syz-manager: add a check for 'localhost'

[p13l13d13]

If localhost is not resolvable on a system, we might hang without any logging (for example in vm/vmimpl/UnusedTCPPort) or fail later. Since syzkaller uses localhost (and not something like 127.0.0.1) it's way better to notify user earlier if their localhost is not configured correctly.

---

P.S. I was running syzkaller on a system with empty /etc/hosts, so I've stuck inside UnusedTCPPort, because of unsolvable 'localhost'.

I guess fixing this ONLY inside the problematic function is not enough, so I've added a top-level check. However, I'm not familiar with the codebase, so please tell me if I need to add this check to more locations or somewhere else.

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[a-nogikh]

To be honest, not having the localhost host seems to be such a non-typical situation that it feels just wrong for syz-manager to be validating for cases like this. There are infinite ways for the host system to be misconfigured.

That being said, I agree that the syzkaller should not have silently hung in this situation.

We could try to make UnusedTCPPort() check availability on all interfaces at the same time (":" + port instead of "localhost:" + port), but that still leaves a corner case of having 0 networking interfaces on the system :)

We could use 127.0.0.1 instead of localhost, but there's again a potential corner case of a machine without ipv4 support. Though syzkaller would anyway fail on such a machine since it already uses 127.0.0.1 in a number of places.

We could modify UnusedTCPPort() to look more carefully at the returned error and only continue if the error is about the port being busy.

There are other places that rely on localhost, but AFAIK they all should already be more explicit, e.g. qemu VMs will fail on start and you'll see some error messages in the logs.

[p13l13d13]

Well, I think that ::1 and 127.0.0.1 are pretty much standard ways to access loopback interfaces. If 127.0.0.1 and ::1 are not accessible on a machine, then it has a rare custom configuration and operators/developers should know about it, thus it's not a syzkaller's problem.

IMHO, UnusedTCPPort should be fixed anyway, cause there is potential for locking in it.

So, my proposal is:

1. I fix the UnusedTCPPort to exit on some errors.
2. I replace 'localhost' with 127.0.0.1.
3. (opt) use ::1 if it's an IPv6 machine. Though, that would require a bigger rework with a helper function, etc.

[tarasmadan]

#3579 fixed the failures on machines with IPv6 only.
Please don't change to 127.0.0.1.

[p13l13d13]

Replaced with #5658