3 ms·

> will see a problem immediately because it's untrusted which is weird You might be right, but I don't think your version of "immediately" matches up with the

> will see a problem immediately because it's untrusted which is weird You might be right, but I don't think your version of "immediately" matches up with the reality of code. How many cert library function calls do you think happen before a TLS client is able to decide that there is no trusted path to the cert's authority and decide that it's "weird"? Hundreds or thousands wouldn't surprise me -- you've got to parse a file format, follow links -- if the vulnerability's in one of those functions, then there's your exploit. You even have to go out to the filesystem, check whether the self-signed CA is installed in the OS trust store, which would make it trusted. It's really easy for me to imagine such a vuln somewhere in cert trust verification.

tialaramex 7y ago
I meant on unaffected platforms, say, Firefox on Linux. It'd be weird if I came to, say, Hacker News and now it had this untrusted cert (which is actually a worm attacking Windows PCs) instead of a normal one. I agree that on an affected platform all bets are off, but deploying to a web server means you don't easily get to pick who your visitors are.
- cjbprime 7y ago
  Oh yes, that's true. (Though now I wonder if cert negotiation can be dependent on useragent under SNI. It could certainly be dependent on TCP fingerprint characteristics.) Agree with you that codesigning certs are a more likely problem space for this bug.
  - atxlurker 7y ago
    I don’t think this is ultimately the vulnerability, but even without SNI, it’s arbitrary to dynamically link to a file, script, png, etc... hosted on a sub domain with the bad cert.
    - cjbprime 7y ago
      Oh yeah! So it's like: * client asks for cert * you give it to them * client tells you the page they want and their useragent * if you think they're vulnerable based on what you've learned about them, you add <img src="https://vulnerable.subdomain/"> https://vulnerable.subdomain/"> to the response. Neat suggestion. Thanks! Agree we've moved well outside of tomorrow's likely actual vuln.
- ikeboy 7y ago
  Just load a hidden image from a subdomain, silently fails if on an unaffected platform.
  - Thorrez 7y ago
    And you could also avoid even creating the iframe if the User Agent isn't Windows.
ct520 7y ago
Ugh man. I seen custom rolled where it frys to follow links building cert Chain validation. This sounds like a nightmare

tialaramex 7y ago

I meant on unaffected platforms, say, Firefox on Linux. It'd be weird if I came to, say, Hacker News and now it had this untrusted cert (which is actually a worm attacking Windows PCs) instead of a normal one. I agree that on an affected platform all bets are off, but deploying to a web server means you don't easily get to pick who your visitors are.

cjbprime 7y ago
Oh yes, that's true. (Though now I wonder if cert negotiation can be dependent on useragent under SNI. It could certainly be dependent on TCP fingerprint characteristics.) Agree with you that codesigning certs are a more likely problem space for this bug.
- atxlurker 7y ago
  I don’t think this is ultimately the vulnerability, but even without SNI, it’s arbitrary to dynamically link to a file, script, png, etc... hosted on a sub domain with the bad cert.
  - cjbprime 7y ago
    Oh yeah! So it's like: * client asks for cert * you give it to them * client tells you the page they want and their useragent * if you think they're vulnerable based on what you've learned about them, you add <img src="https://vulnerable.subdomain/"> https://vulnerable.subdomain/"> to the response. Neat suggestion. Thanks! Agree we've moved well outside of tomorrow's likely actual vuln.
ikeboy 7y ago
Just load a hidden image from a subdomain, silently fails if on an unaffected platform.
- Thorrez 7y ago
  And you could also avoid even creating the iframe if the User Agent isn't Windows.

ct520 7y ago

Ugh man. I seen custom rolled where it frys to follow links building cert Chain validation. This sounds like a nightmare