Ffmpeg has an exceptionally terrible track record when it comes to security. People have been throwing fuzzers at it for as long as I remember and coming back with a nearly inexhaustible supply of memory corruption bugs. Here's an effort by one Googler a decade ago:
So, while it's a demo of the capabilities of LLMs, this should not be at all surprising. Ffmpeg is absolutely not something you should be running outside of a sandbox if you're touching any untrusted or user-supplied content. I know that people do, and these people are taking unreasonable risks.
> Ffmpeg is absolutely not something you should be running outside of a sandbox if you're touching any untrusted or user-supplied content.
I agree. I work for InstaVM which essentially gives you sandboxes - so I can share some perspective from the other side.
The trend is that people are building AI agents and these agents almost always have a chat box, so prompt injections are always a threat apart from the usual hallucinations or wrong code generated by the llm. Not everyone wants to give latest and great AI models to their users due to cost so end up with something like Gpt-4o and rm rf-ing the whole thing at times. At this point you have to use an isolated environment to guard against these.
From what I understand gstreamer is more about building complex pipelines and plugins, ffmpeg is better at playing some obscure 20 year old video format extremely efficiently so you can watch it compiled for a potato.
That's not really true. Ffmpeg is a Swiss army knife for anything related to digital multimedia (old and new). It is broken into a few libraries but doesn't really have plugins.
Gstreamer has a different model, chaining together plugins. Lots of overlap, but I think Gstreamer only has real traction because some silicon vendors use it.
Apr Fools Day really is the shittiest day to be online. For one thing, practical jokes/pranks are just gussied-up asshole behavior. For another thing, nerds generally SUCK at information-delivery pranks, which is what the Internet is full of on Apr 1.
If there was a nearly inexhaustible supply of Indian security researchers emailing you a nearly inexhaustible supply of LLM slop daily, there is a point where you or I would stop caring too.
ffmpeg is Free Software. You are also free not to use it.
Oddly enough, despite all these endless grievances, no one has come up with a better or more capable tool, certainly not one that is freely available.
Evidently no one cares either, because most implementations of ffmpeg I've seen typically run it as root "because we have to". Don't worry we use Docker bro.
> AI slop is a real problem and annoying. Just because it exists does not mean every vulnerability report is AI slop.
Ok but who is going to sift through it all to triage the good bits when you're working on something for free?
> Ffmpeg devs are free not to care, but then they cant complain when they start to get a bad reputation
Who gives a shit about reputation when you're the only game in town?
There is nothing out there that even attempts to approximate an ffmpeg clone. They are the Swiss army knife of media encoding and all complainers have produced are plastic sporks.
> Ok but who is going to sift through it all to triage the good bits when you're working on something for free?
Its like anything else in open source. Maintainers will do so if they care. Maybe they decide they don't care. That is always their decision to make but there are consequences for the project. Maybe those consequences make sense. Being a maintainer is all about making cost-benefit trade offs.
> Who gives a shit about reputation when you're the only game in town?
Its up to the maintainers whether they care or not. It depends on what they value.
Ultimately if maintainers make decisions that are at odds with what their userbase want, someone eventually forks and people vote with their feet.
Today it's an industry driven by unscrupulous clout-chasers and a commitment to quantity over quality.
There is a difference between going through patches and pull requests vs. the endless stream of LLM-assisted bullshit that has started cluttering security inboxes in the last few years.
>The reach of this bug is what makes it serious. Any deployment that points FFmpeg at an attacker-influenced RTSP URL is exposed: media ingest pipelines fetching user-supplied stream URLs, surveillance and CCTV systems pulling RTSP feeds, and transcoding services processing remote AV1-over-RTP sources
Wow this is actually pretty serious - I'm even surprised its being published. There are several services where I can imagine this is exploitable today.
Some people might suggest it’s crucial to publish if you’re aware of a serious vulnerability, so that people using the software in a vulnerable way can take steps to mitigate the risk.
Speed and security are not good bedfellows. Combine that with really shitty standards and dozens of years of development...
Oh, and licensing. Licensing is the real killer. I could just write my own mp3 decoder easily (the format not the file type) but I'm not gonna risk my company getting sued into the ground by doing that.
> At this point the corrupted free pointer is called, and control of the instruction pointer is ours.
Very serious, though in practice it doesn't sound like this bug achieves arbitrary RCE on its own (especially in the presence of ASLR). You would need there to be some writable and executable page of memory lying around.
The article glosses over this, but it looks like the next variable in the struct is conveniently the first parameter to the function, so you can run arbitrary code with system() or whatever. But, yeah, you would need some other exploit to defeat ASLR.
Even if this isn't as big a deal as this [advertisement for a security company] seems, it is a reminder that every application you release does have a security hole somewhere, and a script kiddie can now find it 5 minutes after release for $2 in credit. If you're not red-teaming your code before release, hackers are doing it after.
I understand why it's poorly understood. It's a snappy term, and people assume it means "bad" and nothing else because that's all you can get from the context. However, since most people also don't know the difference between a vulnerability and an exploit, they won't understand the definition of a zero-day when they read it.
But I'm still going to complain if a security vulnerability research company is using the term incorrectly in their own press copy. It makes them look amateurish.
What about VLC's own built-in versions of decoding libraries (I think, from the FFmpeg project)? Is there a scenario here where we may have to deal with malicious MP4 files?
Is the future of defense-against-foreign-agents-on-my-codebase to subtly hide prompt injections into one’s codebase that would defeat agents to find security bugs ?
If the attackers of ffmpeg need to be using such those authors’ services to find RCE in popular tools to attack, what the ffmpeg team needs to defeat attackers is to reduce efficiency of such tools depthfirst
I've been using ffmpeg for a very long time, both personally and for services I've built. Fabrice Bellard is a genius, and the developers who have taken it so far have made the world measurably richer.
But I can't think of a program more worthy of sandboxing when run with untrusted input than ffmpeg. It's a huge amount of C dealing with the most complicated video and audio codecs, which is notoriously impossible to get completely right.
But it's not actually that big of a problem. I run ffmpeg inside a VM or gVisor, and the end result is usually a video file that I'm perfectly willing to play in my browser, where it gets decoded in yet another sandbox because this shit is hard.
What do you mean "video file that I'm perfectly willing to play in my browser". Isn't it safe to assume that no video file can escape the browser decoding sandbox?
I'm not up-to-speed with the current state of sandboxing in browsers, but in principle it's (on modern operating systems) not especially hard for them to sandbox the decoding into a separate process with basically no privileges beyond rendering a video stream. It's a bit trickier if we're only considering demuxing and delegating decoding to the hardware, but that's a much smaller attack surface.
A manually run ffmpeg on the command line does nothing to restrict its privileges, and its security model has very little interest in doing so, while browsers very much have.
Inflated use of the term zero-day, while none of the described vulnerabilities is actually a zero-day. But it sounds and clicks good..
thank you for the PoC.
Most of them involve very weird and unlikely scenarios and bad security practices or access to the ffmpeg binaries and being allowed to run arbitrary commands at an elevated permission.
In and of itself there's not a massive issue from what I can see, they're entry vectors that can lead to worse situations.
That's not to say they're not serious but if a Russian hacking group is using one of them it's in conjunction with other exploits or security flaws. Which is common in practice when it comes to decoding.
"No way to prevent this" say users of only language where this regularly happens, etc, etc. Several of these bugs do not appear to be in hot code and would have been detected by a language with saner behaviour.
Help me understand: depthfirst seems to be bigging up their “security agent” here, but is it not just prompt engineering + writing skill files? What goes into producing a “security agent” beyond this? Feels like they’re really gussying up a process that is ultimately just LLM usage
https://security.googleblog.com/2014/01/ffmpeg-and-thousand-...
So, while it's a demo of the capabilities of LLMs, this should not be at all surprising. Ffmpeg is absolutely not something you should be running outside of a sandbox if you're touching any untrusted or user-supplied content. I know that people do, and these people are taking unreasonable risks.
I agree. I work for InstaVM which essentially gives you sandboxes - so I can share some perspective from the other side.
The trend is that people are building AI agents and these agents almost always have a chat box, so prompt injections are always a threat apart from the usual hallucinations or wrong code generated by the llm. Not everyone wants to give latest and great AI models to their users due to cost so end up with something like Gpt-4o and rm rf-ing the whole thing at times. At this point you have to use an isolated environment to guard against these.
Different cases really I think both are good.
Gstreamer has a different model, chaining together plugins. Lots of overlap, but I think Gstreamer only has real traction because some silicon vendors use it.
Security is the punch line for ffmpeg.
https://nitter.net/ffmpeg/status/2039115531744334180
goddamn, and this is a project that prides itself on having had-written assembly in it
Do you have an example?
ffmpeg is Free Software. You are also free not to use it.
Oddly enough, despite all these endless grievances, no one has come up with a better or more capable tool, certainly not one that is freely available.
Evidently no one cares either, because most implementations of ffmpeg I've seen typically run it as root "because we have to". Don't worry we use Docker bro.
Actual well written vulnerability reports are not the same as slop.
AI slop is a real problem and annoying. Just because it exists does not mean every vulnerability report is AI slop.
Ffmpeg devs are free not to care, but then they cant complain when they start to get a bad reputation.
Ok but who is going to sift through it all to triage the good bits when you're working on something for free?
> Ffmpeg devs are free not to care, but then they cant complain when they start to get a bad reputation
Who gives a shit about reputation when you're the only game in town?
There is nothing out there that even attempts to approximate an ffmpeg clone. They are the Swiss army knife of media encoding and all complainers have produced are plastic sporks.
Its like anything else in open source. Maintainers will do so if they care. Maybe they decide they don't care. That is always their decision to make but there are consequences for the project. Maybe those consequences make sense. Being a maintainer is all about making cost-benefit trade offs.
> Who gives a shit about reputation when you're the only game in town?
Its up to the maintainers whether they care or not. It depends on what they value.
Ultimately if maintainers make decisions that are at odds with what their userbase want, someone eventually forks and people vote with their feet.
Today it's an industry driven by unscrupulous clout-chasers and a commitment to quantity over quality.
There is a difference between going through patches and pull requests vs. the endless stream of LLM-assisted bullshit that has started cluttering security inboxes in the last few years.
Yes, there are security issues but quite a few are not ffmpeg itself related - the input is pretty shabby or at least not exactly easy to deal with!
Obviously, they could do with some assistance and I'm sure you and I will both dive in with equal zeal.
Wow this is actually pretty serious - I'm even surprised its being published. There are several services where I can imagine this is exploitable today.
(There are a number of reasons for this, not least being that C makes it very easy to ship partially initialized memory over the wire.)
Oh, and licensing. Licensing is the real killer. I could just write my own mp3 decoder easily (the format not the file type) but I'm not gonna risk my company getting sued into the ground by doing that.
Very serious, though in practice it doesn't sound like this bug achieves arbitrary RCE on its own (especially in the presence of ASLR). You would need there to be some writable and executable page of memory lying around.
I understand why it's poorly understood. It's a snappy term, and people assume it means "bad" and nothing else because that's all you can get from the context. However, since most people also don't know the difference between a vulnerability and an exploit, they won't understand the definition of a zero-day when they read it.
But I'm still going to complain if a security vulnerability research company is using the term incorrectly in their own press copy. It makes them look amateurish.
is it the difference between a knife and a stab wound?
If the attackers of ffmpeg need to be using such those authors’ services to find RCE in popular tools to attack, what the ffmpeg team needs to defeat attackers is to reduce efficiency of such tools depthfirst
But I can't think of a program more worthy of sandboxing when run with untrusted input than ffmpeg. It's a huge amount of C dealing with the most complicated video and audio codecs, which is notoriously impossible to get completely right.
But it's not actually that big of a problem. I run ffmpeg inside a VM or gVisor, and the end result is usually a video file that I'm perfectly willing to play in my browser, where it gets decoded in yet another sandbox because this shit is hard.
It's 'safe to assume' it's not. It's emphatically not safe to assume any mitigation is perfect.
Why would that be safe to assume? If that were a reasonable assumption, you could just as well assume that it's safe to run ffmpeg.
A manually run ffmpeg on the command line does nothing to restrict its privileges, and its security model has very little interest in doing so, while browsers very much have.
And get hardware acceleration working...
LLM constantly confidently giving me this same sounding script with a "the root cause" and how it "is simple" while being completely incorrect.
In and of itself there's not a massive issue from what I can see, they're entry vectors that can lead to worse situations.
That's not to say they're not serious but if a Russian hacking group is using one of them it's in conjunction with other exploits or security flaws. Which is common in practice when it comes to decoding.
What about "ls"?
Another option is WASM or WASM-style sandboxes if using another process is undesirable.
But are the compiler+OS that runs the ffmpeg executable really a sandbox ?