tmate/libssh/doc/shell.dox

/**
@page libssh_tutor_shell Chapter 3: Opening a remote shell
@section opening_shell Opening a remote shell

We already mentioned that a single SSH connection can be shared
between several "channels". Channels can be used for different purposes.

This chapter shows how to open one of these channels, and how to use it to
start a command interpreter on a remote computer.


@subsection open_channel Opening and closing a channel

The ssh_channel_new() function creates a channel. It returns the channel as
a variable of type ssh_channel.

Once you have this channel, you open a SSH session that uses it with
ssh_channel_open_session().

Once you don't need the channel anymore, you can send an end-of-file
to it with ssh_channel_close(). At this point, you can destroy the channel
with ssh_channel_free().

The code sample below achieves these tasks:

@code
int shell_session(ssh_session session)
{
  ssh_channel channel;
  int rc;

  channel = ssh_channel_new(session);
  if (channel == NULL)
    return SSH_ERROR;

  rc = ssh_channel_open_session(channel);
  if (rc != SSH_OK)
  {
    ssh_channel_free(channel);
    return rc;
  }

  ...

  ssh_channel_close(channel);
  ssh_channel_send_eof(channel);
  ssh_channel_free(channel);

  return SSH_OK;
}
@endcode


@subsection interactive Interactive and non-interactive sessions

A "shell" is a command interpreter. It is said to be "interactive"
if there is a human user typing the commands, one after the
other. The contrary, a non-interactive shell, is similar to
the execution of commands in the background: there is no attached
terminal.

If you plan using an interactive shell, you need to create a
pseud-terminal on the remote side. A remote terminal is usually referred
to as a "pty", for "pseudo-teletype". The remote processes won't see the
difference with a real text-oriented terminal.

If needed, you request the pty with the function ssh_channel_request_pty().
Then you define its dimensions (number of rows and columns)
with ssh_channel_change_pty_size().

Be your session interactive or not, the next step is to request a
shell with ssh_channel_request_shell().

@code
int interactive_shell_session(ssh_channel channel)
{
  int rc;

  rc = ssh_channel_request_pty(channel);
  if (rc != SSH_OK) return rc;

  rc = ssh_channel_change_pty_size(channel, 80, 24);
  if (rc != SSH_OK) return rc;

  rc = ssh_channel_request_shell(channel);
  if (rc != SSH_OK) return rc;

  ...

  return rc;
}
@endcode


@subsection read_data Displaying the data sent by the remote computer

In your program, you will usually need to receive all the data "displayed"
into the remote pty. You will usually analyse, log, or display this data.

ssh_channel_read() and ssh_channel_read_nonblocking() are the simplest
way to read data from a channel. If you only need to read from a single
channel, they should be enough.

The example below shows how to wait for remote data using ssh_channel_read():

@code
int interactive_shell_session(ssh_channel channel)
{
  int rc;
  char buffer[256];
  int nbytes;

  rc = ssh_channel_request_pty(channel);
  if (rc != SSH_OK) return rc;

  rc = ssh_channel_change_pty_size(channel, 80, 24);
  if (rc != SSH_OK) return rc;

  rc = ssh_channel_request_shell(channel);
  if (rc != SSH_OK) return rc;

  while (ssh_channel_is_open(channel) &&
         !ssh_channel_is_eof(channel))
  {
    nbytes = ssh_channel_read(channel, buffer, sizeof(buffer), 0);
    if (nbytes < 0)
      return SSH_ERROR;

    if (nbytes > 0)
      write(1, buffer, nbytes);
  }

  return rc;
}
@endcode

Unlike ssh_channel_read(), ssh_channel_read_nonblocking() never waits for
remote data to be ready. It returns immediately.

If you plan to use ssh_channel_read_nonblocking() repeatedly in a loop,
you should use a "passive wait" function like usleep(3) in the same
loop. Otherwise, your program will consume all the CPU time, and your
computer might become unresponsive.


@subsection write_data Sending user input to the remote computer

User's input is sent to the remote site with ssh_channel_write().

The following example shows how to combine a nonblocking read from a SSH
channel with a nonblocking read from the keyboard. The local input is then
sent to the remote computer:

@code
/* Under Linux, this function determines whether a key has been pressed.
   Under Windows, it is a standard function, so you need not redefine it.
*/
int kbhit()
{
    struct timeval tv = { 0L, 0L };
    fd_set fds;

    FD_ZERO(&fds);
    FD_SET(0, &fds);

    return select(1, &fds, NULL, NULL, &tv);
}

/* A very simple terminal emulator:
   - print data received from the remote computer
   - send keyboard input to the remote computer
*/
int interactive_shell_session(ssh_channel channel)
{
  /* Session and terminal initialization skipped */
  ...

  char buffer[256];
  int nbytes, nwritten;

  while (ssh_channel_is_open(channel) &&
         !ssh_channel_is_eof(channel))
  {
    nbytes = ssh_channel_read_nonblocking(channel, buffer, sizeof(buffer), 0);
    if (nbytes < 0) return SSH_ERROR;
    if (nbytes > 0)
    {
      nwritten = write(1, buffer, nbytes);
      if (nwritten != nbytes) return SSH_ERROR;

    if (!kbhit())
    {
      usleep(50000L); // 0.05 second
      continue;
    }

    nbytes = read(0, buffer, sizeof(buffer));
    if (nbytes < 0) return SSH_ERROR;
    if (nbytes > 0)
    {
      nwritten = ssh_channel_write(channel, buffer, nbytes);
      if (nwritten != nbytes) return SSH_ERROR;
    }
  }

  return rc;
}
@endcode

Of course, this is a poor terminal emulator, since the echo from the keys
pressed should not be done locally, but should be done by the remote side.
Also, user's input should not be sent once "Enter" key is pressed, but 
immediately after each key is pressed. This can be accomplished
by setting the local terminal to "raw" mode with the cfmakeraw(3) function.
cfmakeraw() is a standard function under Linux, on other systems you can
recode it with:

@code
static void cfmakeraw(struct termios *termios_p)
{
    termios_p->c_iflag &= ~(IGNBRK|BRKINT|PARMRK|ISTRIP|INLCR|IGNCR|ICRNL|IXON);
    termios_p->c_oflag &= ~OPOST;
    termios_p->c_lflag &= ~(ECHO|ECHONL|ICANON|ISIG|IEXTEN);
    termios_p->c_cflag &= ~(CSIZE|PARENB);
    termios_p->c_cflag |= CS8;
}
@endcode

If you are not using a local terminal, but some kind of graphical
environment, the solution to this kind of "echo" problems will be different.


@subsection select_loop A more elaborate way to get the remote data

*** Warning: ssh_select() and ssh_channel_select() are not relevant anymore,
    since libssh is about to provide an easier system for asynchronous
    communications. This subsection should be removed then. ***

ssh_channel_read() and ssh_channel_read_nonblocking() functions are simple,
but they are not adapted when you expect data from more than one SSH channel,
or from other file descriptors. Last example showed how getting data from
the standard input (the keyboard) at the same time as data from the SSH
channel was complicated. The functions ssh_select() and ssh_channel_select()
provide a more elegant way to wait for data coming from many sources.

The functions ssh_select() and ssh_channel_select() remind of the standard
UNIX select(2) function. The idea is to wait for "something" to happen:
incoming data to be read, outcoming data to block, or an exception to
occur. Both these functions do a "passive wait", i.e. you can safely use
them repeatedly in a loop, it will not consume exaggerate processor time
and make your computer unresponsive. It is quite common to use these
functions in your application's main loop.

The difference between ssh_select() and ssh_channel_select() is that 
ssh_channel_select() is simpler, but allows you only to watch SSH channels.
ssh_select() is more complete and enables watching regular file descriptors
as well, in the same function call.

Below is an example of a function that waits both for remote SSH data to come,
as well as standard input from the keyboard:

@code
int interactive_shell_session(ssh_session session, ssh_channel channel)
{
  /* Session and terminal initialization skipped */
  ...

  char buffer[256];
  int nbytes, nwritten;

  while (ssh_channel_is_open(channel) &&
         !ssh_channel_is_eof(channel))
  {
    struct timeval timeout;
    ssh_channel in_channels[2], out_channels[2];
    fd_set fds;
    int maxfd;

    timeout.tv_sec = 30;
    timeout.tv_usec = 0;
    in_channels[0] = channel;
    in_channels[1] = NULL;
    FD_ZERO(&fds);
    FD_SET(0, &fds);
    FD_SET(ssh_get_fd(session), &fds);
    maxfd = ssh_get_fd(session) + 1;

    ssh_select(in_channels, out_channels, maxfd, &fds, &timeout);

    if (out_channels[0] != NULL)
    {
      nbytes = ssh_channel_read(channel, buffer, sizeof(buffer), 0);
      if (nbytes < 0) return SSH_ERROR;
      if (nbytes > 0)
      {
        nwritten = write(1, buffer, nbytes);
        if (nwritten != nbytes) return SSH_ERROR;
      }
    }

    if (FD_ISSET(0, &fds))
    {
      nbytes = read(0, buffer, sizeof(buffer));
      if (nbytes < 0) return SSH_ERROR;
      if (nbytes > 0)
      {
        nwritten = ssh_channel_write(channel, buffer, nbytes);
        if (nbytes != nwritten) return SSH_ERROR;
      }
    }
  }

  return rc;
}
@endcode


@subsection x11 Using graphical applications on the remote side

If your remote application is graphical, you can forward the X11 protocol to
your local computer.

To do that, you first declare that you accept X11 connections with
ssh_channel_accept_x11(). Then you create the forwarding tunnel for
the X11 protocol with ssh_channel_request_x11().

The following code performs channel initialization and shell session
opening, and handles a parallel X11 connection:

@code
int interactive_shell_session(ssh_channel channel)
{
  int rc;
  ssh_channel x11channel;

  rc = ssh_channel_request_pty(channel);
  if (rc != SSH_OK) return rc;

  rc = ssh_channel_change_pty_size(channel, 80, 24);
  if (rc != SSH_OK) return rc;

  rc = ssh_channel_request_x11(channel, 0, NULL, NULL, 0);
  if (rc != SSH_OK) return rc;

  rc = ssh_channel_request_shell(channel);
  if (rc != SSH_OK) return rc;

  /* Read the data sent by the remote computer here */
  ...
}
@endcode

Don't forget to set the $DISPLAY environment variable on the remote
side, or the remote applications won't try using the X11 tunnel:

@code
$ export DISPLAY=:0
$ xclock &
@endcode

*/