eth/downloader: ignore headers not requested

[holiman]

This is a bit of work in progress.
Explanation here: #22209 (comment)

[holiman]

What this PR does, is ignore 'spurious' headers that arrive while we're waiting for some specific headers (specifically: during ancestor search).
So whereas previously we would reject the peer, we now don't. One consequence, is that we'll wait out the entire timeout. I'm not 100% sure what that timeout is -- in one of the downloader tests, e.g. TestShiftedHeaderAttack65Full, it's a full minute.

So does this mean that now, instead of rejecting quickly from a malicious peer, we now 'hang' for a full minute? Well, one can then switch it, and say: the attacker can already perform that attack, simply by not answering our header-request.

So regardless of what we do, the starvation/timeout attack is present. But if we roll with this PR, we might be able to continue the sync without stop/rollback/resync in case it's just a matter of a misdelivered header.

[holiman]

Hm, this PR unintentionally surfaced another bug in the downloader, which now causes certiain tests to hang.

https://github.com/ethereum/go-ethereum/blob/master/eth/downloader/downloader.go#L973:L978

If we request header N, and the remote peer delivers M, (e.g. N-1), and both of those are outside of the range of blocks that we have locally, then we''ll happily accept that and continue the algorithm.

The reason the tests hang (e.g. TestFailedSyncProgress65Full) is that it expects the ancestor search to find genesis, but this PR makes the findAncestor algo error out, since the remote part doesn't deliver the correct numbers.

[holiman]

Context

[holiman]

After triage: this is the right approach, need to fix up the tests and also take a look at why/of the block filter in the fetcher does not filter out any fetcher-initiated headers

[holiman]

I'll revisit this once the req-id downloader changes are in

[s1na]

I was going through old PRs and noticed this PR and #22249 were put on hold until req ids were in. Since req ids are now merged I'll go ahead and remove the labels. Just a friendly reminder that they're up for grabs.

[holiman]

Can be closed now that we have request ids