curl https://some-url/ | sh
I see this all over the place nowadays, even in communities that, I would think, should be security conscious. How is that safe? What’s stopping the downloaded script from wiping my home directory? If you use this, how can you feel comfortable?
I understand that we have the same problems with the installed application, even if it was downloaded and installed manually. But I feel the bar for making a mistake in a shell script is much lower than in whatever language the main application is written. Don’t we have something better than “sh” for this? Something with less power to do harm?
Damn that’s bad misinformation. Its a security nightmare
So tell me: if I download and run a bash script over https, or a .deb file over https and then install it, why is the former a “security nightmare” and the latter not?
Both are a security nightmare, if you’re not verifying the signature.
You should verify the signature of all things you download before running it. Be it a bash script or a .deb file or a .AppImage or to-be-compiled sourcecode.
Best thing is to just use your Repo’s package manager. Apt will not run anything that isn’t properly signed by a package team members release PGP key.
I have to assume that we’re in this situation because because the app does not exist in our distro’s repo (or homebrew or whatever else). So how do you go about this verification? You need a trusted public key, right? You wouldn’t happen to be downloading that from the same website that you’re worried might be sending you compromised scripts or binaries? You wouldn’t happen to be downloading the key from a public keyserver and assuming it belongs to the person whose name is on it?
This is such a ridiculously high bar to avert a “security nightmare”. Regular users will be better off ignoring such esoteric suggestions and just looking for lots of stars on GitHub.
No, you download the key from many distinct domains and verify it matches before TOFU
Ah yes, so straightforward.
Fortunately package managers already do this for you. Open a bug report to add to apt. Easy.
For example: A compromised host could detect whether you are downloading the script or piping it.
I’m confident that if the host is compromised I’m screwed regardless.
No it isn’t. What could a Bash script do that the executable it downloads couldn’t do?
It’s not just protection against security, but also human error.
https://github.com/MrMEEE/bumblebee-Old-and-abbandoned/issues/123
https://hackaday.com/2024/01/20/how-a-steam-bug-once-deleted-all-of-someones-user-data/
Just because I trust someone to write a program in a modern language they are familier in, doesn’t mean I trust them to write an install script in bash, especially given how many footguns bash has.
Hilarious, but not a security issue. Just shitty Bash coding.
And I agree it’s easier to make these mistakes in Bash, but I don’t think anyone here is really making the argument that curl | bash is bad because Bash is a shitty error-prone language (it is).
Definitely the most valid point I’ve read in this thread though. I wish we had a viable alternative. Maybe the Linux community could work on that instead of moaning about it.
It absolutely is a security issue. I had a little brain fart, but what I meant to say was “Security isn’t just protection from malice, but also protection from mistakes”.
Let’s put it differently:
This is a common sentiment people say about C, and I have a the same opinion about it. I would rather we use systems in place that don’t give people the opportunity to make mistakes.
Viable alternative for what? Packaging.
I personally quite like the systems we have. The “install anything from the internet” is exactly how Windows ends up with so much malware. The best way to package software for users is via a package manager, that not only puts more eyes on the software, but many package managers also have built in functionality that makes the process more reliable and secure. For example signatures create a chain of trust. I really like Nix as a distro-agnostic package manager, because due to the unique way they do things, it’s impossible for one package’s build process to interfere with another.
If you want to do “install anything from the internet” it’s best to do it with containers and sandboxing. Docker/podman for services, and Flatpak for desktop apps, where it’s pretty easy to publish to flathub. Both also seem to be pretty easy, and pretty popular — I commonly find niche things I look at ship a docker image.
The issue with C is it lets you make mistakes that commonly lead to security vulnerabilities - allowing a malicious third party to do bad stuff.
The Bash examples you linked are not security vulnerabilities. They don’t let malicious third parties do anything. They done have CVEs, they’re just straight up data loss bugs. Bad ones, sure. (And I fully support not using Bash where feasible.)
A viable way to install something that works on all Linux distros (and Mac!), and doesn’t require root.
The reason people use curl | bash is precisely so they don’t have to faff around making a gazillion packages. That’s not a good answer.
Nix portable installations, Soar.
Developers shouldn’t be making packages. They do things like vendor and pin dependencies, which lead to security and stability issues later down the line. See my other comment where I do a quick look at some of these issues.
Soar does look cool, but it requires you to install it first so that adds an extra installation step. Extra installation friction is very bad. And guess how they tell you to install Soar? 😄
Ok so how do these packages magically get made? Am I just supposed to not distribute my software? And don’t say “distros will do it” because that’s clearly a shit non-solution. No distro has made a package for any of the software I’ve written.
I hate this, but I’ve mentioned before in the other threads related to this that I make an exception for package managers due to their ability to install packages themselves.
Systems that protect people mean bureaucracy. And bureaucracy means slowness. It means many niche libraries or programs won’t get packaged. It means that it won’t get updated to the latest version immediately either, even if they receive security updates.
But as a consequence of these systems, Debian 12 remained entirely untouched by the XZ backdoor, when almost every other distribution was hit. That’s a pretty big deal.
As a consequence of a lack of these systems, many Windows programs are still floating around with vulnerable versions of curl, having included the software into their “package” but never bothering to update it.
I care more about the security of the users than the feelings of the developers. It’s that simple. Developers are a tiny fraction of total computer users. The needs of the many outweigh the wants of the few.
You’re telling me that you dont verify the signatures of the binaries you download before running them too?!? God help you.
I download my binaries with apt, which will refuse to install the binary if the signature doesn’t match.
No because there’s very little point. Checking signatures only makes sense if the signatures are distributed in a more secure channel than the actual software. Basically the only time that happens is when software is distributed via untrusted mirror services.
Most software I install via curl | bash is first-party hosted and signatures don’t add any security.
All publishing infrastructure shouldn’t be trusted. Theres countless historical examples of this.
Use crypto. It works.
Crypto is used. It is called TLS.
You have to have some trust of publishing infrastructure, otherwise how do you know your signatures are correct?
TLS is a joke because of X.509.
We dont need to trust any publishing infrastructure because the PGP private keys don’t live on the publishing infrastructure. We solved this issue in the 90s
If you think PGP solved anything at all you’re living in a fantasy land lol
By definition nothing
The point you appear to be making is “everything is insecure so nothing is” and the point others are making is “everything is insecure so everything is”
No, the point I am making is there are no additional security implications from executing a Bash script that someone sends you over executing a binary that they send you. I don’t know how to make that clearer.