Peter Qiu | d788102 | 2015-01-21 12:27:41 -0800 | [diff] [blame^] | 1 | // Copyright 2015 The Chromium OS Authors. All rights reserved. |
| 2 | // Use of this source code is governed by a BSD-style license that can be |
| 3 | // found in the LICENSE file. |
| 4 | |
| 5 | #include "shill/active_link_monitor.h" |
| 6 | |
| 7 | #include <string> |
| 8 | #include <vector> |
| 9 | |
| 10 | #include <base/bind.h> |
| 11 | #include <base/strings/stringprintf.h> |
| 12 | #include <base/strings/string_util.h> |
| 13 | |
| 14 | #include "shill/arp_client.h" |
| 15 | #include "shill/arp_packet.h" |
| 16 | #include "shill/connection.h" |
| 17 | #include "shill/device_info.h" |
| 18 | #include "shill/event_dispatcher.h" |
| 19 | #include "shill/logging.h" |
| 20 | #include "shill/metrics.h" |
| 21 | #include "shill/net/ip_address.h" |
| 22 | #include "shill/net/shill_time.h" |
| 23 | |
| 24 | using base::Bind; |
| 25 | using base::Unretained; |
| 26 | using std::string; |
| 27 | |
| 28 | namespace shill { |
| 29 | |
| 30 | namespace Logging { |
| 31 | static auto kModuleLogScope = ScopeLogger::kLink; |
| 32 | static string ObjectID(Connection *c) { return c->interface_name(); } |
| 33 | } |
| 34 | |
| 35 | const int ActiveLinkMonitor::kDefaultTestPeriodMilliseconds = 5000; |
| 36 | const int ActiveLinkMonitor::kFailureThreshold = 5; |
| 37 | const int ActiveLinkMonitor::kFastTestPeriodMilliseconds = 200; |
| 38 | const int ActiveLinkMonitor::kMaxResponseSampleFilterDepth = 5; |
| 39 | const int ActiveLinkMonitor::kUnicastReplyReliabilityThreshold = 10; |
| 40 | |
| 41 | ActiveLinkMonitor::ActiveLinkMonitor(const ConnectionRefPtr &connection, |
| 42 | EventDispatcher *dispatcher, |
| 43 | Metrics *metrics, |
| 44 | DeviceInfo *device_info, |
| 45 | const FailureCallback &failure_callback, |
| 46 | const SuccessCallback &success_callback) |
| 47 | : connection_(connection), |
| 48 | dispatcher_(dispatcher), |
| 49 | metrics_(metrics), |
| 50 | device_info_(device_info), |
| 51 | failure_callback_(failure_callback), |
| 52 | success_callback_(success_callback), |
| 53 | arp_client_(new ArpClient(connection->interface_index())), |
| 54 | test_period_milliseconds_(kDefaultTestPeriodMilliseconds), |
| 55 | broadcast_failure_count_(0), |
| 56 | unicast_failure_count_(0), |
| 57 | broadcast_success_count_(0), |
| 58 | unicast_success_count_(0), |
| 59 | is_unicast_(false), |
| 60 | gateway_supports_unicast_arp_(false), |
| 61 | response_sample_count_(0), |
| 62 | response_sample_bucket_(0), |
| 63 | time_(Time::GetInstance()) { |
| 64 | } |
| 65 | |
| 66 | ActiveLinkMonitor::~ActiveLinkMonitor() { |
| 67 | Stop(); |
| 68 | } |
| 69 | |
| 70 | bool ActiveLinkMonitor::Start(int test_period) { |
| 71 | StopMonitorCycle(); |
| 72 | return StartInternal(test_period); |
| 73 | } |
| 74 | |
| 75 | void ActiveLinkMonitor::Stop() { |
| 76 | SLOG(connection_.get(), 2) << "In " << __func__ << "."; |
| 77 | // Stop current cycle. |
| 78 | StopMonitorCycle(); |
| 79 | |
| 80 | // Clear stats accumulated from previous monitor cycles. |
| 81 | local_mac_address_.Clear(); |
| 82 | gateway_mac_address_.Clear(); |
| 83 | broadcast_success_count_ = 0; |
| 84 | unicast_success_count_ = 0; |
| 85 | broadcast_failure_count_ = 0; |
| 86 | unicast_failure_count_ = 0; |
| 87 | is_unicast_ = false; |
| 88 | gateway_supports_unicast_arp_ = false; |
| 89 | response_sample_bucket_ = 0; |
| 90 | response_sample_count_ = 0; |
| 91 | } |
| 92 | |
| 93 | int ActiveLinkMonitor::GetResponseTimeMilliseconds() const { |
| 94 | return response_sample_count_ ? |
| 95 | response_sample_bucket_ / response_sample_count_ : 0; |
| 96 | } |
| 97 | |
| 98 | bool ActiveLinkMonitor::IsGatewayFound() const { |
| 99 | return !gateway_mac_address_.IsZero(); |
| 100 | } |
| 101 | |
| 102 | bool ActiveLinkMonitor::StartInternal(int probe_period_milliseconds) { |
| 103 | test_period_milliseconds_ = probe_period_milliseconds; |
| 104 | if (test_period_milliseconds_ > kDefaultTestPeriodMilliseconds) { |
| 105 | LOG(WARNING) << "Long test period; UMA stats will be truncated."; |
| 106 | } |
| 107 | |
| 108 | if (!device_info_->GetMACAddress( |
| 109 | connection_->interface_index(), &local_mac_address_)) { |
| 110 | LOG(ERROR) << "Could not get local MAC address."; |
| 111 | metrics_->NotifyLinkMonitorFailure( |
| 112 | connection_->technology(), |
| 113 | Metrics::kLinkMonitorMacAddressNotFound, |
| 114 | 0, 0, 0); |
| 115 | Stop(); |
| 116 | return false; |
| 117 | } |
| 118 | |
| 119 | if (!StartArpClient()) { |
| 120 | LOG(ERROR) << "Failed to start ARP client."; |
| 121 | metrics_->NotifyLinkMonitorFailure( |
| 122 | connection_->technology(), |
| 123 | Metrics::kLinkMonitorClientStartFailure, |
| 124 | 0, 0, 0); |
| 125 | Stop(); |
| 126 | return false; |
| 127 | } |
| 128 | |
| 129 | if (gateway_mac_address_.IsEmpty()) { |
| 130 | gateway_mac_address_ = ByteString(local_mac_address_.GetLength()); |
| 131 | } |
| 132 | send_request_callback_.Reset( |
| 133 | Bind(&ActiveLinkMonitor::SendRequest, Unretained(this))); |
| 134 | // Post a task to send ARP request instead of calling it synchronously, to |
| 135 | // maintain consistent expectation in the case of send failures, which will |
| 136 | // always invoke failure callback. |
| 137 | dispatcher_->PostTask(send_request_callback_.callback()); |
| 138 | return true; |
| 139 | } |
| 140 | |
| 141 | void ActiveLinkMonitor::StopMonitorCycle() { |
| 142 | StopArpClient(); |
| 143 | send_request_callback_.Cancel(); |
| 144 | timerclear(&sent_request_at_); |
| 145 | } |
| 146 | |
| 147 | void ActiveLinkMonitor::AddResponseTimeSample(int response_time_milliseconds) { |
| 148 | SLOG(connection_.get(), 2) << "In " << __func__ << " with sample " |
| 149 | << response_time_milliseconds << "."; |
| 150 | metrics_->NotifyLinkMonitorResponseTimeSampleAdded( |
| 151 | connection_->technology(), response_time_milliseconds); |
| 152 | response_sample_bucket_ += response_time_milliseconds; |
| 153 | if (response_sample_count_ < kMaxResponseSampleFilterDepth) { |
| 154 | ++response_sample_count_; |
| 155 | } else { |
| 156 | response_sample_bucket_ = |
| 157 | response_sample_bucket_ * kMaxResponseSampleFilterDepth / |
| 158 | (kMaxResponseSampleFilterDepth + 1); |
| 159 | } |
| 160 | } |
| 161 | |
| 162 | // static |
| 163 | string ActiveLinkMonitor::HardwareAddressToString(const ByteString &address) { |
| 164 | std::vector<string> address_parts; |
| 165 | for (size_t i = 0; i < address.GetLength(); ++i) { |
| 166 | address_parts.push_back( |
| 167 | base::StringPrintf("%02x", address.GetConstData()[i])); |
| 168 | } |
| 169 | return JoinString(address_parts, ':'); |
| 170 | } |
| 171 | |
| 172 | bool ActiveLinkMonitor::StartArpClient() { |
| 173 | if (!arp_client_->StartReplyListener()) { |
| 174 | return false; |
| 175 | } |
| 176 | SLOG(connection_.get(), 4) << "Created ARP client; listening on socket " |
| 177 | << arp_client_->socket() << "."; |
| 178 | receive_response_handler_.reset( |
| 179 | dispatcher_->CreateReadyHandler( |
| 180 | arp_client_->socket(), |
| 181 | IOHandler::kModeInput, |
| 182 | Bind(&ActiveLinkMonitor::ReceiveResponse, Unretained(this)))); |
| 183 | return true; |
| 184 | } |
| 185 | |
| 186 | void ActiveLinkMonitor::StopArpClient() { |
| 187 | arp_client_->Stop(); |
| 188 | receive_response_handler_.reset(); |
| 189 | } |
| 190 | |
| 191 | bool ActiveLinkMonitor::AddMissedResponse() { |
| 192 | SLOG(connection_.get(), 2) << "In " << __func__ << "."; |
| 193 | AddResponseTimeSample(test_period_milliseconds_); |
| 194 | |
| 195 | if (is_unicast_) { |
| 196 | if (gateway_supports_unicast_arp_) { |
| 197 | ++unicast_failure_count_; |
| 198 | } |
| 199 | unicast_success_count_ = 0; |
| 200 | } else { |
| 201 | ++broadcast_failure_count_; |
| 202 | broadcast_success_count_ = 0; |
| 203 | } |
| 204 | |
| 205 | if (unicast_failure_count_ + broadcast_failure_count_ >= kFailureThreshold) { |
| 206 | LOG(ERROR) << "Link monitor has reached the failure threshold with " |
| 207 | << broadcast_failure_count_ |
| 208 | << " broadcast failures and " |
| 209 | << unicast_failure_count_ |
| 210 | << " unicast failures."; |
| 211 | failure_callback_.Run(Metrics::kLinkMonitorFailureThresholdReached, |
| 212 | broadcast_failure_count_, |
| 213 | unicast_failure_count_); |
| 214 | Stop(); |
| 215 | return true; |
| 216 | } |
| 217 | is_unicast_ = !is_unicast_; |
| 218 | return false; |
| 219 | } |
| 220 | |
| 221 | void ActiveLinkMonitor::ReceiveResponse(int fd) { |
| 222 | SLOG(connection_.get(), 2) << "In " << __func__ << "."; |
| 223 | ArpPacket packet; |
| 224 | ByteString sender; |
| 225 | if (!arp_client_->ReceivePacket(&packet, &sender)) { |
| 226 | return; |
| 227 | } |
| 228 | |
| 229 | if (!packet.IsReply()) { |
| 230 | SLOG(connection_.get(), 4) << "This is not a reply packet. Ignoring."; |
| 231 | return; |
| 232 | } |
| 233 | |
| 234 | if (!connection_->local().address().Equals( |
| 235 | packet.remote_ip_address().address())) { |
| 236 | SLOG(connection_.get(), 4) << "Response is not for our IP address."; |
| 237 | return; |
| 238 | } |
| 239 | |
| 240 | if (!local_mac_address_.Equals(packet.remote_mac_address())) { |
| 241 | SLOG(connection_.get(), 4) << "Response is not for our MAC address."; |
| 242 | return; |
| 243 | } |
| 244 | |
| 245 | if (!connection_->gateway().address().Equals( |
| 246 | packet.local_ip_address().address())) { |
| 247 | SLOG(connection_.get(), 4) |
| 248 | << "Response is not from the gateway IP address."; |
| 249 | return; |
| 250 | } |
| 251 | |
| 252 | struct timeval now, elapsed_time; |
| 253 | time_->GetTimeMonotonic(&now); |
| 254 | timersub(&now, &sent_request_at_, &elapsed_time); |
| 255 | |
| 256 | AddResponseTimeSample(elapsed_time.tv_sec * 1000 + |
| 257 | elapsed_time.tv_usec / 1000); |
| 258 | |
| 259 | if (is_unicast_) { |
| 260 | ++unicast_success_count_; |
| 261 | unicast_failure_count_ = 0; |
| 262 | if (unicast_success_count_ >= kUnicastReplyReliabilityThreshold) { |
| 263 | SLOG_IF(Link, 2, !gateway_supports_unicast_arp_) |
| 264 | << "Gateway is now considered a reliable unicast responder. " |
| 265 | "Unicast failures will now count."; |
| 266 | gateway_supports_unicast_arp_ = true; |
| 267 | } |
| 268 | } else { |
| 269 | ++broadcast_success_count_; |
| 270 | broadcast_failure_count_ = 0; |
| 271 | } |
| 272 | |
| 273 | if (!gateway_mac_address_.Equals(packet.local_mac_address())) { |
| 274 | const ByteString &new_mac_address = packet.local_mac_address(); |
| 275 | if (!IsGatewayFound()) { |
| 276 | SLOG(connection_.get(), 2) << "Found gateway at " |
| 277 | << HardwareAddressToString(new_mac_address); |
| 278 | } else { |
| 279 | SLOG(connection_.get(), 2) << "Gateway MAC address changed."; |
| 280 | } |
| 281 | gateway_mac_address_ = new_mac_address; |
| 282 | } |
| 283 | |
| 284 | is_unicast_ = !is_unicast_; |
| 285 | |
| 286 | // Stop the current cycle, and invoke the success callback. All the |
| 287 | // accumulated stats regarding the gateway are not cleared. |
| 288 | StopMonitorCycle(); |
| 289 | success_callback_.Run(); |
| 290 | } |
| 291 | |
| 292 | void ActiveLinkMonitor::SendRequest() { |
| 293 | SLOG(connection_.get(), 2) << "In " << __func__ << "."; |
| 294 | |
| 295 | // Timeout waiting for ARP reply and exceed the failure threshold. |
| 296 | if (timerisset(&sent_request_at_) && AddMissedResponse()) { |
| 297 | return; |
| 298 | } |
| 299 | |
| 300 | ByteString destination_mac_address(gateway_mac_address_.GetLength()); |
| 301 | if (!IsGatewayFound()) { |
| 302 | // The remote MAC addess is set by convention to be all-zeroes in the |
| 303 | // ARP header if not known. The ArpClient will translate an all-zeroes |
| 304 | // remote address into a send to the broadcast (all-ones) address in |
| 305 | // the Ethernet frame header. |
| 306 | SLOG_IF(Link, 2, is_unicast_) << "Sending broadcast since " |
| 307 | << "gateway MAC is unknown"; |
| 308 | is_unicast_ = false; |
| 309 | } else if (is_unicast_) { |
| 310 | destination_mac_address = gateway_mac_address_; |
| 311 | } |
| 312 | LOG(INFO) << "IsGatway " << IsGatewayFound() << " unicast: " << is_unicast_; |
| 313 | ArpPacket request(connection_->local(), connection_->gateway(), |
| 314 | local_mac_address_, destination_mac_address); |
| 315 | if (!arp_client_->TransmitRequest(request)) { |
| 316 | LOG(ERROR) << "Failed to send ARP request. Stopping."; |
| 317 | failure_callback_.Run(Metrics::kLinkMonitorTransmitFailure, |
| 318 | broadcast_failure_count_, |
| 319 | unicast_failure_count_); |
| 320 | Stop(); |
| 321 | return; |
| 322 | } |
| 323 | |
| 324 | time_->GetTimeMonotonic(&sent_request_at_); |
| 325 | |
| 326 | dispatcher_->PostDelayedTask(send_request_callback_.callback(), |
| 327 | test_period_milliseconds_); |
| 328 | } |
| 329 | |
| 330 | } // namespace shill |