The FitBit you can buy everywhere nowadays. However, regretfully they closed the protocol for outsiders. Thankfully I live in Holland where figuring out how a device works for purposes of interfacing with it, is one of my rights. :-) Complaints can be send to the Dutch government. Of course, I hope more people are gonna buy the FitBit if I manage to open it to the general public, just as happened with the Kinect which is used by many roboticists now over the entire world. Interesting by the way to read up on that story. The guy behind it, Johnny Chung Lee, ran the Adafruit OpenKinect contest as a covert operation (the hack was a joke).
My background is robotics, embedded programming, wireless sensor networks. At the startup DoBots we are working on very cheap Bluetooth Low-Energy solutions for home automation for example. I am not gonna tell how I reverse engineer things, but you can assume it is always quite simple. Just reading log files, BLE sniffing, dumping /proc/kcore
, perhaps JTAG so now and then, but that's seldomly necessary. :-) As Theodore Watson describes it, I am not a hacker, I'm just a tinkerer.
- Kyle Machulis (openyou) has the code to transfer data from the old versions of the FitBit, which was ANT based, the Ultra. All the current trackers, the One, the Zip, the Flex, and the Force use BLE and are encrypted (partly).
- Benoît Allard uses the USB log files of the dongle given with the FitBit on a Windows machine to understand some information. He uses the sources from sansneural for example to know enough to be able to send the
MEGA_DUMP
packet to the FitBit server. - Chris Wade tells he can pair, send commands and receive data, but if this is only (see code) triggering a micro-dump, an authentication request, then this is quite useless.
The data is encoded with different opcodes as also described by Benoît Allard. The data packets of 20 bytes are SLIP encoded as also explained by him. All control packets start with 0xC0
, so if you have a data packet that starts with 0xC0
it is mapped to a two-byte code 0xDB 0xDC
. And this of course also means that 0xDB
needs to be remapped, namely to 0xDB 0xDD
.
The MEGA_DUMP
packet is of the form (see Google doc):
2802 0000 0100 1502 0000 68FD 9E2C 0F07
...
The header of the dumped data I get is similar to that document.
2802 0000 0100 6B02 0000 1522 242F 1507
...
# next packet
2802 0000 0100 7002 0000 1522 242F 1507
...
Here 1522 242F 1507
is my device id. The value 6B02
and 7002
is a counter expressing the total number of bytes communication back and forth between my laptop and the Fitbit. You can find a larger dump at pastebin. If you can use it to make me look obese on the public part of the Fitbit website, or if you can decode it, I will give you some nice beers! :-)
The MICRO_DUMP
has a different opcode:
3002 0000 0100 5302 0000 1522 242F 1507
Last 4 bytes seem to denote the size (big-endian).
During a connecting there are packets sent at regular intervals (every minute so it seems by default). This packet is built up like this:
02 0C 20 17 00 13 00 04 00 | 1B 11 00
The 02
here means that it is an ACL data packet type, part of Bluetooth HCI. The ACL packet comes with a connection handle 0C
. The following two bytes 17 00
stands for the total data length, 23 in decimal notation. The bytes 13 00
indicates the length of the L2CAP part of the packet. And 04 00
indicates the attribute protocol. The byte 01
is a "handle value notification" opcode, and 11 00
for the handle itself. After this header follows the data itself:
02 0C 20 17 00 13 00 04 00 | 1B 11 00 || F0 F2 D7 53 | 21 01 00 00 | 7C 67 03 00 | 4B 06 00 00
The data F0 F2 D7 53
is in big-endian notation 0x53D7F2F0 = 1406661360 and is a timestamp. The next four bytes 0x00000121
stand for 289 steps, the next four 0x0003677C
stands for 223100 as a distance (0.2231 km). The next byte 0x4B
is a counter and probably this is true for the last four bytes as a whole. So, this was easy. :-) Note, that to get this type of data you first are required to go through an authentication procedure, so you can't just run hcitool -le*
and expect everything to be done.
The good thing is that the data is now encrypted. The bad thing is that FitBit did not send everybody the encryption key to get to their own data. For example, it is not possible to see the number of steps you take of your own BLE device that you proudly build yourself using a nRF51822 chip from Nordic. Pity! If you want to get smart homes, smart buildings, smart sports centra, it is important that you can give them access to your data, without parties in the way that think they have the right to your data. In Dutch baas in eigen buik was a campaign about feminist rights. Likewise baas over eigen data is a world-wide right that can improve our lives without denying companies to profit. But I digress...
There are references to AES and MD5 as clear from pastebin, but the data I get out of it 360501887459dbfa
and 94f36456-8644-4b54-adc0-33b94d4de419360501887459dbfa
are no keys... The prefix 94f36456-8644-4b54-adc0-33b94d4de419
looks more like some kind of internal id, perhaps used to store data locally, but encrypted.
I started from the assumption that there is encryption going on in the dongle or the several devices that talk with the Fitbit. This is probably wrong... This idea I got because steps were still updated on the smartphone after being disconnected from the server. However, I now know that these steps are part of the 'live data stream' that is not encrypted. All other information goes to the server first, and is then made available through public REST APIs. If this is the case, it might very well be that they only locations where encryption/decryption takes place, is on the Fitbit itself and on Fitbit's servers... Not promising...
The NFC chip in the Fitbit Flex, is from NXP Semiconductors. The IC is type NTAG203(F) and is a Type 2 Tag. It has a text record #1 which is the device id, but it in a different order than the 28 02
messages, the text is 07 15 22 24 2F 15
instead of 15 22 24 2F 15 07
. It must be tough for them to maintain all those inconsistencies! There is also an Android Application record, this opens up the Fitbit Mobile app when in proximity to the tag (which for me only worked after installing the FitTap app as well. Apart from the ID the NFC chip does not contain much interesting information.
The FitBit name is property of that company.
- Author: Anne van Rossum
- License: LGPLv3, get the License text yourself, please
- Copyrights: Me
- Date: July, 2014
Please, fork and submit issues.