showcase [C] The DPDK Examples

[Christian]

A small repository I will be using to store my progress and test programs from the DPDK, a kernel bypass library very useful for fast packet processing. The DPDK is perhaps one of the fastest libraries used with network packet processing.

 

This repository uses my DPDK Common project in an effort to make things simpler.

 

WARNING - I am still adding more examples as time goes on and I need to test new functionality/methods.

 

Requirements

• The DPDK - Intel's Data Plane Development Kit which acts as a kernel bypass library which allows for fast network packet processing (one of the fastest libraries out there for packet processing).
• The DPDK Common - A project written by me aimed to make my DPDK projects simpler to setup/run.

 

Building The DPDK
If you want to build the DPDK using default options, the following should work assuming you have the requirements such as ninja and meson.

 

# Clone the DPDK repository.
git clone https://github.com/DPDK/dpdk.git

# Change directory.
cd dpdk/

# Use meson build.
meson build

# Change directory to build/.
cd build

# Run Ninja.
ninja

# Run Ninja install as root via sudo.
sudo ninja install

# Link libraries and such.
sudo ldconfig

All needed header files from the DPDK will be stored inside of /usr/local/include/.

 

You may receive ninja and meson using the following.

 

# Update via `apt`.
sudo apt update

# Install Python PIP (version 3).
sudo apt install python3 python3-pip

# Install meson. Pip3 is used because 'apt' has an outdated version of Meson usually.
sudo pip3 install meson

# Install Ninja.
sudo apt install ninja-build

Building The Source Files
You may use git and make to build the source files inside of this repository.

 

git clone --recursive https://github.com/gamemann/The-DPDK-Examples.git
cd The-DPDK-Examples/
make

Executables will be built inside of the build/ directory by default.

 

EAL Parameters
All DPDK applications in this repository supports DPDK's EAL paramters. These may be found here.

 

This is useful for specifying the amount of l-cores and ports to configure for example.

 

Examples
Drop UDP Port 8080 (Tested And Working)
In this DPDK application, any packets arriving on UDP destination port 8080 will be dropped. Otherwise, if the packet's ethernet header type is IPv4 or VLAN, it will swap the source/destination MAC and IP addresses along with the UDP source/destination ports then send the packet out the TX path (basically forwarding the packet from where it came).

 

In additional to EAL parameters, the following is available specifically for this application.

 

-p --portmask => The port mask to configure (e.g. 0xFFFF).
-P --portmap => The port map to configure (in '(x, y),(b,z)' format).
-q --queues => The amount of RX and TX queues to setup per port (default and recommended value is 1).
-x --promisc => Whether to enable promiscuous on all enabled ports.
-s --stats => If specified, will print real-time packet counter stats to stdout.

Here's an example.

 

./dropudp8080 -l 0-1 -n 1 -- -q 1 -p 0xff -s

Simple Layer 3 Forward (Tested And Working)
In this DPDK application, a simple routing hash table is created with the key being the destination IP address and the value being the MAC address to forward to.

 

Routes are read from the /etc/l3fwd/routes.txt file in the following format.

 

<ip address> <mac address in xx:xx:xx:xx:xx:xx>

The following is an example.

 

10.50.0.4 ae:21:14:4b:3a:6d
10.50.0.5 d6:45:f3:b1:a4:3d

When a packet is processed, we ensure it is an IPv4 or VLAN packet (we offset the packet data by four bytes in this case so we can process the rest of the packet without issues). Afterwards, we perform a lookup with the destination IP being the key on the route hash table. If the lookup is successful, the source MAC address is replaced with the destination MAC address (packets will be going out the same port they arrive since we create a TX buffer and queue) and the destination MAC address is replaced with the MAC address the IP was assigned to from the routes file mentioned above. Otherwise, the packet is dropped and the packet dropped counter is incremented.

 

In additional to EAL parameters, the following is available specifically for this application.

 

-p --portmask => The port mask to configure (e.g. 0xFFFF).
-P --portmap => The port map to configure (in '(x, y),(b,z)' format).
-q --queues => The amount of RX and TX queues to setup per port (default and recommended value is 1).
-x --promisc => Whether to enable promiscuous on all enabled ports.
-s --stats => If specified, will print real-time packet counter stats to stdout.

Here's an example.

 

./simple_l3fwd -l 0-1 -n 1 -- -q 1 -p 0xff -s

Rate Limit (Tested And Working)
In this application, if a source IP equals or exceeds the packets per second or bytes per second specified in the command line, the packets are dropped. Otherwise, the ethernet and IP addresses are swapped along with the TCP/UDP ports and the packet is forwarded back out the TX path.

 

Packet stats are also included with the -s flag.

 

The following command line options are supported.

 

-p --portmask => The port mask to configure (e.g. 0xFFFF).
-P --portmap => The port map to configure (in '(x, y),(b,z)' format).
-q --queues => The amount of RX and TX queues to setup per port (default and recommended value is 1).
-x --promisc => Whether to enable promiscuous on all enabled ports.
-s --stats => If specified, will print real-time packet counter stats to stdout.
--pps => The packets per second to limit each source IP to.
--bps => The bytes per second to limit each source IP to.

Here's an example:

 

./ratelimit -l 0-1 -n 1 -- -q 1 -p 0xff -s

NOTE - This application supports LRU recycling via a custom function I made in the DPDK Common project, check_and_del_lru_from_hash_table(). Make sure to define USE_HASH_TABLES before including the DPDK Common header file when using this function.

 

Least Recently Used Test (Tested And Working)
This is a small application that implements a manual LRU method for hash tables. For a while I've been trying to get LRU tables to work from these libraries. However, I had zero success in actually getting the table initialized.

 

Therefore, I decided to keep using these libraries instead and implement my own LRU functionality. I basically use the rte_hash_get_key_with_position() function to retrieve the oldest key to delete. However, it appears the new entry is inserted at the position that was most recently deleted so you have to keep incrementing the position value up to the max entries of the table. With that said, once the position value exceeds the maximum table entries, you need to set it back to 0.

 

No command line options are needed, but EAL parameters are still supported. Though, they won't make a difference.

 

Here's an example:

 

./ratelimit

 

Credits

• @Christian

 

GitHub Repository & Source Code