naiveproxy/net/socket/transport_client_socket_pool.h

298 lines
11 KiB
C
Raw Permalink Normal View History

2018-12-10 05:59:24 +03:00
// Copyright (c) 2012 The Chromium Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
#ifndef NET_SOCKET_TRANSPORT_CLIENT_SOCKET_POOL_H_
#define NET_SOCKET_TRANSPORT_CLIENT_SOCKET_POOL_H_
#include <memory>
#include <string>
#include "base/macros.h"
#include "base/memory/ref_counted.h"
#include "base/time/time.h"
#include "base/timer/timer.h"
#include "net/base/host_port_pair.h"
#include "net/base/net_export.h"
#include "net/dns/host_resolver.h"
#include "net/socket/client_socket_pool.h"
#include "net/socket/client_socket_pool_base.h"
#include "net/socket/connection_attempts.h"
#include "net/socket/socket_tag.h"
namespace net {
class ClientSocketFactory;
class SocketPerformanceWatcherFactory;
class NetLog;
class NetLogWithSource;
typedef base::Callback<int(const AddressList&, const NetLogWithSource& net_log)>
OnHostResolutionCallback;
class NET_EXPORT_PRIVATE TransportSocketParams
: public base::RefCounted<TransportSocketParams> {
public:
// CombineConnectAndWrite currently translates to using TCP FastOpen.
// TCP FastOpen should not be used if the first write to the socket may
// be non-idempotent, as the underlying socket could retransmit the data
// on failure of the first transmission.
enum CombineConnectAndWritePolicy {
COMBINE_CONNECT_AND_WRITE_DEFAULT, // Default policy, don't combine.
COMBINE_CONNECT_AND_WRITE_DESIRED, // Combine if supported by socket.
};
// |host_resolution_callback| will be invoked after the the hostname is
// resolved. If |host_resolution_callback| does not return OK, then the
// connection will be aborted with that value. |combine_connect_and_write|
// defines the policy for use of TCP FastOpen on this socket.
TransportSocketParams(
const HostPortPair& host_port_pair,
bool disable_resolver_cache,
const OnHostResolutionCallback& host_resolution_callback,
CombineConnectAndWritePolicy combine_connect_and_write);
const HostResolver::RequestInfo& destination() const { return destination_; }
const OnHostResolutionCallback& host_resolution_callback() const {
return host_resolution_callback_;
}
CombineConnectAndWritePolicy combine_connect_and_write() const {
return combine_connect_and_write_;
}
private:
friend class base::RefCounted<TransportSocketParams>;
~TransportSocketParams();
HostResolver::RequestInfo destination_;
const OnHostResolutionCallback host_resolution_callback_;
CombineConnectAndWritePolicy combine_connect_and_write_;
DISALLOW_COPY_AND_ASSIGN(TransportSocketParams);
};
// TransportConnectJob handles the host resolution necessary for socket creation
// and the transport (likely TCP) connect. TransportConnectJob also has fallback
// logic for IPv6 connect() timeouts (which may happen due to networks / routers
// with broken IPv6 support). Those timeouts take 20s, so rather than make the
// user wait 20s for the timeout to fire, we use a fallback timer
// (kIPv6FallbackTimerInMs) and start a connect() to a IPv4 address if the timer
// fires. Then we race the IPv4 connect() against the IPv6 connect() (which has
// a headstart) and return the one that completes first to the socket pool.
class NET_EXPORT_PRIVATE TransportConnectJob : public ConnectJob {
public:
// For recording the connection time in the appropriate bucket.
enum RaceResult {
RACE_UNKNOWN,
RACE_IPV4_WINS,
RACE_IPV4_SOLO,
RACE_IPV6_WINS,
RACE_IPV6_SOLO,
};
// TransportConnectJobs will time out after this many seconds. Note this is
// the total time, including both host resolution and TCP connect() times.
static const int kTimeoutInSeconds;
// In cases where both IPv6 and IPv4 addresses were returned from DNS,
// TransportConnectJobs will start a second connection attempt to just the
// IPv4 addresses after this many milliseconds. (This is "Happy Eyeballs".)
static const int kIPv6FallbackTimerInMs;
TransportConnectJob(
const std::string& group_name,
RequestPriority priority,
const SocketTag& socket_tag,
ClientSocketPool::RespectLimits respect_limits,
const scoped_refptr<TransportSocketParams>& params,
base::TimeDelta timeout_duration,
ClientSocketFactory* client_socket_factory,
SocketPerformanceWatcherFactory* socket_performance_watcher_factory,
HostResolver* host_resolver,
Delegate* delegate,
NetLog* net_log);
~TransportConnectJob() override;
// ConnectJob methods.
LoadState GetLoadState() const override;
void GetAdditionalErrorState(ClientSocketHandle* handle) override;
// Rolls |addrlist| forward until the first IPv4 address, if any.
// WARNING: this method should only be used to implement the prefer-IPv4 hack.
static void MakeAddressListStartWithIPv4(AddressList* addrlist);
// Record the histograms Net.DNS_Resolution_And_TCP_Connection_Latency2 and
// Net.TCP_Connection_Latency and return the connect duration.
static void HistogramDuration(
const LoadTimingInfo::ConnectTiming& connect_timing,
RaceResult race_result);
private:
enum State {
STATE_RESOLVE_HOST,
STATE_RESOLVE_HOST_COMPLETE,
STATE_TRANSPORT_CONNECT,
STATE_TRANSPORT_CONNECT_COMPLETE,
STATE_NONE,
};
void OnIOComplete(int result);
int DoLoop(int result);
int DoResolveHost();
int DoResolveHostComplete(int result);
int DoTransportConnect();
int DoTransportConnectComplete(int result);
// Not part of the state machine.
void DoIPv6FallbackTransportConnect();
void DoIPv6FallbackTransportConnectComplete(int result);
// Begins the host resolution and the TCP connect. Returns OK on success
// and ERR_IO_PENDING if it cannot immediately service the request.
// Otherwise, it returns a net error code.
int ConnectInternal() override;
void CopyConnectionAttemptsFromSockets();
scoped_refptr<TransportSocketParams> params_;
HostResolver* resolver_;
std::unique_ptr<HostResolver::Request> request_;
ClientSocketFactory* const client_socket_factory_;
State next_state_;
std::unique_ptr<StreamSocket> transport_socket_;
AddressList addresses_;
std::unique_ptr<StreamSocket> fallback_transport_socket_;
std::unique_ptr<AddressList> fallback_addresses_;
base::TimeTicks fallback_connect_start_time_;
base::OneShotTimer fallback_timer_;
SocketPerformanceWatcherFactory* socket_performance_watcher_factory_;
int resolve_result_;
// Used in the failure case to save connection attempts made on the main and
// fallback sockets and pass them on in |GetAdditionalErrorState|. (In the
// success case, connection attempts are passed through the returned socket;
// attempts are copied from the other socket, if one exists, into it before
// it is returned.)
ConnectionAttempts connection_attempts_;
ConnectionAttempts fallback_connection_attempts_;
DISALLOW_COPY_AND_ASSIGN(TransportConnectJob);
};
class NET_EXPORT_PRIVATE TransportClientSocketPool : public ClientSocketPool {
public:
typedef TransportSocketParams SocketParams;
TransportClientSocketPool(
int max_sockets,
int max_sockets_per_group,
HostResolver* host_resolver,
ClientSocketFactory* client_socket_factory,
SocketPerformanceWatcherFactory* socket_performance_watcher_factory,
NetLog* net_log);
~TransportClientSocketPool() override;
// ClientSocketPool implementation.
int RequestSocket(const std::string& group_name,
const void* resolve_info,
RequestPriority priority,
const SocketTag& socket_tag,
RespectLimits respect_limits,
ClientSocketHandle* handle,
CompletionOnceCallback callback,
const NetLogWithSource& net_log) override;
void RequestSockets(const std::string& group_name,
const void* params,
int num_sockets,
const NetLogWithSource& net_log) override;
void SetPriority(const std::string& group_name,
ClientSocketHandle* handle,
RequestPriority priority) override;
void CancelRequest(const std::string& group_name,
ClientSocketHandle* handle) override;
void ReleaseSocket(const std::string& group_name,
std::unique_ptr<StreamSocket> socket,
int id) override;
void FlushWithError(int error) override;
void CloseIdleSockets() override;
void CloseIdleSocketsInGroup(const std::string& group_name) override;
int IdleSocketCount() const override;
int IdleSocketCountInGroup(const std::string& group_name) const override;
LoadState GetLoadState(const std::string& group_name,
const ClientSocketHandle* handle) const override;
std::unique_ptr<base::DictionaryValue> GetInfoAsValue(
const std::string& name,
const std::string& type,
bool include_nested_pools) const override;
base::TimeDelta ConnectionTimeout() const override;
// HigherLayeredPool implementation.
bool IsStalled() const override;
void AddHigherLayeredPool(HigherLayeredPool* higher_pool) override;
void RemoveHigherLayeredPool(HigherLayeredPool* higher_pool) override;
ClientSocketFactory* client_socket_factory() {
return client_socket_factory_;
}
protected:
// Methods shared with WebSocketTransportClientSocketPool
void NetLogTcpClientSocketPoolRequestedSocket(
const NetLogWithSource& net_log,
const scoped_refptr<TransportSocketParams>* casted_params);
private:
typedef ClientSocketPoolBase<TransportSocketParams> PoolBase;
class TransportConnectJobFactory
: public PoolBase::ConnectJobFactory {
public:
TransportConnectJobFactory(
ClientSocketFactory* client_socket_factory,
HostResolver* host_resolver,
SocketPerformanceWatcherFactory* socket_performance_watcher_factory,
NetLog* net_log)
: client_socket_factory_(client_socket_factory),
socket_performance_watcher_factory_(
socket_performance_watcher_factory),
host_resolver_(host_resolver),
net_log_(net_log) {}
~TransportConnectJobFactory() override {}
// ClientSocketPoolBase::ConnectJobFactory methods.
std::unique_ptr<ConnectJob> NewConnectJob(
const std::string& group_name,
const PoolBase::Request& request,
ConnectJob::Delegate* delegate) const override;
base::TimeDelta ConnectionTimeout() const override;
private:
ClientSocketFactory* const client_socket_factory_;
SocketPerformanceWatcherFactory* socket_performance_watcher_factory_;
HostResolver* const host_resolver_;
NetLog* net_log_;
DISALLOW_COPY_AND_ASSIGN(TransportConnectJobFactory);
};
PoolBase base_;
ClientSocketFactory* const client_socket_factory_;
DISALLOW_COPY_AND_ASSIGN(TransportClientSocketPool);
};
} // namespace net
#endif // NET_SOCKET_TRANSPORT_CLIENT_SOCKET_POOL_H_