platform_system_core/adb/transport.h

394 lines
13 KiB
C
Raw Normal View History

/*
* Copyright (C) 2011 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#ifndef __TRANSPORT_H
#define __TRANSPORT_H
#include <sys/types.h>
adb: fix two device offline problems. When device goes offline, user usually has to manually replug the usb device. This patch tries to solve two offline situations, all because when adb on host is killed, the adbd on device is not notified. 1. When adb server is killed while pushing a large file to device, the device is still reading the unfinished large message. So the device thinks of the CNXN message as part of the previous unfinished message, so it doesn't reply and the device is in offline state. The solution is to add a write_msg_lock in atransport struct. And it kicks the transport only after sending a whole message. By kicking all transports before exit, we ensure that we don't write part of a message to any device. So next time we start adb server, the device should be waiting for a new message. 2. When adb server is killed while pulling a large file from device, the device is still trying to send the unfinished large message. So adb on host usually reads data with EOVERFLOW error. This is because adb on host is reading less than one packet sent from device. The solution is to use buffered read on host. The max packet size of bulk transactions in USB 3.0 is 1024 bytes. By preparing an at least 1024 bytes buffer when reading, EOVERFLOW no longer occurs. And teach adb host to ignore wrong messages. To be safe, this patch doesn't change any logic on device. Bug: http://b/32952319 Test: run python -m unittest -q test_device.DeviceOfflineTest Test: on linux/mac/windows with bullhead, ryu. Change-Id: Ib149d30028a62a6f03857b8a95ab5a1d6e9b9c4e
2017-03-11 01:01:01 +01:00
#include <atomic>
#include <chrono>
#include <condition_variable>
#include <deque>
#include <functional>
#include <list>
#include <memory>
adb: fix two device offline problems. When device goes offline, user usually has to manually replug the usb device. This patch tries to solve two offline situations, all because when adb on host is killed, the adbd on device is not notified. 1. When adb server is killed while pushing a large file to device, the device is still reading the unfinished large message. So the device thinks of the CNXN message as part of the previous unfinished message, so it doesn't reply and the device is in offline state. The solution is to add a write_msg_lock in atransport struct. And it kicks the transport only after sending a whole message. By kicking all transports before exit, we ensure that we don't write part of a message to any device. So next time we start adb server, the device should be waiting for a new message. 2. When adb server is killed while pulling a large file from device, the device is still trying to send the unfinished large message. So adb on host usually reads data with EOVERFLOW error. This is because adb on host is reading less than one packet sent from device. The solution is to use buffered read on host. The max packet size of bulk transactions in USB 3.0 is 1024 bytes. By preparing an at least 1024 bytes buffer when reading, EOVERFLOW no longer occurs. And teach adb host to ignore wrong messages. To be safe, this patch doesn't change any logic on device. Bug: http://b/32952319 Test: run python -m unittest -q test_device.DeviceOfflineTest Test: on linux/mac/windows with bullhead, ryu. Change-Id: Ib149d30028a62a6f03857b8a95ab5a1d6e9b9c4e
2017-03-11 01:01:01 +01:00
#include <mutex>
#include <string>
#include <thread>
#include <unordered_set>
#include <android-base/macros.h>
#include <android-base/thread_annotations.h>
#include <openssl/rsa.h>
#include "adb.h"
#include "adb_unique_fd.h"
typedef std::unordered_set<std::string> FeatureSet;
const FeatureSet& supported_features();
// Encodes and decodes FeatureSet objects into human-readable strings.
std::string FeatureSetToString(const FeatureSet& features);
FeatureSet StringToFeatureSet(const std::string& features_string);
// Returns true if both local features and |feature_set| support |feature|.
bool CanUseFeature(const FeatureSet& feature_set, const std::string& feature);
// Do not use any of [:;=,] in feature strings, they have special meaning
// in the connection banner.
extern const char* const kFeatureShell2;
// The 'cmd' command is available
extern const char* const kFeatureCmd;
extern const char* const kFeatureStat2;
// The server is running with libusb enabled.
extern const char* const kFeatureLibusb;
// adbd supports `push --sync`.
extern const char* const kFeaturePushSync;
// adbd supports installing .apex packages.
extern const char* const kFeatureApex;
// adbd has b/110953234 fixed.
extern const char* const kFeatureFixedPushMkdir;
TransportId NextTransportId();
// Abstraction for a non-blocking packet transport.
struct Connection {
Connection() = default;
virtual ~Connection() = default;
void SetTransportName(std::string transport_name) {
transport_name_ = std::move(transport_name);
}
using ReadCallback = std::function<bool(Connection*, std::unique_ptr<apacket>)>;
void SetReadCallback(ReadCallback callback) {
CHECK(!read_callback_);
read_callback_ = callback;
}
// Called after the Connection has terminated, either by an error or because Stop was called.
using ErrorCallback = std::function<void(Connection*, const std::string&)>;
void SetErrorCallback(ErrorCallback callback) {
CHECK(!error_callback_);
error_callback_ = callback;
}
virtual bool Write(std::unique_ptr<apacket> packet) = 0;
virtual void Start() = 0;
virtual void Stop() = 0;
std::string transport_name_;
ReadCallback read_callback_;
ErrorCallback error_callback_;
static std::unique_ptr<Connection> FromFd(unique_fd fd);
};
// Abstraction for a blocking packet transport.
struct BlockingConnection {
BlockingConnection() = default;
BlockingConnection(const BlockingConnection& copy) = delete;
BlockingConnection(BlockingConnection&& move) = delete;
// Destroy a BlockingConnection. Formerly known as 'Close' in atransport.
virtual ~BlockingConnection() = default;
// Read/Write a packet. These functions are concurrently called from a transport's reader/writer
// threads.
virtual bool Read(apacket* packet) = 0;
virtual bool Write(apacket* packet) = 0;
// Terminate a connection.
// This method must be thread-safe, and must cause concurrent Reads/Writes to terminate.
// Formerly known as 'Kick' in atransport.
virtual void Close() = 0;
};
struct BlockingConnectionAdapter : public Connection {
explicit BlockingConnectionAdapter(std::unique_ptr<BlockingConnection> connection);
virtual ~BlockingConnectionAdapter();
virtual bool Write(std::unique_ptr<apacket> packet) override final;
virtual void Start() override final;
virtual void Stop() override final;
bool started_ GUARDED_BY(mutex_) = false;
bool stopped_ GUARDED_BY(mutex_) = false;
std::unique_ptr<BlockingConnection> underlying_;
std::thread read_thread_ GUARDED_BY(mutex_);
std::thread write_thread_ GUARDED_BY(mutex_);
std::deque<std::unique_ptr<apacket>> write_queue_ GUARDED_BY(mutex_);
std::mutex mutex_;
std::condition_variable cv_;
std::once_flag error_flag_;
};
struct FdConnection : public BlockingConnection {
explicit FdConnection(unique_fd fd) : fd_(std::move(fd)) {}
bool Read(apacket* packet) override final;
bool Write(apacket* packet) override final;
void Close() override;
private:
unique_fd fd_;
};
struct UsbConnection : public BlockingConnection {
explicit UsbConnection(usb_handle* handle) : handle_(handle) {}
~UsbConnection();
bool Read(apacket* packet) override final;
bool Write(apacket* packet) override final;
void Close() override final;
usb_handle* handle_;
};
// Waits for a transport's connection to be not pending. This is a separate
// object so that the transport can be destroyed and another thread can be
// notified of it in a race-free way.
class ConnectionWaitable {
public:
ConnectionWaitable() = default;
~ConnectionWaitable() = default;
// Waits until the first CNXN packet has been received by the owning
// atransport, or the specified timeout has elapsed. Can be called from any
// thread.
//
// Returns true if the CNXN packet was received in a timely fashion, false
// otherwise.
bool WaitForConnection(std::chrono::milliseconds timeout);
// Can be called from any thread when the connection stops being pending.
// Only the first invocation will be acknowledged, the rest will be no-ops.
void SetConnectionEstablished(bool success);
private:
bool connection_established_ GUARDED_BY(mutex_) = false;
bool connection_established_ready_ GUARDED_BY(mutex_) = false;
std::mutex mutex_;
std::condition_variable cv_;
DISALLOW_COPY_AND_ASSIGN(ConnectionWaitable);
};
enum class ReconnectResult {
Retry,
Success,
Abort,
};
class atransport {
public:
// TODO(danalbert): We expose waaaaaaay too much stuff because this was
// historically just a struct, but making the whole thing a more idiomatic
// class in one go is a very large change. Given how bad our testing is,
// it's better to do this piece by piece.
using ReconnectCallback = std::function<ReconnectResult(atransport*)>;
atransport(ReconnectCallback reconnect, ConnectionState state)
: id(NextTransportId()),
kicked_(false),
connection_state_(state),
connection_waitable_(std::make_shared<ConnectionWaitable>()),
connection_(nullptr),
reconnect_(std::move(reconnect)) {
// Initialize protocol to min version for compatibility with older versions.
// Version will be updated post-connect.
protocol_version = A_VERSION_MIN;
max_payload = MAX_PAYLOAD;
}
atransport(ConnectionState state = kCsOffline)
: atransport([](atransport*) { return ReconnectResult::Abort; }, state) {}
virtual ~atransport();
adb: fix two device offline problems. When device goes offline, user usually has to manually replug the usb device. This patch tries to solve two offline situations, all because when adb on host is killed, the adbd on device is not notified. 1. When adb server is killed while pushing a large file to device, the device is still reading the unfinished large message. So the device thinks of the CNXN message as part of the previous unfinished message, so it doesn't reply and the device is in offline state. The solution is to add a write_msg_lock in atransport struct. And it kicks the transport only after sending a whole message. By kicking all transports before exit, we ensure that we don't write part of a message to any device. So next time we start adb server, the device should be waiting for a new message. 2. When adb server is killed while pulling a large file from device, the device is still trying to send the unfinished large message. So adb on host usually reads data with EOVERFLOW error. This is because adb on host is reading less than one packet sent from device. The solution is to use buffered read on host. The max packet size of bulk transactions in USB 3.0 is 1024 bytes. By preparing an at least 1024 bytes buffer when reading, EOVERFLOW no longer occurs. And teach adb host to ignore wrong messages. To be safe, this patch doesn't change any logic on device. Bug: http://b/32952319 Test: run python -m unittest -q test_device.DeviceOfflineTest Test: on linux/mac/windows with bullhead, ryu. Change-Id: Ib149d30028a62a6f03857b8a95ab5a1d6e9b9c4e
2017-03-11 01:01:01 +01:00
int Write(apacket* p);
void Kick();
bool kicked() const { return kicked_; }
adb: fix two device offline problems. When device goes offline, user usually has to manually replug the usb device. This patch tries to solve two offline situations, all because when adb on host is killed, the adbd on device is not notified. 1. When adb server is killed while pushing a large file to device, the device is still reading the unfinished large message. So the device thinks of the CNXN message as part of the previous unfinished message, so it doesn't reply and the device is in offline state. The solution is to add a write_msg_lock in atransport struct. And it kicks the transport only after sending a whole message. By kicking all transports before exit, we ensure that we don't write part of a message to any device. So next time we start adb server, the device should be waiting for a new message. 2. When adb server is killed while pulling a large file from device, the device is still trying to send the unfinished large message. So adb on host usually reads data with EOVERFLOW error. This is because adb on host is reading less than one packet sent from device. The solution is to use buffered read on host. The max packet size of bulk transactions in USB 3.0 is 1024 bytes. By preparing an at least 1024 bytes buffer when reading, EOVERFLOW no longer occurs. And teach adb host to ignore wrong messages. To be safe, this patch doesn't change any logic on device. Bug: http://b/32952319 Test: run python -m unittest -q test_device.DeviceOfflineTest Test: on linux/mac/windows with bullhead, ryu. Change-Id: Ib149d30028a62a6f03857b8a95ab5a1d6e9b9c4e
2017-03-11 01:01:01 +01:00
// ConnectionState can be read by all threads, but can only be written in the main thread.
ConnectionState GetConnectionState() const;
void SetConnectionState(ConnectionState state);
void SetConnection(std::unique_ptr<Connection> connection);
std::shared_ptr<Connection> connection() {
std::lock_guard<std::mutex> lock(mutex_);
return connection_;
}
const TransportId id;
size_t ref_count = 0;
bool online = false;
TransportType type = kTransportAny;
// Used to identify transports for clients.
std::string serial;
std::string product;
std::string model;
std::string device;
std::string devpath;
bool IsTcpDevice() const { return type == kTransportLocal; }
#if ADB_HOST
std::shared_ptr<RSA> NextKey();
void ResetKeys();
#endif
char token[TOKEN_SIZE] = {};
size_t failed_auth_attempts = 0;
std::string serial_name() const { return !serial.empty() ? serial : "<unknown>"; }
std::string connection_state_name() const;
void update_version(int version, size_t payload);
int get_protocol_version() const;
size_t get_max_payload() const;
const FeatureSet& features() const {
return features_;
}
bool has_feature(const std::string& feature) const;
// Loads the transport's feature set from the given string.
void SetFeatures(const std::string& features_string);
void AddDisconnect(adisconnect* disconnect);
void RemoveDisconnect(adisconnect* disconnect);
void RunDisconnects();
// Returns true if |target| matches this transport. A matching |target| can be any of:
// * <serial>
// * <devpath>
// * product:<product>
// * model:<model>
// * device:<device>
//
// If this is a local transport, serial will also match [tcp:|udp:]<hostname>[:port] targets.
// For example, serial "100.100.100.100:5555" would match any of:
// * 100.100.100.100
// * tcp:100.100.100.100
// * udp:100.100.100.100:5555
// This is to make it easier to use the same network target for both fastboot and adb.
bool MatchesTarget(const std::string& target) const;
// Notifies that the atransport is no longer waiting for the connection
// being established.
void SetConnectionEstablished(bool success);
// Gets a shared reference to the ConnectionWaitable.
std::shared_ptr<ConnectionWaitable> connection_waitable() { return connection_waitable_; }
// Attempts to reconnect with the underlying Connection.
ReconnectResult Reconnect();
private:
std::atomic<bool> kicked_;
// A set of features transmitted in the banner with the initial connection.
// This is stored in the banner as 'features=feature0,feature1,etc'.
FeatureSet features_;
int protocol_version;
size_t max_payload;
// A list of adisconnect callbacks called when the transport is kicked.
std::list<adisconnect*> disconnects_;
adb: fix two device offline problems. When device goes offline, user usually has to manually replug the usb device. This patch tries to solve two offline situations, all because when adb on host is killed, the adbd on device is not notified. 1. When adb server is killed while pushing a large file to device, the device is still reading the unfinished large message. So the device thinks of the CNXN message as part of the previous unfinished message, so it doesn't reply and the device is in offline state. The solution is to add a write_msg_lock in atransport struct. And it kicks the transport only after sending a whole message. By kicking all transports before exit, we ensure that we don't write part of a message to any device. So next time we start adb server, the device should be waiting for a new message. 2. When adb server is killed while pulling a large file from device, the device is still trying to send the unfinished large message. So adb on host usually reads data with EOVERFLOW error. This is because adb on host is reading less than one packet sent from device. The solution is to use buffered read on host. The max packet size of bulk transactions in USB 3.0 is 1024 bytes. By preparing an at least 1024 bytes buffer when reading, EOVERFLOW no longer occurs. And teach adb host to ignore wrong messages. To be safe, this patch doesn't change any logic on device. Bug: http://b/32952319 Test: run python -m unittest -q test_device.DeviceOfflineTest Test: on linux/mac/windows with bullhead, ryu. Change-Id: Ib149d30028a62a6f03857b8a95ab5a1d6e9b9c4e
2017-03-11 01:01:01 +01:00
std::atomic<ConnectionState> connection_state_;
#if ADB_HOST
std::deque<std::shared_ptr<RSA>> keys_;
#endif
// A sharable object that can be used to wait for the atransport's
// connection to be established.
std::shared_ptr<ConnectionWaitable> connection_waitable_;
// The underlying connection object.
std::shared_ptr<Connection> connection_ GUARDED_BY(mutex_);
// A callback that will be invoked when the atransport needs to reconnect.
ReconnectCallback reconnect_;
std::mutex mutex_;
DISALLOW_COPY_AND_ASSIGN(atransport);
};
/*
* Obtain a transport from the available transports.
* If serial is non-null then only the device with that serial will be chosen.
* If transport_id is non-zero then only the device with that transport ID will be chosen.
* If multiple devices/emulators would match, *is_ambiguous (if non-null)
* is set to true and nullptr returned.
* If no suitable transport is found, error is set and nullptr returned.
*/
atransport* acquire_one_transport(TransportType type, const char* serial, TransportId transport_id,
bool* is_ambiguous, std::string* error_out,
bool accept_any_state = false);
void kick_transport(atransport* t);
void update_transports(void);
// Iterates across all of the current and pending transports.
// Stops iteration and returns false if fn returns false, otherwise returns true.
bool iterate_transports(std::function<bool(const atransport*)> fn);
void init_reconnect_handler(void);
void init_transport_registration(void);
void init_mdns_transport_discovery(void);
std::string list_transports(bool long_listing);
atransport* find_transport(const char* serial);
void kick_all_tcp_devices();
void kick_all_transports();
void register_transport(atransport* transport);
void register_usb_transport(usb_handle* h, const char* serial,
const char* devpath, unsigned writeable);
/* Connect to a network address and register it as a device */
void connect_device(const std::string& address, std::string* response);
/* cause new transports to be init'd and added to the list */
bool register_socket_transport(unique_fd s, std::string serial, int port, int local,
atransport::ReconnectCallback reconnect, int* error = nullptr);
// This should only be used for transports with connection_state == kCsNoPerm.
void unregister_usb_transport(usb_handle* usb);
bool check_header(apacket* p, atransport* t);
void close_usb_devices();
void close_usb_devices(std::function<bool(const atransport*)> predicate);
void send_packet(apacket* p, atransport* t);
asocket* create_device_tracker(bool long_output);
#endif /* __TRANSPORT_H */