Subscription services: hate them or loathe them, they sadly exist. More and more companies are moving towards them. It started with paying for subscriptions to watch movies and TV shows on Netflix, then free services added subscriptions to avoid ads, and then companies that sell physical devices started adding subscriptions to use them.
BMW wanted to include heated seats with a subscription, but after backlash, they dropped it. But many companies have them like 8sleep, their sleep mattress for $2,500 and then require a subscription service to use it. But no company has gone further in this business model than Whoop. They don't even "sell" their device, instead, they sell membership for either one or two years and you get the device for "free".
Whoop has a few sensors: heart rate, temperature, and blood oxygen sensors, but temperature and blood oxygen are only measured once a day, during sleep. Its alarm can only be set to ring only once a day instead of for multiple periods.
The first step is general reconnaissance, how does device communicates with the app, what happens on the device and what happens on the phone...
What can the device do:
Let's see what can we learn from the device with a BLE scanner
We can see it has heart rate service, which means without any additional reverse engineering we can get heart rate from a device. We can assume most of the data between the device and app is exchanged using CUSTOM SERVICE
. This service has a few characteristics:
Characteristics | Name | Writable | Readable | Notifiable | Handle |
---|---|---|---|---|---|
0x61080002 | CMD_TO_STRAP | + | - | - | 0x0010 |
0x61080003 | CMD_FROM_STRAP | - | - | + | 0x0012 |
0x61080004 | EVENTS_FROM_STRAP | - | - | + | 0x0015 |
0x61080005 | DATA_FROM_STRAP | - | - | + | 0x0018 |
0x61080007 | MEMFAULT | - | - | + | 0x001b |
(Names were gotten from the decompiled apk)
Since CMD_TO_STRAP
is the only writable characteristic, and its name is CMD_TO_STRAP
we can assume it is used to send commands to the device.
How can we see what data is sent to the device, there are multiple methods to do this:
First is using adb bugreport to do this you first must enable it in Developer Settings
under Bluetooth HCI snoop logging
.
Bluetooth HCI logs can be extracted with the following commands:
adb bugreport logs
unzip logs.zip
wireshark FS/data/log/bt/btsnoop_hci.log
This is a good way to extract many logs and then extract data from them to check for correctness and get information being used but I prefer using Wireshark directly on the device while using the app, to do this enable connect a phone to compute and check if adb
can see it using:
adb devices -l
If you see your device then it is connected after that just run Wireshark
, sudo is required
sudo wireshark
Select Android Bluetooth Btsnoop ...
After that filter for packets that use Bluetooth Attribute Protocol
with btatt
. Now that we only see btatt
packages go over an app and you will see data getting exchanged between device and phone.
After opening the app, we can see the following in Wireshark:
We can see that the app sends data to the device using CMD_TO_STRAP
, then we receive one notification on CMD_FROM_STRAP
and many notifications on DATA_FROM_STRAP
Data being sent to CMD_TO_STRAP
is
aa0800a823461600699b4cfb
Before we look at data being sent from the device let's try to see what data we send to the device. To help we can add filter btatt.handle == 0x10
, to test commands let's try to connect to the device from a computer.
There are multiple ways to connect to the device both from your phone and from your computer. Some of them are hcitool
and gatttool
, but I prefer using Python and pygatt
.
Scanning for devices:
from pygatt import GATTToolBackend, BLEAddressType
adapter = GATTToolBackend(hci_device='hci0')
adapter.start()
for device in adapter.scan(timeout=5):
print(device)
This will scan for devices for 5 seconds and print them out, to connect to the device:
device = adapter.connect('XX:XX:XX:XX:XX:XX', address_type=BLEAddressType.random)
Let's go to an app set an alarm to time and see what data the device sends:
aa10005723 6d 4201 d0366566 00000000 f62deb81 # 7:00 Exact time
aa10005723 6e 4201 0c376566 00000000 1023ccef # 7:01 Exact time
aa10005723 6f 4201 207d6566 00000000 fea1e060 # 12:00 Exact time
aa10005723 70 4201 50116566 00000000 f226a8bd # 4:20 Exact time
aa10005723 81 4201 f07e6666 00000000 7037c2a4 # Peak 06:20
aa10005723 82 4201 f07e6666 00000000 7151203d # Perform 06:20
aa10005723 83 4201 f07e6666 00000000 b18eaefc # In the Green 06:20
Let's examine the packages sent:
aa10005723
It seems that sleep goal is handled on a phone, and if sleep is in green range alarm new alarm time gets sent to the device. Now let's try to set an alarm to ring in 10 seconds. First, let's see if it checks package count and checksum
unix_time = int(time.time()) + 10
unix = struct.pack('<I', unix_time).hex()
package = f"aa10005723704201{unix}00000000f226a8bd"
device.char_write("61080002-8d6d-82b8-614a-1c8cb0f8dcc6", bytearray.fromhex(package), wait_for_response=False)
And it doesn't work, now let's see what kind of checksum is used:
aa100057236d4201d036656600000000 f62deb81
aa100057236e42010c37656600000000 1023ccef
aa100057236f4201207d656600000000 fea1e060
aa100057237042015011656600000000 f226a8bd
Go to crccalc.com, paste data and select hex
for Input and CRC-32
And no luck, let's try to reverse engineer checksum, @colinoflynn created a python package to reverse engineer CRCs, crcbeagle so first clone it and enter it:
git clone https://github.com/colinoflynn/crcbeagle
cd crcbeagle
touch examine.py
And then copy the following script to examine.py
, in your example put your packages in the data
from crcbeagle.crcbeagle import CRCBeagle
crc = CRCBeagle()
def hex_to_int(hex):
return [i for i in bytearray.fromhex(hex)]
data = """
aa100057236d4201d036656600000000f62deb81
aa100057236e42010c376566000000001023ccef
aa100057236f4201207d656600000000fea1e060
aa100057237042015011656600000000f226a8bd
""".split('\n')
data = [i for i in data if i]
checksums = [i[len(i) - 8:] for i in data]
data = [i[:len(i) - 8] for i in data]
crc.search(
[hex_to_int(i) for i in data],
[hex_to_int(i) for i in checksums]
)
Output provides code for checksums:
import struct
from crccheck.crc import Crc32Base
crc = Crc32Base
def my_crc(message):
crc._poly = 0x4C11DB7
crc._reflect_input = True
crc._reflect_output = True
crc._initvalue = 0x0
crc._xor_output = 0xF43F44AC
output_int = crc.calc(message)
output_bytes = struct.pack("<I", output_int)
output_list = list(output_bytes)
return (output_int, output_bytes, output_list)
m = [170, 16, 0, 87, 35, 109, 66, 1, 208, 54, 101, 102, 0, 0, 0, 0]
output = my_crc(m)
print(hex(output[0]))
Now let's add a checksum to the package we send:
package = f"aa10005723704201{unix}00000000"
checksum = my_crc(bytearray.fromhex(package))
package = f"{package}{checksum}"
device.char_write("61080002-8d6d-82b8-614a-1c8cb0f8dcc6", bytearray.fromhex(package), wait_for_response=False)
And no go, sometimes it throws an error and sometimes it just doesn't, so it is possible that this is the error sending and our package is good, so let's send data from the BLE scanner
And it works. So why it doesn't work with a computer, I don't know... I have tried to do it with another library bleak
, and it doesn't work, so the next try is with gatttool
:
sudo gatttool -i hci0 -t random -b XX:XX:XX:XX:XX:XX --char-write -a 0x0010 -n aa10005723704201126d6566000000003d59d8fd
And this "works", it works randomly but it is good enough for me.
With Wireshark opened, toggle BROADCAST HEART RATE
, these are the packages sent:
aa0800a823070e00c7e40f08 # Off
aa0800a823080e016c935474 # On
aa0800a823090e00cdc99102 # Off
After running enable_notifications.py
with notifications off:
python3 enable_notifications.py --address XX:XX:XX:XX:XX:XX 00002a37-0000-1000-8000-00805f9b34fb
It gets ERROR: could not find device with address 'XX:XX:XX:XX:XX:XX'
, but after having notifications enabled it connects and prints heart rate, so after running:
sudo gatttool -i hci0 -t random -b XX:XX:XX:XX:XX:XX --char-write -a 0x0010 -n aa0800a823070e00c7e40f08
and rerunning enable_notifications.py
it fails, and after running it works:
sudo gatttool -i hci0 -t random -b XX:XX:XX:XX:XX:XX --char-write -a 0x0010 -n aa0800a823080e016c935474
This means that the packet count is not checked and used at all.
After starting an activity packet gets sent and after it, notifications on DATA_FROM_STRAP
get received every second. These packages have the same header and format as packages for Broadcast heart rate.
aa0800a823 8c 03 01 7d5ec627 # Starting
aa0800a823 8d 03 00 dc040351 # Exiting
aa0800a823 90 03 01 6904fa32 # Start
aa0800a823 91 03 00 c85e3f44 # Exit
aa0800a823 8f 03 00 b2d08752 # Sleep start
aa0800a823 07 0e 00 c7e40f08 # Heart Rate Broadcast Off
aa0800a823 08 0e 01 6c935474 # Heart Rate Broadcast On
aa0800a823 09 0e 00 cdc99102 # Heart Rate Broadcast Off
aa0800a823 05 03 00 e44e25be # Health Monitor off
aa0800a823 06 03 01 2bc064cb # Health Monitor on
aa0800a823 66 74 01 3ae4cde3 # -||-
aa0800a823 91 45 01 dd861b95 # Alarm off
aa0800a823 0e 16 00 1147c585 # This command seems to retrieve data
Packet structure
0x03
is used to start/end activity and recording0x0e
is used to enable/disable the broadcast heart rate, it seems that the app can't read this value from the device, so if in the app it is set to enabled, it sends 01
every once in a while0x14
always sent with value 00
0x16
is used to trigger data retrieval on DATA_FROM_STRAP
, this is also a command that gets sent after the alarm is tapped, value is always 00
0x23
gets sent during the sync0x24
is used to get some string from the device0x43
IDK0x45
IDK0x4c
is used to get the device name0x73
returns notification on CMD_FROM_STRAP
with 01
0x74
returns notification on CMD_FROM_STRAP
with 01
, notification sometimes has data in string and sometimes it is empty0x75
IDK0x76
IDKAfter listening on DATA_FROM_STRAP
with
python3 enable_notifications.py --address XX:XX:XX:XX:XX:XX 61080005-8d6d-82b8-614a-1c8cb0f8dcc6
And running:
sudo gatttool -i hci0 -t random -b XX:XX:XX:XX:XX:XX --char-write -a 0x0010 -n aa0800a8238c03017d5ec627
We receive a notification every second and after running, they stop:
sudo gatttool -i hci0 -t random -b XX:XX:XX:XX:XX:XX --char-write -a 0x0010 -n aa0800a8238d0300dc040351
This is the data received after starting an activity:
Header Unix S HR RR RR data Checksum
aa1800ff2802 ad896566 f065 42 01 67060000000000000101 3ba00d4d
aa1800ff2802 ae896566 f860 43 00 00000000000000000101 5025f793
aa1800ff2802 af896566 085c 42 00 00000000000000000101 add7df13
aa1800ff2802 b0896566 1057 42 00 00000000000000000101 24b22179
aa1800ff2802 b1896566 1852 42 00 00000000000000000101 e0a905b8
aa1800ff2802 b2896566 284d 42 00 00000000000000000101 d943226a
aa1800ff2802 b3896566 3848 43 00 00000000000000000101 28364865
aa1800ff2802 b4896566 4043 43 00 00000000000000000101 9c2e99ba
aa1800ff2802 b5896566 503e 43 00 00000000000000000101 ebb8a1ce
aa1800ff2802 b6896566 5039 43 00 00000000000000000101 d1635459
aa1800ff2802 b7896566 6834 43 00 00000000000000000101 ccefa569
aa1800ff2802 b8896566 702f 43 00 00000000000000000101 8770ff99
aa1800ff2802 b9896566 802a 43 00 00000000000000000101 d2299e02
aa1800ff2802 ba896566 8825 44 00 00000000000000000101 3cd6a988
aa1800ff2802 bb896566 9020 44 00 00000000000000000101 94f13f2f
aa1800ff2802 bc896566 a01b 44 00 00000000000000000101 c8ed7785
aa1800ff2802 bd896566 b016 44 00 00000000000000000101 3b07bb4b
These are also the same packages that are received when Health Monitor
gets opened.
aa4800f323b8780167656e6572616c5f61625f74657374000000000000000000000000000000000032000...00625af8c
aa4800f323b9780173696770726f635f31305f7365635f647000000000000000000000000000000032000...0458e3fdb
aa4800f323ba780173696770726f635f70646166000000000000000000000000000000000000000032000...09334bc6a
aa4800f323bb7801656e61626c655f7231395f7061636b657473000000000000000000000000000032000...05db99544
aa4800f32346780167656e6572616c5f61625f74657374000000000000000000000000000000000032000...05c04a6f4
aa4800f32347780173696770726f635f31305f7365635f647000000000000000000000000000000032000...01faf36a3
aa4800f32348780173696770726f635f70646166000000000000000000000000000000000000000032000...02cb908fb
aa4800f323497801656e61626c655f7231395f7061636b657473000000000000000000000000000032000...0e23421d5
aa4800f323b5780167656e6572616c5f61625f74657374000000000000000000000000000000000032000...0bb6d53e8
aa4800f323b6780173696770726f635f31305f7365635f647000000000000000000000000000000032000...00908317e
aa4800f323b7780173696770726f635f70646166000000000000000000000000000000000000000032000...02e7c400e
aa4800f323b87801656e61626c655f7231395f7061636b657473000000000000000000000000000032000...0f4932608
aa4800f32328780167656e6572616c5f61625f74657374000000000000000000000000000000000032000...06418c0fc
aa4800f32329780173696770726f635f31305f7365635f647000000000000000000000000000000032000...027b350ab
aa4800f3232a780173696770726f635f70646166000000000000000000000000000000000000000032000...0f109d31a
aa4800f3232b7801656e61626c655f7231395f7061636b657473000000000000000000000000000032000...03f84fa34
These packets consist of header aa4800f323
, u16
number, string and 32-bit checksum, decoded:
aa4800f323 47224 general_ab_test2 0625af8c
aa4800f323 47480 sigproc_10_sec_dp2 458e3fdb
aa4800f323 47736 sigproc_pdaf2 9334bc6a
aa4800f323 47992 enable_r19_packets2 5db99544
aa4800f323 18040 general_ab_test2 5c04a6f4
aa4800f323 18296 sigproc_10_sec_dp2 1faf36a3
aa4800f323 18552 sigproc_pdaf2 2cb908fb
aa4800f323 18808 enable_r19_packets2 e23421d5
aa4800f323 46456 general_ab_test2 bb6d53e8
aa4800f323 46712 sigproc_10_sec_dp2 0908317e
aa4800f323 46968 sigproc_pdaf2 2e7c400e
aa4800f323 47224 enable_r19_packets2 f4932608
aa4800f323 10360 general_ab_test2 6418c0fc
aa4800f323 10616 sigproc_10_sec_dp2 27b350ab
aa4800f323 10872 sigproc_pdaf2 f109d31a
aa4800f323 11128 enable_r19_packets2 3f84fa34
I am not sure what this is, it gets sent randomly
To see how to retrieve data from the device, I turned off Bluetooth overnight and recorded packages during sync, the first thing I noticed is that the app asks for the journal then after sending requests it fails, and the app asks for the journal again after sync and after that, it succeeds.
After the phone connects to the device, a bunch of notifications are received on DATA_FROM_STRAP
the last one received has a following format:
unix don't care batch N checksum
aa1c00ab31 18 02 f65c7066 804043000000 2e470100 04000000000000 7f873cf3
aa1c00ab31 19 02 fb5c7066 704143000000 2e470100 04000000000000 f277ceb0
aa1c00ab31 1a 02 005d7066 684243000000 2e470100 04000000000000 6b7b573f
aa1c00ab31 1b 02 055d7066 684343000000 2e470100 04000000000000 0b876f87
The important part is batch N
which needs to be sent in the following format to trigger the next burst of notifications:
aa10005723{packet count}1701{batch N}00000000{checksum}
After this notifications on DATA_FROM_STRAP
start being received, these packets are very large(96 bytes):
First 31 bytes:
Header Time? Unix? Something HR RR RR data
aa5c00f02f0c 07 8bb7 0900 c8326966 e03c8054cc01 58 01 b902000000000000
aa5c00f02f0c 07 8cb7 0900 c9326966 f0378054cc01 58 01 b502000000000000
aa5c00f02f0c 07 8db7 0900 ca326966 f8328054cc01 58 02 b802b90200000000
aa5c00f02f0c 07 8eb7 0900 cb326966 082e8054cc01 58 01 ce02000000000000
aa5c00f02f0c 07 8fb7 0900 cc326966 10298054cc01 58 01 c102000000000000
aa5c00f02f0c 07 90b7 0900 cd326966 20248054cc01 58 02 df02d30200000000
aa5c00f02f0c 07 91b7 0900 ce326966 281f8054cc01 57 01 f802000000000000
aa5c00f02f0c 07 92b7 0900 cf326966 381a8054cc01 57 01 fb02000000000000
Remaining 65 bytes, I have no idea what it is:
00004563f880c9143cae47afbe5ccf5b3ef6a4753f00008e46ae47afbe5ccf5b3ef6a4753fe50146021c03440282016004010c020c3000000000000002 2dcbe388
0000cf6bff00e8e23b5248afbe66265c3e0078753f0000c8465248afbe66265c3e0078753fe50146022003460282016004010c020c3000000000000002 4d49cb2e
0000f96dff00a7e73b140eafbe298c5b3e715d753f00001046140eafbe298c5b3e715d753fe50146022003470282016004010c020c3000000000000002 833a28fd
0000a16eff0070b43b1fedaebe3d9a5b3ecd34753f00005ac61fedaebe3d9a5b3ecd34753fe50146022003450282016004010c020c3000000000000002 149f6d4f
00008b63ff0041a43b14e6aebea4b05b3e4819753f0000c0c514e6aebea4b05b3e4819753fe50146021e03450282016004010c020c3000000000000002 a4d38c63
00004661ff001bb03b8fcaaebe1fd55b3e0008753f0000b2c68fcaaebe1fd55b3e0008753fe50146021d03440287016004010c020c3000000000000002 2891f236
0000d161ff00f3a43ba480aebe859b5b3ed7e7743f0000e046a480aebe859b5b3ed7e7743fe50146021d03450287016004010c020c3000000000000002 c197f209
0000f961ff0041fd3b7bb4aebeae475b3e1fd9743f000082467bb4aebeae475b3e1fd9743fe50146021d03440287016004010c020c3000000000000002 aec41896
0000a16aff0061c63b48a1aebe3d0a5c3e66aa743f0000d84548a1aebe3d0a5c3e66aa743fe50146021c03440287016004010c020c3000000000000002 dfafea59
00008c6aff00c8b43bec71aebe33c35b3e9ac1743f000050c5ec71aebe33c35b3e9ac1743fe50146021c03440287016004010c020c3000000000000002 76ae1006
And then two Enable
commands get sent at the end:
aa0800a823 15 73 01 f4a43bfa
aa0800a823 16 74 01 6a8c3cb7
When the erase device button is pressed following package gets sent:
aa10005723cf19fefefefefefefefe002f8744f6
aa10005723d219fefefefefefefefe00e30e2693
aa10005723d319fefefefefefefefe0023d1a852
For reboot device following packages get sent:
aa0800a823d41d003c2e2fe6
aa0800a823da1d003603b1ec
aa0800a823df1d00ddc17aea
aa0800a823e21d001eb7c9c6
On EVENTS_FROM_STRAP
multiple packages get received, I am not sure what these packages have, the unix timestamp in them doesn't represent any start or end
header unix checksum
aa2400fa30 b0 0300 2e316966 901f 140002 e900 0000 e90e 0000 01 01 0f 03 01 00 2f 01 000000 699d4a60
aa2400fa30 64 0300 6a316966 d02e 140002 f100 0000 ed0e 0000 01 01 01 04 01 00 2e 01 000000 2d6beb1e
aa2400fa30 28 0300 a6316966 703d 140002 f900 0000 f00e 0000 01 01 37 04 01 00 2d 01 000000 50e4148e
aa2400fa30 f3 0300 e2316966 584c 140002 0101 0000 f20e 0000 01 01 31 05 01 00 2c 01 000000 0a6472e1
aa2400fa30 57 0300 1e326966 a05a 140002 0901 0000 f60e 0000 01 01 23 06 01 00 2b 01 000000 b6aff508
aa2400fa30 6b 0300 5a326966 5869 140002 1101 0000 fb0e 0000 01 01 15 07 01 00 2b 01 000000 b584c4fc
aa2400fa30 76 0300 96326966 a877 140002 1901 0000 4a0f 0000 01 01 07 08 01 00 2a 01 000000 f6cfde23
aa2400fa30 87 0300 d3326966 c805 140002 2101 0000 480f 0000 01 01 35 08 01 00 29 01 000000 7586b95f
aa2400fa30 a1 0300 0f336966 d013 140002 2901 0000 4b0f 0000 01 01 2f 09 01 00 28 01 000000 b7d66926
aa2400fa30 ac 0300 4b336966 f020 140002 3101 0000 480f 0000 01 01 21 0a 01 00 27 01 000000 5b22a14b
aa2400fa30 b7 0300 87336966 682e 140002 3901 0000 4f0f 0000 01 01 13 0b 01 00 26 01 000000 bc513013
aa2400fa30 c5 0300 c3336966 f03b 140002 4101 0000 500f 0000 01 01 05 0c 01 00 25 01 000000 2849050a
The same goes for these packets:
header unix checksum
aa10005730 5b 2100 3f326966 6854 0000 b0b2435b
aa10005730 65 2200 45326966 a866 0000 093b5aa6
aa10005730 66 1800 48326966 3012 0000 ef5360f0
aa10005730 81 1800 ba326966 2873 0000 9b28989f
aa10005730 6a 2100 e45e6866 e068 0000 323b066b
aa10005730 6e 2200 e85e6866 2843 0000 1489a6e0
aa10005730 70 1800 135f6866 f810 0000 f05e2f2f
aa10005730 75 1500 1c5f6866 b003 0000 55066632
aa10005730 76 0700 1c5f6866 a005 0000 ac1a4bdc
All packages received on EVENTS_FROM_STRAP
have the header, counter, some 8-bit number, 00
, unix timestamp and then some data, but none of the unix times correspond to any activity starts or endings.
Part 2 with blood oxygen and temperature measuring coming soon