Chroot in OpenSSH

bsdphx writes "OpenSSH developers Damien Miller and Markus Friedl have recently added a nifty feature to make life easier for admins. Now you can easily lock an SSH session into a chroot directory, restrict them to a built-in sftp server and apply these settings per user. And it's dead simple to do. If you need to allow semi-trusted people on your computers, then you want this bad!"

Why bother?

[whoever57]

Didn't we just read that chroot "jails" are not secure?

Re:Why bother?

[bsdphx]

Understanding the issues is better than parroting what you've heard from random sources. Given the OpenBSD and OpenSSH track record for security it's obvious they have some serious clues about security.

Re:

[khellendros1984]

I think English speakers are slowly dropping use of "ly" on the ends of adjectives. If so, it's one of those things that will slowly happen more and more, until that's just the way it's done.

Re:

[bytesex]

'If so, it's one of those things that will slow happen more and more, until that's just the way it's done.'

Did that sound good to you ? (Ironically, the word 'good' should really have been 'well' in the previous sentence)

Re:

[IpalindromeI]

Ironically, the word 'good' should really have been 'well' in the previous sentence

Did it change the meaning to something other than what was intended? Because that's what ironic means.

Making the same mistake yourself while correcting someone else is not ironic. It's just humorously coincidental. Please watch this episode [wikipedia.org] of Futurama for further education.

Re:Why bother?

[Wesley Felter]

Didn't we just read that chroot "jails" are not secure?

I've read those arguments and find them confusing. Sure, root can break out of a chroot, but what about non-root users?

Re:

[Brian Gordon]

I don't really know anything about chroot security either, but the first thing that occurs to me is that there's way too much code running/setuid'd as root or in kernel/driver space to just change the filesystem interface and expect that to be secure. You can't possibly lock down every interface in all of this code without a more pervasive solution.. especially since we can barely (read: we can't) keep code listening on internet-accessable ports secure from buffer/integer overflows and other tricky busine

The idea

[warrax_666]

is that cron and ls aren't in the chrooted filesystem. That's why they're (supposedly) more secure than just running the daemons "from" /.

Re:Why bother?

[illegibledotorg]

Giving someone a shell and putting them in a chroot crafted to look and function like a full system is one thing.

Giving someone an SFTP session and chrooting them into a subdirectory is another thing.

The feature added in this commit was arguably intended for the latter purpose given the additional changes to the SFTP subsystem that were included. There are countless tutorials and patches and scripts that are available to achieve chrooted SFTP-only access, but now it's been implemented in the core of OpenSSH. In my eyes, this solution is not only a "cleaner" solution to the problem, but it's probably more secure too.

Re:

[mike_sucks]

"Giving someone an SFTP session and chrooting them into a subdirectory is another thing."

Yep, that's precisely why my ears perked up upon hearing this, for hosting providers that want to provide secure remote file access, this is awesome!

/Mike

Re:

[schon]

that's precisely why my ears perked up upon hearing this, for hosting providers that want to provide secure remote file access, this is awesome

My question is: what took so long?

The (now defunct) commercial SSH has had this feature for almost 10 years.

Re:

[mike_sucks]

Yeah, good question. I dunno - maybe everyone was standing around asking "when is someone going to do this?" and using FTP(S) in the mean time. I might have been. ;)

/mike

Re:Why bother?

[jandrese]

They are probably better than giving semi-trusted users full filesystem access, even if they aren't perfect security. It's not even that chroot is inherently broken, it's just that people were using it incorrectly (setting it suid or letting the user become root inside of their jail). Most of the complaints seem to be "the user managed to get root and broke out of the jail", which is a problem with whatever allowed your user to become root in the first place, not the jail itself.

Basically, to break out of a chroot you need to be root. If you're root, then you've already defeated the security on the box anyway. Don't let untrusted users become root.

Tell us more.

[Valdrax]

I'd like to hear more about methods of breaking out of a chroot jail without becoming root during the process.

Re:

[jandrese]

Yeah, every method I've seen starts with:

Step 1: Become root

Once you are root, there are dozens of ways to break out of a jail (all the way from modifying kernel memory structures directly to rewriting inodes to installing a kernel module that grants you access to whatever you need, etc...

Re:Tell us more.

[sjames]

In the right circumstances, 2 non-root users can conspire to break out of jail if one is chrooted below the other.

Let's say A is chrooted to /home/sorta-trusted and B to /home/sorta-trusted/not-so-much.

A diropens his / and creates a unix socket in /not-so-much. B opens the socket in his /. Now, A passes his fd to his / to B. B then does fdchdir on the fd and he's out of jail. Now B can break A out.

The moral is, never use nested chroot jails!

Re:

[Just Some Guy]

A diropens his / and creates a unix socket in /not-so-much. B opens the socket in his /. Now, A passes his fd to his / to B. B then does fdchdir on the fd and he's out of jail. Now B can break A out.

That's pretty interesting! In practice, though, what would that get them? B would now have access to A's chroot (possibly - file permissions might be too restrictive anyway). If A trusts B that much, they could just share A's login.

Not saying this isn't serious, but I'm too tired to think of a useful attack vector at this moment. Am I missing something?

Re:

[sjames]

It's far worse, at least in the Linux kernel (and quite probably other Unix as well but I haven't studied them). The linux kernel assumes that the PWD is at or below the chroot. When a system call parses a pathname, it substitutes the chroot for a leading /, and when walking through a .. in a pathname, it checks that the current directory in the walk isn't already the chroot before following .. up.

So, once B gets to A's chroot (/home/sorta-trusted), it can access the real / as ../.. because now, the ..'s

Re:

[Just Some Guy]

You know, I've had two "oh crap!" moments involving computer security. The first happened when I learned about SQL injection. The second occurred just now when I read your letter. Thanks for the followup!

Re:Why bother?

[parcel]

Didn't we just read that chroot "jails" are not secure?

You may want to take a look at http://www.openbsd.org/faq/faq10.html#httpdchroot [openbsd.org], especially the section titled "Should I use the chroot feature?".

I imagine something similar would be forthcoming regarding OpenSSH specifically.

Re:Why bother?

[jjohnson]

They're not secure for root users. That was the issue identified in the recent "read" you mention--someone was pointing out that root can break out in a particular way, and the kernel devs responded that 1) that was by design, and 2) locking that down left an infinite number of other ways to break out if you're root.

For regular user accounts, a properly configured chroot jail is still a very useful security tool.

Re:

[ianezz]

Didn't we just read that chroot "jails" are not secure?

Only when you have full shell access. This patch is just about confining sftp file transfers via chroot(2) for some users without the burden of setting up a full chrooted environment. Sounds really sweet.

FTP servers have been doing this for years

[caseih]

The purpose of this feature doesn't seem to be to restrict what a shell user can do. Rather, if I read this correctly, it restricts what files a user can access via sftp. Without this feature, a user can sftp in, and then cd to / or any other folder that he has rights too. This chroot feature lets the admin limit the root to, say, his home directory, or some other folder such as a virtual web root or something.

It's only natural that this same chroot feature would be added to sftp.

Re:

[jjohnson]

You misunderstood that the whole debate was about what a root user could do from within a chroot jail. A properly limited user account can be secured in a chroot jail (i.e., not simply chrooting but configuring a jailed environment).

The devs were correct that chroot alone was not designed for the purpose of security and can't be used alone to provide any.

Bad to the bone

[Anonymous Coward]

I want it good. Perhaps you meant badly?

Have you seen that new Ferrari? That's one bad piece of machinery!

(However, in this case you're correct in your usage.)
 

Re:

[nuzak]

I love the PowerGlove, it's so Bad!

Oh thank god

[Just Some Guy]

Now I can finally switch some customers from FTP to SFTP. Thanks for making this hugely useful change!

Anyone know if SFTP logging will be added any time soon? That's the last missing feature i always have to manually patch in.

Re:

[pembo13]

All we need now is some form of virtual user system that can be mapped to a real, unprivileged user, preferably with a flexible auth system.

Re:

[Just Some Guy]

For real? My sarcasm meter is out of whack today so I can't tell.

Re:

[pembo13]

For real

Re:

[Just Some Guy]

OK, just checking. It's been a long day. :-)

Anyway, would something like Kerberos fill the bill for you? I use it to navigate around our network, and it's not that hard to do once you get over the initial learning curve. If you're stucking working with Windows, you can even auth against an Active Directory domain.

Re:Oh thank god - vsftpd v sftpd vs. vs ftpd.

[anon mouse-cow-aard]

I really love how vsftpd works.

This is a "Very Secure FTP Daemon" I would love for it to be configured exactly the same, but
the transport protocol would be Sech-file-xfer draft protocol (SFTP)

vssftpd ?

how about just a protocol option in the config of vsftpd...

Re:

[David_W]

That's the last missing feature i always have to manually patch in.

I'm still hankering for tab completion in SFTP myself... maybe someday.

Re:

[PylonHead]

Use lftp

It will let you connect to sftp servers, and have a sane command line experience. It also has many nifty mirroring commands.

Why is this news?

[Abattoir]

I was doing openssh(+sftp) with chroot on Solaris 2.6 several years ago. Does this have some Ubuntu GUI to make it easy or something?

Re:

[Doctor Crumb]

This is news because the chroot and sftp server are now built in to the openssh binaries, so you don't have to manually set up the chroot. While there's no GUI, it is in fact now easier to set up such a thing.

Re:

[LurkerXXX]

Ubuntu GUI? Where do you get Ubuntu from? This is from the folks who make OpenSSH. And OpenSSH is made by the OpenBSD folks, not Ubuntu.

This simply makes it much easier to do what many folks have been setting up manually for a long time. I'm hoping they'll follow it up with SCP support as well. Time to buy another CD set to support the project :)

Nothing New

[NYFreddie]

This isn't really anything new. This functionality has existed as a patch for a while. It's still nice to see that it's finally being integrated into the main tree, though.

Does This Mean

[ajs318]

Does this mean that I can give users shell access, by placing (hard links to) a stripped-down busybox and ash in $HOME/bin, and they won't be able to access anything outside the chroot environment? That could be sweet.

all that for sftp?

[sgt scrub]

It is cool tech but not the way I would do things. WebDav with ApacheSSL properly installed is lots safer. IMHO there should never be user accounts on a machine, other than root and the person administrating the box.

Re:all that for sftp?

[evilviper]

WebDav with ApacheSSL properly installed is lots safer.

Why? No privilege separation. A MUCH bigger code base.

Not to mention fewer standalone programs.

IMHO there should never be user accounts on a machine,

Why not? The user security model is reliable and time tested. It does not require reinventing the "user". It does not depend on one program handling it's own system of virtual permissions correctly. It does not depend on the security of a large program that users directly interact with.

I can see ample reasons sftp is safer.

Re:

[evilviper]

You're welcome to break into any of my systems using a user account that has been disabled, and has no permission to access anything. eg. user "nobody". That's precisely how I would set up a sftp group.

Re:

[weicco]

So you are running Apache as root? Scary.

RSSH ?

[ElberethZone]

This looks similar to the features in RSSH http://www.pizzashack.org/rssh/ [pizzashack.org]...

Re:

[Just Some Guy]

it doesn't, unless you want not to allow users to run anything?

Since SFTP stands for Secure File Transfer Protocol, it's not really designed to allow you to execute commands.