During configuration, trade-offs are made between security, ease-of-use, and legacy compatibility. There are a wide variety of options covering network and protocol support, authentication, and user environment, that obscure the individual option's impact to the whole. This section includes some recommendations, along with the consequences of their usage.
Only the Solaris Secure Shell software (Solaris 9 OE) and OpenSSH (3.5p1) versions that were current at the time this article was written were used. Not all of the options are covered. Consult the vendor documentation for more information on the other options and on the options presented here.
Server configuration concerns how the daemon presents itself on the network, what protocols are offered, and what authentication methods are allowed. Specific recommendations are given for each topic. Recommendations specific to a particular Secure Shell implementation have been noted.
Two major versions of the Secure Shell protocol exist. Protocol 1 has been deprecated due to vulnerabilities, such as packet insertion and password length determination. Whenever possible, use Protocol 2. Unfortunately, many legacy clients support only Protocol 1. If this protocol must be enabled, consult the Legacy Support recommendations later in this article. Consider migrating to clients that support Protocol 2, as soon as reasonably possible.
By default, the sshd(1M) daemon listens on all network interfaces on its bound ports. For workstations or other systems where accessibility is desired on all interfaces, this is not a problem. For architectures such as the Service Delivery Network, in which management traffic is limited to a particular interface, it is a problem. Limit network access with the ListenAddress keyword, where access is limited by a particular IP address, not a network interface.
# Listen only to the management network. ListenAddress 192.168.0.10
To further narrow down what the daemon will listen to, use either a host-based firewall, such as the SunScreen™ software, or TCP Wrappers.
For a reference on traffic limited architectures, consult the Sun BluePrints OnLine article "Building Secure N-Tier Environments" (October 2000).
Occasionally, connections are temporarily suspended when a route is downed, a machine crashes, a connection is hijacked, or a man-in-the-middle attack is attempted. TCP keep alives should be sent to detect any of these cases. The server will disconnect the connection if TCP keep alives fail and return allocated resources. Regular disconnects can aggravate users on faulty networks.
Optionally, you can use compression on the encrypted data streams. This results in bandwidth savings for compressible data, such as interactive logins or log files, at the expense of more CPU resources. For uncompressible data, such as encrypted or compressed files, the extra CPU time is wasted and decreases performance. For singular Secure Shell sessions, this is not a problem. For a file server, the extra load could impact performance. In this case, turn compression off to prevent misconfigured clients from driving up the system load.
# Transfering ASCII data such as interactive logins or log files Compression yes # Transfering random data such as compressed or encrypted files # Prevents performance issues and reduces CPU load Compression no
Privilege separation is a feature only in OpenSSH. The sshd(1M) daemon is split into two parts: a privileged process to deal with authentication and process creation and an unprivileged process to deal with incoming network connections. After successful authentication, the privileged process spawns a new process with the privileges of the authenticated user. The goal is to prevent compromise from an error in the network facing process. Unfortunately, privilege separation is not entirely compatible with pluggable authentication modules (PAM), SunShield™ Basic Security Module (BSM) auditing. Also, some OpenSSH features are disabled. If this functionality is desired, consult the vendor documentation.
# OpenSSH only UsePrivilegeSeparation no
Login Grace Time
The default login grace time is the time a connection can exist before successfully authenticating. The default of 600 seconds for the Solaris Secure Shell software and 120 seconds for later OpenSSH versions is too long. Reduce the time to sixty seconds.
Password and Public Key Authentication
Passwords are not always appropriate. The stronger identity method might be required. An identity is a public-key cryptographic, two-factor authentication system. When passwords are deemed sufficient, do not allow empty passwords. They are trivial to guess.
PasswordAuthentication yes PermitEmptyPasswords no PubKeyAuthentication yes DSAAuthentication yes
Superuser (root) Logins
Neither the Solaris Secure Shell software nor OpenSSH honor the values set in the /etc/default/login file. To prevent network superuser (root) logins, this must be explicitly denied. By default, the Solaris Secure Shell software denies superuser logins; while, OpenSSH allows them. This forces system administrators to log in as unprivileged users, then change users (su) to the superuser. Only if the system has no user accounts and the appropriate host protection is in place should you enable superuser logins.
Banners, Mail, and Message-of-the-Day
Some sites require that a banner be displayed after a user connects to a system, but before logging in. You can accomplish this with the Banner keyword. Set Banner to /etc/issue so that only one banner file exists for the entire system.
In the Solaris OE, the interactive login shell is expected to display the message-of-the-day (MOTD) and to check for new mail. Using some versions of OpenSSH, this causes the MOTD display and mail check to be done twice. Set these keywords to no to eliminate the duplication.
CheckMail no PrintMotd no
Connection and X11 Forwarding
Secure Shell can tunnel TCP and X connections through encrypted connections established between the client and server. Tunneling the traffic is referred to as forwarding. The forwarding occurs at the application level and is not completely transparent to the applications being forwarded. The applications need some configuration to use the tunnel.
Data is protected only while it is in the tunnel between the client and server. After that, it is normal network traffic in the clear.
Tunneled traffic bypasses firewalls and intrusion detection systems. Allowing connection (TCP port) forwarding enables remote users safer access to email or the corporate Web server. X forwarding allows system administrators to run GUI applications remotely, such as the Sun™ Management Center software. This might not be functionality you want your users setting up. You can inconvenience users by turning off the functionality, but as long as they have shell access, they can run their own forwarders. Use role-based access control (RBAC) to explicitly limit what you want your users to do in this case.
If port forwarding is enabled, disable GatewayPorts and educate your users. GatewayPorts allows machines, other than the client, to access the forwarded ports in the tunnel. This effectively bypasses any firewall usage. Again, users could run their own private forwarders on their client machines to defeat the server restrictions. Consider placing an intrusion detection sensor on the private network side of a Secure Shell bastion host to detect problem traffic.
AllowTCPForwarding yes GatewayPorts no X11DisplayOffset 10 X11Forwarding yes XAuthLocation /usr/X/bin/xauth
User Access Control Lists
User access control lists (ACL) can be specified in the server configuration file. You can either specifically allow or deny individual users or groups. No other part of the Solaris OE honors this ACL. The default is to allow access to anyone with a valid account. This can be a method to limit particular user's access in NIS environments, without resorting to custom pluggable authentication modules. Use only one of the following four ACL keywords (AllowGroups, AllowUsers, DenyGroups, or DenyUsers) in the server configuration file.
# Allow only the sysadmin staff AllowGroups staff # Or prevent unauthorized users. DenyUsers Cheng Atkinson
User File Permissions
If users have left their home directories or .ssh files world writable by accident or trickery, an intruder could insert identities allowing password free access or alter the known_hosts file, allowing man-in-the-middle attacks. Enabling StrictModes, the sshd(1M) daemon will not allow a login. When on, users can be easily confused because they will not know why they cannot log in. No different error message is presented to them. For sites that will disable StrictModes, consider eliminating public-key-based authentication to prevent user account compromise. The consequence is the elimination of password-free logins for users or automated jobs.
For OpenSSH only, UseLogin specifies that the OpenSSH sshd(1M) daemon call login(1) instead of performing the initial login tasks itself for interactive sessions. This feature is used to work around OpenSSH's lack of support for SunShield BSM auditing. Enabling this feature should allow interactive users to be audited and prevent corruption of the user's cron audit file. This will not affect non-interactive (remote job execution) usage auditing, which will still not work. This feature disables X11 and port forwarding and is not compatible with privilege separation.
If legacy clients must be supported, strengthen the default configuration as much as possible. Default to Protocol 2 for the clients that support it. Disable all of the various rhosts style authentication methods. Increase the server key size and decrease the ephemeral key regeneration interval to minimize offline factoring attacks against the keys.
# Enable protocol 1 but default to protocol 2. Protocol 2,1 # Legacy support options - protocol 1 only HostKey /etc/ssh/ssh_host_key IgnoreRhosts yes IgnoreUserKnownHosts yes KeyRegenerationInterval 1800 RhostsAuthentication no RhostsRSAAuthentication no RSAAuthentication yes ServerKeyBits 1024
Client configuration concerns protocol support, identity management, and host options assignment. Specific recommendations are given for each topic.
Host Option Assignment
Configuration options can be assigned to a specific host or to all hosts by using the Host keyword. The value is matched to what the user types on the command line, not the actual hostname of the server. An asterisk (*) is used to set global default options. Options assigned to a specific host have precedence over the global default options.
# Only for a specific host Host legacy Protocol 1 # For all hosts Host * Protocol 2
Data compression can be used on the encrypted data stream to save bandwidth. Set to off by default, you should enable it for interactive sessions or for transferring easily compressible data. The compression cost is asymmetric in that compressing the data is more computationally expensive than decompression. Client-side CPU cycles are generally cheaper than server-side CPU cycles. Avoid attempting to compress already compressed or encrypted data to avoid needlessly raising the CPU load on the server.
# For interactive sessions, low bandwidth links, # or easily compressable files Compression yes
Enable TCP keep alives to detect downed connections. See "Keep Alives" on page 5 for the server recommendations.
Always use Protocol 2 when possible. See "Protocol Support" on page 4 for the server recommendation.
rlogin and rsh
The rlogin and rsh protocols should not be used. Prevent the client from attempting to execute rsh in case a Secure Shell connection is refused.
FallBackToRsh no UseRsh no
Verify server identity both by its host key and IP address. For higher levels of identity assurance, set StrictHostKeyChecking to yes and distribute hosts keys out-of-band. This is impractical when users frequently encounter new hosts. Set StrictHostKeyChecking to ask, and train the users to verify the offered host key with the stored host key on the server.
CheckHostIP yes # only access one host StrictHostKeyChecking yes # access a variety of hosts StrictHostKeyChecking ask