OpenSSL security advisory: Infinite loop reachable when parsing certificates(openssl.org)
openssl.org
OpenSSL security advisory: Infinite loop reachable when parsing certificates
https://www.openssl.org/news/secadv/20220315.txt
24 comments
I wouldn't have thought this an issue if you're behind an AWS WAF or LB? (and assuming AWS is patched already)
This vulnerability affects parsing maliciously crafted certificates, so it will mostly affect clients. If your app is fetching data from a 3rd party and validating its certificate, it may be vulnerable, regardless of how you are fronting requests to your site.
> This vulnerability affects parsing maliciously crafted certificates, so it will mostly affect clients.
Actually, it is the opposite.
You seem to be unaware of the fact that servers do receive certificates from the clients which are then parsed.
Which is already mentioned in the advisory document:
Actually, it is the opposite.
You seem to be unaware of the fact that servers do receive certificates from the clients which are then parsed.
Which is already mentioned in the advisory document:
"Thus vulnerable situations include:
- TLS clients consuming server certificates
- TLS servers consuming client certificates <---- here
- Hosting providers taking certificates or private keys from customers
- Certificate authorities parsing certification requests from subscribers
- Anything else which parses ASN.1 elliptic curve parameters"That's not how the TLS handshake works. The TLS server must be configured to request a certificate from the client in order for the client to know that it needs to send a client certificate to the server, and that server-side configuration is disabled for ~99%+ of endpoints.
TLS server implementations should be aborting the TLS connection for violating the TLS Handshake state machine if a client attempts to send a client certificate when it wasn't requested.
So while this bug affects both clients and servers, 100% of clients are parsing the server's TLS cert during the TLS handshake, but less than ~1% of servers are parsing a client's certificate during a handshake.
TLS server implementations should be aborting the TLS connection for violating the TLS Handshake state machine if a client attempts to send a client certificate when it wasn't requested.
So while this bug affects both clients and servers, 100% of clients are parsing the server's TLS cert during the TLS handshake, but less than ~1% of servers are parsing a client's certificate during a handshake.
There is very little reason to DOS a client and a lot of reasons to attack servers.
There is a huge number of public facing services that implement mutual auth and all those are potentially vulnerable to DOS. While clients can just decide to not connect to a web service that causes their browser to malfunction (and why have you connected there in the first place?), services are usually not at liberty to ignore a client at this stage.
So yes, those servers that do request client certificate are targets and my point still stands that servers are much more affected than the clients.
What would be an affected client? You keep connecting to this infected website that causes your browser to die? Somebody embedded some tracking on their page that now points to an infected website? Everybody will just move on and it is hard to say you are very much affected by this problem.
Whereas if you are a service and you are affected you absolutely need to implement a fix.
There is a huge number of public facing services that implement mutual auth and all those are potentially vulnerable to DOS. While clients can just decide to not connect to a web service that causes their browser to malfunction (and why have you connected there in the first place?), services are usually not at liberty to ignore a client at this stage.
So yes, those servers that do request client certificate are targets and my point still stands that servers are much more affected than the clients.
What would be an affected client? You keep connecting to this infected website that causes your browser to die? Somebody embedded some tracking on their page that now points to an infected website? Everybody will just move on and it is hard to say you are very much affected by this problem.
Whereas if you are a service and you are affected you absolutely need to implement a fix.
correction: literally 99.99999% of endpoints. they use password^W SMS authentication instead. no seriously, how is the only one good thing about X.509 (authentication via public key) the only unused part (to be fair, if anyone used it, wed have a whole new 10 episodes of vuln disclosures).
> correction: literally 99.99999% of endpoints.
You made up a number with no grounding in reality because of your bias due to being "general public".
For corporate services it is actually quite common to use client certificates and mutual auth. Also popular with VPNs.
You might not be aware of this because corporations do not want to deal with people who do not know or can be forced to know how to generate signing request.
This is different when you control both the service and the users of the service and you have something valuable to protect.
As an example, I worked with credit card terminals and these used mutual auth with properly managed client certificates.
You wouldn't call DOS on all terminals and ATMS "insignificant".
You made up a number with no grounding in reality because of your bias due to being "general public".
For corporate services it is actually quite common to use client certificates and mutual auth. Also popular with VPNs.
You might not be aware of this because corporations do not want to deal with people who do not know or can be forced to know how to generate signing request.
This is different when you control both the service and the users of the service and you have something valuable to protect.
As an example, I worked with credit card terminals and these used mutual auth with properly managed client certificates.
You wouldn't call DOS on all terminals and ATMS "insignificant".
No, I was purely focused on public web. As for your corporate services, those are all insecure as hell despite whaever tech they use. Anything that's remotely hidden from public in any way historically was uncovered to have non stop, gaping, and obvious security holes, even after being corrected 1-5 times. This is a result of the way businesses are run as miniature reactionary states ("just ban people in the firewall brother. call the police").
For an intermediary like a WAF, load-balancer, or CDN to help it needs to do all of the certificate processing and keep your backend app from seeing the certificate at all. If you are terminating HTTPS at the edge and using HTTP internally that's very likely but if you're using an NLB instead of an ALB, it's not. Similarly, you'd also want to see if there's any application-level path where a certificate would be included in an uploaded file but that's far less common.
Depends if s2n is vulnerable to the same bug. Of course AWS doesn't protect you against application bugs so if your application has an SSRF vulnerability (presumably otherwise unexploitable) an attacker might be able to DoS you.
They don't mention S/MIME or things like signed binaries or SAML assertions but those seem plausibly vulnerable as well.
They don't mention S/MIME or things like signed binaries or SAML assertions but those seem plausibly vulnerable as well.
I think nearly everything at AWS uses s2n and not OpenSSL these days
Has s2n been fips certified? I thought some things (govcloud?) were still using openssl due to fips requirements.
I'm not sure what they are using right now, but they currently have the "AWS-LC Cryptographic Module" listed as being in process for validation which may be sufficient for s2n.
I know there are other AWS features/services that are lined up behind that validation for FIPS support.
I know there are other AWS features/services that are lined up behind that validation for FIPS support.
Dare I ask if LibreSSL is affected?
yes, fixed here: https://cvsweb.openbsd.org/src/lib/libcrypto/bn/bn_sqrt.c?re...
they removed a lot of cruft, but this is needed
they removed a lot of cruft, but this is needed
how is it even possible to have the question of whether there can be an infinite loop during establishing a data channel? just make the user provide the public key and do a handshake [1]. X.509 is literally a non-practical academic idea like CSS. software is bad because everyone accepts bad ideas.
1. yes, having things done the right way where they need to be done makes sense. think of the user after. for instance, you could just make his stupid cloud device aggregate all the keys he needs automatically
1. yes, having things done the right way where they need to be done makes sense. think of the user after. for instance, you could just make his stupid cloud device aggregate all the keys he needs automatically
> X.509 is literally a non-practical academic idea
Non-practical but https has been running on it for decades now?
Yeah there are issues with it and as someone that's worked with it a lot at the profile design and issuing level, I think it's high time we came up with a replacement and ditched ASN.1 entirely. But ...
It's hard to see it as a non-practical, academic idea when it's out there, deeply embedded in the internet, everywhere.
> just make the user provide the public key
Doesn't solve the problems X.509 addresses. There's a lot more going on than a method to exchange keys.
Non-practical but https has been running on it for decades now?
Yeah there are issues with it and as someone that's worked with it a lot at the profile design and issuing level, I think it's high time we came up with a replacement and ditched ASN.1 entirely. But ...
It's hard to see it as a non-practical, academic idea when it's out there, deeply embedded in the internet, everywhere.
> just make the user provide the public key
Doesn't solve the problems X.509 addresses. There's a lot more going on than a method to exchange keys.
> It's hard to see it as a non-practical, academic idea when it's out there, deeply embedded in the internet, everywhere.
It is non-practical, has no application in reality, and the direct consequence of using such a tool is massive security problems being disclosed every year.
> Doesn't solve the problems X.509 addresses.
What problem does X.509 address? Not only is it needlessly centralized, but your connection can be intercepted as long as one single CA anywhere is compromised. The "solution" to this is to log what certs each client got, reducing their privacy because this would allow us to detect when an attack happened, after it already happened (followed by a barrage of petty arguments like "w-well, it would disincentivize an attacker"). Also, a domain name that sounds kind of like the company you want to connect to is not proof that you have the right domain name. You still have to ask them for the right domain name, at which point you may as well have just received their public key. The alternative is to do this dog shit thing you guys keep doing where you complain about Google search not being regulated enough and enact more ad-hoc laws many of which backfire. Imagine search engines having to do work to find the "real" domain for some company. This is 10th level stupidity, we could not have got here without several other generations of stupidity becoming normalized prior.
It is non-practical, has no application in reality, and the direct consequence of using such a tool is massive security problems being disclosed every year.
> Doesn't solve the problems X.509 addresses.
What problem does X.509 address? Not only is it needlessly centralized, but your connection can be intercepted as long as one single CA anywhere is compromised. The "solution" to this is to log what certs each client got, reducing their privacy because this would allow us to detect when an attack happened, after it already happened (followed by a barrage of petty arguments like "w-well, it would disincentivize an attacker"). Also, a domain name that sounds kind of like the company you want to connect to is not proof that you have the right domain name. You still have to ask them for the right domain name, at which point you may as well have just received their public key. The alternative is to do this dog shit thing you guys keep doing where you complain about Google search not being regulated enough and enact more ad-hoc laws many of which backfire. Imagine search engines having to do work to find the "real" domain for some company. This is 10th level stupidity, we could not have got here without several other generations of stupidity becoming normalized prior.
Both X.509 and CSS have had wild commercial success. What are you talking about? They're literally practiced so how can they be impractical?
>provide the public key
Which format?
What algorithim?
How do you prevent it from being sent to the wrong domain and enabling MITM phishing?
X.509 has a lot of issues but they've thought through a number of things. Whatever you replace it with would have similar problems. Even if you "just sent a key" it would have to be parsed in some way (Say it's RSA, you'd still have to pull the module and exponent out before you could do anything with it, if the channel is text you'd probably do some base64 decoding etc.)
Which format?
What algorithim?
How do you prevent it from being sent to the wrong domain and enabling MITM phishing?
X.509 has a lot of issues but they've thought through a number of things. Whatever you replace it with would have similar problems. Even if you "just sent a key" it would have to be parsed in some way (Say it's RSA, you'd still have to pull the module and exponent out before you could do anything with it, if the channel is text you'd probably do some base64 decoding etc.)
Standards are good.
> How do you prevent it from being sent to the wrong domain and enabling MITM phishing?
TOFU. Petnames.
> it would have to be parsed in some way
Encoding a key is trivial, even with an ad-hoc serialization format. Nothing near what X.509 does.
> How do you prevent it from being sent to the wrong domain and enabling MITM phishing?
TOFU. Petnames.
> it would have to be parsed in some way
Encoding a key is trivial, even with an ad-hoc serialization format. Nothing near what X.509 does.
Is BoringSSL affected?
Not to the same degree.
See: https://github.com/google/boringssl/commit/c7a3c46574e7fc323...
In particular: " In OpenSSL, this loop resulted in a DoS vulnerability, CVE-2022-0778. BoringSSL is mostly unaffected by this. In particular, this case is not reachable in BoringSSL from certificate and other ASN.1 elliptic curve parsing code. Any impact in BoringSSL is limited to:
- Callers of EC_GROUP_new_curve_GFp that take untrusted curve parameters - Callers of BN_mod_sqrt that take untrusted moduli"
In particular: " In OpenSSL, this loop resulted in a DoS vulnerability, CVE-2022-0778. BoringSSL is mostly unaffected by this. In particular, this case is not reachable in BoringSSL from certificate and other ASN.1 elliptic curve parsing code. Any impact in BoringSSL is limited to:
- Callers of EC_GROUP_new_curve_GFp that take untrusted curve parameters - Callers of BN_mod_sqrt that take untrusted moduli"
So it’s a timing attack?
:D :D :D
:D :D :D