shill: LinkMonitor: Schedule periodic ARP requests
The LinkMonitor performs periodic unicast and broadcast ARP requests
for a connection's gateway IP address. It creates a weighted
average of response time, and can call a failure callback function
whenever the sum of consecutive unicast and broadcast response
failures exceed a threshold.
BUG=chromium-os:32600
TEST=Unit tests
Change-Id: Id5a66ad9ca80c51f227ef4811eb8c3885a70c1b8
Reviewed-on: https://gerrit.chromium.org/gerrit/28610
Commit-Ready: Paul Stewart <pstew@chromium.org>
Reviewed-by: Paul Stewart <pstew@chromium.org>
Tested-by: Paul Stewart <pstew@chromium.org>
diff --git a/link_monitor.cc b/link_monitor.cc
index e9ab5b0..2485ce5 100644
--- a/link_monitor.cc
+++ b/link_monitor.cc
@@ -4,31 +4,253 @@
#include "shill/link_monitor.h"
-#include <base/logging.h>
+#include <vector>
+#include <base/bind.h>
+#include <base/logging.h>
+#include <base/stringprintf.h>
+#include <base/string_util.h>
+
+#include "shill/arp_client.h"
+#include "shill/arp_packet.h"
+#include "shill/byte_string.h"
#include "shill/connection.h"
+#include "shill/device_info.h"
#include "shill/event_dispatcher.h"
-#include "shill/sockets.h"
+#include "shill/ip_address.h"
+#include "shill/scope_logger.h"
+#include "shill/shill_time.h"
+
+using base::Bind;
+using base::Unretained;
+using std::string;
namespace shill {
+const unsigned int LinkMonitor::kTestPeriodMilliseconds = 5000;
+const unsigned int LinkMonitor::kFailureThreshold = 5;
+const unsigned int LinkMonitor::kMaxResponseSampleFilterDepth = 5;
+
LinkMonitor::LinkMonitor(const ConnectionRefPtr &connection,
EventDispatcher *dispatcher,
+ DeviceInfo *device_info,
const FailureCallback &failure_callback)
: connection_(connection),
dispatcher_(dispatcher),
+ device_info_(device_info),
failure_callback_(failure_callback),
- sockets_(new Sockets()) {}
+ broadcast_failure_count_(0),
+ unicast_failure_count_(0),
+ is_unicast_(false),
+ response_sample_count_(0),
+ response_sample_bucket_(0),
+ time_(Time::GetInstance()) {}
-LinkMonitor::~LinkMonitor() {}
+LinkMonitor::~LinkMonitor() {
+ Stop();
+}
bool LinkMonitor::Start() {
- NOTIMPLEMENTED();
- return false;
+ Stop();
+
+ if (!device_info_->GetMACAddress(
+ connection_->interface_index(), &local_mac_address_)) {
+ LOG(ERROR) << "Could not get local MAC address.";
+ Stop();
+ return false;
+ }
+ gateway_mac_address_ = ByteString(local_mac_address_.GetLength());
+ send_request_callback_.Reset(
+ Bind(&LinkMonitor::SendRequestTask, Unretained(this)));
+ return SendRequest();
}
void LinkMonitor::Stop() {
- NOTIMPLEMENTED();
+ SLOG(Link, 2) << "In " << __func__ << ".";
+ local_mac_address_.Clear();
+ gateway_mac_address_.Clear();
+ arp_client_.reset();
+ broadcast_failure_count_ = 0;
+ unicast_failure_count_ = 0;
+ is_unicast_ = false;
+ response_sample_bucket_ = 0;
+ response_sample_count_ = 0;
+ receive_response_handler_.reset();
+ send_request_callback_.Cancel();
+ timerclear(&sent_request_at_);
+}
+
+unsigned int LinkMonitor::GetResponseTimeMilliseconds() {
+ return response_sample_count_ ?
+ response_sample_bucket_ / response_sample_count_ : 0;
+}
+
+void LinkMonitor::AddResponseTimeSample(
+ unsigned int response_time_milliseconds) {
+ SLOG(Link, 2) << "In " << __func__ << " with sample "
+ << response_time_milliseconds << ".";
+ response_sample_bucket_ += response_time_milliseconds;
+ if (response_sample_count_ < kMaxResponseSampleFilterDepth) {
+ ++response_sample_count_;
+ } else {
+ response_sample_bucket_ =
+ response_sample_bucket_ * kMaxResponseSampleFilterDepth /
+ (kMaxResponseSampleFilterDepth + 1);
+ }
+}
+
+// static
+string LinkMonitor::HardwareAddressToString(const ByteString &address) {
+ std::vector<string> address_parts;
+ for (size_t i = 0; i < address.GetLength(); ++i) {
+ address_parts.push_back(
+ base::StringPrintf("%02x", address.GetConstData()[i]));
+ }
+ return JoinString(address_parts, ':');
+}
+
+bool LinkMonitor::CreateClient() {
+ arp_client_.reset(new ArpClient(connection_->interface_index()));
+
+ if (!arp_client_->Start()) {
+ return false;
+ }
+ receive_response_handler_.reset(
+ dispatcher_->CreateReadyHandler(
+ arp_client_->socket(),
+ IOHandler::kModeInput,
+ Bind(&LinkMonitor::ReceiveResponse, Unretained(this))));
+ return true;
+}
+
+bool LinkMonitor::AddMissedResponse() {
+ SLOG(Link, 2) << "In " << __func__ << ".";
+ AddResponseTimeSample(kTestPeriodMilliseconds);
+
+ if (is_unicast_) {
+ ++unicast_failure_count_;
+ } else {
+ ++broadcast_failure_count_;
+ }
+
+ if (unicast_failure_count_ + broadcast_failure_count_ >= kFailureThreshold) {
+ LOG(ERROR) << "Link monitor has reached the failure threshold with "
+ << broadcast_failure_count_
+ << " broadcast failures and "
+ << unicast_failure_count_
+ << " unicast failures.";
+ failure_callback_.Run();
+ Stop();
+ return true;
+ }
+ is_unicast_ = !is_unicast_;
+ return false;
+}
+
+void LinkMonitor::ReceiveResponse(int fd) {
+ SLOG(Link, 2) << "In " << __func__ << ".";
+ ArpPacket packet;
+ ByteString sender;
+ if (!arp_client_->ReceiveReply(&packet, &sender)) {
+ return;
+ }
+
+ if (!connection_->local().Equals(packet.local_ip_address())) {
+ SLOG(Link, 4) << "Response is not for our IP address.";
+ return;
+ }
+
+ if (!local_mac_address_.Equals(packet.local_mac_address())) {
+ SLOG(Link, 4) << "Response is not for our MAC address.";
+ return;
+ }
+
+ if (!connection_->gateway().Equals(packet.remote_ip_address())) {
+ SLOG(Link, 4) << "Response is not from the gateway IP address.";
+ return;
+ }
+
+ struct timeval now, elapsed_time;
+ time_->GetTimeMonotonic(&now);
+ timersub(&now, &sent_request_at_, &elapsed_time);
+
+ AddResponseTimeSample(elapsed_time.tv_sec * 1000 +
+ elapsed_time.tv_usec / 1000);
+
+ receive_response_handler_.reset();
+ arp_client_.reset();
+
+ if (is_unicast_) {
+ unicast_failure_count_ = 0;
+ } else {
+ broadcast_failure_count_ = 0;
+ }
+
+ if (!gateway_mac_address_.Equals(packet.remote_mac_address())) {
+ const ByteString &new_mac_address = packet.remote_mac_address();
+ if (gateway_mac_address_.IsZero()) {
+ SLOG(Link, 2) << "Found gateway at "
+ << HardwareAddressToString(new_mac_address);
+ } else {
+ SLOG(Link, 2) << "Gateway MAC address changed.";
+ }
+ gateway_mac_address_ = new_mac_address;
+ }
+
+ is_unicast_ = !is_unicast_;
+}
+
+bool LinkMonitor::SendRequest() {
+ SLOG(Link, 2) << "In " << __func__ << ".";
+ if (!arp_client_.get()) {
+ if (!CreateClient()) {
+ LOG(ERROR) << "Failed to start ARP client.";
+ Stop();
+ return false;
+ }
+ } else if (AddMissedResponse()) {
+ // If an ARP client is still listening, this means we have timed
+ // out reception of the ARP reply.
+ return false;
+ } else {
+ // We already have an ArpClient instance running. These aren't
+ // bound sockets in the conventional sense, and we cannot distinguish
+ // which request (from which trial, or even from which component
+ // in the local system) an ARP reply was sent in response to.
+ // Therefore we keep the already open ArpClient in the case of
+ // a non-fatal timeout.
+ }
+
+ ByteString destination_mac_address(gateway_mac_address_.GetLength());
+ if (gateway_mac_address_.IsZero()) {
+ // The remote MAC addess is set by convention to be all-zeroes in the
+ // ARP header if not known. The ArpClient will translate an all-zeroes
+ // remote address into a send to the broadcast (all-ones) address in
+ // the Ethernet frame header.
+ SLOG_IF(Link, 2, is_unicast_) << "Sending broadcast since "
+ << "gateway MAC is unknown";
+ is_unicast_ = false;
+ } else if (is_unicast_) {
+ destination_mac_address = gateway_mac_address_;
+ }
+
+ ArpPacket request(connection_->local(), connection_->gateway(),
+ local_mac_address_, destination_mac_address);
+ if (!arp_client_->TransmitRequest(request)) {
+ LOG(ERROR) << "Failed to send ARP request. Stopping.";
+ Stop();
+ return false;
+ }
+
+ time_->GetTimeMonotonic(&sent_request_at_);
+
+ dispatcher_->PostDelayedTask(send_request_callback_.callback(),
+ kTestPeriodMilliseconds);
+ return true;
+}
+
+void LinkMonitor::SendRequestTask() {
+ SendRequest();
}
} // namespace shill