 115 | "though this may harm performance." ); |
| 116 | |
| 117 | return array; |
| 118 | } |
| 119 | |
| 120 | kj::ArrayPtr<const word> initMessageBuilderFromFlatArrayCopy( |
| 121 | kj::ArrayPtr<const word> array, MessageBuilder& target, ReaderOptions options) { |
| 122 | FlatArrayMessageReader reader(array, options); |
| 123 | target.setRoot(reader.getRoot<AnyPointer>()); |
| 124 | return kj::arrayPtr(reader.getEnd(), array.end()); |
| 125 | } |
| 126 | |
| 127 | kj::Array<word> messageToFlatArray(kj::ArrayPtr<const kj::ArrayPtr<const word>> segments) { |
| 128 | kj::Array<word> result = kj::heapArray<word>(computeSerializedSizeInWords(segments)); |
| 129 | |
| 130 | _::WireValue<uint32_t>* table = |
| 131 | reinterpret_cast<_::WireValue<uint32_t>*>(result.begin()); |
| 132 | |
| 133 | // We write the segment count - 1 because this makes the first word zero for single-segment |
| 134 | // messages, improving compression. We don't bother doing this with segment sizes because |
| 135 | // one-word segments are rare anyway. |
| 136 | table[0].set(segments.size() - 1); |
| 137 | |
| 138 | for (uint i = 0; i < segments.size(); i++) { |
| 139 | table[i + 1].set(segments[i].size()); |
| 140 | } |
| 141 | |
| 142 | if (segments.size() % 2 == 0) { |
| 143 | // Set padding byte. |
| 144 | table[segments.size() + 1].set(0); |
| 145 | } |
| 146 | |
| 147 | word* dst = result.begin() + segments.size() / 2 + 1; |
| 148 | |
| 149 | for (auto& segment: segments) { |
| 150 | memcpy(dst, segment.begin(), segment.size() * sizeof(word)); |
| 151 | dst += segment.size(); |
| 152 | } |
| 153 | |
| 154 | KJ_DASSERT(dst == result.end(), "Buffer overrun/underrun bug in code above." ); |
| 155 | |
| 156 | return kj::mv(result); |
| 157 | } |
| 158 | |
| 159 | size_t computeSerializedSizeInWords(kj::ArrayPtr<const kj::ArrayPtr<const word>> segments) { |
| 160 | KJ_REQUIRE(segments.size() > 0, "Tried to serialize uninitialized message." ); |
| 161 | |
| 162 | size_t totalSize = segments.size() / 2 + 1; |
| 163 | |
| 164 | for (auto& segment: segments) { |
| 165 | totalSize += segment.size(); |
| 166 | } |
| 167 | |
| 168 | return totalSize; |
| 169 | } |
| 170 | |
| 171 | // ======================================================================================= |
| 172 | |
| 173 | InputStreamMessageReader::InputStreamMessageReader( |
| 174 | kj::InputStream& inputStream, ReaderOptions options, kj::ArrayPtr<word> scratchSpace) |
| 175 | : MessageReader(options), inputStream(inputStream), readPos(nullptr) { |
| 176 | _::WireValue<uint32_t> firstWord[2]; |
| 177 | |
| 178 | inputStream.read(firstWord, sizeof(firstWord)); |
| 179 | |
| 180 | uint segmentCount = firstWord[0].get() + 1; |
| 181 | uint segment0Size = segmentCount == 0 ? 0 : firstWord[1].get(); |
| 182 | |
| 183 | size_t totalWords = segment0Size; |
| 184 | |
| 185 | // Reject messages with too many segments for security reasons. |
| 186 | KJ_REQUIRE(segmentCount < 512, "Message has too many segments." ) { |
| 187 | segmentCount = 1; |
| 188 | segment0Size = 1; |
| 189 | break; |
| 190 | } |
| 191 | |
| 192 | // Read sizes for all segments except the first. Include padding if necessary. |
| 193 | KJ_STACK_ARRAY(_::WireValue<uint32_t>, moreSizes, segmentCount & ~1, 16, 64); |
| 194 | if (segmentCount > 1) { |
| 195 | inputStream.read(moreSizes.begin(), moreSizes.size() * sizeof(moreSizes[0])); |
| 196 | for (uint i = 0; i < segmentCount - 1; i++) { |
| 197 | totalWords += moreSizes[i].get(); |
| 198 | } |
| 199 | } |
| 200 | |
| 201 | // Don't accept a message which the receiver couldn't possibly traverse without hitting the |
| 202 | // traversal limit. Without this check, a malicious client could transmit a very large segment |
| 203 | // size to make the receiver allocate excessive space and possibly crash. |
| 204 | KJ_REQUIRE(totalWords <= options.traversalLimitInWords, |
| 205 | "Message is too large. To increase the limit on the receiving end, see " |
| 206 | "capnp::ReaderOptions." ) { |
| 207 | segmentCount = 1; |
| 208 | segment0Size = kj::min(segment0Size, options.traversalLimitInWords); |
| 209 | totalWords = segment0Size; |
| 210 | break; |
| 211 | } |
| 212 | |
| 213 | if (scratchSpace.size() < totalWords) { |
| 214 | // TODO(perf): Consider allocating each segment as a separate chunk to reduce memory |
| 215 | // fragmentation. |
| 216 | ownedSpace = kj::heapArray<word>(totalWords); |
| 217 | scratchSpace = ownedSpace; |
| 218 | } |
| 219 | |
| 220 | segment0 = scratchSpace.slice(0, segment0Size); |
| 221 | |
| 222 | if (segmentCount > 1) { |
| 223 | moreSegments = kj::heapArray<kj::ArrayPtr<const word>>(segmentCount - 1); |
| 224 | size_t offset = segment0Size; |
| 225 | |
| 226 | for (uint i = 0; i < segmentCount - 1; i++) { |
| 227 | uint segmentSize = moreSizes[i].get(); |
| 228 | moreSegments[i] = scratchSpace.slice(offset, offset + segmentSize); |
| 229 | offset += segmentSize; |
| 230 | } |
| 231 | } |
| 232 | |
| 233 | if (segmentCount == 1) { |
| 234 | inputStream.read(scratchSpace.begin(), totalWords * sizeof(word)); |
| 235 | } else if (segmentCount > 1) { |
| 236 | readPos = scratchSpace.asBytes().begin(); |
| 237 | readPos += inputStream.read(readPos, segment0Size * sizeof(word), totalWords * sizeof(word)); |
| 238 | } |
| 239 | } |
| 240 | |
| 241 | InputStreamMessageReader::~InputStreamMessageReader() noexcept(false) { |
| 242 | if (readPos != nullptr) { |
| 243 | unwindDetector.catchExceptionsIfUnwinding([&]() { |
| 244 | // Note that lazy reads only happen when we have multiple segments, so moreSegments.back() is |
| 245 | // valid. |
| 246 | const byte* allEnd = reinterpret_cast<const byte*>(moreSegments.back().end()); |
| 247 | inputStream.skip(allEnd - readPos); |
| 248 | }); |
| 249 | } |
| 250 | } |
| 251 | |
| 252 | kj::ArrayPtr<const word> InputStreamMessageReader::getSegment(uint id) { |
| 253 | if (id > moreSegments.size()) { |
| 254 | return nullptr; |
| 255 | } |
| 256 | |
| 257 | kj::ArrayPtr<const word> segment = id == 0 ? segment0 : moreSegments[id - 1]; |
| 258 | |
| 259 | if (readPos != nullptr) { |
| 260 | // May need to lazily read more data. |
| 261 | const byte* segmentEnd = reinterpret_cast<const byte*>(segment.end()); |
| 262 | if (readPos < segmentEnd) { |
| 263 | // Note that lazy reads only happen when we have multiple segments, so moreSegments.back() is |
| 264 | // valid. |
| 265 | const byte* allEnd = reinterpret_cast<const byte*>(moreSegments.back().end()); |
| 266 | readPos += inputStream.read(readPos, segmentEnd - readPos, allEnd - readPos); |
| 267 | } |
| 268 | } |
| 269 | |
| 270 | return segment; |
| 271 | } |
| 272 | |
| 273 | void readMessageCopy(kj::InputStream& input, MessageBuilder& target, |
| 274 | ReaderOptions options, kj::ArrayPtr<word> scratchSpace) { |
| 275 | InputStreamMessageReader message(input, options, scratchSpace); |
| 276 | target.setRoot(message.getRoot<AnyPointer>()); |
| 277 | } |
| 278 | |
| 279 | // ------------------------------------------------------------------- |
| 280 | |
| 281 | void writeMessage(kj::OutputStream& output, kj::ArrayPtr<const kj::ArrayPtr<const word>> segments) { |
| 282 | KJ_REQUIRE(segments.size() > 0, "Tried to serialize uninitialized message." ); |
| 283 | |
| 284 | KJ_STACK_ARRAY(_::WireValue<uint32_t>, table, (segments.size() + 2) & ~size_t(1), 16, 64); |
| 285 | |
| 286 | // We write the segment count - 1 because this makes the first word zero for single-segment |
| 287 | // messages, improving compression. We don't bother doing this with segment sizes because |
| 288 | // one-word segments are rare anyway. |
| 289 | table[0].set(segments.size() - 1); |
| 290 | for (uint i = 0; i < segments.size(); i++) { |
| 291 | table[i + 1].set(segments[i].size()); |
| 292 | } |
| 293 | if (segments.size() % 2 == 0) { |
| 294 | // Set padding byte. |
| 295 | table[segments.size() + 1].set(0); |
| 296 | } |
| 297 | |
| 298 | KJ_STACK_ARRAY(kj::ArrayPtr<const byte>, pieces, segments.size() + 1, 4, 32); |
| 299 | pieces[0] = table.asBytes(); |
| 300 | |
| 301 | for (uint i = 0; i < segments.size(); i++) { |
| 302 | pieces[i + 1] = segments[i].asBytes(); |
| 303 | } |
| 304 | |
| 305 | output.write(pieces); |
| 306 | } |
| 307 | |
| 308 | // ======================================================================================= |
| 309 | |
| 310 | StreamFdMessageReader::~StreamFdMessageReader() noexcept(false) {} |
| 311 | |
| 312 | void writeMessageToFd(int fd, kj::ArrayPtr<const kj::ArrayPtr<const word>> segments) { |
| 313 | kj::FdOutputStream stream(fd); |
| 314 | writeMessage(stream, segments); |
| 315 | } |
| 316 | |
| 317 | void readMessageCopyFromFd(int fd, MessageBuilder& target, |
| 318 | ReaderOptions options, kj::ArrayPtr<word> scratchSpace) { |
| 319 | kj::FdInputStream stream(fd); |
| 320 | readMessageCopy(stream, target, options, scratchSpace); |
| 321 | } |
| 322 | |
| 323 | } // namespace capnp |
| 324 | |