User login

Shane Alcock's Blog

09

Nov

2018

Libtrace 4.0.5 has been released.

This release is a collection of bug fixes and performance improvements for libtrace 4.0.4.

Major changes in this release:
* Fixed bug where only one client could subscribe to an nDAG multicast group concurrently.
* Fixed bad payload length calculations on outgoing packets when IP length field is populated by hardware.
* Fixed build error when compiling with DPDK support.
* Improved performance when decoding ETSI records via etsilive:
* Improved ERF / DAG packet processing performance by caching the framing length.

The full list of changes in this release can be found in the libtrace ChangeLog.

You can download the new version of libtrace from the libtrace website.

02

Jul

2018

Libtrace 4.0.4 has been released.

New features in this release include:
* Added reference counting API for keeping track of packets shared across multiple threads.
* Added new input format for receiving live streams of packets encoded using the ETSI LI standard.
* Libpacketdump can now also decode ETSI-encoded packets.
* Added CLI option to tracereplay to "speed up" replays by a given factor, i.e. set to 2 to replay at double speed.
* Added new trace_flush_output() API function to force a libtrace output to flush any buffered output.

There are also plenty of bug fixes in this release, including:
* Fixed problems with trying to have two ring: inputs active at the same time.
* Fixed missing subseconds in timestamps from pcapng: packets.
* Fixed bug that was causing poor ndag: performance.
* Fixed bug that prevented trace_pstop() from working correctly on some live inputs.
* tracereplay will now skip meta-data records rather than trying to replay them and failing.
* Fixed some buffer overrun issues in the pcapng reading code.

The full list of changes in this release can be found in the libtrace ChangeLog.

You can download the new version of libtrace from the libtrace website.

02

Mar

2018

Libtrace 4.0.3 has been released.

New features in this release include:
* A new capture format (dpdkndag:), which allows nDAG records to be intercepted and decoded using a DPDK interface.
* The message queue data structure API is now publicly exported, so can be used in user code.
* The toeplitz hasher data structure API is now publicly exported, so can be used in user code.
* Added a new API function: trace_get_perpkt_thread_id(), which will return the ID number of the running packet processing thread.
* Upgraded DAG code to use the 64 bit API, so libtrace will work with large streams.

The following bug fixes are also included:
* ERF provenance records will no longer cause libtrace to halt.
* Captures from GRE tunnel interfaces should now work correctly.
* Packets captured using DPDK will no longer lose any payload after the first 1024 bytes.
* Fixed a couple of nDAG packet corruption problems.
* Ensure all key fields are correctly initialised when doing DPDK output.
* Fixed assertion failure when libwandio has an unexpected error.

We've also further improved the performance of the nDAG format.

The full list of changes in this release can be found in the libtrace ChangeLog.

You can download the new version of libtrace from the libtrace website.

18

Dec

2017

Continued trying to get better performance out of the ndag protocol. Most of my time this week was actually spent trying to resolve some issues with the DPDK packet counters, which were not proving to be particularly accurate or useful. After a lot of debugging, I realised that my DPDK packet generator software was reading and clearing the stats for all of the DPDK interfaces, not just the one it was transmitting on. Once I blacklisted the interface I was using to receive reflected ndag packets on from the packet generator, my numbers started making sense again.

Eventually I found that the best way to improve performance was to just enable jumbo frames and therefore send less multicast packets. With jumbo frames, I've been able to capture, encapsulate, export and subsequently receive a sustained 10G with no real issues.

On the OpenLI side, I've improved the API around my decoding code so that I can easily perform the main tasks that libtrace and libpacketdump will require. Specifically, I now have API functions for getting the timestamp, record length and start of the IP capture field from an ETSI-encoded packet. I've also added a "get next field as a string" function for generating nice tracepktdump output. Hopefully with all of this in place, I should be able to integrate the decoder into libtrace by the end of the year.

08

Dec

2017

Separated common code that was shared between the telescope prototype and the new wdcapsniffer into its own file so that there is less repeated code to maintain. After a couple of extra bug fixes, I've managed to get my libprotoident daily monitor code working again and now using the ndag export from wdcapsniffer as the packet source. This will help me confirm that the code is generally stable and doesn't drop packets, as I should notice fairly quickly if my libprotoident reports are empty or have bogus data in them.

Added a dpdkndag capture format to libtrace which intercepts ndag multicast on the wire using dpdk, strips the IP, UDP and ndag headers and converts the contained ERF records into libtrace packets. The idea is that this would be faster than joining the multicast group and waiting for ndag packets to work their way through the network stack. This has turned out to be the case, although it is still not enough for a client to keep up with anything more than capturing ~6.5 Gbps.

Started my OpenLI work by developing my own ETSI-LI decoder. To support this, I've written a simple DER decoder which supports most of the primitives that are present in the ETSI standards. I've also written some code to model the ETSI structural hierarchy. I can now decode an example ETSI-encoded bytestream by walking both the hierarchy and the bytestream, seeing which fields are present and interpreting them according to the type defined in the matching hierarchy entry.

08

Dec

2017

Continued polishing up my presentation on the proposed OpenLI project. The meeting itself was held on Thursday -- people seemed pretty happy with my design and thinking thus far and the project is now all scheduled to start next week. Had a good chat with Neil from the Police about the ETSI standards and some of the gotchas that I'll need to think about when writing my code.

Finished up the initial libwdcap code and used it to write a wdcapsniffer program that exports via nDAG. Spent some time testing and tweaking the wdcapsniffer on the 10g dev machines before rolling the new and improved version out onto the probe VM.

28

Nov

2017

Started working towards rolling nDAG out onto our own capture environment, so that I can observe how it performs in a slightly more realistic scenario. Had to work around a few environment limitations, such as my VM disk needing to be resized and multicast being heavily rate limited on the path between the probe and my client. I've also started working on a libwdcap that will provide all of our old snapping and anonymisation capabilities, as I won't be able to export full unencrypted packets off the probe.

Started working on an actual design and development plan for the ETSI project, including plenty of architecture diagrams. Put together some slides describing the plan and proposed architecture for presentation at the first project meeting this coming Thursday.

20

Nov

2017

Implemented a few more nDAG optimisations. I'm now using recvmmsg and sendmmsg to receive and send batches of messages using a single system call -- this saves us quite a lot of pthread cancellation status changes when doing I/O. Sender can now comfortably handle 7.5 Gbps without dropping any records. Client is still stuck around 5 Gbps receiving, but I still have one or two ideas up my sleeve.

Tested and released a new version of libtrace, which includes both nDAG read and pcap-ng read support.

Started experimenting some more with the ETSI capture format. Started adding an etsilive: input format to libtrace for testing purposes. Played with cyberprobe as a possible independent source of ETSI-encoded packets to develop my decoder against, but it turned out that their encoding was not as compliant as I had hoped. I found two errors in their encoding of the first field alone.

15

Nov

2017

Libtrace 4.0.2 has been released today.

This release adds two new input formats: pcapng and nDAG.

This release also includes the following bug fixes:
* Fixed bad IPv6 fragment offset calculation that would randomly cause IPv6 packets to be incorrectly parsed.
* Fixed bug where SIGINT would cause parallel programs to assert fail.
* Fixed compilation issues caused by inconsistent BPF presence checks.
* Fixed errors in IPv6 and SCTP libpacketdump decoders.
* Fixed bug where a short cryptopan key would cause traceanon to assert fail.
* Fixed compilation error when building traceanon against new versions of libssl.
* Fixed bug where DPDK would not be detected if it had been built as a shared library.
* Fixed bug where the wrong parallel read function would be used by libtrace.

The full list of changes in this release can be found in the libtrace ChangeLog.

You can download the new version of libtrace from the libtrace website.

13

Nov

2017

Continued working on tweaking nDAG to both improve performance and add some handy features such as the ability for clients to recognise when an nDAG monitor has restarted and therefore may have missed some packets. Still got one or two ideas on how to improve performance further, so will try those out before merging the code back into mainline libtrace.

Started thinking a bit more about how my ETSI monitor is going to work and how much of it will intersect with libtrace. Will probably need to add an etsilive: read format to libtrace with suitable libpacketdump decoders to help with testing and validation, so that seems like a useful starting point.

Added a feature to my daily libprotoident analysis program to tell me what proportion of traffic on the campus network remains unidentified.