Reverse engineering sounds like black magic to most people, even developers. But what if it's the only way? I tried it myself and I would like to share with you how this process looked like.
While ago I decided to buy myself a synthesizer. I chose Arturia Microbrute. Product itself is pretty great, but there was one problem. Arturia decided to release its software only for Windows and OS X, so as a Linux user I could not use my Microbrute to the fullest. I had no other choice than to write an application myself, but in order to do that, I had to figure out how Arturia’s software and Microbrute communicate
First of all, I needed a Windows machine (or a Windows running on a VirtualBox) with USBPcap installed and an Arturia’s Microbrute application. To analyze captured packets I used, obviously, Wireshark. I knew that it communicates through USB, so I had to reverse engineer the protocol… and that’s where the fun begun.
I set up a capture process on a Windows machine as described on USBPcap’s website. Step 2 contains a really useful tip I omitted at first: after identifying the root hub you will sniff on, disconnect the device, start sniffing and then connect the device. This will capture the USB enumeration (more on USB enumeration can be found in chapter 4 here) packets that are used for proper packet dissection in Wireshark. So I started sniffing, connected the device, launched the application and changed some settings.
Getting hands dirty
A quick look at the captured packets in Wireshark and I discovered that Microbrute uses MIDI SysEx protocol. It’s nothing special for a synthesizer, after all. Knowing that, I could take a closer look. I used
sysex filter to analyze captured packets and show only important data. First column represents the source, second the host and last one is the SysEx data.
$ tshark -r file_with_captured_packets.pcapng -2 -R 'sysex' -T fields \ -e usb.src -e usb.dst -e usbaudio.sysex.reassembled.data host 2.5.2 f0:7e:7f:06:01:f7 2.5.5 host f0:7e:01:06:02:00:20:6b:04:00:02:01:01:00:03:02:f7 host 2.5.2 f0:00:20:6b:05:01:00:00:06:f7 2.5.5 host f0:00:20:6b:05:01:00:01:05:00:00:00:00:00:00:00:00:00:f7 host 2.5.2 f0:00:20:6b:05:01:01:00:08:f7 2.5.5 host f0:00:20:6b:05:01:01:01:07:00:00:00:00:00:00:00:00:00:f7 host 2.5.2 f0:00:20:6b:05:01:02:00:35:f7 2.5.5 host f0:00:20:6b:05:01:02:01:34:01:00:00:00:00:00:00:00:01:f7 host 2.5.2 f0:00:20:6b:05:01:03:00:10:f7 2.5.5 host f0:00:20:6b:05:01:03:01:0f:01:00:00:00:00:00:00:00:01:f7 host 2.5.2 f0:00:20:6b:05:01:04:00:2f:f7 2.5.5 host f0:00:20:6b:05:01:04:01:2e:00:00:00:00:00:00:00:00:00:f7 host 2.5.2 f0:00:20:6b:05:01:05:00:0c:f7 2.5.5 host f0:00:20:6b:05:01:05:01:0b:00:00:00:00:00:00:00:00:00:f7 host 2.5.2 f0:00:20:6b:05:01:06:00:0e:f7 2.5.5 host f0:00:20:6b:05:01:06:01:0d:00:00:00:00:00:00:00:00:00:f7 host 2.5.2 f0:00:20:6b:05:01:07:00:12:f7 2.5.5 host f0:00:20:6b:05:01:07:01:11:02:01:00:00:00:00:00:00:00:f7 host 2.5.2 f0:00:20:6b:05:01:08:00:33:f7 2.5.5 host f0:00:20:6b:05:01:08:01:32:00:00:00:00:00:00:00:00:00:f7 host 2.5.2 f0:00:20:6b:05:01:09:00:2d:f7 2.5.5 host f0:00:20:6b:05:01:09:01:2c:02:01:00:00:00:00:00:00:00:f7 host 2.5.2 f0:00:20:6b:05:01:0a:00:39:f7 2.5.5 host f0:00:20:6b:05:01:0a:01:38:04:02:00:00:00:00:00:00:00:f7 host 2.5.2 f0:00:20:6b:05:01:0b:00:37:f7 2.5.5 host f0:00:20:6b:05:01:0b:01:36:03:01:00:00:00:00:00:00:01:f7 host 2.5.2 f0:00:20:6b:05:01:0c:00:2b:f7 2.5.5 host f0:00:20:6b:05:01:0c:01:2a:01:00:00:00:00:00:00:00:01:f7 host 2.5.2 f0:00:20:6b:05:01:0d:00:3d:f7 2.5.5 host f0:00:20:6b:05:01:0d:01:3c:00:00:00:00:00:00:00:00:00:f7 host 2.5.2 f0:00:20:6b:05:01:0e:01:0b:01:f7 host 2.5.2 f0:00:20:6b:05:01:0f:01:0b:02:f7 host 2.5.2 f0:00:20:6b:05:01:10:01:0b:00:f7 host 2.5.2 f0:00:20:6b:05:01:11:01:11:01:f7 [...some more similar packets here...] host 2.5.2 f0:00:20:6b:05:01:57:01:0d:01:f7 host 2.5.2 f0:00:20:6b:05:01:58:01:0d:00:f7 host 2.5.2 f0:00:20:6b:05:01:59:01:0f:00:f7 host 2.5.2 f0:00:20:6b:05:01:5a:01:0f:01:f7
I noticed that the packets can be divided into two groups. In the first one, for every outgoing packet there is one packet sent back by Microbrute. The second group only contains outgoing packets. Doing a live capture, I found out right away that the first group appears just after running the application (so this is possibly reading the current state of the device), while the packets from the second group appear after changing any of the device’s parameter (so they are used to alter the state of the device). At first, I focused on the second group, as it seemed easier to reverse engineer.
Writing to Microbrute
I started a live capture, applied a
sysex filter in Wireshark and observed what happened. I captured a couple of packets – the first group. Then, when I started changing values in the application, I noticed that one change of a parameter corresponds to one packet. I added a comment to each packet to specify what parameter was changed and to what value. Then, I inspected the captured packets using
thsark (the last column shows the comment added during live capture session):
$ tshark -r microbrute2comments.pcapng -2 -R 'sysex' -T fields \ -e usb.src -e usb.dst -e usbaudio.sysex.reassembled.data -e frame.comment [first group of packets here] host 2.5.2 f0:00:20:6b:05:01:0e:01:0b:01:f7 Set note priority to "low" host 2.5.2 f0:00:20:6b:05:01:0f:01:0b:02:f7 Set note priority to "high host 2.5.2 f0:00:20:6b:05:01:10:01:0b:00:f7 Set note priority to "last" host 2.5.2 f0:00:20:6b:05:01:11:01:34:02:f7 Set seq retrig to "none" host 2.5.2 f0:00:20:6b:05:01:12:01:34:00:f7 Set seq retrig to "reset" host 2.5.2 f0:00:20:6b:05:01:13:01:34:01:f7 Set seq retrig to "legato"
And this was something I could work on! I dissected an example packet: I knew it was MIDI SysEx and this packets was expected to start with
f0 and end with
f7. I knew that SysEx message should contain manufacturer ID. Those numbers can be found here. Arturia has been assigned the number
0x00 0x20 0x6b and in fact: those are the three bytes following
f0. The next byte,
0x05, is a number assigned to a specific device (in our case Microbrute). When I examined the sixth and eighth bytes, I noticed that they are always set to
0x01. The seventh byte is incremented by one with each consecutive SysEx command sent to Microbrute, so it serves as a counter.
The ninth and tenth bytes are the crucial ones. The ninth byte is always set to the same value if the value of the same parameter is changed (e.g.
0x0b for Note priority or
0x34 for Seq retrig), so it has to be the code which defines the parameter. The tenth byte then must hold the value of a changed parameter. For example: for note priority we can easily assign
0x00 to “Last”,
0x01 to “Low” and
0x02 to “High”. The SysEx message sent to Microbrute has the following format:
XX is the current value of the counter,
YY is the byte defining the parameter and
ZZ is the byte carrying the value. Then, I had to patiently click through all of the options to be able to assign proper values to all parameters and values.
Reading from Microbrute
After I had sorted out how to change Microbrute settings, I had to focus on reverse engineering the first part of the packets. This would allow me to read the current Microbrute’s state. Take a look at the first part of the sniffed packets:
$ tshark -r microbrute -2 -R 'sysex' -T fields \ -e usb.src -e usb.dst -e usbaudio.sysex.reassembled.data host 2.4.2 f0:7e:7f:06:01:f7 2.4.5 host f0:7e:01:06:02:00:20:6b:04:00:02:01:01:00:03:02:f7 host 2.4.2 f0:00:20:6b:05:01:00:00:06:f7 2.4.5 host f0:00:20:6b:05:01:00:01:05:00:00:00:00:00:00:00:00:00:f7 host 2.4.2 f0:00:20:6b:05:01:01:00:08:f7 2.4.5 host f0:00:20:6b:05:01:01:01:07:00:00:00:00:00:00:00:00:00:f7 host 2.4.2 f0:00:20:6b:05:01:02:00:35:f7 2.4.5 host f0:00:20:6b:05:01:02:01:34:01:00:00:00:00:00:00:00:01:f7 host 2.4.2 f0:00:20:6b:05:01:03:00:10:f7 2.4.5 host f0:00:20:6b:05:01:03:01:0f:01:00:00:00:00:00:00:00:01:f7 host 2.4.2 f0:00:20:6b:05:01:04:00:2f:f7 2.4.5 host f0:00:20:6b:05:01:04:01:2e:00:00:00:00:00:00:00:00:00:f7 host 2.4.2 f0:00:20:6b:05:01:05:00:0c:f7 2.4.5 host f0:00:20:6b:05:01:05:01:0b:00:00:00:00:00:00:00:00:00:f7 host 2.4.2 f0:00:20:6b:05:01:06:00:0e:f7 2.4.5 host f0:00:20:6b:05:01:06:01:0d:00:00:00:00:00:00:00:00:00:f7 host 2.4.2 f0:00:20:6b:05:01:07:00:12:f7 2.4.5 host f0:00:20:6b:05:01:07:01:11:02:01:00:00:00:00:00:00:00:f7 host 2.4.2 f0:00:20:6b:05:01:08:00:33:f7 2.4.5 host f0:00:20:6b:05:01:08:01:32:00:00:00:00:00:00:00:00:00:f7 host 2.4.2 f0:00:20:6b:05:01:09:00:2d:f7 2.4.5 host f0:00:20:6b:05:01:09:01:2c:02:01:00:00:00:00:00:00:00:f7 host 2.4.2 f0:00:20:6b:05:01:0a:00:39:f7 2.4.5 host f0:00:20:6b:05:01:0a:01:38:04:02:00:00:00:00:00:00:00:f7 host 2.4.2 f0:00:20:6b:05:01:0b:00:37:f7 2.4.5 host f0:00:20:6b:05:01:0b:01:36:03:01:00:00:00:00:00:00:01:f7 host 2.4.2 f0:00:20:6b:05:01:0c:00:2b:f7 2.4.5 host f0:00:20:6b:05:01:0c:01:2a:01:00:00:00:00:00:00:00:01:f7 host 2.4.2 f0:00:20:6b:05:01:0d:00:3d:f7 2.4.5 host f0:00:20:6b:05:01:0d:01:3c:00:00:00:00:00:00:00:00:00:f7
It is easy to notice that the first packet is significantly shorter than the fourteen following packets sent by host. Again, when I consulted MIDI SysEx specification, I came to the conclusion that it is the request for the device to identify itself (
0x06 0x01 – the fourth and fifth bytes of the first packet sent to the device – is described in the table as “General information: identify request”, while
0x06 0x02 – the fourth and fifth bytes sent by the device to the host – is described in the table as “General information: identify reply”).
I have failed in reverse engineering the meaning of the response (
f0:7e:01:06:02:00:20:6b:04:00:02:01:01:00:03:02:f7), but it is always the same, regardless of Microbrute’s state, so it cannot be used to respond with the current state. This must mean that the following 14 packets pairs (request-reply) are used to obtain Microbrute’s state… wait, there are 14 parameters in the configuration – that’s a perfect match!
Taking a closer look at the packets pairs, I noticed that they contain the “usual” stuff:
0xf0 as the first byte, then three bytes of Arturia’s identifier (
0x00 0x20 0x6b), a byte of Microbrute’s identifier (
0x05), probably a delimiter (
0x01) and a counter. The first seven bytes sorted out. The eighth byte is always
0x00 in the request and
0x01 in the response.
For the request, there is only one byte left (except for the usual ending byte:
0xf7). Having compared the ninth byte in request and response packets, I realized that in the request it is always greater by 1 than in the response. Moreover the value of this byte in the response is the same as the one used to specify the parameter in the requests used to set its values!
For example: the value identifying “Note priority” parameter is
0x0b. This means that it is required to set the ninth byte of the request to
0x0c. Requests for reading Microbrute data are fully reverse engineered now. I just needed to change the parameters a couple of times, request the data from Microbrute and sniff the packets to see how they changed. Doing that, I observed that in the response, the tenth byte was changing which indicated it holds the value of a given parameter. The eight following bytes, from the eleventh to the eighteenth, are quite weird, as they usually have value
0x00, only sometimes something different. Maybe it’s some kind of checksum, but for now they seem insignificant.
The request to obtain Microbrute’s parameter’s value has the following format:
XX is the counter and
YY is the parameter code incremented by 1, while the response has the following format:
XX is the counter,
YY is the parameter code (not incremented!) and
ZZ is the parameter’s value.
I prepared the full documentation of the communication between Microbrute and the application and it can be found here.