I'm obviously not communicating my point well. Let's try this:
A backdoor executes in a remote machine. It allows attackers to access that machine.
A rootkit executes in a "remote" privileged context. It allows attackers to access that privileged context. It's in this context that I refer to escalation; it allows the attacker in a non-priviledged context access to a privileged context; aka escalation. And yes, the actual escalation already happened in the past, when the rootkit was installed. However, a non-priviledged user is still gaining illicit access to a privileged context at the moment that the rootkit is utilized.
Also, at this point I think we're splitting semantic hairs that don't really matter, aside from pedantry.
Well I do appreciate you trying, but I disagree with you that it is pedantry that doesn't matter. This conversation is the best possible example of why we can't allow the corruption of previously well defined words - it causes confusion for no good reason.
A backdoor doesn't need to be remote and the user isn't necessarily an attacker. It is simply a secret method of access that the designer put in place, it isn't designed for end-user use. It is almost always security through obscurity, and it is always a bad idea. It can be activated in a variety of ways: port knocking, hardcoded passwords, preinstalled remote software, shorting ground to some magic pin, an undocumented serial terminal, etc.
A rootkit doesn't need to be remote and the user isn't necessarily an attacker. It doesn't need to have any functionality for user interaction - which means no "escalation" occurs (It could simply scan memory for passwords and log them to a file). It runs above user space, and can therefor be completely hidden (but it isn't always, see DTrace). It runs with the same privileges as the OS that it is part of. That is important to keep in mind, the rootkit becomes part of the running OS - that could mean any of the OSes running in your tower (CPU, HD firmware, BIOS, etc).
Your definitions work fine in a vacuum, but they quickly fall apart in real world usage. For example, by your definition: a remotely accessible privileged service is a rootkit, because an unprivileged internet user can interact with it - accessing data and executing code in the service's privileged context. 'sudo nginx' is not a rootkit.
> A rootkit doesn't need to be remote and the user isn't necessarily an attacker.
No one said a rootkit needs to be remote. (I used "remote" in quotes just to align it to the backdoor.) And in the context of security, it is definitely an attack. If there's not a user executing unauthorized commands, then it's simply installed and authorized software.
> It doesn't need to have any functionality for user interaction...
This is true, and I can see how some of my statements were maybe a bit more specific about this than they needed to be. The point is still to give an attacker a context with elevated permissions; it need not be an interactive context.
> It runs with the same privileges as the OS that it is part of.
This I still think is overly restrictive. I don't think running in ring 0/1/2 with the kernel and drivers is a necessary component; having "root" access such that it can invoke kernel functionality necessary to achieve its goals is sufficient. Now, it may use "root" access to modify kernel files and drivers, which is perhaps what you're referring to and where the line blurs and pedantry beings. If "root" access gives you unfettered access to the system, including modifying kernel executable files, then there is basically no difference between "root" and ring 0.
> For example, by your definition: a remotely accessible privileged service is a rootkit, because an unprivileged internet user can interact with it - accessing data and executing code in the service's privileged context. 'sudo nginx' is not a rootkit.
More pedantry. Clearly intended and authorized access to a service is just normal operation. This is why I'm very explicit about the usage being unauthorized and label the user an "attacker".
I think that we're as close as we'll get to agreeing - the main contention being the "attacker" part. I'd describe Symantic's data protection product as a rootkit, regardless of who installed it. I'd say the same for the LKMs that I've installed in honeypots.
I think you are spot on for rootkit, but you are absolutely wrong on backdoor.
You say "This conversation is the best possible example of why we can't allow the corruption of previously well defined words - it causes confusion for no good reason." when YOU(and others like you) are the one corrupting the meaning of backdoor.
Backdoor has meant for ages to be a way to access a computer/program while bypassing the normal authentication method, whether added by the designer or by someone else. You are trying to redefine it to mean only methods of bypassing normal authentication added by the designer. If you find it confusing that both types of backdoor are backdoors, then make up a new word that can be considered a subtype of backdoor don't try to coop an existing word and change its meaning.
> ... whether added by the designer or by someone else.
Your exception seems to hing on the word designer. I'd describe the individual responsible placing the backdoor as the designer. So if you place a modified version of /usr/sbin/sshd, then you've designed the backdoor for that system. I see no redefinition.
> Calling BO a backdoor is a major corruption of the word, as you loose the only word for describing intentionally weakened security - so that you may describe a thing which already has several more explicitly defining names: malware, trojan, dropper, etc.
Thinking in that context, it sounded like you were arguing further for the fact that backdoors should only be describing intentionally weakened security. Have you changed your mind about that?
No. Unlike a rootkit, context really matters in the case of a backdoor - not so much the implementation means. BO is no more a backdoor than vnc or sshd. Now if Dell decides to secretly package BO in their product line, then it is a backdoor.
> ...backdoors should only be describing intentionally weakened security.
I can't think of a backdoor that does not meet that description, do you have anything in mind?
A backdoor executes in a remote machine. It allows attackers to access that machine.
A rootkit executes in a "remote" privileged context. It allows attackers to access that privileged context. It's in this context that I refer to escalation; it allows the attacker in a non-priviledged context access to a privileged context; aka escalation. And yes, the actual escalation already happened in the past, when the rootkit was installed. However, a non-priviledged user is still gaining illicit access to a privileged context at the moment that the rootkit is utilized.
Also, at this point I think we're splitting semantic hairs that don't really matter, aside from pedantry.