[security] Introducing signing support to MySensors

[Proyectos Integrasoft]

@Anticimex And as for the radio. I followed the connection guide that comes out at https://www.mysensors.org/build/serial_gateway, even watching the video. I do not know if you mean another guide? If so, could you give me the link? Thank you for your collaboration.

[Anticimex]

@Proyectos-Integrasoft I assume you use an official release first of all. That personalization is more complicated than the one used on beta/development branch.
Then you are first expected to generate the keys (like you say). These keys are printed on the serial console. You then copy those into the personalizer sketch and reconfigure the personalizer to store the keys you have set. And then you run the personalizer to use those keys.
You can of course skip the generation step and set the hmac key manually using the personalizer. The only requirement is the size of the key (32 bytes) and that it is identical on all nodes.

[Proyectos Integrasoft]

@Anticimex
Could you please give me the link of the last official release? to verify that is the one that I have. When you say "copy" the keys in the sketch personalizer, are you referring to PERSONALIZE_SOFT? And what do you mean by configuring the sketch personalizer? Could you additionally tell me how it would be done manually? (example of sketch)

[Anticimex]

@Proyectos-Integrasoft I am not sure where to start. I assume you are familiar with c code? The signing solution available in the latest official release (which you find on github, I believe is 2.1.1) require at least fundamental understanding of how to adjust sketch code.
The documentation gives the exact lines to change.
There is, like I said, a step by step guide, and if you follow it you should end up with a properly personalized device. In this case, that is of less importance since you currently do not have a stable enough radio link to use security since you get NACKs for full size payloads (so neither signing nor encryption will work).
So you will have to make that work and get rid of the NACKs, before we should start worrying about personalization.
And like I said before, that is not a signing related issue. You will get the same problem if you try to send full size payloads of any kind. Just try to disable signing and send full size payloads.

[Proyectos Integrasoft]

@Anticimex
Okay, so I'll start by tackling things step by step ... How can I avoid getting NACKs in my log?

[Anticimex]

@Proyectos-Integrasoft as I said; ensuring good decoupling, a stable power supply (measure that to confirm). Also, counterfeit RF24 chips are all over the place that perform under par. There are quite a few threads here on this topic. So please post such questions in those, it is somewhat off topic here
Also, setting proper power levels can have a huge impact on the performance.

[ahmedadelhosni]

Hi @Anticimex Actually there is something that I can not understand regarding cryptograhpy. I want to know how other products like Fibaro, Smartthings, etc handles the security ?

Here in our library the SW is not a good idea, why ? I thought beacuse someone can dump the memory .. but is it that easy ? Can't we lock the code and memory ? Also in the hardware ATSHA solution, someone can easily take the chip and intercept our network and sniff it or even send commands as it is explained in the documentation and that's why we don't use security for public nodes as it usualy reports states. But can't we lock the chip ? and by some way only the atmega can communicate with it to get the key by some way ?

I read online that some people are using private and public keys .. if this is the case, then the private key is offcourse saved in the memory. How do they handle this problem ?

Do they use AES , SHA ? which encyption way ?

Also the nRF52, I tried to read a lot and they use private and public keys i guess.

lots of questions and I am confused but I want to know how do they handle protection for public nodes.

Can you please explain this to me ?

Thanks.

[Anticimex]

@ahmedadelhosni some devices, like the atmega, doss not support locking the memory, so the software based signing is inherently insecure in terms of hw theft.
Atsha204a based signing protection specifically against this because the personalizer locks the chip from readout. It is not possible to extract the hmac key from the atsha204a memory and the key is never transmitted OTA (unless you deploy the personalizer OTA).

[Anticimex]

@ahmedadelhosni and regarding reusing a node/atsha204a for attack purpose, we have whitelisting to protect against that. The serials used for whitelisting are also never send OTA (again, unless you send the personalizer OTA).

[ahmedadelhosni]

@Anticimex

1- So if we have a microcontroller that supports locking the memory then the problem is solved ? I know that SAM is being introduced now, Does it support this ?

2- what is then the purpose of locking the ATSHA if we can't extract the HMAC which we depend on it ?

Thanks.

[Anticimex]

@ahmedadelhosni
We lock the atsha to make sure it can't be readable.
It does not matter that samd supports locking or not. The atmega328p does not. For now, we have a security scheme that supports any target, so we have to have a system that works for all.
For MySensors v3, an entirely new security scheme is under consideration. But it will mean dropping support for the atmga328p as it is not powerful enough.
As for what others do, I suggest you ask them
Security can be implemented in many ways. Each with drawbacks and benefits. The one currently in use is a scheme that can work on basically any target with reasonable security and performance. It has drawbacks, yes, but at the time of implementation, these were considered acceptable.
For the future, more sophisticated schemes can be used which are easier to use, arguably more secure but more complex in terms of computational power and protocol. The core team is investigating various solutions.

[ahmedadelhosni]

@Anticimex Sorry but I didn't understand the benefit of locking the ATSHA to be unreadable ?
I know we do not lock it so that we can read the HMAC and use it during verification, but what is the usage of a locked ATSHA ?

[Anticimex]

@ahmedadelhosni what do you mean? All cryptography is performed inside the chip. The hmac key never leaves the chip after it has been programmed and locked. Thats the whole point with the atsha204a.

[ahmedadelhosni]

@Anticimex aha okay I understand a bit now. So we put s special hmac that does all cryptography jobs then it gives us something that is used for transmision?

Looks like i have to read the datasheet also

[Anticimex]

@ahmedadelhosni I'd suggest you start by reading the documentation on signing linked at the very top of this post. It explains in detail how the signing security is implemented.

[ahmedadelhosni]

@Anticimex yeah I read it several times before but maybe didnt pay attention to tge technical stuff

[Anticimex]

@ahmedadelhosni not really needed to be able to use it, but it hopefully helps in understanding it

[ahmedadelhosni]

@Anticimex yeah I know. I have already managed to use Siging in my network and it works.

I just wanted to understand more about how the code works and the technical stuff.

Thanks.

[Anticimex]

@ahmedadelhosni You mean this?

[ahmedadelhosni]

@Anticimex Yeah actually I have read this post like 20 times before but I guess I begun to really understand the "technical" stuff today.

So basically what I understood is that we have a HMAC Key, which is generated and is saved in all devices, this is when we do step 1 'generate key' and step 2 'personalize'. Thus when the gateway needs to send to node X, it send to node X asking for a nonce from the ATSHA on Node 2 board. Then node 2 sends the nonce over the air. THe gateway then uses this nonce to produce signed message by first applying SHA then use the HMAC key to produce the signed data. Then the signed data is transmitted over the air to the node X again which does the same operations again and verify that the nonce produces the same signed message in small period of time ( to avoid replay block attacks)

Is my understanding correct ?

[Anticimex]

@ahmedadelhosni yes

[ahmedadelhosni]

@Anticimex finally I understood it. Actually I don't like using just the code without fully understanding the implementation. Thanks for support.

I will come with more questions maybe

[Anticimex]

@ahmedadelhosni Feel free to ask, but it is all documented. If what I say does not correspond to what the documentation says, please let me know so it can be improved.

[ahmedadelhosni]

@Anticimex The documentation is great regarding how to enable the signing and make the nodes work. My questions were related more to technical stuff.

Actually I still have a problem that I shall only use the hardware in private places. I know we have whitelist but I don't like the idea of having to re program node to add or revoke other stolen nodes.

If I need to put a motion sensor outside then I will have to make sure that all other nodes inside my house accept messages from only my gateway for example. Because if this node is stolen I don't want someone to send same commands to my private nodes.

What do you suggest to solve this ?
Do I have to set all private nodes to accept signed data from gateway only ?

[Anticimex]

@ahmedadelhosni The documentation DO cover the technical aspects as well. I just linked directly to that chapter. What is missing from it?
What do you mean with "only use the hardware in private places"? If you only do that, then you don't need to revoke anything since the nodes most likely won't be stolen, right? Revoking is intended for "exposed" nodes.
You typically only need to reprogram your gateway since normally that is the node all other nodes talk to, and hence is the node that would carry the whitelist.
If you put a motion sensor outside, then enable signing and whitelisting on your nodes and then you can inform each and every node exactly what devices are permitted to communicate to it. If your other nodes only typically talk to your gateway, just have the gateway serial in their whitelist and they won't accept other nodes even if they would have the correct HMAC key to base their signatures on.
Even so, if a node is stolen I would highly recommend you change HMAC key because an attacker could dig out the serial of your gateway if it is in a whitelist in the stolen node and use that to spoof a gateway.

[Anticimex]

@ahmedadelhosni in your specific case, I would suggest, in order for you to not risk exposing your gateway serial to a thief, that you don't use whitelisting at the node at risk (the one outside). Then it won't reveal anything about your security if stolen, assuming you use an atsha204a.
Your gateway on the other hand, has a whitelist of every node in your system (if you so choose), so you can, as soon as you notice your node being stolen, remove its entry on the gateway, and it would be rejected. The attacker would then have to try to guess its way into figuring out a serial that match any other node the gateway accept in order to be able to "get in".
That way, your hmac, at least in theory, will still be secure and usable (and you shouldn't need to redo personalization on your network).

[ahmedadelhosni]

@Anticimex This seems a good solution.

I have to points to discuss here please.

First : I know that we have an API to specify that node 4 shall send this message to node 7 for example. In our library, does this communication happens without passing by the gateway ?
If for example in order to reach node 7, a repeater node 6 shall be used in between. Thus node 4, send to the gateway then to node 6 then to node 7 ?

In our case when we revoke our stolen node from the Gateway which is now node 4. will the message pass first to the gateway or if the attacker knows node's 7 serial, then node 4 sends it directly to 7 ?

Actually I guess it may pass by the gateway but I am not famailar how is the look up table implemented.

[Anticimex]

@ahmedadelhosni no, if you directly target another node, I do not believe it will pass through the gateway.
I would say that if something like that is desired, you target your controller and have the controller relay the message to the other node. Everything to/from the controller pass through the gateway. That becomes controller specific behavior though.

[ahmedadelhosni]

@Anticimex Second : Don't you think that the whitelisting need to be more robust ?

I mean that I don't like the idea of reflashing. Why don't you implement an API that can be used to securely add or revoke serials during run time ?

Also another idea which I would like to discuss. Maybe when a node is started, it sends it's serial number securely to the gateway and it is added to the whitelist for example.

The whole idea is that I don't really know how do other commerial products handle security for private and public nodes. All I know is usually you scan a QR code which is on the box. Do you have any idea ?

[ahmedadelhosni]

@Anticimex Actually something now came to my mind. Can't the attacker flash a gateway sketch easier and control all nodes now ? He has a trusted ATSHA with HMAC.
Am I missing something ?

[Anticimex]

@ahmedadelhosni signing has no (and will not have) a dependency to encryption so serial will never be sent OTA.
For MySensors v3 a complete new security scheme will obsolete the current one, so I won't make significant changes to the existing framework. Of course pull requests are always welcome for review.

[Anticimex]

@ahmedadelhosni nodes still need to be included and your existing gateway needs to be blocked out since they will have conflicting addresses.
But as I already said, if you get your devices stolen, the recommendation is to distrust your network and replace the keys.
These are the limitations when implementing security for systems as limited in resources as the atmga328p.

[Anticimex]

@ahmedadelhosni also, if your public node is stolen and your gateway had a whitelist (and your other nodes has whitelists) the attacker would not know the serials of your other nodes, and therefore not be able to sign messages to them (assuming your other nodes require signed and whitelist enabled messages).

[ahmedadelhosni]

@Anticimex aha so you mean that since our message frame contains the payload (not encrypted ) + signature so it is not applicable to send it OTA ?

So do you have any documentation for tracking 3.0 progress ?

[Anticimex]

@ahmedadelhosni no, I mean that I don't want features relating to signing to depend on encryption. Things like serials and hmac keys must never be sent OTA even with encryption enabled since the current encryption available (at least in SW) is very weak due to protocol limitations.

Progress on v3 is currently not moving because the core team is busy with other things.
We want to make a security solution that is robust, easy to use and properly secure so we do not want to rush anything. We are well under way with deciding the core principles but I will not publish the working documents because we quite frankly do not have time to handle questions from the public. Especially when we have not finalized the design.

Rest assured fully qualified people are investigating and discussing the matter. In due course the results will become public as we are working with an open source project so anything concrete will show on github.

[ahmedadelhosni]

@Anticimex Great. Thanks for your time.

[Anticimex]

@ahmedadelhosni don't mention it. I strive to be as transparent as possible when it comes to security. And please, please let me know if there is anything missing from the documentation or things that can be improved in it.

We keep the next level security somewhat under wraps for now since we are not completely sure ourselves on how it is supposed to work just yet. Once we decide on that, we may publish something to get general feedback from anyone who might have input on the design (if so it will be a highly technical document) but most likely this will be a development process where anyone can test and evaluate it on a branch (like the development branch today) long before anything becomes an official release.
We will try to document it continuously as it evolves once we get to start doing actual code.

[skywatch]

@Anticimex said in [security] Introducing signing support to MySensors:

@ahmedadelhosni

For MySensors v3, an entirely new security scheme is under consideration. But it will mean dropping support for the atmga328p as it is not powerful enough.

Eeekk... Does that mean any nodes based on the pro-mini etc will no longer work with v3.0 signing? Or will there be backwards compatibility to still use these with ATSHA204 as we do today?

[Anticimex]

@skywatch The security solution known as "signing" and "encryption" today will still be available in v3, but will then be referred as "legacy" signing. It will be considered obsolete and only bugfixing will take place, but it will still be supported. Also for the current newer devices, but it might not be ported to upcoming platforms with higher performence.

[skywatch]

That was a fast reply! - Thank you, at least I can carry on building now

Where are the latest docs for signing and encryption for ver 2.2.0-rc.1? A few links on the site no longer work and I am having trouble finding what to do (eg HW signing, attach ATSHA204 like this, then do this, then do the other thing and in your sketch do this etc.....) You get the idea!

[Anticimex]

@skywatch you have the links at the very top of this thread

[skywatch]

Wouldn't you believe it? - I just found this....

https://www.mysensors.org/apidocs-beta/group__MySigninggrpPub.html

[Anticimex]

@skywatch

[skywatch]

Lots to read today!

[skywatch]

Just had a quick look and that is a good explaination and example code too - This should be on the main site as it is much clearer and more user friendly than the current content.... Just my 0.02€

[Anticimex]

@skywatch it is intended to replace the main site documentation. But as master and development branches differ significantly currently, we are awaiting v2.2.0 release first.

[skywatch]

No problems. I have enough to learn a lot from now!

Just curious though as to what would be replacing the AT328p for v3.0 and signing and encryption - thinking to maybe order some now and get used to them..... Raspberry pi zeros everywhere or ...?

[Anticimex]

@skywatch nRF5 is a versatile, powerful option to an atmga328p plus nrf24 radio.

[skywatch]

I just looked - built in aes and lots of memory..... looks interesing.
Downsied is the higher price, but will have some good uses I am sure.

[alowhum]

Is there any news on Security 3.0?

[Anticimex]

@alowhum progress on that is best viewed here: https://github.com/mysensors/MySensors/issues/1118

Progress is quite slow at the moment because I am busy with other things of personal nature, like buying a house and moving into it, so don't expect much to happen this year (unless anyone would like to help out :))