Hacker News Reader

Comment by princekolt | Hacker News Reader

There has been a number of cases of installers from trusted developers being infected lately. (For example Transmission being infected twice...)

On our side (developers) we need to be careful with this idea that "we will know" when something is wrong and be more careful when deploying software. It would also be nice if some form of tool could be used to test a binary to make sure it only contains what it should contain (sort of a whitelist of symbol names compared to the source files, idk...) I'm sure something along these lines probably exists for some different purpose.

View on HN

First of all, this is why reproducible builds (getting bit-for-bit identical binaries independent of the machine used to build the software) is something we should be putting much more work into. The Debian folks are doing an amazing job there.

Another important point is that distributions have already solved effectively all of these problems. We have automated building and signing systems that mean that installation and upgrades are done in ways that are not vulnerable even to fairly sophisticated attacks. You can build packages locally if you want to verify them, and modifying a package after it has been built invalidates the signature that all modern package managers require before installing a package. As part of the openSUSE project we even have a free-to-use (and free as in freedom) build project called the Open Build Service[1] which allows you to build packages (with automated dependency update rebuilds) for many different distributions (Arch Linux, Debian, Ubuntu, Fedora, RHEL and obviously openSUSE and SLES).

I get that distributions aren't "sexy" but it's getting quite frustrating seeing all these communities make the same mistakes that distributions made (and learned from) more than 20 years ago.

[Disclaimer: I work for SUSE and am an openSUSE community member.]

[1]: https://build.opensuse.org/

That's ok for the NSA and a few large enterprises to check certain binaries but you can't ask any start up or small company to recompile themselves every version of every OS / Software / Library that they use. It's not really a practical solution (even assuming everything is open source).

As I posted above, we have a build service that does this automatically for you. It takes maybe 30 minutes to create a few spec files, and now it's supported by a bunch of different distributions. You can then point your users at the repo (or mirror the repo if you prefer). Also OBS lets you create forks of projects that just inherit the unmodified packages (and said forks can also be cross-host so you don't need to self-host a whole distribution).

I agree that doing everything I mentioned manually is hard, that's again why I said that distributions have solved this problem and made it easy.

But who would use the build service? The consumer of the software or the publisher?

[edit] also that protects you against malware injecting binaries in an executable when compiling it, but not from malware injecting code into the source code of the executable.

> But who would use the build service? The consumer of the software or the publisher?

Publisher. A user could (if they were really paranoid) rebuild the packages locally, with two or three commands.

> Also that protects you against malware injecting binaries in an executable when compiling it, but not from malware injecting code into the source code of the executable.

You can download the source code that OBS used (both as a src RPM generated by the builder and the OBS repo that the builder was given read-only access to), and OBS supports cryptographic signatures of the originating source (with gpg-offline keys to avoid WoT attacks). If your developers are using sane source control practices (use GPG keys for every commit, but especially tags) then you are protected against that too.

Of course, reproducible builds is something that would solve this problem even better (protecting against attacks on OBS that cause it to add source that are not in the repo). As a side point, our threat model doesn't fully trust the nodes compiling the software so such attacks are fairly limited in scope (but I'm not a developer of OBS so I'm really not the right person to be asked these questions).

compile times are fast enough now that if the tools were friendly enough I don't know it would be so inconvenient to the end user anymore.

is there any distros out there now that are source based like Gentoo with more friendly defaults?

There's NixOS. By default it uses a binary cache, but it'd be easy to disable that if you want to.

And reproducible builds are a work in progress.

And why exactly can't the source be compromised?

I don't know what they're planning to do, but we could need only a few parties doing the verifications. If a build yields the wrong binary then the release is flagged and nobody gets it. Apt should check against the expected results.

Many things could go wrong with this (mainly attacks on the expected results db, it should be replicated) but the idea should work.

The thing is if the threat is any machine on the publisher side being infected, then it sorts of need to be verified by the consumer of the software, a bit like compiling a checksum when you download software.

Debian packages are signed. It is hard to make software distribution completely secure. But, one could imagine creating pretty strong guarantees. Imagine that you have N completely different parties/machines (isolated, on different infrastructure, different user accounts).

- (N-1) machines build a new Debian package.

- The (N-1) results are fetched by the fourth party.

- The Nth party checks whether the build results are identical.

- The Nth party signs the result with GPG on a machine that is not connected.

- The signed package is distributed.

An attacker would need to compromise either: (1) N-1 build machines; (2) the offline machine used for signing packages; or (3) the upstream source.

I consider (3) to be the most serious thread. But in this scenario, only the distribution packagers need to inspect the source changes. In the common scenario where you download from a vendor (e.g. the Transmission or CCleaner website), every user has to inspect a binary blob.

Another possibility would be to have strong sandboxing for applications. An application could still participate in DDoS attacks, etc. But it would at least not encrypt/destroy your data.

tl;dr: reproducible builds are an extremely important development and Debian (and other distributions participating in this initiative) should be commended for their work!

Right, and this is why we need reproducible builds.

> There has been a number of cases of installers from trusted developers being infected lately.

Here a few other cases

* https://www.pcworld.com/article/3104180/software/pc-nuking-m...

* https://www.reddit.com/r/Windows10/comments/6hvzzt/psa_my_ch...

* https://lifehacker.com/many-browser-extensions-have-become-a...

* http://www.zdnet.com/article/android-malware-in-google-play-...

* https://arstechnica.com/security/2015/04/google-kills-200-ad...

* https://arstechnica.com/security/2017/01/ciscos-webex-chrome...

Here another high profile case

* https://nakedsecurity.sophos.com/2016/02/22/worlds-biggest-l...

I'm not sure how exactly these infections work, but one method would be to infect the developers' PCs. In which case you essentially can't trust anything. You'd need some kind of byzantine fault tolerance (mandatory multi-person code review?) to be sure nothing like this ever happens

What makes this scary is that, as far as I know, pretty much no software has that kind of security, and there are several pieces of widely used software that always update automatically (sometimes for good reason, sometimes not so much).

You can enable AppLocker and have explicit control on what executes and what not by creating rules. I know quite a few companies that enforce its use in their employees' PCs.

As an aside, AppLocker was trivially bypassable for several years -- there were two different APIs that allowed you to set an "ignore AppLocker" flag. We used to use it in high-school to play games (or in my case, run gvim and some other development tools).

I think that there needs to be a more complete solution than just "secure the developers machines". You need to have peer-review, where the developers sign commits to approve them.

The solution would be code review + reproducable builds. This is how Debian tries to prevent these kinds of attacks.

> mandatory multi-person code review?

Isn't this pretty much mandatory with SOX compliance anyway?

My first thought would be a little process running on a server not on your network that pulls the download every so often, say once a day, and texts you if it is different. You'd expect a text after you push your new version, but if you get one out of the blue, then you'd know something was up.

Binary analysis for each software you install might be too cumbersome for most developers.

I suggest you use something like "Little Snitch" for mac which warns you when software makes inside/outside connections.

It might not be the best, but it's definitely something that works to mitigate some hacks.

Thing is, you can't even trust little snitch these days[1] :(

At least a binary check after compilation+signing (by the developer) should improve security a little bit.

[1]: https://objective-see.com/blog/blog_0x21.html

I like to compliment Little Snitch with XFence (formerly known as Little Flocker).

You can think of it as a firewall for your filesystem and devices.

I use Little Snitch too. I had a Little Flocker license, but the rules were quite painful to maintain. Especially if you are also using the Terminal and command-line apps.

That creates an interesting conflict between automatic updates (you want to be patched against the latest security vulnerabilities) and manual, more infrequent updates (you are less exposed to a compromised update directly and automatically infecting your machine).

>Transmission being infected twice

Could you please elaborate? I only recall one instance of compromised Transmission installer.

https://www.reddit.com/r/torrents/comments/50dm8e/transmissi...

Reply on news.ycombinator.com