Skip to content

Latest commit

 

History

History
66 lines (39 loc) · 2.61 KB

analysis_plan.md

File metadata and controls

66 lines (39 loc) · 2.61 KB

Attack and Analysis plan

This document helps us structure our research

Reading list

See README.md

Deployment scenarios

We will analyse small and medium scale operators and try to identify possible traps for misconfiguring and map out residual privacy leaks in such scenarios. Where possible, we will recommend mitigation strategies.

Split mode is out of scope of this document, since it will mostly be used by very large setups (such as CDNs).

List of possible deployment scenarios:

  • small to medium sized website hoster (world4you.com, Hetzner, ...)
  • universities, NRENs

Question: what is the distribution number of domains / host? What if we remove the CDNs? How does that look like? What is the effort for the smaller hosts (hosters) vs. benefit for them? Does ECH pay off for them? If not, it will not be widely deployed in these scenarios.

Question: which scenarios does Stephen see?

Attacks / security considerations per deployment scenario

Per scenario, list all possible attack possibilities and if possible how to mitigate them/counter them.

Example:

  • As a univeristy, I will deploy ECH for ... I need to do: 1 2 3...
  • As a hoster I will deploy ECH for ... I need to do 1 2 3

Game-theoretic considerations

If I have work with it, if I risk breaking it, and I don't have a direct benefit myself (but just protect the client's client hello), then what is my benefit? Why should I do it?

Global scenario considerations

Remaining metadata

review the metadata in a properly functioning ECH interaction and explore remaining avenues for de-anonymisation, filtering, blocking and censorship.

.RU case ...

"The Russian internet regulator also recommended local website owners switch to domestic CDN services"

other ways to block ECH Config

DNS

Is there something like certificate pinning?

Other downgrade attacks?

Other de-anonymization techniques based on correlating data sets