Hi folks,
The Linux kernel RNG currently pretends to care about the "premature
next" RNG threat model. I'm wondering whether this is sensible and
corresponds to anything real.
"Premature next" is the scenario in which:
- Attacker compromises the current state of a fully initialized RNG with
a wild 'n crazy kernel infoleak.
- New bits of entropy are added directly to the key used to generate the
/dev/urandom stream, without any buffering or pooling.
- Attacker then, somehow having read access to /dev/urandom, samples RNG
output and brute forces the individual new bits that were added.
- Result: the RNG never "recovers" from the initial compromise, a
so-called violation of what academics term "post-compromise security".
(Note that this is a different scenario from "premature first", which
relates to boot-time concerns; this email isn't about "premature
first".)
There are other varied scenarios to this, usually involving some
combination of:
a) Attacker has access to /dev/urandom output continuously or at some
interesting interval.
b) Attacker controls one or more entropy sources providing some subset
of varying size of those new bits of entropy.
The Linux kernel currently pretends to mitigate this for scenario (a) at
least, using "entropy estimation". The idea is that it waits until 256
estimated "bits" of new entropy are pooled before mixing them into the
key used to generate the /dev/urandom stream. Never mind the fact that
entropy estimation is an impossible proposition and thus flawed, it
certainly does nothing in the way of (b), since there's just one pool.
The NT kernel is a bit more robust, by way of their Fortuna RNG, in
which there are several pools, and entropy sources round-robin into
those pools. When it's time to reseed, the first pool is used every
time, the second pool is used every other time, the third pool is used
every third time, the forth pool is used every forth time, and so on. In
theory this should handle both (a) and (b) without needing entropy
estimation, and this sort of scheduler prompts interesting questions for
academics with regards to different types of scheduling (a random walk
instead of round-robin? sounds like a paper topic.) and trying to
characterize the rate of inputs (continuous? sporadic? modelable?).
While the above "problem" maps pretty clearly to things academics are
interested in -- post-compromise security for a system with a clear
model and various creative solutions -- I'm wondering whether any of
this matters in the real world. From conversations over the last several
months with various security experts and cryptographers, including those
who work on the "premature next" problem, the impression I get is that
nobody actually thinks this matters back on planet Earth, even from
people who write papers on it.
So the purpose of this email is to solicit feedback on whether anybody
can think of a plausible scenario in which it does matter. If it does
matter, the next step will be to determine how much it matters exactly,
in order for me to gauge the cost-benefit ratio of mitigating the issue
more robustly in the kernel (e.g. Fortuna requires non-zero code
complexity; does the benefit outweigh the cost of such complexity?). On
the other hand, if nobody can think of any reason why this matters, then
there are some nice improvements that I'm eager to make in a different
direction.
To review, this attack model concerns:
- An attacker who compromises the RNG at one point in time via a kernel
infoleak.
- After that infoleak, the attacker somehow no longer has access to the
system, but can prevent the RNG from recovering from the compromise by
having pretty rapid access to /dev/urandom (and possibly also having
compromised zero or more entropy sources).
The questions are thus:
1) When does an attacker with a kernel infoleak exercise it just once,
and then attempt to maintain the leak with some sort of network access
to lots of /dev/urandom output (from, e.g., large nonces)?
2) Or, if it's a local user attacker, when does that attacker infoleak
once, but rather than just running the exploit again, cats /dev/urandom
continuously?
3) More broadly speaking, what kernel infoleak is actually acceptable to
the degree that anybody would feel okay in the first place about the
system continuing to run after it's been compromised?
Looking forward to hearing opinions on this. There's certainly a lot to
split hairs about above -- incomplete/inaccurate description of the
"premature next" model, what Fortuna actually achieves, my entropy
estimation remark, and so forth -- but hopefully this at least throws
enough things at the board to begin the discussion.
Is "premature next" just an academic exercise, rather than a real world
RNG concern?
Regards,
Jason
Jason A. Donenfeld <[email protected]> wrote:
> The Linux kernel RNG currently pretends to care about the "premature
> next" RNG threat model. I'm wondering whether this is sensible and
> corresponds to anything real.
>
> "Premature next" is the scenario in which:
> - Attacker compromises the current state of a fully initialized RNG with
> a wild 'n crazy kernel infoleak.
> - New bits of entropy are added directly to the key used to generate the
> /dev/urandom stream, without any buffering or pooling.
So don't do that, then. Keep a separate input pool/buffer and put
only hashed outputs from ir into the output pool.
> - Attacker then, somehow having read access to /dev/urandom, samples RNG
> output and brute forces the individual new bits that were added.
> - Result: the RNG never "recovers" from the initial compromise, a
> so-called violation of what academics term "post-compromise security".
Use chunks big enough for "catastrophic reseeding", impractical to
brute force, at least 64 bits & preferably larger.
> Fortuna requires non-zero code
> complexity; does the benefit outweigh the cost of such complexity?
I'd say certainly not.
> The questions are thus:
> ...
> 3) More broadly speaking, what kernel infoleak is actually acceptable to
> the degree that anybody would feel okay in the first place about the
> system continuing to run after it's been compromised?
If we have a good entropy source -- e.g. running on an Intel CPU
& consider their RNG instruction trustworthy, or in a VM & trust
the host -- then we should be able to guarantee recovery at the
next reseeding. Just dump at least 128 bits from that source
into the input pool before hashing.
The interesting question is whether & how soon we can guarantee
recovery if no such source is available, we rely only on entropy
gathering from interrupts, with or without the gcc latent entropy
plugin.
> Is "premature next" just an academic exercise, rather than a real world
> RNG concern?
No.
Hi Ted,
That's a useful analysis; thanks for that.
On Sat, Apr 30, 2022 at 05:49:55PM -0700, tytso wrote:
> On Wed, Apr 27, 2022 at 03:58:51PM +0200, Jason A. Donenfeld wrote:
> >
> > 3) More broadly speaking, what kernel infoleak is actually acceptable to
> > the degree that anybody would feel okay in the first place about the
> > system continuing to run after it's been compromised?
>
> A one-time kernel infoleak where this might seem most likely is one
> where memory is read while the system is suspended/hibernated, or if
> you have a VM which is frozen and then replicated. A related version
> is one where a VM is getting migrated from one host to another, and
> the attacker is able to grab the system memory from the source "host"
> after the VM is migrated to the destination "host".
You've identified ~two places where compromises happen, but it's not an
attack that can just be repeated simply by re-running `./sploit > state`.
1) Virtual machines:
It seems like after a VM state compromise during migration, or during
snapshotting, the name of the game is getting entropy into the RNG in a
usable way _as soon as possible_, and not delaying that. This is
Nadia's point. There's some inherent tension between waiting some amount
of time to use all available entropy -- the premature next requirement
-- and using everything you can as fast as you can because your output
stream is compromised/duplicated and that's very bad and should be
mitigated ASAP at any expense.
[I'm also CC'ing Tom Risenpart, who's been following this thread, as he
did some work regarding VM snapshots and compromise, and what RNG
recovery in that context looks like, and arrived at pretty similar
points.]
You mentioned virtio-rng as a mitigation for this. That works, but only
if the data read from it are actually used rather quickly. So probably
/waiting/ to use that is suboptimal.
One of the things added for 5.18 is this new "vmgenid" driver, which
responds to fork/snapshot notifications from hypervisors, so that VMs
can do something _immediately_ upon resumption/migration/etc. That's
probably the best general solution to that problem.
Though vmgenid is supported by QEMU, VMware, Hyper-V, and hopefully soon
Firecracker, there'll still be people that don't have it for one reason
or another (and it has to be enabled manually in QEMU with `-device
vmgenid,guid=auto`; perhaps I should send a patch adding that to some
default machine types). Maybe that's their problem, but I take as your
point that we can still try to be less bad than otherwise by using more
entropy more often, and not delaying as the premature next model
requirements would have us do.
2) Suspend / hibernation:
This is kind of the same situation as virtual machines, but the
particulars are a little bit different:
- There's no hypervisor giving us new seed material on resumption like
we have with VM snapshots and vmgenid; but
- We also always know when it happens, because it's not transparent to
the OS, so at least we can attempt to do something immediately like
we do with the vmgenid driver.
Fortunately, most systems that are doing suspend or hibernation these
days also have a RDRAND-like thing. It seems like it'd be a good idea
for me to add a PM notifier, mix into the pool both
ktime_get_boottime_ns() and ktime_get(), in addition to whatever type
info I get from the notifier block (suspend vs hibernate vs whatever
else) to account for the amount of time in the sleeping state, and then
immediately reseed the crng, which will pull in a bunch of
RDSEED/RDRAND/RDTSC values. This way on resumption, the system is always
in a good place.
I did this years ago in WireGuard -- clearing key material before
suspend -- and there are some details around autosuspend (see
wg_pm_notification() in drivers/net/wireguard/device.c), but it's not
that hard to get right, so I'll give it a stab and send a patch.
> But if the attacker can actually obtain internal state from one
> reconstituted VM, and use that to attack another reconstituted VM, and
> the attacker also knows what the nonce or time seed that was used so
> that different reconstituted VMs will have unique CRNG streams, this
> might be a place where the "premature next" attack might come into
> play.
This is the place where it matters, I guess. It's also where the
tradeoff's from Nadia's argument come into play. System state gets
compromised during VM migration / hibernation. It comes back online and
starts doling out compromised random numbers. Worst case scenario is
there's no RDRAND or vmgenid or virtio-rng, and we've just got the good
old interrupt handler mangling cycle counters. Choices: A) recover from
the compromise /slowly/ in order to mitigate premature next, or B)
recover from the compromise /quickly/ in order to prevent things like
nonce reuse.
What is more likely? That an attacker who compromised this state at one
point in time doesn't have the means to do it again elsewhere in the
pipeline, will use a high bandwidth /dev/urandom output stream to mount
a premature next attack, and is going after a high value target that
inexplicably doesn't have RDRAND/vmgenid/virtio-rng enabled? Or that
Nadia's group (or that large building in Utah) will get an Internet tap
and simply start looking for repeated nonces to break?
Jason
resending in plain text... (hope got it right)
On Mon, May 9, 2022 at 11:15 AM Yevgeniy Dodis <[email protected]> wrote:
>
> Hi Jason and all.
>
> Thank you for starting this fascinating discussion. I generally agree with everything Jason said. In particular, I am not
> 100% convinced that the extra cost of the premature next defense is justified.(Although Windows and MacOS are adamant it is
> worth it :).)
>
> But let me give some meta points to at least convince you this is not as obvious as Jason makes it sound.
>
> 1) Attacking RNGs in any model is really hard. Heck, everybody knew for years that /dev/random is a mess
> (and we published it formally in 2013, although this was folklore knowledge), but in all these years nobody
> (even Nadya's group :)) managed to find a practical attack. So just because the attack seems far-fetched, I do not think we should
> lower our standards and do ugly stuff. Otherwise, just leave /dev/random the way it was before Jason started his awesome work.
>
> 2) As Jason says, there are two distinct attack vectors needed to make the premature next attack.
> A) compromising the state
> B) (nearly) continuously observing RNG outputs
>
> I agree with Jason's point that finding places where
> -- A)+B) is possible, but
> --- A)+A) is not possible,
> is tricky. Although Nadya kind of indicated a place like that. VM1 and VM2 start with the same RNG state (for whatever
> reason). VM1 is insecure, so can leak the state via A). VM2 is more secure, but obviously allows for B) through system
> interface. This does not seem so hypothetical for me, especially in light of my mega-point 1) above -- almost any real-world
> RNG attack is hard.
>
> But I want to look at it from a different angle here. Let's ask if RNGs should be secure against A) or B) individually.
>
> I think everybody agrees protection from B) is a must. This is the most basic definition of RNG! So let's just take itas
> an axiom.
>
> Protection against A) is trickier. But my read of Jason's email is that all his criticism comes exactly from this point.
> If your system allows for state compromise, you have bigger problems than the premature next, etc. But let's ask ourselves
> the question. Are we ready to design RNGs without recovery from state compromise? I believe nobody on this list would
> be comfortable saying "yes". Because this would mean we don;t need to accumulate entropy beyond system start-up.
> Once we reach the point of good initial state, and state compromise is not an issue, just use straight ChaCha or whatever other
> stream cipher.
>
> The point is, despite all arguments Jason puts, we all would feel extremely uncomfortable/uneasy to let continuous
> entropy accumulation go, right?
>
> This means we all hopefully agree that we need protection against A) and B) individually.
>
> 3) Now comes the question. If we want to design a sound RNG using tools of modern cryptography, and we allow
> the attacker an individual capability to enforce A) or B) individually, are we comfortable with the design where we:
> * offer protection against A)
> * offer protection against B)
> * do NOT offer protection against A)+B), because we think it's too expensive given A)+B) is so rare?
>
> I do not have a convincing answer to this question, but it is at least not obvious to me. On a good note, one worry
> we might have is how to even have a definition protecting A), protecting B), but not protecting A)+B).
> Fortunately, our papers resolve this question (although there are still theoretical annoyances which I do not
> want to get into in this email). So, at least from this perspective, we are good. We have a definition with
> exactly these (suboptimal) properties.
>
> Anyway, these are my 2c.
> Thoughts?
>
> Yevgeniy
>
> On Sun, May 1, 2022 at 7:17 AM Jason A. Donenfeld <[email protected]> wrote:
>>
>> Hi Ted,
>>
>> That's a useful analysis; thanks for that.
>>
>> On Sat, Apr 30, 2022 at 05:49:55PM -0700, tytso wrote:
>> > On Wed, Apr 27, 2022 at 03:58:51PM +0200, Jason A. Donenfeld wrote:
>> > >
>> > > 3) More broadly speaking, what kernel infoleak is actually acceptable to
>> > > the degree that anybody would feel okay in the first place about the
>> > > system continuing to run after it's been compromised?
>> >
>> > A one-time kernel infoleak where this might seem most likely is one
>> > where memory is read while the system is suspended/hibernated, or if
>> > you have a VM which is frozen and then replicated. A related version
>> > is one where a VM is getting migrated from one host to another, and
>> > the attacker is able to grab the system memory from the source "host"
>> > after the VM is migrated to the destination "host".
>>
>> You've identified ~two places where compromises happen, but it's not an
>> attack that can just be repeated simply by re-running `./sploit > state`.
>>
>> 1) Virtual machines:
>>
>> It seems like after a VM state compromise during migration, or during
>> snapshotting, the name of the game is getting entropy into the RNG in a
>> usable way _as soon as possible_, and not delaying that. This is
>> Nadia's point. There's some inherent tension between waiting some amount
>> of time to use all available entropy -- the premature next requirement
>> -- and using everything you can as fast as you can because your output
>> stream is compromised/duplicated and that's very bad and should be
>> mitigated ASAP at any expense.
>>
>> [I'm also CC'ing Tom Risenpart, who's been following this thread, as he
>> did some work regarding VM snapshots and compromise, and what RNG
>> recovery in that context looks like, and arrived at pretty similar
>> points.]
>>
>> You mentioned virtio-rng as a mitigation for this. That works, but only
>> if the data read from it are actually used rather quickly. So probably
>> /waiting/ to use that is suboptimal.
>>
>> One of the things added for 5.18 is this new "vmgenid" driver, which
>> responds to fork/snapshot notifications from hypervisors, so that VMs
>> can do something _immediately_ upon resumption/migration/etc. That's
>> probably the best general solution to that problem.
>>
>> Though vmgenid is supported by QEMU, VMware, Hyper-V, and hopefully soon
>> Firecracker, there'll still be people that don't have it for one reason
>> or another (and it has to be enabled manually in QEMU with `-device
>> vmgenid,guid=auto`; perhaps I should send a patch adding that to some
>> default machine types). Maybe that's their problem, but I take as your
>> point that we can still try to be less bad than otherwise by using more
>> entropy more often, and not delaying as the premature next model
>> requirements would have us do.
>>
>> 2) Suspend / hibernation:
>>
>> This is kind of the same situation as virtual machines, but the
>> particulars are a little bit different:
>>
>> - There's no hypervisor giving us new seed material on resumption like
>> we have with VM snapshots and vmgenid; but
>>
>> - We also always know when it happens, because it's not transparent to
>> the OS, so at least we can attempt to do something immediately like
>> we do with the vmgenid driver.
>>
>> Fortunately, most systems that are doing suspend or hibernation these
>> days also have a RDRAND-like thing. It seems like it'd be a good idea
>> for me to add a PM notifier, mix into the pool both
>> ktime_get_boottime_ns() and ktime_get(), in addition to whatever type
>> info I get from the notifier block (suspend vs hibernate vs whatever
>> else) to account for the amount of time in the sleeping state, and then
>> immediately reseed the crng, which will pull in a bunch of
>> RDSEED/RDRAND/RDTSC values. This way on resumption, the system is always
>> in a good place.
>>
>> I did this years ago in WireGuard -- clearing key material before
>> suspend -- and there are some details around autosuspend (see
>> wg_pm_notification() in drivers/net/wireguard/device.c), but it's not
>> that hard to get right, so I'll give it a stab and send a patch.
>>
>> > But if the attacker can actually obtain internal state from one
>> > reconstituted VM, and use that to attack another reconstituted VM, and
>> > the attacker also knows what the nonce or time seed that was used so
>> > that different reconstituted VMs will have unique CRNG streams, this
>> > might be a place where the "premature next" attack might come into
>> > play.
>>
>> This is the place where it matters, I guess. It's also where the
>> tradeoff's from Nadia's argument come into play. System state gets
>> compromised during VM migration / hibernation. It comes back online and
>> starts doling out compromised random numbers. Worst case scenario is
>> there's no RDRAND or vmgenid or virtio-rng, and we've just got the good
>> old interrupt handler mangling cycle counters. Choices: A) recover from
>> the compromise /slowly/ in order to mitigate premature next, or B)
>> recover from the compromise /quickly/ in order to prevent things like
>> nonce reuse.
>>
>> What is more likely? That an attacker who compromised this state at one
>> point in time doesn't have the means to do it again elsewhere in the
>> pipeline, will use a high bandwidth /dev/urandom output stream to mount
>> a premature next attack, and is going after a high value target that
>> inexplicably doesn't have RDRAND/vmgenid/virtio-rng enabled? Or that
>> Nadia's group (or that large building in Utah) will get an Internet tap
>> and simply start looking for repeated nonces to break?
>>
>> Jason
Jason A. Donenfeld writes:
> Right, VMs are super problematic, but for that, there's now this
> "vmgenid" driver, where the hypervisor actually gives a 128-bit seed to
> guests when they're resumed, so that we can immediately reseed, which
> should pretty comprehensively handle that situation.
Hmmm. If an application initializes its own RNG state from /dev/urandom,
and is then cloned, and then generates an ECDSA nonce from the RNG
state, and then uses this nonce to sign a message that's different
across the clones, how is disaster averted?
Given the goal of sending money to cryptographers, I'm pretty sure we
want the answer to be a security-audit nightmare, so let me suggest the
following idea. There's SIGWINCH to notify processes about window-size
changes, so there should also be a signal for RNG changes, which should
be called SIGRINCH, and there should be a different mechanism to address
RNG output cloning inside the kernel, and there should be endless papers
on Grinch Attacks, including papers that sort of prove security against
Grinch Attacks, and deployment of software that's sort of protected
against Grinch Attacks, and fear of the bad PR from abandoning anything
labeled as protection, because, hey, _maybe_ the protection accomplishes
something, and it's not as if anyone is going to be blamed for whatever
damage is caused by the systems-level effect of the added complexity.
---D. J. Bernstein
P.S. Yes, yes, I know the name "Grinch Attack" has been used before.
On Tue, 2022-05-10 at 22:09 +0200, Jason A. Donenfeld wrote:
> For 5.19 (or at this point, more likely 5.20), there's a userspace
> notifier in store, maybe, if I can figure out how to do it right.
> There's a pretty bikesheddy thread here on what shape that interface
> should take: https://lore.kernel.org/lkml/[email protected]/
> But basically there are some details about how an async interface should
> work, and what the virtual hardware future, if any, looks like for a
> memory mapped race-free polling interface. Plus some considerations on
> how much we should care etc.
Perhaps it might be simpler to add an "epoch" number or similar exposed
via something like a vDSO that proper user space implementations can
then check before returning numbers from PRNGs and will immediately
reseed from /dev/random if it ever changes (rare event).
It is much simpler to poll for this information in user space crypto
libraries than using notifications of any kind given libraries are
generally not in full control of what the process does.
This needs to be polled fast as well, because the whole point of
initializing a PRNG in the library is that asking /dev/urandom all the
time is too slow (due to context switches and syscall overhead), so
anything that would require a context switch in order to pull data from
the PRNG would not really fly.
Note that a generic "epoch" exposed via vDSO would be useful beyond
RNGs, as it would be usable by any other user space that needs to know
that "I have been cloned and I need to do something now" and would be
able to use it immediately w/o the kernel needing to grow any other
specific interface for it.
HTH.
Simo.
--
Simo Sorce
RHEL Crypto Team
Red Hat, Inc
Hi Simo,
On Tue, May 10, 2022 at 05:33:34PM -0400, Simo Sorce wrote:
> On Tue, 2022-05-10 at 22:09 +0200, Jason A. Donenfeld wrote:
> > For 5.19 (or at this point, more likely 5.20), there's a userspace
> > notifier in store, maybe, if I can figure out how to do it right.
> > There's a pretty bikesheddy thread here on what shape that interface
> > should take: https://lore.kernel.org/lkml/[email protected]/
> > But basically there are some details about how an async interface should
> > work, and what the virtual hardware future, if any, looks like for a
> > memory mapped race-free polling interface. Plus some considerations on
> > how much we should care etc.
>
> Perhaps it might be simpler to add an "epoch" number or similar exposed
> [...]
Could you send these ideas to the bikeshed thread that I linked rather
than this premature next one. I think it'd be a good idea to keep Alex
and Lennart looped in to that discussion, since they represent userspace
projects that probably care about it, and not bog this one down with
systems programming things pretty far from premature next threat model
stuff. It's a bikeshed thing, after all...
Thanks,
Jason
Hi Tom,
On Wed, May 11, 2022 at 08:26:08PM +0000, Thomas Ristenpart wrote:
> To me the high-level design features that seems to check all the
> boxes, including importantly simplicity:
>
> 1) A single pool where opportunistic entropy measurements (interrupt
> timings, etc.) are folded in and that is used to generate outputs.
>
> 2) An explicit “generate entropy” routine that attempts to quickly
> generate a large amount of entropy. Use this to (re)initialize the
> state upon system events like boot and VM resumption. The CPU jitter
> dance type mechanisms are a good bet, though someone should probably
> check that these work on low-end systems.
>
> Also I would advocate always folding in other sources of entropy
> (e.g., RDRAND) when available, performance allowing, in both 1 and 2.
> Given the above discussion, I don’t think it’s very important, but an
> extension of the above to provide some limitation of premature next
> concerns would be:
RDRAND is currently mixed in during system boot and during reseeding
(which now happens after sleep resumption too, as of [1]).
Specifically, reseeding takes this HKDF-like form:
τ = blake2s(key=last_key, input₁ ‖ input₂ ‖ … ‖ inputₙ)
κ₁ = blake2s(key=τ, RDSEED ‖ 0x0)
κ₂ = blake2s(key=τ, RDSEED ‖ 0x1)
last_key = κ₁
crng_seed = κ₂
where RDSEED here represents 256 bits of RDSEED, RDRAND, or RDTSC,
depending on what's available on the platform, operating as a sort of
salt parameter. When RDSEED or RDRAND is available, this matches your
suggestion. When only RDTSC is available, it's maybe jittery, but not
very much because it's just called in a tight loop, which brings us to
your next suggestion:
> 3) Periodically call 2. For example, when a CPU is otherwise idle.
> This would have same effect as Fortuna-style approaches without adding
> new buffers, etc.
For systems without RDSEED or RDRAND, doing jitter entropy periodically
would at least be /something/ significant in the service of "solving"
the premature next "problem" (in addition to the more significant VM
problem in the absence of vmgenid). Your suggestion of an explicit
"generate entropy" function that can be called periodically is a similar
to Linus' point when he introduced jitter entropy, titling the commit,
"try to actively add entropy rather than passively wait for it" [2].
It's a good point. If we have a way of generating entropy directly instead of
passively waiting for it, something complicated like Fortuna isn't even
necessary. (As a silly side note: since Fortuna only claims to
/eventually/ recover from a compromise but can't tell you when, such
jitter could be done once a week and still, on paper, accomplish the
same theoretical goal...)
Jitter might not be available on all architectures that Linux supports,
though it likely is available for most deployed systems out there. And
for 5.19, I've fixed some cycle counter fallback things so that it
should hopefully be available a few more places it wasn't before.
Similarly, RDSEED/RDRAND isn't available everywhere, but it is available
most places these days.
But on the other hand, it appears that none of us really thinks that
premature next is a real problem worth complicating designs over. So
maybe we can just say that it is nice when the silicon in one way or
another helps with premature next, but maybe not an explicit must have.
So where does that leave us?
- Systems with RDSEED/RDRAND don't have premature next, due to the above
KDF salt. This is probably the majority of systems out there these
days. This also applies to the sleep resumption notification (and the
vmgenid one), and I suspect that most systems with S3 or S0ix or
whatever else these days also probably have RDRAND.
- Systems with viable jitter entropy could be in a position to not have
premature next too, if we added periodic jitter entropy calls per your
suggestion (3). Though, the jitter dance as it currently exists involves
hammering on the scheduler a bit and spiking latency, so I'm not
totally sure this is really worth it to do beyond boot time. It'd need
a little bit more specific engineering, anyhow, to get the details
right on it.
- Systems with no viable jitter nor RDSEED/RDRAND would need something
like Fortuna, which doesn't seem worth it at all, given the
discussion. These machines are probably in the first percentile of
deployed systems too, and probably should be using something like
seedrng [3] to initialize the RNG anyway. Plus, are these systems even
fast enough to make condition B) viable to an attacker?
> Details would need to be worked out, of course. Hope this was helpful
> and apologies that it got long,
Very helpful, thank you. The key takeaways for me are:
- Premature next remains not a real world problem, per the reasons you
and others cited.
- Entropy *generation* makes most of those concerns disappear anyway,
without the complexities and security issues associated with entropy
long- or multi- *pooling*.
Jason
[1] https://git.kernel.org/crng/random/c/7edc59743da5
[2] https://git.kernel.org/crng/random/c/50ee7529ec45
[3] https://git.zx2c4.com/seedrng/about/
Am Thu, May 12, 2022 at 01:47:06PM +0200 schrieb Jason A. Donenfeld:
> But on the other hand, it appears that none of us really thinks that
> premature next is a real problem worth complicating designs over. So
> maybe we can just say that it is nice when the silicon in one way or
> another helps with premature next, but maybe not an explicit must have.
> So where does that leave us?
>
> - Systems with RDSEED/RDRAND don't have premature next, due to the above
> KDF salt. This is probably the majority of systems out there these
> days. This also applies to the sleep resumption notification (and the
> vmgenid one), and I suspect that most systems with S3 or S0ix or
> whatever else these days also probably have RDRAND.
... and most of these systems have TPM chips with a RNG, which is (alas)
usually only used at system startup, as that hw_rng device sets its quality
to 0 (meaning untrusted). So there's also room for improvement involving
these hw rng devices.
Thanks,
Dominik