Brian Silverman | f59fe3f | 2020-09-22 21:04:09 -0700 | [diff] [blame] | 1 | #include "aos/events/logging/lzma_encoder.h" |
| 2 | |
| 3 | #include "glog/logging.h" |
| 4 | |
Austin Schuh | be91b34 | 2022-06-27 00:53:45 -0700 | [diff] [blame^] | 5 | DEFINE_int32(lzma_threads, 1, "Number of threads to use for encoding"); |
| 6 | |
Brian Silverman | f59fe3f | 2020-09-22 21:04:09 -0700 | [diff] [blame] | 7 | namespace aos::logger { |
| 8 | namespace { |
| 9 | |
Austin Schuh | 3bd4c40 | 2020-11-06 18:19:06 -0800 | [diff] [blame] | 10 | // Returns true if `status` is not an error code, false if it is recoverable, or |
| 11 | // otherwise logs the appropriate error message and crashes. |
Austin Schuh | ed292dc | 2020-12-22 22:32:59 -0800 | [diff] [blame] | 12 | bool LzmaCodeIsOk(lzma_ret status, std::string_view filename = "") { |
Brian Silverman | f59fe3f | 2020-09-22 21:04:09 -0700 | [diff] [blame] | 13 | switch (status) { |
| 14 | case LZMA_OK: |
| 15 | case LZMA_STREAM_END: |
Austin Schuh | 3bd4c40 | 2020-11-06 18:19:06 -0800 | [diff] [blame] | 16 | return true; |
Brian Silverman | f59fe3f | 2020-09-22 21:04:09 -0700 | [diff] [blame] | 17 | case LZMA_MEM_ERROR: |
| 18 | LOG(FATAL) << "Memory allocation failed:" << status; |
| 19 | case LZMA_OPTIONS_ERROR: |
| 20 | LOG(FATAL) << "The given compression preset or decompression options are " |
| 21 | "not supported: " |
| 22 | << status; |
| 23 | case LZMA_UNSUPPORTED_CHECK: |
| 24 | LOG(FATAL) << "The given check type is not supported: " << status; |
| 25 | case LZMA_PROG_ERROR: |
| 26 | LOG(FATAL) << "One or more of the parameters have values that will never " |
| 27 | "be valid: " |
| 28 | << status; |
| 29 | case LZMA_MEMLIMIT_ERROR: |
| 30 | LOG(FATAL) << "Decoder needs more memory than allowed by the specified " |
| 31 | "memory usage limit: " |
| 32 | << status; |
| 33 | case LZMA_FORMAT_ERROR: |
Austin Schuh | ed292dc | 2020-12-22 22:32:59 -0800 | [diff] [blame] | 34 | if (filename.empty()) { |
| 35 | LOG(FATAL) << "File format not recognized: " << status; |
| 36 | } else { |
| 37 | LOG(FATAL) << "File format of " << filename |
| 38 | << " not recognized: " << status; |
| 39 | } |
Brian Silverman | f59fe3f | 2020-09-22 21:04:09 -0700 | [diff] [blame] | 40 | case LZMA_DATA_ERROR: |
Brian Silverman | 517431e | 2021-11-10 12:48:58 -0800 | [diff] [blame] | 41 | VLOG(1) << "Compressed file is corrupt: " << status; |
Austin Schuh | 3bd4c40 | 2020-11-06 18:19:06 -0800 | [diff] [blame] | 42 | return false; |
Brian Silverman | f59fe3f | 2020-09-22 21:04:09 -0700 | [diff] [blame] | 43 | case LZMA_BUF_ERROR: |
Brian Silverman | 517431e | 2021-11-10 12:48:58 -0800 | [diff] [blame] | 44 | VLOG(1) << "Compressed file is truncated or corrupt: " << status; |
Austin Schuh | 3bd4c40 | 2020-11-06 18:19:06 -0800 | [diff] [blame] | 45 | return false; |
Brian Silverman | f59fe3f | 2020-09-22 21:04:09 -0700 | [diff] [blame] | 46 | default: |
| 47 | LOG(FATAL) << "Unexpected return value: " << status; |
| 48 | } |
| 49 | } |
| 50 | |
| 51 | } // namespace |
| 52 | |
Austin Schuh | be91b34 | 2022-06-27 00:53:45 -0700 | [diff] [blame^] | 53 | LzmaEncoder::LzmaEncoder(const uint32_t compression_preset, size_t block_size) |
Brian Silverman | f59fe3f | 2020-09-22 21:04:09 -0700 | [diff] [blame] | 54 | : stream_(LZMA_STREAM_INIT), compression_preset_(compression_preset) { |
| 55 | CHECK_GE(compression_preset_, 0u) |
| 56 | << ": Compression preset must be in the range [0, 9]."; |
| 57 | CHECK_LE(compression_preset_, 9u) |
| 58 | << ": Compression preset must be in the range [0, 9]."; |
| 59 | |
Austin Schuh | be91b34 | 2022-06-27 00:53:45 -0700 | [diff] [blame^] | 60 | if (FLAGS_lzma_threads <= 1) { |
| 61 | lzma_ret status = |
| 62 | lzma_easy_encoder(&stream_, compression_preset_, LZMA_CHECK_CRC64); |
| 63 | CHECK(LzmaCodeIsOk(status)); |
| 64 | } else { |
| 65 | lzma_mt mt_options; |
| 66 | memset(&mt_options, 0, sizeof(mt_options)); |
| 67 | mt_options.threads = FLAGS_lzma_threads; |
| 68 | mt_options.block_size = block_size; |
| 69 | // Compress for at most 100 ms before relinquishing control back to the main |
| 70 | // thread. |
| 71 | mt_options.timeout = 100; |
| 72 | mt_options.preset = compression_preset_; |
| 73 | mt_options.filters = nullptr; |
| 74 | mt_options.check = LZMA_CHECK_CRC64; |
| 75 | lzma_ret status = lzma_stream_encoder_mt(&stream_, &mt_options); |
| 76 | CHECK(LzmaCodeIsOk(status)); |
| 77 | } |
| 78 | |
Brian Silverman | f59fe3f | 2020-09-22 21:04:09 -0700 | [diff] [blame] | 79 | stream_.avail_out = 0; |
| 80 | VLOG(2) << "LzmaEncoder: Initialization succeeded."; |
| 81 | } |
| 82 | |
| 83 | LzmaEncoder::~LzmaEncoder() { lzma_end(&stream_); } |
| 84 | |
| 85 | void LzmaEncoder::Encode(flatbuffers::DetachedBuffer &&in) { |
| 86 | CHECK(in.data()) << ": Encode called with nullptr."; |
| 87 | |
| 88 | stream_.next_in = in.data(); |
| 89 | stream_.avail_in = in.size(); |
| 90 | |
| 91 | RunLzmaCode(LZMA_RUN); |
| 92 | } |
| 93 | |
| 94 | void LzmaEncoder::Finish() { RunLzmaCode(LZMA_FINISH); } |
| 95 | |
| 96 | void LzmaEncoder::Clear(const int n) { |
| 97 | CHECK_GE(n, 0); |
| 98 | CHECK_LE(static_cast<size_t>(n), queue_size()); |
| 99 | queue_.erase(queue_.begin(), queue_.begin() + n); |
| 100 | if (queue_.empty()) { |
| 101 | stream_.next_out = nullptr; |
| 102 | stream_.avail_out = 0; |
| 103 | } |
| 104 | } |
| 105 | |
| 106 | std::vector<absl::Span<const uint8_t>> LzmaEncoder::queue() const { |
| 107 | std::vector<absl::Span<const uint8_t>> queue; |
| 108 | if (queue_.empty()) { |
| 109 | return queue; |
| 110 | } |
| 111 | for (size_t i = 0; i < queue_.size() - 1; ++i) { |
| 112 | queue.emplace_back( |
| 113 | absl::MakeConstSpan(queue_.at(i).data(), queue_.at(i).size())); |
| 114 | } |
| 115 | // For the last buffer in the queue, we must account for the possibility that |
| 116 | // the buffer isn't full yet. |
| 117 | queue.emplace_back(absl::MakeConstSpan( |
| 118 | queue_.back().data(), queue_.back().size() - stream_.avail_out)); |
| 119 | return queue; |
| 120 | } |
| 121 | |
| 122 | size_t LzmaEncoder::queued_bytes() const { |
| 123 | size_t bytes = queue_size() * kEncodedBufferSizeBytes; |
| 124 | // Subtract the bytes that the encoder hasn't filled yet. |
| 125 | bytes -= stream_.avail_out; |
| 126 | return bytes; |
| 127 | } |
| 128 | |
| 129 | void LzmaEncoder::RunLzmaCode(lzma_action action) { |
| 130 | CHECK(!finished_); |
| 131 | |
| 132 | // This is to keep track of how many bytes resulted from encoding this input |
| 133 | // buffer. |
| 134 | size_t last_avail_out = stream_.avail_out; |
| 135 | |
| 136 | while (stream_.avail_in > 0 || action == LZMA_FINISH) { |
| 137 | // If output buffer is full, create a new one, queue it up, and resume |
| 138 | // encoding. This could happen in the first call to Encode after |
| 139 | // construction or a Reset, or when an input buffer is large enough to fill |
| 140 | // more than one output buffer. |
| 141 | if (stream_.avail_out == 0) { |
| 142 | queue_.emplace_back(); |
| 143 | queue_.back().resize(kEncodedBufferSizeBytes); |
| 144 | stream_.next_out = queue_.back().data(); |
| 145 | stream_.avail_out = kEncodedBufferSizeBytes; |
| 146 | // Update the byte count. |
| 147 | total_bytes_ += last_avail_out; |
| 148 | last_avail_out = stream_.avail_out; |
| 149 | } |
| 150 | |
| 151 | // Encode the data. |
| 152 | lzma_ret status = lzma_code(&stream_, action); |
Austin Schuh | 3bd4c40 | 2020-11-06 18:19:06 -0800 | [diff] [blame] | 153 | CHECK(LzmaCodeIsOk(status)); |
Brian Silverman | f59fe3f | 2020-09-22 21:04:09 -0700 | [diff] [blame] | 154 | if (action == LZMA_FINISH) { |
| 155 | if (status == LZMA_STREAM_END) { |
| 156 | // This is returned when lzma_code is all done. |
| 157 | finished_ = true; |
| 158 | break; |
| 159 | } |
| 160 | } else { |
| 161 | CHECK(status != LZMA_STREAM_END); |
| 162 | } |
| 163 | VLOG(2) << "LzmaEncoder: Encoded chunk."; |
| 164 | } |
| 165 | |
| 166 | // Update the number of resulting encoded bytes. |
| 167 | total_bytes_ += last_avail_out - stream_.avail_out; |
| 168 | } |
| 169 | |
Austin Schuh | cd36842 | 2021-11-22 21:23:29 -0800 | [diff] [blame] | 170 | LzmaDecoder::LzmaDecoder(std::unique_ptr<DataDecoder> underlying_decoder, |
| 171 | bool quiet) |
Tyler Chatow | 2015bc6 | 2021-08-04 21:15:09 -0700 | [diff] [blame] | 172 | : underlying_decoder_(std::move(underlying_decoder)), |
Austin Schuh | cd36842 | 2021-11-22 21:23:29 -0800 | [diff] [blame] | 173 | stream_(LZMA_STREAM_INIT), |
| 174 | quiet_(quiet) { |
Brian Silverman | f59fe3f | 2020-09-22 21:04:09 -0700 | [diff] [blame] | 175 | compressed_data_.resize(kBufSize); |
| 176 | |
| 177 | lzma_ret status = |
| 178 | lzma_stream_decoder(&stream_, UINT64_MAX, LZMA_CONCATENATED); |
Austin Schuh | 3bd4c40 | 2020-11-06 18:19:06 -0800 | [diff] [blame] | 179 | CHECK(LzmaCodeIsOk(status)) << "Failed initializing LZMA stream decoder."; |
Brian Silverman | f59fe3f | 2020-09-22 21:04:09 -0700 | [diff] [blame] | 180 | stream_.avail_out = 0; |
| 181 | VLOG(2) << "LzmaDecoder: Initialization succeeded."; |
| 182 | } |
| 183 | |
| 184 | LzmaDecoder::~LzmaDecoder() { lzma_end(&stream_); } |
| 185 | |
| 186 | size_t LzmaDecoder::Read(uint8_t *begin, uint8_t *end) { |
| 187 | if (finished_) { |
| 188 | return 0; |
| 189 | } |
| 190 | |
| 191 | // Write into the given range. |
| 192 | stream_.next_out = begin; |
| 193 | stream_.avail_out = end - begin; |
| 194 | // Keep decompressing until we run out of buffer space. |
| 195 | while (stream_.avail_out > 0) { |
| 196 | if (action_ == LZMA_RUN && stream_.avail_in == 0) { |
| 197 | // Read more bytes from the file if we're all out. |
Tyler Chatow | 2015bc6 | 2021-08-04 21:15:09 -0700 | [diff] [blame] | 198 | const size_t count = underlying_decoder_->Read(compressed_data_.begin(), |
| 199 | compressed_data_.end()); |
Brian Silverman | f59fe3f | 2020-09-22 21:04:09 -0700 | [diff] [blame] | 200 | if (count == 0) { |
| 201 | // No more data to read in the file, begin the finishing operation. |
| 202 | action_ = LZMA_FINISH; |
| 203 | } else { |
| 204 | stream_.next_in = compressed_data_.data(); |
| 205 | stream_.avail_in = count; |
| 206 | } |
| 207 | } |
| 208 | // Decompress the data. |
| 209 | const lzma_ret status = lzma_code(&stream_, action_); |
| 210 | // Return if we're done. |
| 211 | if (status == LZMA_STREAM_END) { |
| 212 | CHECK_EQ(action_, LZMA_FINISH) |
| 213 | << ": Got LZMA_STREAM_END when action wasn't LZMA_FINISH"; |
| 214 | finished_ = true; |
| 215 | return (end - begin) - stream_.avail_out; |
| 216 | } |
Austin Schuh | 3bd4c40 | 2020-11-06 18:19:06 -0800 | [diff] [blame] | 217 | |
| 218 | // If we fail to decompress, give up. Return everything that has been |
| 219 | // produced so far. |
Tyler Chatow | 2015bc6 | 2021-08-04 21:15:09 -0700 | [diff] [blame] | 220 | if (!LzmaCodeIsOk(status, filename())) { |
Austin Schuh | 3bd4c40 | 2020-11-06 18:19:06 -0800 | [diff] [blame] | 221 | finished_ = true; |
Brian Silverman | 517431e | 2021-11-10 12:48:58 -0800 | [diff] [blame] | 222 | if (status == LZMA_DATA_ERROR) { |
Austin Schuh | cd36842 | 2021-11-22 21:23:29 -0800 | [diff] [blame] | 223 | if (!quiet_ || VLOG_IS_ON(1)) { |
| 224 | LOG(WARNING) << filename() << " is corrupted."; |
| 225 | } |
Brian Silverman | 517431e | 2021-11-10 12:48:58 -0800 | [diff] [blame] | 226 | } else if (status == LZMA_BUF_ERROR) { |
Austin Schuh | cd36842 | 2021-11-22 21:23:29 -0800 | [diff] [blame] | 227 | if (!quiet_ || VLOG_IS_ON(1)) { |
| 228 | LOG(WARNING) << filename() << " is truncated or corrupted."; |
| 229 | } |
Brian Silverman | 517431e | 2021-11-10 12:48:58 -0800 | [diff] [blame] | 230 | } else { |
| 231 | LOG(FATAL) << "Unknown error " << status << " when reading " |
| 232 | << filename(); |
| 233 | } |
Austin Schuh | 3bd4c40 | 2020-11-06 18:19:06 -0800 | [diff] [blame] | 234 | return (end - begin) - stream_.avail_out; |
| 235 | } |
Brian Silverman | f59fe3f | 2020-09-22 21:04:09 -0700 | [diff] [blame] | 236 | } |
| 237 | return end - begin; |
| 238 | } |
| 239 | |
Tyler Chatow | 2015bc6 | 2021-08-04 21:15:09 -0700 | [diff] [blame] | 240 | ThreadedLzmaDecoder::ThreadedLzmaDecoder( |
Austin Schuh | cd36842 | 2021-11-22 21:23:29 -0800 | [diff] [blame] | 241 | std::unique_ptr<DataDecoder> underlying_decoder, bool quiet) |
| 242 | : decoder_(std::move(underlying_decoder), quiet), decode_thread_([this] { |
Tyler Chatow | 7df6083 | 2021-07-15 21:18:36 -0700 | [diff] [blame] | 243 | std::unique_lock lock(decode_mutex_); |
| 244 | while (true) { |
| 245 | // Wake if the queue is too small or we are finished. |
| 246 | continue_decoding_.wait(lock, [this] { |
| 247 | return decoded_queue_.size() < kQueueSize || finished_; |
| 248 | }); |
| 249 | |
| 250 | if (finished_) { |
| 251 | return; |
| 252 | } |
| 253 | |
| 254 | while (true) { |
| 255 | CHECK(!finished_); |
| 256 | // Release our lock on the queue before doing decompression work. |
| 257 | lock.unlock(); |
| 258 | |
| 259 | ResizeableBuffer buffer; |
| 260 | buffer.resize(kBufSize); |
| 261 | |
| 262 | const size_t bytes_read = |
| 263 | decoder_.Read(buffer.begin(), buffer.end()); |
| 264 | buffer.resize(bytes_read); |
| 265 | |
| 266 | // Relock the queue and move the new buffer to the end. This should |
| 267 | // be fast. We also need to stay locked when we wait(). |
| 268 | lock.lock(); |
| 269 | if (bytes_read > 0) { |
| 270 | decoded_queue_.emplace_back(std::move(buffer)); |
| 271 | } else { |
| 272 | finished_ = true; |
| 273 | } |
| 274 | |
| 275 | // If we've filled the queue or are out of data, go back to sleep. |
| 276 | if (decoded_queue_.size() >= kQueueSize || finished_) { |
| 277 | break; |
| 278 | } |
| 279 | } |
| 280 | |
| 281 | // Notify main thread in case it was waiting for us to queue more |
| 282 | // data. |
| 283 | queue_filled_.notify_one(); |
| 284 | } |
| 285 | }) {} |
| 286 | |
| 287 | ThreadedLzmaDecoder::~ThreadedLzmaDecoder() { |
| 288 | // Wake up decode thread so it can return. |
| 289 | { |
| 290 | std::scoped_lock lock(decode_mutex_); |
| 291 | finished_ = true; |
| 292 | } |
| 293 | continue_decoding_.notify_one(); |
| 294 | decode_thread_.join(); |
| 295 | } |
| 296 | |
| 297 | size_t ThreadedLzmaDecoder::Read(uint8_t *begin, uint8_t *end) { |
| 298 | std::unique_lock lock(decode_mutex_); |
| 299 | |
| 300 | // Strip any empty buffers |
| 301 | for (auto iter = decoded_queue_.begin(); iter != decoded_queue_.end();) { |
| 302 | if (iter->size() == 0) { |
| 303 | iter = decoded_queue_.erase(iter); |
| 304 | } else { |
| 305 | ++iter; |
| 306 | } |
| 307 | } |
| 308 | |
| 309 | // If the queue is empty, sleep until the decoder thread has produced another |
| 310 | // buffer. |
| 311 | if (decoded_queue_.empty()) { |
| 312 | continue_decoding_.notify_one(); |
| 313 | queue_filled_.wait(lock, |
| 314 | [this] { return finished_ || !decoded_queue_.empty(); }); |
| 315 | if (finished_ && decoded_queue_.empty()) { |
| 316 | return 0; |
| 317 | } |
| 318 | } |
| 319 | // Sanity check if the queue is empty and we're not finished. |
| 320 | CHECK(!decoded_queue_.empty()) << "Decoded queue unexpectedly empty"; |
| 321 | |
| 322 | ResizeableBuffer &front_buffer = decoded_queue_.front(); |
| 323 | |
| 324 | // Copy some data from our working buffer to the requested destination. |
| 325 | const std::size_t bytes_requested = end - begin; |
| 326 | const std::size_t bytes_to_copy = |
| 327 | std::min(bytes_requested, front_buffer.size()); |
| 328 | memcpy(begin, front_buffer.data(), bytes_to_copy); |
| 329 | front_buffer.erase_front(bytes_to_copy); |
| 330 | |
| 331 | // Ensure the decoding thread wakes up if the queue isn't full. |
| 332 | if (!finished_ && decoded_queue_.size() < kQueueSize) { |
| 333 | continue_decoding_.notify_one(); |
| 334 | } |
| 335 | |
| 336 | return bytes_to_copy; |
| 337 | } |
| 338 | |
Brian Silverman | f59fe3f | 2020-09-22 21:04:09 -0700 | [diff] [blame] | 339 | } // namespace aos::logger |