rnhmjoj/qutebrowser
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Add suckless surf ML post to notes.

Browse Source
This commit is contained in:
Florian Bruhin 2015-01-08 06:41:44 +01:00
parent b12b83f98c
commit 1086c33ec2
1 changed files with 114 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

114
doc/notes
View File

@ -80,3 +80,117 @@ some cases passing the url to some cli program).
I've also noticed the lack of completion. For example, on "o" pentadactyl will
show sites (e.g. from history) that can be completed. I think I've been spoiled
by pentadactyl having completion for just about everything.
suckless surf ML post
=====================
From: Ben Woolley <tautolog_AT_gmail.com>
Date: Wed, 7 Jan 2015 18:29:25 -0800
Hi all,
This patch is a bit of a beast for surf. It is intended to be applied after
the disk cache patch. It breaks some internal interfaces, so it could
conflict with other patches.
I have been wanting a browser to implement a complete same-origin policy,
and have been investigating how to do this in various browsers for many
months. When I saw how surf opened new windows in a separate process, and
was so simple, I knew I could do it quickly. Over the last two weeks, I
have been developing this implementation on surf.
The basic idea is to prevent browser-based tracking as you browse from site
to site, or origin to origin. By "origin" domain, I mean the "first-party"
domain, the domain normally in the location bar (of the typical browser
interface). Each origin domain effectively gets its own browser profile,
and a browser process only ever deals with one origin domain at a time.
This isolates origins vertically, preventing cookies, disk cache, memory
cache, and window.name vulnerabilities. Basically, all known
vulnerabilities that google and Mozilla cite as counter-examples when they
explain why they haven't disabled third-party cookies yet.
When you are on msnbc.com, the tracking pixels will be stored in a cookie
file for msnbc.com. When you go to cnn.com, the tracking pixels will be
stored in a cookie file for cnn.com. You will not be tracked between them.
However, third-party cookies, and the caching of third party resources will
still work, but they will be isolated between origin domains. Instead of
blocking cookies and cache entries, they are "double-keyed", or *also*
keyed by origin.
There is a unidirectional communication channel, however, from one origin
to the next, through navigation from one origin to the next. That is, the
query string is passed from one origin to the next, and may embed
identifiers. One example is an affiliate link that identifies where the
lead came from. I have implemented what I call "horizontal isolation", in
the form of an "Origin Crossing Gate".
Whenever you follow a link to a new domain, or even are just redirected to
a new domain, a new window/tab is opened, and passed the referring origin
via -R. The page passed to -O, for example -O originprompt.html, is an HTML
page that is loaded in the new origin's context. That page tells you the
origin you were on, the new origin, and the full link, and you can decide
to go just to the new origin, or go to the full URL, after reviewing it for
tracking data.
Also, you may click links that store your trust of that relationship with
various expiration times, the same way you would trust geolocation requests
for a particular origin for a period of time. The database used is actually
the new origin's cookie file. Since the origin prompt is loaded in the new
origin's context, I can set a cookie on behalf of the new origin. The
expiration time of the trust is the expiration time of the cookie. The
cookie implementation in webkit automatically expires the trust as part of
how cookies work. Each time you cross an origin, the origin crossing page
checks the cookie to see if trust is still established. If so, it will use
window.location.replace() to continue on automatically. The initial page
renders blank until the trust is invalidated, in which case the content of
the gate is made visible.
However, the new origin is technically able to mess with those cookies, so
a website could set trust for an origin crossing. I have addressed that by
hashing the key with a salt, and setting the real expiration time as the
value, along with an HMAC to verify the contents of the value. If the
cookie is messed with in any way, the trust will be disabled, and the
prompt will appear again. So it has a fail-safe function.
I know it seems a bit convoluted, but it just started out as a nice little
rabbit hole, and I just wanted to get something workable. At first I
thought using the cookie expiration time was convenient, but then when I
realized that I needed to protect the cookie, things got a bit hairy. But
it works.
Each profile is, by default, stored in ~/.surf/origins/$origin/
The interesting side effect is that if there is a problem where a website
relies on the cross-site cookie vulnerability to make a connection, you can
simply make a symbolic link from one origin folder to another, and they
will share the same profile. And if you want to delete cookies and/or cache
for a particular origin, you just rm -rf the origin's profile folder, and
don't have to interfere with your other sites that are working just fine.
One thing I don't handle are cross-origins POSTs. They just end up as GET
requests right now. I intend to do something about that, but I haven't
figured that out yet.
I have only been using this functionality for a few days myself, so I have
absolutely no feedback yet. I wanted to provide the first implementation of
the management of identity as a system resource the same way that things
like geolocation, camera, and microphone resources are managed in browsers
and mobile apps.
Currently, Mozilla and Tor have are working on third-party tracking issues
in Firefox.
https://blog.mozilla.org/privacy/2014/11/10/introducing-polaris-privacy-initiative-to-accelerate-user-focused-privacy-online/
Up to this point, Tor has provided a patch that double-keys cookies with
the origin domain, but no other progress is visible. I have seen no
discussion of how horizontal isolation is supposed to happen, and I wanted
to show people that it can be done, and this is one way it can be done, and
to compel the other browser makers to catch up, and hopefully the community
can work toward a standard *without* the tracking loopholes, by showing
people what a *complete* solution looks like.
Thank you,
Ben Woolley
Patch: http://lists.suckless.org/dev/att-25070/0005-same-origin-policy.patch