opsec A. J. Campling
Internet-Draft 419 Consulting Limited
Intended status: Informational P. Vixie
Expires: 27 July 2024 Red Barn
D. Wright
UK Safer Internet Centre
A. Taddei
S. Edwards
Broadcom
24 January 2024
Encrypted Client Hello Deployment Considerations
draft-campling-ech-deployment-considerations-08
Abstract
(Editorial note: to be updated as the text in the main body of the
document is finalised) This document is intended to inform the
community about the impact of the deployment of the proposed
Encrypted Client Hello (ECH) standard that encrypts Server Name
Indication (SNI) and other data. Data encapsulated by ECH (ie data
included in the encrypted ClientHelloInner) is of legitimate interest
to on-path security actors including those providing inline malware
detection, parental controls, content filtering to prevent access to
malware and other risky traffic, mandatory security controls etc.
The document includes observations on current use cases for SNI data
in a variety of contexts. It highlights how the use of that data is
important to the operators of both public and private networks and
shows how the loss of access to SNI data will cause difficulties in
the provision of a range of services to end-users, including the
potential weakening of cybersecurity defences. Some mitigations are
identified that may be useful for inclusion by those considering the
adoption of support for ECH in their software.
Status of This Memo
This Internet-Draft is submitted in full conformance with the
provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF). Note that other groups may also distribute
working documents as Internet-Drafts. The list of current Internet-
Drafts is at https://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months
and may be updated, replaced, or obsoleted by other documents at any
time. It is inappropriate to use Internet-Drafts as reference
material or to cite them other than as "work in progress."
This Internet-Draft will expire on 27 July 2024.
Copyright Notice
Copyright (c) 2024 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents (https://trustee.ietf.org/
license-info) in effect on the date of publication of this document.
Please review these documents carefully, as they describe your rights
and restrictions with respect to this document. Code Components
extracted from this document must include Revised BSD License text as
described in Section 4.e of the Trust Legal Provisions and are
provided without warranty as described in the Revised BSD License.
Table of Contents
1. Introduction
1.1. Background
1.2. Scope, objectives and limits of this document
2. General considerations about the encryption of the Client Hello
2.1. About encrypting the Server Name Indication (SNI)
2.2. Why are middleboxes using the SNI?
2.3. The case of Proxies
2.4. Why relying on the SNI and not the DNS
2.5. About the unreliability of the SNI
2.5.1. With TLS 1.2
2.5.2. With TLS 1.3
2.5.3. With TLS 1.3 with ECH extension
3. The Education Sector
3.1. Context
3.2. Why Content Filtering Matters to Schools
3.3. Mitigations
3.4. Implications
4. Impact of ECH in private network contexts (Enterprises or other
organizations)
4.1. Context
4.1.1. The main requirements
4.1.2. A degrading threat landscape
4.2. Additional considerations
4.3. Implications
4.3.1. Examples of regulatory implications
4.3.2. Impact of ECH deployment on Network Security Operations
4.3.3. Specific implications for SMBs
5. Public Network Service Providers
5.1. Context
5.2. Mitigations
5.2.1. Current approaches and procedures
5.2.2. The blocking use case
5.3. Implications
6. General issues
6.1. Threat Detection
6.2. Endpoint security limits
6.3. Network management
6.4. Future operational deployment issues due to the
introduction of the Client Facing servers themselves
6.5. Migration issues
7. Potential further development of this work
7.1. Potential development of this document.
7.2. Potential development outside of the scope of this document
8. Conclusion
9. Security Considerations
10. IANA Considerations
11. References
11.1. Normative References
11.2. Informative References
Appendix A. Acknowledgment
Appendix B. Initial data illustrating SNI unreliability
B.1. Data collections
B.2. Consumer Network Traffic
B.3. Corporate Customer Traffic
B.4. Take away observations
Contributors
Authors' Addresses
1. Introduction
1.1. Background
In order to establish its handshake, the TLS protocol needs to start
with a first handshake message called the Client Hello. As this
handshake message is in clear text, it exposes metadata, e.g. the
Server Name Indication (SNI) which allow middleboxes on path to make
policy decisions, in particular but not only for security reasons.
As part of a wider initiative to achieve end-to-end encryption, a
proposed extension to TLS 1.3 called Encrypted Client Hello (ECH)
[I-D.draft-ietf-tls-esni] is attempting to encrypt all the remaining
metadata in the clear.
There are use cases where encryption of the SNI data may be a useful
precaution to reduce the risk of pervasive monitoring that offers
some benefits (e.g Enterprises offering services for their own
customers will appreciate that their customers' privacy be better
protected). However ECH presents challenges for other use cases
(e.g. Enterprises needing network security controls for compliance
reasons).
The Internet was envisaged as a network of networks, each able to
determine what data to transmit and receive from their peers.
Developments like ECH mark a fundamental change in the architecture
of the Internet, allowing opaque paths to be established from
endpoints to commercial services, some potentially without the
knowledge or permission of the device owners. This change should not
be undertaken lightly given both the architectural impact on the
Internet and potentially adverse security implications for end users.
Given these implications, it certainly should not be undertaken
without either the knowledge of or consultation with end users, as
outlined in [RFC8890].
Whilst it is reasonable to counter that VPNs also establish opaque
paths, a primary difference is that the use of a VPN is a deliberate
act by the user, rather than a choice made by client software,
potentially without either the knowledge and/or consent of the end-
user or device owner.
[RFC7258] discusses the critical need to protect users' privacy when
developing IETF specifications and also recognises that making
networks unmanageable to mitigate pervasive monitoring is not an
acceptable outcome.
[RFC8404] discusses current security and network operations as well
as management practices that may be impacted by the shift to
increased use of encryption to help guide protocol development in
support of manageable and secure networks. As [RFC8404] notes, "the
implications for enterprises that own the data on their networks or
that have explicit agreements that permit the monitoring of user
traffic are very different from those for service providers who may
be accessing content in a way that violates privacy considerations".
1.2. Scope, objectives and limits of this document
This document considers the implications of ECH for private, edge and
public networks using the examples of education establishments,
enterprises and public operators. It addresses the limitations of
[RFC8744] by providing more information about the issues posed by the
introduction of ECH due to the loss of visibility of SNI data on
private networks building on the report from a roundtable discussion
[ECH_Roundtable].
The objective of this document is to detail some operational impacts
of ECH. It will focus specifically on the impact of encrypting the
SNI data by ECH, but it should be noted that other elements in the
client hello may also be relevant for some on-path security methods.
The data encapsulated by ECH is of legitimate interest to on-path
security actors including those providing inline malware detection,
firewalls, parental controls, content filtering to prevent access to
malware and other risky traffic, mandatory security controls (e.g.
Data Loss Prevention) etc. Beyond network security, there are
various operational impacts of different types e.g. network
management, content filtering, etc.
Whilst this document identifies operational issues:
* it does not consider solutions nor question the development of the
ECH proposal itself
* it doesn't attempt to be exhaustive,
* it will start by focusing on one category of middleboxes: proxies.
2. General considerations about the encryption of the Client Hello
2.1. About encrypting the Server Name Indication (SNI)
[RFC8744] describes the general problem of encrypting the Server Name
Identification (SNI) TLS extension. The document includes a brief
description of what it characterises as "unanticipated" usage of SNI
information (section 2.1) as well as a brief (two paragraph)
assessment of alternative options in the event that the SNI data is
encrypted (section 2.3).
The text in [RFC8744] suggests that most of the unanticipated SNI
usage "could also be implemented by monitoring DNS traffic or
controlling DNS usage", although it does then acknowledge the
difficulties posed by encrypted DNS protocols. It asserts, with
limited evidence, that "most of 'the unanticipated usage' functions
can, however, be realized by other means", although without
considering or quantifying the affordability, operational complexity,
technical capability of affected parties or privacy implications that
might be involved. It is unclear from the document whether any
stakeholders that may be impacted by the encryption of SNI data have
been consulted; it certainly does not appear to be the case that any
such consultation has taken place.
The characterisation of "unanticipated usage" of SNI data could be
taken to imply that such usage was not approved and therefore
inappropriate in some manner. The reality is that the development of
the Internet has many examples of permissionless innovation and so
this "unanticipated usage" of SNI data should not be dismissed as
lacking in either importance or validity.
2.2. Why are middleboxes using the SNI?
For middleboxes to be able to perform their job they need to identify
the destination of the requested communication. Before TLS1.3 a
middlebox could rely on, at least, 3 metadata sources: The
certificate, the DNS name and the SNI. A middlebox may have used
some or all of these metadata to determine the destination in the
best possible way. Yet, as part of the current initiative to
complete end-to-end encryption, the certificate was encrypted into
TLS1.3, then DoH/DoT/DoQ are encrypting the DNS flow to its resolver
making it harder for middleboxes to use this information. Even if
the DNS data can be accessed, it can be misleading in some situations
(does it point to the real destination, or just the site hosting
server name, or a proxy?) and the SNI was invented to overcome some
of the limitations of the DNS data by providing additional
information. However, the SNI in itself may be unreliable which is
why middleboxes start by non-trusting it until they have validated
the information that it provides.
2.3. The case of Proxies
A proxy server is a server application that acts as an intermediary
between a client requesting a resource and the server providing that
resource. Instead of connecting directly, the client directs the
request to the proxy server which evaluates the request before
performing the required network activity. Proxies are used for
various purposes including load balancing, privacy and security.
Proxies can be used explicity or transparently.
In explicit proxy model, proxies are accessed by configuring a user's
application or network settings, with traffic diverted to the proxy
rather than the target destination.
With "transparent" proxying, the proxy intercepts packets directed to
the destination, making it seem as though the request is handled by
the target destination itself.
A key advantage of transparent proxies is that they work without
requiring the configuration of user devices or software. They are
commonly used by organisations to provide content filtering for
devices that they don't own that are connected to their networks.
For example, some education environments use transparent proxies to
implement support for “bring your own device” (BYOD) without needing
to load software on third- party devices.
Transparent proxies use SNI data to understand whether a user is
accessing inappropriate content without the need to inspect data
beyond the SNI field. Because of this, encryption of the SNI field,
as is the case with ECH, will disrupt the use of transparent proxies,
requiring far more intrusive data inspection to be undertaken
instead.
2.4. Why relying on the SNI and not the DNS
As per the Introduction section of [RFC8744] "More and more services
are colocated on multiplexed servers," the SNI was introduced to
allow "direct connections to the appropriate service implementation".
By design, a proxy cannot rely only on the DNS to ensure establishing
the connection, the SNI is simply required by design.
2.5. About the unreliability of the SNI
SNI is by design not reliable therefore prompting the question for
why and how middelboxes are using it in the first place. Before
anything, in Security, in general, unreliability is a useful source
of information in itself.
Referring to [RFC6066], TLS extenisons, including SNI, are designed
to be backward compatible. This means that if the server doesn't
recognize the SNI value, the TLS handshake should continue anyway.
In other terms, SNI value can 1) be empty or 2) have an alternative
name which is different from the real name of the destination server
without impacting the establishment of the TLS session. This can
also be easily exploited by bad actors indeed to bypass security
middle boxes. E.g A malware could just be coded to provide an SNI
value that is mapped to finance or healthcare categories to bypass
inspection (it is not that easy, but there is a way to do it).
2.5.1. With TLS 1.2
User client generates “ClientHello” with SNI in plain text to the
destination server. If accepted, the server responds to the user
client request with the “ServerHello” message containing the intended
server certificate alongside other encryption-related information in
the plain text mode as well. Middleboxes can then see/inspect 1/ the
SNI in the “ClientHello” and 2/ the server certificate details in the
“ServerHello” message. Middleboxes can perform selective inspection
based on the destination service details (service requested by the
user client) extracted from the server certificate. I would like to
note that depending on the middlebox design, it is more about a
correlation of different source of information to confirm the right
destination service. But still only information within the server
certificate is reliable to perform accurate web categorization and
then do selective inspection of any kind of web/content filtering.
2.5.2. With TLS 1.3
TLS 1.3 improves significantly over the TLS 1.2 in terms of security
and privacy aspects. More specifically, in terms of privacy, it
overcomes the plain text server certificate exchange issue by masking
the server’s host identity through the encrypted server certificate.
So as middleboxes inspection capabilities are designed based on the
server certificate, all vendors worked on how to adapt their
capabilities to support TLS 1.3 (e.g ServerHello encryption but not
only).
So how do providers support TLS 1.3 inspection then?
The most common technique is first to get the SNI from the
ClientHello (which is still shared in plain text format.) and then
replays / establish a new full TLS session initiated from the proxy
server itself to the destination server to retrieve the server
certificate details before determining the web category; Once
identified, selective inspection can be performed on the real TLS
session initiated by the user client.
As the SNI is not reliable, proxies accept the SNI asis but do
without trusting it, then they perform checks at various level to
verify this SNI and they step by step enrich the evaluation,
therefore bringing more possibilites to interpret which policy to
apply. This could end up with the proxy deciding to block the
connection, or the proxy to let the connection happened with a
verified or corrected SNI.
See Appendix for some initial data gathering on the reliability of
the SNI.
2.5.3. With TLS 1.3 with ECH extension
The entire legacy ClientHello message (Inner ClientHello) is
encrypted, encapsulated and sent as part of the new ClientHello
wrapper message (Outer ClientHello); So middleboxes cannot identity
the destination service anymore and cannot replay the TLS session to
the destination server because it has no clue about it. (May be DNS
knows but not all the time, see other issue here created by myself).
So TLS ECH has an impact on enterprise security and compliance
(including selective inspection) not because of SNI (which is not
reliable) encryption but because there is no other information about
the destination server is available and that can be used to fetch and
retrieve the server certificate (Which is indeed reliable) to perform
web categorization.
We don’t care about SNI and its reliability, we need just to know the
destination service to get the destination server certificate
details.
3. The Education Sector
3.1. Context
Focusing specifically on the education sector, the primary issue
caused by ECH is that it is likely to circumvent the safeguards
applied to protect children through content filtering, whether in the
school or home environments, adding to adverse impacts already
introduced through the use of encrypted DNS protocols such as DNS
over HTTPS [RFC8484].
Content filtering that leverages SNI information is used by education
establishments to protect children from exposure to malicious, adult,
extremist and other content that is deemed either age-inappropriate
or unsuitable for other reasons. Any bypassing of content filtering
by client software on devices will be problematic and may compromise
duties placed on education establishments. For example: schools in
England and Wales have obligations to provide "appropriate filtering
systems" [KCSE]; schools in the US use Internet filters and implement
other measures to protect children from harmful online content as a
condition for the receipt of certain federal funding, especially
E-rate funds [CIPA].
3.2. Why Content Filtering Matters to Schools
The impact that ineffective content filtering can have on an
educational institutions should not be underestimated. For example,
a coroner in the UK in 2021 ruled that a school's failure to prevent
a pupil from accessing harmful material online on its equipment
contributed to her taking her own life [Coroner]. In this particular
instance, the filtering software installed at the school was either
faulty or incorrectly configured but the case highlights the harmful
risks posed if the filtering is bypassed by client software using
ECH.
3.3. Mitigations
Whilst it may be possible for schools to overcome some of the issues
ECH raises by adopting similar controls to those used by enterprises,
it should be noted that most schools have a very different budget for
IT compared to enterprises and usually have very limited technical
support capabilities. Therefore, even where technical solutions
exist that may allow them to continue to meet their compliance
obligations, affordability and operational expertise will present
them with significant difficulties.
Absent funding and technical expertise, schools will need to consider
the best way forward that allows them to remain compliant. If client
software does not allow ECH to be disabled, any such software that
implements support for ECH may need to be removed from school devices
and replaced, assuming that suitable alternatives are available.
This will have a negative impact on budgets and may be operationally
challenging if institutions have made a significant investment in the
deployment and use of particular applications and technologies.
There are instances where policies in education establishments allow
for the use of equipment not owned by the institution, including
personal devices and the devices of contractors and site visitors.
These devices are unlikely to be configured to use the institution's
proxy but can nevertheless connect to the school network using a
transparent proxy (see below). Transparent proxies used for
filtering will typically use SNI data to understand whether a user is
accessing inappropriate data, so encrypting the SNI field will
disrupt the use of these transparent proxies.
3.4. Implications
In the event that transparent proxies are no longer effective,
institutions will either have to require more invasive software to be
installed on third party devices before they can be used along with
ensuring they have the capability to comprehend and adequately manage
these technologies or will have to prevent those devices from
operating. Neither option is desirable.
4. Impact of ECH in private network contexts (Enterprises or other
organizations)
4.1. Context
4.1.1. The main requirements
Enterprises and Organizations need to protect themselves for a vast
number of reasons, mainly:
* Reduce their Risks. And in particular as part of any Cyber
Resilience strategy.
* Protect their Reputation. The term Reputation includes many
aspects way beyond the traditional enterprises and organization
assets (data, etc.).
* Comply to a growing diverse set of Policies, Regulations,
Certifications, Labeling and Guidelines. These requirements are
growing in both scope and complexity as they are added to by
various national and regional bodies around the world.
4.1.2. A degrading threat landscape
In addition, the general threat landscape which was already very
large (see [I-D.draft-mcfadden-smart-threat-changes]), has
significantly increased in three ways:
* COVID crisis generally accelerated the overall attack landscape.
Indeed as the crisis forced many enterprises and organizations to
accelerate their digital transformation, it increased the
opportunity for cyber criminals and nation states to launch more
attacks, leverage innovations to their advantage, better select
their targets, increase their efficiency and increase their
rewards, in particular with Ransomware based attacks.
* The Supply Chain is under stress as per the [SOLARWIND] attack
* Nation State attacks are continuing to evolve, for example as
noted to those linked to the current Ukraine crisis.
Attacks are now damaging enterprises and other organizations with
ransomware being the number 1 issue by a considerable margin. The
attacks are increasing in severity, to the extent that this is now
being measured at macroscopic level in some countries:
* EUR1B loss of revenue for French organizations from January to
August 2022 [LOSSINREVENUE]
* Loss in capitalisation between 1-5% [LOSSINCAP]
* Degradation by credit notation agencies [LOSSINCREDITSCORE]
Another implication from the COVID crisis is the acceleration of BYOD
with the current reliance on remote working. This has created two
side effects for remote employees, contractors and third parties that
need to connect to one or more enterprise networks on a temporary
basis:
* need to use a VPN access to the corporate network, which brings
all the benefits (e.g. protected access to corporate network) and
risks that VPNs may open (e.g. lateral movement when the end point
is compromised),
* need to access a cloud proxy which requires an agent to be
installed on the device to steer the traffic to the right place.
4.2. Additional considerations
In such circumstances, requiring software or custom configurations to
be installed on those devices may be problematic (see
[I-D.draft-taddei-smart-cless-introduction]).
This is why network security solutions are required and this is why
the use of ECH to prevent access to the SNI data makes it impossible
for blue teams to defend (see the next sections for details).
Finally there is a global shortage of cybersecurity personnel. Any
expansion of technical requirements, for example to mitigate the
operational challenges through the introduction of ECH, will
exacerbate the problem.
All the above conditions are weighing on capabilities to defend,
both:
* Directly: a lack of visibility on a key meta data like the SNI
will cause significant issues to enterprises and organizations
* Indirectly: should ECH happen and should alternative be provided,
managing migrations to any alternative not requiring access to the
SNI, in these conditions, is undesirable from a timing, resources,
capacities and risks perspectives.
4.3. Implications
4.3.1. Examples of regulatory implications
Regulators are accelerating their lawfare capabilities at accelerated
pace and new legislations is impacting on the actions of enterprises
with increased precision. The EU GDPR had ripple effects such as
requiring Financial Institutions to use selective decrypt in order to
implement Data Loss Prevention. The recent indication that US
regulators are in the process of levying fines of $200m each on a
number of institutions because they were unable to track all
communications by their employees using WhatsApp or Signal ,
[Bloomberg], creates new auditability constraints. It is with
growing concern that an ECH enabled ecosystem may clash with future
regulatory requirements.
4.3.2. Impact of ECH deployment on Network Security Operations
Enterprises approach to endpoint control varies significantly
depending on size, use cases and a vast number of other factors.
For example, large enterprises generally exert control over their
endpoints, yet to the limits of some use cases they may need to
implement, e.g. BYOD. The latter was accelerated, as per above, due
to COVID forcing more flexibility in the extended work force
(employees, contractors, etc.).
On the contrary, small and some medium businesses may not be in the
position to control their endpoints to the same extent (see specific
implications for SMBs section below).
As some Browser makers made the use of ECH optional, this gives a
first approach for enterprises to disable ECH for their employees.
However this doesn't provide an holistic solution. Indeed
enterprises will need to consider a number of issues:
* Browsers which do not offer an option to disable ECH
* Browsers that will make ECH non optional in the future
* Non-browsers applications which are designed to use libraries
enforcing ECH, without any option to disable it
* All the range of BYOD use cases where enterprises do not control
the endpoint
* Adversaries leveraging ECH e.g. to hide their command and control
communications, e.g. in Ransomware cases.
Whilst, disabling ECH wherever possible provides one approach to
mitigate ECH deployment issues, as per above, other mitigations
approaches need to be offered to enterprises.
(Editor's note: we need to describe how to strip the RRs to force a
global disabling of ECH, yet mindful it might not be sufficient if an
adversary finds a way to not use the enterprise DNS resolver)
4.3.2.1. Reminders on Network Security
Network Security is a set of security capabilities which is
articulated as part of a defense strategy, e.g. Defense In Depth
[NIST-DID], Zero Trust, SASE/SSE, etc. and can trigger and enable
other security capabilities such as sandboxing, Data Loss Prevention,
Cloud Access Service Broker (CASB), etc. One constituency is a Web
Proxy, combining both a TLS proxy and an application level (HTTP)
proxy.
In the same way that [I-D.draft-ietf-opsec-ns-impact] showed the
impact of TLS1.3 on operational security, a loss of visibility of the
SNI as indicator of compromise (see
[I-D.draft-ietf-opsec-indicators-of-compromise]) has two main
implications
4.3.2.2. Implications from loss of Meta Data
The loss of visibility of the SNI, at TLS level, will prevent
transparent proxies from applying corporate policies to manage risk
and compliancy. Typical examples:
* categories of compromised sites cannot be applied anymore,
exposing employees and their organisations to potential
cybersecurity risks; alternative approaches to block access to
theses sites need to be found
* corporate lists of excluded sites for compliance or policy reasons
need alternatives ways to be blocked.
4.3.2.3. Implications from loss of Selective Decrypt
TLS proxies also have the ability to selectively intercept, avoiding
any visibility into or modification of the original application
protocol payload - but such selective intercept relies heavily on
knowledge of the origin content server hostname, which can be
extracted in plaintext from the TLS ClientHello SNI (server name)
field.
This capability allows the application proxy, in particular an HTTPS
proxy to engage efficiently specific security controls, e.g. Data
Loss Prevention, Sandboxing, etc.
The loss of SNI visibility will make it more difficult for corporate
user flows to be intercepted, with it becoming impossible for BYOD
use cases.
This will create inefficiencies, will require more resources and will
increase security risks. It will also be counter productive for
privacy as it may require the proxy to decrypt the whole TLS
connection.
4.3.3. Specific implications for SMBs
Small and Medium Business (SMBs) form a particularly vulnerable
subset of enterprises and organizations and span from Small Office
Home Office (SOHO, sometimes a one person business) to Medium
Business with strong variations depending on the country (a 50
employee company is considered the upper range of SMB business in
developing countries while it is up to 25,000 in some developed
countries).
Similarly to the above education use case and irrespective of
definitions, many SMBs have very limited in-house capabilities to
defend themselves, with security often outsourced to Managed Security
Service Providers (typically network operators, mid range and small
service providers).
5. Public Network Service Providers
5.1. Context
In Public Networks the national, regional and international
legislator has to balance between freedom of access to the
information on the one hand, and safety of the Internet and the
protection of other fundamental rights on the other hand.
There are 2 main approaches:
* First, there are countries which do not have any specific
legislation on the issue of blocking, filtering and takedown of
illegal Internet content: there is no legislative or other
regulatory system put in place by the state with a view to
defining the conditions and the procedures to be respected by
those who engage in the blocking, filtering or takedown of online
material. In the absence of a specific or targeted legal
framework, several countries rely on an existing "general" legal
framework that is not specific to the Internet to conduct what is,
generally speaking, limited blocking or takedown of unlawful
online material. Here the approach has been differentiated in
relying on self regulation from the private sector or limited
political or legislative intervention to specific areas.
* The other approach has been to set up a legal framework
specifically aimed at the regulation of the Internet and other
digital media, including the blocking, filtering and removal of
Internet content. Such legislation typically provides for the
legal grounds on which blocking or removal may be warranted, the
administrative or judicial authority which has competence to take
appropriate action and the procedures to be followed.
5.2. Mitigations
5.2.1. Current approaches and procedures
In relation to specific areas where the public interest has to be
protected more strongly, such as child abuse crimes, terrorism,
criminality and national security, many states have a framework for
the urgent removal of Internet content regarding the above materials
without the need of a court order. In such circumstances,
administrative authorities, police authorities or public prosecutors
are given specific powers to order Internet access providers to block
access without advance judicial authority. It is common to see such
orders requiring action on the part of the Internet access provider
within 24 hours, and without any notice being given to the content
provider or host themselves.
Particularly in relation to material concerning child abuse and other
serious crimes, many countries adopt a “list” system, whereby a
central list of blocked URLs or domain names are maintained and
updated by the relevant administrative authority. This is notified
to the relevant Internet access providers, who are required to ensure
that blocking is enforced. Additionally in some states the
authorities can request the removal of content that infringes
intellectual property, privacy or defamation rights. In this case
the removal need to be requested by a court order.
Generally speaking, the grounds relied on broadly correspond to the
interests protected under Article 10(2) of the European Convention of
Human Rights (ECHR), namely: the protection of national security,
territorial integrity or public safety, the prevention of disorder or
crime, the protection of health or morals, the protection of the
reputation or rights of others, and the prevention of the disclosure
of information received in confidence. From the methodology we have
to distinguish between blocking or takedown of content.
* The blocking, filtering or prevention of access to Internet
content are generally technical measures intended to restrict
access to information or resources typically hosted in another
jurisdiction. Such action is normally taken by the Internet
access provider through hardware or software products that block
specific targeted content from being received or displayed on the
devices of customers of the Internet access provider.
* Takedown or removal of Internet content, on the other hand, will
instead broadly refer to demands or measures aimed at the website
operator (or "host") to remove or delete the offending website
content or sub content.
In these considerations we will refer to blocking only.
5.2.2. The blocking use case
This can be achieved through a number of techniques, including the
blocking of the Domain Name System (DNS), the analysis of the SNI
field or the Uniform Resource Locator (URL). Given the increasing
adoption of encryption techniques often a mixture of the above
techniques is needed.
In particular for the most serious crimes such as child abuse or
national security many countries adopt a “list” methodology, where a
central list of blocked Domains or URLs is maintained by the
authorities and updated on a regular basis (daily or even hourly) and
shared with Public Network Operators that have to enforce the
blocking.
In many jurisdictions there are legal consequences for the Operator
not complying with the blocking order.
Technically the blocking can be implemented using techniques that
have been adapted over time as new technologies have been introduced.
Historically depending on the content of the list the technique have
been based on DNS or proxy blocking.
DNS is effective on Domains (the whole domain is blocked), while
proxy is effective either on Domain (for encrypted traffic) or URL
(for unencrypted traffic).
Given that nowadays the vast majority of traffic is encrypted, the
capability of blocking based on URL is limited to a small portion of
traffic and proxy blocking is as effective as that based on the DNS.
Theoretically DNS blocking would be the preferred option for
operators given the more limited investments necessary to implement
blocking of the Domains, but given the increased usage of external
encrypted DNS services DNS blocking is becoming less effective and
operators need to use SNI analysis as well in order to fulfil legal
obligations.
5.3. Implications
The adoption of ECH will cause additional problems and limit the
possibility of implementing operators fulfilling their legal blocking
obligations, exposing the population to illegal content related to
crimes such as Child Sex Abuse Material (CSAM), malware and other
malicious content, and possibly even content deemed to be detrimental
to National Security.
In addition, operators that do not fulfil their legal obligations may
be exposed to legal or regulatory remedies.
6. General issues
6.1. Threat Detection
[RFC8404] identifies a number of issues arising from increased
encryption of data, some of which apply to ECH. For example, it
notes that an early trigger for DDoS mitigation involves
distinguishing attacker traffic from legitimate user traffic; this
become more difficult if traffic sources are obscured.
The various indicators of compromise (IoCs) are documented in
[I-D.draft-ietf-opsec-indicators-of-compromise], which also describes
how they are used effectively in cyber defence. For example, section
4.1.1 of the document describes the importance of IoCs as part of a
defence- in-depth strategy; in this context, SNI is just one of the
range of indicators that can be used to build up a resilient defence
(see section 3.1 in the same document on IoC types and the 'pyramid
of pain').
In the same Internet-Draft, section 6.1 expands on the importance of
the defence in depth strategy. In particular, it explains the role
that domains and IP addresses can play, especially where end-point
defences are compromised or ineffective, or where endpoint security
isn't possible, such as in BYOD, IoT and legacy environments. SNI
data plays a role here, in particular where DNS data is unavailable
because it has been encrypted; if SNI data is lost too, alongside
DNS, defences are weakened and the attack surface increased.
6.2. Endpoint security limits
(Editorial note: Elaborate on endpoint security complications as
[I-D.draft-taddei-smart-cless-introduction] as well as [MAGECART]
[MITB] [MITB-MITRE] [MALVERTISING] showed that in some cases, the
only way to detect an attack is through the use of network-based
security. The loss of visibility of the SNI data will make it much
harder to detect attacks. The endpoints components (operating
system, applications, browsers, etc.) cannot be judge and jury.)
6.3. Network management
(Editorial note: this is a placeholder for future issues)
6.4. Future operational deployment issues due to the introduction of
the Client Facing servers themselves
(Editorial note: this is a placeholder for future issues;
* Consolidation considerations - the use of ECH may accelerate the
move of content away from standalone servers and on to CDNs,
reducing infrastructure resilience.
* What happens if Client Facing servers are controlled by malicious
actors?
* The Client Facing servers are acting as a new category of
middleboxes. In this shift left movement, until the attack
surface is minimal and complexities are removed, you have to rely
on third parties for inspection. In these conditions, on which
basis can they be more trusted than any other middleboxes? Is
this creating a concentration problem?
)
6.5. Migration issues
(Editorial note: this is a placeholder for future issues;
* If ECH is enforced what are the solutions to all the above
problems and what are the migration paths?
)
7. Potential further development of this work
7.1. Potential development of this document.
This section lists potential development of this work in particular
for the General Issues section.
* There are need for further clarifications from the ECH draft, e.g.
The link between the Client Facing and the backend servers are not
clear enough and need further description. It can't be just 'left
to the implementation'. The action is still underway and feedback
to the TLS working group will be provided.
* Will there be any impact to the DNS by adding so many new RRs?
7.2. Potential development outside of the scope of this document
This document infers a number of ideas that could be relevant for
other groups and in other deliverables. In particular regarding what
type of solutions could be considered
* There is a need to address the apparent disconnect between user
privacy and security, it should be possible to provide both rather
than one compromising the other.
* What prevents a Client Facing server providing security solutions
to protect the data path?
* Given some of the many challenges there is the opportunity to
review the current ECH proposal from the perspective of a
respectful inspection protocol.
8. Conclusion
Access to SNI data is sometimes necessary in order for institutions,
including those in the education and finance sectors, to discharge
their compliance obligations. The introduction of ECH in client
software poses operational challenges that could be overcome on
devices owned by those institutions if policy settings are supported
within the software that allows the ECH functionality to be disabled.
(Editorial note: these two below paragraph need revision towards the
end of the development of this draft)
Third-party devices pose an additional challenge, primarily because
the use of ECH will render transparent proxies inoperable. The most
likely solution is that institutions will require the installation of
full proxies and certificates on those devices before they are
allowed to be connected to the host networks. They may alternatively
determine that such an approach is impractical and instead withdraw
the ability for network access by third-party devices.
An additional option that warrants further consideration is the
development of a standard that allows a network to declare its policy
regarding ECH and other such developments. Clients would then have
the option to continue in setting up a connection if they are happy
to accept those policies, or to disconnect and try alternative
network options if not. Such a standard is outside of the scope of
this document but may provide a mechanism that allows the interests
and preferences of client software, end-users and network operators
to be balanced.
9. Security Considerations
In addition to introducing new operational and financial issues, the
introduction of SNI encryption poses new challenges for threat
detection which this document outlines.
This I-D should help improve security in deployments of ECH.
10. IANA Considerations
This document has no IANA actions.
11. References
11.1. Normative References
[RFC6066] Eastlake 3rd, D., "Transport Layer Security (TLS)
Extensions: Extension Definitions", RFC 6066,
DOI 10.17487/RFC6066, January 2011,
<https://www.rfc-editor.org/rfc/rfc6066>.
[RFC8484] Hoffman, P. and P. McManus, "DNS Queries over HTTPS
(DoH)", RFC 8484, DOI 10.17487/RFC8484, October 2018,
<https://www.rfc-editor.org/rfc/rfc8484>.
11.2. Informative References
[Bloomberg]
Spezzati, S., Robinson, M., and L. Beyoud, "Wall Street's
Record Fines Over WhatsApp Use Were Years in the Making",
16 August 2022, <https://www.bloomberg.com/news/
articles/2022-08-16/wall-street-sticker-shock-whatsapp-
fines-were-years-in-making>.
[CIPA] FCC, "Children's Internet Protection Act (CIPA)", 30
December 2019, <https://www.fcc.gov/consumers/guides/
childrens-internet-protection-act/>.
[Coroner] Henderson, "Prevention of future deaths report", 26
November 2021, <https://www.judiciary.uk/publications/
frances-thomas-prevention-of-future-deaths-report/>.
[ECH_Roundtable]
419 Consulting, "Encrypted Client Hello - Notes from an
ECH Roundtable", 18 August 2021,
<https://419.consulting/encrypted-client-hello/>.
[I-D.draft-ietf-opsec-indicators-of-compromise]
Paine, K., Whitehouse, O., Sellwood, J., and A. S,
"Indicators of Compromise (IoCs) and Their Role in Attack
Defence", Work in Progress, Internet-Draft, draft-ietf-
opsec-indicators-of-compromise-04, 3 February 2023,
<https://datatracker.ietf.org/doc/html/draft-ietf-opsec-
indicators-of-compromise-04>.
[I-D.draft-ietf-opsec-ns-impact]
Cam-Winget, N., Wang, E., Danyliw, R., and R. DuToit,
"Impact of TLS 1.3 to Operational Network Security
Practices", Work in Progress, Internet-Draft, draft-ietf-
opsec-ns-impact-04, 26 January 2021,
<https://datatracker.ietf.org/doc/html/draft-ietf-opsec-
ns-impact-04>.
[I-D.draft-ietf-tls-esni]
Rescorla, E., Oku, K., Sullivan, N., and C. A. Wood, "TLS
Encrypted Client Hello", Work in Progress, Internet-Draft,
draft-ietf-tls-esni-17, 9 October 2023,
<https://datatracker.ietf.org/doc/html/draft-ietf-tls-
esni-17>.
[I-D.draft-mcfadden-smart-threat-changes]
McFadden, M., "BCP72 - A Problem Statement", Work in
Progress, Internet-Draft, draft-mcfadden-smart-threat-
changes-04, 22 January 2022,
<https://datatracker.ietf.org/doc/html/draft-mcfadden-
smart-threat-changes-04>.
[I-D.draft-taddei-smart-cless-introduction]
Taddei, A., Wueest, C., Roundy, K. A., and D. Lazanski,
"Capabilities and Limitations of an Endpoint-only Security
Solution", Work in Progress, Internet-Draft, draft-taddei-
smart-cless-introduction-03, 13 July 2020,
<https://datatracker.ietf.org/doc/html/draft-taddei-smart-
cless-introduction-03>.
[KCSE] DfE, "Keeping children safe in education 2021", 1 November
2021, <https://419.consulting/encrypted-client-hello/>.
[LOSSINCAP]
Neyret, A. and Autorité des Marchés Financiers, "La
cybercriminalité boursière – définition, cas et
perspectives", 10 October 2019, <https://www.amf-
france.org/sites/default/files/2020-02/etude-sur-la-
cybercriminalite-boursiere-_-definition-cas-et-
perspectives.pdf>.
[LOSSINCREDITSCORE]
Deloitte, "Beneath the surface of a cyberattack – A deeper
look at business impacts", 2016,
<https://www2.deloitte.com/content/dam/Deloitte/global/
Documents/Risk/gx-risk-gra-beneath-the-surface.pdf>.
[LOSSINREVENUE]
ANOZR WAY, "BAROMÈTRE ANOZR WAY DU RANSOMWARE", 4
September 2022, <https://anozrway.com/wp-
content/uploads/dlm_uploads/2022/09/ANOZR-WAY_Barometre-
Ransomware_edition-septembre-2022.pdf>.
[MAGECART] Wikipedia, "Magecart", 3 April 2022,
<https://en.wikipedia.org/wiki/Web_skimming#Magecart>.
[MALVERTISING]
Wikipedia, "Malvertising", 2 June 2022,
<https://en.wikipedia.org/wiki/Malvertising>.
[MITB] OWASP, "Man-in-the-browser attack", n.d.,
<https://owasp.org/www-community/attacks/Man-in-the-
browser_attack>.
[MITB-MITRE]
MITRE, "Browser Session Hijacking - T1185", 25 February
2022, <https://attack.mitre.org/techniques/T1185/>.
[NIST-DID] NIST, "Glossary - defense-in-depth", n.d.,
<https://csrc.nist.gov/glossary/term/defense_in_depth#:~:t
ext=Definition(s)%3A,and%20missions%20of%20the%20organizat
ion.>.
[RFC7258] Farrell, S. and H. Tschofenig, "Pervasive Monitoring Is an
Attack", BCP 188, RFC 7258, DOI 10.17487/RFC7258, May
2014, <https://www.rfc-editor.org/rfc/rfc7258>.
[RFC8404] Moriarty, K., Ed. and A. Morton, Ed., "Effects of
Pervasive Encryption on Operators", RFC 8404,
DOI 10.17487/RFC8404, July 2018,
<https://www.rfc-editor.org/rfc/rfc8404>.
[RFC8744] Huitema, C., "Issues and Requirements for Server Name
Identification (SNI) Encryption in TLS", RFC 8744,
DOI 10.17487/RFC8744, July 2020,
<https://www.rfc-editor.org/rfc/rfc8744>.
[RFC8890] Nottingham, M., "The Internet is for End Users", RFC 8890,
DOI 10.17487/RFC8890, August 2020,
<https://www.rfc-editor.org/rfc/rfc8890>.
[SOLARWIND]
Symantec, a Division of Broadcom Software Group,
"SolarWinds (Sunburst) Attack What You Need to Know",
December 2020, <https://symantec.broadcom.com/en/
solarwinds-sunburst-attacks>.
Appendix A. Acknowledgment
In memory of Simon Edwards who passed away in the night of 8th-9th of
January 2023.
In addition to the authors, this document is the product of an
informal group of experts including the people listed in the
Contributors list in Appendix.
Appendix B. Initial data illustrating SNI unreliability
B.1. Data collections
In this appendix a couple data sets were collected from SSL Session
logs from a Symantec SSLV. The goal was to see how prevalent are TLS
sessions being established where the Server Name Indicator (SNI) was
incorrect as compared to the Subject Alternative Name (SAN) contained
within the Server Certificate. Applications and browsers that are
establishing these mismatched connections have TLS Hygiene issues.
In other words these sessions are being improperly established. None
of the traffic was for nefarious means. However, an improperly
defined SNI can be used to fool inspection devices to bypass security
rules and measures.
The first set was based on consumer traffic which includes Internet
of Things, Social Media, and Corporate access traffic. The dataset
of session log entries were over 63K event entries over a 24 hour
period.
The second dataset was from a telecommunications customer with Proxy
Offloading. The log entries were for a period of 24 hours and
contained over 5M log events entries. Since this customer is using a
Symantec Edge Proxy aligned with SSLV. The session data is for
explicit clients. Guest or Internet of things type traffic is a much
lower percentage of total traffic. However the existence of
Mismatched SNIs persisted.
B.2. Consumer Network Traffic
For consumer based network traffic Mismatched SNIs were very
prevalent. Actually out of new sessions the majority of the sessions
were with mismatched SNIs as compared to properly matched SNIs.
These were the result of many many short lived TLS sessions that
persistently 'phone home'. 22% of all traffic was mismatched as
compared to only 4% was properly matched. The rest of the log
activity was non session related.
The services that were in the top 20 offenders for mismatched SNIs
includes Google, Apple, Adware, IoT. Google DNS dominated the counts
at 5.3% of all mismatched sessions. A close second at 4.8% was
Samsung Smart things. But, if I add up common services like Google,
Apple, Adware, and the remainder: 13%, 9.8%, 8%, and 10%
respectively.
For matched traffic Amazon Alexa stood tall at 25%, with a runner up
for Broadcom Cloud Proxy at 7.9%. Both Google and Apple did have a
minority of properly established sessions. In other words some of
their stuff is clean.
B.3. Corporate Customer Traffic
Since the traffic was proxy traffic the session hygiene was much
better. The majority of new TLS sessions were properly established
with matching SNIs/SANs. The vast majority of this traffic was VPN
based. Likely masking out consumer like traffic invisible within the
VPN tunnels. For instance there was 0.8% Google.com traffic
reported. But, there was a small percentage ~1.3% Google API traffic
that was mismatched. I would have expected more search engine
activity.
Looking across the distribution of domain names for mismatched
sessions. 29.6% of the traffic was related to Corporate applications.
These applications could be updated and corrected. The next runner
up coming in at 7.4% was for Akamai which also could be updated.
Office applications and the remainder each individually accounted for
2.4% of the traffic.
B.4. Take away observations
* IoT and API based traffic is by far the largest offender as
compared to Browser based initiated sessions.
* Long-lived TLS session counts were dwarfed by the chatter of the
API calls using short lived yet pervasively reporting. For
instance there were new sessions at a rate of every 20 seconds per
IoT device.
* The consumer mismatched sessions were all using TLS v1.3. Which
reaffirmed the need to decrypt TLS v1.3 traffic. These sessions,
if established without TLS interception, may have gone unreported
by NGFW, which makes policy decisions on SNI vs SAN. Conversely,
the corporate traffic was a close split between TLS v1.3 and v1.2.
* The presence of VPN tunnels masked a clearer picture of the
corporate traffic usage.
* SNI Mismatches are more prevalent in the wild than first thought.
* SNI Mismatches has a side cause which policy rules have to be
enumerated twice for category matching. And the second category
matching is more intensive since it has to enumerate the entire
SAN list which can be very large.
* Fixing the session hygiene for corporate owned applications could
potentially improve performance of the security stack.
Contributors
Eric Chien
Broadcom
Email: Eric.Chien@broadcom.com
URI: https://www.linkedin.com/in/eric-chien-66b4b258/
Eric contributed to the analysis of the Man in the Browser attacks.
Gianpaolo Scalone
Vodafone
Email: gianpaolo-angelo.scalone@vodafone.com
URI: https://www.linkedin.com/in/gianpaoloscalone/
Contributed the research on the conflicts of ECH with local
legislations to block.
Daniel Engberg
Skandinaviska Enskilda Banken AB (SEB)
Email: daniel.engberg@seb.se
URI: https://www.linkedin.com/in/daniel-engberg-1561aaa/
Validate the issues for his organization.
Celine Leroy
Eight Advisory
Email: celine.leroy@8advisory.com
URI: https://www.linkedin.com/in/celine-leroy-1a534252/
Thank you to Céline for her work on cybersecurity financial impacts
on enterprises.
Daniel Engberg
Skandinaviska Enskilda Banken AB (SEB)
Email: daniel.engberg@seb.se
URI: https://www.linkedin.com/in/daniel-engberg-1561aaa/
Validate the issues for his organization.
Gianpiero Tavano
Broadcom
Email: Gianpiero.Tavano@broadcom.com
URI: https://www.linkedin.com/in/gianpiero-tavano-5b975383/
Review the text, provided feedback and reminded us on the budgetary
issues
Roelof duToit
Broadcom
Email: roelof.dutoit@broadcom.com
URI: https://www.linkedin.com/in/roelof-du-toit-a66831/
Roelof contributed many things including research, former I-D, text,
the newly setup github, etc.
Diego Lopez
Telefonica
Email: diego.r.lopez@telefonica.com
URI: https://www.linkedin.com/in/dr2lopez/
Diego contributed in several aspects including MCPs.
Gary Tomic
Broadcom
Email: gary.tomic@broadcom.com
URI: https://www.linkedin.com/in/garytomic/
Gary contributed many things including research, keep us on scope,
critique for when issues where not impacted by ECH as we initially
thought.
Bob Blair
Broadcom
Email: bob.blair@broadcom.com
URI: https://www.linkedin.com/in/bob-blair-8b7273/
Bob contributed to several reviews, many calls, and the whole
appendix A.
Pascal Paisant
BNP Paribas
Email: pascal.paisant@bnpparibas.com
URI: https://www.linkedin.com/in/pascal-paisant-727a531/
Pascal contributed to several parts, in particular in the general SNI
section, on enterprises section and on migration issues.
Zied Turki
ZT Consulting
Email: zied.turki@ztconsulting.fr
URI: https://www.linkedin.com/in/zied-turki/
Zied contributed to several parts, in particular the rationale on SNI
unreliability.
Authors' Addresses
Andrew Campling
419 Consulting Limited
Email: Andrew.Campling@419.Consulting
URI: https://www.419.Consulting/
Paul Vixie
Red Barn
Email: paul@redbarn.org
URI: http://www.redbarn.org/
David Wright
UK Safer Internet Centre
Email: david.wright@swgfl.org.uk
URI: https://saferinternet.org.uk/
Arnaud Taddei
Broadcom
1320 Ridder Park Dr
San Jose, CA 95131
United States of America
Phone: 41795061129
Email: Arnaud.Taddei@broadcom.com
URI: https://www.linkedin.com/in/arnaudtaddei/
Simon Edwards
Broadcom
1320 Ridder Park Dr
San Jose, CA 95131
United States of America
Email: Simon.Edwards@broadcom.com
URI: https://www.linkedin.com/in/simononsecurity/