Rust Programming Language Rust Hungary Meetup, 4 Feb 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: One Factor Authentication (2020-02-04) - Rust Hungary Meetup

Description

Stefan talks about his One Factor Authentication (1FA) Proof of Concept (PoC).

Building a new Login System that uses one security factor less than what most of the big tech sites are trying to force on you?
Why would anyone want a new system that is less secure?

To answer these questions we have to look at where we get our security and talk about operational and long-term security and safety.

About Stefan:
Stefan started with Rust in 2015, has multiple Rust services in production, gives talks and workshops, maintains a couple of crates and currently works on his Computer Science Master.
https://estada.ch/

Meetup link:
https://rust-lang.hu/

Chat link:
https://matrix.to/#/#hungary:rustch.at?via=rustch.at

A

Hi guys yeah, I'm liv, I'm in it. Okay, hi guys after a big break. uh Welcome back here. I'm really glad to be here, and I see um how many of you are here with us. uh My name is juji. You've already met me before I am the organizer of the raspita.

A

I work at nimox crafty for three years now we are a recruitment company from 1999 and we organize and practice and participate participate.

A

Last fest also, I uh hoping a lot to have clients who are looking for us developers in the near future, uh and I hope it's not a dream um in budapest, um but now uh I would like to speak about stefan schindler. uh Thank thank you. Stefan to accept my invitation to give a talk on our meetup. um We already met also stefan and hello,.

A

He was the last meetup speaker in budapest before as well, and he is one of the last fast organizer, so you can ask him also about the online rust fest from november.

A

um It's a really fresh uh story, but now he is going to speak about one factor: authentication um yeah! So thank you. Stefan, you are on the stage.

B

Thank you very happy to be back. Welcome everyone, um quick about me. I've been doing rust for a little over five years already it's 20 21. um yeah and I started with a chatbot. Actually that was uh the first project that we wanted to migrate from a very often failing python implementation to something that would not constantly require us to restart it, and ever since I've been looking at uh many things and today uh this is less of this is all the things we have in rust, and this is how it's done in raspberries.

B

It's also like the whole other part over here. That will need both. We need to connect these two worlds, so we can have a system that actually works sushi. Yes, thank you very much, so the one factor, authentication, prototype or proof of concept.

B

I proposed that in fall to colleagues at the university, and they said: okay, uh let's have a look at it and then we dug into this project.

B

um It was with uh marco schadler and stefan kophamer credits to them, uh also for working with me, and we discovered that there's a multitude of one factor, authentication or, let's not say one factor but like passwordless authentication, because that was my original pitch. So um what is the things we do so we'll go over this quick history and then we'll look into what do we need and also look into the proof of concept.

B

Then I will quickly talk to you why I built in a container and then we'll have a lot of time for your questions.

B

So what do we need? What gives us security uh used to be just knowing where stuff is right in the very early internet in like the 70s, so few computers existed globally that we could just say if we know where something is we can have access to it? The internet wasn't a thing for many things. There was many competing standards like ip and netlink and ibm stuff and apple stuff, and none of them were compatible.

B

So if you wanted to reach something you had to be on the physical location, so why bother and also how many people knew what what to do with the computer, so user's names?

B

That was the thing that didn't exist in the beginning, but over time interconnectivity grew and we added more and more stuff until we are here today where everything is connected. Even your toaster can talk to your fridge, whether that's a good thing or not, is not a debate, but that's where we are. So what gives us security today on its entropy, more or less? How much guessing does someone we don't want in our system have to do to get in, although we don't want that.

B

So what alternatives do we have to uh passwords right? Because that was the standard thing at some point, people settled on this like yeah, um username password good enough right, five character. Passwords should be enough or these what android had with the maze where you had this graphical password or biometrics. Magnetics is easy right. You don't have to remember just have to look the same all the time and why just use like one password for all the services.

B

Why not just have one service that does it your all your passwords, so we have the federated single sign-ons. There's many services of delt um and we'll talk about this later so.

B

This xkcd, probably many of you, have seen it already. Let me quickly scroll to the number xkcd number.

B

Wait, I'm sorry. I have skipped all of the slides, quick history. There we go. This is what I wanted to show. So um so the five most prominent hacks in 2020 were probably zoom.

B

uh They didn't exactly say what was hacked, but apparently lots of data was lost, live journal I have never heard of, but apparently it was password stopping, which means they gained passwords and username combinations from one side and use them. On some other side, twitter was hacked more than once.

B

I think the funniest one was when an employee did something to ex-president trump's twitter account, uh wishbone and instagram, of course, attractive targets, but yeah.

B

So another fun story is the most common passwords in 2020 or these and to my as a surprise, four zeros isn't on the list anymore, probably because a lot of programmers forbid this password. But these aren't that hard. uh However, we can see in the top 20 there's always at least six digits or six characters. So there's some improvement. I think the last survey I've seen was from 2015, where we had four digits so yeah, it's something and we covered this already.

B

So this is hard to read, but basically authentication is. We want to know that the the system is talking to us actually belongs to the person that it says.

B

So we need something that the human can carry with them. Either knowledge uh something in possession like a smart card or a feature like a biometric feature or a behavior like the typing, the kind of typing. You do, there's a very interesting article, written about two years ago, where they analyzed.

B

How easy is it to identify people by letting them type a sentence just from the rhythm of the people, but biometrics- and this is this slide- we have all of these kinds of biometrics right. um They all have the same problem.

B

We're not machines, we're not always the same right. If we, let's say make face id, we have to be fuzzy as engineers, because people sometimes are tired. Sometimes they stand in the rain.

B

Sometimes they just age and aging is a real problem, because how can you differentiate two siblings right?

B

How can you differentiate an older from a younger sister or twins heart problems, and then sometimes can you even distinguish parents from their children because when they dress similarly and look similar, we have to be fussy, and then there is an amount of data that we want and an amount of data we don't want, and the overlap of this cannot be too small, because otherwise people will just say it's broken.

B

It doesn't work all the time, so we have to be a little fuzzy and give little headroom, and this headroom is what attackers can use to infiltrate systems which they should not have. Access to. Another thing is with diametrics. You cannot change them or it's really hard to change them. I mean you can sometimes change your biometrical behavior, for instance, if you have a door that identifies you by the kind of walk you go into right. What happens when you break your leg?

B

Suddenly it changes. The system doesn't recognize you anymore. What? If the system is trained to recognize your broken leg?

B

Can someone just tape their legs? So it looks the same to the system.

B

And in the end, what happens if the data is leaked because at some point enough data about people will be leaked? Can we change biometrics? We cannot. We can just switch to a new thing. So when we use our fingerprints, we can then maybe use iris scans, but at some point someone will make an iris database maybe build it up from the passport pictures database, but then well we can use ears or nails or lips, but do you really want to wait for and then in 20 years you?

B

The government is like oh well um now for more security. You have to kiss um some pad every time you want to use your passport, for instance, for boarding a train or something um so biometrics yeah it works for now, but I wouldn't trust it moving on. Let's see if this works correctly. Yes, so single sign-on, the most common in the public internet nowadays is probably all out second version.

B

It has multiple flows and you can revoke access.

B

The application has to handle revoking access during a session. But apart from that, that's pretty good. There's a an older standard. That's used at universities a lot, it's called shibolet. um It basically does the same, and I think I have a picture for that. No, we don't have a picture for that, no going back.

B

So um what all of these services have in common? Is the client requests access to your server and your server says? No. You have to authorize. First. Here is a list of identity providers. uh Without this could be google or gitlab or github, or something um with siebelet.

B

This has to be a university that the application trusts and, in the end it doesn't really matter because they all sign a random token. They hand over to the client. The client goes to the identity provider or chooses one from the list. Goes there and says: okay, I want to be authorized for this application.

B

The identity provider can then decide whether or not this request can be granted.

B

um Not always I mean with google your bit pretty much allowed to try to log in anywhere which simulate, for instance, you're only allowed to use university services. You cannot use it for your bank or or something like that if the login usually with username and password, is successful at the identity provider, the identity provider will then sign this token as well hand. It back to the client and the client can then take the signed and present it to the application, which then can verify the identity profile.

B

The the identity provider signed the token as well, and then let the user in and also knows what rules to apply, because the only thing the identity provider says is this: is this user in these groups do with it? What you want so.

B

Coming back to this, why should we do that? Why use something as complicated as a single sign or any third party service? um Let me just oop.

B

I'm sorry for that, so this is xkcd936.

B

um Passwords are hard because they need to be very secure. How much security do we need? We need roughly 60 bits of security. Well, how much do we need? How many different choices do we have? So this is the example with the english alphabet, so this is just uh lowercase letters and um some punctuations and spaces gives us uh 28 uh bits of entropy per character.

B

Sorry, is that right, yeah? I think that's right!

B

No wait! Oh, I can't remember 10 to 10 yes, 28 character, so in in theory we would need.

B

Four uh characters, so it should be sufficient for breaking through this thing, but you all know that it doesn't don't work. um What's the current recommendation well, according to ovas, it's 12 characters for a password, and it should look like this, like the one on the top left, uh troubadour and free, but many people use common substitutions like leadspeak many people just use their dog's name, their birthday or just the same password for all the things, even if that password is password, as we have seen before.

B

um So this comic proposes that we use our full words right. We just spell them out, and this gives us way more entropy. This is pretty great right because it's easier to remember, but what I would change nowadays is. Why does this have just to be the know? The nouns we can use all of the sentence like full with adjectives?

B

You can have this as a password right, the sentence it's a beautiful day and I like ice cream as a sentence with spaces and the dot at the end is a great password which you can remember far more easy than whatever uh this cryptic uh thing is plus it's like 50 characters. Long. So that's great there's some disadvantages like there is a couple of systems, especially in banking, that still store plain text passwords. Nowadays I found that out yesterday when I called one of the banks.

B

That holds an account for me where they said so now. We would like to have the third fourth and the fifth character of your password for authentication and you're. Just like great I'm, I'm really thankful for using a password manager, because that password is basically burned. It's obviously stored in clear text in their database.

B

So that brings us.

B

B

Working ah there's another slide missing, okay, so quick checklist. What do we need? We need passwords that are stored safely, aka, salted and hashed. There are some people that go further and say: okay, we need passwords that are also pepper. Pepper means that the hash is enhanced with uh another value that's stored in the source code. The idea is: if someone is able to leak your password database table, then they don't have access to the pepper which is stored in the source code. Ideally it's true sometimes in the cloud.

B

Sometimes it's not it's just another measure, but we want at least having salted hashes and then we would prefer to actually store even less right. If we can, then we just store an identifier that will hand off to an other service and then we're back to single sign-on, um but then we need to be very friendly to the users, because people will not like us. If we just say hey here is like a 128-bit identifier, why don't you use that? As your username?

B

I mean it's basically a password, so it doesn't matter what you choose for a password, but remember this 128 bit value, so they still want to use email and we can use the email for notifications and stuff like that as well and in the end, let me show you this.

B

It should look like this more or less right. We have a login form that just asks for the email address and the people will be able to click login, but I'm getting ahead of myself.

B

So, let's go through the development stage, I'm really sorry for the bad readability. So this is where the where stefan's rust tips come in.

B

I assume many of you have used cargo for building rust project. It's a package manager. We can also have add-ins. So this one is the cargo watch add-in you can install it with cargo, install cargo dash watch.

B

It will install a binary in your user's home folder, which then extends cargo, and you can say, cargo watch and then with x we can say different actions. So the first action is check.

B

Check means it compiles all your dependencies in debug mode and then tries all the steps but the linking with your code, so the dependencies have to build and your project would most likely build. It will be syntactically correct if this stage passes you get all the fancy error messages of rust and it's really fast. So cargo build takes with this project, roughly 30 seconds, which is very long for immediate feedback from the compiler with check it's done in less than one second, which is great, I mean after the initial build.

B

The initial build depends on on the 300 something dependencies I have, but after the initial build check is really fast, then we have fmt format. This will format the last source code in a standard way.

B

This is great because when I can type the ras code fast, don't have to spend time like indenting stuff. If it's synthetic syntactically correct, which the check step made sure of format will format it always the same. So the source code will always look the same, and my editor is smart enough to notice the file, change and just reload the file.

B

Unless I change something, then it will be like hey something has changed: you can either overwrite it or load the changes from the disk.

B

Then we have test step. uh This depends if you have actually have tests in your project, but uh for the default. It's it's. A great idea just run all the tests inside the rust source code, and this is run every time. You change some code, but every time I press ctrl s, it will check format and then test the thing. And finally, if all the tests pass, it will run the project, since this is using actix this project, the actix, is smart enough to let itself be killed.

B

Once the cargo watch detects a change, that's really handy, so we don't have to kill the web server all the time we can just have the browser open with our website the terminal open for the error, messages that are from either the compiler or own application, and then just edit in a fourth window.

B

B

What did I use, as I mentioned before, I use actics with some of its add-ons.

B

I use actics id sorry actics identity as the session storage, but this just handles that all the clients that have any contact with my service get.

B

They get the cookie, that's I don't remember pretty long at least 64 bits, random value and just store that set all the flags make sure it's safe that the javascript vm cannot access it, and I can I fixed that I just looked at the code.

B

It's a 128-bit, random value, that's assigned per client and the clients will just bring that id back and then I have uh my old little database, aka hashmap. That gives me back the session state for each step.

B

Then terra tera is a templating engine. It's pretty smooth. It can write almost html. It's based on the ideas from the django framework, so you can have small loops.

B

If else and imports, so you can actually layer the template, so I have a base template then all the other sides I have use the base template or almost all of them, use the base template and then just insert blocks of html into the rendered output.

B

Third day. I use 30 year, of course, through the framework, but then I wouldn't have it directly. I use 30 for configuration, so all of the configuration that it's done for this project is stored in adjacent files and they're just loaded on startup.

B

um I have three special kinds: certimillis and compo the compound duration. They allow me to store the standard, uh duration and standard instant in a json value which the standard um certainly does not implement. Then I have mail gun because a mail transport can be hard. It worked on. My first tests didn't work with another male provider, so I switched to having a remote api. I'm not 100 sure if I would stay with mail gun, but it's one of them. It works.

B

You can use any mail provider with ease that has an http interface, you preferably a rest interface, so you can have a token that's stored in some configuration file and then just whenever you have a request, pass that on to the rest service um and then, of course, rand for random numbers. And then this has been cut off, sadly, lazy, static, enveloper and identicon identicon.

B

I will show you later, if I don't forget, um just little extra touch like give the user some um some personality like hey, like I don't have space for your long email address all the time, but hey here is a picture that uh identifies you that you can remember, maybe uh by the way almost forgot about this. Thanks for the slide, I had to extend this crate and uh thanks uh to it being open source.

B

I was able to do that because for the university prototype I needed to use european services because we have to be gdpr compliant, so I was able to extend this crate and it was actually merged and it's released in the version 0.10.

B

um If you want to use that just update, it's not micrate, someone was kind enough to share moving on so building.

B

We have our source code now it works on on my machine and then I deploy it to the server and it doesn't even start. So. What is this it's classic case so? Well, I have arch on my local system. There's a debian installation on the server libsy doesn't work. This is a very common error. If you deploy without containers, there's two ways around this one: you can build a container if your server supports containers. You can just point the server to your container registry, tell it to download the container and be done with that.

B

In my case that just doesn't work now. I want to have a binary, that's naked in the file system path of the linux, so I'm going to use raster for the compiler cargo for all the testing before and then docker.

B

I want to switch to podman, but it's been a busy year and in the end the goal is to have gitlab with the gitlab registry ready. So it presents the container and then I can just clone the container, pull it down onto the server and let it run and injecting the configuration with uh overlays.

B

That's a very cool thing, so you can just tell um your binary hey.

B

This is this is the path of the the configuration file and then point that into a folder and that folder can be overlaid from your kubernetes or k3s environment. So you have a shared volume. That's like read. Only to the container read write for you. All. Your configuration is separate from the binary. If you update the binary configuration is untouched and reused. It's pretty great.

B

In our case, we need to have a binary in the end lucky for us, rust gives us static binaries, so we just have one file and all the static resources, but rsync is fine for that. So, let's have a look.

B

uh I use a make file for that. Those of you who are longer in my team, maybe remember them. There's a large debate or there used to be a large debate about cmake versus make.

B

In my opinion, make files are just more convenient way than writing a bash script, because I can have multiple targets in one file. I don't have to do argument, parsing all the time or weird switch statements.

B

So this is that um so building in a container first have to touch source. Why? I will tell you later, then we have this docker run.

B

We want to use a container, that's removed after the run, so we don't want to have lingering containers on our local system, because I mean this laptop has one terabyte disk, but with each container weighing around 500 megabytes, I would run out at some point- and this way I don't have to worry about this user means that we map the current user and group to what's inside the container, the root user that is working for the most part and um yeah.

B

We still need to uh touch it up. As we see in the next line, then we map the current project into user source. My app this is just the convention that the rust container uses and then we tell it okay. Yes, we want you to work inside this container inside this folder, with this image rust latest from the docker hub and then run this command cargo build release.

B

This is the normal cargo build command just running inside the container. Why do we do that? Because this container here is built based on the debian base image, so all the libraries match up, so the debian based image on my server that's used is the same as for this container, hence the the binary build in the end. Sadly, this doesn't take all the files, because the artifact at the end and some files are still created as the user. Instead of my local user, I have to change ownership of the release folder and the binary.

B

Also, you can use variables like this in the makefile and then we'll deploy directly we'll just ssh into the system, use systemd to control. Sorry to stop the service. There's a dollar sign missing here in the presentation.

B

Sorry, for that, it's in it's in the real make file and then we'll rsync all of the files we'll have our mail config, that's synchronized! So we have the secrets that are for the email configuration we'll synchronize them over the release binary the templates. This is all the things the terra needs and static. That's the javascript. The pictures and whatnot and they'll put them onto the host with the username in the the home. Folder of that user and then in the end, we'll say make system be restart.

B

The systemd restart is another rule that will just ssh into the host and restart the service. Again.

B

B

Remember this touch from before. Well, this was because we have build rs, build rs is used for a very handy little feature, it's imprinting, the git version and the git tag into our binary.

B

You can then see that on the page, let me just quickly reload this.

B

See down here version 4.0 also, I did just do a login sorry for spoiling the demo. um How is it done? Build rs is run every time. Cargo is run and it will send will start a command. It will use the git command with uh describe and the arguments for the described modules are tag and dirty which will give us um either.

B

If we're on the tag itself, it will just give us the name of the git tag and, if we're above, if we like, committed onto the version, it will show us. This is version, let's say: zero, four zero on this git hash, and if we have uncommitted changes in the tree, it will add dash dirty on the end of that we run the command and then, if it's successful, we'll map it to standard out, um sometimes people clone the code than they have get.

B

Sometimes they just download the zip version and don't have git installed. Then this fails with a bit of a cryptic error.

B

So the second step is, if any of the previous things fail, I will just use what's stored in cargo, the package version of cargo and append unknown because we don't know which commit we are dash dirty.

B

It may not be true that it's dirty, but usually when people download a zip file of source code, they make changes to it. So it's safe to assume it's dirty in some way.

B

Up until now, this has all been internal.

B

Nothing has been handed over to cargo.

B

So now we have to output information to cargo, and this is just done with print line, so we say cargo rust, so we'll we want cargo to add an environment, uh variable called git version and that's equal to our string, which is either what we parsed from the git command or our fallback.

B

And then we just did this every time we change something. So, for instance, if we change the html or a javascript file or css file, this build process is run again and the output is different. Therefore it has to be linked again, which again takes 30 seconds. So if I'm developing my ui components, I would rather have a fast reload of the server than waiting for inch one character in an html file. So we add so we add this rerun.

B

If n changed and then say, git version equal, the same thing only rerun if this environment variable changed.

B

It's not working perfectly, so that's why you have to touch the build rs source code, because, if build rs is being has been modified more recent than anything else, it will be rerun again. It will trigger this change and then it will actually build with the current version has been a learning for me.

B

I'm not sure how we can improve that, but yeah future works future works. So now, demo time uh as we have seen before, this is the demo. You can try it for yourself right now. Actually I opened up the verification, so everybody can try it. Let's quickly see if anyone did.

B

A

B

Me all right, um so what I did is just enter my email address up there and press login and then this is not great, it's a proof of concept, but it says what happened suppose were we able to send email? Because sometimes our third party email provider is not able to be reached right? We cannot reach it. We cannot find it. Maybe they have tls errors this time it worked and they responded with success.

B

They said be cute, thank you, and this is the id of that email. Now let me check on my email.

B

I got a little mailbag.

B

And it's really slow- and I show you this: do you see that okay, there's apparently a little problem with the stream yard? Doesn't work no matter, so we will just open a new tab with this email address, so it sends us an email that guesses, our name. It just uses the part in front of the app and then says, click this link to access your account, and this is the link I copy paste it in here and then, when we do, it says hooray.

B

Your current session state is logged in welcome to the system and now because this is the first time I've done this can go to our profile and see we did the login from this address, which is the ipv version. 6 of this laptop, and here we can see the identicon the little gimmick for people this.

A

B

Based on the email address so now, if we log out and log back in it sends us another email, it says: hey it's cued again and after I get the other email address.

B

Oh, I accidentally deleted it very sorry about that.

B

Clicking on the link again gives us back our access and it's not just a scam, so we can also just go back. This is the old email you can just press on the main page since it's cookie based, we are logged in again.

B

B

This was the demo, um do you have any questions.

B

See someone has shared the link in the chat already enjoy the demo.

A

B

I think juji, and I will be here in case there's, questions.

B

A

Last year,.

B

B

Oh no she's gone.

A

I am okay, I'm here again so.

B

A

Question maybe to stefan about anything.

B

B

One question that might come up: if you see this as a recording.

B

I might have to shut down the service in the near future because I might run out of the free quota of emails. I have on this account so.

A

Maybe you have a question here.

A

Oh, I just came into my room.

B

B

Oh, I see someone is typing.

A

B

I didn't pay too much attention during the talk. Did someone ask a question during or.

A

A

But there is a question here.

A

How much time did you spend on it till it worked, that's build, dot, rs beat heck was interesting. How much time did you spend on it still? First yeah.

B

Too much um how much was.

B

It I want to say almost a day stretched over two, so I came across um builders and was just like yeah.

B

This is cool, I will add it, and then I was developing on the binary so that that was quick um and then the next day I was working on the html and I noticed that it was really slow and- and it took me quite a while to figure out- um why was it rebuilding all the time until I disabled the builders again and then learning about this hack, yeah four hours, at least maybe five and, as I said, it's a bit too reliable.

B

So it just stops rebuilding uh everything until you have some change either in the source code or the builder s itself. So that's why the makefile has to update the builder s every time you do a release build because sometimes it happens that it does not include the new git version or the git tag.

B

Yeah, I hope this answers the question.

A

Thank you. Stefan there's another question here.

A

So is this meant to be a third-party service that other web apps use for authentication like sso, or is this project meant to be a library for old forest projects.

B

So this is a very good question. uh It can go either way.

B

One would have to refactor the login procedure into an actics module, so you can just plug it in as a middleware which I didn't do, because uh I didn't have a real application right. The application is showing your login attempts and when you did it.

B

So for sso it's.

B

If you're, if you trust a connection between your service and the one fa service, then you can just take the credentials from the user and forward them to the other service, like any man in the middle and then when the user submits the email that this was verified.

B

You could react to that and then go back and and let the user do what they're supposed to do or not.

B

In the state, that's currently in it's more on the path of a library, because we don't have any of the cryptographic implementations that we get from or out or shivolet. um Why is this still useful? It's very useful, because how much security do we actually gain from building a system that has an internal database for passwords and and all the logins and and stuff like that? We still have to manage the roles right with this system.

B

But all these systems that use email address are as safe as the email provider and um if you have a reset account link somewhere in the application, it doesn't really matter where you are. If someone uh guesses, the email password or is, has access to the email account of a victim, they can log in any service because they can reset the password.

B

So the question really is: do we actually need to force to the user to remember yet another password, or can we just say, hey look um the session. Timeout is quite high. There's a button up here for logging out.

B

If you don't want this session anymore, but in the end, the security level is not increased by having a special password for the service we might as well just use the email address, one more thing which isn't implemented because the project run out of time is: let's say we are in a non-trusted environment like you're on holiday somewhere in a hotel, and they have this wonky computer. But you want to access this service.

B

um The next step is, um we type in our email, address, click login, and it will show us okay, it's pending, and then you take out your smartphone open the email on your smartphone and then authorize the session on the not trusted computer. So you can log in another system.

B

I currently disable this mode because it didn't match the requirements we had for the project at university, but in theory it's absolutely doable and the found uh founding the the basic structure of the code is already there. But we have websockets and we can have the browser just have the websocket connection and as soon as the smartphone validates the email address, we can then go back and authorize the user on the non-trusted device.

B

Longer answer this time I did see.

A

B

Yeah is this the next question.

A

Yeah, what if the email arrives only after some minutes.

B

uh That can happen from the emails in the token, in the email in this demo is valid for 30 minutes currently for a production application. I would make this configurable, but in the end, that's a risk. We take right as soon as we have a third party and that third party is unreliable uh with email, there's more than one third party. Actually right, we have the email gateway we have.

B

Everything in between could be multiple hops for the system, like maybe someone uses that university email then this is one connection, then they have to forward it to some other email address and then maybe they have this their business account and they have forwarded that to somewhere else.

B

So it's multiple hops until it reaches the client and the client still has to synchronize uh yeah, no risk, no fun.

A

Okay, the next question is: do you not think that we are really a bit too much on email? There is a risk of someone's email, getting hacked and potentially being logged out of every service on the internet.

B

That is correct. There is a there's, a talk from tom scott, where he talks about the day that google went down and he proposes this exact scenario. Actually, with a little twist, he asks the questions. What have happens if you don't need a password for any google account anymore, but if some high level employee on purpose makes everything accessible on google, um it's the same question yeah you depend in the end. It's a it's a trade-off right. You depend on either all the things doing in-house you do all the things in-house.

B

You have to be doing them right. um If you have one error in in your chain of security, the chain is broken and when we include services or systems from outside our full control, we'll rely on them so um yeah.

B

If someone loses their email account today. This is really bad um and I'm pretty sure for most people. There is such a single point of of like trust right. It's usually their first or their second email address.

B

It's probably for for some people, it's it may be msn or hotmail like something really old uh for younger people is probably gmail because or apple mail, and they got their first smartphone and just use that for everything so yeah you can help avoid that a little bit by using uh two single sign-on services that you can authenticate uh crossover right. So you could build yourself a path out of this a little bit, if you say okay, I use single sign-on for google that uses microsoft and vice versa.

B

So you can log in with one and of the others into the account, but that doesn't solve the problem. If the email provider decides to delete your account for some reason, maybe the government issues.

B

What's it called like in an order to delete or hold your account, then you're done. That's true.

B

Yeah so yeah, the question is: what's more likely to happen, that people forget the password that they use a weak password or that the government decides to block their email address.

B

Yes, I think that's the answer.

A

Yeah here we are the next question correctly.

B

A

Lips is also a viable solution for a different lips version problem. Maybe.

B

I used to build with muscle in the past. Yes, uh I didn't do it for this project, because why did I not do.

B

B

Oh yes, there was a rumor that there are some problems with um accessing the event loops with actics and I never investigated it but um yeah it's it might work, definitely possible.

B

um In the end I had systemd, which does all the nice things. So in the event, I have a panic in the rust program. um Systemd will restart the service automatically. I get uh log capturing and lock rotation from systemd all of these things, and it was at that point where I was just like. I could either build in a container for mosul or I could just build in a container for the standard, tbn environment and then uh it's already there.

B

So that's the question. Oh that's the answer.

A

Okay, so before the next question, uh I have to say goodbye, so we passed the um so. Thank you, stefan very much for your presentation. It was really really good and I really hope that we are going to meet in this year next time um as another meetup and um thank you, everybody guys who saw us and- um and if you have any further questions you can, uh you can write it down into the chat. Stefan will be around and answer your questions.

A

uh So thank you again and have a nice evening.

B