Differences between revisions 138 and 139
Revision 138 as of 2008-05-16 09:20:00
Size: 34284
Editor: ?Dererk
Comment: Added Kerberos (MIT and Heimdal) advisory
Revision 139 as of 2008-05-16 10:42:38
Size: 34387
Editor: ?Justin''''''Mason
Comment: GeoTrust RapidSSL keys are not covered by free reissuance policy :(
Deletions are marked like this. Additions are marked like this.
Line 347: Line 347:
 * GeoTrust: https://certs.tucows.com/geotrust_agreements/refund.htm (Available throughout the lifetime of the certificate. Tucows/OpenSRS in this case, but the instructions are generic to any GeoTrust client.)  * GeoTrust: https://certs.tucows.com/geotrust_agreements/refund.htm (Available throughout the lifetime of the certificate. Tucows/OpenSRS in this case, but the instructions are generic to any GeoTrust client.  Note however that RapidSSL keys are excluded from this, unless you've paid for "reissuance insurance")

/!\ As this page is changing rapidly, please be careful not to introduce conflicts; respecting edit locks would be a good idea and please do not use the GUI editor, as it introduces huge numbers of whitespace changes which make it harder to keep track of what is changing on this page (Changing one line with the GUI editor can make a 17KB diff as it arbitrarily reformats the text). Also don't check the "remove trailing whitespace" checkbox as it also introduces many changes.


  • ?TableOfContents

In [http://www.debian.org/security/2008/dsa-1571 Debian Security Advisory 1571] (New openssl packages fix predictable random number generator), the Debian Security Team disclosed a vulnerability in the openssl package that makes many cryptographic keys that are used for authentication (e.g. through SSH) or signing (e.g. web server certificates) potentially vulnerable.

End User Summary

The scope of the problem includes:

  • weak keys for both clients and servers (see section "Identifying Weak Keys below")
  • all key types that were generated using openssl (this includes RSA and DSA keys)
  • compromise of other keys or passwords that were transmitted over an encrypted link that was set up using weak keys. Note that this last point means that passwords transmitted over ssh to a server with a weak dsa server key could be compromised too; see the [http://lists.debian.org/debian-devel-announce/2008/05/msg00003.html Debian project's reaction to this].

The following cryptographic tools are unaffected:

  • cryptsetup (neither LUKS nor the regular dm-crypt use openssl, the openssl keyscript - which is not used in any default installations - does use openssl, but only to encrypt the key, not to actually generate the key that is used to encrypt the partition, the encryption of the key may therefore be less strong than expected but the key itself is not)

  • GNUTLS
  • GnuPG

Identifying Weak Keys

Characteristics of potentially vulnerable keys:

  • Generated since 2006-09-17
  • Generated with Etch, Lenny or Sid (Sarge is not vulnerable)
  • Generated using 'openssl', 'ssh-keygen', or 'openvpn --keygen' (GnuPG and GNUTLS are not affected)

Additionally, simply the use of DSA keys may have compromised them. A strong key (i.e., generated with a 'good' OpenSSL) but used locally on a machine with a 'bad' OpenSSL must be considered to be compromised. This is due to an 'attack' on DSA that allows the secret key to be found if the nonce used in the signature is reused or known.

If you have used the vulnerable openssl version to generate an SSL key and then created a CSR from that key and sent it to a SSL certificate provider, then the certificate should be considered vulnerable.

A full list of 'weak' keys has been generated by the metasploit project: http://metasploit.com/users/hdm/tools/debian-openssl/

Applications/protocols known to use these keys:

  • OpenSSH (both server and user keys)
  • OpenVPN
  • OpenSWAN/StrongSWAN
  • DNSSEC
  • key material for X.509
  • encfs
  • Tor
  • postfix, exim4, sendmail and other MTAs when using SSL/TLS
  • cyrus imapd
  • courier imap/pop3
  • dovecot with imaps/pops support
  • apache2 (ssl certs)
  • dropbear
  • cfengine
  • puppet
  • xrdp
  • tinc
  • vsftpd SSL certificates for FTPS
  • proftpd SSL/TLS certificates for FTPS
  • ftpd-ssl SSL certificates for FTPS
  • telnetd-ssl SSL certificates for SSL-Telnet
  • ?DomainKeys (DK) and DKIM

To fix this, first aptitude update && aptitude upgrade to install the new version of the openssl and libssl0.9.8 packages (the vulnerability is fixed in version 0.9.8c-4etch3 for etch and version 0.9.8g-9 for lenny/sid). You probably want to also pick up the new openssh packages that include the blacklist of known weak keys, but you will need to aptitude dist-upgrade for that in order to install the new openssh-blacklist package.

If you choose not to use the above aptitude command, note that all of the following packages must be upgraded (they all come from the same source package):

  • openssl
  • libssl0.9.8
  • libssl-dev

Then, regenerate and distribute any potentially vulnerable keys. Instructions for how to regenerate the keys for these applications are below. You can also test to see if keys are vulnerable using the dowkd.pl utility as described below.

How weak?

Is there any information on how weak these keys are? Are we talking brute-forceable in minutes, or days, or millennia? -- AsheeshLaroia

  • Very weak, according to [http://www.derkeiler.com/Mailing-Lists/Full-Disclosure/2008-05/msg00416.html this post] a key can be broken in around 20min of bruteforcing. You might like to look at things like [http://packages.debian.org/fail2ban fail2ban] as well to either mitigate or at least ignore script kiddies who are trying this.

  • It looks likely that the only remaining source of entropy in the generated keys comes from the PID of the process. This is 16 bits, typically much less effective entropy. So there may, in fact, be just a few thousand possibilities for a specific key size. Looks like 'dowkd.pl' lists about 262,000 entries. -- ?MarshRay

    • Are you telling me that the only source of entropy is the process' PID, and some uninitialized memory that may even be filled in a predictable manner by the execution trace leading up to use of the buffer? -- ?KenBloom

      • Under correct circumstances, OpenSSL was, indeed, adding so-called uninitialized memory to the entropy pool, but this was not the primary source of entropy (provided externally through ssleay_add_rand). Please see below for more detail. -- ?NathanielFilardo

    • A perl script is available that tests keys for weakness. This perl script is <10 MiB in size and contains a pre-generated list of keys. This suggests that the key search space is very, very small. -- ["Mentor"]

    • As pointed out on the metasploit page, there are only 2^15 unique keys generated. This is *reall* small. -- ["Mentor"]

Application Details

BIND9

boxbackup

Upstream has published a first impact analysis of key material created on system with insufficient random PRNG. You can read the details here: http://lists.warhead.org.uk/pipermail/boxbackup/2008-May/004476.html

From that post:

If the PRNG in your OpenSSL was insufficiently random, you need to:

  • Regenerate all affected certificates, which have been generated or

signed on an affected system

  • Regenerate all the data keys (*-?FileEncKeys.raw)

  • Destroy the data stored on your server to an appropriate level of

security (overwrite with zeros at the least, more if you're paranoid)

  • Upload everything again
  • Take appropriate measures under the assumption that you have been

storing your data in plain text on a public server without authentication.

(ie, start from scratch, destroying all trace of the backed up data, and take other measures to mitigate the exposure of your secrets)

Cfengine

For each cfengine host, remove the old keys and generate new keys:

rm /var/lib/cfengine2/ppkeys/localhost.priv
rm /var/lib/cfengine2/ppkeys/localhost.pub
/usr/sbin/cfkey
Then restart cfservd:
/etc/init.d/cfengine2 restart

Once the keys are regenerated, exchange keys between hosts as necessary to reestablish 2-way trusts.

courier imap/pop3

Follow the "Generic PEM Generation" instructions and add a openssl gendh >> mysite.pem.

Or

rm /etc/courier/imapd.pem
dpkg-reconfigure courier-imap-ssl

and let dpkg generate back an imapd.pem file.

csync2

As described in /usr/share/doc/csync2/README.Debian

/etc/csync2_ssl*
openssl genrsa -out /etc/csync2_ssl_key.pem 1024
openssl req -new -key /etc/csync2_ssl_key.pem -out /etc/csync2_ssl_cert.csr
openssl x509 -req -days 600 -in /etc/csync2_ssl_cert.csr \
        -signkey /etc/csync2_ssl_key.pem -out /etc/csync2_ssl_cert.pem

Then the keys must be distributed out to each host prior to running csync2 again.

cyrus imapd

To find out which certificates/keys are in use, see the directives whose names contain "key_file" or "cert_file" in /etc/imapd.conf.

Generate new private keys and certificates as shown in "Generic PEM Generation" above, then restart the service:

invoke-rc.d cyrus21 restart

if you are using version 2.1 (if you are using another version, please examine /etc/init.d for the correct service name)

dovecot

Generate a new PEM as shown in "Generic PEM Generation" above (making sure that it is placed in the correct place according to your dovecot configuration, then restart dovecot:  invoke-rc.d dovecot restart .

dropbear

If you have /etc/ssh/*host* keys, either remove them, or follow the openssh instructions above, first, before updating dropbear's keys. Dropbear's postinst imports any existing openssh keys if they exist.

rm /etc/dropbear/*_host_key
dpkg-reconfigure dropbear

Note that keys that have been generated by Dropbear itself are not vulnerable (it uses libtomcrypt rather than OpenSSL).

exim4

If TLS is in use, generate a new PEM using /usr/share/doc/exim4-base/examples/exim-gencert --force to get a self-signed certificate. Per default it has a three years expiration duration.

ftpd-ssl

rm -f /etc/ftpd-ssl/ftpd.pem /etc/ssl/certs/ftpd.pem
dpkg-reconfigure ftpd-ssl

Generic PEM Generation

This is just a reminder for those who generate PEM encoded certificates. Your site probably has other policies in place about how to manage keys which you should follow. Additionally, you may need to get the certificates signed again by a 3rd party Certificate Authority rather than by using a self-signed certificate as shown below:

cd /etc/ssl/private
openssl genrsa 1024 > mysite.pem
cd /etc/ssl/certs
openssl req -new -key ../private/mysite.pem -x509 -days 9999 -out mysite.pem

OpenSSH (Server)

See also [http://www.debian.org/security/key-rollover/#openssh Official Key-Rollover page].

Updated packages for openssh that have a blacklist of known weak keys are now available See [http://www.debian.org/security/2008/dsa-1576 DSA 1576] for more information. Installing these packages on hosts with weak keys will cause the ssh server to regenerate its keys. Weak user keys being used for a connection will also be rejected where possible.

/!\ Note that you will have to use aptitude dist-upgrade (or apt-get dist-upgrade) to install these packages rather than just upgrade because this update will cause the new package openssh-blacklist to be installed.

You can also update your openssh-server keys manually.

rm /etc/ssh/ssh_host_*
dpkg-reconfigure openssh-server

/!\ Note that in either case, your users will see a "IT IS POSSIBLE THAT SOMEONE IS DOING SOMETHING NASTY!" warning when they next log on to your ssh server because the key has changed. They will need to edit $HOME/.ssh/known_hosts to remove the offending line before continuing; checking that the key fingerprint is correct, of course (the fingerprint of your new key can be found with ssh-keygen -l -f /etc/ssh/ssh_host_dsa_key). You can remove the key from known_hosts by running "ssh-keygen -R hostname"

Also, note that your existing ssh connections shouldn't be interrupted.

OpenSSH (Client)

See also [http://www.debian.org/security/key-rollover/#openssh Official Key-Rollover page].

You will need to have a list of the openssh keys that you currently have and where they have been copied to. For each key that is vulnerable:

cd ~/.ssh
ssh-keygen -t rsa -f filename
ssh-copy-id -i filename hostname 

Replacing rsa by dsa if you prefer dsa keys and replacing filename and hostname with appropriate values.

Also remember to remove compromized keys from you .ssh/authorized_keys file!

OpenSWAN / StrongSWAN

See [http://www.debian.org/security/key-rollover/#openswan Official Key-Rollover page].

OpenVPN

Backup /etc/openvpn/openvpn.secretkey.* (I placed them in a new subdirectory called vulnerable-DSA-1571-1)

recreate the keys using: openvpn --genkey --secret openvpn.secretkey.system1-system2 

copy the shared secret keys to the other hosts

restart the vpn with /etc/init.d/openvpn force-reload on each host.

postfix

Generate a new PEM as shown in "Generic PEM Generation" above (making sure that it is placed in the correct place according to your postfix configuration, then restart postfix:  invoke-rc.d postfix restart .

puppet

There are two methods to handle puppet certificates, one is via capistrano, the second is manually.

The capistrano method is detailed here: http://reductivelabs.com/trac/puppet/wiki/RegenerateSSL

The manual steps are as follows:

  1. You need to wipe and regenerate your CA info:
    /etc/init.d/puppetmaster stop
    rm $vardir/ssl/*
    /etc/init.d/puppetmaster start 
    However, if you are running mongrel, instead of starting puppetmaster from the initscript, you will need to first stop the front-end web listener (apache, nginx, etc.) and then do the following:

 puppetmasterd --daemonize ; sleep 30 ; pkill -f 'ruby /usr/sbin/puppetmasterd' 

The above is necessary because for some reason when running with mongrel, puppetmaster will not regenerate its CA.

  1. Wipe all the client certs
    /etc/init.d/puppet stop
    rm $vardir/ssl/* 
  2. Have each client request a new cert:
    puppetd --onetime --debug --ignorecache --no-daemonize
  3. Once all the requests have rolled in, you can sign them all at once:
    puppetca --sign --all
  4. Start up your puppet clients:
    /etc/init.d/puppet start

You could also enable autosign temporarily, if you are comfortable with that.

ssl-cert

For the snakeoil certificate /etc/ssl/certs/ssl-cert-snakeoil.pem, just call:

make-ssl-cert generate-default-snakeoil --force-overwrite

telnetd-ssl

rm -f /etc/telnetd-ssl/telnetd.pem /etc/ssl/certs/telnetd.pem
dpkg-reconfigure telnetd-ssl

tinc

Remove all suspect public and private key files:

  1. Remove rsa_key.priv.
  2. Edit all files in the hosts/ directory and remove the public key blocks.

Generate a new public/private key pair:

tincd -n <netname> -K

Exchange your host config file with the new public key with other members of your VPN. Do not forget to restart your tinc daemons.

Tor Onion Router / Hidden Service Keys

Please see the vulnerability announcement at http://archives.seul.org/or/talk/May-2008/msg00049.html.

Upgrading to 0.1.2.19-3 or 0.2.0.26-rc-1 is recommended. If you run a relay these versions will force new server keys (/var/lib/tor/keys/secret_*.) being generated.

If you are hidden service provider, please delete your hidden service's private key. This will change your hidden service's host name and may require reconfiguring your servers.

If you are running 0.2.0.x, an upgrade is highly recommended -- 3 of the 6 v3 authority servers have compromised keys; however 0.1.2.x (using v2 authorities) is OK.

encfs

Analysis of the impact is still ongoing. Details known so far: encfs uses the RNG from libssl to create an internal encryption key with some post-processing applied. Checking of that key against a pre-calculated blacklist like with dowkd.pl might be possible when it's ready.

In the meantime, encrypted filesystems that might be created with a broken version of libssl shall be considered vulnerable against offline attacks. Any copy of encrypted data located in unthrusted environment might become readable, sooner or later. If the date of creation is unknown, checking of the originating version in output of the encfsctl command against the changelog of encfs package may provide some hints but is not aboslutely reliable either because there was no strong dependency on certain openssl version.

To regain security, try following:

  • Create another encfs file system with encfs after upgrading to the latest (fixed) openssl library
  • Mount the original FS and copy the data into the new location
  • Umount the old filesystem. Destroy the contents of the old encrypted files using the shred command. (If the files have been moved rather than copied than make sure that the remaining space on the particular block device or filesystem is rewritten with random data).

xrdp

rm /etc/xrdp/rsakeys.ini
/etc/init.d/xrdp restart

Kerberos (MIT and Heimdal)

If you were using MIT Kerberos, you would be fine so far as I can tell since MIT Kerberos has its own crypto layer and its own randomness functions. Heimdal uses OpenSSL as its crypto layer. Given that, it may well be possible to brute-force the session key of any captured GSSAPI-encrypted traffic and decrypt it retroactively.

If you're using Heimdal, you should also change all long-term random keys (such as any key in generated keytab files) that were generated using the vulnerable version of OpenSSL.

This can be done using cpw -r <principal> within kadmin. Take into account that this principals have been randomed-key assigned and should be regenerated as well.

kadmin/admin
kadmin/hprop
kadmin/changepw
changepw/kerberos
krbtgt/YOUR.REALM
host/SOMEHOST@YOUR.REALM
host/SOMEOTHERHOST@YOUR.REALM

Keys based on user passwords should be fine.

SSL Certificate Reissuance

If you paid good money to have a vulnerable key signed by a Certificate Authority (CA), chances are your CA can re-issue a certificate for free, provided all information in the CSR is identical to the original CSR. Create a new key with a non-vulnerable OpenSSL installation, re-create the CSR with the same information as your original (vulnerable) key's CSR, and submit it to your CA according to their reissuance policy

  • ?GeoTrust: https://certs.tucows.com/geotrust_agreements/refund.htm (Available throughout the lifetime of the certificate. Tucows/OpenSRS in this case, but the instructions are generic to any ?GeoTrust client. Note however that RapidSSL keys are excluded from this, unless you've paid for "reissuance insurance")

  • Thawte: http://www.thawte.com/reissue/ (Available throughout the lifetime of the certificate.)

  • ?VeriSign: Unknown

  • ?GoDaddy: http://help.godaddy.com/topic/234/article/867 (Only possible within 30 days of the initial order. ?GoDaddy calls the process "re-keying", while they call the act of sending you the same signed certificate as your original order a "reissuance".)

  • ipsCA: Generate a new CSR as if you are purchasing a new certificate, follow through the procedure up until you get to the point where you are required to pay with your credit card. At that point contact support via their email and let them know that you are requesting a revocation and re-issue and include the ticket number of your new CSR request.
  • [https://www.cacert.org/ CAcert]: This is a cost free certification authority. Simply revoke your old certificates and add new ones. (The key has to be created on a fixed machine and ONLY the certification request has to be uploaded!) At the moment the certificate generation will take some time as it seems that many users are re issue there certificate.

  • [https://www.digicert.com Digicert]: Login to Your account to re-issue (free).

Testing keys using ssh-vulnkey

In the latest openssh-client packages for etch, ssh-vulnkey is included along with a blacklist of known weak keys. This tool appears to be similar to dowkd.pl (below) although it is restricted to checking ssh keys (not that dowkd.pl can do much more than that anyway).

Run ssh-vulnkey as a user to check your host and your own keys.

Run ssh-vulnkey -a as root to check all users keys (as well as the host key).

Sample output:

ssh-vulnkey -a
Not blacklisted: 2048 fa:2e:1d:a6:84:64:a1:80:c4:31:68:5a:b0:1a:cb:fe /etc/ssh/ssh_host_rsa_key.pub
Not blacklisted: 1024 f4:34:04:85:58:a0:6b:0a:a1:b9:2d:3b:e6:19:5a:76 /etc/ssh/ssh_host_dsa_key.pub
COMPROMISED: 2048 5c:10:8a:c0:55:8c:1f:d9:4b:05:f0:35:0a:0d:2f:5c /home/someuser/.ssh/authorized_keys
Not blacklisted: 2048 a7:b4:3e:41:18:cb:f7:68:5e:4f:ae:30:14:d2:17:fd /home/someuser/.ssh/authorized_keys

/!\ Note: about output is one line per key not linewrapped as shown here in the wiki

/!\ Note: openssl-blacklist-0.1.1 provided for Debian Etch does not include 4096bit RSA fingerprints. Here's an unofficial package I created with them included (another thanks to #debian): http://love.hole.fi/atte/openssh-blacklist/

Testing keys using dowkd.pl

The security team has prepared a utility for you to test your keys to see if they have a fingerprint that matches a list of known-weak finger prints. Download [http://security.debian.org/project/extra/dowkd/dowkd.pl.gz dowkd.pl.gz ] ([http://security.debian.org/project/extra/dowkd/dowkd.pl.gz.asc gpg signature]):

cd /tmp
wget http://security.debian.org/project/extra/dowkd/dowkd.pl.gz
wget http://security.debian.org/project/extra/dowkd/dowkd.pl.gz.asc
gpg --keyserver subkeys.pgp.net --recv-keys 02D524BE
gpg --verify dowkd.pl.gz.asc
gunzip dowkd.pl.gz

/!\ Note that dowkd.pl produces many false positives and that the authors suggest that it may also produce false negatives. On that basis, you might decide that it is simply easier to regenerate your keys as described above.

  • Can we get a version of this that also checks 4096 bit keys, or a version that can crank out new databases?

Use dowkd.pl in the following ways:

SSH Server Host Key

Key is OK:

perl dowkd.pl host localhost
# localhost SSH-2.0-OpenSSH_4.3p2 Debian-9
# localhost SSH-2.0-OpenSSH_4.3p2 Debian-9

Key is weak:

perl dowkd.pl host localhost
# localhost SSH-2.0-OpenSSH_4.3p2 Debian-9
# localhost SSH-2.0-OpenSSH_4.3p2 Debian-9
localhost: weak key
localhost: weak key

A remote check based on the keys generated by HD Moore (http://metasploit.com/users/hdm/tools/debian-openssl/ ) is available at http://itsecurity.net. debian_ssh_scan_v3 now includes fingerprints of all weak DSA 1024, RSA 2048 and RSA 4096 bit keys.

root@box:~/debian_ssh_scan_v2# ./debian_ssh_scan_v2.py 10.128.15.110
65536 fingerprints loaded.
10.128.15.110:22 sshd fingerprint 9cf71acb1b0dff0dceef4f755f721e9d VULNERABLE (RSA 2048 bit key, pid 5252)

SSH User Key

To check one user (someuser):

perl dowkd.pl user someuser
/home/someuser/.ssh/authorized_keys:1: warning: unparsable line
/home/someuser/.ssh/authorized_keys:3: warning: unparsable line
/home/someuser/.ssh/authorized_keys:4: weak key
/home/someuser/.ssh/authorized_keys:5: weak key

Notes:

  • The complaints about lines 1 and 3 are because these lines are comment lines (starting with #) in the ~/.ssh/authorized_keys files (which are allowed, see man 8 sshd for the syntax).

  • Line 2 has a non-weak dsa key that should be replaced anyway according to the information provided above.
  • Lines 4 and 5 are rsa keys that are reported as being weak (note that both rsa and dsa keys are affected by this vulnerability)

To check all users:

perl dowkd.pl user
  • The script does not check the ~/.ssh/authorized_keys2 files, but it should. To fix this, add the following two lines to the from_user() subroutine:

$file = "$dir/.ssh/authorized_keys2";
from_ssh_key_file $file if -r $file;

PEM keys

The dowkd.pl tool is unable to parse PEM files at present.

The Ubuntu team has created a package which will verify if PEM files are vulnerable. I used it against PEM keys created on a known vulnerable system and the utility highlighted them as vulnerable. The package is called "openssl-blacklist". The package is available [http://security.ubuntu.com/ubuntu/pool/main/o/openssl-blacklist/openssl-blacklist_0.1-0ubuntu0.8.04.1_all.deb here]. It can be installed on etch using dpkg -i --force-all.

openssl-blacklist repackaged for Debian: attachment:openssl-blacklist_0.1-0%7Edebian-1_all.deb . Built for etch/stable, but should install cleanly on any later debian versions. I'm not uploading this to Debian proper at present, because I haven't checked whether anyone else is already doing that. For the (signed) sources, see http://xillion.org/openssl-blacklist/ . -- PaulCannon

While the dowkd.pl tool is unable to parse PEM files, there is a way to convert PEM files to SSH-style public keys, which the tool can parse. I've not verified this method completely, it comes from information found at: http://webjob.sourceforge.net/Files/Recipes/openssl-convert-ssl-key-to-ssh-keypair.txt

cp /etc/apache/ssl.key/your.keyfile.key ~/.ssh/id_ssl
ssh-keygen -y -f ~/.ssh/id_ssl  > ~/.ssh/id_ssl.pub
./dowkd.pl file ~/.ssh/id_ssl.pub

I have as yet not determined how accurate this method is, but it seems to work reliably. Someone more familiar with the workings of ssh-keygen can confirm the reliability of this method. Scripting this test should be easy enough.

  • If the code for generating the key in your PEM file used OpenSSL's random number generator even slightly differently than ssh-keygen from openssh, it's likely to have keys that are equally-weak but not contained in that blacklist. I wouldn't recommend drawing any conclusions from this method. -- ?MarshRay

  • Can you tell us how you tested this? We are unable to reproduce your results as none of the keys we generate with the known-weak version of openssl are flagged as being weak by dowkd.pl or dowkd.pl with RichiH's patch (below). Details of how those certificates were generated are below; are you doing the same? -- StuartPrescott

someone who actually understands this tool and its output, please insert details here!

  • I don't know what kind of information is being requested, but it seems fairly easy. On first invocation of the tool, it generates a DB file from the "blacklist" contained in the __DATA__ section of the script. If the DB exists, it is opened for reading. Once the DB exists, it uses ssh-keygen to retrieve the fingerprint of the public key presented to it by the executor of the script, and compares the fingerprint that it retrieves to the blacklist. If there is a match, then the script prints 'weak key'.

    • This is perhaps irrelevant now -- an earlier version of this tool didn't say "weak key" but just printed the key name. It also completely failed to parse correctly formed authorized_key files. Perhaps it has now been sufficiently refined as to now be understandable. This page has been a work in progress throughout the day and getting the bones of the content here so that it could then be added to by others was the priority. -- StuartPrescott

I created a patch to parse PEM files. fw told me to look at the Modulus of the RSA key, so I am doing that. Even though I tested more than 20 certificates generated on an affected machine, I did not find a single weak key. This could mean that

  • the list is incomplete
  • I am looking in the wrong place
  • the certificates are not, in fact, affected by this. Note that I did not generate them myself as I do not have any vulnerable hosts left

As I manually verified the results against, I do not think there is an error in my patch. It is worth noting that none of the certs showed any hits when I converted them into SSH format. Please test this and report back to me!

To test this for yourself, do this

for i in $(seq 1 100)
do
    openssl genrsa 1024 > server$i.key 2>/dev/null
    echo -e "\n\n\n\n\n\n\n\n\n\n" | \
            openssl req -new -key server$i.key -days 365 -out server$i.req 2>/dev/null
    openssl x509 -req -days 365 -in server$i.req -signkey server$i.key -out server$i.crt 2>/dev/null
    chmod 600 server$i*
    ssh-keygen -y -f server$i.key > server$i.pub 2>/dev/null
    perl dowkd.pl file server$i.pub
    perl dowkd.pl pem server$i.crt
done

and please post your results.

  • When we were throwing test cases around on #debian (approx 2008-05-14 02:00 UTC for those who have logs), I threw together this script to run on an unpatched etch box. None of the keys were marked as weak by dowkd.pl when tested by either method even though the keys were generated by a known weak version of openssl. I can only conclude that these methods of testing this class of key for weakness must be incorrect. -- StuartPrescott

    • I think the existing blacklist works only for keys generated with ssh-keygen and a keylength oh 2048 bits. I tried to produce vulnerable keys on a vulenrable system with ssh-keygen and did not get any weak keys as per dowkd.pl for 1024 bits while 2048 bits worked every time. I also tried to generate 2048 bit keys with openssl genrsa and had no success either. I had a look at the generated keys and found that ssh-keygen uses 35 as public exponent while openssl uses 65535. So probably the blacklists don't fit again. Unfortunately this means we cannot use the script for checking keys generated with openssl until we get updated blacklists or a description how these were generated. -- Joachim Ring

Here is a small patch of mine checking also authorized_keys2 and known_hosts: attachment:dowkd.pl.ke.patch.gz (also available [http://www.ethgen.de/~klaus/dowkd.pl.ke.patch.gz here]) -- Klaus Ethgen

I have hacked a little shell script using the tables from openssl-blacklist from ubuntu which allows the user to check the ssl key of a webhost from remote by just calling "chksslkey hostname". You can find it here: attachment:chksslkey.tar.bz2 -- Michael Holzt

Scope of the blacklists

The current official blacklists cover RSA-2048 and DSA-1024 keys as generated on 32-bit little-endian, 64-bit little-endian and 32-bit big-endian systems. That's sufficient to cover users who used the default key length, but not if some of your users decided they wanted a longer keylength.

  • I have made a blacklist for RSA-4096 keys as generated on 32/64-bit little-endian systems and posted it at http://www.red-bean.com/~maxb/ . The 64-bit keys I generated, the 32-bit ones I took from the metasploit site linked earlier on this page. The format is full fingerprints:

    • suitable for appending to dowkd.pl, if you also edit the body of the script so that it tries to check 4096 bit keys too
    • or, if you remove the first 12 characters from each line, suitable for use with ssh-vulnkey
    (There are also RSA-1024 and RSA-1023 versions there, potentially useful for the paranoid evaluation of older keys.)

    --?MaxBowsher

Technical Summary

(If you want to add more technical details that an end-user doesn't need to know or isn't likely to understand, please add them here rather than making the above summary impossible for the average user to understand.)

Causes

It is important to understand that this problem was caused by trying to remove valgrind warnings related to the use of uninitialised memory within the openssl libraries. This was done to try to make it easier to debug C applications that use the openssl libraries which is a good thing to do.

A discussion of why this change was made can be found at [http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=363516 #363516] and also on the [http://marc.info/?t=114651088900003&r=1&w=2 openssl-dev list]. Judging from the [http://www.links.org/?p=327#comment-176642 discussion there], the main culprit seems to be a misunderstanding about which is the right list to ask this question on, followed by misleading answers from the list.

A bit more detail

In an effort to clear up confusion about this bug, here's a bit more technical description. This was caused by an overzealous, well-intentioned elimination of code that was believed to have no impact on security. (Please do note from the above links that this was discussed with the !OpenSSL team and that no objections were raised at the time.)

So here's the problem: the Debian maintainer wanted very much to get rid of valgrind errors while using OpenSSL; certainly a noble cause, right? As you can see here, there are two identical lines,  MD_Update(&m, buf, j);  in SSL's md_rand.c file that were commented out way back in 2006: http://svn.debian.org/viewsvn/pkg-openssl/openssl/trunk/rand/md_rand.c?rev=141&view=diff&r1=141&r2=140&p1=openssl/trunk/rand/md_rand.c&p2=/openssl/trunk/rand/md_rand.c

The second of these is in  ssleay_rand_bytes  where  buf  is used as an output buffer. It had already been marked as a bad idea when -DPURIFY was in effect, because Purify (and valgrind, naturally) dislike this use of an output buffer as input. This use of  MD_Update  is dubious but shouldn't hurt as long as the mixing function of the PRNG is "sufficiently good". The removal of this call to MD_Update should not meaningfully alter the entropy available in the pool -- that is, OpenSSL does not depend upon uninitialized memory for its correct operation.

The first call, however, is in  ssleay_rand_add  where  buf  is used as an INPUT buffer, to add entropy to the pool. Failing to call  MD_Update  there means that the pool will never actually get the entropy intended for it.

Acknowledgements

This document was put together by the folks on [:] to help with user support. It would have been nice to have had the chance to prepare this (and the key rollover page) in the 5 days between the commit being made to pkg-openssl and the DSA being released. Particular credit to themill, dondelelcaro, stew, Maulklin, iobound for reading, writing and suggesting bits for this page. No thanks to the people who keep using the wiki's GUI editor and keep generating 15 to 20 KB diffs from when it randomly reformats the entire page when you only change one line.

See Also

http://www.debian.org/security/key-rollover/


CategorySystemSecurity