Unix security

A core security feature in these systems is the file system permissions. All files in a typical Unix filesystem have permissions set enabling different access to a file.

Permissions on a file are commonly set using the chmod command and seen through the ls command. For example:

-r-xr-xr-x  1 root  wheel  745720 Sep  8  2002 /bin/sh

Unix permissions permit different users access to a file. Different user groups have different permissions on a file.

More advanced Unix filesystems include the Access Control List concept which allows permissions to be granted to multiple users or groups. An Access Control List may be used to grant permission to additional individual users or groups. For example:

/pvr [u::rwx,g::r-x,o::r-x/u::rwx,u:sue:rwx,g::r-x,m::rwx,o::r-x]

In this example, which is from the chacl command on the Linux operating system, the user sue is granted write permission to the /pvr directory.

Users under Unix style operating systems often belong to managed groups with specific access permissions. This enables users to be grouped by the level of access they have to this system. Many Unix implementations add an additional layer of security by requiring that a user be a member of the wheel user privileges group in order to access the su command.[1]

Sudo command on Ubuntu to temporarily assume root privileges

Most Unix and Unix-like systems have an account or group which enables a user to exact complete control over the system, often known as a root account. If access to this account is gained by an unwanted user, this results in a complete breach of the system. A root account however is necessary for administrative purposes, and for the above security reasons the root account is seldom used for day to day purposes (the sudo program is more commonly used), so usage of the root account can be more closely monitored.

Root access "as it should be" can be visualised by those familiar with the Superman stories using the following analogy:

Using a root account is rather like being Superman; an administrator's regular user is more like Clark Kent. Clark Kent becomes Superman for only as long as necessary, in order to save people. He then reverts to his "disguise". Root access should be used in the same fashion. The Clark Kent disguise doesn't really restrict him though, as he is still able to use his super powers. This is analogous to using the sudo program.

Selecting a strong password and guarding it properly are probably the most important things a user can do to improve Unix security. In Unix systems, the essential information about users is stored under the file /etc/passwdThis file keeps track of the users registered in the system and their main definitions. Passwords, or more correctly, the hash of the password, can also be stored in the same place. The entries in /etc/passwd occupy exactly one line each, and have the following form:

nickname:password_hash:UserID:GroupID:Complete_Name:home_dir:shell_bin

An example would be:

xfze:$1zuW2nX3sslp3qJm9MYDdglEApAc36r/:1000:100:Daniel Ernesto Ortiz Costa:/home/xfze:/bin/bash

Since all users must have read access to the /etc/passwd file to do many common tasks (ls -l /home will use /etc/passwd to map UIDs to login names for example), anyone could also read the password hashes of other users. To solve this problem, the file /etc/shadow was created to store the password hashes, with only root having read access. Under password shadowing, the 2nd field (password hash) is replaced by an 'x' which tells the system to retrieve the corresponding user's password via the /etc/shadow file.

The /etc/shadow file often only contains values for the first two fields:

xfze:$1zuW2nX3sslp3qJm9MYDdglEApAc36r/:::::

The remaining fields in the /etc/shadow file include:

1. The minimum number of days between password changes
2. The maximum number of days until the password must be changed
3. The number of days of warning given before the password must be changed
4. The number of days after the password must be changed when the account becomes unusable
5. The date (expressed as the number of days since January 1st, 1970) when the account is expired

These fields may be used to improve Unix security by enforcing a password security policy.

Administrators should delete old accounts promptly.

• su, sudo, ssh only, no remote root logins

Operating systems, like all software, may contain bugs in need of fixing or may be enhanced with the addition of new features. Patching the operating system in a secure manner requires that the software come from a trustworthy source and not have been altered since it was packaged. Common methods for verifying that operating system patches have not been altered include the use of the digital signature of a cryptographic hash, such as a SHA-256 based checksum, or the use of read-only media.

From a security standpoint, the specific packaging method, such as the dpkg package manager or the RPM Package Manager format originally from Debian and Red Hat Linux, respectively is not as important as the use of features which ensure the integrity of the patch itself.

Source distributions include the ability to examine the code for suspicious content. The drawback, absent an accompanying cryptographic hash value, is that the user must be able to perform a security analysis of the code themselves.

Linux distributions which use the RPM Package Manager format for providing base functionality and software updates make use of cryptographic hashes and GPG to ensure content integrity. The hash values are packaged with the RPM file and verified when the package is installed.

Linux distributions which use the Debian .deb package format for providing base functionality and software updates make use of GPG signatures to ensure content integrity. A signature is computed when the package is constructed and verified later when the package is installed.

Regardless of the vendor or distribution, all software distributions should provide a mechanism for verifying that the software is legitimate and has not been modified since it was originally packaged.

File system security within UNIX and Unix-like systems is based on 9 permission bits, set user and group ID bits, and the sticky bit, for a total of 12 bits. These permissions apply almost equally to all filesystem objects such as files, directories and devices.

The 9 permission bits are divided into three groups of three bits each. The first group describes the permissions of the file owner, the second group describes the permissions of the GID assigned to the file, which by default is the group associated with the file owner or the directory containing the file when it is set-GID, and the third group describes the permissions associated with any process which does not have the same user ID as the file. Each group of three bits contains a bit indicating the read, write or execute access is granted. In the case of directories, execute access is interpreted as the permission to perform a filename lookup within the directory.

The set user ID and set group ID bits, commonly abbreviated set-UID and set-GID respectively, are used to change the identity of the process which executes a file having either or both of those bits set. A file having the set-UID permission bit set will cause a process which executes that file to temporarily switch the effective user ID to that of the file owner. A file having the set-GID permission bit set will cause a process which executes that file to temporarily switch the effective group ID to that of the file group. A process may then alternate between the effective user or group ID which it inherited from the file and the real user or group ID which it inherited when the user logged on to the system. This provides a mechanism by which a process may limit the access rights it possesses to those code regions which require those access rights. This is a form of a security technique known as privilege separation and improves program security by limiting the unintended or undesirable actions of a processes.

A directory having the set-GID permission bit set will cause a newly created file to have an initial file group value equal to the file group of the directory. This provides a mechanism whereby a subsystem, such as the system's mail subsystem, can create files which have a common file group value so that set-GID processes within that subsystem are then able to read or write the file.

The sticky bit, formally known as the save text on swap bit, derives its name from its original purpose. Originally the sticky bit caused a process's initial memory image to be stored as a contiguous image on the disk drive which was used to store real memory pages when they were not in use. This improved the performance of commonly executed commands by making the initial memory image readily available. Modern UNIX systems no longer perform that function when the bit is set, but the name has been preserved nonetheless. In the case of files, the sticky-bit may be used by the system to indicate the style of file locking to be performed. In the case of directories, the sticky bit prevents any process, other than one which has super-user privileges or one having an effective user ID of the file owner, from deleting a file within that directory. The sticky bit is most commonly used on publicly writable directories, such as the various temporary working space directories on the system.

Root squash[2][3] is a special mapping of the remote superuser (root) identity when using identity authentication (local user is the same as remote user). Under root squash, a client's uid 0 (root) is mapped to 65534 (nobody). It is primarily a feature of NFS but may be available on other systems as well.

Root squash is a technique to avoid privilege escalation on the client machine via suid executables Setuid. Without root squash, an attacker can generate suid binaries on the server that are executed as root on other client, even if the client user does not have superuser privileges. Hence it protects client machines against other malicious clients. It does not protect clients against a malicious server (where root can generate suid binaries), nor does it protect the files of any user other than root (as malicious clients can impersonate any user).

iptables is the current user interface for interacting with Linux kernel netfilter functionality. It replaced ipchains. Other Unix like operating systems may provide their own native functionality and other open source firewall products exist. More detailed information about iptables is contained elsewhere. A brief discussion is contained here in order to describe how iptables may be used to configure a Linux firewall.

netfilter provides a state-full packet filter which can be configured according to network interface, protocol, source and/or destination address, source and/or destination port and the state of the packet. A network packet traverses several chains between the time it is received by a network interface and the time it is accepted by the host or forwarded to another host. The common chains are INPUT, OUTPUT and FORWARD. The INPUT chain is traversed for all packets as they are received by a network interface, regardless of whether they are to be accepted by the host or forwarded to another host. The OUTPUT chain is traversed for all packets as they are transmitted by a network interface. The FORWARD chain is traversed for those packets are being routed through the host from one network interface to another, such as is the case for a multi-homed system (a system with more than one physical network interface).

Each of the built-in chains has a default policy which defines what action is taken for a packet which reaches the end of the chain. Packet traversal ends when a rule matches the packet and has an action of ACCEPT, DROP, REJECT or RETURN.

The simplest iptables firewall consists of rules for each desired service, followed by a rule which indicates that any packets which reach this rule are dropped. A system which only permitted, for example, incoming email traffic would have a rule which accepted connections on the SMTP port, and then dropped others. A rule would be required which indicated that all established connections were also permitted so that outgoing connections would receive responses from other systems.

The following example shows a simple packet filter for the INPUT chain for the above described example:

Chain INPUT (policy DROP 0 packets, 0 bytes)
 pkts bytes target     prot opt   in        out                 source               destination
    0     0 ACCEPT     all—any    any     anywhere             anywhere            state ESTABLISHED
    0     0 ACCEPT     tcp—any    any     anywhere             anywhere            tcp dpt:smtp
    0     0 LOG        all—any    any     anywhere             anywhere            LOG level warning
    0     0 DROP       all—any    any     anywhere             anywhere

The addition of an explicit DROP action ensures that the packets are discarded should the default policy of the INPUT chain accidentally be changed to ACCEPT.

There is less need for an OUTPUT chain and the default policy of the OUTPUT chain can safely be set to ACCEPT. In some instances it may be desirable for a firewall to limit certain outgoing connections to a certain set of approved systems. This is known as egress filtering and may be used to prevent viruses within the firewall from escaping to other systems. For example, it may be the policy of a network to limit outgoing email connections to a single authorized email servers as a way of combating e-mail spam. This could be achieved by the following example:

Chain OUTPUT (policy ACCEPT)
 pkts bytes target     prot opt    in     out                  source              destination
    0     0 DROP       tcp—any    any    !server               anywhere            tcp dpt:smtp

There is no need to include any other rules in this example as the default policy for the OUTPUT chain is ACCEPT. This rule assumes that the host which is acting as the firewall will not be sending email itself, such as to the email server. This is a good assumption as typically a firewall system contains the minimal amount of system code needed to act as a firewall.

A more restrictive OUTPUT chain would contain permissive (ACCEPT) entries for those services which may be accessed outside the firewall and then a restrictive (DROP) policy for the chain itself.

• Practical UNIX and Internet Security, Simson Garfinkel and Gene Spafford, O'Reilly & Associates, 2003.