2 ms·
I think we are mostly in agreement, but the problem is that if the attacker can perform a MITM attack then a self signed certificate unless it's explicitly veri
by dogma1138 11y ago
I think we are mostly in agreement, but the problem is that if the attacker can perform a MITM attack then a self signed certificate unless it's explicitly verified is worth less.
I've said that we need to add additional levels of trust to the current PKI/SSL cert system preferably adding DNS records with the hash of the certificates and enforcing DNSSec globally this isn't the best solution but it's just enough of an easy hack that it will work for a good while.
What i do not want to see is people thinking that if we just spam SSL certs and stick them on everything it will some how help us, this is the classical definition of security theater.
When cost reduction isn't sufficient to thwart any effective surveillance, and worse using self signed certs trains people to just click accept, and free SSL certificates make it easier to gain valid certificates in a fraudulent manner things will only become worse not better.
While it's true that authentication and encryption are separate domains with how TLS/SSL is implemented currently they are tied by the hip, if you can't trust the authentication part of SSL the encryption part becomes meaningless to all but the most primitive types of passive sniffing.
If you want to separate them then lets kill TLS all together since PKI for the most part have failed, we can use other pure key exchange mechanism like well implemented DH without all the certificate nonsense.
- pdkl95 11y ago> self signed certificate unless it's explicitly verified is worth less. Of course it's worth less; the point is that it's not worthless, and has greater value than plaintext. > if the attacker can perform a MITM attack Then you communication is still protected from other attackers. > What i do not want to see is people thinking that if we just spam SSL certs and stick them on everything That's exactly the thing that needs to happen. PHK was right when he pointed dout that it would be trivial for Apache to simply generate a self-signed cert on first use if no other cert exists, removing a lot of the plaintext on the the internet. > it will some how help us "some how"? The point was it provides very good protection against passive analytics (aka "eavesdropping", "spying"). > this is the classical definition of security theater. Nonsense, nobody has ever claimed self-signed certificates were "safe", we only said they were better than plaintext. Why would you accept self-signed certificates as any kind of authentication? > self signed certs trains people to just click accept sigh That dialog may be the biggest reason we're still using plaintext. I needs to die yesterday, for the very reason you say: it gives people bad habits. That kind of scary popup needs to be reserved for invalid certificates, which actually warrant that kind of attention. Given BULLRUN's large budget and how easy it is to manipulate software projects[1], it is somewhat suspicious that there just happens to be a scary popup if someone tries to encrypt without caring about authentication. Nobody said anything about giving self-signed certificates any kind of trust. This is about removing plaintext. Unfortunately, there has been culture that badly confuses authentication with encryption (privacy) when they are entirely separate matters. > meaningless to all but the most primitive types of passive sniffing. Which is exactly the set of problems I'm talking about. Why do you want to continue allowing passive sniffing? We can trivially put a stop to this kind of nearly-free eavesdropping, but because those solutions don't address more complicated types of attacks (by design! use authentication when you can!), security people tend to be hostile to the idea. Meanwhile, as the search for a perfect solution continues, incredible amounts of passive sniffing is de facto accepted as the status quo. [1] https://www.youtube.com/watch?v=fwcl17Q0bpk https://www.youtube.com/watch?v=fwcl17Q0bpk
- dogma1138 11y agoWould you stop putting words into my mouth? I don't think that running plain text services is warranted, but i also do not want to see us rush into a horrible system that will not provide any sufficient protection against surveillance while forming the mindset of people into a false pretense of security. If we want security we should design a better system not force horrid implementation of already bad system. What you are doing is like putting the people that we should moderate the use of antibiotics and anti-malaria drugs because of resistances in the same camps as anti-vaccine nut jobs. I want a better secure Internet, not one that is flooded with out of control certificates and poor implementations which will only make things worse and much harder to untangle in the long run. And no self-signed certificates aren't better than plain text, unless i have received public key / thumbprint of the certificate through a channel I trust and could verify the sender of that information to me a self signed certificate is equivalent of plain text because the moment I have to accept that connection on good faith i cannot trust that the party which I am communicating it is the party I intended to communicate with or that no one else has the key. Under those circumstances I will not transfer any information that I would not transfer over plain text period, if you do so well do it at your own peril.
- pdkl95 11y ago> I don't think that running plain text services is warranted So stop the fear mongering over self-signed certs so plaintext can simply end. > while forming the mindset of people into a false pretense of security Then stop conflating self-signed certificates with a trust system. I'm talking about a replacement for plaintext; this doesn't affect the handling of authenticated certificates in any way. Since you still seem to not not understand this, let me be explicit: I'm saying we need protect against passive eavesdropping at the network level. I never said you should lie to the use* and claim their connection had any additional security. These fears about a "false pretense of security" are not relevant, as a move from plaintext to opportunistic encryption with self-signed certificates shouldn't be visible to the user. > I cannot trust that the party which I am communicating it is the party I intended to communicate with or that no one else has the key. Obviously. I never claimed otherwise. This isn't relevant, and if you want to protect against this threat model, you obviously need proper authentication (which I recommend whenever possible). Currently a lot communication on the internet has authentication risks AND allows anybody with a packet sniffer to log their data for next to no cost? You to believe it's a bad idea to convert those plaintext situations into something that has the same authentication risks while only betraying the contents to a single eavesdropper, who must be a MitM. There is no technical reason not to do this. As you obviously know because you referenced the certificate warning popup, self-signed certificates are trivially identifiable, so there is no reason to confuse them with normal certificates. Absolutely nothing needs to change in how this is presented to the user, so we aren't creating any sort of false sense of security. > Under those circumstances I will not transfer any information that I would not transfer over plain text period, Good for you. What does that have to do with the fact that much of the internet is still filled with the plaintext of people who don't understand this kind of technical topic? Why are you resisting shielding their communications from passive eavesdropping? This attitude that security must protect against all threats (or at least the treats YOU care about) is a major problem. What do you think is going to happen? It is magical thinking to assume that everyone will suddenly start using full, proper crypto (they can do that right now). We've been trying that for years and the result was a lot of people staying with plaintext.
- Karunamon 11y agoand worse using self signed certs trains people to just click accept This is a thing that happens right here, right now. We treat all of the following things as equivalent big scary warnings from the browser regardless of how bad they are or what the impact on security actually is. * Expiration date d+1day (security impact: NIL) * Name doesn't match the cert * CA not trusted * Certificate is self-signed * Certificate was generated using old/broken crypto Weve've got a couple decades of retraining to do because of this boneheaded, "wolf-crying" design.
- dogma1138 11y ago* Expiration date d+1day (security impact: NIL) > not much impact unless the certificate age is tied to the time it might take some one to refactor a key * Name doesn't match the cert > Can't trust that you communicating with the correct party, worth as much as plain text to me * CA not trusted > Unless you can explicitly verity the end certificate and the chain = plain text to me * Certificate is self-signed > same as the above * Certificate was generated using old/broken crypto > depending on the exact issue government agencies will most likely be able to break the encryption, if it's more easy attacks like BEAST or CRIME so can resourceful non-governmental organizations and individuals. This isn't just crying wolf we need a new and better system, one that ensures that the key exchange is secure, and one that ensures that who ever you speaking too is the actual intended precipitant and not a system which currently if one of those fails both fail, and more often than not can easily fail on both on it's own out of the gate.
- Karunamon 11y agoIt is exactly crying wolf because AFAIK, okay, cert expiration - you don't have to rekey when you go get a new one, you just get a new one. The benefit seems to be commercial, rather than technical. Name doesn't match: I've encountered a legitimate (as opposed to an obvious misconfiguration) error along these lines precisely zero times in well over a decade. CA untrusted: CA trust is worthless - the browser doesnt differentiate between "the person who holds the cert had control of the domain when we generated it" and "we made sure this person exists and checked their identity". (Unless you jump into the total scam that is EV certs...) Would you kindly stop banging on about "As good as plain text?". because as has been explained to you ad nauseum, in the worst case it becomes you vs one attacker who holds the key to the bogus cert, instead of you vs anyone on the local wifi network. The number of attackers is unbounded in the latter case, QED, it is strictly a better security posture. SSL as it exists today is fine for keeping a hacker out of your connection. It was not designed nor implemented for a post-Snowden world.