TS-2026-009: Insecure argument handling in Tailscale SSH permitted root access
tailscale.com
tailscale.com
(If you had SSH access to a host in your Tailscale ACL, you could log in as `-i` and get a root login.)
There's Paramiko, but Python is still a huge liability in memory-constrained systems.
But we don't trust Tailscale. Just look at the thousands of unresolved Github issues, many of which are actually quite important/useful but have been ignored for months and years.
We very much fear at $work that there are vulnerabilities in the Tailscale product awaiting discovery. Especially as, AFAIK, Tailscale have never had a formal security audit on their software.
So we install it on hardened bastion hosts in an old-school "jump host" model. So people can still get access to where they need to be, but we don't need to install Tailscale's unaudited shit on every single server / vm / etc.
And we only use Tailscale as a glorified VPN, we don't use their hundreds of extra random features like this SSH one.
In terms of SSH, we use old-school OpenSSH and SSH certificates. Its really not that difficult and its really not expensive, you can do offline signing with Yubikeys, no need for expensive HSMs.
And these published audits of the `tailscale` software are where ?
Even half-serious VPN providers like Mullvad publish in public their regular security audits of their app and infrastructure.
There is zero reason Tailscale cannot do the same.
And frankly, given the nature of this vulnerability, "insecure argument handling" I'm not entirely sure it has been audited ? Or if it has, they should be looking for a new auditor ASAP !
I agree. Its annoying that lots of places have recently been blanket-banning Mullvad IP ranges.
Their sister-company Tilitis[1] is also doing interesting things with the Tkey product.
The present version has limitations due to the original security model but the up-and-coming version has a revised security model to make things a bit more "real-world" useful whilst not making any security trade-offs.
Spoiler alert: This applies to all software. This is why security preaches defense in depth.
Thank you Mr/Mrs Pedant.
However this is a VPN product I am talking about which has an integral part in the defence in depth of which you speak.
Therefore it is only right and proper that I, and others, should be able to trust them to a greater degree. Never 100% of course, as you say. But if you are selling me a security product then you should be able to demonstrate you have put some damn effort into securing it.
"insecure argument handling" in a security product is not a good look. It stinks of sloppy coding practices followed up by a lack of security auditing.
Fair question. The short answer is my wording was a little off.
We do have pure SSH bastion hosts. Those are OpenBSD-based with a few tweaks, but basically they are locked-down and heavily monitored to within an inch of their life.
The reason Tailscale is there is mostly so the non-techies can access intranet portals and such-like.
But Tailscale does have a useful party-trick for the techies too, in that you can do DNS-suffix based routing (without needing to use Tailscale's DNS).
So they can automagically hop via a Tailscale bastion host to "*.$region.rds.amazonaws.com" or whatever. You just specify the DNS suffix and anything on that wildcard will work. It therefore enables us to further harden access to that stuff to a source IP of the Tailscale bastion hosts (in addition to the usual security stuff, of course).
Tailscale takes care of the automagic load-balancing/re-routing to the Tailscale bastion hosts, so we just have a bunch installed in random geographically-dispersed places and as long as >0 are up and running the Techies will always have a route to where they need to be.
Obviously the above is over-simplified summary before the pedants start picking it apart. ;)
Another concern I have is whether a compromise of Tailscale's own infra could let an attacker just add itself to my network. Apparently the "Tailnet Lock" feature mitigates this, but it is off by default. If I was an APT, compromising Tailscale would be priority number 1!
Yeah, we use "tailnet lock" to sort of cover that. AFAIK its the only option available.
I say "sort of" because "tailnet lock" is a bit half-assed in its design and implementation.
For example, you cannot sign new nodes from mobile devices (e.g. iOS). And as we know, locking down iOS is easier than a full desktop machine. So that's a bit of a missed opportunity. And as usual, there's a Tailscale Github issue open for it for years ...
And in practice, the "tailnet lock" addition is done through the `tailscale` software on the desktop/laptop. Its click a button or run a CLI without further authentication/authorisation required. So basically anyone could do it. And the keys are, of course, open to exfiltration, you can't use a Yubikey or anything like that. And you can't require multi-signer. Complete joke really.
Also if you use "tailnet lock" with the Tailscale Mullvad integration, its a one-way street. You can sign Mullvad nodes so they can be used as exit nodes, but it is impossible to revoke signatures from obsolete/replaced Mullvad nodes (with tailscale clients, you simply delete them from the org which means they have to re-auth and re-join under a new ID ... but you can't do that with the Mullvad integration, so the 'ghost' exit node could easily return). Your only option is to remove the signing node that signed the Mullvad nodes and start again from scratch. Another joke.
So, yeah, "better than nothing" is the way I would summarise "tailscale lock".
You need to generate a QR code then scan it from the signing mobile device, which opens a secret menu option to sign (fine just brings up a confirmation dialog).
Incredibly annoying but perhaps more secure vs the threat of randomly tapping at prompts
Interesting, thanks for pointing out the insane workflow.
Although looking around, footguns still remain in terms of relying too much on iOS and Tailnet Lock, e.g. https://github.com/tailscale/tailscale/issues/20475
Crowdstrike would be at a much higher priority. They already have a cloud connected root console to all the machines in entire corporations. Compromising your network is small potatoes compared to that.
Am aware of them but IIRC they are both unaudited which kind of brings us back to square one ? We would still end up running them at arms-length as we do with Tailscale at the moment.
Isn't Headscale server-side only ?
Also a bit strange that "Tailscale vs Netbird" doesn't feature more prominently on their "Compare Netbird" page. It is hidden behind "Load More". ;D
> Isn't Headscale server-side only ?
Yes. You're still running the native Tailscale client code on the hosts, which this evidence reveals can't be as trusted as Tailscale would like us to believe.
I also wouldn't trust Headscale fully. It had a critical defect at some point that, IIRC, would allow an attacker to rotate the key of a registered node without auth to a value chosen by the attacker. And its primary maintainer is a member of the Tailscale team, apparently maintained with full approval of their employer and with reasonable transparency between the projects, but nonetheless the overlap is a little close for comfort.
Frankly, as you've said, I don't trust any of these solutions to be anything more than a convenient way to jump onto a bastion or another host of minimal consequence to get into the network and jump onwards.
This is a poor measure of quality. I've spent considerable time knee-deep in these issues in particular and the vast, vast majority of them are feature requests, bug reports awaiting more information from the submitter, or bug reports that cannot easily be reproduced (likely conflicts with other software). On the whole, Tailscale does an excellent job of "just working" in practically every possible environment. They do an incredible job thriving in the diverse ecosystem of software networking and their work will naturally never be remotely done. There will always be gaps, it's the nature of the beast. I'm not aware of any other product that does a better job here.
If you've spent any time dealing with enterprise software you'd know that there is an infinite firehose of these sort of issues. We're lucky that Tailscale keeps these public. Many other vendors track these sort of issues privately.
If there are particular issues which jeopardize the security posture of Tailscale deployments that have been open for a significant amount of time, my clients and I would love to know. Please share!
> We very much fear at $work that there are vulnerabilities in the Tailscale product awaiting discovery. Especially as, AFAIK, Tailscale have never had a formal security audit on their software.
I can't take this seriously. If you were a customer you could, you know, ask them? Or inspect their SOC2 documents?
I absolutely guarantee they undergo regular formal security audits. There's no question.
> So we install it on hardened bastion hosts in an old-school "jump host" model. So people can still get access to where they need to be, but we don't need to install Tailscale's unaudited shit on every single server / vm / etc.
Bastion hosts are a terrible model in 2026. I can't take this approach seriously.
> we don't need to install Tailscale's unaudited shit on every single server / vm / etc.
You never need to do this. Simply create an exit node into your subnet, and everything you want becomes routable.
It sounds to me like your architecture struggles with a zero-trust network approach if you view this as a blocker. I've got some slots available if you need a consultation!
> In terms of SSH, we use old-school OpenSSH and SSH certificates. Its really not that difficult and its really not expensive, you can do offline signing with Yubikeys, no need for expensive HSMs.
It's expensive in terms of engineer-hours, especially compared to Tailscale. It's also easy to get wrong, and end up with the same vulnerability as TFA, or worse.
But my main issue is that if you see this as equivalent it's telling me that you're not really the target audience. Your set-up is far more vulnerable than the out-of-the-box experience you get with Tailscale. If you don't mind, then I'm happy for you.
Lastly, I don't work for Tailscale and I'm not affiliated with them in any way beyond being a happy user that has solved a lot of problems very easily with their product. I highly recommend it to practically everyone. It's great.
I'm having a hard time taking your comment seriously. It just seems like non-constructive FUD.
Well, they clearly don't if they have an "insecure argument handling" vulnerability.
As others here have said already here, its an "venerable and ancient class of bugs".
Its the sort of thing that should be picked up by modern defensive programming that includes fuzz testing.
And it is CERTAINLY the sort of thing that should be flagged by any competent security audit. "insecure argument handling" is bread-and-butter for security auditors.
> I can't take this seriously. If you were a customer you could, you know, ask them? Or inspect their SOC2 documents?
SOC2, ISO27001 and all that shit is not the same thing.
As I said, anyone serious who is proud of having had their software audited as clean would publish their reports in public. Nothing to hide. And it strengthens your case with customers.
It can always be a suitably redacted management summary written by the auditor. That's what everyone else does.
The finding in TFA was the result of a security audit.
Don't you fucking dare.
Might I point you to the words "We would like to thank Anthropic and Ada Logics for reporting this issue.".
It was not commissioned by Tailscale. It was DONE BY OTHERS AND REPORTED TO TAILSCALE. Just like the fucking disclosure tells you.
Tailscale should not be relying on the random goodwill of others to do random audits of unknown coverage at random intervals.
That is not a serious approach to security.
They should be commissioning their own, paid out of their own pocket, at regular intervals, and publishing the results.
> SOC2, ISO27001 and all that shit is not the same thing.
Hard agree. SOC2/ISO27001 are a thing that might be useful, but are mostly a framework dreamt up by auditors and other people who like wearing suits (and making lots of money).
They have some normative controls. But a SOC2 audit is not going to survive contact with an engineer who can push code and whose job depends on the company they work for making money. Where they have to ship product features, and don't have weeks to justify every line of code they write.
A SOC2 audit is not going to survive contact with a script kiddie in their bedroom (or, a nation-state APT) who has plenty of time and an LLM to help them find vulnerabilities or weird edge cases they can compose to attack a system.
These frameworks might serve a purpose, but corporate 'compliance' departments have been lulled into a false sense of security that a satisfactory SOC2 audit means the product is secure and obviates them from asking more technical details or for a fuller audit. It's not. That doesn't matter in a lot of cases, but probably does for a network security solution - I would absolutely be asking deeper questions if I was deploying any of these products en masse.
But then again, the insurers probably asked for a SOC2 certificate, so I guess like most things in life, it's not about whether the systems are secure - it's about whose insurance is ultimately covering the loss.
Same, you are not alone. The Tailscale VPN stuff just works. None of their competitors can claim this that I'm aware of. We tried several other products and none of them were as reliable and 'just worked'. I imagine they spend a lot of time just keeping that stuff working.
Are you having issues with it when it's always connected?
Meanwhile with wireguard: It just works. Every time. Every where. Unless someone blocks UDP.
Its better than it used to be.
But the fundamental problem is that the Wireguard app is a simple GUI around `wireguard-go` built as a static C library via cgo. But Tailscale uses a fork of `wireguard-go` and then adds control client, DERP, NAT traversal etc. on top of it.
So there's quite a lot of "bloat" on top of the Wireguard code in Tailscale iOS and therefore your problem might not be Wireguard vs Wireguard implementation question but something happening elsewhere in the Tailscale code.
Even then, it's obvious you are running SSH, and they can fingerprint the OS, and external logging shows which machines are connecting, and hence have said SSH keys. If they have SSH open they become targets; if they come from CGNAT, carriers can leak location via CGNAT behaviour.
In contrast tailscale makes this much harder.
Always try to use actual API/system calls (in this case getpwnam) instead of calling sub-processes.
I also run self-hosted Wireguard. Initially on a Debian box, nowadays it is integrated into my router (admittedly, this is closed source). For around 6 years at this point.
The whole thing could not be easier and simpler. It has never randomly broken on me. It is fast. It is free. There is no middle man, no vendor.
I never understood the popularity of Tailscale, though that is on me. I'm sure it is a great product, I just never tried it, do not seem the target audience.
What confuses me is the often accompanying, sometimes aggressive anti-selfhosting stance in these sorts of threads. I do not see this in other topics, e.g. someone mentioning they run Jellyfin isn't met with "why not Plex?". Where does that come from? We are on HackerNews, not ProductShillNews, aren't we? I guess self hosting Wireguard is too boring to warrant any further discussion? The VPN equivalent of a Toyota Corolla.
i have my homelab only reachable via tailscale and can access everything i would ever want on the go that way. it was a matter of 15 min to get it all working.
Where Tailscale comes into its own is automatic managing of mesh networking (like an “sdwan” solution). The other thing it excels at is firewall busting - if you have a firewall (with or without address translation) which only allows outgoing traffic to be established (with UDP timeouts for session) then Tailscale also works in a similar way to turn/stun.
If I needed that capability then I’d be looking at Headscale. I don’t need it though.
Remember that this is hackernews, not slashdot. Where the community used to be far smaller and the technology far smaller it was quite normal for everyone to understand basic building blocks of ip addresses, use open source software, wear t-shirts threatening to replace people with a small shell script etc.
It’s not the same community, many people here have no real understanding of computer fundamentals, but instead have expertise in specific narrow areas. They also have little interest in things like free software, but do have an interest in building a new billion dollar company to sell to a behemoth.
Some would consider that an anti-feature. Firewalls are not to be busted. Nothing good lies at the extreme end of working around overly strict policies. Change the policy instead.
> I guess self hosting Wireguard is too boring to warrant any further discussion?
It's popular because you don't have to deal with NAT punching. It "just works", all the time. And Wireguard is not too boring, it's just not enough on its own.
I'm all for self-hosting and this is exactly why I prefer to use Tailscale and not have to manage jump-hosts and STUN points on some cloud, given that I won't be able to make it as reliable as Tailscale and as cheap as Tailscale (effectively $0). So this is literally the only tradeoff I made while self-hosting everything else.
Can you talk an $elderly_relative through a wireguard installation on the phone so they can join your VPN?
I've done it with Tailscale and it's a breeze.
Others report no issues but I had massive drain on iOS even with only 4 connections open.
Native wireguard is unnoticeable.
Does it do the opposite of that?
I ditched wireguard for tailscale for the ease of managing it. I'd much rather run my own independently but CBA with the config editing hassle.
Is the proper fix not restricting users not possible in these poorly designed ancient systems?
Similarly re another issue: why not just fix the permission issues instead of restricting users?
> Tailscale now disallows the use of UIDs or numeric-only usernames via SSH to avoid this ambiguity
My guess is they hastily threw together something hacky in early development, and forgot to replace it with a real, safe solution later.
It's implemented in libc. So you need to link to libc. Tailscale is a Go binary, and they probably prefer it to be statically-linked. glibc NSS implementation also REQUIRES you to load `.so` so you just can't emulate it in Go.
Then, "link to libc". Which libc? glibc? musl?
Yes there is, and you answered in the next line, it is implemented in libc.
If you want to check authentication use libc don't try to implement crypto and authentication yourself.
https://github.com/tailscale/tailscale/blob/e4144230f410204a...
// userLookupGetent uses "getent" to look up users so that even with static
// tailscaled binaries without cgo (as we distribute), we can still look up
// PAM/NSS users which the standard library's os/user without cgo won't get
// (because of no libc hooks). If "getent" fails, userLookupGetent falls back
// to the standard library.If you are writing go, you usually want to set CGO_ENABLED=0 by default, to avoid inadvertently introducing nonportable code. In this way, only the pure Go implementations are used and there is no need to link (statically or dynamically) against a libc implementation to compile and run your programs.
To clarify: I’m not against CGO, just introducing a glibc dependency by default. I would only introduce it when I needed it.
Instead, they just sneaked the change upon the world.
1.98.9 version exists! That's not the question. It should already have been made available for Linux distros assuming this resource from Tailscale is accurate https://pkgs.tailscale.com/stable/?v=1.98.9
Edit: Their changelog also mentions the version: https://tailscale.com/changelog#all
This project is not associated with Tailscale Inc.
However, one of the active maintainers for Headscale is employed by Tailscale and he is allowed to spend work hours contributing to the project. Contributions from this maintainer are reviewed by other maintainers.
it seems anthropic also use tailscale or it's just being discovered by the mythos model?
Tailscale can also just apply to be their own CNA and issue CVEs for their products themselves, eliminating any such issues entirely.
As single tailnet+single user, perhaps it's just okay
This is incredibly bad engineering, on level of a SQL injection, in 21st century. Something a highschool student experimenting with scripting could come up with, but not a supposedly professional software company.
No ED25519 host key is known for $host and you have requested strict checking.Really? That's the fix?
A proper fix is to use "--" to separate arguments.
Refactoring external invocations to use safe argument handling is a better way to fix it. Along with tests that exercise weird names.
The username policy fixes this issue for good, regardless of whatever you write in the future, or whatever new mechanism is introduced.
It’s a restriction for sure, but it’s not a nonsense restriction? Who would have a username starting with a hyphen? I didn’t even know it was possible until today.
The better fix would be to not have the username pass through a parser looking for cli flags in the first place.
If so, this is kind of an understandably ugly problem, though there is still a better option than shelling out -- systemd-userdb.
Your answer is mostly correct, except that when you tug on that thread the shelf comes off the wall, the plaster comes with it, and then it cracks the water pipes on the way to the floor.
Still, requesting the information from systemd-userdb would be a better way of doing this if it's really necessary.
EDIT: Yeah, some older comments confirm that was their reasoning. https://github.com/tailscale/tailscale/commit/1fc1077052dfa7...
A better fix is to call “getent passwd” with no user controlled arguments and then parse the resulting list. This gets rid of the input sanitization problem entirely.
Reference: https://gitlab.alpinelinux.org/alpine/aports/-/blob/master/m...
Edit: my bad, it's between 008 and 007 ?=
s/bug/backdoor/gIf their scope grows, and they run so much as root, it won't be their last.