gensio - General Stream I/O

This is gensio (pronounced gen'-see-oh), a framework for giving a consistent view of various stream (and packet) I/O types. You create a gensio object (or a gensio), and you can use that gensio without having to know too much about what is going on underneath. You can stack gensio on top of another one to add protocol functionality. For instance, you can create a TCP gensio, stack SSL on top of that, and stack Telnet on top of that. It supports a number of network I/O and serial ports. It also supports sound interfaces. gensios that stack on other gensios are called filters.

You can do the same thing with receiving ports. You can set up a gensio accepter to accept connections in a stack. So in our previous example, you can setup TCP to listen on a specific port and automatically stack SSL and Telnet on top when the connection comes in, and you are not informed until everything is ready.

gensio works on Linux, BSDs, MacOS, and Windows. On Windows, it gives you a single-threaded capable (but also multi-thread capable) event-driven interface (with blocking interfaces available) to simplify programming with lots of I/Os. It goes a long way to making writing portable I/O driven code easy.

A very important feature of gensio is that it makes establishing encrypted and authenticated connections much easier than without it. Beyond basic key management, it's really no harder than TCP or anything else. It offers extended flexibility for controlling the authentication process if needed. It's really easy to use.

Note that the gensio(5) man page has more details on individual gensio types.

For instructions on building this from source, see the "Building" section at the end.

gensio tools

A couple of tools are available that use gensios, both as an example and for trying things out. These are:

gensiot

A tool for making basic gensio connections. You can create any arbitrary gensio setup you like. See gensiot(1) for details.

gtlsshd

An sshd-like daemon that uses certauth, ssl, and SCTP or TCP gensios for making connections. It uses standard PAM authentication and uses ptys. See gtlsshd(8) for details.

There is a item in FAQ.rst named "How to run gtlsshd on Windows", see that and the Building on Windows section below for more details, as there are a few tricky things you have to handle.

gtlssh

An ssh-like program that can connect to gtlsshd. It can also be used with ser2net to make establishing encrypted and authenticated connections easier. See gtlssh(1) for details.

Available gensios

The following gensios are available in the library:

sctp

Normal SCTP communication. Streams and out of bound data are supported. End of message demarcation is not supported because it doesn't currently work on Linux.

tcp

Normal TCP communication. Out of bound data is supported.

unix

Unix stream domain socket.

unixdgram

Unix datagram domain socket. This is sort-of connection oriented.

unixseq

Unix seqpacket domain socket. Probably only works on Linux.

udp

Sort-of connection oriented UDP.

stdio

Access to either the calling program's stdio, or the ability to run a program and connect to its stdin, stdout, and stderr. NOTE: Do not use this for file I/O. Use the file gensio.

file

Used for accessing files. Allows both input and output file, and streams the data to/from the files. No accepter available.

pty

Run a program in a PTY and use the gensio to communicate with its tty. No accepter available.

serialdev

Connect to a device. You can use "sdev" as shorthand for "serialdev". It can hook to termios type devices, like ptys and /dev/tty, more than just serial ports. No accepter available.

dev

Connects to devices (like serialdev does) but does not do any serial port processing. It also has a write-only option for talking to printer ports or other write-only devices. It also has a rdonly option for talking to read-only devices. No accepter available.

ipmisol

Connect to a remote over IPMI SOL. Full serial port capabilities are available. No accepter available, unfortunately.

dummy

An accepter that doesn't do anything except look like an accepter to the user. Useful in some situations where an accepter is expected but you don't need to do anything.

echo

A gensio that echos everything that is sent to it. Useful for testing. No accepter available.

telnet

A filter gensio that implements the telnet protocol. It can do full serial support with RFC2217.

ssl

Implement SSL/TLS as a gensio filter. It supports client authentication, too.

certauth

A user authentication protocol implemented as a gensio filter.

mux

A channel multiplexer. You can create channels on top of it using open_channel(). Channels work as normal gensio, so you can have a number of gensios running on top of a single gensio. It also has end-of-message demarcation and obviously full flow-control capability individually on each channel. If you just need a gensio with end-of-message demarcation, you can use this as without creating channels.

msgdelim

Converts an unreliable stream interface into an unreliable packet interface. This is primarily so a reliable packet interface like relpkt can run over a serial port. It does not support streaming of data, so it's not very useful by itself.

relpkt

Converts an unreliable packet interface to a reliable packet interface (that also supports streaming). Made for running over msgdelim. It will run over UDP, but it's not ideal for that because it doesn't do all the internet-friendly flow control and such that SCTP and TCP do.

trace

A transparent gensio that allows the data read and/or written to be sent to a file, either as raw data or as human-readable hex data. It can also be used to block data flowing in one or both directions.

perf

A gensio that can send/receive data on top of a stack of gensios and measure the throughput on the channel. The received data from perf is information about the channel throughput.

conacc

A gensio accepter that takes a gensio stack string as a parameter. This lets you use a gensio as an accepter. When conacc is started, it opens the gensio, and when the gensio opens it reports a new child for the accepter. When the child closes it attempts to open the child again and go through the process again (unless accepts have been disabled on conacc).

Why would you want to use this? Say in ser2net you wanted to connect one serial port to another. You could have a connection like:

connection: &con0
  accepter: conacc,serialdev,/dev/ttyS1,115200
  connector: serialdev,/dev/ttyS2,115200

And it would connect /dev/ttyS1 to /dev/ttyS2. Without conacc, you could not use serialdev as an accepter. It would also let you use gtlsshd on a serial port if you wanted encrypted authenticated logins over a serial port. If you ran gtlsshd with the following:

gtlsshd --notcp --nosctp --oneshot --nodaemon --other_acc
   'conacc,relpkt(mode=server),msgdelim,/dev/ttyUSB1,115200n81'

You could connect with:

gtlssh --transport 'relpkt,msgdelim,/dev/ttyUSB2,115200n81' USB2

This creates a reliable packet transport over a serial port. The mode=server is required to make relpkt run as the server, since it would normally run as a client since it is not being started as an accepter. The ssl gensio (which runs over the transport) requires reliable communication, so it won't run directly over a serial port.

xlt

This gensio allows character translations to be done on data flowing through this filter. It's primarily to convert carraige returns and line feeds.

mdns

This gensio uses mDNS to lookup a service (protocol type, network type, port, address) and then connect to that service. If you have a program like ser2net that advertise mDNS service, you don't have to worry about finding port numbers and such, it's all handled for you.

keepopen

This gensio presents an always open connection to the upper layer and keeps the lower layer connection open. If it closes, it re-opens it.

script

This gensio executes an external program with the external program's stdio connected to the child of this gensio. Once the external program terminates, this gensio will report that it is open and pass all the data through. This can be used to run scripts to set things up on a connection before hooking to the parent gensio.

sound

A gensio that provides access to sound devices and files. It's a little complicated, read the docs in gensio.5

afskmdm

Yes, it looks like a jumble of letters.

A filter gensio that sits on top of the sound gensio and does an Audio Frequency Shift Keying modem, like is used on AX.25 amateur radio.

kiss

An amateur radio protocol for talking to TNCs. This is used by AX25 in many cases.

ax25

An amateur radio protocol for packet radio. To fully use this you would need to write code, since it uses channels and oob data for unnumbered information, but you can do basic things with just gensiot if all you need is one communication channel. For instance, if you wanted to chat with someone over the radio, and the kiss port is on 8001 on both machines, on the accepting machine you can run:

gensiot -i 'stdio(self)' -a \
    'ax25(laddr=AE5KM-1),kiss,conacc,tcp,localhost,8001'

which will hook to the TNC and wait for a connection on address AE5KM-1. Then you could run:

gensiot -i 'stdio(self)' \
    'ax25(laddr=AE5KM-2,addr="0,AE5KM-1,AE5KM-2"),kiss,tcp,localhost,8001'

on the other machine. This will connect to the other machine over TNC 0 with the given address. Then anything you type in one will appear on the other, a line at a time. Type "Ctrl-D" to exit. The 'stdio(self)' part turns off raw mode, so it's a line at a time and you get local echo. Otherwise every character you types would send a packet and you couldn't see what you were typing.

To hook to the N5COR-11 AX.25 BBS system, you would do:

gensiot -i 'xlt(nlcr),stdio(self)' \
  'ax25(laddr=AE5KM-2,addr="0,N5COR-11,AE5KM-2"),kiss,tcp,localhost,8001'

Most BBS systems use CR, not NL, for the new line, so the xlt gensio is used to translate incoming these characters.

Of course, this being gensio, you can put any workable gensio underneath ax25 that you would like. So if you want to play around or test without a radio, you could do ax25 over UDP multicast. Here's the accepter side:

gensiot -i 'stdio(self)' -a \
'ax25(laddr=AE5KM-1),conacc,'\
'udp(mcast="ipv4,224.0.0.20",laddr="ipv4,1234",nocon),'\
'ipv4,224.0.0.20,1234'

and here's the connector side:

gensiot -i 'stdio(self)' \
'ax25(laddr=AE5KM-2,addr="0,AE5KM-1,AE5KM-2"),'\
'udp(mcast="ipv4,224.0.0.20",laddr="ipv4,1234",nocon),'\
'ipv4,224.0.0.20,1234'

kiss is not required because UDP is already a packet-oriented media. Or you can use the greflector program to create a simulated radio situation. On the machine "radiopi2", run:

greflector kiss,tcp,1234

which will create a program that will reflect all received input to all other connections. Then on the accepter side:

gensiot -i 'stdio(self)' -a \
'ax25(laddr=AE5KM-1),kiss,conacc,tcp,radiopi2,1234'

and the connecting side:

gensiot -i 'stdio(self)' \
'ax25(laddr=AE5KM-2,addr="0,AE5KM-1,AE5KM-2"),kiss,tcp,radiopi2,1234'

The test code uses the reflector for some testing, since it's so convenient to use.

ratelimit

Limit the data throughput for a gensio stack.

cm108gpio

Allow a GPIO on a CMedia CM108 or equivalent sound device to be controlled. Used with afskmdm for keying a transmitter.

These are all documented in detail in gensio(5). Unless otherwise stated, these all are available as accepters or connecting gensios.

Creating Your Own Gensios

You can create your own gensios and register them with the library and stack them along with the other gensios.

The easiest way to do this is to steal code from a gensio that does kind of what you want, then modify it to create your own gensio. There is, unfortunately, no good documentation on how to do this.

The include file include/gensio/gensio_class.h has the interface between the main gensio library and the gensio. The gensio calls all come through a single function with numbers to identify the function being requested. You have to map all these to the actual operations. This is somewhat painful, but it makes forwards and backwards compatibility much easier.

Creating your own gensio this way is fairly complex. The state machine for something like this can be surprisingly complex. Cleanup is the hardest part. You have to make sure you are out of all callbacks and no timers might be called back in a race condition at shutdown. Only the simplest gensios (echo, dummy), strange gensios (conadd, keepopen, stdio), and gensios that have channels (mux, ax25) directly implement the interface. Everything else uses include/gensio/gensio_base.h. gensio_base provides the basic state machine for a gensio. It has a filter portion (which is optional) and a low-level (ll) portion, which is not.

The filter interface has data run through it for the processing. This is used for things like ssl, certauth, ratelimit, etc. Filter gensios would use this. These all use gensio_ll_gensio (for stacking a gensio on top of another gensio) for the ll.

Terminal gensios each have their own ll and generally no filter. For lls based on a file descriptor (fd), gensio_ll_fd is used. There is also an ll for IPMI serial-over-lan (ipmisol) and for sound. Most of the terminal gensios (tcp, udp, sctp, serial port, pty) use the fd ll, obviously.

Once you have a gensio, you can compile it as a module and stick it in $(moduleinstalldir)/<version>. Then the gensio will just pick it up and use it. You can also link it in with your application and do the init function from your application.

mDNS support

The mdns gensio has already been discussed, but the gensio library provides an easy to use mDNS interface. The include file for it is in gensio_mdns.h, and you can use the gensio_mdns(3) man page to get more information on it.

To make an mdns connection using gensiot, say you have ser2net set up with mdns enabled like:

connection: &my-port
  accepter: telnet(rfc2217),tcp,3001
  connector: serialdev,/dev/ttyUSB1,115200N81
  options:
    mdns: true

then you can connection to it with gensiot:

gensiot 'mdns,my-port'

gensiot will find the server, port, and whether telnet and rfc2217 are enabled and make the connection.

In addition, there is an gmdns tool that lets you do queries and advertising, and gtlssh can do mDNS queries to find services. If you have secure authenticated logins for ser2net, and you enable mdns on ser2net, like:

connection: &access-console
  accepter: telnet(rfc2217),mux,certauth(),ssl,tcp,3001
  connector: serialdev,/dev/ttyUSBaccess,115200N81
  options:
    mdns: true

it makes the setup very convenient, as you can just do:

gtlssh -m access-console

That's right, you can just directly use the connection name, no need to know the host, whether telnet or rfc2217 is enabled, or what the port is. You still have to set up the keys and such on the ser2net server, of course, per those instructions.

General Concepts

gensio has an object oriented interface that is event-driven. Synchronous interfaces are also available. You deal with two main objects in gensio: a gensio and a gensio accepter. A gensio provides a communication interface where you can connect, disconnect, write, receive, etc.

A gensio accepter lets you receive incoming connections. If a connection comes in, it gives you a gensio.

The interface is event-driven because it is, for the most part, completely non-blocking. If you open a gensio, you give it a callback that will be called when the connection is up, or the connection fails. Same for close. A write will return the number of bytes accepted, but it may not take all the bytes (or even any of the bytes) and the caller must account for that.

The open and close interfaces have a secondary blocking interface for convenience. These end in _s. This is for convenience, but it's not necessary and use of these must be careful because you can't really use them from callbacks.

Speaking of callbacks, data and information coming from gensio to the user is done with a function callback. Read data, and when the gensio is ready for write data comes back in a callback. A similar interface is used for calling from the user to the gensio layer, but it is hidden from the user. This sort of interface is easily extensible, new operations can be easily added without breaking old interfaces.

The library provides several ways to create a gensio or gensio accepter. The main way is str_to_gensio() and str_to_gensio_accepter(). These provide a way to specify a stack of gensios or accepters as a string and build. In general, you should use this interface if you can.

In general, interfaces that are not performance sensitive are string based. You will see this in gensio_control, and in auxiliary data in the read and write interface to control certain aspects of the write.

The library also provides ways to set up your gensios by individually creating each one. In some situations this might be necessary, but it limits the ability to use new features of the gensio library as it gets extended.

If a gensio supports multiple streams (like SCTP), stream numbers are passed in the auxdata with "stream=n". Streams are not individually flow controlled.

Channels, on the other hand, are separate flows of data over the same connection. Channels are represented as separate gensios, and they can be individually flow controlled.

Include Files

There are a few include files you might need to deal with when using gensios:

gensio.h

The main include files for gensios and gensio accepters.

sergensio.h

Serial port handling gensios and gensio accepters.

gensio_os_funcs.h

The definition for an OS handler.

argvutils.h

Many gensio functions take an argv array, this is utilities for dealing with argvs.

gensio_selector.h

A definition for a default OS handler.

These are for the most part documented in the man pages.

For creating your own gensios, the following include files are available for you:

gensio_class.h

The main include file for creating your own gensio.

sergensio_class.h

The main include file for creating your own serial port gensio.

gensio_base.h

This handles a lot of the boiler plate for a gensio. Most of the standard gensios use this. It splits the gensio function into an optional filter, and a lower layer interface called an ll.

gensio_ll_fd.h

An ll that provides most of the boilerplate for dealing with a file descriptor.

gensio_ll_gensio.h

An ll that provides all that is necessary for stacking a gensio on top of another gensio. The filter gensios (telnet, ssl, etc.) use this as the ll.

Each include file has lots of documentation about the individual calls and handlers.

Errors

gensio has it's own set of errors to abstract it from the OS errors (named GE_xxx) and provide more flexibility in error reporting. These are in the gensio_err.h include file (automatically included from gensio.h) and may be translated from numbers to a meaningful string with gensio_err_to_str(). Zero is defined to be not an error.

If an unrecognized operating system error occurs, GE_OSERR is returned and a log is reported through the OS handler log interface.

OS Handler

One slightly annoying thing about gensio is that it requires you to provide an OS handler (struct gensio_os_funcs) to handle OS-type functions like memory allocation, mutexes, the ability to handle file descriptors, timers and time, and a few other things.

The library does provide several OS handlers. You can call gensio_alloc_os_funcs() to allocate a default one for your system (POSIX or Windows). You can see that man page for more details. This will generally be the best performing option you have for your system.

For POSIX systems, OS handlers for glib and TCL are available, allocated with gensio_glib_funcs_alloc() and gensio_tcl_funcs_alloc(). These really don't work very well, especially from a performance point of view, the APIs for glib and TCL are not well designed for what gensio does. TCL can only support single-threaded operation. glib multithreaded operation only has one thread at a time waiting for I/O. But they do work, and the tests are run with them. These are not available on Windows because of poor abstractions on glib and because of lack of motivation on TCL.

But if you are using something else like X Windows, etc that has it's own event loop, you may need to adapt one for your needs. But the good thing is that you can do this, and integrate gensio with pretty much anything.

There is also a waiter interface that provides a convenient way to wait for things to occur while running the event loop. This is how you generally enter the event loop, because it provides a convenient way to signal when you are done and need to leave the loop.

Documentation for this is in:

include/gensio/gensio_os_funcs.h

Threads

The gensio library fully supports threads and is completely thread-safe. However, it uses signals on POSIX system, and COM on Windows systems, so some setup is required.

The "main" thread should call gensio_os_proc_setup() at startup and call gensio_os_proc_cleanup() when it is complete. This sets up signals and signal handlers, thread local storage on Windows, and other sorts of things.

You can spawn new threads from a thread that is already set up using gensio_os_new_thread(). This gives you a basic OS thread and is configured properly for gensio.

If you have a thread created by other means that you want to use in gensio, as long as the thread create another thread and doesn't do any blocking functions (any sort of wait, background processing, functions that end in _s like read_s, etc.) you don't have to set them up. That way, some external thread can write data, wake another thread, or do things like that.

If an external thread needs to do those things, it should call gensio_os_thread_setup().

Signals

As mentioned in the threads section, the gensio library on Unix uses signals for inter-thread wakeups. I looked hard, and there's really no other way to do this cleanly. But Windows has a couple of signal-like things, too, and these are available in gensio, also.

If you use gensio_alloc_os_funcs(), you will get an OS funcs using the passed in signal for IPC. You can pass in GENSIO_OS_FUNCS_DEFAULT_THREAD_SIGNAL for the signal if you want the default, which is SIGUSR1. The signal you use will be blocked and taken over by gensio, you can't use it.

gensio also provides some generic handling for a few signals. On Unix, it will handle SIGHUP through the gensio_os_proc_register_reload_handler() function.

On Windows and Unix you can use gensio_os_proce_register_term_handler(), which will handle termination requests (SIGINT, SIGTERM, SIGQUIT on Unix) and gensio_os_proc_register_winsize_handler() (SIGWINCH on Unix). How these come in through Windows is a little messier, but invisible to the user.

All the callbacks from from a waiting routine's wait, not from the signal handler. That should simplify your life a lot.

You can see the man pages for more details on all of these.

Creating a gensio

Connecting gensios

To create a gensio, the general way to do this is to call str_to_gensio() with a properly formatted string. The string is formatted like so:

<type>[([<option>[,<option[...]]])][,<type>...][,<end option>[,<end option>]]

The end option is for terminal gensios, or ones that are at the bottom of the stack. For instance, tcp,localhost,3001 will create a gensio that connects to port 3001 on localhost. For a serial port, an example is serialdev,/dev/ttyS0,9600N81 will create a connection to the serial port /dev/ttyS0.

This lets you stack gensio layers on top of gensio layers. For instance, to layer telnet on top of a TCP connection:

telnet,tcp,localhost,3001

Say you want to enable RFC2217 on your telnet connection. You can add an option to do that:

telnet(rfc2217=true),tcp,localhost,3001

When you create a gensio, you supply a callback with user data. When events happen on a gensio, the callback will be called so the user could handle it.

gensio accepters

A gensio accepter is similar to a connecting gensio, but with str_to_gensio_accepter() instead. The format is the same. For instance:

telnet(rfc2217=true),tcp,3001

will create a TCP accepter with telnet on top. For accepters, you generally do not need to specify the hostname if you want to bind to all interfaces on the local machine.

Using a gensio

Once you have created a gensio, it's not yet open or operational. To use it, you have to open it. To open it, do:

struct gensio *io;
int rv;

rv = str_to_gensio("tcp,localhost,3001", oshnd,
                   tcpcb, mydata, &io);
if (rv) { handle error }
rv = gensio_open(io, tcp_open_done, mydata);
if (rv) { handle error }

Note that when gensio_open() returns, the gensio is not open. You must wait until the callback (tcp_open_done() in this case) is called. After that, you can use it.

Once the gensio is open, you won't immediately get any data on it because receive is turned off. You must call gensio_set_read_callback_enable() to turn on and off whether the callback (tcpcb in this case) will be called when data is received.

When the read handler is called, the buffer and length is passed in. You do not have to handle all the data if you cannot. You must update the buflen with the number of bytes you actually handled. If you don't handle data, the data not handled will be buffered in the gensio for later. Not that if you don't handle all the data, you should turn off the read enable or the event will immediately called again.

If something goes wrong on a connection, the read handler is called with an error set. buf and buflen will be NULL in this case.

For writing, you can call gensio_write() to write data. You may use gensio_write() at any time on an open gensio. gensio_write() may not take all the data you write to it. The count parameter passes back the number of bytes actually taken in the write call.

You can design your code to call gensio_set_write_callback_enable() when you have data to send and the gensio will call the write ready callback and you can write from the callback. This is generally simpler, but enabling and disabling the write callback adds some overhead.

A more efficient approach is to write data whenever you need to and have the write callback disabled. If the write operation returns less than the full request, the other end has flow-controlled and you should enable the write callback and wait until it is called before sending more data.

In the callbacks, you can get the user data you passed in to the create call with gensio_get_user_data().

Note that if you open then immediately close a gensio, this is fine, even if the open callback hasn't been called. The open callback may or may not be called in that case, though, so it can be difficult to handle this properly.

Synchronous I/O

You can do basic synchronous I/O with gensios. This is useful in some situations where you need to read something inline. To do this, call:

err = gensio_set_sync(io);

The given gensio will cease to deliver read and write events. Other events are delivered. Then you can do:

err = gensio_read_s(io, &count, data, datalen, &timeout);
err = gensio_write_s(io, &count, data, datalen, &timeout);

Count is set to the actual number of bytes read/written. It may be NULL if you don't care (though that doesn't make much sense for read).

Timeout may be NULL, if so then wait for forever. If you set a timeout, it is updated to the amount of time left.

Note that signals will cause these to return immediately, but no error is reported.

Reads will block until some data comes in and returns that data. It does not wait until the buffer is full. timeout is a timeval, the read will wait that amount of time for the read to complete and return. A timeout is not an error, the count will just be set to zero.

Writes block until the whole buffer is written or a timeout occurs. Again, the timeout is not an error, the total bytes actually written is returned in count.

Once you are done doing synchronous I/O with a gensio, call:

err = gensio_clear_sync(io);

and delivery through the event interface will continue as before. You must not be in a synchronous read or write call when calling this, the results will be undefined.

Note that other I/O on other gensios will still occur when waiting for synchronous I/O

There is not currently a way to wait for multiple gensios with synchronous I/O. If you are doing that, you should really just use the event-driven I/O. It's more efficient, and you end up doing the same thing in the end, anyway.

Using a gensio accepter

Like a gensio, a gensio accepter is not operational when you create it. You must call gensio_acc_startup() to enable it:

struct gensio_accepter *acc;
int rv;

rv = str_to_gensio_accepter("tcp,3001", oshnd,
                            tcpacccb, mydata, &acc);
if (rv) { handle error }
rv = gensio_startup(acc);
if (rv) { handle error }

Note that there is no callback to the startup call to know when it's enabled, because there's no real need to know because you cannot write to it, it only does callbacks.

Even after you start up the accepter, it still won't do anything until you call gensio_acc_set_accept_callback_enable() to enable that callback.

When the callback is called, it gives you a gensio in the data parameter that is already open with read disabled. A gensio received from a gensio acceptor may have some limitations. For instance, you may not be able to close and then reopen it.

gensio accepters can do synchronous accepts using gensio_acc_set_sync() and gensio_acc_accept_s. See the man pages on those for details.

Logging

struct gensio_os_funcs has a vlog callback for handling internal gensio logs. These are called when something of significance happens but gensio has no way to report an error. It also may be called to make it easier to diagnose an issue when something goes wrong.

Serial I/O

The gensio and gensio accepter classes each have subclasses for handling serial I/O and setting all the parameters associated with a serial port.

You can discover if a gensio (or any of its children) is a serial port by calling gensio_to_sergensio(). If that returns NULL, it is not a sergensio and none of it's children are sergensios. If it returns non-NULL, it returns the sergensio object for you to use. Note that the gensio returned by sergensio_to_gensio() will be the one passed in to gensio_to_sergensio(), not necessarily the gensio that sergensio is directly associated with.

A sergensio may be a client, meaning that it can set serial settings, or it may be a server, meaning that it will receive serial settings from the other end of the connection.

Most sergensios are client only: serialdev (normal serial port), ipmisol, and stdio accepter. Currently only telnet has both client and server capabilities.

Python Interface

NOTE: The python interface described here is deprecated. Use the one in c++/swig/pygensio now.

You can access pretty much all of the gensio interface through python, though it's done a little differently than the C interface.

Since python is fully object oriented, gensios and gensio accepters are first-class objects, along with gensio_os_funcs, sergensios, and waiters.

Here's a small program:

import gensio

class Logger:
    def gensio_log(self, level, log):
        print("***%s log: %s" % (level, log))

class GHandler:
    def __init__(self, o, to_write):
        self.to_write = to_write
        self.waiter = gensio.waiter(o)
        self.readlen = len(to_write)

    def read_callback(self, io, err, buf, auxdata):
        if err:
            print("Got error: " + err)
            return 0
        print("Got data: " + buf);
        self.readlen -= len(buf)
        if self.readlen == 0:
            io.read_cb_enable(False)
            self.waiter.wake()
        return len(buf)

    def write_callback(self, io):
        print("Write ready!")
        if self.to_write:
            written = io.write(self.to_write, None)
            if (written >= len(self.to_write)):
                self.to_write = None
                io.write_cb_enable(False)
            else:
                self.to_write = self.to_write[written:]
        else:
            io.write_cb_enable(False)

    def open_done(self, io, err):
        if err:
            print("Open error: " + err);
            self.waiter.wake()
        else:
            print("Opened!")
            io.read_cb_enable(True)
            io.write_cb_enable(True)

    def wait(self):
        self.waiter.wait_timeout(1, 2000)

o = gensio.alloc_gensio_selector(Logger())
h = GHandler(o, "This is a test")
g = gensio.gensio(o, "telnet,tcp,localhost,2002", h)
g.open(h)

h.wait()

The interface is a pretty direct translation from the C interface. A python representation of the interface is in swig/python/gensiodoc.py, you can see that for documentation.

C++

The C++ interface is documented in c++/README.rst.

pygensio

The new pygensio interface is a cleaner implementation using swig directors instead of hand-coded callbacks into python. See the README.rst in c++/swig/pygensio. There are also glib and tcl OS_Funcs in the glib and tcl directories.

GO

The full C++ interface is available to Go programs through swig and swig directors. See c++/swig/go/README.rst for details.

Building

This is a normal autoconf system, nothing special. Note that if you get this directly from git, you won't have the build infrastructure included. There is a script named "reconf" in the main directory that will create it for you.

If you don't know about autoconf, the INSTALL file has some info, or google it.

To fully build gensio, you need the following:

• swig - For python and go bindings
• python dev libraries - For python bindings
• go language installed in the path
• openssl dev libraries and executable - For all the crypto
• openipmi dev libraries - For IPMI serial over lan, if you want that. Note that you need a pretty recent one, really 2.0.31 or newer.
• libsctp dev library - For sctp support
• pkgconfig - If you want gensio to install its pkgconfig files.
• avahi dev - If you want gensio to have mdns support.
• pam dev - For support of logins with gtlsshd
• libwrap - for tcpd
• glib dev - for the glib os funcs
• tcl dev - for the tcl os funcs
• alsa dev - for sound (on Linux)
• udev dev - for cm108 GPIO soundcard support (on Linux)

The following sets everything except openipmi up on ubuntu 20.04:

sudo apt install gcc g++ git swig python3-dev libssl-dev pkg-config

libavahi-client-dev avahi-daemon libtool autoconf automake make libsctp-dev libpam-dev libwrap0-dev libglib2.0-dev tcl-dev libasound2-dev libudev-dev

On Redhat, libwrap is gone, so you won't be using that, and swig doesn't appear to be available, so you will have to build that yourself with at least go and python support. Here's the command for Redhat-like systems:

sudo yum install gcc gcc-c++ git python3-devel swig openssl-devel

pkg-config avahi-devel libtool autoconf automake make lksctp-tools-devel pam-devel glib2-devel tcl-devel alsa-lib-devel systemd-devel

You might have to do the following to enable access to the development packages:

sudo dnf config-manager --set-enabled devel

And get the SCTP kernel modules, you might have to do:

sudo yum install kernel-modules-extra

To use Go language, you must get a version of swig 4.1.0 or greater. You may have to pull a bleeding edge version out of git and use that.

Handling python installation configuration is a bit of a pain. By default the build scripts will put it wherever the python program expects installed python programs to be. A normal user generally doesn't have write access to that directory.

To override this, you can use the --with-pythoninstall and --with-pythoninstalllib configure options or you can set the pythoninstalldir and pythoninstalllibdir environment variables to where you want the libraries and modules to go.

Note that you may need to set --with-uucp-locking to your lockdir (on older systems it's /var/lock, which is the default. On newer it might be /run/lock/lockdev. You might also need to be a member of dialout and lock groups to be able to open serial devices and/or locks.

go language support requires go to be installed and in the path.

Dynamic vs Built In gensios

As I continued to add gensios to the library, like crypto, mdns, sound, IPMI, sctp, etc. the number of dependencies in the library was getting out of control. Why should you be loading libasound, or libOpenIPMI, if you don't need it? Plus, though the library supported adding your own gensios through a programmatic API, it had no standard way to add them for the system so you could write your own gensio and let everyone on the system use it.

The gensio library supports loading gensios dynamically or building them in to the library. By default if you create shared libraries, then all gensios are compiled as modules for dynamic loading and installed in a place that makes it possible. If you do not create shared libraries, all gensios are built in to the library. But you can override this behaviour.

To set all gensios to be built in to the library, you can add "--with-all-gensios=yes" on the configure command line and it will build them in to the library.

You can also set them to all be dynamically loaded by adding "--with-all-gensios=dynamic", but this is the default.

You can also disable all gensios by default by specifying "--with-all-gensios=no". Then no gensios will be built by default. This is useful if you only want a few gensios, you can turn all of them off then enable then ones you want.

To set how individual gensios are built, you do "--with-<gensio>=x" where x is "no (don't build), yes (build into library) or dynamic (dynamically loaded executable). For instance, if you only wanted to build the tcp gensio into the library and make the rest dynamic, you could set up for all dynamic gensios and then add "--with-net=yes".

These modules are put by default into $(moduleinstalldir) (specified with --with-moduleinstall on the configure line) which defaults to $(pkglibexecdir) (which is generally /usr/libexec/gensio).

Note that dynamic loading is always available, even if you build in all the gensios in the library. So you can still add your own gensios by adding then to the proper directory.

Gensios will be loaded first from the environment variable LD_LIBRARY_PATH, then from GENSIO_LIBRARY_PATH, then from the default location.

Building on MacOS

MacOS, being a sort of *nix, builds pretty cleanly with Homebrew (https://brew.sh). You have to, of course, install all the libraries you need. Most everything works, with the following exceptions:

* cm108gpio
* sctp
* uucp locking

The built-in DNSSD code is used for MDNS, so avahi is not required.

flock locking for serial ports works, so uucp locking really isn't required.

openipmi should work, but it is not available in homebrew so you would have to build it yourself.

Building on FreeBSD

Install the necessary software:

pkg install gcc portaudio autoconf automake libtool mDNSResponder swig

go python3 gmake

You have to use gmake to compile it, for some reason the standard make on BSD doesn't accept the "c++" variable in a list of requirements. The following don't work and are not compiled:

* sctp
* ipmisol
* cm108gpio

Add the following to /etc/rc.conf:

mdnsd_enable=YES

And reboot or start the service.

The pty gensio fails the oomtest (oomtest 14), there seems to be something up with the BSD PTYs. I'm seeing a 07 character inserted into the data stream in cases. I haven't spent too much time on it, though, but since this is heavily tested on Linux and MacOS, I don't think the problem is in the gensio code.

Building on Windows

The gensio library can be built under Windows using mingw64. The following things don't work:

* sctp
* pam
* libwrap
* ipmisol

You also don't need to install alsa, it uses the Windows sound interface for sound.

The cm108gpio uses native windows interfaces, so udev is not required.

The Windows built-in MDNS interfaces are used, so you don't need avahi or DNSSD. You will need to install the pcre library if you want regular expressions in it.

You need to get msys2 from https://msys2.org. Then install autoconf, automake, libtool, git, make, and swig as host tools:

pacman -S autoconf automake libtool git make swig

You have to install the mingw-w64-x86_64-xxx version of all the libraries or the mingw-w64-i686-xxx version of all the libraries. 32-bit is not well tested:

pacman -S mingw-w64-x86_64-gcc \
  mingw-w64-x86_64-python3 \
  mingw-w64-x86_64-pcre \
  mingw-w64-x86_64-openssl

for mingw64, or for ucrt64:

pacman -S mingw-w64-ucrt-x86_64-gcc \
  mingw-w64-ucrt-x86_64-python3 \
  mingw-w64-ucrt-x86_64-pcre \
  mingw-w64-ucrt-x86_64-openssl

For go, install go from https://go.dev and log out and log back in. It should then be in the PATH, but if it's not, you will need to add it to the PATH. I haven't gotten go working on on mingw32, but I haven't tried a 32-bit version of go.

For gtlsshd, --sysconfdir has no meaning on Windows. Instead, the sysconf dir is relative to the patch of the executable, in ../etc/gtlssh. So if gtlsshd is in:

C:/Program Files/Gensio/bin/gtlsshd

the sysconfdir will be:

C:/Program Files/Gensio/etc/gtlssh

For standard installation, you can run:

../configure --sbindir=/Gensio/bin --libexecdir=/Gensio/bin \
   --mandir=/Gensio/man --includedir=/Gensio/include \
   --with-pythoninstall=/Gensio/python3 --prefix=/Gensio

and when you run "make install DESTDIR=..." and you set DESTDIR to where you want it to go, like "C:/Program Files". Then you can add that to the PATH using the control panel. To use gtlsshd, you create an etc/gtlsshd directory in the Gensio directory. You must set the permissions on this directory so only System and Administrators have access, like:

PS C:\Program Files (x86)\Gensio\etc> icacls gtlssh
gtlssh NT AUTHORITY\SYSTEM:(OI)(CI)(F)
       BUILTIN\Administrators:(OI)(CI)(F)

Otherwise gtlsshd will fail with an error about permissions on the key. You can set these permission on the .key file instead of the directory, but you will have to set it again every time you generate a new key.

For using the Inno Setup Compiler, do "make install DESTDIR=$HOME/install" and then run Inno on gensio.iss. It will create an executable installer for installing Gensio.

Then you need to remove the .la files from the install directory, as they screw up linking with other things:

rm $HOME/install/Gensio/lib/*.la

Running Tests

There are a number of tests for gensios. They all run on Linux if you have the serialsim kernel module. Besides the serial port ones, they run on other platforms as the gensios are supported on that platform.

The serial port tests require the serialsim kernel module and python interface. These are at https://github.com/cminyard/serialsim and allow the tests to use a simulated serial port to read modem control line, inject errors, etc.

You can get by without serialsim if you have three serial devices: one hooked in echo mode (RX and TX tied together) and two serial devices hooked together do I/O on one device goes to/comes from the other. This should work on non-Linux platforms. Then set the following environment variables:

export GENSIO_TEST_PIPE_DEVS="/dev/ttyxxx:/dev/ttywww"
export GENSIO_TEST_ECHO_DEV="/dev/ttyzzz"

It will not be able to test modemstate or rs485.

They also require the ipmi_sim program from the OpenIPMI library at https://github.com/cminyard/openipmi to run the ipmisol tests.

To run the tests, you need to enable some internal debugging to get the full effect. You generally want to run something like:

./configure --enable-internal-trace CFLAGS='-g -Wall'

You can turn on -O3 in the CFLAGS, too, if you like, but it makes debugging harder.

There are two basic types of tests. The python tests are functional tests testing both the python interface and the gensio library. Currently they are ok, but there is plenty of room for improvement. If you want to help, you can write tests.

The oomtest used to be an out of memory tester, but has morphed into something more extensive. It spawns a gensiot program with specific environment variables to cause it to fail at certain points, and to do memory leak and other memory checks. It writes data to the gensiot through its stdin and receives data on stdout. Some tests (like serialdev) use an echo. Other tests make a separate connection over the network and data flows both into stdin and comes back over the separate connection, and flows into the separate connection and comes back via stdout. oomtest is multi-threaded and the number of threads can be controlled. oomtest has found a lot of bugs. It has a lot of knobs, but you have to look at the source code for the options. It needs to be documented, if someone would like to volunteer...

Fuzzing

To set up for fuzzing, install afl, then configure with the following:

mkdir Zfuzz; cd Zfuzz
../configure --enable-internal-trace=yes --disable-shared --with-go=no \
    CC=afl-gcc CXX=afl-g++

Or use clang, if available:

../configure --enable-internal-trace=yes --disable-shared --with-go=no \
    CC=afl-clang-fast CXX=afl-clang-fast++ LIBS='-lstdc++'

I'm not sure why the LIBS thing is necessary above, but I had to add it to get it to compile.

Then build. Then "cd tests" and run "make test_fuzz_xxx" where xxx is one of: certauth, mux, ssl, telnet, or relpkt. You will probably need to adjust some things, afl will tell you. Note that it will run forever, you will need to ^C it when you are done.

The makefile in tests/Makefile.am has instructions on how to handle a failure to reproduce for debugging.

Code Coverage

Running code coverage on the library is pretty easy. First you need to configure the code to enable coverage:

mkdir Ocov; cd Ocov
../configure --enable-internal-trace=yes \
    CC='gcc -fprofile-arcs -ftest-coverage' \
    CXX='g++ -fprofile-arcs -ftest-coverage'

The compile and run "make check".

To generate the report, run:

gcovr -f '.*/.libs/.*' -e '.*python.*'

This will generate a summary. If you want to see the coverage of individual lines in a file, you can do:

cd lib
gcov -o .libs/ *.o

You can look in the individual .gcov files created for information about what is covered. See the gcov docs for detail.

At the time of writing, I was getting about 74% code coverage, So that's really pretty good. I'll be working to improve that, mostly through improved functional testing.

ser2net is used for testing some things, primarily the serial port configuration (termios and rfc2217). You can build ser2net against the gcov version of the gensio library and run "make check" in ser2net to get coverage on those parts. With that, I'm seeing about 76% coverage, so it doesn't add much to the total.

It would be nice to be able to combine this with fuzzing, but I'm not sure how to do that. afl does it's own thing with code coverage. There appears to be a afl-cov package that somehow integrated gcov, but I haven't looked into it.