| 1 | /*++ |
| 2 | |
| 3 | Copyright (c) Microsoft. All rights reserved. |
| 4 | |
| 5 | Module Name: |
| 6 | |
| 7 | telemetry.cpp |
| 8 | |
| 9 | Abstract: |
| 10 | |
| 11 | This file contains the telemetry agent implementation. |
| 12 | |
| 13 | --*/ |
| 14 | |
| 15 | #include "common.h" |
| 16 | #include <dirent.h> |
| 17 | #include <errno.h> |
| 18 | #include <libgen.h> |
| 19 | #include <stdio.h> |
| 20 | #include <string.h> |
| 21 | #include <unistd.h> |
| 22 | #include <linux/connector.h> |
| 23 | #include <linux/netlink.h> |
| 24 | #include <linux/cn_proc.h> |
| 25 | #include <string> |
| 26 | #include <lxwil.h> |
| 27 | #include "util.h" |
| 28 | #include "mountutil.h" |
| 29 | #include "SocketChannel.h" |
| 30 | #include "message.h" |
| 31 | |
| 32 | namespace { |
| 33 | |
| 34 | #pragma GCC diagnostic push |
| 35 | #pragma GCC diagnostic ignored "-Wgnu-variable-sized-type-not-at-end" |
| 36 | |
| 37 | union messageBuffer |
| 38 | { |
| 39 | struct |
| 40 | { |
| 41 | nlmsghdr netlinkHeader; |
| 42 | cn_msg connectorMessage; |
| 43 | proc_cn_mcast_op operation; |
| 44 | } Send; |
| 45 | |
| 46 | struct |
| 47 | { |
| 48 | nlmsghdr netlinkHeader; |
| 49 | cn_msg connectorMessage; |
| 50 | proc_event event; |
| 51 | } Receive; |
| 52 | }; |
| 53 | |
| 54 | #pragma GCC diagnostic pop |
| 55 | |
| 56 | // Map containing the binaries and commands that are filesystem-intensive that we want to warn users against using in DrvFs. |
| 57 | const std::map<std::string, std::string> g_drvFsUsageMap = { |
| 58 | {"git", "clone"}, {"node", "/usr/bin/npm install"}, {"cargo", "build"}}; |
| 59 | |
| 60 | bool g_drvFsUsageEnabled = true; |
| 61 | |
| 62 | } // namespace |
| 63 | |
| 64 | std::pair<std::string, bool> GetProcessInformation(int pid) |
| 65 | { |
| 66 | // N.B. Procfs files may no longer be present for short-lived processes that exit before the |
| 67 | // process creation notification can be processed. |
| 68 | const std::string procPidPath = std::format("/proc/{}", pid); |
| 69 | wil::unique_fd fd{open(std::format("{}/cmdline", procPidPath).c_str(), O_RDONLY)}; |
| 70 | if (!fd) |
| 71 | { |
| 72 | return {}; |
| 73 | } |
| 74 | |
| 75 | // /proc/pid/cmdline contains all the arguments separated by NULL characters. |
| 76 | LX_MINI_INIT_TELEMETRY_MESSAGE message{}; |
| 77 | message.Header.MessageSize = sizeof(message); |
| 78 | message.Header.MessageType = LX_MINI_INIT_TELEMETRY_MESSAGE::Type; |
| 79 | |
| 80 | std::string commandLine(256, '\0'); |
| 81 | auto bytesRead = TEMP_FAILURE_RETRY(read(fd.get(), commandLine.data(), commandLine.size())); |
| 82 | if (bytesRead <= 0) |
| 83 | { |
| 84 | return {}; |
| 85 | } |
| 86 | |
| 87 | commandLine.resize(bytesRead - 1); |
| 88 | const auto* executable = basename(commandLine.data()); |
| 89 | |
| 90 | bool showDrvfsNotification = false; |
| 91 | |
| 92 | // Determine if the DrvFs perf notification should be displayed. |
| 93 | if (g_drvFsUsageEnabled) |
| 94 | { |
| 95 | // Check the if the binary name and first argument are in the list of scenarios. |
| 96 | auto found = g_drvFsUsageMap.find(executable); |
| 97 | if (found != g_drvFsUsageMap.end()) |
| 98 | { |
| 99 | auto length = strlen(executable); |
| 100 | if (bytesRead > length + 1) |
| 101 | { |
| 102 | auto argument = std::string_view(commandLine.data(), bytesRead).substr(length + 1); |
| 103 | if (strcmp(argument.data(), found->second.c_str()) == 0) |
| 104 | { |
| 105 | // Determine if the current working directory is a DrvFs mount. |
| 106 | std::error_code errorCode; |
| 107 | auto cwd = std::filesystem::read_symlink(std::format("{}/cwd", procPidPath), errorCode); |
| 108 | if (!errorCode) |
| 109 | { |
| 110 | auto mountInfo = std::format("{}{}", procPidPath, MOUNT_INFO_FILE_NAME); |
| 111 | size_t prefixLength; |
| 112 | auto drvFsPrefix = UtilFindMount(mountInfo.c_str(), cwd.c_str(), false, &prefixLength); |
| 113 | if (!drvFsPrefix.empty()) |
| 114 | { |
| 115 | showDrvfsNotification = true; |
| 116 | g_drvFsUsageEnabled = false; |
| 117 | } |
| 118 | } |
| 119 | } |
| 120 | } |
| 121 | } |
| 122 | } |
| 123 | |
| 124 | return std::make_pair(executable, showDrvfsNotification); |
| 125 | } |
| 126 | |
| 127 | unsigned int StartTelemetryAgent() |
| 128 | { |
| 129 | try |
| 130 | { |
| 131 | constexpr auto flushPeriod = std::chrono::minutes(30); |
| 132 | |
| 133 | // The telemetry agent is only supported on VM mode. |
| 134 | if (!UtilIsUtilityVm()) |
| 135 | { |
| 136 | return 1; |
| 137 | } |
| 138 | |
| 139 | // Initialize logging. |
| 140 | InitializeLogging(false); |
| 141 | |
| 142 | // Open and bind a netlink socket. |
| 143 | wil::unique_fd fd = socket(PF_NETLINK, SOCK_DGRAM, NETLINK_CONNECTOR); |
| 144 | THROW_LAST_ERROR_IF(!fd); |
| 145 | |
| 146 | sockaddr_nl address{}; |
| 147 | address.nl_family = AF_NETLINK; |
| 148 | address.nl_groups = CN_IDX_PROC; |
| 149 | address.nl_pid = getpid(); |
| 150 | THROW_LAST_ERROR_IF(bind(fd.get(), reinterpret_cast<sockaddr*>(&address), sizeof(address)) < 0); |
| 151 | |
| 152 | // Fill in the netlink header and connector message and send a message. |
| 153 | messageBuffer buffer{}; |
| 154 | buffer.Send.netlinkHeader.nlmsg_len = sizeof(buffer.Send); |
| 155 | buffer.Send.netlinkHeader.nlmsg_type = NLMSG_DONE; |
| 156 | buffer.Send.netlinkHeader.nlmsg_pid = getpid(); |
| 157 | buffer.Send.connectorMessage.id.idx = CN_IDX_PROC; |
| 158 | buffer.Send.connectorMessage.id.val = CN_VAL_PROC; |
| 159 | buffer.Send.connectorMessage.len = sizeof(buffer.Send.operation); |
| 160 | buffer.Send.operation = proc_cn_mcast_op::PROC_CN_MCAST_LISTEN; |
| 161 | auto bytes = send(fd.get(), &buffer, sizeof(buffer.Send), 0); |
| 162 | THROW_LAST_ERROR_IF(bytes != sizeof(buffer.Send)); |
| 163 | |
| 164 | // Set the receive timeout to 1 minute so the thread has an opportunity to flush even when no events are received. |
| 165 | timeval tv{}; |
| 166 | tv.tv_sec = 10; |
| 167 | THROW_LAST_ERROR_IF(setsockopt(fd.get(), SOL_SOCKET, SO_RCVTIMEO, &tv, sizeof(tv)) < 0); |
| 168 | |
| 169 | wsl::shared::SocketChannel channel({STDOUT_FILENO}, "Telemetry"); |
| 170 | |
| 171 | std::map<std::string, size_t> events; |
| 172 | std::optional<std::string> drvfsNotifyCommand; |
| 173 | |
| 174 | // Schedule the next flush in 30 seconds so that some events are captured even if WSL shuts down quickly. |
| 175 | auto nextFlush = std::chrono::steady_clock::now() + std::chrono::seconds(30); |
| 176 | |
| 177 | // Begin reading netlink messages. |
| 178 | for (;;) |
| 179 | { |
| 180 | memset(&buffer, 0, sizeof(buffer)); |
| 181 | sockaddr_nl fromAddress{}; |
| 182 | fromAddress.nl_family = AF_NETLINK; |
| 183 | fromAddress.nl_groups = CN_IDX_PROC; |
| 184 | fromAddress.nl_pid = 1; |
| 185 | socklen_t addressLength = sizeof(fromAddress); |
| 186 | bytes = TEMP_FAILURE_RETRY(recvfrom(fd.get(), &buffer, sizeof(buffer), 0, reinterpret_cast<sockaddr*>(&fromAddress), &addressLength)); |
| 187 | |
| 188 | if (bytes <= 0) |
| 189 | { |
| 190 | THROW_LAST_ERROR_IF(errno != ETIMEDOUT && errno != EAGAIN); |
| 191 | } |
| 192 | else |
| 193 | { |
| 194 | for (nlmsghdr* netlinkHeader = &buffer.Receive.netlinkHeader; NLMSG_OK(netlinkHeader, bytes); |
| 195 | netlinkHeader = NLMSG_NEXT(netlinkHeader, bytes)) |
| 196 | { |
| 197 | if ((netlinkHeader->nlmsg_type == NLMSG_ERROR) || (netlinkHeader->nlmsg_type == NLMSG_OVERRUN)) |
| 198 | { |
| 199 | break; |
| 200 | } |
| 201 | |
| 202 | if (netlinkHeader->nlmsg_type == NLMSG_NOOP) |
| 203 | { |
| 204 | continue; |
| 205 | } |
| 206 | |
| 207 | // For exec events, log app usage telemetry. |
| 208 | auto event = reinterpret_cast<proc_event*>(((cn_msg*)NLMSG_DATA(netlinkHeader))->data); |
| 209 | if (event->what == PROC_EVENT_EXEC) |
| 210 | { |
| 211 | auto [executable, showNotification] = GetProcessInformation(event->event_data.exec.process_pid); |
| 212 | if (showNotification) |
| 213 | { |
| 214 | drvfsNotifyCommand = executable; |
| 215 | } |
| 216 | |
| 217 | // Make sure the name doesn't contain a '/' so it doesn't break our message format. |
| 218 | if (!executable.empty() && executable.find("/") == std::string::npos) |
| 219 | { |
| 220 | auto it = events.find(executable); |
| 221 | if (it == events.end()) |
| 222 | { |
| 223 | events.emplace(std::move(executable), 1); |
| 224 | } |
| 225 | else |
| 226 | { |
| 227 | it->second++; |
| 228 | } |
| 229 | } |
| 230 | } |
| 231 | |
| 232 | if (netlinkHeader->nlmsg_type == NLMSG_DONE) |
| 233 | { |
| 234 | break; |
| 235 | } |
| 236 | } |
| 237 | } |
| 238 | |
| 239 | // Regularly flush messages back to the service. |
| 240 | |
| 241 | auto now = std::chrono::steady_clock::now(); |
| 242 | if (drvfsNotifyCommand.has_value() || now > nextFlush) |
| 243 | { |
| 244 | if (!events.empty()) |
| 245 | { |
| 246 | std::stringstream content; |
| 247 | |
| 248 | if (drvfsNotifyCommand.has_value()) |
| 249 | { |
| 250 | content << drvfsNotifyCommand.value() << "/1/"; |
| 251 | } |
| 252 | |
| 253 | for (auto [executable, count] : events) |
| 254 | { |
| 255 | // Having an extra ',' at the end makes parsing simpler. |
| 256 | content << executable << '/' << count << '/'; |
| 257 | } |
| 258 | |
| 259 | wsl::shared::MessageWriter<LX_MINI_INIT_TELEMETRY_MESSAGE> message; |
| 260 | message->ShowDrvFsNotification = drvfsNotifyCommand.has_value(); |
| 261 | message.WriteString(content.str()); |
| 262 | |
| 263 | channel.SendMessage<LX_MINI_INIT_TELEMETRY_MESSAGE>(message.Span()); |
| 264 | |
| 265 | events.clear(); |
| 266 | drvfsNotifyCommand = {}; |
| 267 | } |
| 268 | |
| 269 | nextFlush = now + flushPeriod; |
| 270 | } |
| 271 | } |
| 272 | |
| 273 | return 0; |
| 274 | } |
| 275 | catch (...) |
| 276 | { |
| 277 | LOG_CAUGHT_EXCEPTION(); |
| 278 | return 1; |
| 279 | } |
| 280 | |
| 281 | return 0; |
| 282 | } |