nfqueue 0.9.0

Netfilter NFQUEUE high-level bindings
Documentation
# nfqueue-bindings-rs

[![Build Status](https://travis-ci.org/chifflier/nfqueue-rs.svg?branch=master)](https://travis-ci.org/chifflier/nfqueue-rs)

nfqueue-rs is a wrapper library for
[libnetfilter-queue](http://www.netfilter.org/projects/libnetfilter_queue/). The goal is to provide a library to gain access to packets queued by the kernel packet filter.

It is important to note that these bindings will not follow blindly libnetfilter_queue API. For ex., some higher-level wrappers will be provided for the open/bind/create mechanism (using one function call instead of three).

**The API is not yet stable.**

To use the library, a program must
- open a queue
- bind to a network family (`AF_PACKET` for IPv4)
- provide a callback function, which will be automatically called when a packet is received. The callback must return a verdict
- create the queue, providing the queue number (which must match the `--queue-num` from the iptables rules, see below
- run a loop, waiting for events. The program should also provide a clean way to exit the loop (for ex on `SIGINT`)

## Example

See [nfq-example](src/bin/nfq-example.rs) for a minimal example.

## IPtables

You must add rules in netfilter to send packets to the userspace queue.
The number of the queue (--queue-num option in netfilter) must match the
number provided to create_queue().

Example of iptables rules:

    iptables -A OUTPUT --destination 1.2.3.4 -j NFQUEUE --queue-num 0

Of course, you should be more restrictive, depending on your needs.

## Privileges

nfqueue-rs does not require root privileges, but needs to open a netlink socket and send/receive packets to the kernel.

You have several options:
- Use the CAP_NET_ADMIN capability in order to allow your application to receive from and to send packets to kernel-space:
```setcap 'cap_net_admin=+ep' /path/to/program```
- Run your program as `root` and drop privileges