Multiplexing over C file descriptors. [just an exercise]

Each packet transmitted over a multiplexed file descriptor consists of a 4-byte length prefix, followed by one byte indicating the channel ID the data is destined for. So, for example, the byte sequence 01 23 45 67 for channel 13 is put on the wire as:

00 00 00 05 0d 01 23 45 67

The problem addressed by this library is buffering of currently unused channels, as well as concurrency. One might have different threads that want to process data from different channels (but the same file descriptor), and they have to make sure that a read operation from one of the threads does not result in data loss if the incoming data stems from the wrong channel.

Data is stored in dynamically resizing buffers (only previously "activated" channels are observed), and access is synchronized using a single mutex.

You can wrap any file descriptor in a multiplexer, using multiplex_new, before sending data to a given channel (0-255) using multiplex_send:

#include <multiplex.h>
...
Multiplex * m = multiplex_new(sockfd);
multiplex_send(m, channelId, buffer, length);
...

Before data can be received from a channel, it has to be enabled using multiplex_enable or multiplex_enable_range. Afterwards, multiplex_receive can wait for data destined for a given channel ID:

#include <multiplex.h>
...
Multiplex * m = multiplex_new(sockfd);
int r = 0;
char buffer[1024];

while (r != CHANNEL_CLOSED) {
    r = multiplex_receive(m, timeout, channelId, buffer, 0, 1024);
    if (r >= 0) { /* Process r bytes of Data */ }
}
...

The receive call returns either the number of bytes read, CHANNEL_CLOSED, CHANNEL_TIMEOUT or CHANNEL_IGNORED (if data was received but on a different channel).

© 2013 Yannick Scherer

This code is released under the MIT License.