|
1 | | -# Browsers |
| 1 | +# Browsers and ECH |
2 | 2 |
|
3 | | -browsers are permissive |
4 | | -if dns is misconfigured, browser tries without ECH |
| 3 | +Modern web browsers are notably permissive toward emerging standards, often prioritizing user functionality over enforcing new security features if the potential impact is too disruptive. With their quick adoption of new technologies and fast release cycles, they effectively serve as experimental platforms for evaluating and implementing new protocols in real-world environments. |
5 | 4 |
|
6 | | -explore how FF and other browsers behave with famility safety software exactly, so far not tested |
| 5 | +The leading browsers Firefox, Chrome, Edge and further Brave, Opera, Vivaldi adopted DoH as their default setting, reverting back to Do53 should a DoH connection fail to establish. Such failures can arise from active downgrade attacks, where malicious entities intercept and manipulate traffic. Consequently, the implementation of ECH can be thwarted if an attacker holds sway over any part of the network path between the user and the intended target. |
7 | 6 |
|
8 | | -## Non-pure-HTTPS traffic |
| 7 | +For the implementation of ECH, attention must not only be paid to pure HTTPS traffic but also to other communication channels such as WebRTC and network proxies, as neglecting ECH on these channels can introduce ways for de-anonymization. |
9 | 8 |
|
10 | | -### Jitsi |
| 9 | +## Browsers' Policy Enforcement Power |
11 | 10 |
|
12 | | -- UDP/10000 Jitsi video bridge |
13 | | -- UDP/3478 STUN server (coturn) |
14 | | -- TCP/5349 fall-back connection for video/audio communication (in case the UDP connection did not work) |
15 | | -- TCP/443 (HTTPS) only for the UI, protected with ECH by default |
| 11 | +In the past, Browsers and the CA/Browser forum have repeatedly shown that they can enforce new policies towards network and website operators, pushing them to fast adjustments to not risk their website's reputations, such as: |
16 | 12 |
|
17 | | -### Network Proxy |
| 13 | +- Starting in 2015, Browsers gradually marked unencrypted HTTP connections as unsafe, pushing all website operators to use HTTPS. |
| 14 | +- Previous to 2011, certificate lifetimes of up to 10 years were allowed by standards and accepted by browsers. Since 2020, a year is the maximum allowed lifetime. Discussions on further reductions are ongoing. |
| 15 | +- Other examples include: Deprecation of SHA-1 (2014-2017) and Deprecation of RSA Keys (since 2010), Distrust of CAs (Symantec, 2017–2018), Mixed Content Blocking, [Requirements to CAs](https://cabforum.org/baseline-requirements/) such as Certificate Transparency, OCSP, `SubjectAltName`, Domain Validation Methods, Deprecation of Certificates for Internal Server Names (2011-2016), Deprecation of TLS protocols and cipher suites |
18 | 16 |
|
19 | | -SOCKS, HTTPS |
| 17 | +The reasons for this effectiveness lie in the huge market share of a small number of browsers and, on the operators' side, the high reputation risk: Website operators simply cannot afford security warnings or site breakage. |
20 | 18 |
|
21 | | -### OCSP and CRL |
| 19 | +These policies are increasingly getting tighter, pushing the operators to more automation and more complex environments. This, in turn makes the operation less efficient pushing many of them to centralized service operators. This counteracts the original intention of a decentralized Internet. |
22 | 20 |
|
23 | | -Downloaded via HTTP but signed by the CA |
24 | | -Blocking the access to the lists is possible with packet inspection due to the traffic being unencrypted |
25 | | -Browser soft-fail by default |
26 | | -OCSP is dead: <https://letsencrypt.org/2022/09/07/new-life-for-crls/> |
27 | | -Work ongoing to fix these issues |
| 21 | +**Therefore, we emphasize that the CA/Browser forum must not use its power to enforce ECH** because it pushes users to DoH. |
28 | 22 |
|
29 | | -### QUIC and SPDY |
| 23 | +## DoH server oligarchy |
30 | 24 |
|
31 | | -should work |
| 25 | +Firefox and Edge use Coudflare's DoH server, and Chrome uses Google's DoH server. |
| 26 | +Just two DoH servers provide DoH services to the majority of browser users. |
| 27 | +This imbalance implies various problem areas: |
32 | 28 |
|
33 | | -## Firefox |
| 29 | +- Privacy: Despite their emphasis that they won't save the query data, these policies can change at any time. Users are locked into a trust dependency without opt-in. |
| 30 | +- Jurisdiction and Geopolitics: The DoH servers are operated by companies in a single country. Their legal system can force them to share the query data at any time, impacting users **worldwide** without any possibility for them to notice. |
| 31 | +- Market dominance: Just two operators control the majority of all DoH traffic. This creates a huge market dominance. In the future, this can lead to them effectively taking over the role of Domain registrars. |
34 | 32 |
|
35 | | -timing issue FF: no new information |
| 33 | +As DoH is required for ECH, these problems are worsened by ECH. |
| 34 | +ECH aims to defend users' privacy, but its dependency on DoH thwarts this goal. |
36 | 35 |
|
37 | | -### Deliberately disabled |
| 36 | +## OCSP and CRL |
38 | 37 |
|
39 | | -The Interop report mentions that ECH is disabled deliberately when local interception software is detected, and also with ISP- or state-level censorship |
| 38 | +Online Certificate Status Protocol (OCSP) and Certificate Revocation Lists (CRL) are core components in the validation of digital WebPKI certificates. The data is typically transmitted over HTTP but verified through signing by the Certificate Authority (CA). |
| 39 | +Access to these lists can be obstructed through packet inspection due to the unencrypted nature of the traffic. Browsers generally employ a soft-fail approach for CRL validations by default, meaning the information about a revoked certificate may not reach the client. If an adversary has the capability to interfere at this stage, they may effectively disrupt other, more important connections as well. |
| 40 | + |
| 41 | +OCSP has recently been called into question and may be nearing obsolescence: <https://letsencrypt.org/2022/09/07/new-life-for-crls/>. As a result, the CRLs' importance increases again. |
| 42 | + |
| 43 | +This study found no relevant interplay of ECH on CRL and OCSP mechanisms. |
0 commit comments