Using SSH keys from untrusted clients

We all know and love OpenSSH’s scriptability. For example:

# Burn file.iso from 'host' locally without using disk space
ssh host cat file.iso | cdrecord driveropts=burnfree speed=4 - 

# Create a uptime high score list 
for host in hostone hosttwo hostthree hostfour
do 
    echo "$(ssh -o BatchMode=yes $host "cut -d\  -f 1 /proc/uptime" \
                 || echo "0 host is unavailable: ") $host"
done | sort -rn 

The former is something you’d just do from your own box, since you need to be physically present to insert the CD anyways. But what if you want to automate the latter—commands that repetedly poll or invoke something—from a potentially untrustworthy box?

Preferably, you’d use something other than ssh. Perhaps an entry in inetd that invokes the app, or maybe a cgi script (potentially over SSL and with password protection). But let’s say that for whichever reason (firewalls, available utilities, application interfaces) that you do want to use ssh.

In those cases, you won’t be there to type in a password or unlock your ssh keys, and you don’t want someone to just copy the passwordless key and run their own commands.

OpenSSH has a lot of nice features, and some of them relate to limiting what a user can do with a key. If you generate a passwordless key pair with ssh-keygen, you can add the following to .ssh/authorized_keys:

command="uptime" ssh-rsa AAAASsd+olg4(rest of public key follows)

Select the key to use with ssh -i key .... This will make sure that anyone authenticated with this key pair will only be able to run “uptime” and not any other commands (including scp/sftp). This seems clever enough, but we’re not entirely out of the woods yet. SSH supports more than running commands.

Someone might use your key to forward spam via local port forwarding, or they could open a bunch of ports on your remote host and spoof services with remote port forwarding.

Some less well documented authorized_keys options will help:

#This is really just one line: 
command="uptime",   
from="192.168.1.*",
no-port-forwarding,
no-x11-forwarding,
no-pty ssh-rsa AAAASsd+olg4(rest of public key follows)

Now we’ve disabled port forwarding including socks, x11 forwarding (shouldn’t matter, but hey), PTY allocation (due to DoS). And for laughs, we’ve limited the allowed clients to a subnet of IPs.

Clients can still hammer the service, and depending on the command, that could cause DoS. However, we’ve drastically reduced the risks of handing out copies of the key.

Two classic ways of getting owned

There are two classic ways that Linux newbies open themselves up for pranks and shenanigans (or worse): double-su and startx. The double-su will not cause any holes that a crafty conman couldn’t already have arranged, but the startx trick can actually be a serious back door.

The double-su is when you su twice from some other user’s shell. Imagine, if you will, that Vidar just called over the admin of the company’s server, pointed to top where a process is running un-niced at 99% and has racked up hours and hours of cpu time. Vidar makes a big fuss about this, so the admin says “fine, move over”, and does the following at Vidar’s terminal:


vidar@kelvin ~ $ su
Password:
root@kelvin:/home/vidar# renice 19 3156
3156: old priority 0, new priority 19
root@kelvin:/home/vidar# su vidar
vidar@kelvin ~ $

He then scampers off to lunch. Spotted the problem? “su” doesn’t switch to another user’s account; UNIX/Linux doesn’t allow non-root users to do that, even if they have the password. Instead, it starts another shell on top of the old one. Then the admin run su again, creating a third shell on top of the other two. Now, when Vidar exits the third shell, he finds himself back at the second one, with full root access:


vidar@kelvin ~ $ exit
exit
root@kelvin:/home/vidar# echo "Want to buy: Baggy pants and a more suitable job. Love, your admin" >> /etc/issue
root@kelvin:/home/vidar# exit
exit
vidar@kelvin ~ $

The admin clearly should have ended his su-session with exit rather than su originaluser Of course, the real issue here is using “su” on untrusted hardware and software.

If Vidar was evil, he could just as easily have set up a software or hardware keylogger, a spoofed su or simply used strace. This is the reason why the double-su is more of a prank opportunity than an exploit.

Now, startx, on the other hand…! Some users, mostly for leetness, like to log in in text mode and then “startx” to start X, instead of a graphical login. What most of these don’t consider, is that both the shell and startx are still running on the virtual console it was started on.

If the user dutifully locks the screen before attending wetware chores, you can hit Ctrl-Alt-F1 to get to this shell, Ctrl-Z and bg. You now have a shell running as this user. If that isn’t enough, you can killall xscreensaver and Ctrl-Alt-F7. You now have an unlocked X session:


vidar@kelvin ~ $ startx
^Z
[1]+ Stopped startx
vidar@kelvin ~ $ bg
[1]+ startx &
vidar@kelvin ~ $ killall xscreensaver
vidar@kelvin ~ $ clear; exit;

This user should at least have used startx & exit to log off the virtual console when X started.

So how serious is this hole? It depends on how far you’re willing to go. Sure, with physical access you can try all sorts of things, like rebooting with a livecd. If you know there’s a bios password you can’t clear, you can take the disk out. If the disk is encrypted, you can try a cold boot attack. But surely by then, the user’s back and is trying to figure out why you’re pouring liquid nitrogen into his hardware.

It might have been easier to hit him over the head before he locked the screen in the first pace.

More seriously, proper startx usage turns getting your stuff from a trivial act of stealthy espionage into a violent crime or an invasive and time consuming thousand-euro procedure. Don’t underestimate that.

If you can think of any other classical security no-nos being reinvented by every new generation of Linux users, do comment!

“Linux ate my ram!”

About once a week I hear some poor newbie scream in terror as he discovers that his box is just seconds away from a gruesome death with barely a few megabytes of memory left. How could this have happened, it was fine when I booted it this morning, why does a Linux box need 2GB of memory just to run Apache, more bloated than Vista, etc, etc.

Then you explain about the wonders of disk caching, and invariably the first question is always “How do I disable it?”

All in all, it’s a lot of repetition.

To avoid this, I registered LinuxAteMyRam.com which features a big flashing “Don’t Panic” sign and answers the most frequent questions as reassuringly as possible.

The goal is to allow people to appreciate Linux’s disk cache for the brilliant, unobtrusive and effective optimization it is, so it skips over some details like the swappiness setting.

If you have thoughts or suggestions, do comment.

Password generation traps

Generating a random password is simple, but generating a secure one is harder if you don’t know what you’re doing. When I looked through the password generation algorithm at work, I found (and fixed) several vulnerabilities and bugs, one of which allowed a remote attacker to crack any known account with a generated password in a couple of minutes.

The password generator itself was just inadequate, but an API misunderstanding made it extremely severe:

The following method signature was used for the password generation:

public static string GenerateReadablePassword(int length, int seed) 

On reading “int seed”, your head should be ringing with warning bells. More on that later. The real kicker was the invocation:

string Password = GenerateReadablePassword(10, DateTime.Now.Millisecond);

If you’re familiar with the C# API, you’re likely rolling on the floor about now. Otherwise, the part about DateTime.Now will give you the shivers. And then you realize that this wouldn’t compile without a cast unless DateTime.Now.Millisecond is an int or narrower, which would make for a pretty lousy timestamp. Oh yes… This is the number of milliseconds into the current second.

A new account got one of 1000 possible passwords, for an effective “two characters, no uppercase or symbols” policy.

Even if this had been the proper epoch time in milliseconds, it still wouldn’t have been secure. There are only 86 400 000 milliseconds in a day, and only 28 million of them are during office hours. If you can narrow it down to a specific hour, you have 3 million possible ones. And if you can hear the beep of the user’s e-mail client as the (plain-text) password e-mail is received, you’re back down to a few thousand.

Going from a timestamp to an actual high quality pseudo-random seed is better, but it still doesn’t win any prizes. You then have a best case of 232 possible passwords regardless of how long you specify the password to be.

Another example of this, included in our source code, was a copy of the first google hit for “generate random password C#”. I won’t link to it in case I increase its standing. It’s a code sample that claims that it “Generates random password, which complies with the strong password rules”, and here is an excerpt:

        // Generate 4 random bytes.
        RNGCryptoServiceProvider rng = new RNGCryptoServiceProvider();
        rng.GetBytes(randomBytes);

        // Convert 4 bytes into a 32-bit integer value.
        int seed = (randomBytes[0] & 0x7f) << 24 |
                    randomBytes[1]         << 16 |
                    randomBytes[2]         <<  8 |
                    randomBytes[3];

        // Now, this is real randomization.
        Random  random  = new Random(seed);

No, this is not real randomization, this is 31 bits of high quality randomness sprinkled with a snake oil vinaigrette. Any password of any length generated by this function will not be more secure than a proper 5-character alphanumeric password. You have a secure PRNG right there! Use it!

Project: Screenshot diary

So to try something new, I’ll write about a little scripting project you can try for laughs and learning. If you find this too basic, you can browse the “Advanced Linux-related things” category (and there’s an RSS feed for just those posts as well).

Now, if you know that someone is taking your picture, you try to smile and look natural (but invariably fail, with a strained smile and rigid pose as if you were caught grave robbing). The equivalent in screenshots is to either close all your apps (if you like your background image) or run a bunch of random ones (if you don’t), and then opening the program menu two or three levels. Judging from most screenshots you see, people are constantly contemplating which of their many lovely apps to run next:
Screenshot just as described Screenshot just as described Screenshot just as described

How about this for an idea: Take a new screenshot at random intervals while you actually use the desktop.

Not only do you always have a natural looking screenshot if anyone should ask, but you get basically a little timelapse of your activity. I set up such a system in 2004, and it’s more fun than it should be to flip through them all!

To do this, we’ll make a script and stick it in the crontab. Since cron can only run things at fixed intervals, we’ll use short intervals and make the script randomly choose if it should take a screenshot or not. When it does, it’ll put it with a timestamp into some directory.

Open ~/bin/takeshot. First the shebang, and in 99 out of 100 cases, we’ll just exit:

#!/bin/bash 

if (( RANDOM % 100 ))   
then
	exit 0
fi

Let’s define a good place for our screenshots:

directory=~/screenshots

Since the crontab runs independently of our X11 session, we have to specify which display to use. :0 is the first display, which on a single user box is probably the only one:

export DISPLAY=:0

While the screensavers are very nice, I don’t really want screenshots of them. Xscreensaver comes with a tool that can be used to check if the screen is currently blanked. For simplicity we use the short-hand && notation rather than a full if statement:

xscreensaver-command -cycle 2>&1 | grep -q 'cycling' && exit 0

This only works for xscreensaver, not for other screen saver packages such as xlock or the KDE screensavers. Feel free to skip.

Now let’s create the output directory if it doesn’t exist, and define the filename to use:

mkdir -p "$directory"
output="${directory}/shot$(date +%Y%m%d%H%M%S).png"

This gives us a filename with the current date and time, such as ~/screenshots/shot20090314232310.png.

Now to actually take the screenshot. There are tons of utilities for this, but the two main ones are ‘import’ from ImageMagick; and xwd (from X.org) plus NetPBM to convert it. Import is simpler to use, but I’m a fan of NetPBM for its modularity. Plus NetPBM produces png files that are half the size of ImageMagick’s. Here are both ways:

# Using ImageMagick 
import -win root "$output"
## NetPBM Alternative: 
# xwd -root | anytopnm | pnmtopng > "$output"

Now chmod +x ~/bin/takeshot and try running it a few times (you might want to temporarily delete the zeroes in “100” to speed things up). Check that the screenshots are there.

Now add it to cron. Run crontab -e and add

*/10 * * * * ~/bin/takeshot &> /dev/null

Save and exit whichever editor crontab -e invoked for you.

The script should now be taking a screenshot on average every 100*10 minutes, or 17 hours of actual use time. You can adjust either factor up and down (or make an even more clever scheme) to get more or less screenshots.

To summarise the script:

#!/bin/bash 

if (( RANDOM % 100 ))   
then
	exit 0
fi
directory=~/screenshots
export DISPLAY=:0

xscreensaver-command -cycle 2>&1 | grep -q 'cycling' && exit 0

mkdir -p "$directory"
output="${directory}/shot$(date +%Y%m%d%H%M%S).png"

# Using ImageMagick 
import -win root "$output"
## NetPBM Alternative: 
# xwd -root | anytopnm | pnmtopng > "$output"

Here are some random screenshots of mine from different years and wms:

Bunch of terminals in Fluxbox Bunch of terminals in KDE Bunch of terminals in Ion3 Bunch of terminals in Ion3, now widescreen

Multithreading for performance in shell scripts

Now that everyone and their grandmother have at least two cores, you can double the efficiency by distributing the workload. However, multithreading support in pure shell scripts is terrible, even though you often do things that can take a while, like encoding a bunch of chip tunes to ogg vorbis:

mkdir ogg
for file in *.mod
do
	xmp -d wav -o - "$file" | oggenc -q 3 -o "ogg/$file.ogg"
done

This is exactly the kind of operation that is conceptually trivial to parallelize, but not obvious to implement in a shell script. Sure, you could run them all in the background and wait for them, but that will give you a load average equal to the number of files. Not fun when there are hundreds of files.

You can run two (or however many) in the background, wait and then start two more, but that’ll give terrible performance when the jobs aren’t of roughly equal length, since at the end, the longest running job will be blocking the other eager cores.

Instead of listing ways that won’t work, I’ll get to the point: GNU (and FreeBSD) xargs has a -P for specifying the number of jobs to run in parallel!

Let’s rewrite that conversion loop to parallelize

mod2ogg() { 
	for arg; do xmp -d wav -o - "$arg" | oggenc -q 3 -o "ogg/$arg.ogg" -; done
}
export -f mod2ogg
find . -name '*.mod' -print0 | xargs -0 -n 1 -P 2 bash -c 'mod2ogg "$@"' -- 

And if we already had a mod2ogg script, similar to the function just defined, it would have been simpler:

find . -name '*.mod' -print0 | xargs -0 -n 1 -P 2 mod2ogg

Voila. Twice as fast, and you can just increase the -P with fancier hardware.

I also added -n 1 to xargs here, to ensure an even distribution of work. If the work units are so small that executing the command starts becoming a sizable portion of it, you can increase it to make xargs run mod2ogg with more files at a time (which is why it’s a loop in the example).