/dev/random and /dev/urandom radomness seeding in Tails

/dev/random and /dev/urandom are special Linux devices that provide access from user land to the Linux kernel Cryptographically Secure Pseudo Random Number Generator (CSPRNG). This generator is used for almost every security protocol, like TLS/SSL key generation, choosing TCP sequences, ASLR offsets, and GPG key generation [1]. In order for this CSPRNG to be really cryptographically secure, it's recommended to seed it with a 'good' entropy source, even though The Linux kernel collects entropy from several sources, for example keyboard typing, mouse movement, among others.

Because of the Tails nature of being amnesic, and run from different type of live devices (from DVDs to USB sticks), special care must be taken to ensure the system still gets enough entropy and boots with enough randomness. This is not easy in the Tails context, where the system is almost always booting the same way. Even the squashfs file is ordered to optimize boot time.

Although these problem have been documented since a long time (see [7] and [8]), there's not much done to tackle the problem. We looked at notes and research from LiveCD OS's and supply them here for completements sake. Whonix has a wiki page with some notes, and Qubes has tickets about this ([3],[4],[5] and [6]).

Current situation

See the related design document

Tails do not ship /var/lib/urandom/random-seed in the ISO, since it means shipping a fixed known value for every Tails installation which means its entropy contribution is zero, and breaks reproducibility of the ISO image.

Without this random seed, systemd-random-seed won't write anything to /dev/urandom, so we rely purely on the kernel CSPRNG and current system entropy to get /dev/urandom. It's commonly admitted to be quite good, but given the Live nature of Tails, and the fact that good cryptography is a must, we may want to add additional measures to ensure any Tails system has enough entropy.

Tails ships Haveged and rngd since a while. Still there are concerns about Haveged's reliability to provide cryptographically secure randomness, and rngd is only really useful when random generator devices are used.

Taking other measures to seed the Linux Kernel CSPRNG with good material is something worst spending efforts on.

Use cases

Tails is used in different ways with different live devices. That requires different solutions, depending on how and what the Tails OS is installed.

DVD

This may be the most difficult, since all that the user is running is the plain ISO we provide. In there, there's no seed at all, and no way for the users to add one.

On the other hand, that's not the installation method we want to support the most, and probably not the most used when people want to secure other communication types than HTTPS (e.g persistence is very usefull for OpenPGP key storage and usage, chat account configuration, ...).

So we may eventually just document somewhere to users that they MUST NOT use this type of installation if they want to rely on good cryptograpy for their communications and key generation, or that they should wait after having interacting a long (but hard to define) time with the system so that it had time to collect entropy, and does not rely on the CSPRNG, Haveged and rngd only.

We could also add some kind of notification to users when entropy gets too low, or just saying them that the way they use Tails is not compatible with strong cryptography.

Intermediary USB

This type of installation is supposed to be used when people are installing Tails from another OS (except Debian and Ubuntu, where they can use the Tails installer). In most case, this means having a bit by bit copy of the Tails ISO on the USB stick, except for Windows where we ask to use the Universal USB Installer

In this case the situation is pretty much the same than with the DVD one. No seed, and adding one is very difficult if not impossible (except with the Windows installation where we may ask upstream to implement that in the Universal USB Installer, but well...).

That's also not really the way we encourge users to use Tails, so as with DVD there's maybe no point to fix the situation here, and the same workaround could be applied (document it).

Final USB

That's supposed to be the standard way to use Tails.

Note that in this case, there are two situations: booting this installation with persistence enabled, and without.

It is worth noting too that the first time this Tails installation is booted, most of the time the first step is to configure persistence, which means creating an encrypted partition. At this step though, there is at the moment probably very little entropy, so this may weaken the LUKS volume encryption.

Virtual Machines

That's a way to use Tails, and one of the worst cases: it is of public knowledge that entropy in VMs is very poor. It's not really clear how the entropy gathering daemons we have would help, but there are mechanisms now in libvirt to pass randomness from the host using the Virtio RNG feature (even if it may not be enough by itself).

Proposed solutions

Persist entropy pool seeds #7675

We hope to improve this situation for users who enable the persistent storage option by storing a seed from the previous session to help bootstrap with some "well" generated randomness.

Storing it in the persistent partition will be implemented using a default (hidden to the user) persistence setting. But it does not solve the problem for the first time Tails is booted, which is likely when the encrypted persistence partition is created.

Use the Tails installer to create a better seed #11897

Tails installer can be used on Debian and Ubuntu, and is the tool people running OSX or Windows are told to use to install their final Tails USB stick with, by using an intermediary Tails to create the final USB.

Tails installer could store a seed in the FAT filesystem of the system partition. That would workaround this first boot problem not handled by the persistence option.

We can't sadly update this seed while running Tails, as mounting RW the system FAT partition during a Tails session does not work. So the question whether updating it or not is open.

If we want to do so, we'll have to update it at the system shutdown. This will mean remount this partition, write the new random seed, then unmount it and start the shutdown of the system. Obviously we can do this only in normal shutdown process, and will have to avoid it in emergency shutdown mode.

We may alternatively not update it, and use it only when the persistence is not enabled. That would still be a unique source of entropy per Tails installation, so that would be a better situation that the current one.

One drawback: this would break the ability to verify this system partition with a simple shasum operation.

Use stronger/more entropy collectors #5650

As already stated, Tails run Haveged, and rngd (since 2.6 for the later).

We may want to add other sources though, given there are concerns about Haveged, and rngd starts only when a hardware RNG is detected, which is not so often the case.

XXX: It would be nice to have a study (read: a survey of packages, etc) of all the useful entropy gathering daemons that might be of use on a Tails system.

An evaluation of some of them has been done already

Possible candidates:

  • entropy gathering daemon: not packaged into Debian.
  • twuewand: used by Finnix LiveCD (so made for this kind of environment), packaged into Ubuntu only.
  • timer entropy daemon: not packaged into Debian
  • randomsound: probably a bad idea in the Tails context as we're discussing a Greeter option to deactivate the microphone.

Block booting till enough entropy has been gathered

One way to ensure Tails is booting with enough entropy would be to block during the boot if the system is lacking of it.

But this brings questions about how to interact correctly with the users, as blocking without notifications would be terrible UX. Also Tails boot time is a bit long already, and this may grow it quite a bit more again.

XXX: So before going on, we need a bit more data about the state of the entropy when Tails boot, specially now that we have several entropy collector daemons. It may very well be that this case do not happen anymore. And if it is, we need to know on average how much time that blocking would last. [Sycamoreone] #11758

Regulary check available entropy and notify if low

An idea that has been mentioned several time is to have a service that check if the available entropy is high enough, and notify the user if it's not the case. One downside, is that observing the entropy pool costs randomness, so this may have to be implemented with care or is worth discussing/researching the costs/benefits.

Related tickets

This is about #7642, #7675, #6116, #11897 and friends.

References