Samba 3.0.0 - 3.0.25rc3 are subject for Remote Command Injection Vulnerability (CVE-2007-2447), allows remote attackers to execute arbitrary commands by specifying a username containing shell meta characters.

TL;DR Link to heading

The root cause is passing unfiltered user input provided via MS-RPC calls to /bin/sh when invoking non-default "username map script" configuration option in smb.conf, so no authentication is needed to exploit this vulnerability.

Source Link to heading


/* first try the username map script */

if ( *cmd ) {
        char **qlines;
        pstring command;
        int numlines, ret, fd;

        pstr_sprintf( command, "%s \"%s\"", cmd, user );

        DEBUG(10,("Running [%s]\n", command));
        ret = smbrun(command, &fd);
        DEBUGADD(10,("returned [%d]\n", ret));

        if ( ret != 0 ) {
                if (fd != -1)
                return False;

The smbrun() function is responsible for executing system commands!


#ifndef __INSURE__
        /* close all other file descriptors, leaving only 0, 1 and 2. 0 and
           2 point to /dev/null from the startup code */
        int fd;
        for (fd=3;fd<256;fd++) close(fd);


        /* not reached */
        return 1;

PoC Link to heading

To exploit this vulnerability, all you need to do is to change the username for authentication to :

username = "/=`nohup mkdir /tmp/foo`"

In this example we just execute a mkdir command to create a folder in the /tmp directory. For educational/research purposes I create a python script to exploit this vulnerability and gain a reverse shell.

Screenshot Link to heading