Openwall GNU/*/Linux - a small security-enhanced Linux distro for servers
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Fri,  3 Apr 2015 22:06:09 -0400 (EDT)
Subject: Re: Request CVE for LinuxNode - DoS vulnerability

Hash: SHA1


> our package ax25-node

Here are some comments that may help to determine the correct number
of CVE IDs for this report.

> The SIGQUIT routine fails to close the app leaving the IP sockets open and in
> some cases DDOS the remote site if a user "ctrl-]+q" out of a telnet session.
> Also the app fails to close and more can be spawned by a crafty malicious user
> thus bringing the system to a point of no memory available.

We found this within the node 0.3.2 download:

  Node is intended to be called from ax25d or inetd.

  /etc/inetd.conf could have something like this in it:

    # Set up LinuxNode to listen at telnet port
    telnet  stream  tcp     nowait  root    /usr/bin/node     node

As far as we can tell, "if a user "ctrl-]+q" out of a telnet session"
means that the client side is attempting to close the TCP connection
in an unsupported way (the supported way is the "bye" command). We
think that "leaving the IP sockets open" means that the TCP connection
transitions to CLOSE_WAIT for a while, and thus TCP data is still
being transmitted outbound. (The specific TCP state isn't critical -
the continued data transmission is what is of interest.) For node
(unlike, say, a generic telnetd), this is a relatively important
vulnerability because a TCP connection usually occurs over a very
low-bandwidth radio path. We think "DDOS the remote site" does not
really mean DDOS, and instead means "occasionally sends TCP packets
from one source IP address to one destination IP address." Also, for
purposes of defining the vulnerability, it isn't a self-DoS in which
the client can omit the bye command and thereby suffer a loss of
bandwidth that the client otherwise could have used for something
else. Instead, the client has an "attacker" role in which it can omit
the bye command and thereby cause the server to waste bandwidth that
the server otherwise could have used for something else.

777013 adds "Found this was the situation in ALL linux-based ax25 node
packages. Patched this in URONode and informed ax25-node upstream in
2007." As far as we can tell, there are no independent codebases, and
thus any CVE ID would be shared among all related packages. The
SIGQUIT routine in node ends up doing:

  void logout(char *reason)
          log(L_LOGIN, "%s @ %s logged out: %s",, User.ul_name, reason);
          Nodecmds = NULL;

whereas the SIGQUIT routine in URONode ends up doing:

  void node_logout(char *reason)
    node_log(LOGLVL_LOGIN, "%s @ %s logged out",, User.ul_name);
    node_log(LOGLVL_LOGIN, "%s %s", NodeId, reason);
    Nodecmds = NULL;

showing that "derivative work" is a reasonable conclusion.

The only problem being reported in the SIGQUIT routine is that it is
implemented incorrectly in a way that causes TCP connections to spend
too much time in an undesired state, such as CLOSE_WAIT. The report is
not about other issues in the signal handler such as issues.

The remaining concern is "more can be spawned by a crafty malicious
user thus bringing the system to a point of no memory available." This
suggests that there was an intended protection mechanism within node
itself for cases where multiple clients wished to connect at the same
time. (In other words, node was not letting inetd take responsibility
for controlling the number of simultaneous invocations of the
service.) Also, "crafty malicious user" suggests that there was an
unstated way to bypass this protection mechanism. If so, then this
would need a different CVE ID than the one for the
bandwidth-consumption problem.

So, the questions are:

1. Is the above reasonable, i.e., there was (at one time) a single
vulnerability affecting both node and URONode in which a client could
use "quit" within telnet, and thereby cause the server to waste
network bandwidth on a radio path?

2. If the "crafty malicious user" attack could occur separately, are
there any available details about what "crafty" action would occur?
Also, did this issue affect at least one older version of URONode in
addition to node?

- -- 
CVE assignment team, MITRE CVE Numbering Authority
M/S M300
202 Burlington Road, Bedford, MA 01730 USA
[ PGP key available through ]
Version: GnuPG v1.4.14 (SunOS)


Powered by blists - more mailing lists

Your e-mail address:

Please check out the Open Source Software Security Wiki, which is counterpart to this mailing list.

Powered by Openwall GNU/*/Linux - Powered by OpenVZ