| 1 | //===-- DataFileCache.cpp -------------------------------------------------===// |
| 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 | |
| 9 | #include "lldb/Core/DataFileCache.h" |
| 10 | #include "lldb/Core/Module.h" |
| 11 | #include "lldb/Core/ModuleList.h" |
| 12 | #include "lldb/Host/FileSystem.h" |
| 13 | #include "lldb/Symbol/ObjectFile.h" |
| 14 | #include "lldb/Utility/DataEncoder.h" |
| 15 | #include "lldb/Utility/LLDBLog.h" |
| 16 | #include "lldb/Utility/Log.h" |
| 17 | #include "llvm/Support/CachePruning.h" |
| 18 | |
| 19 | using namespace lldb_private; |
| 20 | |
| 21 | |
| 22 | llvm::CachePruningPolicy DataFileCache::GetLLDBIndexCachePolicy() { |
| 23 | static llvm::CachePruningPolicy policy; |
| 24 | static llvm::once_flag once_flag; |
| 25 | |
| 26 | llvm::call_once(flag&: once_flag, F: []() { |
| 27 | // Prune the cache based off of the LLDB settings each time we create a |
| 28 | // cache object. |
| 29 | ModuleListProperties &properties = |
| 30 | ModuleList::GetGlobalModuleListProperties(); |
| 31 | // Only scan once an hour. If we have lots of debug sessions we don't want |
| 32 | // to scan this directory too often. A timestamp file is written to the |
| 33 | // directory to ensure different processes don't scan the directory too |
| 34 | // often. This setting doesn't mean that a thread will continually scan the |
| 35 | // cache directory within this process. |
| 36 | policy.Interval = std::chrono::hours(1); |
| 37 | // Get the user settings for pruning. |
| 38 | policy.MaxSizeBytes = properties.GetLLDBIndexCacheMaxByteSize(); |
| 39 | policy.MaxSizePercentageOfAvailableSpace = |
| 40 | properties.GetLLDBIndexCacheMaxPercent(); |
| 41 | policy.Expiration = |
| 42 | std::chrono::hours(properties.GetLLDBIndexCacheExpirationDays() * 24); |
| 43 | }); |
| 44 | return policy; |
| 45 | } |
| 46 | |
| 47 | DataFileCache::DataFileCache(llvm::StringRef path, llvm::CachePruningPolicy policy) { |
| 48 | m_cache_dir.SetPath(path); |
| 49 | pruneCache(Path: path, Policy: policy); |
| 50 | |
| 51 | // This lambda will get called when the data is gotten from the cache and |
| 52 | // also after the data was set for a given key. We only need to take |
| 53 | // ownership of the data if we are geting the data, so we use the |
| 54 | // m_take_ownership member variable to indicate if we need to take |
| 55 | // ownership. |
| 56 | |
| 57 | auto add_buffer = [this](unsigned task, const llvm::Twine &moduleName, |
| 58 | std::unique_ptr<llvm::MemoryBuffer> m) { |
| 59 | if (m_take_ownership) |
| 60 | m_mem_buff_up = std::move(m); |
| 61 | }; |
| 62 | llvm::Expected<llvm::FileCache> cache_or_err = |
| 63 | llvm::localCache(CacheNameRef: "LLDBModuleCache" , TempFilePrefixRef: "lldb-module" , CacheDirectoryPathRef: path, AddBuffer: add_buffer); |
| 64 | if (cache_or_err) |
| 65 | m_cache_callback = std::move(*cache_or_err); |
| 66 | else { |
| 67 | Log *log = GetLog(mask: LLDBLog::Modules); |
| 68 | LLDB_LOG_ERROR(log, cache_or_err.takeError(), |
| 69 | "failed to create lldb index cache directory: {0}" ); |
| 70 | } |
| 71 | } |
| 72 | |
| 73 | std::unique_ptr<llvm::MemoryBuffer> |
| 74 | DataFileCache::GetCachedData(llvm::StringRef key) { |
| 75 | std::lock_guard<std::mutex> guard(m_mutex); |
| 76 | |
| 77 | const unsigned task = 1; |
| 78 | m_take_ownership = true; |
| 79 | // If we call the "m_cache_callback" function and the data is cached, it will |
| 80 | // call the "add_buffer" lambda function from the constructor which will in |
| 81 | // turn take ownership of the member buffer that is passed to the callback and |
| 82 | // put it into a member variable. |
| 83 | llvm::Expected<llvm::AddStreamFn> add_stream_or_err = |
| 84 | m_cache_callback(task, key, "" ); |
| 85 | m_take_ownership = false; |
| 86 | // At this point we either already called the "add_buffer" lambda with |
| 87 | // the data or we haven't. We can tell if we got the cached data by checking |
| 88 | // the add_stream function pointer value below. |
| 89 | if (add_stream_or_err) { |
| 90 | llvm::AddStreamFn &add_stream = *add_stream_or_err; |
| 91 | // If the "add_stream" is nullptr, then the data was cached and we already |
| 92 | // called the "add_buffer" lambda. If it is valid, then if we were to call |
| 93 | // the add_stream function it would cause a cache file to get generated |
| 94 | // and we would be expected to fill in the data. In this function we only |
| 95 | // want to check if the data was cached, so we don't want to call |
| 96 | // "add_stream" in this function. |
| 97 | if (!add_stream) |
| 98 | return std::move(m_mem_buff_up); |
| 99 | } else { |
| 100 | Log *log = GetLog(mask: LLDBLog::Modules); |
| 101 | LLDB_LOG_ERROR(log, add_stream_or_err.takeError(), |
| 102 | "failed to get the cache add stream callback for key: {0}" ); |
| 103 | } |
| 104 | // Data was not cached. |
| 105 | return std::unique_ptr<llvm::MemoryBuffer>(); |
| 106 | } |
| 107 | |
| 108 | bool DataFileCache::SetCachedData(llvm::StringRef key, |
| 109 | llvm::ArrayRef<uint8_t> data) { |
| 110 | std::lock_guard<std::mutex> guard(m_mutex); |
| 111 | const unsigned task = 2; |
| 112 | // If we call this function and the data is cached, it will call the |
| 113 | // add_buffer lambda function from the constructor which will ignore the |
| 114 | // data. |
| 115 | llvm::Expected<llvm::AddStreamFn> add_stream_or_err = |
| 116 | m_cache_callback(task, key, "" ); |
| 117 | // If we reach this code then we either already called the callback with |
| 118 | // the data or we haven't. We can tell if we had the cached data by checking |
| 119 | // the CacheAddStream function pointer value below. |
| 120 | if (add_stream_or_err) { |
| 121 | llvm::AddStreamFn &add_stream = *add_stream_or_err; |
| 122 | // If the "add_stream" is nullptr, then the data was cached. If it is |
| 123 | // valid, then if we call the add_stream function with a task it will |
| 124 | // cause the file to get generated, but we only want to check if the data |
| 125 | // is cached here, so we don't want to call it here. Note that the |
| 126 | // add_buffer will also get called in this case after the data has been |
| 127 | // provided, but we won't take ownership of the memory buffer as we just |
| 128 | // want to write the data. |
| 129 | if (add_stream) { |
| 130 | llvm::Expected<std::unique_ptr<llvm::CachedFileStream>> file_or_err = |
| 131 | add_stream(task, "" ); |
| 132 | if (file_or_err) { |
| 133 | llvm::CachedFileStream *cfs = file_or_err->get(); |
| 134 | cfs->OS->write(Ptr: (const char *)data.data(), Size: data.size()); |
| 135 | if (llvm::Error err = cfs->commit()) { |
| 136 | Log *log = GetLog(mask: LLDBLog::Modules); |
| 137 | LLDB_LOG_ERROR(log, std::move(err), |
| 138 | "failed to commit to the cache for key: {0}" ); |
| 139 | } |
| 140 | return true; |
| 141 | } else { |
| 142 | Log *log = GetLog(mask: LLDBLog::Modules); |
| 143 | LLDB_LOG_ERROR(log, file_or_err.takeError(), |
| 144 | "failed to get the cache file stream for key: {0}" ); |
| 145 | } |
| 146 | } |
| 147 | } else { |
| 148 | Log *log = GetLog(mask: LLDBLog::Modules); |
| 149 | LLDB_LOG_ERROR(log, add_stream_or_err.takeError(), |
| 150 | "failed to get the cache add stream callback for key: {0}" ); |
| 151 | } |
| 152 | return false; |
| 153 | } |
| 154 | |
| 155 | FileSpec DataFileCache::GetCacheFilePath(llvm::StringRef key) { |
| 156 | FileSpec cache_file(m_cache_dir); |
| 157 | std::string filename("llvmcache-" ); |
| 158 | filename += key.str(); |
| 159 | cache_file.AppendPathComponent(component: filename); |
| 160 | return cache_file; |
| 161 | } |
| 162 | |
| 163 | Status DataFileCache::RemoveCacheFile(llvm::StringRef key) { |
| 164 | FileSpec cache_file = GetCacheFilePath(key); |
| 165 | FileSystem &fs = FileSystem::Instance(); |
| 166 | if (!fs.Exists(file_spec: cache_file)) |
| 167 | return Status(); |
| 168 | return fs.RemoveFile(file_spec: cache_file); |
| 169 | } |
| 170 | |
| 171 | CacheSignature::CacheSignature(lldb_private::Module *module) { |
| 172 | Clear(); |
| 173 | UUID uuid = module->GetUUID(); |
| 174 | if (uuid.IsValid()) |
| 175 | m_uuid = uuid; |
| 176 | |
| 177 | std::time_t mod_time = 0; |
| 178 | mod_time = llvm::sys::toTimeT(TP: module->GetModificationTime()); |
| 179 | if (mod_time != 0) |
| 180 | m_mod_time = mod_time; |
| 181 | |
| 182 | mod_time = llvm::sys::toTimeT(TP: module->GetObjectModificationTime()); |
| 183 | if (mod_time != 0) |
| 184 | m_obj_mod_time = mod_time; |
| 185 | } |
| 186 | |
| 187 | CacheSignature::CacheSignature(lldb_private::ObjectFile *objfile) { |
| 188 | Clear(); |
| 189 | UUID uuid = objfile->GetUUID(); |
| 190 | if (uuid.IsValid()) |
| 191 | m_uuid = uuid; |
| 192 | |
| 193 | std::time_t mod_time = 0; |
| 194 | // Grab the modification time of the object file's file. It isn't always the |
| 195 | // same as the module's file when you have a executable file as the main |
| 196 | // executable, and you have a object file for a symbol file. |
| 197 | FileSystem &fs = FileSystem::Instance(); |
| 198 | mod_time = llvm::sys::toTimeT(TP: fs.GetModificationTime(file_spec: objfile->GetFileSpec())); |
| 199 | if (mod_time != 0) |
| 200 | m_mod_time = mod_time; |
| 201 | |
| 202 | mod_time = |
| 203 | llvm::sys::toTimeT(TP: objfile->GetModule()->GetObjectModificationTime()); |
| 204 | if (mod_time != 0) |
| 205 | m_obj_mod_time = mod_time; |
| 206 | } |
| 207 | |
| 208 | enum SignatureEncoding { |
| 209 | eSignatureUUID = 1u, |
| 210 | eSignatureModTime = 2u, |
| 211 | eSignatureObjectModTime = 3u, |
| 212 | eSignatureEnd = 255u, |
| 213 | }; |
| 214 | |
| 215 | bool CacheSignature::Encode(DataEncoder &encoder) const { |
| 216 | if (!IsValid()) |
| 217 | return false; // Invalid signature, return false! |
| 218 | |
| 219 | if (m_uuid) { |
| 220 | llvm::ArrayRef<uint8_t> uuid_bytes = m_uuid->GetBytes(); |
| 221 | encoder.AppendU8(value: eSignatureUUID); |
| 222 | encoder.AppendU8(value: uuid_bytes.size()); |
| 223 | encoder.AppendData(data: uuid_bytes); |
| 224 | } |
| 225 | if (m_mod_time) { |
| 226 | encoder.AppendU8(value: eSignatureModTime); |
| 227 | encoder.AppendU32(value: *m_mod_time); |
| 228 | } |
| 229 | if (m_obj_mod_time) { |
| 230 | encoder.AppendU8(value: eSignatureObjectModTime); |
| 231 | encoder.AppendU32(value: *m_obj_mod_time); |
| 232 | } |
| 233 | encoder.AppendU8(value: eSignatureEnd); |
| 234 | return true; |
| 235 | } |
| 236 | |
| 237 | bool CacheSignature::(const lldb_private::DataExtractor &data, |
| 238 | lldb::offset_t *offset_ptr) { |
| 239 | Clear(); |
| 240 | while (uint8_t sig_encoding = data.GetU8(offset_ptr)) { |
| 241 | switch (sig_encoding) { |
| 242 | case eSignatureUUID: { |
| 243 | const uint8_t length = data.GetU8(offset_ptr); |
| 244 | const uint8_t *bytes = (const uint8_t *)data.GetData(offset_ptr, length); |
| 245 | if (bytes != nullptr && length > 0) |
| 246 | m_uuid = UUID(llvm::ArrayRef<uint8_t>(bytes, length)); |
| 247 | } break; |
| 248 | case eSignatureModTime: { |
| 249 | uint32_t mod_time = data.GetU32(offset_ptr); |
| 250 | if (mod_time > 0) |
| 251 | m_mod_time = mod_time; |
| 252 | } break; |
| 253 | case eSignatureObjectModTime: { |
| 254 | uint32_t mod_time = data.GetU32(offset_ptr); |
| 255 | if (mod_time > 0) |
| 256 | m_obj_mod_time = mod_time; |
| 257 | } break; |
| 258 | case eSignatureEnd: |
| 259 | // The definition of is valid changed to only be valid if the UUID is |
| 260 | // valid so make sure that if we attempt to decode an old cache file |
| 261 | // that we will fail to decode the cache file if the signature isn't |
| 262 | // considered valid. |
| 263 | return IsValid(); |
| 264 | default: |
| 265 | break; |
| 266 | } |
| 267 | } |
| 268 | return false; |
| 269 | } |
| 270 | |
| 271 | uint32_t ConstStringTable::Add(ConstString s) { |
| 272 | auto [pos, inserted] = m_string_to_offset.try_emplace(Key: s, Args&: m_next_offset); |
| 273 | if (inserted) { |
| 274 | m_strings.push_back(x: s); |
| 275 | m_next_offset += s.GetLength() + 1; |
| 276 | } |
| 277 | return pos->second; |
| 278 | } |
| 279 | |
| 280 | static const llvm::StringRef kStringTableIdentifier("STAB" ); |
| 281 | |
| 282 | bool ConstStringTable::Encode(DataEncoder &encoder) { |
| 283 | // Write an 4 character code into the stream. This will help us when decoding |
| 284 | // to make sure we find this identifier when decoding the string table to make |
| 285 | // sure we have the rigth data. It also helps to identify the string table |
| 286 | // when dumping the hex bytes in a cache file. |
| 287 | encoder.AppendData(data: kStringTableIdentifier); |
| 288 | size_t length_offset = encoder.GetByteSize(); |
| 289 | encoder.AppendU32(value: 0); // Total length of all strings which will be fixed up. |
| 290 | size_t strtab_offset = encoder.GetByteSize(); |
| 291 | encoder.AppendU8(value: 0); // Start the string table with an empty string. |
| 292 | for (auto s: m_strings) { |
| 293 | // Make sure all of the offsets match up with what we handed out! |
| 294 | assert(m_string_to_offset.find(s)->second == |
| 295 | encoder.GetByteSize() - strtab_offset); |
| 296 | // Append the C string into the encoder |
| 297 | encoder.AppendCString(data: s.GetStringRef()); |
| 298 | } |
| 299 | // Fixup the string table length. |
| 300 | encoder.PutU32(offset: length_offset, value: encoder.GetByteSize() - strtab_offset); |
| 301 | return true; |
| 302 | } |
| 303 | |
| 304 | bool StringTableReader::(const lldb_private::DataExtractor &data, |
| 305 | lldb::offset_t *offset_ptr) { |
| 306 | llvm::StringRef identifier((const char *)data.GetData(offset_ptr, length: 4), 4); |
| 307 | if (identifier != kStringTableIdentifier) |
| 308 | return false; |
| 309 | const uint32_t length = data.GetU32(offset_ptr); |
| 310 | // We always have at least one byte for the empty string at offset zero. |
| 311 | if (length == 0) |
| 312 | return false; |
| 313 | const char *bytes = (const char *)data.GetData(offset_ptr, length); |
| 314 | if (bytes == nullptr) |
| 315 | return false; |
| 316 | m_data = llvm::StringRef(bytes, length); |
| 317 | return true; |
| 318 | } |
| 319 | |
| 320 | llvm::StringRef StringTableReader::Get(uint32_t offset) const { |
| 321 | if (offset >= m_data.size()) |
| 322 | return llvm::StringRef(); |
| 323 | return llvm::StringRef(m_data.data() + offset); |
| 324 | } |
| 325 | |
| 326 | |