Dynamic firewall with FirewallD
FirewallD is a service daemon with a D-BUS interface that provides a dynamic managed firewall.
Why A Firewall Daemon
The current firewall model is static and every change requires a complete firewall restart. This includes also to unload the firewall netfilter kernel modules and to load the modules that are needed for the new configuration. The unload of the modules is breaking stateful firewalling and established connections.
The firewall daemon on the other hand manages the firewall dynamically and applies changes without restarting the whole firewall. Therefore there is no need to reload all firewall kernel modules. But using a firewall daemon requires that all firewall modifications are done with that daemon to make sure that the state in the daemon and the firewall in kernel are in sync. The firewall daemon can not parse firewall rules added by the ip*tables and ebtables command line tools.
The daemon provides information about the current active firewall settings via D-BUS and also accepts changes via D-BUS using PolicyKit authentication methods. SELinux access restrictions are also planned.
Applications, daemons and the user can request to enable a firewall feature over D-BUS. A feature could either be one of the predefined firewall features like services, port and protocol combinations, trusted interfaces/hosts/network areas, port/packet forwarding, masquerading, icmp blocking or even a custom rule. The feature can be enabled for a certain amount of time or can be disabled by again.
New chains for virtualization, network settings, services, ports, masquerading, port forwarding, icmp filtering and virtualization are added to make the firewall setup more flexible, safe and robust. Adding a rule with the firewall daemon to one of these chains will most likely not interefere with rules of other chains. The order of the chains and how they are used is fixed.
The netfilter firewall helpers, that are for example used for amanda, ftp, samba and tftp services, are also handled by the daemon as long as they are part of a predefined service. Loading of additional helpers is not part of the current interface. For some of the helpers onloading is only possible after all connections that are handled by the module are closed. Therefore connection tracking information is important here and needs to get into account.
Static Firewall (system-config-firewall)
The actual static firewall model with system-config-firewall will still be available and usable, but not at the same time as the daemon is running. The user or admin can decide which firewall solution should be used.
At install time, firstboot or first network usage, there will be a selector for the firewall solution to use. If the firewall daemon will be used, it will disable the ip*tables services and also add the disabled option for system-config-firewall. The configuration will stay intact and s-c-fw can be enabled simply with "lokkit --enabled" again.
The firewall daemon is independent to system-config-firewall, but should not be used at the same time.
Planned and Proposed Features
The D-BUS interface gives information about the firewall state and makes it possible to enable, disable and query firewall settings.
Firewall Abstraction Model
Adding an abstraction layer on top of ip*tables and ebtables firewall rules makes adding rules simple and more intuitive. The abstraction layer needs to be powerful, but also simple, which makes this not an easy task. A firewall language has to gen invented for this. Firewall rules have a fixed position and querying generic information about access state, access policies for ports and other firewall features is possible.
System and Session (User) Configuration
The firewall system configuration is the base configuration for the whole system and can be altered with the session (user) configuration. The session configuration can be limited by policies and the system administrator.
Permanent and Temporary Configuration
Permanent firewall settings are settings, that still exist after a reboot and service restart. Temporary settings will be lost after reboot and service restart. A firewall service reload will restore all settings, permanent and temporary.
The try applet firewall-applet visualizes the firewall state and also problems with the firewall for the user. It can also be used to configure settings the user is allowed to change.
Graphical Configuration Tool
The configuration tool firewall-config is the main configuration tool for the firewall daemon. It supports all features of the firewall, but firewall settings of libvirt and also NetworkManager are readonly.
It also makes it possible for the administrator to change system and user policies.
Command Line client
firewall-cmd provides (most of) the configuration features of the graphical tool for the command line.
Support for ebtables
ebtables support is needed to fullfill all needs of the libvirt daemon and to prevent access problems between ip*tables and ebtables on kernel netfilter level. All these commands are accessing the same structures and therefore they should not be used at the same time.
Support for conntrack
Conntrack is needed to be able to terminate established connections for features that get disabled. For some use cases it might not be good to terminate the connection: Enabling of a firewall service for a limited time to establish a persistent external connection.
User interaction mode
This is a special mode of in the firewall the user or admin can enable. All requests of applications to alter the firewall are directed to the user to get notified and granted or denied. It is possible to set a time limit for the acceptance of a connection and to limit it to hosts, networks or connections. It can be saved to behave the same in the future without notification.
An additional feature of this mode is direct external connection attempts on preselected services or ports to the user with the same features as the application initiated requests. The limitation on services and ports will also limit the amount of requests sent to the user.
Network zones: Network security model
Network zones define the level of trust for network connections. A public WIFI network connection for example should be untrusted, a wired home network connection should be fairly trusted.
The network security model can be selected initially at installation time, firstboot or when the first network connection gets established. The model describes the trust level of the whole network environment, the host is connected to, and also defines what to do with new connections.
There are different initial zones:
- Home / Work - Public - Connection specific
The home or work zone has the highest trust level. All incoming connections are allowed. The public zone on the other hand is fully untrusted. No incoming connection is allowed. The connection specific model requires that the user tunes the trust level of a connection according to the needs. The default is untrusted.
The user or administrator is able to define new zones or adapt initial zones to change the behaviour according to the needs.
The network security model makes it possible to have one trust level for all connections or to have several connections with different trust levels.
User policy support
The administrator can define which users are able to use the User Interaction Mode and can also limit the firewall features, that can be used with it.
Port metadata information (proposed by Lennart Poettering)
To have a port independent metadata information would be good to have. The current model with a static assignment of ports and protocols from /etc/services is not a good solution and is not reflecting current use cases. Ports in applications or services are dynamic and therefore the port itself does not describe the use case.
This metadata information could be used to form simple rules for the firewall. Here are some examples:
allow external access to file sharing applications or services allow external access to music sharing applications or services allow external access to all sharing applications or services allow external access to torrent file sharing applications or services allow external access to http web services
The metadata information here could not only be application specific, but also a group of use cases. For example the "all sharing" group or the "file sharing" group could match all sharing or file sharing applications, for example torrent file sharing. These are examples, therefore it might be that they are not useful.
There are two possible solutions to get metadata information in the firewall:
The first is to add it to netfilter (kernel space). This has the advantage, that it can be used by everyone, but also limits the use. To get user or system specific information into account, all these need to be implemented in kernel space also.
The other one would be to add this to a firewall daemon. These abtract rules could be used together with informations like the trust level of the network connections, the user decision to share with as specific person/host or the hard rule of the administrator to forbid sharing completely.
The second solution would have the advantage that new metadata groups or changes in incorporation of trust levels, user preferences or administrator rules would not require to push a new kernel. Adding these kind of abtract rules to a firewall daemon would make it much more flexible. Even new security levels woule be easy to add without kernel updates.
Netfilter firewalls are always susceptible to rule ordering issues, because a rule does not have a fixed position in a chain. The position can change if other rules are added or removed in a position before that rule.
In the static firewall model a firewall change is recreating a clean and sane firewall setup limited to the features directly supported by system-config-firewall / lokkit. Firewall rules created by other applications are not integrated and s-c-fw / lokkit does not know about them if the customs rules file feature is not in use. Default chains are used and there is no safe way to add and remove rules without interfering with others.
The dynamic model has additional chains for the firewall features. These specific chains are called in a defined ordering and rules added to a chain could not interfere with reject or drop rules in chains that were called before. This makes it possible to have a more sane firewall configuration.
Here are example rules created by the daemon in the filter table with ssh and ipp-client enabled inthe public zone:
*filter :INPUT ACCEPT [0:0] :FORWARD ACCEPT [0:0] :OUTPUT ACCEPT [0:0] :FORWARD_ZONES - [0:0] :FORWARD_direct - [0:0] :INPUT_ZONES - [0:0] :INPUT_direct - [0:0] :IN_ZONE_public - [0:0] :IN_ZONE_public_allow - [0:0] :IN_ZONE_public_deny - [0:0] :OUTPUT_direct - [0:0] -A INPUT -m conntrack --ctstate INVALID -j REJECT --reject-with icmp-host-prohibited -A INPUT -m conntrack --ctstate RELATED,ESTABLISHED -j ACCEPT -A INPUT -i lo -j ACCEPT -A INPUT -j INPUT_direct -A INPUT -j INPUT_ZONES -A INPUT -p icmp -j ACCEPT -A INPUT -j REJECT --reject-with icmp-host-prohibited -A FORWARD -m conntrack --ctstate INVALID -j REJECT --reject-with icmp-host-prohibited -A FORWARD -m conntrack --ctstate RELATED,ESTABLISHED -j ACCEPT -A FORWARD -i lo -j ACCEPT -A FORWARD -j FORWARD_direct -A FORWARD -j FORWARD_ZONES -A FORWARD -p icmp -j ACCEPT -A FORWARD -j REJECT --reject-with icmp-host-prohibited -A OUTPUT -j OUTPUT_direct -A IN_ZONE_public -j IN_ZONE_public_deny -A IN_ZONE_public -j IN_ZONE_public_allow -A IN_ZONE_public -j REJECT --reject-with icmp-host-prohibited -A IN_ZONE_public_allow -p tcp -m tcp --dport 22 -j ACCEPT -A IN_ZONE_public_allow -p tcp -m tcp --dport 631 -j ACCEPT
Used is a deny/allow model to have a clear behaviour and at best no rule interferences. Icmp blocks for example will go to the IN_ZONE_public_deny chain if set for the public zone and will be handled before the rules in the IN_ZONE_public_allow chain.
This model makes it more easy to add or remove rules from a specifig block without interfering with accept or drop rules from another block.
At the moment there are sysctl settings that are not properly applied. This happens if the module providing the setting is not loaded at boot time when rc.sysinit runs or it the module gets reloaded at runtime. Another example is net.ipv4.ip_forward, which is needed for example for specific firewall settings, libvirt and also user/admin changes. If there are two apps or daemons enabling ip_forwarding only if needed, then it could happen that one of them is turning it off again without knowing that there is anotherone, that still needs it tuned on.
The sysctl daemon could solve this by having an internal use count for settings, that will make it possible to turn it off or go to the previous setting again if the requester reverted the request to change it.
These phases are planned and might change over time according to requests and needs.
Phase 1: Fedora 15
Fedora 15 will contain the proof of concept implementation with some enhancements as a preview and test version. It will not be installed and enabled by default.
The list below contains the actual state in the GIT repo (2010-12-22):
- Nearly all system-config-firewall features are supported. Custom rule files are not supported, but simple custom rules with the custom interface
- iptables, ip6tables and ebtables are supported
- Simple tray applet showing the firewall state, can also enable and disable firewall services
- Command line client
- Tray applet and command line client can not handle libvirt and NetworkManager rules
- No permanent firewall features (rules will be lost with service restart and reboot, but service reload keeps rules)
- System, but no session (user) configuration
- The firewall-config utility is not functional
- sysctl changes for ip_forwarding are not implemented
- Access conflicts of ip*tables and ebtables rules are not solved
- firewalld check in system-config-firewall version 1.2.28
Phase 2: Fedora 17 (planned)
- D-BUS interface cleanup and extensions
- Finalize firewall-applet and firewall-config
- Permanent and temporary firewall rules
- Zone support
Phase 3: Fedora 18 (planned)
- User Interaction Mode
- User Policy Support
- Enhanced interface for services like libvirt and also applications
- Maybe: Reimplementation of the server in C with system and session configuration
Phase 4: Fedora 19 (planned)
- Port Metadata Information (requires bigger system changes)