Skip to content

Latest commit

 

History

History
177 lines (125 loc) · 6.73 KB

README.md

File metadata and controls

177 lines (125 loc) · 6.73 KB
Raw

Advanced network examples in C for Linux

Collection of advanced Linux network examples in C to demonstrate non-obvious but useful features of the Linux/POSIX network API:

To try out the examples you need

  • gcc
  • make
  • netcat (commandline tool nc)

To build the examples just execute

$ make

NOTE: Currently there is only one example. Maybe there will be more in the future.

Example 1: Get local IP address of a TCP/UDP connection

When writing network servers it's often a good choice to listen for incoming connections and packages on all available interfaces. Binding the socket to only a specific interface or IP address is possible, but then the program logic has to deal with changes in the network configuration. Interfaces can go up and down or vanished completely and IP addresses can be removed and added. Using the kernel feature to listen on all available interfaces makes the code simpler and easier to read.

So you write your server code and bind the server socket to a specific port and to the unspecific IP address 'in6addr_any' to listen on all available interfaces. Of course the code sets the socket option 'IPV6_V6ONLY' to false to serve IPv6 and IPv4 (with IPv4-mapped IPv6 addresses) connections.

But maybe you have a new problem now. When the protocol that your server implements allows to link other resources on the your host, think of an HTTP URL and link, but does not allow relative links, e.g. because it's a HTTP redirect to another port and protocol on your host, the server has to know the local IP address or hostname.

Mostly our device or server does not have a fully qualified domain name and you don't want to force your users to configure one. Additionally you know that there is not single IP address of a host. It depends on the network interface (ethernet, wireless, ...) and may change over time (DHCP and IPv6 autoconfigure).

So the question remains: How to know the local IP address of your server that can be included into a response message to the client and this local IP address is reachable for the client?

Infact on the network protocol level (IP, TCP and UDP) answering this question is easy. A TCP and UDP connection consists of the tuple: (client IP address, client port, server IP address, server port). When the server's OS kernel has an open TCP connection or receives an UDP packet, the kernel knows the local server IP address that the client was connecting to! The information is available on the network protocol level. The open questions is: How to get this information in your server program?

In the C language the OS API exists. For a TCP socket you can use getsockname(), see man 2 getsockname. The file tcp-server.c contains example code. For UDP packets you can use the socket option IPV6_RECVPKTINFO, see man 7 ipv6. See example in file udp-server.c.

For the Java language it seems that there is no API for that. E.g. the stack overflow question

Stack Overflow: Java, DatagramPacket receive, how to determine local ip interface
https://stackoverflow.com/a/31966181

purposed a really bad solution for that problem. To get the local IP address, the code should open a new connection back to the client and then read out the local IP address via sock.getLocalAddress().

Pointed conclusion: Only the C system APIs are the real APIs. Everything else is just an incomplete wrapper.

Try out the TCP example

First choose a local listing port (here '8080') and start the TCP server in a shell:

$ ./tcp-server 8080
# is running

It starts and waits for incoming TCP connections. Now open a second terminal window and use the netcat utility to connect to the server. You should try different local IP addresses (IPv4 and IPv6):

$ nc localhost 8080
$ nc 127.0.0.1 8080
$ nc 192.168.178.26 8080
$ nc -6 ::1 8080
$ nc -6 2001:aaaa:bbbb:0:cccc:eeee:eeee:ffff 8080

In the terminal window, in which the server is running, you should see

$ ./tcp-server 8080
peer ::ffff:127.0.0.1 (47920)  ->  local ::ffff:127.0.0.1 ( 8080)
peer ::ffff:127.0.0.1 (47922)  ->  local ::ffff:127.0.0.1 ( 8080)
peer ::ffff:192.168.178.26 (45114)  ->  local ::ffff:192.168.178.26 ( 8080)
peer ::1 (47116)  ->  local ::1 ( 8080)
peer 2001:aaaa:bbbb:0:cccc:eeee:eeee:ffff (48490)  ->  local 2001:aaaa:bbbb:0:cccc:eeee:eeee:ffff ( 8080)

The server prints a line for every connection. A line contains the IP address and port of the connecting peer/client and then the local interface IP address and port of the server.

Now use a second machine on your network to connect to the same server.

other-host$ nc 192.168.178.26 8080

The server should print

$ ./tcp-server 8080
peer ::ffff:192.168.178.50 (42231)  ->  local ::ffff:192.168.178.26 ( 8080)

The address 192.168.178.50 should be the IP address of your other-host machine. The IP address 192.168.178.26 is the IP address of the local interface on your server machine.

Try out the UDP example

Similar to the TCP example above. Start the server program:

$ ./udp-server 8080
# is running

In a second terminal window start netcat, type test and then press the enter key:

$ nc -u ::1 8080
test<press enter>
$

In the shell running the server program you can see now:

$ ./udp-server 8080
peer ::1 (47029)  ->  local ::1 ( 8080)
Server 5 received bytes of data from the client.

The server program prints two lines for every connection. The first line contains the peer/remote/client IP address and port and then the local interface IP address and port. The second line is a diagnostic message showing the count of bytes in the received UDP packet of the client. Here the client has sent five bytes, four printable characters and a newline character.

After you tried an UDP connection on the localhost over the loopback interface, you should try netcat with different local IP addresses (IPv4 and IPv6) and from another machine on your network.

License

The C source code in this repository is unlicensed. See the text in file UNLICENSE for details. It means that you are free to do what every you want with the source code (e.g. also use it in proprietary software), without any obligations, but also without any warrenty of any kind. Again see the text in file UNLICENSE for the exact terms. The code is intended to be copied&pasted in your own applications.

Unlicense Yourself: Set Your Code Free

Further reading

Documentation and interesting articles about networking:

Title: ipv6 - Linux IPv6 protocol implementation
$ man 7 ipv6
http://man7.org/linux/man-pages/man7/ipv6.7.html

Title: How TCP Sockets Work
https://eklitzke.org/how-tcp-sockets-work

Title: How TCP backlog works in Linux
http://veithen.io/2014/01/01/how-tcp-backlog-works-in-linux.html