5 ms·

Wll thats pretty awesome :) I'm hoping we don't have to patch everything tomorrow but.. ;) On a side note - a lot of libs start to implement their own crypto b

by zobzu 12y ago

Wll thats pretty awesome :) I'm hoping we don't have to patch everything tomorrow but.. ;) On a side note - a lot of libs start to implement their own crypto because "OpenSSL code sux its terrible and buggy!" impression. This scares me. These won't get much exposure and will most likely be full of implementation bugs that nobody will ever review - except bad guys and the NSA.

derefr 12y ago
OpenSSL isn't the be-all-and-end-all of crypto. Simple crypto can be done simply, as it is in e.g. Tarsnap. If all you need is to pair your clients with your servers, a little bit of glue on top of NaCl is a whole lot easier to trust than the whole TLS ecosystem. What OpenSSL is, is a magic wand that can set up encrypted tunnels between peers that don't have pre-existing mutual key pairing, by making use of well-known X.509 certificates and the CA infrastructure. Don't try to make that kind of wand yourself; 99.999% of such wands are really misidentified Wands of +5 Foot Shooting.
- toast0 12y ago
  Given that we know OpenSSL is a Wand of +3 Foot Shooting[1], seeing what kind of modifier you can roll doesn't seem so terrible. :) [1] I rolled my own CA bundle this week, turns out I need to include 4 1024-bit certs because the endpoints I hit are chaining to them for legacy clients, and OpenSSL didn't understand to stop when it got to a cert it knew until 1.0.2 which was released in January. Given what happened the last time I upgraded OpenSSL versions, I'm not inclined to upgrade right away. Seriously, the week after we upgraded to OpenSSL 1.0.1, heartbleed came out; serves us right for wanting to support TLS 1.1 and 1.2 and PFS.
  - yuhong 12y ago
    One of the more unfortunate cases is the Equifax one which expires in 2018.
nailer 12y ago
> On a side note - a lot of libs start to implement their own crypto because "OpenSSL code sux its terrible and buggy!" impression. I don't think they'd do that - it's be far easier to grab libressl or boringssl. If you're aware that openssl is buggy, you're probably aware that writing your own crypto when you're not a cryptographer is worse.

by zobzu 12y ago

derefr 12y ago

OpenSSL isn't the be-all-and-end-all of crypto. Simple crypto can be done simply, as it is in e.g. Tarsnap. If all you need is to pair your clients with your servers, a little bit of glue on top of NaCl is a whole lot easier to trust than the whole TLS ecosystem. What OpenSSL is, is a magic wand that can set up encrypted tunnels between peers that don't have pre-existing mutual key pairing, by making use of well-known X.509 certificates and the CA infrastructure. Don't try to make that kind of wand yourself; 99.999% of such wands are really misidentified Wands of +5 Foot Shooting.

toast0 12y ago
Given that we know OpenSSL is a Wand of +3 Foot Shooting[1], seeing what kind of modifier you can roll doesn't seem so terrible. :) [1] I rolled my own CA bundle this week, turns out I need to include 4 1024-bit certs because the endpoints I hit are chaining to them for legacy clients, and OpenSSL didn't understand to stop when it got to a cert it knew until 1.0.2 which was released in January. Given what happened the last time I upgraded OpenSSL versions, I'm not inclined to upgrade right away. Seriously, the week after we upgraded to OpenSSL 1.0.1, heartbleed came out; serves us right for wanting to support TLS 1.1 and 1.2 and PFS.
- yuhong 12y ago
  One of the more unfortunate cases is the Equifax one which expires in 2018.

nailer 12y ago

> On a side note - a lot of libs start to implement their own crypto because "OpenSSL code sux its terrible and buggy!" impression. I don't think they'd do that - it's be far easier to grab libressl or boringssl. If you're aware that openssl is buggy, you're probably aware that writing your own crypto when you're not a cryptographer is worse.