DOKK / manpages / debian 12 / gvpe / gvpe.conf.5.en
GVPE.CONF(5) GNU Virtual Private Ethernet GVPE.CONF(5)

gvpe.conf - configuration file for the GNU VPE daemon

   # global options for all nodes
   udp-port = 407
   mtu = 1492
   ifname = vpn0
   # first node is named branch1 and is at 1.2.3.4
   node = branch1
   hostname = 1.2.3.4
   # second node uses dns to resolve the address
   node = branch2
   hostname = www.example.net
   udp-port = 500       # this host uses a different udp-port
   # third node has no fixed ip address
   node = branch3
   connect = ondemand

The gvpe config file consists of a series of lines that contain variable = value pairs. Empty lines are ignored. Comments start with a # and extend to the end of the line. They can be used on their own lines, or after any directives. Whitespace is allowed around the = sign or after values, but not within the variable names or values themselves.

All settings are applied "in order", that is, later settings of the same variable overwrite earlier ones.

The only exceptions to the above are the following directives:

Introduces a node section. The nodename is used to select the right configuration section and is the same string as is passed as an argument to the gvpe daemon.

Multiple node statements with the same node name are supported and will be merged together.

This statement switches back to the global section, which is mainly useful if you want to include a second config file, e..g for local customisations. To do that, simply include this at the very end of your config file: 

   global
   include local.conf
You can prefix any configuration directive with on and a nodename. GVPE will will only "execute" it on the named node, or (if the nodename starts with !) on all nodes except the named one.

Example: set the MTU to 1450 everywhere, loglevel to noise on branch1, and connect to ondemand everywhere but on branch2.

   mtu = 1450
   on branch1 loglevel = noise
   on !branch2 connect = ondemand
Reads the specified file (the path must not contain whitespace or = characters) and evaluate all config directives in it as if they were spelled out in place of the include directive.

The path is a printf format string, that is, you must escape any % by doubling it, and you can have a single %s inside, which will be replaced by the current nodename.

Relative paths are interpreted relative to the GVPE config directory.

Example: include the file local.conf in the config directory on every node.

   include local.conf

Example: include a file conf/nodename.conf

   include conf/%s.conf

Usually, a config file starts with a few global settings (like the UDP port to listen on), followed by node-specific sections that begin with a node = nickname line.

Every node that is part of the network must have a section that starts with node = nickname. The number and order of the nodes is important and must be the same on all nodes. It is not uncommon for node sections to be completely empty - if the default values are right.

Node-specific settings can be used at any time. If used before the first node section they will set the default values for all following nodes.

GLOBAL SETTINGS

Global settings will affect the behaviour of the running gvpe daemon, that is, they are in some sense node-specific (config files can set different values on different nodes using on), but will affect the behaviour of the gvpe daemon and all connections it creates.

Tells GVPE to chroot(2) to the specified path after reading all necessary files, binding to sockets and running the if-up script, but before running node-up or any other scripts.

The special path / instructs GVPE to create (and remove) an empty temporary directory to use as new root. This is most secure, but makes it impossible to use any scripts other than the if-up one.

These two options tell GVPE to change to the given user and/or group id after reading all necessary files, binding to sockets and running the if-up script.

Other scripts, such as node-up, are run with the new user id or group id.

Alternative to chuid and chgid: Sets both chuid and chgid to the user and (primary) group ids of the specified user (for example, nobody).
The DNS server to forward DNS requests to for the DNS tunnel protocol (default: 127.0.0.1, changing it is highly recommended).
The port where the dns-forw-host is to be contacted (default: 53, which is fine in most cases).
Sets whether the DNS transport forwarding server preserves case (DNS servers have to, but some access systems are even more broken than others) (default: true).

Normally, when the forwarding server changes the case of domain names then GVPE will automatically set this to false.

The maximum number of outstanding DNS transport requests (default: 100). GVPE will never issue more requests then the given limit without receiving replies. In heavily overloaded situations it might help to set this to a low number (e.g. 3 or even 1) to limit the number of parallel requests.

The default should be working OK for most links.

The DNS transport uses the minimum request latency (min_latency) seen during a connection as it's timing base. This factor (default: 0.5, must be > 0) is multiplied by min_latency to get the maximum sending rate (= minimum send interval), i.e. a factor of 1 means that a new request might be generated every min_latency seconds, which means on average there should only ever be one outstanding request. A factor of 0.5 means that GVPE will send requests twice as often as the minimum latency measured.

For congested or picky DNS forwarders you could use a value nearer to or exceeding 1.

The default should be working OK for most links.

The minimum send interval (= maximum rate) that the DNS transport will use to send new DNS requests. GVPE will not exceed this rate even when the latency is very low. The default is 0.01, which means GVPE will not send more than 100 DNS requests per connection per second. For high-bandwidth links you could go lower, e.g. to 0.001 or so. For congested or rate-limited links, you might want to go higher, say 0.1, 0.2 or even higher.

The default should be working OK for most links.

Factor to multiply the min_latency (see dns-overlap-factor) by to get request timeouts. The default of 8 means that the DNS transport will resend the request when no reply has been received for longer than eight times the minimum (= expected) latency, assuming the request or reply has been lost.

For congested links a higher value might be necessary (e.g. 30). If the link is very stable lower values (e.g. 2) might work nicely. Values near or below 1 makes no sense whatsoever.

The default should be working OK for most links but will result in low throughput if packet loss is high.

if-up = relative-or-absolute-path
Sets the path of a script that should be called immediately after the network interface is initialized (but not necessarily up). The following environment variables are passed to it (the values are just examples).

Variables that have the same value on all nodes:

The configuration base directory.
The network interface to initialize.
The interface type (native or tincd) and the subtype (usually the OS name in lowercase) that this GVPE was configured for. Can be used to select the correct syntax to use for network-related commands.
The MTU to set the interface to. You can use lower values (if done consistently on all nodes), but this is usually either inefficient or simply ineffective.
The number of nodes in this GVPE network.

Variables that are node-specific and with values pertaining to the node running this GVPE:

The value of the configuration directive if-up-data.
The MAC address the network interface has to use.

Might be used to initialize interfaces on platforms where GVPE does not do this automatically. Please see the gvpe.osdep(5) man page for platform-specific information.

The nickname of the node.
The numerical node ID of the node running this instance of GVPE. The first node mentioned in the config file gets ID 1, the second ID 2 and so on.

In addition, all node-specific variables (except NODEID) will be available with a postfix of _nodeid, which contains the value for that node, e.g. the MAC_1 variable contains the MAC address of node #1, while the NODENAME_22 variable contains the name of node #22.

Here is a simple if-up script:

   #!/bin/sh
   ip link set $IFNAME up
   [ $NODENAME = branch1 ] && ip addr add 10.0.0.1 dev $IFNAME
   [ $NODENAME = branch2 ] && ip addr add 10.1.0.1 dev $IFNAME
   ip route add 10.0.0.0/8 dev $IFNAME

More complicated examples (using routing to reduce ARP traffic) can be found in the etc/ subdirectory of the distribution.

Sets the tun interface name to the given name. The default is OS-specific and most probably something like tun0.
Should the tun/tap device be made persistent, that is, should the device stay up even when gvpe exits? Some versions of the tunnel device have problems sending packets when gvpe is restarted in persistent mode, so if the connections can be established but you cannot send packets from the local node, try to set this to off and do an ifconfig down on the device.
Sets the protocol number to be used for the rawip protocol. This is a global option because all nodes must use the same protocol, and since there are no port numbers, you cannot easily run more than one gvpe instance using the same protocol, nor can you share the protocol with other programs.

The default is 47 (GRE), which has a good chance of tunneling through firewalls (but note that gvpe's rawip protocol is not GRE compatible). Other common choices are 50 (IPSEC, ESP), 51 (IPSEC, AH), 4 (IPIP tunnels) or 98 (ENCAP, rfc1241).

Many versions of Linux seem to have a bug that causes them to reorder packets for some ip protocols (GRE, ESP) but not for others (AH), so choose wisely (that is, use 51, AH).

The http-proxy-* family of options are only available if gvpe was compiled with the --enable-http-proxy option and enable tunneling of tcp connections through a http proxy server.

http-proxy-host and http-proxy-port should specify the hostname and port number of the proxy server. See http-proxy-loginpw if your proxy requires authentication.

Please note that gvpe will still try to resolve all hostnames in the configuration file, so if you are behind a proxy without access to a DNS server better use numerical IP addresses.

To make best use of this option disable all protocols except TCP in your config file and make sure your routers (or all other nodes) are listening on a port that the proxy allows (443, https, is a common choice).

If you have a router, connecting to it will suffice. Otherwise TCP must be enabled on all nodes.

Example:

   http-proxy-host = proxy.example.com
   http-proxy-port = 3128       # 8080 is another common choice
   http-proxy-auth = schmorp:grumbeere
The port where your proxy server listens.
The optional login and password used to authenticate to the proxy server, separated by a literal colon (:). Only basic authentication is currently supported.
Sets the keepalive probe interval in seconds (default: 60). After this many seconds of inactivity the daemon will start to send keepalive probe every 3 seconds until it receives a reply from the other end. If no reply is received within 15 seconds, the peer is considered unreachable and the connection is closed.
Set the logging level. Connection established messages are logged at level info, notable errors are logged with error. Default is info.
Sets the maximum MTU that should be used on outgoing packets (basically the MTU of the outgoing interface) The daemon will automatically calculate maximum overhead (e.g. UDP header size, encryption blocksize...) and pass this information to the if-up script.

Recommended values are 1500 (ethernet), 1492 (pppoe), 1472 (pptp).

This value must be the minimum of the MTU values of all nodes.

This advanced option, when set to a nonzero value (default: 0), tries to set the netfilter mark (or fwmark) value on all sockets gvpe uses to send packets.

This can be used to make gvpe use a different set of routing rules. For example, on GNU/Linux, the if-up could set nfmark to 1000 and then put all routing rules into table 99 and then use an ip rule to make gvpe traffic avoid that routing table, in effect routing normal traffic via gvpe and gvpe traffic via the normal system routing tables:

   ip rule add not fwmark 1000 lookup 99
Sets a command (default: none) that should be called whenever a connection is established (even on rekeying operations). Note that node-up/down scripts will be run asynchronously, but execution is serialised, so there will only ever be one such script running.

In addition to all the variables passed to if-up scripts, the following environment variables will be set (values are just examples):

The name of the remote node.
The node id of the remote node.
The "socket info" of the target node, protocol dependent but usually in the format protocol/ip:port.
The numerical IP address of the remote node (gvpe accepts connections from everywhere, as long as the other node can authenticate itself).
The protocol port used by the other side, if applicable.
Node-up scripts get called with STATE=up, node-change scripts get called with STATE=change and node-down scripts get called with STATE=down.

Here is a nontrivial example that uses nsupdate to update the name => ip mapping in some DNS zone:

   #!/bin/sh
   {
     echo update delete $DESTNODE.lowttl.example.net. a
     echo update add $DESTNODE.lowttl.example.net. 1 in a $DESTIP
     echo
   } | nsupdate -d -k $CONFBASE:key.example.net.
Same as node-change, but gets called whenever something about a connection changes (such as the source IP address).
Same as node-up, but gets called whenever a connection is lost.
The path to the pid file to check and create (default: LOCALSTATEDIR/run/gvpe.pid). The first %s is replaced by the nodename - any other use of % must be written as %%.
Sets the path (relative to the config directory) to the private key (default: hostkey). This is a printf format string so every % must be doubled. A single %s is replaced by the hostname, so you could use paths like hostkeys/%s to be able to share the same config directory between nodes.

Since only the private key file of the current node is used and the private key file should be kept secret per-node to avoid spoofing, it is not recommended to use this feature this way though.

Sets the rekeying interval in seconds (default: 3607). Connections are reestablished every rekey seconds, making them use a new encryption key.
The random device used to initially and regularly seed the random number generator (default: /dev/urandom). Randomness is of paramount importance to the security of the algorithms used in gvpe.

On program start and every seed-interval, gvpe will read 64 octets.

Setting this path to the empty string will disable this functionality completely (the underlying crypto library will likely look for entropy sources on it's own though, so not all is lost).

The number of seconds between reseeds of the random number generator (default: 3613). A value of 0 disables this regular reseeding.
The configuration serial number. This can be any string up to 16 bytes length. Only when the serial matches on both sides of a connection will the connection succeed. This is not a security mechanism and eay to spoof, this mechanism exists to alert users that their config is outdated.

It's recommended to specify this is a date string such as 2013-05-05 or 20121205084417.

The exact algorithm is as this: if a connection request is received form a node with an identical serial, then it succeeds normally.

If the remote serial is lower than the local serial, it is ignored.

If the remote serial is higher than the local serial, a warning message is logged.

NODE SPECIFIC SETTINGS

The following settings are node-specific, that is, every node can have different settings, even within the same gvpe instance. Settings that are set before the first node section set the defaults, settings that are set within a node section only apply to the given node.

Allow direct connections to this node. See deny-direct for more info.
For the current node, this specified whether it will accept compressed packets, and for all other nodes, this specifies whether to try to compress data packets sent to this node (default: yes). Compression is really cheap even on slow computers, has no size overhead at all and will only be used when the other side supports compression, so enabling this is often a good idea.
Sets the connect mode (default: always). It can be always (always try to establish and keep a connection to the given node), never (never initiate a connection to the given host, but accept connections), ondemand (try to establish a connection when there are outstanding packets in the queue and take it down after the keepalive interval) or disabled (node is bad, don't talk to it).

Routers will automatically be forced to always unless they are disabled, to ensure all nodes can talk to each other.

Deny direct connections to the specified node (or all nodes when * is given). Only one node can be specified, but you can use multiple allow-direct and deny-direct statements. This only makes sense in networks with routers, as routers are required for indirect connections.

Sometimes, a node cannot reach some other nodes for reasons of network connectivity. For example, a node behind a firewall that only allows connections to/from a single other node in the network. In this case one should specify deny-direct = * and allow-direct = othernodename (the other node must be a router for this to work).

The algorithm to check whether a connection may be direct is as follows:

1. Other node mentioned in an allow-direct? If yes, allow the connection.

2. Other node mentioned in a deny-direct? If yes, deny direct connections.

3. Allow the connection.

That is, allow-direct takes precedence over deny-direct.

The check is done in both directions, i.e. both nodes must allow a direct connection before one is attempted, so you only need to specify connect limitations on one node.

The DNS domain suffix that points to the DNS tunnel server for this node.

The domain must point to a NS record that points to the dns-hostname, i.e.

   dns-domainname = tunnel.example.net
   dns-hostname   = tunnel-server.example.net

Corresponds to the following DNS entries in the example.net domain:

   tunnel.example.net.         NS tunnel-server.example.net.
   tunnel-server.example.net.  A  13.13.13.13
The address to bind the DNS tunnel socket to, similar to the hostname, but for the DNS tunnel protocol only. Default: 0.0.0.0, but that might change.
The port to bind the DNS tunnel socket to. Must be 53 on DNS tunnel servers.
See gvpe.protocol(7) for a description of the DNS transport protocol. Avoid this protocol if you can.

Enable the DNS tunneling protocol on this node, either as server or as client. Support for this transport protocol is only available when gvpe was compiled using the --enable-dns option.

See gvpe.protocol(7) for a description of the ICMP transport protocol.

Enable the ICMP transport using ICMP packets of type icmp-type on this node.

See gvpe.protocol(7) for a description of the RAW IP transport protocol.

Enable the RAW IPv4 transport using the ip-proto protocol (default: no).

See gvpe.protocol(7) for a description of the TCP transport protocol.

Enable the TCPv4 transport using the tcp-port port (default: no). Support for this transport protocol is only available when gvpe was compiled using the --enable-tcp option.

See gvpe.protocol(7) for a description of the UDP transport protocol.

Enable the UDPv4 transport using the udp-port port (default: no).

Forces the address of this node to be set to the given DNS hostname or IP address. It will be resolved before each connect request, so dyndns should work fine. If this setting is not specified and a router is available, then the router will be queried for the address of this node. Otherwise, the connection attempt will fail.

Note that DNS resolving is done synchronously, pausing the daemon. If that is an issue you need to specify IP addresses.

Sets the type value to be used for outgoing (and incoming) packets sent via the ICMP transport.

The default is 0 (which is echo-reply, also known as "ping-reply"). Other useful values include 8 (echo-request, a.k.a. "ping") and 11 (time-exceeded), but any 8-bit value can be used.

The value specified using this directive will be passed to the if-up script in the environment variable IFUPDATA.
Whether to inherit the TOS settings of packets sent to the tunnel when sending packets to this node (default: yes). If set to yes then outgoing tunnel packets will have the same TOS setting as the packets sent to the tunnel device, which is usually what you want.
If true, designates a node as a low-power node. Low-power nodes use larger timeouts and try to reduce cpu time. Other nodes talking to a low-power node will also use larger timeouts, and will use less aggressive optimisations, in the hope of reducing load. Security is not compromised.

The typical low-power node would be a mobile phone, where wakeups and encryption can significantly increase power drain.

The maximum interval in seconds (default: 3600, one hour) between retries to establish a connection to this node. When a connection cannot be established, gvpe uses exponential back-off capped at this value. It's sometimes useful to set this to a much lower value (e.g. 120) on connections to routers that usually are stable but sometimes are down, to assure quick reconnections even after longer downtimes.
Expire packets that couldn't be sent after this many seconds (default: 60). Gvpe will normally queue packets for a node without an active connection, in the hope of establishing a connection soon. This value specifies the maximum lifetime a packet will stay in the queue, if a packet gets older, it will be thrown away.
The maximum number of packets that will be queued (default: 512) for this node. If more packets are sent then earlier packets will be expired. See max-ttl, above.
Sets the router priority of the given node (default: 0, disabled).

If some node tries to connect to another node but it doesn't have a hostname, it asks a router node for it's IP address. The router node chosen is the one with the highest priority larger than 1 that is currently reachable. This is called a mediated connection, as the connection itself will still be direct, but it uses another node to mediate between the two nodes.

The value 0 disables routing, that means if the node receives a packet not for itself it will not forward it but instead drop it.

The special value 1 allows other hosts to route through the router host, but they will never route through it by default (i.e. the config file of another node needs to specify a router priority higher than one to choose such a node for routing).

The idea behind this is that some hosts can, if required, bump the router-priority setting to higher than 1 in their local config to route through specific hosts. If router-priority is 0, then routing will be refused, so 1 serves as a "enable, but do not use by default" switch.

Nodes with router-priority set to 2 or higher will always be forced to connect = always (unless they are disabled).

Similar to udp-port (default: 655), but sets the TCP port number.
Sets the port number used by the UDP protocol (default: 655, not officially assigned by IANA!).

The default (or recommended) directory layout for the config directory is:

The config file.
The if-up script
If used the node up or node-down scripts.
The (default path of the) private key of the current host.
The public keys of the other nodes, one file per node.

gvpe(5), gvpe(8), gvpectrl(8).

Marc Lehmann <gvpe@schmorp.de>

2016-11-12 3.0