| |||||||||||||
NVIDIA to provide documentation for NouveauNVIDIA to provide documentation for NouveauPosted Sep 24, 2013 16:38 UTC (Tue) by flussence (subscriber, #85566)In reply to: NVIDIA to provide documentation for Nouveau by yokem_55 Parent article: NVIDIA to provide documentation for Nouveau
I thought they'd kept it around as a basic userspace-modesetting driver, with the intent being to allow users to boot to a basic desktop and install their "real" driver. It seems the last release was over 12 months ago though so I assume you're correct.
(Log in to post comments)
NVIDIA to provide documentation for Nouveau Posted Sep 24, 2013 17:01 UTC (Tue) by mjg59 (subscriber, #23239) [Link]
Userspace modesetting won't work under most distributions when Secure Boot is enabled.
NVIDIA to provide documentation for Nouveau Posted Sep 24, 2013 20:42 UTC (Tue) by rahvin (subscriber, #16953) [Link]
Is there a quick explanation why? That seems very strange to me.
NVIDIA to provide documentation for Nouveau Posted Sep 24, 2013 20:44 UTC (Tue) by mjg59 (subscriber, #23239) [Link]
Userspace modesetting means giving userspace the ability to arbitrarily program hardware with a DMA engine which means now userspace can turn your kernel into a bootloader and that's kind of the point.
NVIDIA to provide documentation for Nouveau Posted Sep 24, 2013 22:38 UTC (Tue) by rahvin (subscriber, #16953) [Link]
So a mode setting kernel driver could potentially allow a user (might require privilege escalation depending on setup I suppose) to boot arbitrary software?
NVIDIA to provide documentation for Nouveau Posted Sep 24, 2013 22:43 UTC (Tue) by mjg59 (subscriber, #23239) [Link]
Kernel drivers are supposed to prevent userspace performing any kind of operation that would touch arbitrary kernel memory. One that didn't do so would clearly be a mechanism to boot arbitrary software, yes.
NVIDIA to provide documentation for Nouveau Posted Sep 25, 2013 7:23 UTC (Wed) by paulj (subscriber, #341) [Link]
Given there are Secure Boot bootloaders that allow one to load arbitrary operating systems, then what does it matter if such an OS then allows privileged user-space to run?
Unless you mean that, ultimately, Secure Boot is intended to disallow users from running arbitrary code on the hardware, and allow only code limited to approved distributors to be run?
NVIDIA to provide documentation for Nouveau Posted Sep 25, 2013 9:20 UTC (Wed) by lsl (subscriber, #86508) [Link]
Those boot loaders are supposed to perform a check for physical presence. If there's someone right in front of the physical hardware he could already smash it to pieces or install some hardware bug, so why not let him boot an arbitrary kernel?
The kernel can't reasonably perform such checks before letting an user space driver access the graphics hardware (and with it, the entire system).
NVIDIA to provide documentation for Nouveau Posted Sep 25, 2013 12:13 UTC (Wed) by farnz (guest, #17727) [Link] It all depends on the bootloader you're using, though. I can easily imagine that Red Hat and SuSE enterprise customers are planning to use bootloaders that check the signature of the next thing they load, so that there's a chain of blame all the way through the system from power on to running kernel that lets them trace the sign-off on any exploitable code (and thus work out if there's something they could do differently to avoid an exploit of their employee's CAD workstations or whatever). I'd expect such customers to follow James Bottomley's guide to replacing all keys trusted by the firmware with ones under the customer's control, so it wouldn't matter that you can get the LF bootloader that's been signed by Microsoft's key, as that key is not trusted by the firmware.
NVIDIA to provide documentation for Nouveau Posted Sep 25, 2013 15:18 UTC (Wed) by PaXTeam (subscriber, #24616) [Link]
that blog url seems to be dead but you can still read the entry on the main blog page. more importantly though, why would you trust the firmware to manage your keys properly when you don't trust the keys that come with it? the source of these is the same manufacturer that you either trust or don't trust, but you can't have it both ways ;).
NVIDIA to provide documentation for Nouveau Posted Sep 26, 2013 18:48 UTC (Thu) by farnz (guest, #17727) [Link] That's why I referred to a chain of blame, not a chain of trust - there's a good chance that the firmware isn't secure enough to be trustworthy. If that's the case (e.g. because it executes code signed by a Microsoft key when you've replaced that with your own key), the chain of blame ends at the firmware, and hence at the manufacturer.
NVIDIA to provide documentation for Nouveau Posted Sep 26, 2013 19:45 UTC (Thu) by PaXTeam (subscriber, #24616) [Link]
my comment was about your second paragraph, not the first one. you pointed at a blog post that explains how to replace untrusted keys and then you said that this way one would avoid trust issues coming from MS signed boot loaders. but this statement assumes that the key replacement process is trusted - whereas you start with the assumption that the UEFI code doing the replacement is not which is a contradiction ;). or in other words: your advice of replacing untrusted keys but keeping untrusted UEFI code around doesn't make sense.
NVIDIA to provide documentation for Nouveau Posted Sep 26, 2013 20:06 UTC (Thu) by mjg59 (subscriber, #23239) [Link]
You replace the certainty of Microsoft (or anyone with access to their keys) being able to run code on your system with a possibility that they're able to do so. If you don't trust Microsoft (or anyone who might be able to force them to hand over keys), that still seems like an improvement. You're obviously at the mercy of your firmware vendor, but that's always going to be true.
NVIDIA to provide documentation for Nouveau Posted Sep 26, 2013 21:35 UTC (Thu) by PaXTeam (subscriber, #24616) [Link]
what exactly is being improved? the user's sense of security or actual security? because in the world of security only the latter matters (and yes, i don't exactly have a high opinion about much of the security industry ;) and i don't see how that's possible with replacing keys (=trusting that they actually get replaced) all the while not trusting the very code responsible for, among others, the key replacement.
NVIDIA to provide documentation for Nouveau Posted Sep 26, 2013 21:42 UTC (Thu) by mjg59 (subscriber, #23239) [Link]
You can't verify that your CPU will perform all cryptographic operations correctly, so what's the point in any cryptography at all? Even if you believe that some sufficiently powerful actor can still compromise your system via a firmware flaw or backdoor, removing Microsoft's keys means that the majority of existing signed binaries won't run on your system. That means an attacker won't be able to use a flaw in one of the thousands of Microsoft-signed binaries to attack your system.
NVIDIA to provide documentation for Nouveau Posted Sep 29, 2013 0:42 UTC (Sun) by PaXTeam (subscriber, #24616) [Link]
i don't understand, since when are CPU level backdoors part of the threat model of UEFI/Secure Boot?
also you keep saying what the removal of the MS key means but you don't explain why the UEFI code to accomplish this can be so much more trusted than trusting what the MS key will sign.
NVIDIA to provide documentation for Nouveau Posted Sep 29, 2013 5:03 UTC (Sun) by mjg59 (subscriber, #23239) [Link]
Security tends to be built on the assumption that lower layers aren't actively trying to screw with you. We trust ssh because the layers we can examine appear to be correct, even though the CPU could potentially have determined that we were generating a host key and given us one from a list maintained by the NSA.
Secure Boot is the same. It's possible that the UEFI code *is* hostile to us. Secure Boot is irrelevant here - hostile firmware can modify our OS even without Secure Boot being involved. But the firmware is produced by one of a range of different firmware vendors, and has in turn been modified by our system vendor. The probability that the NSA (or any other state agency) has a backdoor that exists in every firmware implementation is slim - there's too many people who have access to the source code (including random board vendors in Taiwan) to be able to guarantee secrecy.
However, the security of the firmware is irrelevant if you don't believe that Microsoft's key can be trusted. If your firmware trusts Microsoft's key then the NSA (or some other hostile body with access to Microsoft's key) can sign a bootloader that your firmware will trust and then use that to compromise the rest of your OS. Removing Microsoft's key removes that avenue of attack. Someone who wants to compromise your system can no longer simply go to Microsoft. They instead have to identify the specific firmware that you're running, locate a back door or vulnerability and then deliver an attack that's specifically tailored to you. Removing Microsoft's key doesn't make it impossible for someone to attack your boot process, but it does make it harder. That seems like an improvement in security.
The source code to AMI's Secure Boot implementation is out in the wild, thanks to Jetway leaving it on an insecure FTP site. I'm sure someone could do a reasonable audit.
NVIDIA to provide documentation for Nouveau Posted Oct 1, 2013 13:07 UTC (Tue) by PaXTeam (subscriber, #24616) [Link]
1. the CPU cannot determine anything of this sort in general, it'd be equivalent to solving the halting problem (cf. Rice's theorem). if you think otherwise i think the whole world would be dying to know how to pull off such a feat ;).
2. Secure Boot (and its key management in particular) was proposed here to solve the problem of not booting untrusted code, i.e., the entire trust was placed in these keys all the while ignoring the fact that the UEFI code is in the same 'trust domain'. i'm glad you actually agree with me that separating the two is non-sensical.
3. your statement about the NSA's capabilities requires justification. some food for thought: backdooring doesn't require source code (and especially not permanent source changes). backdooring doesn't require the cooperation of firmware producers (also the food chain is longer than that). backdooring can be targeted. etc.
4. the security of the firmware becomes relevant *exactly* when it comes to key management. how else would you know whether the firmware does as asked (removes the MS key for good) or just pretends to do so but still accepts some blob containing some secret magic signed by the supposedly revoked key? one more time: you *cannot* separate the trust in UEFI code from the trust in the keys it manages. so any 'improvement in security' based on removing the MS key is just a feeling at best, not actual security.
5. availability of some source code on the net has exactly zero relevance to the code running as your machine's UEFI firmware. you can audit it all you want, it won't give you all the backdoors that could exist in the actual binary stored in flash (and to be honest, it'd be rather dumb to risk exposing anything of this sort in this kind of source code anyway).
NVIDIA to provide documentation for Nouveau Posted Oct 1, 2013 15:26 UTC (Tue) by mjg59 (subscriber, #23239) [Link]
1. It doesn't need to know in the general case - if you know your target's crypto library then you just push a microcode update that recognises the specific series of instructions it executes in that path.
2. Secure Boot places the root of trust in the firmware, so yes, if you can't trust the firmware then you can't trust anything above that. But like I said, that's true even without Secure Boot.
3. There are at least 4 common firmware implementations that were independently developed. They're built with different compilers. This code is distributed to a much larger number of board vendors, each of whom then rebuilds it with their own choice of compiler.
Could a security agency compromise all of these? It's theoretically possible, but it doesn't seem like the easiest avenue of attack. The number of firmware implementations is larger than the number of operating systems that run on top of them - backdoor Windows and Linux and you have the same benefits for much less effort. That's not to say that firmware is secure and trustworthy, or that individuals won't be targeted, just that Secure Boot probably isn't the easiest avenue of attack.
4. Why would you do it that way? It'd be far too easy for a user to verify - remove the Microsoft key, check whether Windows boots. As you suggest, the obvious thing to do would be to have some additional embedded key that's checked regardless of whether or not the Microsoft key is present. So sure, removing the Microsoft key doesn't secure you against a security agency who's managed to compromise your firmware. But it *does* protect you against attacks where someone's found an exploit in something that was signed by Microsoft. Reducing your attack surface is an improvement in security.
5. So someone should just rebuild that source code and check whether it matches the binaries that Jetway ship. Matching obviously doesn't guarantee the absence of a backdoor, but a failure to do so would be a pretty strong indication that something's up.
NVIDIA to provide documentation for Nouveau Posted Oct 1, 2013 21:14 UTC (Tue) by PaXTeam (subscriber, #24616) [Link]
1. matching the crypto library code (something microcode cannot do anyway) won't do any good because that code can occur in several other contexts than your target sshd and it can also be executed for arbitrary other purposes and the CPU can't tell which is which without solving the halting problem.
second, if you have the ability to force arbitrary microcode updates on your target then you're already root there for all intents and purposes, so you have easier ways to backdoor their system.
2. but according to local wisdom here Secure Boot is supposed to give us, well, a secure boot process if only we used our own keys. turns out we're no better off than trusting trust once again.
3. have you seen the Snowden leaks? do you know how many companies got compromised and/or compelled into cooperation by the NSA and other agencies? so yes, i'm not exactly impressed by '4 common firmware implementations', it's a very *small* industry actually (and what does building with different compilers matter? nothing?). in fact, if i were the NSA, i would have my men planted there long ago and the 'how do i compromise them' question would simply become 'when and what do you want me to do?'. as for which is more numerous, i bet there're more different vmlinux and heck, even ntoskrnl images out there than UEFI firmware updates.
4. that's not how it'd work, obviously a Windows boot despite (supposedly) not having the MS key would be a dead giveaway. however it's possible to still accept code signed by the (supposedly removed) MS key if there's an additional condition - no need for an extra key at all, just some embedded secret that only the backdoor owner would know (and whoever else reverse engineers it of course). in fact MS would not even have to be complicit here, they'd just sign such an image in good faith without being aware of the secret payload that'd trigger the 'accept this despite being signed by the removed MS key' logic in the backdoored UEFI firmware.
as for 'an exploit in something that was signed by MS' you probably didn't mean that, but rather an exploitable bug as i find it unlikely that an actual exploit embedded in the to-be-signed code would pass their processes whereas bugs (exploitable or not) slip by all the time. with this understanding it seems now that the worry about the MS key is not that someone abuses it to sign something bad (which is what i was going about before) but that otherwise well meaning code gets signed and then exploited due to its bugs. this is a legitimite concern and removing the MS key would indeed help here... except this problem applies equally well to any other key as well and considering where MS stands with its SDLC and other processes in the industry (read: mostly above everyone else) i think the users are worse off trusting anybody else's keys, key signing processes and software development capabilities than those of MS. perhaps a sad piece of truth for free software but as far as i'm concerned, this is the reality. so if the advice here is still that users should become their own CA and use their own keys for Secure Boot then my bet is that the likes of the NSA (and even less resourceful actors) will still have a field day owning their systems.
5. just a few points above you were bringing up different compilers as one of the reasons why universal backdooring would be so much harder for powerful and skilled actors (it isn't, but that's not the point here) and now you're suggesting that much less resourceful end users should try to gain confidence in their firmware by trying to second guess the exact toolchain and build environment their vendor used. sorry, this doesn't add up to a good argument ;).
NVIDIA to provide documentation for Nouveau Posted Oct 1, 2013 22:23 UTC (Tue) by raven667 (subscriber, #5198) [Link]
You guys seem to be talking about two different things and I'm not sure any effective communication is going on here. In one case you have a general statement that firmware, especially firmware that is capable of being updated, can harbor persistent threats, backdoors, etc. EFI provides standard capabilities, like network support, which may make these kinds of threats easier to design or more useful. These risk of these kinds of threats though are unchanged whether SecureBoot(tm) exists or not and so to rope in Secure Boot into the discussion is to muddy the waters about two different threats/risks and two different responses.
NVIDIA to provide documentation for Nouveau Posted Oct 2, 2013 18:28 UTC (Wed) by mjg59 (subscriber, #23239) [Link]
I do wonder if we're somehow talking past one another. We appear to agree that there's no fundamental reason to trust firmware - it's possible that it's deliberately backdoored, and it's almost certainly buggy in exploitable ways. When we build any kind of secure system we have to assume that the firmware isn't actively malicious, in the same way that we have to assume that everything else under our stack isn't actively malicious.
But Secure Boot isn't about protecting us from the firmware. It never has been. It's about limiting the set of objects that your firmware will run. Now obviously if a sufficiently powerful actor has leaned on your firmware vendor then they may be able to run arbitrary code on your firmware, but why bother? They could just have the firmware include some SMM code that'd trigger in specific circumstances and modify arbitrary addresses in your running OS.
Obviously Secure Boot does nothing to protect you against such actors, but that doesn't mean it adds nothing to security. Microsoft have signed literally hundreds of binaries. Fedora have signed significantly fewer than that, and all the ones signed by Fedora have also been signed by Microsoft. Removing the Microsoft key and only trusting the Fedora one clearly improves security, if only because you'll no longer be able to boot the Ubuntu grub that'll happily boot unsigned kernels. Perhaps you weren't aware that Microsoft is effectively the global signing authority for UEFI binaries?
NVIDIA to provide documentation for Nouveau Posted Sep 25, 2013 16:05 UTC (Wed) by mjg59 (subscriber, #23239) [Link]
The machine owner should (through a demonstration of physical presence) be able to do whatever they want. Someone who doesn't have physical presence shouldn't be able to modify the booted kernel, even if they have privileged access.
NVIDIA to provide documentation for Nouveau Posted Sep 25, 2013 23:05 UTC (Wed) by dlang (✭ supporter ✭, #313) [Link]
this works for personal devices like laptops, but not for servers
NVIDIA to provide documentation for Nouveau Posted Sep 25, 2013 23:07 UTC (Wed) by mjg59 (subscriber, #23239) [Link]
Yes, which is why we have IPMI.
NVIDIA to provide documentation for Nouveau Posted Sep 26, 2013 2:32 UTC (Thu) by dlang (✭ supporter ✭, #313) [Link]
two things here.
not all systems that are deployed as servers have IMPI
trading a system that requires root access to do thing for a system that puts it's console on the network exposted to attackers (especially where that console is 'secured' by vendor proprietary code) doesn't seem like a win to me.
IPMI Posted Sep 26, 2013 15:06 UTC (Thu) by tialaramex (subscriber, #21167) [Link]
"not all systems that are deployed as servers have IMPI"
Sure, it's totally acceptable to choose no lights out management if you have 24/7 hands-on. The 24/7 hands-on people are physically present and meet that constraint.
The practice of calling a cheap desktop PC in a closet a "server" has plenty of other problems long before you get to remote management.
We have drifted far off topic.
NVIDIA to provide documentation for Nouveau Posted Sep 26, 2013 15:13 UTC (Thu) by mjg59 (subscriber, #23239) [Link]
If you don't deploy using IPMI then you clearly have physical access during initial configuration. And if you've left the IPMI network connected to the rest of your network, you're doing it very, very wrong. But since nobody in the server market seems to be talking about shipping with Secure Boot enabled by default, you can do your key installation in any way you want.
NVIDIA to provide documentation for Nouveau Posted Sep 26, 2013 22:48 UTC (Thu) by jmorris42 (subscriber, #2203) [Link]
> The machine owner should (through a demonstration of physical
> presence) be able to do whatever they want.
Nope. Physical presence does not equal ownership. We circulate laptops as library material. Do they get to do whatever they want? Oh heck no. And if the security tape over the screws is tampered with we fine em to cover our time reauditing the system.
How about a lab computer for library patrons. They are sitting at the console, so they install a new OS? Not on my systems they ain't, at last not without a screwdriver and some way to distract the staff.
NVIDIA to provide documentation for Nouveau Posted Sep 27, 2013 4:43 UTC (Fri) by mjg59 (subscriber, #23239) [Link]
Does physical ownership in your environment equal root? If so, you have other problems. If not, you have nothing to worry about.
|
|||||||||||||