lib/net/ssh/connection/channel.rb



require 'net/ssh/loggable'
require 'net/ssh/connection/constants'
require 'net/ssh/connection/term'

module Net; module SSH; module Connection

  # The channel abstraction. Multiple "channels" can be multiplexed onto a
  # single SSH channel, each operating independently and seemingly in parallel.
  # This class represents a single such channel. Most operations performed
  # with the Net::SSH library will involve using one or more channels.
  #
  # Channels are intended to be used asynchronously. You request that one be
  # opened (via Connection::Session#open_channel), and when it is opened, your
  # callback is invoked. Then, you set various other callbacks on the newly
  # opened channel, which are called in response to the corresponding events.
  # Programming with Net::SSH works best if you think of your programs as
  # state machines. Complex programs are best implemented as objects that
  # wrap a channel. See Net::SCP and Net::SFTP for examples of how complex
  # state machines can be built on top of the SSH protocol.
  #
  #   ssh.open_channel do |channel|
  #     channel.exec("/invoke/some/command") do |ch, success|
  #       abort "could not execute command" unless success
  #
  #       channel.on_data do |ch, data|
  #         puts "got stdout: #{data}"
  #         channel.send_data "something for stdin\n"
  #       end
  #
  #       channel.on_extended_data do |ch, type, data|
  #         puts "got stderr: #{data}"
  #       end
  #
  #       channel.on_close do |ch|
  #         puts "channel is closing!"
  #       end
  #     end
  #   end
  #
  #   ssh.loop
  #
  # Channels also have a basic hash-like interface, that allows programs to
  # store arbitrary state information on a channel object. This helps simplify
  # the writing of state machines, especially when you may be juggling
  # multiple open channels at the same time.
  #
  # Note that data sent across SSH channels are governed by maximum packet
  # sizes and maximum window sizes. These details are managed internally
  # by Net::SSH::Connection::Channel, so you may remain blissfully ignorant
  # if you so desire, but you can always inspect the current maximums, as
  # well as the remaining window size, using the reader attributes for those
  # values.
  class Channel
    include Constants, Loggable

    # The local id for this channel, assigned by the Net::SSH::Connection::Session instance.
    attr_reader :local_id

    # The remote id for this channel, assigned by the remote host.
    attr_reader :remote_id

    # The type of this channel, usually "session".
    attr_reader :type

    # The underlying Net::SSH::Connection::Session instance that supports this channel.
    attr_reader :connection

    # The maximum packet size that the local host can receive.
    attr_reader :local_maximum_packet_size

    # The maximum amount of data that the local end of this channel can
    # receive. This is a total, not per-packet.
    attr_reader :local_maximum_window_size

    # The maximum packet size that the remote host can receive.
    attr_reader :remote_maximum_packet_size

    # The maximum amount of data that the remote end of this channel can
    # receive. This is a total, not per-packet.
    attr_reader :remote_maximum_window_size

    # This is the remaining window size on the local end of this channel. When
    # this reaches zero, no more data can be received.
    attr_reader :local_window_size

    # This is the remaining window size on the remote end of this channel. When
    # this reaches zero, no more data can be sent.
    attr_reader :remote_window_size

    # A hash of properties for this channel. These can be used to store state
    # information about this channel. See also #[] and #[]=.
    attr_reader :properties

    # The output buffer for this channel. Data written to the channel is
    # enqueued here, to be written as CHANNEL_DATA packets during each pass of
    # the event loop. See Connection::Session#process and #enqueue_pending_output.
    attr_reader :output #:nodoc:

    # The list of pending requests. Each time a request is sent which requires
    # a reply, the corresponding callback is pushed onto this queue. As responses
    # arrive, they are shifted off the front and handled.
    attr_reader :pending_requests #:nodoc:

    # Instantiates a new channel on the given connection, of the given type,
    # and with the given id. If a block is given, it will be remembered until
    # the channel is confirmed open by the server, and will be invoked at
    # that time (see #do_open_confirmation).
    #
    # This also sets the default maximum packet size and maximum window size.
    def initialize(connection, type, local_id, max_pkt_size = 0x8000, max_win_size = 0x20000, &on_confirm_open)
      self.logger = connection.logger

      @connection = connection
      @type       = type
      @local_id   = local_id

      @local_maximum_packet_size = max_pkt_size
      @local_window_size = @local_maximum_window_size = max_win_size

      @on_confirm_open = on_confirm_open

      @output = Buffer.new

      @properties = {}

      @pending_requests = []
      @on_open_failed = @on_data = @on_extended_data = @on_process = @on_close = @on_eof = nil
      @on_request = {}
      @closing = @eof = @sent_eof = @local_closed = @remote_closed = false
    end

    # A shortcut for accessing properties of the channel (see #properties).
    def [](name)
      @properties[name]
    end

    # A shortcut for setting properties of the channel (see #properties).
    def []=(name, value)
      @properties[name] = value
    end

    # Syntactic sugar for executing a command. Sends a channel request asking
    # that the given command be invoked. If the block is given, it will be
    # called when the server responds. The first parameter will be the
    # channel, and the second will be true or false, indicating whether the
    # request succeeded or not. In this case, success means that the command
    # is being executed, not that it has completed, and failure means that the
    # command altogether failed to be executed.
    #
    #   channel.exec "ls -l /home" do |ch, success|
    #     if success
    #       puts "command has begun executing..."
    #       # this is a good place to hang callbacks like #on_data...
    #     else
    #       puts "alas! the command could not be invoked!"
    #     end
    #   end
    def exec(command, &block)
      send_channel_request("exec", :string, command, &block)
    end

    # Syntactic sugar for requesting that a subsystem be started. Subsystems
    # are a way for other protocols (like SFTP) to be run, using SSH as
    # the transport. Generally, you'll never need to call this directly unless
    # you are the implementor of something that consumes an SSH subsystem, like
    # SFTP.
    #
    #   channel.subsystem("sftp") do |ch, success|
    #     if success
    #       puts "subsystem successfully started"
    #     else
    #       puts "subsystem could not be started"
    #     end
    #   end
    def subsystem(subsystem, &block)
      send_channel_request("subsystem", :string, subsystem, &block)
    end

    # Syntactic sugar for setting an environment variable in the remote
    # process' environment. Note that for security reasons, the server may
    # refuse to set certain environment variables, or all, at the server's
    # discretion. If you are connecting to an OpenSSH server, you will
    # need to update the AcceptEnv setting in the sshd_config to include the
    # environment variables you want to send.
    #
    #   channel.env "PATH", "/usr/local/bin"
    def env(variable_name, variable_value, &block)
      send_channel_request("env", :string, variable_name, :string, variable_value, &block)
    end

    # A hash of the valid PTY options (see #request_pty).
    VALID_PTY_OPTIONS = { :term        => "xterm",
                          :chars_wide  => 80,
                          :chars_high  => 24,
                          :pixels_wide => 640,
                          :pixels_high => 480,
                          :modes       => {} }

    # Requests that a pseudo-tty (or "pty") be made available for this channel.
    # This is useful when you want to invoke and interact with some kind of
    # screen-based program (e.g., vim, or some menuing system).
    #
    # Note, that without a pty some programs (e.g. sudo, or subversion) on
    # some systems, will not be able to run interactively, and will error
    # instead of prompt if they ever need some user interaction.
    #
    # Note, too, that when a pty is requested, user's shell configuration
    # scripts (.bashrc and such) are not run by default, whereas they are
    # run when a pty is not present.
    #
    #   channel.request_pty do |ch, success|
    #     if success
    #       puts "pty successfully obtained"
    #     else
    #       puts "could not obtain pty"
    #     end
    #   end
    def request_pty(opts={}, &block)
      extra = opts.keys - VALID_PTY_OPTIONS.keys
      raise ArgumentError, "invalid option(s) to request_pty: #{extra.inspect}" if extra.any?

      opts = VALID_PTY_OPTIONS.merge(opts)

      modes = opts[:modes].inject(Buffer.new) do |memo, (mode, data)|
        memo.write_byte(mode).write_long(data)
      end
      # mark the end of the mode opcode list with a 0 byte
      modes.write_byte(0)

      send_channel_request("pty-req", :string, opts[:term],
        :long, opts[:chars_wide], :long, opts[:chars_high],
        :long, opts[:pixels_wide], :long, opts[:pixels_high],
        :string, modes.to_s, &block)
    end

    # Sends data to the channel's remote endpoint. This usually has the
    # effect of sending the given string to the remote process' stdin stream.
    # Note that it does not immediately send the data across the channel,
    # but instead merely appends the given data to the channel's output buffer,
    # preparatory to being packaged up and sent out the next time the connection
    # is accepting data. (A connection might not be accepting data if, for
    # instance, it has filled its data window and has not yet been resized by
    # the remote end-point.)
    #
    # This will raise an exception if the channel has previously declared
    # that no more data will be sent (see #eof!).
    #
    #   channel.send_data("the password\n")
    def send_data(data)
      raise EOFError, "cannot send data if channel has declared eof" if eof?
      output.append(data.to_s)
    end

    # Returns true if the channel exists in the channel list of the session,
    # and false otherwise. This can be used to determine whether a channel has
    # been closed or not.
    #
    #   ssh.loop { channel.active? }
    def active?
      connection.channels.key?(local_id)
    end

    # Runs the SSH event loop until the channel is no longer active. This is
    # handy for blocking while you wait for some channel to finish.
    #
    #   channel.exec("grep ...") { ... }
    #   channel.wait
    def wait
      connection.loop { active? }
    end

    # True if close() has been called; NOTE: if the channel has data waiting to
    # be sent then the channel will close after all the data is sent. See
    # closed?() to determine if we have actually sent CHANNEL_CLOSE to server.
    # This may be true for awhile before closed? returns true if we are still
    # sending buffered output to server.
    def closing?
      @closing
    end

    # True if we have sent CHANNEL_CLOSE to the remote server.
    def local_closed?
      @local_closed
    end

    def remote_closed?
      @remote_closed
    end

    def remote_closed!
      @remote_closed = true
    end

    # Requests that the channel be closed. It only marks the channel to be closed
    # the CHANNEL_CLOSE message will be sent from event loop
    def close
      return if @closing
      @closing = true
    end

    # Returns true if the local end of the channel has declared that no more
    # data is forthcoming (see #eof!). Trying to send data via #send_data when
    # this is true will result in an exception being raised.
    def eof?
      @eof
    end

    # Tells the remote end of the channel that no more data is forthcoming
    # from this end of the channel. The remote end may still send data.
    # The CHANNEL_EOF packet will be sent once the output buffer is empty.
    def eof!
      return if eof?
      @eof = true
    end

    # If an #on_process handler has been set up, this will cause it to be
    # invoked (passing the channel itself as an argument). It also causes all
    # pending output to be enqueued as CHANNEL_DATA packets (see #enqueue_pending_output).
    def process
      @on_process.call(self) if @on_process
      enqueue_pending_output

      if @eof and not @sent_eof and output.empty? and remote_id and not @local_closed
        connection.send_message(Buffer.from(:byte, CHANNEL_EOF, :long, remote_id))
        @sent_eof = true
      end

      if @closing and not @local_closed and output.empty? and remote_id
        connection.send_message(Buffer.from(:byte, CHANNEL_CLOSE, :long, remote_id))
        @local_closed = true
        connection.cleanup_channel(self)
      end
    end

    # Registers a callback to be invoked when data packets are received by the
    # channel. The callback is called with the channel as the first argument,
    # and the data as the second.
    #
    #   channel.on_data do |ch, data|
    #     puts "got data: #{data.inspect}"
    #   end
    #
    # Data received this way is typically the data written by the remote
    # process to its +stdout+ stream.
    def on_data(&block)
      old, @on_data = @on_data, block
      old
    end

    # Registers a callback to be invoked when extended data packets are received
    # by the channel. The callback is called with the channel as the first
    # argument, the data type (as an integer) as the second, and the data as
    # the third. Extended data is almost exclusively used to send +stderr+ data
    # (+type+ == 1). Other extended data types are not defined by the SSH
    # protocol.
    #
    #   channel.on_extended_data do |ch, type, data|
    #     puts "got stderr: #{data.inspect}"
    #   end
    def on_extended_data(&block)
      old, @on_extended_data = @on_extended_data, block
      old
    end

    # Registers a callback to be invoked for each pass of the event loop for
    # this channel. There are no guarantees on timeliness in the event loop,
    # but it will be called roughly once for each packet received by the
    # connection (not the channel). This callback is invoked with the channel
    # as the sole argument.
    #
    # Here's an example that accumulates the channel data into a variable on
    # the channel itself, and displays individual lines in the input one
    # at a time when the channel is processed:
    #
    #   channel[:data] = ""
    #
    #   channel.on_data do |ch, data|
    #     channel[:data] << data
    #   end
    #
    #   channel.on_process do |ch|
    #     if channel[:data] =~ /^.*?\n/
    #       puts $&
    #       channel[:data] = $'
    #     end
    #   end
    def on_process(&block)
      old, @on_process = @on_process, block
      old
    end

    # Registers a callback to be invoked when the server acknowledges that a
    # channel is closed. This is invoked with the channel as the sole argument.
    #
    #   channel.on_close do |ch|
    #     puts "remote end is closing!"
    #   end
    def on_close(&block)
      old, @on_close = @on_close, block
      old
    end

    # Registers a callback to be invoked when the server indicates that no more
    # data will be sent to the channel (although the channel can still send
    # data to the server). The channel is the sole argument to the callback.
    #
    #   channel.on_eof do |ch|
    #     puts "remote end is done sending data"
    #   end
    def on_eof(&block)
      old, @on_eof = @on_eof, block
      old
    end

    # Registers a callback to be invoked when the server was unable to open
    # the requested channel. The channel itself will be passed to the block,
    # along with the integer "reason code" for the failure, and a textual
    # description of the failure from the server.
    #
    #   channel = session.open_channel do |ch|
    #     # ..
    #   end
    #
    #   channel.on_open_failed { |ch, code, desc| ... }
    def on_open_failed(&block)
      old, @on_open_failed = @on_open_failed, block
      old
    end

    # Registers a callback to be invoked when a channel request of the given
    # type is received. The callback will receive the channel as the first
    # argument, and the associated (unparsed) data as the second. The data
    # will be a Net::SSH::Buffer that you will need to parse, yourself,
    # according to the kind of request you are watching.
    #
    # By default, if the request wants a reply, Net::SSH will send a
    # CHANNEL_SUCCESS response for any request that was handled by a registered
    # callback, and CHANNEL_FAILURE for any that wasn't, but if you want your
    # registered callback to result in a CHANNEL_FAILURE response, just raise
    # Net::SSH::ChannelRequestFailed.
    #
    # Some common channel requests that your programs might want to listen
    # for are:
    #
    # * "exit-status" : the exit status of the remote process will be reported
    #   as a long integer in the data buffer, which you can grab via
    #   data.read_long.
    # * "exit-signal" : if the remote process died as a result of a signal
    #   being sent to it, the signal will be reported as a string in the
    #   data, via data.read_string. (Not all SSH servers support this channel
    #   request type.)
    #
    #     channel.on_request "exit-status" do |ch, data|
    #       puts "process terminated with exit status: #{data.read_long}"
    #     end
    def on_request(type, &block)
      old, @on_request[type] = @on_request[type], block
      old
    end

    # Sends a new channel request with the given name. The extra +data+
    # parameter must either be empty, or consist of an even number of
    # arguments. See Net::SSH::Buffer.from for a description of their format.
    # If a block is given, it is registered as a callback for a pending
    # request, and the packet will be flagged so that the server knows a
    # reply is required. If no block is given, the server will send no
    # response to this request. Responses, where required, will cause the
    # callback to be invoked with the channel as the first argument, and
    # either true or false as the second, depending on whether the request
    # succeeded or not. The meaning of "success" and "failure" in this context
    # is dependent on the specific request that was sent.
    #
    #   channel.send_channel_request "shell" do |ch, success|
    #     if success
    #       puts "user shell started successfully"
    #     else
    #       puts "could not start user shell"
    #     end
    #   end
    #
    # Most channel requests you'll want to send are already wrapped in more
    # convenient helper methods (see #exec and #subsystem).
    def send_channel_request(request_name, *data, &callback)
      info { "sending channel request #{request_name.inspect}" }
      fail "Channel open not yet confirmed, please call send_channel_request(or exec) from block of open_channel" unless remote_id
      msg = Buffer.from(:byte, CHANNEL_REQUEST,
        :long, remote_id, :string, request_name,
        :bool, !callback.nil?, *data)
      connection.send_message(msg)
      pending_requests << callback if callback
    end

    public # these methods are public, but for Net::SSH internal use only

      # Enqueues pending output at the connection as CHANNEL_DATA packets. This
      # does nothing if the channel has not yet been confirmed open (see
      # #do_open_confirmation). This is called automatically by #process, which
      # is called from the event loop (Connection::Session#process). You will
      # generally not need to invoke it directly.
      def enqueue_pending_output #:nodoc:
        return unless remote_id

        while output.length > 0
          length = output.length
          length = remote_window_size if length > remote_window_size
          length = remote_maximum_packet_size if length > remote_maximum_packet_size

          if length > 0
            connection.send_message(Buffer.from(:byte, CHANNEL_DATA, :long, remote_id, :string, output.read(length)))
            output.consume!
            @remote_window_size -= length
          else
            break
          end
        end
      end

      # Invoked when the server confirms that a channel has been opened.
      # The remote_id is the id of the channel as assigned by the remote host,
      # and max_window and max_packet are the maximum window and maximum
      # packet sizes, respectively. If an open-confirmation callback was
      # given when the channel was created, it is invoked at this time with
      # the channel itself as the sole argument.
      def do_open_confirmation(remote_id, max_window, max_packet) #:nodoc:
        @remote_id = remote_id
        @remote_window_size = @remote_maximum_window_size = max_window
        @remote_maximum_packet_size = max_packet
        connection.forward.agent(self) if connection.options[:forward_agent] && type == "session"
        forward_local_env(connection.options[:send_env]) if connection.options[:send_env]
        @on_confirm_open.call(self) if @on_confirm_open
      end

      # Invoked when the server failed to open the channel. If an #on_open_failed
      # callback was specified, it will be invoked with the channel, reason code,
      # and description as arguments. Otherwise, a ChannelOpenFailed exception
      # will be raised.
      def do_open_failed(reason_code, description)
        if @on_open_failed
          @on_open_failed.call(self, reason_code, description)
        else
          raise ChannelOpenFailed.new(reason_code, description)
        end
      end

      # Invoked when the server sends a CHANNEL_WINDOW_ADJUST packet, and
      # causes the remote window size to be adjusted upwards by the given
      # number of bytes. This has the effect of allowing more data to be sent
      # from the local end to the remote end of the channel.
      def do_window_adjust(bytes) #:nodoc:
        @remote_maximum_window_size += bytes
        @remote_window_size += bytes
      end

      # Invoked when the server sends a channel request. If any #on_request
      # callback has been registered for the specific type of this request,
      # it is invoked. If +want_reply+ is true, a packet will be sent of
      # either CHANNEL_SUCCESS or CHANNEL_FAILURE type. If there was no callback
      # to handle the request, CHANNEL_FAILURE will be sent. Otherwise,
      # CHANNEL_SUCCESS, unless the callback raised ChannelRequestFailed. The
      # callback should accept the channel as the first argument, and the
      # request-specific data as the second.
      def do_request(request, want_reply, data) #:nodoc:
        result = true

        begin
          callback = @on_request[request] or raise ChannelRequestFailed
          callback.call(self, data)
        rescue ChannelRequestFailed
          result = false
        end

        if want_reply
          msg = Buffer.from(:byte, result ? CHANNEL_SUCCESS : CHANNEL_FAILURE, :long, remote_id)
          connection.send_message(msg)
        end
      end

      # Invokes the #on_data callback when the server sends data to the
      # channel. This will reduce the available window size on the local end,
      # but does not actually throttle requests that come in illegally when
      # the window size is too small. The callback is invoked with the channel
      # as the first argument, and the data as the second.
      def do_data(data) #:nodoc:
        update_local_window_size(data.length)
        @on_data.call(self, data) if @on_data
      end

      # Invokes the #on_extended_data callback when the server sends
      # extended data to the channel. This will reduce the available window
      # size on the local end. The callback is invoked with the channel,
      # type, and data.
      def do_extended_data(type, data)
        update_local_window_size(data.length)
        @on_extended_data.call(self, type, data) if @on_extended_data
      end

      # Invokes the #on_eof callback when the server indicates that no
      # further data is forthcoming. The callback is invoked with the channel
      # as the argument.
      def do_eof
        @on_eof.call(self) if @on_eof
      end

      # Invokes the #on_close callback when the server closes a channel.
      # The channel is the only argument.
      def do_close
        @on_close.call(self) if @on_close
      end

      # Invokes the next pending request callback with +false+ as the second
      # argument.
      def do_failure
        if callback = pending_requests.shift
          callback.call(self, false)
        else
          error { "channel failure recieved with no pending request to handle it (bug?)" }
        end
      end

      # Invokes the next pending request callback with +true+ as the second
      # argument.
      def do_success
        if callback = pending_requests.shift
          callback.call(self, true)
        else
          error { "channel success recieved with no pending request to handle it (bug?)" }
        end
      end

    private

      # Runs the SSH event loop until the remote confirmed channel open
      # experimental api
      def wait_until_open_confirmed
        connection.loop { !remote_id }
      end

      # Updates the local window size by the given amount. If the window
      # size drops to less than half of the local maximum (an arbitrary
      # threshold), a CHANNEL_WINDOW_ADJUST message will be sent to the
      # server telling it that the window size has grown.
      def update_local_window_size(size)
        @local_window_size -= size
        if local_window_size < local_maximum_window_size/2
          connection.send_message(Buffer.from(:byte, CHANNEL_WINDOW_ADJUST,
            :long, remote_id, :long, 0x20000))
          @local_window_size += 0x20000
          @local_maximum_window_size += 0x20000
        end
      end

      # Gets an +Array+ of local environment variables in the remote process'
      # environment.
      # A variable name can either be described by a +Regexp+ or +String+.
      #
      #   channel.forward_local_env [/^GIT_.*$/, "LANG"]
      def forward_local_env(env_variable_patterns)
        Array(env_variable_patterns).each do |env_variable_pattern|
          matched_variables = ENV.find_all do |env_name, _|
            case env_variable_pattern
            when Regexp then env_name =~ env_variable_pattern
            when String then env_name == env_variable_pattern
            end
          end
          matched_variables.each do |env_name, env_value|
            self.env(env_name, env_value)
          end
        end
      end
  end

end; end; end