A few minutes of preparation and planning ahead before putting your systems online can help to protect them and the data stored on them.
nosuid
option in
/etc/fstab
for partitions that are writable by others than root. You
may also wish to use nodev
and noexec
on users' home partitions,
as well as /var
, thus prohibiting execution of programs, and
creation of character or block devices, which should never be
necessary anyway. /etc/exports
with the most restrictive access possible. This means
not using wildcards, not allowing root write access, and exporting
read-only wherever possible.umask
to be as restrictive as
possible. See
umask settings.unlimited
as is the
default. You can control the per-user limits using the
resource-limits PAM module and /etc/pam.d/limits.conf
. For example,
limits for group users
might look like this:
@users hard core 0
@users hard nproc 50
@users hard rss 5000
This says to prohibit the creation of core files, restrict the number of processes to 50, and restrict memory usage per user to 5M.
/var/log/wtmp
and /var/run/utmp
files contain the login records
for all users on your system. Their integrity must be maintained
because they can be used to determine when and from where a user (or
potential intruder) has entered your system. These files should
also have 644
permissions, without affecting normal system
operation.
/etc/passwd
or /etc/shadow
).
See the chattr
(1) man page for information on the immutable bit.
Find all SUID/SGID programs on your system, and keep track of what they are, so you are aware of any changes which could indicate a potential intruder. Use the following command to find all SUID/SGID programs on your system:
root# find / -type f \( -perm -04000 -o -perm -02000 \)
The Debian distribution runs a job each night to determine what SUID
files exist. It then compares this to the previous night's run. You can
look in /var/log/setuid*
for this log.
You can remove the SUID or SGID permissions on a
suspicious program with chmod
, then restore them back if you
absolutely feel it is necessary.
root# find / -perm -2 ! -type l -ls
and be sure you know why those files are writable. In the normal
course of operation, several files will be world-writable, including some
from /dev
, and symbolic links, thus the ! -type l
which excludes these from the previous find
command.
Unowned files may also be an indication an intruder has accessed your system. You can locate files on your system that have no owner, or belong to no group with the command:
root# find / -nouser -o -nogroup -print
.rhosts
files should be a part of your regular system
administration duties, as these files should not be permitted on your
system. Remember, a cracker only needs one insecure account to
potentially gain access to your entire network. You can locate all
.rhosts
files on your system with the following command:
root# find /home -name .rhosts -print
Finally, before changing permissions on any system files, make sure you understand what you are doing. Never change permissions on a file because it seems like the easy way to get things working. Always determine why the file has that permission before changing it.
The umask
command can be used to determine the default file creation
mode on your system. It is the octal complement of the desired file
mode. If files are created without any regard to their permissions
settings, the user could inadvertently give read or write permission
to someone that should not have this permission. Typical umask
settings include 022
, 027
, and 077
(which is the most
restrictive). Normally the umask is set in /etc/profile
, so it applies
to all users on the system. The file creation mask can be calculated
by subtracting the desired value from 777. In other words, a umask of
777 would cause newly-created files to contain no read, write or execute
permission for anyone. A mask of 666 would cause newly-created files
to have a mask of 111. For example, you may have a line that
looks like this:
# Set the user's default umask
umask 033
Be sure to make root's umask 077
, which will disable read, write, and
execute permission for other users, unless explicitly changed using
chmod
. In this case, newly-created directories would have 744
permissions, obtained by subtracting 033 from 777. Newly-created files
using the 033 umask would have permissions of 644.
If you are using Red Hat, and adhere to their user and group ID
creation scheme (User Private Groups), it is only necessary to use 002
for a umask
. This is due to the fact that the default configuration
is one user per group.
It's important to ensure that your system files are not open for casual editing by users and groups who shouldn't be doing such system maintenance.
Unix separates access control on files and directories according to three characteristics: owner, group, and other. There is always exactly one owner, any number of members of the group, and everyone else.
A quick explanation of Unix permissions:
Ownership - Which user(s) and group(s) retain(s) control of the permission settings of the node and parent of the node
Permissions - Bits capable of being set or reset to allow certain types of access to it. Permissions for directories may have a different meaning than the same set of permissions on files.
Read:
Write:
Execute:
The "sticky bit" also has a different meaning when
applied to directories than when applied to files. If the sticky bit is set on a directory, then
a user may only delete files that the he owns or for which he has
explicit write permission granted, even when he has write access to
the directory. This is designed for directories like /tmp
, which are
world-writable, but where it may not be desirable to allow any user to
delete files at will. The sticky bit is seen as a t
in a long
directory listing.
This describes set-user-id permissions on the file. When the set user ID access mode is set in the owner permissions, and the file is executable, processes which run it are granted access to system resources based on user who owns the file, as opposed to the user who created the process. This is the cause of many "buffer overflow" exploits.
If set in the group permissions, this bit controls the "set group id" status of a file. This behaves the same way as SUID, except the group is affected instead. The file must be executable for this to have any effect.
If you set the SGID bit on a directory (with chmod g+s directory
),
files created in that directory will have their group set to the
directory's group.
You - The owner of the file
Group - The group you belong to
Everyone - Anyone on the system that is not the owner or a member of the group
File Example:
-rw-r--r-- 1 kevin users 114 Aug 28 1997 .zlogin
1st bit - directory? (no)
2nd bit - read by owner? (yes, by kevin)
3rd bit - write by owner? (yes, by kevin)
4th bit - execute by owner? (no)
5th bit - read by group? (yes, by users)
6th bit - write by group? (no)
7th bit - execute by group? (no)
8th bit - read by everyone? (yes, by everyone)
9th bit - write by everyone? (no)
10th bit - execute by everyone? (no)
The following lines are examples of the minimum sets of permissions that are required to perform the access described. You may want to give more permission than what's listed here, but this should describe what these minimum permissions on files do:
-r-------- Allow read access to the file by owner
--w------- Allows the owner to modify or delete the file
(Note that anyone with write permission to the directory
the file is in can overwrite it and thus delete it)
---x------ The owner can execute this program, but not shell scripts,
which still need read permission
---s------ Will execute with effective User ID = to owner
--------s- Will execute with effective Group ID = to group
-rw------T No update of "last modified time". Usually used for swap
files
---t------ No effect. (formerly sticky bit)
Directory Example:
drwxr-xr-x 3 kevin users 512 Sep 19 13:47 .public_html/
1st bit - directory? (yes, it contains many files)
2nd bit - read by owner? (yes, by kevin)
3rd bit - write by owner? (yes, by kevin)
4th bit - execute by owner? (yes, by kevin)
5th bit - read by group? (yes, by users
6th bit - write by group? (no)
7th bit - execute by group? (yes, by users)
8th bit - read by everyone? (yes, by everyone)
9th bit - write by everyone? (no)
10th bit - execute by everyone? (yes, by everyone)
The following lines are examples of the minimum sets of permissions that are required to perform the access described. You may want to give more permission than what's listed, but this should describe what these minimum permissions on directories do:
dr-------- The contents can be listed, but file attributes can't be read
d--x------ The directory can be entered, and used in full execution
paths
dr-x------ File attributes can be read by owner
d-wx------ Files can be created/deleted, even if the directory
isn't the current one
d------x-t Prevents files from deletion by others with write
access. Used on /tmp
d---s--s-- No effect
System configuration files (usually in /etc
) are usually mode 640
(-rw-r-----
), and owned by root. Depending on your site's security
requirements, you might adjust this. Never leave any system files
writable by a group or everyone. Some configuration files, including
/etc/shadow
, should only be readable by root, and directories in /etc
should at least not be accessible by others.
SUID shell scripts are a serious security risk, and for this reason the kernel will not honor them. Regardless of how secure you think the shell script is, it can be exploited to give the cracker a root shell.
Another very good way to detect local (and also network) attacks on
your system is to run an integrity checker like Tripwire
,
Aide
or Osiris
.
These integrety checkers run a number of checksums on all your important
binaries and config files and compares them against a database of former,
known-good values as a reference. Thus, any changes in the files will
be flagged.
It's a good idea to install these sorts of programs onto a floppy, and then physically set the write protect on the floppy. This way intruders can't tamper with the integrety checker itself or change the database. Once you have something like this setup, it's a good idea to run it as part of your normal security administration duties to see if anything has changed.
You can even add a crontab
entry to run the checker from your floppy
every night and mail you the results in the morning. Something like:
# set mailto
MAILTO=kevin
# run Tripwire
15 05 * * * root /usr/local/adm/tcheck/tripwire
will mail you a report each morning at 5:15am.
Integrety checkers can be a godsend to detecting intruders before you would otherwise notice them. Since a lot of files change on the average system, you have to be careful what is cracker activity and what is your own doing.
You can find the open sourced version of Tripwire
at
http://www.tripwire.org,
free of charge. Manuals and support can be purchased.
Aide
can be found at
http://www.cs.tut.fi/~rammer/aide.html.
Osiris
can be found at
http://www.shmoo.com/osiris/.
"Trojan Horses" are named after the fabled ploy in Homer's "Iliad". The idea is that a cracker distributes a program or binary that sounds great, and encourages other people to download it and run it as root. Then the program can compromise their system while they are not paying attention. While they think the binary they just pulled down does one thing (and it might very well), it also compromises their security.
You should take care of what programs you install on your machine. Redhat provides MD5 checksums and PGP signatures on it's RPM files so you can verify you are installing the real thing. Other distributions have similar methods. You should never run any unfamiliar binary, for which you don't have the source, as root! Few attackers are willing to release source code to public scrutiny.
Although it can be complex, make sure you are getting the source for a program from its real distribution site. If the program is going to run as root, make sure either you or someone you trust has looked over the source and verified it.