5 ms·

Thou Shalt Use TLS?

8 points by gthank 16y ago

mike-cardwell 16y ago
I want to use untrusted networks like the WIFI provided at cafes. The only way I can currently do that securely is by using a VPN back to a network I trust. If all web traffic was encrypted end to end, I would be able to use untrusted networks without worrying about all my traffic being sniffed, my sessions being hijacked, and my ISP inserting adverts into the stream. People use SSH rather than Telnet for a very good reason.
- mooism2 16y ago
  If all web traffic is encrypted end to end, then either every web site needs to pay for a tls certificate, or your isp can insert adverts into the stream anyway.
  - agl 16y ago
    You can get certificates for free at startssl.com
    - mooism2 16y ago
      Any idea which browsers their root certificate is in? They don't seem to say in their faq.
  - mmagin 16y ago
    This is one of the problems with how TLS/SSL (as implemented in web browsers) conflates the seperate issues of authentication/trust and encryption.
    - briansmith 16y ago
      How would you implement encryption without authentication? What would stop an attacker from sending confidential information to them, if you allow them to intercept the traffic and supply their own encryption keys? You may be interested in the TLS DH_anon cipher suites and the numerous debates surrounding them.
briansmith 16y ago
I think Mark's comments here are representative of our whole industry's reluctance to recognize our end-users' right to privacy. Once privacy becomes a fundamental assumption, then TLS will be the bare minimum that is implemented. (In fact, most HTTPS stacks leak a lot of information, so additional measures are necessary.) Caching intermediary proxies won't even enter the debate because, while they are a significant optimization, they are completely at odds with this fundamental right. This is only a debate because, historically, we've completely ignored users' fundamental right to privacy due to performance concerns. Now, the folklore of SSL/TLS performance is largely out of date, but people are still building software using those outdated assumptions. Plus, people (especially at Intel, Sun/Oracle, and Google) are working right now to significantly reduce TLS overhead. IMO, it would be negligent to build a public system to be deployed in the next year that doesn't offer users basic protection from eavesdropping. Signed HTTP is no alternative to TLS. Most users want repudiation and privacy, not non-repudiation and non-privacy. In other words, they want their actions to be hidden, and they don't want people to have proof of their actions (signed HTTP requests) available after the fact. (TLS provides some level of repudiation because application data is encrypted using symmetric keys, so either side can forge the other side's traffic.) Mark has a good point that using TLS basically renders (caching) HTTP proxies useless. But, really, HTTP proxies have been holding back the internet for years. It is too easy to make a HTTP proxy that works for HTTP 1.0 and doesn't work for HTTP 1.1, and so that's what everybody has done. That is why we can't deploy HTTP pipelining on the Internet today, even though it was standardized more than a decade ago in the HTTP 1.1 specification. Bad proxies are the reason that every HTTP-based protocol has to be made to work using just GET and POST, instead of (or in addition to) using PUT and DELETE. I'm not going to miss HTTP proxies one bit.

8 points by gthank 16y ago

I want to use untrusted networks like the WIFI provided at cafes. The only way I can currently do that securely is by using a VPN back to a network I trust. If all web traffic was encrypted end to end, I would be able to use untrusted networks without worrying about all my traffic being sniffed, my sessions being hijacked, and my ISP inserting adverts into the stream. People use SSH rather than Telnet for a very good reason.

mooism2 16y ago
If all web traffic is encrypted end to end, then either every web site needs to pay for a tls certificate, or your isp can insert adverts into the stream anyway.
- agl 16y ago
  You can get certificates for free at startssl.com
  - mooism2 16y ago
    Any idea which browsers their root certificate is in? They don't seem to say in their faq.
- mmagin 16y ago
  This is one of the problems with how TLS/SSL (as implemented in web browsers) conflates the seperate issues of authentication/trust and encryption.
  - briansmith 16y ago
    How would you implement encryption without authentication? What would stop an attacker from sending confidential information to them, if you allow them to intercept the traffic and supply their own encryption keys? You may be interested in the TLS DH_anon cipher suites and the numerous debates surrounding them.

briansmith 16y ago

I think Mark's comments here are representative of our whole industry's reluctance to recognize our end-users' right to privacy. Once privacy becomes a fundamental assumption, then TLS will be the bare minimum that is implemented. (In fact, most HTTPS stacks leak a lot of information, so additional measures are necessary.) Caching intermediary proxies won't even enter the debate because, while they are a significant optimization, they are completely at odds with this fundamental right. This is only a debate because, historically, we've completely ignored users' fundamental right to privacy due to performance concerns. Now, the folklore of SSL/TLS performance is largely out of date, but people are still building software using those outdated assumptions. Plus, people (especially at Intel, Sun/Oracle, and Google) are working right now to significantly reduce TLS overhead. IMO, it would be negligent to build a public system to be deployed in the next year that doesn't offer users basic protection from eavesdropping. Signed HTTP is no alternative to TLS. Most users want repudiation and privacy, not non-repudiation and non-privacy. In other words, they want their actions to be hidden, and they don't want people to have proof of their actions (signed HTTP requests) available after the fact. (TLS provides some level of repudiation because application data is encrypted using symmetric keys, so either side can forge the other side's traffic.) Mark has a good point that using TLS basically renders (caching) HTTP proxies useless. But, really, HTTP proxies have been holding back the internet for years. It is too easy to make a HTTP proxy that works for HTTP 1.0 and doesn't work for HTTP 1.1, and so that's what everybody has done. That is why we can't deploy HTTP pipelining on the Internet today, even though it was standardized more than a decade ago in the HTTP 1.1 specification. Bad proxies are the reason that every HTTP-based protocol has to be made to work using just GET and POST, instead of (or in addition to) using PUT and DELETE. I'm not going to miss HTTP proxies one bit.