However, both studies do not cache intermediate certificates as part of their validation strategy, which skews the results.
yeah what an obvious and glaring mistake for them to make. i have no fucking clue what this means and they don't explain
Timeline
Post
Remote status
Context
14it keeps getting more and more surveillance coded lmao
After a brief characterization of the data set, the authors devise a classifier that separates malicious certificates from benign, based on 9 features including the certificate signature algorithm in use, if the subject appears to have default names (e.g.,
ST=somestate), if it is self-signed, if the subject common name is similar to the host name, or a “normal” value of the validity period.
are you self-hosting a server? sorry, you're not a real person
The basic implementation of Perspectives suffers from completeness (only the initial SSL connection is subject to inspection), performance (notary lag, and lack of caching), and privacy issues (leaking browsing history),
not only did they not cache either, LMAOOOOO
which the follow-up Convergence [1] aims to address.
that is the name of a villainous plot
The EFF proposes to solve the problem of malicious CA certificates
malice has never once been mentioned until now but apparently CAs are malicious actors now hell yeah
by using semi-centralized timeline servers utilizing an cryptographic append-only data structure [13].
that's literally just a fucking certificate
This data structure would make MITM attack virtually impossible
you know what MITM attacks need? a data structure. that's how we can solve trust on the web
and clients could automatically use alternative paths to a site when they detect any tampering.
that's not how certificates work, that is the job of network routing. is the EFF going to solve DNS with cryptocurrency too?
the fact that djb linked this paper actively claiming to solve network routing against censorship makes it obvious that my anonymity network needs to exist
The authors evaluate several theoretical scenarios in which such man in the middle attacks might be carried out and propose to solve the problem by displaying a warning if the CA is situated in a different country than the entity for which the certificate was generated.
surely the problem is those conniving foreign CAs. we should have never given them any authority
Generally, we assume the vantage point of a site’s upstream border link where we passively monitor live SSL/TLS 2 traffic to extract a set of features for later offline analysis.
imagine someone telling you a system that enables "extracting a set of features for later offline analysis" constitutes transport layer security
transport layer surveillance
For the remainder of the paper, we will refer to either SSL or TLS as “SSL”.
you will never fucking guess how fucked up the version numbering is on this shit
In particular, we do not record any information that identifies a client system directly.
and you know EXACTLY what's coming next
We however log one-way hashes of the pairs (client, server) and (client, SNI), which allow us identify all sessions involving the same endpoints.
we don't record vertices, we just label the edges! (with the server certificate this constitutes a client identifier, QED)
We use Bro 2.0 [28, 4] to extract the features from live traffic.
TECH BRO?????
Bro detects and parses all SSL connections independent of any transport-layer ports by inspecting their payload. [8]
TECH BRO WITH CONSENT ISSUES????
We spent significant time to optimize our import scripts for handling large numbers of SSL connections.
imagine bragging about this???
At peak times, our data providers upload more than a million new connections per hour.
now it seems more directly like they're talking about some sort of large-scale cryptanalytic process
During bulk imports, we measured the script’s maximum rate at about 100K connections/minute for a single thread, running on a Intel Xeon E5630 CPU. As certificate validation accounts for
most the work, the processing parallelizes well across CPUs.
this is so weird!!!! who is this for?????
At three of the sites, members of our team did the set up after receiving administrative approval.
oh go to hell
Browsers ignore such additional certificates as they stop after the first valid end host. Indeed, we find ourselves unable to display that extra certificate with any of the standard browsers.
is this a problem? is there a solution? what was the intention? this paper has no answers only more questions
Replies
6Specifics of the validation logic tend to be hidden deeply in a browser’s code, with no easy way to split it out, or access from external.
I STILL REMEMBER WHEN MOZILLA INVENTED THE RUST PROGRAMMING LANGUAGE TO SOLVE THIS PROBLEM
Furthermore, browsers themselves can disagree in individual cases.
YEAH UNTIL GOOGLE PAYS EVERYONE OFF TO AGREE WITH THEM USING THE ILLEGAL AND ANTI-COMPETITIVE REVENUE SHARE PROGRAM
transport layer surveillance
Grid traffic accounts for a significant share of our data set. While Grid infrastructure uses SSL extensively, it deploys an independent root hierarchy and hence its certificates, according to our definition relative to the Mozilla set, do not validate. We instead identify Grid activity by separately looking for corresponding certificates.
so "validation" of certificates seems pretty clearly to be a way of "identifying activity". but what the hell is the "Grid"
The Grid infrastructure provides distributed computing environments for data-intensive applications, typically in a scientific setting [9].
terrible paper. why is this only introduced four pages later
the entire fucking problem with this analysis is that what it frames as an impartial and universal measurement of "certificate validation" actually means "is anyone employing public key cryptography in ways we don't know about"
it leads to "analyses" like this https://blog.yossarian.net/2023/05/21/PGP-signatures-on-PyPI-worse-than-useless from the guy i hate who sucks. this same class of argument was used to motivate removing pgp support (i.e. support for public key cryptography) from pypi
The tail of non-validated certificates comes from applications that use self-signed certificates with unusually high lifetimes.
IMAGINE! USING PUBLIC KEY CRYPTOGRAPHY TO REPRESENT YOUR OWN IDENTITY TO OTHERS ON THE INTERNET!
@hipsterelectron DeCentRalizEd TruST mOdeLS WilL nEVer WorK!!!!111 wE MUst deSTRoy tHeM wITH preJUdicE iN ANtICipaTorY oBEdienCE For thiS iMMOvabLe NatURAl LaW!@!!
🤡