require 'net/ssh/loggable'
require 'net/ssh/ruby_compat'
require 'net/ssh/connection/channel'
require 'net/ssh/connection/constants'
require 'net/ssh/service/forward'
require 'net/ssh/connection/keepalive'
require 'net/ssh/connection/event_loop'
module Net
module SSH
module Connection
# A session class representing the connection service running on top of
# the SSH transport layer. It manages the creation of channels (see
# #open_channel), and the dispatching of messages to the various channels.
# It also encapsulates the SSH event loop (via #loop and #process),
# and serves as a central point-of-reference for all SSH-related services (e.g.
# port forwarding, SFTP, SCP, etc.).
#
# You will rarely (if ever) need to instantiate this class directly; rather,
# you'll almost always use Net::SSH.start to initialize a new network
# connection, authenticate a user, and return a new connection session,
# all in one call.
#
# Net::SSH.start("localhost", "user") do |ssh|
# # 'ssh' is an instance of Net::SSH::Connection::Session
# ssh.exec! "/etc/init.d/some_process start"
# end
class Session
include Loggable
include Constants
# Default IO.select timeout threshold
DEFAULT_IO_SELECT_TIMEOUT = 300
# The underlying transport layer abstraction (see Net::SSH::Transport::Session).
attr_reader :transport
# The map of options that were used to initialize this instance.
attr_reader :options
# The collection of custom properties for this instance. (See #[] and #[]=).
attr_reader :properties
# The map of channels, each key being the local-id for the channel.
attr_reader :channels #:nodoc:
# The map of listeners that the event loop knows about. See #listen_to.
attr_reader :listeners #:nodoc:
# The map of specialized handlers for opening specific channel types. See
# #on_open_channel.
attr_reader :channel_open_handlers #:nodoc:
# The list of callbacks for pending requests. See #send_global_request.
attr_reader :pending_requests #:nodoc:
class NilChannel
def initialize(session)
@session = session
end
def method_missing(sym, *args)
@session.lwarn { "ignoring request #{sym.inspect} for non-existent (closed?) channel; probably ssh server bug" }
end
end
# Create a new connection service instance atop the given transport
# layer. Initializes the listeners to be only the underlying socket object.
def initialize(transport, options={})
self.logger = transport.logger
@transport = transport
@options = options
@channel_id_counter = -1
@channels = Hash.new(NilChannel.new(self))
@listeners = { transport.socket => nil }
@pending_requests = []
@channel_open_handlers = {}
@on_global_request = {}
@properties = (options[:properties] || {}).dup
@max_pkt_size = (options.key?(:max_pkt_size) ? options[:max_pkt_size] : 0x8000)
@max_win_size = (options.key?(:max_win_size) ? options[:max_win_size] : 0x20000)
@keepalive = Keepalive.new(self)
@event_loop = options[:event_loop] || SingleSessionEventLoop.new
@event_loop.register(self)
end
# Retrieves a custom property from this instance. This can be used to
# store additional state in applications that must manage multiple
# SSH connections.
def [](key)
@properties[key]
end
# Sets a custom property for this instance.
def []=(key, value)
@properties[key] = value
end
# Returns the name of the host that was given to the transport layer to
# connect to.
def host
transport.host
end
# Returns true if the underlying transport has been closed. Note that
# this can be a little misleading, since if the remote server has
# closed the connection, the local end will still think it is open
# until the next operation on the socket. Nevertheless, this method can
# be useful if you just want to know if _you_ have closed the connection.
def closed?
transport.closed?
end
# Closes the session gracefully, blocking until all channels have
# successfully closed, and then closes the underlying transport layer
# connection.
def close
info { "closing remaining channels (#{channels.length} open)" }
channels.each { |id, channel| channel.close }
begin
loop(0.1) { channels.any? }
rescue Net::SSH::Disconnect
raise unless channels.empty?
end
transport.close
end
# Performs a "hard" shutdown of the connection. In general, this should
# never be done, but it might be necessary (in a rescue clause, for instance,
# when the connection needs to close but you don't know the status of the
# underlying protocol's state).
def shutdown!
transport.shutdown!
end
# preserve a reference to Kernel#loop
alias :loop_forever :loop
# Returns +true+ if there are any channels currently active on this
# session. By default, this will not include "invisible" channels
# (such as those created by forwarding ports and such), but if you pass
# a +true+ value for +include_invisible+, then those will be counted.
#
# This can be useful for determining whether the event loop should continue
# to be run.
#
# ssh.loop { ssh.busy? }
def busy?(include_invisible=false)
if include_invisible
channels.any?
else
channels.any? { |id, ch| !ch[:invisible] }
end
end
# The main event loop. Calls #process until #process returns false. If a
# block is given, it is passed to #process, otherwise a default proc is
# used that just returns true if there are any channels active (see #busy?).
# The # +wait+ parameter is also passed through to #process (where it is
# interpreted as the maximum number of seconds to wait for IO.select to return).
#
# # loop for as long as there are any channels active
# ssh.loop
#
# # loop for as long as there are any channels active, but make sure
# # the event loop runs at least once per 0.1 second
# ssh.loop(0.1)
#
# # loop until ctrl-C is pressed
# int_pressed = false
# trap("INT") { int_pressed = true }
# ssh.loop(0.1) { not int_pressed }
def loop(wait=nil, &block)
running = block || Proc.new { busy? }
loop_forever { break unless process(wait, &running) }
begin
process(0)
rescue IOError => e
if e.message =~ /closed/
debug { "stream was closed after loop => shallowing exception so it will be re-raised in next loop" }
else
raise
end
end
end
# The core of the event loop. It processes a single iteration of the event
# loop. If a block is given, it should return false when the processing
# should abort, which causes #process to return false. Otherwise,
# #process returns true. The session itself is yielded to the block as its
# only argument.
#
# If +wait+ is nil (the default), this method will block until any of the
# monitored IO objects are ready to be read from or written to. If you want
# it to not block, you can pass 0, or you can pass any other numeric value
# to indicate that it should block for no more than that many seconds.
# Passing 0 is a good way to poll the connection, but if you do it too
# frequently it can make your CPU quite busy!
#
# This will also cause all active channels to be processed once each (see
# Net::SSH::Connection::Channel#on_process).
#
# TODO revise example
#
# # process multiple Net::SSH connections in parallel
# connections = [
# Net::SSH.start("host1", ...),
# Net::SSH.start("host2", ...)
# ]
#
# connections.each do |ssh|
# ssh.exec "grep something /in/some/files"
# end
#
# condition = Proc.new { |s| s.busy? }
#
# loop do
# connections.delete_if { |ssh| !ssh.process(0.1, &condition) }
# break if connections.empty?
# end
def process(wait=nil, &block)
@event_loop.process(wait, &block)
rescue StandardError
force_channel_cleanup_on_close if closed?
raise
end
# This is called internally as part of #process. It dispatches any
# available incoming packets, and then runs Net::SSH::Connection::Channel#process
# for any active channels. If a block is given, it is invoked at the
# start of the method and again at the end, and if the block ever returns
# false, this method returns false. Otherwise, it returns true.
def preprocess(&block)
return false if block_given? && !yield(self)
ev_preprocess(&block)
return false if block_given? && !yield(self)
return true
end
# Called by event loop to process available data before going to
# event multiplexing
def ev_preprocess(&block)
dispatch_incoming_packets(raise_disconnect_errors: false)
each_channel { |id, channel| channel.process unless channel.local_closed? }
end
# Returns the file descriptors the event loop should wait for read/write events,
# we also return the max wait
def ev_do_calculate_rw_wait(wait)
r = listeners.keys
w = r.select { |w2| w2.respond_to?(:pending_write?) && w2.pending_write? }
[r,w,io_select_wait(wait)]
end
# This is called internally as part of #process.
def postprocess(readers, writers)
ev_do_handle_events(readers, writers)
end
# It loops over the given arrays of reader IO's and writer IO's,
# processing them as needed, and
# then calls Net::SSH::Transport::Session#rekey_as_needed to allow the
# transport layer to rekey. Then returns true.
def ev_do_handle_events(readers, writers)
Array(readers).each do |reader|
if listeners[reader]
listeners[reader].call(reader)
else
if reader.fill.zero?
reader.close
stop_listening_to(reader)
end
end
end
Array(writers).each do |writer|
writer.send_pending
end
end
# calls Net::SSH::Transport::Session#rekey_as_needed to allow the
# transport layer to rekey
def ev_do_postprocess(was_events)
@keepalive.send_as_needed(was_events)
transport.rekey_as_needed
true
end
# Send a global request of the given type. The +extra+ parameters must
# be even in number, and conform to the same format as described for
# Net::SSH::Buffer.from. If a callback is not specified, the request will
# not require a response from the server, otherwise the server is required
# to respond and indicate whether the request was successful or not. This
# success or failure is indicated by the callback being invoked, with the
# first parameter being true or false (success, or failure), and the second
# being the packet itself.
#
# Generally, Net::SSH will manage global requests that need to be sent
# (e.g. port forward requests and such are handled in the Net::SSH::Service::Forward
# class, for instance). However, there may be times when you need to
# send a global request that isn't explicitly handled by Net::SSH, and so
# this method is available to you.
#
# ssh.send_global_request("keep-alive@openssh.com")
def send_global_request(type, *extra, &callback)
info { "sending global request #{type}" }
msg = Buffer.from(:byte, GLOBAL_REQUEST, :string, type.to_s, :bool, !callback.nil?, *extra)
send_message(msg)
pending_requests << callback if callback
self
end
# Requests that a new channel be opened. By default, the channel will be
# of type "session", but if you know what you're doing you can select any
# of the channel types supported by the SSH protocol. The +extra+ parameters
# must be even in number and conform to the same format as described for
# Net::SSH::Buffer.from. If a callback is given, it will be invoked when
# the server confirms that the channel opened successfully. The sole parameter
# for the callback is the channel object itself.
#
# In general, you'll use #open_channel without any arguments; the only
# time you'd want to set the channel type or pass additional initialization
# data is if you were implementing an SSH extension.
#
# channel = ssh.open_channel do |ch|
# ch.exec "grep something /some/files" do |ch, success|
# ...
# end
# end
#
# channel.wait
def open_channel(type="session", *extra, &on_confirm)
local_id = get_next_channel_id
channel = Channel.new(self, type, local_id, @max_pkt_size, @max_win_size, &on_confirm)
msg = Buffer.from(:byte, CHANNEL_OPEN, :string, type, :long, local_id,
:long, channel.local_maximum_window_size,
:long, channel.local_maximum_packet_size, *extra)
send_message(msg)
channels[local_id] = channel
end
class StringWithExitstatus < String
def initialize(str, exitstatus)
super(str)
@exitstatus = exitstatus
end
attr_reader :exitstatus
end
# A convenience method for executing a command and interacting with it. If
# no block is given, all output is printed via $stdout and $stderr. Otherwise,
# the block is called for each data and extended data packet, with three
# arguments: the channel object, a symbol indicating the data type
# (:stdout or :stderr), and the data (as a string).
#
# Note that this method returns immediately, and requires an event loop
# (see Session#loop) in order for the command to actually execute.
#
# This is effectively identical to calling #open_channel, and then
# Net::SSH::Connection::Channel#exec, and then setting up the channel
# callbacks. However, for most uses, this will be sufficient.
#
# ssh.exec "grep something /some/files" do |ch, stream, data|
# if stream == :stderr
# puts "ERROR: #{data}"
# else
# puts data
# end
# end
def exec(command, status: nil, &block)
open_channel do |channel|
channel.exec(command) do |ch, success|
raise "could not execute command: #{command.inspect}" unless success
if status
channel.on_request("exit-status") do |ch2,data|
status[:exit_code] = data.read_long
end
channel.on_request("exit-signal") do |ch2, data|
status[:exit_signal] = data.read_long
end
end
channel.on_data do |ch2, data|
if block
block.call(ch2, :stdout, data)
else
$stdout.print(data)
end
end
channel.on_extended_data do |ch2, type, data|
if block
block.call(ch2, :stderr, data)
else
$stderr.print(data)
end
end
end
end
end
# Same as #exec, except this will block until the command finishes. Also,
# if no block is given, this will return all output (stdout and stderr)
# as a single string.
#
# matches = ssh.exec!("grep something /some/files")
#
# the returned string has an exitstatus method to query it's exit satus
def exec!(command, status: nil, &block)
block_or_concat = block || Proc.new do |ch, type, data|
ch[:result] ||= ""
ch[:result] << data
end
status ||= {}
channel = exec(command, status: status, &block_or_concat)
channel.wait
channel[:result] ||= "" unless block
channel[:result] &&= channel[:result].force_encoding("UTF-8") unless block
StringWithExitstatus.new(channel[:result], status[:exit_code]) if channel[:result]
end
# Enqueues a message to be sent to the server as soon as the socket is
# available for writing. Most programs will never need to call this, but
# if you are implementing an extension to the SSH protocol, or if you
# need to send a packet that Net::SSH does not directly support, you can
# use this to send it.
#
# ssh.send_message(Buffer.from(:byte, REQUEST_SUCCESS).to_s)
def send_message(message)
transport.enqueue_message(message)
end
# Adds an IO object for the event loop to listen to. If a callback
# is given, it will be invoked when the io is ready to be read, otherwise,
# the io will merely have its #fill method invoked.
#
# Any +io+ value passed to this method _must_ have mixed into it the
# Net::SSH::BufferedIo functionality, typically by calling #extend on the
# object.
#
# The following example executes a process on the remote server, opens
# a socket to somewhere, and then pipes data from that socket to the
# remote process' stdin stream:
#
# channel = ssh.open_channel do |ch|
# ch.exec "/some/process/that/wants/input" do |ch, success|
# abort "can't execute!" unless success
#
# io = TCPSocket.new(somewhere, port)
# io.extend(Net::SSH::BufferedIo)
# ssh.listen_to(io)
#
# ch.on_process do
# if io.available > 0
# ch.send_data(io.read_available)
# end
# end
#
# ch.on_close do
# ssh.stop_listening_to(io)
# io.close
# end
# end
# end
#
# channel.wait
def listen_to(io, &callback)
listeners[io] = callback
end
# Removes the given io object from the listeners collection, so that the
# event loop will no longer monitor it.
def stop_listening_to(io)
listeners.delete(io)
end
# Returns a reference to the Net::SSH::Service::Forward service, which can
# be used for forwarding ports over SSH.
def forward
@forward ||= Service::Forward.new(self)
end
# Registers a handler to be invoked when the server wants to open a
# channel on the client. The callback receives the connection object,
# the new channel object, and the packet itself as arguments, and should
# raise ChannelOpenFailed if it is unable to open the channel for some
# reason. Otherwise, the channel will be opened and a confirmation message
# sent to the server.
#
# This is used by the Net::SSH::Service::Forward service to open a channel
# when a remote forwarded port receives a connection. However, you are
# welcome to register handlers for other channel types, as needed.
def on_open_channel(type, &block)
channel_open_handlers[type] = block
end
# Registers a handler to be invoked when the server sends a global request
# of the given type. The callback receives the request data as the first
# parameter, and true/false as the second (indicating whether a response
# is required). If the callback sends the response, it should return
# :sent. Otherwise, if it returns true, REQUEST_SUCCESS will be sent, and
# if it returns false, REQUEST_FAILURE will be sent.
def on_global_request(type, &block)
old, @on_global_request[type] = @on_global_request[type], block
old
end
def cleanup_channel(channel)
if channel.local_closed? and channel.remote_closed?
info { "#{host} delete channel #{channel.local_id} which closed locally and remotely" }
channels.delete(channel.local_id)
end
end
# If the #preprocess and #postprocess callbacks for this session need to run
# periodically, this method returns the maximum number of seconds which may
# pass between callbacks.
def max_select_wait_time
@keepalive.interval if @keepalive.enabled?
end
private
# iterate channels with the posibility of callbacks opening new channels during the iteration
def each_channel(&block)
channels.dup.each(&block)
end
# Read all pending packets from the connection and dispatch them as
# appropriate. Returns as soon as there are no more pending packets.
def dispatch_incoming_packets(raise_disconnect_errors: true)
while packet = transport.poll_message
raise Net::SSH::Exception, "unexpected response #{packet.type} (#{packet.inspect})" unless MAP.key?(packet.type)
send(MAP[packet.type], packet)
end
rescue StandardError
force_channel_cleanup_on_close if closed?
raise if raise_disconnect_errors || !$!.is_a?(Net::SSH::Disconnect)
end
# Returns the next available channel id to be assigned, and increments
# the counter.
def get_next_channel_id
@channel_id_counter += 1
end
def force_channel_cleanup_on_close
channels.each do |id, channel|
channel_closed(channel)
end
end
def channel_closed(channel)
channel.remote_closed!
channel.close
cleanup_channel(channel)
channel.do_close
end
# Invoked when a global request is received. The registered global
# request callback will be invoked, if one exists, and the necessary
# reply returned.
def global_request(packet)
info { "global request received: #{packet[:request_type]} #{packet[:want_reply]}" }
callback = @on_global_request[packet[:request_type]]
result = callback ? callback.call(packet[:request_data], packet[:want_reply]) : false
raise "expected global request handler for `#{packet[:request_type]}' to return true, false, or :sent, but got #{result.inspect}" if result != :sent && result != true && result != false
if packet[:want_reply] && result != :sent
msg = Buffer.from(:byte, result ? REQUEST_SUCCESS : REQUEST_FAILURE)
send_message(msg)
end
end
# Invokes the next pending request callback with +true+.
def request_success(packet)
info { "global request success" }
callback = pending_requests.shift
callback.call(true, packet) if callback
end
# Invokes the next pending request callback with +false+.
def request_failure(packet)
info { "global request failure" }
callback = pending_requests.shift
callback.call(false, packet) if callback
end
# Called when the server wants to open a channel. If no registered
# channel handler exists for the given channel type, CHANNEL_OPEN_FAILURE
# is returned, otherwise the callback is invoked and everything proceeds
# accordingly.
def channel_open(packet)
info { "channel open #{packet[:channel_type]}" }
local_id = get_next_channel_id
channel = Channel.new(self, packet[:channel_type], local_id, @max_pkt_size, @max_win_size)
channel.do_open_confirmation(packet[:remote_id], packet[:window_size], packet[:packet_size])
callback = channel_open_handlers[packet[:channel_type]]
if callback
begin
callback[self, channel, packet]
rescue ChannelOpenFailed => err
failure = [err.code, err.reason]
else
channels[local_id] = channel
msg = Buffer.from(:byte, CHANNEL_OPEN_CONFIRMATION, :long, channel.remote_id, :long, channel.local_id, :long, channel.local_maximum_window_size, :long, channel.local_maximum_packet_size)
end
else
failure = [3, "unknown channel type #{channel.type}"]
end
if failure
error { failure.inspect }
msg = Buffer.from(:byte, CHANNEL_OPEN_FAILURE, :long, channel.remote_id, :long, failure[0], :string, failure[1], :string, "")
end
send_message(msg)
end
def channel_open_confirmation(packet)
info { "channel_open_confirmation: #{packet[:local_id]} #{packet[:remote_id]} #{packet[:window_size]} #{packet[:packet_size]}" }
channel = channels[packet[:local_id]]
channel.do_open_confirmation(packet[:remote_id], packet[:window_size], packet[:packet_size])
end
def channel_open_failure(packet)
error { "channel_open_failed: #{packet[:local_id]} #{packet[:reason_code]} #{packet[:description]}" }
channel = channels.delete(packet[:local_id])
channel.do_open_failed(packet[:reason_code], packet[:description])
end
def channel_window_adjust(packet)
info { "channel_window_adjust: #{packet[:local_id]} +#{packet[:extra_bytes]}" }
channels[packet[:local_id]].do_window_adjust(packet[:extra_bytes])
end
def channel_request(packet)
info { "channel_request: #{packet[:local_id]} #{packet[:request]} #{packet[:want_reply]}" }
channels[packet[:local_id]].do_request(packet[:request], packet[:want_reply], packet[:request_data])
end
def channel_data(packet)
info { "channel_data: #{packet[:local_id]} #{packet[:data].length}b" }
channels[packet[:local_id]].do_data(packet[:data])
end
def channel_extended_data(packet)
info { "channel_extended_data: #{packet[:local_id]} #{packet[:data_type]} #{packet[:data].length}b" }
channels[packet[:local_id]].do_extended_data(packet[:data_type], packet[:data])
end
def channel_eof(packet)
info { "channel_eof: #{packet[:local_id]}" }
channels[packet[:local_id]].do_eof
end
def channel_close(packet)
info { "channel_close: #{packet[:local_id]}" }
channel = channels[packet[:local_id]]
channel_closed(channel)
end
def channel_success(packet)
info { "channel_success: #{packet[:local_id]}" }
channels[packet[:local_id]].do_success
end
def channel_failure(packet)
info { "channel_failure: #{packet[:local_id]}" }
channels[packet[:local_id]].do_failure
end
def io_select_wait(wait)
[wait, max_select_wait_time].compact.min
end
MAP = Constants.constants.each_with_object({}) do |name, memo|
value = const_get(name)
next unless Integer === value
memo[value] = name.downcase.to_sym
end
end
end
end
end