Devil in the details - why package signing matters

Kurt Seifried, [email protected]


This article has been updated on October 24th. I love it when I get to write a happy ending. See section marked "Update".


On October 23rd as I was downloading Red Hat 7.2 release I noticed something strange. While idly checking the package signatures on the files I had two of them weren't signed correctly. At first I thought it might be an issue in GnuPG or rpm, so I ran all the Red Hat 7.1 files through an "rpm -K" and examined the output. Everything there worked fine (as it always has). But when I tried it on the 7.2 files the same error came up (edited for brevity):

[[email protected] RPMS]$ rpm -K *.rpm
pam-0.75-14.i386.rpm: md5 gpg OK
pam-devel-0.75-14.i386.rpm: md5 gpg OK
redhat-release-7.2-1.noarch.rpm: md5 OK
rmt-0.4b22-6.i386.rpm: md5 gpg OK
rpmdb-redhat-7.2-0.20010924.i386.rpm: md5 OK
ttfonts-ko-1.0-10.noarch.rpm: md5 gpg OK

Huh. redhat-release and rpmdb-redhat are not signed correctly? Has been broken into? So I downloaded those files from another server. Same result.

No I start to seriously wonder what is going. Have several ftp servers been broken into, and the files replaced? This seems rather unlikely since every server I checked it was the same, only rpmdb-redhat and redhat-release were not properly signed. This leaves three other possible causes, both undesirable. The first is that someone managed to subvert the process by which mirror sites get files from Red Hat. Via DNS poisoning perhaps, or breaking into the Red Hat distribution server. This is exactly the kind of attack that proper cryptographic signing of packages will defeat however. Even if the attacker pulls of a perfect attack, switches the packages around, and leaves without a trace any end users that tries to verify the package signatures will very quickly find out these are not the original packages that were issued by the vendor. The other option is that Red Hat had a failure in it's security process somewhere, and the packages were not signed, managed to slip through quality control, and were released to the public. Both would indicate a major failure in the security process and practices of Red Hat as a company and be cause for concern by any customer. To make a bad analogy: would you buy a new car if you noticed pieces were missing?

The third option is what worries me the most. Red Hat made a conscious decision not to sign these packages. The actual reason for this escapes me. These packages were signed in the Red Hat 7.1 file distribution (contrary to what Marty Wesley Red Hat OS Product Manager is quoted as saying in an email forwarded to me by Brian McWilliams, a reporter for NewsBytes) [minor spell check fixes]:

While security should always be an important concern, this is not a
security problem. These two packages are not signed and have not been
signed since they existed. Contrary to the report, these
packages were not signed for RHL7.1 either.

rpmdb-redhat = This is a database listing of files included with the Red
Hat distribution. Think of it like an index. It is simply an
informational database, thus the worst any attacker could do is change
the information in the database so that every time a customer performed
an inquiry it replied with an erroneous message.

redhat-release = This is the package that identifies the system as Red
Hat Linux 7.2 (Enigma) when you login. The worst case for us is if this
package is changed, it break Red Hat Network as it uses this file to
determine what version of Red Hat they are using.

As far as the part of the report that addresses exploits when running as
You have to be root to install packages, so this isn't an exploit.
If someone wants to modify software on a machine and they have root
access, there
are far easier ways to cause damage (like deleting every file on
the machine). Why bother installing a bogus rpm?


There are several significant problems and completely incorrect statements in this message. The first of course is that the packages have never been signed since they existed (perhaps he means Red Hat 7.2 betas and not 7.1 and before). The second major problem is the statement "thus the worst any attacker could do is change the information in the database so that every time a customer performed an inquiry it replied with an erroneous message.". This is completely untrue. An attacker could easily add a file to the rpm package that creates a backdoor, installs Trojan software, adds user accounts, or any other number of possibilities. The other attack would be to add a script to the package, this script could add users and groups, create backdoors using installed software like telnet, ssh, bash or login, or could even call rpm to install packages from remote sites. The same goes for the statement about redhat-release. The last section is completely wrong as Marty misses the point of the advisory entirely. The attacker doesn't need root access on your system. He simply needs the subvert the files you are installing from, for example by breaking into an ftp server, or poisoning your DNS cache so that you are redirected to the attackers site instead of a legitimate site. Even if you take the time and care to verify the files signatures you are left with two unsigned packages that cannot be verified. The MD5 sums of the packages are not available in any trusted locations (i.e. Red Hat website), so they cannot be easily used to verify integrity. So what is the user to do? Modify the default install so they do not install these packages (redhat-release is most often installed by default).

Ultimately this practice is extremely dangerous for one simple reason. Cryptographic signing of rpm packages is the primary defense against attackers modifying these packages before you install them. By not signing some packages Red Hat has started the training process that will result in people routinely installing RPM packages that are not correctly signed, such as security updates. This is completely unacceptable. This problem will continue until Red Hat (like most other vendors) religiously signs all packages and encourages end users to check these signatures. Several vendors have already taken this step, making the default in update software to check the signature and reject the upgrade package if it is not signed properly.


Reference links: -

NewsBytes - Beware New Red Hat Linux Release, Expert Warns

NewsBytes - Red Hat Denies Security Flaw in `Enigma'



Well Red Hat has emailed me, and to quote Preston Borwn in two seperate emails:

We have now signed them.  Our MD5SUM files for the ISO images are now
signed as well with our GPG key, to allay any potential fears of mischief.


We will sign these packages going forward.

This makes me very happy and is the right thing to do. Hopefully other vendors will learn from this and not repeat the mistakes made by Red Hat, but if they do hoepfully they will respond as quickly, and ultimately as correctly as Red Hat has.



Last updated 24/10/2001

Copyright Kurt Seifried 2001