ipfixprobe is a high-performance flow exporter. It creates bidirectional flows from packet input and exports them to output interface. The ipfixprobe support vide variety of flow extenstion for application layer protocol information. The flow extension can be turned on with process plugins. We support TLS, QUIC, HTTP, DNS and many more. Just check our documentation.
The RPM packages for RHEL based distributions can be downloaded from our copr repository. Or just simply run:
dnf install -y dnf-plugins-core && dnf copr -y enable @CESNET/NEMEA
dnf install ipfixprobe
-i ARGS
Activate input plugin (-h input for help)-s ARGS
Activate storage plugin (-h storage for help)-o ARGS
Activate output plugin (-h output for help)-p ARGS
Activate processing plugin (-h process for help)-q SIZE
Size of queue between input and storage plugins-b SIZE
Size of input queue packet block-Q SIZE
Size of queue between storage and output plugins-B SIZE
Size of packet buffer-f NUM
Export max flows per second-c SIZE
Quit after number of packets are processed on each interface-P FILE
Create pid file-t PATH
Mount point of AppFs telemetry directory-d
Run as a standalone process-h [PLUGIN]
Print help text. Supported help for input, storage, output and process plugins-V
Show version and exit
Printing general help is done using the -h
parameter. To print help for specific plugins, -h
with parameter is used.
This parameter accepts input
, storage
, process
, output
or name of a plugin (or path to a .so file with plugin).
Here are the examples of various plugins usage:
# Capture from wlp2s0 interface using raw sockets, print flows to console
./ipfixprobe -i 'raw;ifc=wlp2s0' -o 'text'
# Capture from wlp2s0 interface and scale packet processing using 2 instances of plugins, send flow to ifpfix collector using UDP
./ipfixprobe -i 'raw;ifc=wlp2s0;f' -i 'raw;ifc=wlp2s0;f' -o 'ipfix;u;host=collector.example.com;port=4739'
# Capture from a COMBO card using ndp plugin, sends ipfix data to 127.0.0.1:4739 using TCP by default
./ipfixprobe -i 'ndp;dev=/dev/nfb0:0' -i 'ndp;dev=/dev/nfb0:1' -i 'ndp;dev=/dev/nfb0:2'
# Capture from eth0 interface using pcap plugin, split biflows into flows and prints them to console without mac addresses, telemetry data are exposed via the appFs library in /var/run/ipfixprobe directory
./ipfixprobe -i 'pcap;ifc=eth0' -s 'cache;split' -o 'text;m' -t /var/run/ipfixprobe
# Read packets from pcap file, enable 4 processing plugins, sends L7 HTTP extended biflows to unirec interface named `http` and data from 3 other plugins to the `stats` interface
./ipfixprobe -i 'pcap;file=pcaps/http.pcap' -p http -p pstats -p idpcontent -p phists -o 'unirec;i=u:http:timeout=WAIT,u:stats:timeout=WAIT;p=http,(pstats,phists,idpcontent)'
# Read packets using DPDK input interface and 1 DPDK queue, enable plugins for basic statistics, http and tls, output to IPFIX on a local machine
# DPDK EAL parameters are passed in `e, eal` parameters
# DPDK plugin configuration has to be specified in the first input interface.
# The following `dpdk` interfaces are given without parameters; their configuration is inherited from the first one.
# Example for the queue of 3 DPDK input plugins (q=3):
`./ipfixprobe -i "dpdk;p=0;q=3;e=-c 0x1 -a <[domain:]bus:devid.func>" -i dpdk -i dpdk -p http "-p" bstats -p tls -o "ipfix;h=127.0.0.1"`
# Same example for the multiport read from ports 0 and 1, note comma separated ports:
`./ipfixprobe -i "dpdk;p=0,1;q=3;e=-c 0x1 -a <[domain:]bus:devid.func>" -i dpdk -i dpdk -p http "-p" bstats -p tls -o "ipfix;h=127.0.0.1"`
# Read packets using DPDK input interface as secondary process with shared memory (DPDK rings) - in this case, 4 DPDK rings are used
`./ipfixprobe -i 'dpdk-ring;r=rx_ipfixprobe_0;e= --proc-type=secondary' -i 'dpdk-ring;r=rx_ipfixprobe_1' -i 'dpdk-ring;r=rx_ipfixprobe_2' -i 'dpdk-ring;r=rx_ipfixprobe_3' -o 'text'`
- libatomic
- kernel version at least 3.19 when using raw sockets input plugin enabled by default (disable with
--without-raw
parameter for./configure
) - libpcap when compiling with pcap plugin (
--with-pcap
parameter) - netcope-common COMBO cards when compiling with ndp plugin (
--with-ndp
parameter) - libunwind-devel when compiling with stack unwind on crash feature (
--with-unwind
parameter) - nemea when compiling with unirec output plugin (
--with-nemea
parameter) - cloned submodule with googletest framework to enabled optional tests (
--with-gtest
parameter)
To compile DPDK interfaces, make sure you have DPDK libraries (and development files) installed and set the PKG_CONFIG_PATH
environment variable if necessary. You can obtain the latest DPDK at http://core.dpdk.org/download/ Use --with-dpdk
parameter of the configure
script to enable it.
This project uses a standard process of:
git clone --recurse-submodules https://github.com/CESNET/ipfixprobe
cd ipfixprobe
autoreconf -i
./configure
make
sudo make install
Check ./configure --help
for more details and settings.
RPM package can be created in the following versions using --with
parameter of rpmbuild
:
--with pcap
enables RPM with pcap input plugin--with ndp
enables RPM with netcope-common, i.e., ndp input plugin--with nemea
enables RPM with unirec output plugin--without raw
disables RPM with default raw socket input plugin--with unwind
enables RPM with stack unwinding feature
These parameters affect required dependencies of the RPM and build process.
The default configuration of the RPM can be created using simply: make rpm
Alternative versions (described in the following section) can be created by:
- NEMEA version of RPM:
make rpm-nemea
- NDP version of RPM:
make rpm-ndp
We use COPR infrastructure to build and serve RPM packages for EPEL9. It is not possible to pass arguments to rpmbuild, so there is an option in configure to enforce NEMEA dependency:
./configure --enable-coprrpm && make srpm
The output source RPM can be uploaded to copr.
To install ipfixprobe with NEMEA dependency from binary RPM packages, it is possible to follow instructions on: https://copr.fedorainfracloud.org/coprs/g/CESNET/NEMEA/
ipfixprobe
exports statistics and other diagnostic information through a telemetry interface based on appFs library, which leverages the fuse3 library (filesystem in userspace) to allow telemetry data to be accessed and manipulated
through standard filesystem operations.
/var/run/ipfixprobe/
├── output
│ └── ipxRing
│ └── stats
└── pipeline
└── queues
├── 0
│ ├── cache-stats
│ ├── fragmentation-cache-stats
│ ├── input-stats
│ └── parser-stats
├── 1
...
From telemetry stats you can read:
Output Directory:
In the output directory, you can monitor the capacity and current usage of the ipxRing. This helps determine whether the output plugin can export flows quickly enough or if there are bottlenecks caused by insufficient ring capacity.
Example: ipxRing/stats
count: 204
size: 16536
usage: 1.23 (%)
Pipeline Directory:
The pipeline directory provides statistics for all worker queues. Each queue is identified by its ID (e.g., 0, 1) and includes the following files:
-
cache-stats: Provides detailed metrics about flow cache usage and exported flow statistics.
Example:
FlowCacheUsage: 3.81 (%) FlowEndReason:ActiveTimeout: 34666654 FlowEndReason:Collision: 4272143 FlowEndReason:EndOfFlow: 486129363 FlowEndReason:Forced: 58905 FlowEndReason:InactiveTimeout: 2169352600 FlowRecordStats:11-20packets: 178735501 FlowRecordStats:1packet: 1824500140 FlowRecordStats:2-5packets: 376268956 FlowRecordStats:21-50packets: 87971544 FlowRecordStats:51-plusPackets: 55424342 FlowRecordStats:6-10packets: 171579322 FlowsInCache: 39986 TotalExportedFlows: 2694479805
-
fragmentation-cache-stats: Provides metrics related to packet fragmentation.
Example:
firstFragments: 163634416 fragmentedPackets: 395736897 fragmentedTraffic: 0.13 (%) notFoundFragments: 85585913 totalPackets: 314829930486
-
input-stats: Provides metrics on the data received by by the queue.
Example:
received_bytes: 388582006601530 received_packets: 314788702409
-
parser-stats: Provides detailed information about the types of packets processed by the parser.
Example:
ipv4_packets: 193213761481 ipv6_packets: 121566104060 mpls_packets: 0 pppoe_packets: 0 seen_packets: 314791928764 tcp_packets: 301552123188 trill_packets: 0 udp_packets: 12783568334 unknown_packets: 11601117 vlan_packets: 31477986554
The availability of the input and output interfaces depends on the ipfixprobe build settings. By default, we provide RPM package with pcap and raw inputs. The default provided outpus are ipfix and text.
When the project is configured with ./configure --with-nemea
, the flow
exporter supports NEMEA output via TRAP IFC besides the default IPFIX output.
For more information about NEMEA, visit
https://nemea.liberouter.org.
The flow exporter supports compilation with libpcap (./configure --with-pcap
), which allows for receiving packets
from PCAP file or network interface card.
When the project is configured with ./configure --with-ndp
, it is prepared for high-speed packet transfer
from special HW acceleration FPGA cards. For more information about the cards,
visit COMBO cards or contact
us.
There are several currently available output plugins, such as:
ipfix
standard IPFIX RFC 5101unirec
data source for the NEMEA system, the output is in the UniRec format sent via a configurable interface using https://nemea.liberouter.org/trap-ifcspec/text
output in human readable text format on standard output file descriptor (stdout)
The output flow records are composed of information provided by the enabled plugins (using -p
parameter, see Flow Data Extension - Processing Plugins).
See ipfixprobe -h output
for more information and complete list of output plugins and their parameters.
LZ4 compression: ipfix plugin supports LZ4 compression algorithm over tcp. See plugin's help for more information.
Turn off message buffering using buffer=off
option and set timeout=WAIT
on output interfaces.
./ipfixprobe -i 'pcap;file=traffic.pcap' -o 'unirec;i=u:out:timeout=WAIT:buffer=off'