3 ms·

LIFX claims[0] they have addressed this as follows: #1: WiFi credentials are now encrypted #2: We have introduced new security settings in the hardware #3: R

by antsar 8y ago

LIFX claims[0] they have addressed this as follows: #1: WiFi credentials are now encrypted #2: We have introduced new security settings in the hardware #3: Root certificate and RSA private key is now encrypted Can anyone ELI5 how #1 is even possible (in a meaningful way)? Doesn't the bulb need to decrypt this to connect to WiFi? Doesn't it need the key for decryption? And doesn't it have nowhere except onboard storage to retrieve that key from, since it isn't yet connected to WiFi? In that case, this "fix" would have no value besides PR. Am I missing something? [0] https://www.lifx.com/pages/privacy-security https://www.lifx.com/pages/privacy-security

kevin_nisbet 8y ago
It sounds like security theatre to me, however, there are a number of things that can be done to make it fairly difficult but not impossible to dump from an embedded device. IIRC, many embedded processors have a sort of secure enclave, that allows encryption of small amounts of data. The key is basically baked into the CPU, so if someone just takes the flash out of the device and tries to read it somewhere else, the file will need the CPU to be decrypted. I think mostly this is obsolete encryption, but that doesn't necessarily mean it's trivial to break. Some vendors also embed secrets in their software, to "encrypt" on disk. This doesn't tend to be real security, but does add the barrier that someone needs to go through the software to find the hardcoded keys. Shipping hardware should also have things like the JTAG and debug pins disabled, this makes it a lot harder to memory dump the in-memory state or get the device to load a custom image that would export the unencrypted keys in memory. So, there are barriers that can be placed, that does make attacking the hardware harder and out of reach of unsophisticated attacks, but nothing is perfect. In my experience and from what I've heard from the community, it's safer to assume meaningful security measures haven't been taken by the vendors, even if they say they really care about security. Many of the barriers will be like you suggested, encrypted on disk, with the decryption key also on disk.
- jtaft 8y ago
  It's an interesting problem, because all the connected devices would need the attack mitigation. Would be interesting to see a protocol which auto-revokes certificates for devices which do not digitally sign some "ping" message periodically. That way if someone digs through the dumpster, the credentials probably won't be valid.
- kayfox 8y ago
  Many embedded processors these days allow you to disable the JTAG/SWD debugging port. So once that is disabled it becomes quite a bit harder to read data from any of the on chip memory using external devices.
- jdietrich 8y ago
  IIRC, many embedded processors have a sort of secure enclave, that allows encryption of small amounts of data. The key is basically baked into the CPU, so if someone just takes the flash out of the device and tries to read it somewhere else, the file will need the CPU to be decrypted. I think mostly this is obsolete encryption, but that doesn't necessarily mean it's trivial to break. This would be the sensible implementation, but I doubt that LiFX have used a suitable (and more expensive) MCU. https://www.digikey.com/en/articles/techzone/2015/apr/security-is-the-key-to-success-for-mcu-based-iot-applications https://www.digikey.com/en/articles/techzone/2015/apr/securi...
clairity 8y ago
i”d love to understand this as well. it seems to boil down to how to secure the (un)encryption key, since unencrypted secrets in (volatile) storage should disappear when the bulb is removed from power, and other secrets in long-term storage should all be encrypted by the one primary (un)encryption key. how do you secure this key? secure enclave seems to make it harder but not impossible (as kevin_nisbet points out).
IshKebab 8y ago
You have to use a secure element essentially. But there's no way they paid for a chip that includes that functionality because it costs a fair bit of money (on the scale of a smart bulb).

by antsar 8y ago

kevin_nisbet 8y ago

It sounds like security theatre to me, however, there are a number of things that can be done to make it fairly difficult but not impossible to dump from an embedded device. IIRC, many embedded processors have a sort of secure enclave, that allows encryption of small amounts of data. The key is basically baked into the CPU, so if someone just takes the flash out of the device and tries to read it somewhere else, the file will need the CPU to be decrypted. I think mostly this is obsolete encryption, but that doesn't necessarily mean it's trivial to break. Some vendors also embed secrets in their software, to "encrypt" on disk. This doesn't tend to be real security, but does add the barrier that someone needs to go through the software to find the hardcoded keys. Shipping hardware should also have things like the JTAG and debug pins disabled, this makes it a lot harder to memory dump the in-memory state or get the device to load a custom image that would export the unencrypted keys in memory. So, there are barriers that can be placed, that does make attacking the hardware harder and out of reach of unsophisticated attacks, but nothing is perfect. In my experience and from what I've heard from the community, it's safer to assume meaningful security measures haven't been taken by the vendors, even if they say they really care about security. Many of the barriers will be like you suggested, encrypted on disk, with the decryption key also on disk.

jtaft 8y ago
It's an interesting problem, because all the connected devices would need the attack mitigation. Would be interesting to see a protocol which auto-revokes certificates for devices which do not digitally sign some "ping" message periodically. That way if someone digs through the dumpster, the credentials probably won't be valid.
kayfox 8y ago
Many embedded processors these days allow you to disable the JTAG/SWD debugging port. So once that is disabled it becomes quite a bit harder to read data from any of the on chip memory using external devices.
jdietrich 8y ago
IIRC, many embedded processors have a sort of secure enclave, that allows encryption of small amounts of data. The key is basically baked into the CPU, so if someone just takes the flash out of the device and tries to read it somewhere else, the file will need the CPU to be decrypted. I think mostly this is obsolete encryption, but that doesn't necessarily mean it's trivial to break. This would be the sensible implementation, but I doubt that LiFX have used a suitable (and more expensive) MCU. https://www.digikey.com/en/articles/techzone/2015/apr/security-is-the-key-to-success-for-mcu-based-iot-applications https://www.digikey.com/en/articles/techzone/2015/apr/securi...

clairity 8y ago

i”d love to understand this as well. it seems to boil down to how to secure the (un)encryption key, since unencrypted secrets in (volatile) storage should disappear when the bulb is removed from power, and other secrets in long-term storage should all be encrypted by the one primary (un)encryption key. how do you secure this key? secure enclave seems to make it harder but not impossible (as kevin_nisbet points out).

IshKebab 8y ago

You have to use a secure element essentially. But there's no way they paid for a chip that includes that functionality because it costs a fair bit of money (on the scale of a smart bulb).