Password-less login(medium.com)
medium.com
Password-less login
https://medium.com/findworkco/password-less-login-df0354c3f3ee#.gutdzrjjj
63 comments
>The difference here is that if/when someone uses their (unauthorized) access to my e-mail account to reset the password for this service, I will notice the next time I attempt to log in.
I am assuming the indicator would be an email from a service stating password reset but if the person has access to your email account, the obvious thing to do would be delete such an email. There is no difference.
Passwordless login is basically piggy backing on the security of the email account. You'd have to login to a site anyway but you are just logging in to your email which is assumed to be protected better.
I am assuming the indicator would be an email from a service stating password reset but if the person has access to your email account, the obvious thing to do would be delete such an email. There is no difference.
Passwordless login is basically piggy backing on the security of the email account. You'd have to login to a site anyway but you are just logging in to your email which is assumed to be protected better.
I think OP meant that he will clearly realize if someone used his email account to reset his password because next time he tries to log in his old password will no longer work. However with passwordless authentication, someone with access to your email account could issue your account for potentially a long period of time without you noticing. You would still be under the impression that your account is secure.
Con: Greylisting.
My primary email uses greylisting to combat spam. This means that if we receive a mail from an unfamiliar mail server, it's getting an error message saying the mail can not be delivered at the moment, and to come back later. Properly configured servers try again several minutes later. Spam bots seldom try again, so it works well.
I can see how greylisting and tokens that expire quickly can be a problem. Particularly if you have multiple servers for outgoing mails, so that the second mail attempt may come from another IP than the first.
My primary email uses greylisting to combat spam. This means that if we receive a mail from an unfamiliar mail server, it's getting an error message saying the mail can not be delivered at the moment, and to come back later. Properly configured servers try again several minutes later. Spam bots seldom try again, so it works well.
I can see how greylisting and tokens that expire quickly can be a problem. Particularly if you have multiple servers for outgoing mails, so that the second mail attempt may come from another IP than the first.
I'd say that this is a con for graylisting, and not for passwordless access.
Or at least, for both.
When email comes delayed (with invalid token) can you white-list that particular sender? You'd then request another sign-in email that comes promptly this time.
Well this is certainly a problem, but it's mostly limited to the first login, as (IMHO) the mail server(s) should use DKIM so greylisting can safely assume it's receiving mail from the same (legitimate) domain (the actual IP used becomes irrelevant). The first password-less login email could allow for a longer expiration, and it wouldn't be worse than a password-ful first login with email verification anyway.
> You can't leak passwords if you don't store passwords.
That's exactly why you only store password hashes in the DB. Come on, this is a solved problem!
> Pro: less implementation effort
Sure. But how much of your app's total dev effort is it? Nothing.
I hate the UX of having to log in by email.
That's exactly why you only store password hashes in the DB. Come on, this is a solved problem!
> Pro: less implementation effort
Sure. But how much of your app's total dev effort is it? Nothing.
I hate the UX of having to log in by email.
>> You can't leak passwords if you don't store passwords.
> That's exactly why you only store password hashes in the DB. Come on, this is a solved problem!
Isn't leaking hashes dangerous for the users' privacy though (I'm not well-versed in security, so sorry if n00b question)?
> That's exactly why you only store password hashes in the DB. Come on, this is a solved problem!
Isn't leaking hashes dangerous for the users' privacy though (I'm not well-versed in security, so sorry if n00b question)?
Not if you salt and choose a strong hash algorithm. Salting is what keeps a leaked hash from hurting a user's security
Though if I have the user name and the salted hash and also access to the service providers system, can't I use those to initiate a session? I mean getting the cleartext password allows me to start a session through the UI and also try to penetrate the user's other accounts elsewhere (maybe they reused the password), but if I have the salted hash and access to the sytem, I'm still able to get a session there, no?
Salting is to cover password reuse across services. If you already have the keys to the kingdom for a service you can almost certainly access a user's session
When you submit the hashed password it will get re-hashed and no longer match the hashed password in the database.
I'm not sure what you mean by having access to the system. If you have write access to the database you can just change the user's password to whatever you want.
I'm not sure what you mean by having access to the system. If you have write access to the database you can just change the user's password to whatever you want.
> Isn't leaking hashes dangerous for the users' privacy though (I'm not well-versed in security, so sorry if n00b question)?
No, you can't compute the password from the hash.
No, you can't compute the password from the hash.
I'd call this dangerously un-caveated!
https://arstechnica.com/security/2012/12/25-gpu-cluster-crac...
If your users passwords are sufficiently weak (and let's face it - they are) they can likely be brute forced and then matched against leaked hashes, no matter how strong your salt and hash. If you've got good users who only give you "typically" weak passwords, weak unsalted hashes will still expose said passwords.
Use a strong hash. Perhaps run multiple times, bcrypt style. Salt correctly, and per-user. Somehow convince your users to actually use strong, non-reused passwords. Perhaps then there's no danger to leaking the hashes - but I'd perhaps err on the side of caution and not make the password database publicly readable nonetheless.
https://arstechnica.com/security/2012/12/25-gpu-cluster-crac...
If your users passwords are sufficiently weak (and let's face it - they are) they can likely be brute forced and then matched against leaked hashes, no matter how strong your salt and hash. If you've got good users who only give you "typically" weak passwords, weak unsalted hashes will still expose said passwords.
Use a strong hash. Perhaps run multiple times, bcrypt style. Salt correctly, and per-user. Somehow convince your users to actually use strong, non-reused passwords. Perhaps then there's no danger to leaking the hashes - but I'd perhaps err on the side of caution and not make the password database publicly readable nonetheless.
Depends. If you use a nice fast hash like SHA256, yes. If you use a KDF optimized for nonoptimizability like PBKDF2 or scrypt, no.
This implementation is just skipping the charade that passwords are where the security happens.
With almost all current systems the place security really needs to happen is at the email provider. If you can hijack/intercept a users email or hack their account then it's game over with a password anyway, thanks to password resets being (almost) universally done over that channel.
There's an argument that this is no bad thing as security focused email providers could potentially provide very secure access as part of their core business (unlike a startup just trying to get going).
But there's still something bugging me about this implementation I can't put my finger on.
With almost all current systems the place security really needs to happen is at the email provider. If you can hijack/intercept a users email or hack their account then it's game over with a password anyway, thanks to password resets being (almost) universally done over that channel.
There's an argument that this is no bad thing as security focused email providers could potentially provide very secure access as part of their core business (unlike a startup just trying to get going).
But there's still something bugging me about this implementation I can't put my finger on.
E-mail is good enough for me as a channel. But the service should allow me to upload a PGP pub key and use it for all further communication, no exceptions (including password resets, etc.).
That would secure the channel and the website would be able to trust that whoever is clicking the reset link is the owner of the key.
It would also be a nice second factor. You have to have access to the email account and to the private PGP key. And it is a second factor I can fully control, unlike SMS. And it's simple and cheap for the service.
That would secure the channel and the website would be able to trust that whoever is clicking the reset link is the owner of the key.
It would also be a nice second factor. You have to have access to the email account and to the private PGP key. And it is a second factor I can fully control, unlike SMS. And it's simple and cheap for the service.
Thinking on this it would be nice if my email provider could segment the kingdom that is my account.
My normal login only needs to see about 10 days of email, I could re-auth or use a bigger PW to get into the whole archive.
Isn't this implementation kind of a really basic idea of what mozialla persona was getting at? https://wiki.mozilla.org/Identity/Persona_Shutdown_Guideline...
My normal login only needs to see about 10 days of email, I could re-auth or use a bigger PW to get into the whole archive.
Isn't this implementation kind of a really basic idea of what mozialla persona was getting at? https://wiki.mozilla.org/Identity/Persona_Shutdown_Guideline...
The only thing that's bothering me is that it is done through e-mail. I would prefer the token to arrive as a text message to my phone number. I feel like it is safer and faster. However, I understand that it is more expensive option, especially for services where you need to log in multiple times, compared to mobile apps where you log in only once.
SMS (in it's classical form) is highly spoofable as has been proven with TAN codes time-and-time again. It will work most of the time (sms has no guarantee of delivery as far as I know) and is somewhat secure but by no means good in any way.
This relies on email, which is highly secure, smtp is not spoof-able, server and mail origins are traceable and verifiable and there is no way to intercept the message on a malicious router/switch whatever. Ow wait ...
This relies on email, which is highly secure, smtp is not spoof-able, server and mail origins are traceable and verifiable and there is no way to intercept the message on a malicious router/switch whatever. Ow wait ...
How do you encrypt any user data without a password/key? Intelligentsia's iPhone app does this, and I deleted my account because I couldn't figure out how they were protecting my payment info.
I built (essentially) this a few weeks back, for an iOS app. Currently trying to get Apple's App Review team to agree it's a good idea.
Apple has UI in App Store submission that lets you provide App Review with a demo account username/password combo, but only that.
So to highlight another drawback: sometimes being able to share login credentials is a feature. Part of me thinks this ought not to be the case, and I hope I will be able to stick with password-less login.
Apple has UI in App Store submission that lets you provide App Review with a demo account username/password combo, but only that.
So to highlight another drawback: sometimes being able to share login credentials is a feature. Part of me thinks this ought not to be the case, and I hope I will be able to stick with password-less login.
Can't you just whitelist some token for them? If the sharing is the issue you're having with them anyway.
I'm looking at that. I don't know what email they will use to test, and I don't know if they'll look kindly at a special case for reviewers. Do you have any experience with that?
Is it really necessary that the email link is clicked on the same device or to use a typeable token? Clicking the link on any device is enough to confirm to the server that the login attempt should be granted access. The server can then send and authentication token to the browser where the user is logging in.
The downsides I can think of in that case are an attacker attempting to log in at the same time, or -- more generally -- a user thoughtlessly clicking the link when an attacker is trying to log in.
The first problem can be addressed by displaying a second readable token on the login screen which the user should verify is present in the email before they click the link. That is how banks and other high security systems do it here in Norway now when you select authentication via mobile, except they use some sort of mobile communication tied to the sim card instead of email. You get a popup on your phone containing two random words from a dictionary. You confirm that the words match what the browser is currently displaying and then tap "OK".
The second problem can probably be reduced by short token timeouts and using appropriate language in the email to emphasize the implication of clicking the link uncritically.
The downsides I can think of in that case are an attacker attempting to log in at the same time, or -- more generally -- a user thoughtlessly clicking the link when an attacker is trying to log in.
The first problem can be addressed by displaying a second readable token on the login screen which the user should verify is present in the email before they click the link. That is how banks and other high security systems do it here in Norway now when you select authentication via mobile, except they use some sort of mobile communication tied to the sim card instead of email. You get a popup on your phone containing two random words from a dictionary. You confirm that the words match what the browser is currently displaying and then tap "OK".
The second problem can probably be reduced by short token timeouts and using appropriate language in the email to emphasize the implication of clicking the link uncritically.
cons: If the customers email service goes down, then their access goes down.
If your email service goes down, then the service goes down for everyone.
If a major email provider is compromised, then a bunch of your accounts can be compromised.
All public server-to-server email communication takes place over port 25, unsecurred clear text. That seems like a poor choice for anything security related in 2017.
If anything we should be removing insecure protocols like email, telephone and SMS from the authentication chain rather than relying on them even more heavily than we already do.
I hate to suggest it but authentication via something like WhatsApp might be more secure.
If anything we should be removing insecure protocols like email, telephone and SMS from the authentication chain rather than relying on them even more heavily than we already do.
I hate to suggest it but authentication via something like WhatsApp might be more secure.
Server-to-server email is mostly STARTTLS (but without certificate validation) these days, I believe. Not great, but better than you think.
An interesting idea for a few reasons.
Users don't have to remember passwords, which is a strong benefit, but the user's security is now dependent on the security of their mail provider, or their own security practices in relation to the use of that mail provider, and in all likeliness their smart device.
With the ubiquity of smart devices, this works from a UX point of view and could be really streamlined, but I wonder how much the meme at the top of the article is actually betraying - this is the service provider not wanting to be responsible for your security in relation to their service.
I'm not suggesting that this is unreasonable per se, but the ultimate evolution of this model looks like it could (potentially dangerously) increase the value of a smaller surface area (the smart device) and put a fairly huge onus on the end user understanding and being responsible for their security.
And whilst that seems ostensibly ok, we already know that the weakest link in security is the end user...so this could be putting all your eggs in one basket.
Users don't have to remember passwords, which is a strong benefit, but the user's security is now dependent on the security of their mail provider, or their own security practices in relation to the use of that mail provider, and in all likeliness their smart device.
With the ubiquity of smart devices, this works from a UX point of view and could be really streamlined, but I wonder how much the meme at the top of the article is actually betraying - this is the service provider not wanting to be responsible for your security in relation to their service.
I'm not suggesting that this is unreasonable per se, but the ultimate evolution of this model looks like it could (potentially dangerously) increase the value of a smaller surface area (the smart device) and put a fairly huge onus on the end user understanding and being responsible for their security.
And whilst that seems ostensibly ok, we already know that the weakest link in security is the end user...so this could be putting all your eggs in one basket.
Isn't user security already dependent or their mail provider? Password reset mails are pretty common.
> but the user's security is now dependent on the security of their mail provider
Let me put it this way. In many (but not all) cases I would much rather Google be responsible for my security rather than some random startup whose primary objective is to stay alive, not take security seriously.
Let me put it this way. In many (but not all) cases I would much rather Google be responsible for my security rather than some random startup whose primary objective is to stay alive, not take security seriously.
It's funny you say that because I feel the same way as a developer. However, users of my site wanted password-based logins even though I support multiple social logins (Google, FB, Microsoft, etc.). Granted, I use PBKDF2 salting and hashing and I take security seriously, but it's interesting how average people sometimes don't trust social providers or think I'm storing their credentials to that provider.
Password-less login is a bit of a weird naming. I'd call it one-time passwords. And then I'd maybe not use email, since this clutters the inbox. Instead something like Google Authenticator or SMS passwords would be a good solution, the latter maybe being a bit expensive for the service provider.
That would make having an SMS-capable phone nearby a hard requirement to simply login to some service. I am fine with offering additional authentication methods — if they are safe enough for that application — but dismissing passwords as a concept (unless you are scaling up to proper 2FA, e.g., Fido U2F) just means losing users for no good reason.
> since this clutters the inbox
Email can be deleted.
Email can be deleted.
Password-less as in "one password less to remember/invent".
You can easily leak whether someone has an account on your website -- if I know their email address and input it, does it basically tell me? This limits your ability to reasonably error if the email address inputted is wrong.
Still, a lot of "forgot password" forms seem to be subject to this, too.
Also, you can only have one account per-email -- a disadvantage for some types of service, but admittedly probably a good thing for most.
Still, a lot of "forgot password" forms seem to be subject to this, too.
Also, you can only have one account per-email -- a disadvantage for some types of service, but admittedly probably a good thing for most.
This is similar to what Whatsapp (and some other apps) are doing. The only difference is that they are using phone numbers.
Loosely related. Why doesn't my web browser speak IMAP? If I could just enter my IMAP details into my browser settings, then it could silently IDLE on my Inbox, and inform my currently open http://example.com/ tabs when it receives email from an @example.com address.
You'd need to stick it behind a "This website wishes to be notified about email you receive from it" prompt.
But then you'd be able to do all sorts of clever stuff, including streamlining email validation checks and password resets.
Everybody has email. You could create a "Login with Email" function really easily on your website, and it would work for even more users than the "Login with Facebook" option.
You'd need to stick it behind a "This website wishes to be notified about email you receive from it" prompt.
But then you'd be able to do all sorts of clever stuff, including streamlining email validation checks and password resets.
Everybody has email. You could create a "Login with Email" function really easily on your website, and it would work for even more users than the "Login with Facebook" option.
why should the browsers speak every protocol?
Why should everything be done through the browsers, instead of better specialized programs?
Why should protocols designed for one problem be abused to solve other problems?
Why not use client certs instead?
Why should everything be done through the browsers, instead of better specialized programs?
Why should protocols designed for one problem be abused to solve other problems?
Why not use client certs instead?
"Why should the browsers speak every protocol?"
Why should browsers only speak HTTP? Who made that rule?
"Why should everything be done through the browsers, instead of better specialized programs?"
I never said "everything". "Better specialised programs"? What are you talking about? I'm talking about essentially being able to access email from a webpage. There is no "better specialised program" for doing that. I'm not talking about webmail. I'm talking about automated emails.
"Why should protocols designed for one problem be abused to solve other problems?"
What you call "abuse", I call "use".
"Why not use client certs instead?"
This has nothing to do with anything I've mentioned.
Why should browsers only speak HTTP? Who made that rule?
"Why should everything be done through the browsers, instead of better specialized programs?"
I never said "everything". "Better specialised programs"? What are you talking about? I'm talking about essentially being able to access email from a webpage. There is no "better specialised program" for doing that. I'm not talking about webmail. I'm talking about automated emails.
"Why should protocols designed for one problem be abused to solve other problems?"
What you call "abuse", I call "use".
"Why not use client certs instead?"
This has nothing to do with anything I've mentioned.
No better program to access email from anywhere than a browser. Have you heard about MUA-s?
Special mails should be used for authentication which would specially handled by the browser accessing your mailbox. This is abusing the mail protocols in my opinion.
Certificate based auth does have something to with this: an existing, partially supported solution for auth with no magic needed and no need to reinvent the wheel, with hacks and extra complications.
Special mails should be used for authentication which would specially handled by the browser accessing your mailbox. This is abusing the mail protocols in my opinion.
Certificate based auth does have something to with this: an existing, partially supported solution for auth with no magic needed and no need to reinvent the wheel, with hacks and extra complications.
"No better program to access email from anywhere than a browser. Have you heard about MUA-s?"
You understand the difference between accessing email and reading email? I'm talking about JavaScript accessing email, and acting on it. You're talking about a human being reading a message.
"Special mails should be used for authentication which would specially handled by the browser accessing your mailbox. This is abusing the mail protocols in my opinion."
We're already sending registration emails and password reset links. We currently have a system where the end user has to read an email and click a link in it. My suggestion would just allow us to skip the human part and let the JavaScript read it directly, with a perfect fallback built in for people who don't have their browsers configured to have access to their mail.
"Certificate based auth does have something to with this: an existing, partially supported solution for auth with no magic needed and no need to reinvent the wheel, with hacks and extra complications."
Even with certificate based auth, you'd still end up having email involved with registration or password resets in some fashion.
You understand the difference between accessing email and reading email? I'm talking about JavaScript accessing email, and acting on it. You're talking about a human being reading a message.
"Special mails should be used for authentication which would specially handled by the browser accessing your mailbox. This is abusing the mail protocols in my opinion."
We're already sending registration emails and password reset links. We currently have a system where the end user has to read an email and click a link in it. My suggestion would just allow us to skip the human part and let the JavaScript read it directly, with a perfect fallback built in for people who don't have their browsers configured to have access to their mail.
"Certificate based auth does have something to with this: an existing, partially supported solution for auth with no magic needed and no need to reinvent the wheel, with hacks and extra complications."
Even with certificate based auth, you'd still end up having email involved with registration or password resets in some fashion.
You know, there is one thing I don't ever want to happen: to have JS running in my browser access to my emails and act upon that automatically.
This has way too many security implications which don't make it a reasonable idea:
1. If sites can set up such action, then for that (I don't trust anybody's site to have any form of Access to my emails, except my mail providers per se),
2. if the browsers supply these actions, then for that cause (it would be exploited, or if it would be perfectly secure then it would also be perfectly useless).
This has way too many security implications which don't make it a reasonable idea:
1. If sites can set up such action, then for that (I don't trust anybody's site to have any form of Access to my emails, except my mail providers per se),
2. if the browsers supply these actions, then for that cause (it would be exploited, or if it would be perfectly secure then it would also be perfectly useless).
What security implications are there for a tab with the example.com origin to be able to read emails that you receive from an example.com email address? If it's behind a user access prompt, and the API requires that you specify the sender and recipient to monitor for? E.g:
navigator.notifyOnEmail({
from: 'sender',
to: 'recipientaddress',
}, (email) => {
/**
* This callback is run whenever a new email
* arrives from sender to recipientaddress.
* Assuming the browser has access to the
* recipientaddress account via IMAP, and
* assuming the sender address matches the
* browser origin and the user has given the
* origin permission to do this.
*/
});This leaks some information about the user's email address.
If you suspect one of your users is (say) "[email protected]", you can email [email protected].
Then, call notifyOnEmail({from: "[email protected]", to: "[email protected]"}) in every session on your site.
If Jean happened to be logged in, you've just identified her.
Spam filters would probably stop you brute-forcing the address of every visitor, but they might not help if you already have a good guess.
Requiring user approval will stop most of this, although it won't block queries like "are [email protected] and [email protected] the same person?".
Then, call notifyOnEmail({from: "[email protected]", to: "[email protected]"}) in every session on your site.
If Jean happened to be logged in, you've just identified her.
Spam filters would probably stop you brute-forcing the address of every visitor, but they might not help if you already have a good guess.
Requiring user approval will stop most of this, although it won't block queries like "are [email protected] and [email protected] the same person?".
> that you receive from an example.com email address?
How do you know that it really was from that email address?
How do you know that it really was from that email address?
Why do you think that matters? The sending site knows if it sent an email. The only thing it doesn't know is if you received it.
For an account registration or password reset email, the sending site will typically send you an email with a link inside it containing a secret token. If the JavaScript reads that it received an email with one of those links in it, it will post that information back to the server so the server can complete the function it started (account registration or password reset). Obviously, at that point, it will know if the token it was passed was the one it sent.
[edit] Also, as this system is being implemented from scratch, it could easily be decided that emails which don't pass DMARC for example will simply be ignored. Although I argue that is not necessary.
For an account registration or password reset email, the sending site will typically send you an email with a link inside it containing a secret token. If the JavaScript reads that it received an email with one of those links in it, it will post that information back to the server so the server can complete the function it started (account registration or password reset). Obviously, at that point, it will know if the token it was passed was the one it sent.
[edit] Also, as this system is being implemented from scratch, it could easily be decided that emails which don't pass DMARC for example will simply be ignored. Although I argue that is not necessary.
>Claim: With this setup, I can’t login easily with my LastPass, 1Password, etc browser extension.
>Rebuttal: This is a legitimate claim but it’s equally plausible to use an email integrated browser extension.
You could do that, but since this isn't a very common login scheme, these might not exist/be popular yet. Also, some of us use password managers that aren't browser integrated, and that further reduces the chance of me having a decent workflow for logging in to the site.
You could do that, but since this isn't a very common login scheme, these might not exist/be popular yet. Also, some of us use password managers that aren't browser integrated, and that further reduces the chance of me having a decent workflow for logging in to the site.
I'm glad to see this idea gaining traction. Passwords are a lazy solution that is often not needed. This type of login is ideal for websites and services that I might use less than 4 times a year.
I usually log into such services by doing a password reset every time I visit their service, and this technique at least removes the risk of password leaking.
I usually log into such services by doing a password reset every time I visit their service, and this technique at least removes the risk of password leaking.
This is commonly referred to as OTP and I have implemented it on a few recent projects. It is very common in Asian countries.
> Strawperson argument
Will this newspeak fury ever stop?
Will this newspeak fury ever stop?
For WordPress:
https://wordpress.org/plugins/passwordless-login/
https://wordpress.org/plugins/passwordless-login/
Just what Wordpress needs.
>Con: If database is compromised and passwords are reused across sites, then attackers might be able to access more sites
Never heard of hashing or TLS/SRP?
Never heard of hashing or TLS/SRP?
Those alone are not enough, a compromised database storing hashed passwords behind a service that uses TLS would still allow the attackers to try common passwords and hashing algorithms to see what passwords were used.
Password simplicity and reuse (especially across lots of people) is a major problem with password security in data breaches.
Password simplicity and reuse (especially across lots of people) is a major problem with password security in data breaches.
Salting is more relevant in this case.
So, nothing new since Kerberos?
> Rebuttal: This is possible in a password-ful login setup via the forgot password flow. Sometimes there are additional “security questions” but this is typically easily found information.
The difference here is that if/when someone uses their (unauthorized) access to my e-mail account to reset the password for this service, I will notice the next time I attempt to log in.
In an ideal world, one could prevent another from gaining access to an account, service, etc. When you can't prevent something, the next best thing is detection. Here, you've just removed the ability for a user to detect that their account is compromised -- and worse, you're trying to spin it like it's a good thing.
Also, greylisting.
Also, what if I wanted to log in to my account from my phone or a mobile device? I choose not to receive e-mail on my phone (only alerts/notifications) and I'm certainly not the only one.
It sounds like someone pitched this idea at a meeting ("Hey, I know! No passwords!") and the next 10 minutes was spent rationalizing all its problems away before declaring it as the one true solution.