dswd
/
vpncloud
mirror of https://github.com/dswd/vpncloud.git
1
0
Fork 0
Code Issues Releases Wiki Activity
vpncloud/vpncloud.adoc

619 lines
26 KiB
Plaintext
Raw Blame History

vpncloud(1)
===========
== Name
vpncloud - Peer-to-peer VPN
== SYNOPSIS
*vpncloud [options] [--config <file>] [-p <password>] [-l <addr>] [-c <addr>...]*
== OPTIONS
*--config <file>*::
Read configuration options from the specified file. Please see the section
*CONFIG FILES* for documentation on the file format.
If the same option is defined in the config file and as a parameter, the
parameter overrides the config file.
*-t <type>*, *--type <type>*::
Set the type of network. There are two options: *tap* devices process
Ethernet frames *tun* devices process IP packets. [default: *tun*]
*-d <name>*, *--device <name>*::
Name of the virtual device. Any *%d* will be filled with a free number.
[default: *vpncloud%d*]
*--device-path <path>*::
The path of the base device inode, e.g. /dev/net/tun.
*--fix-rp-filter*::
If this option is set, VpnCloud will change the rp_filter settings to protect
against a potential system vulnerability. See *SECURITY* for more info.
*-m <mode>*, *--mode <mode>*::
The mode of the VPN. The VPN can like a router, a switch or a hub. A *hub*
will send all data always to all peers. A *switch* will learn addresses
from incoming data and only send data to all peers when the address is
unknown. A *router* will send data according to known subnets of the
peers and ignore them otherwise. The *normal* mode is switch for tap
devices and router for tun devices. [default: *normal*]
*-l <addr>*, *--listen <addr>*::
The address on which to listen for data. This can be simply a port number
or a full address in form IP:PORT. If the IP is specified as \'\*' or only
a port number is given, then the socket will listen on all IPs (v4 and v6),
otherwise the socket will only listen on the given IP.
Alternatively, a websocket proxy URL (starting with ws://) can be given
here. Please see the section *WEBSOCKET PROXY* for more info.
[default: **3210**]
*-c <addr>*, *--peer <addr>*, *--connect <addr>*::
Address of a peer to connect to. The address should be in the form
*addr:port*. If the node is not started, the connection will be retried
periodically. This parameter can be repeated to connect to multiple peers.
*--claim <subnet>*::
The local subnets to claim. This parameter should be in the form
*address/prefixlen* where address is an IPv4 address, an IPv6 address, or a
MAC address. The prefix length is the number of significant front bits that
distinguish the subnet from other subnets. Example: *10.1.1.0/24*.
*--no-auto-claim*::
Do not automatically claim the IP set on the virtual interface (on TUN
devices).
*-p <password>*, *--password <password>*::
A password to encrypt the VPN data. This parameter must be set unless a
password is given in a config file or a private key is set.
See *SECURITY* for more info.
*--key <key>*, *--private-key <key>*::
A private key to use for encryption. The key must be given as base62 as
generated by *genkey*. See *SECURITY* for more info.
*--public-key <key>*::
A public key matching the given private key. The key must be given as base62
as generated by *genkey*. This argument is purely optional. See *SECURITY*
for more info.
*--trust <key>*, **--trusted-key <key>*::
A public key to trust. Any peer must have a key pair that is trusted by this
node, otherwise it will be rejected. The key must be given as base62 as
generated by *genkey*. This argument can be given multiple times. If it is
not set, only the own public key will be trusted. See *SECURITY* for more
info.
*--algo <method>*, *--algorithm <method>*::
Supported encryption algorithms ("plain", "aes128", "aes256", or "chacha20").
Nodes exchange the supported algorithms and select the one that is fastest on
both ends. This parameter can be given multiple times to enable multiple
algorithms. *Warning:* "plain" means unencrypted and needs to be enabled
explicitly. As default, all algorithms except "plain" are enabled.
*--peer-timeout <secs>*::
Peer timeout in seconds. The peers will exchange information periodically
and drop peers that are silent for this period of time. [default: *300*]
*--keepalive <secs>*::
Interval of peer exchange messages in seconds. The peers will exchange
information periodically to keep connections alive. This setting overrides
how often this will happen. [default: *peer-timeout/2-60*]
*--switch-timeout <secs>*::
Switch table entry timeout in seconds. This parameter is only used in switch
mode. Addresses that have not been seen for the given period of time will
be forgotten. [default: *300*]
*--beacon-store <path|command>*::
Periodically store beacons containing the address of this node in the given
file or via the given command. If the parameter value starts with a pipe
character (*|*), the rest of the value is interpreted as a shell command.
Otherwise the value is interpreted as a file to write the beacon to.
If this parameter is not given, beacon storage is disabled.
Please see the section *BEACONS* for more information.
*--beacon-load <path|command>*::
Periodically load beacons containing the addresses of other nodes from the
given file or via the given command. If the parameter value starts with a
pipe character (*|*), the rest of the value is interpreted as a shell
command. Otherwise the value is interpreted as a file to read the beacon
from.
If this parameter is not given, beacon loading is disabled.
Please see the section *BEACONS* for more information.
*--beacon-interval <secs>*::
Beacon storage/loading interval in seconds. If configured to do so via
*--beacon-store* and *--beacon-load*, the node will periodically store its
beacon and load beacons of other nodes. This parameter defines the interval
in seconds. [default: *3600*]
*--beacon-password <password>*::
An optional password to use to encrypt all beacon data. See the section
*BEACONS* for more information.
*--ip <address>*::
An IP address (plus optional prefix length) for the interface. If this
argument is given, the address (and if a prefix length is given, also the
netmask) is configured on the device and the device is activated.
If also *--ifup* is given, the interface is configured before the ifup
command is executed. Please see *DEVICE SETUP* for more info.
*--ifup <command>*::
A command to setup the network interface. The command will be run (as
parameter to *sh -c*) when the device has been created to configure it.
The name of the allocated device will be available via the environment
variable *IFNAME*.
Please note that this command is executed with the full permissions of the
caller. Please see *DEVICE SETUP* for more info.
*--ifdown <command>*::
A command to bring down the network interface. The command will be run (as
parameter to *sh -c*) to remove any configuration from the device.
The name of the allocated device will be available via the environment
variable *IFNAME*.
Please note that this command is executed with the (limited) permissions of
the user and group given as *--user* and *--group*.
*--pid-file <file>*::
Store the process id in this file when running in the background. If set,
the given file will be created containing the process id of the new
background process. This option is only used when running in background.
*--user <user>*::
*--group <group>*::
Change the user and/or group of the process once all the setup has been
done.
*--log-file <file>*::
If set, print logs also to the given file. The file will be created and
truncated if is exists.
*--stats-file <file>*::
If set, periodically write statistics on peers and current traffic to the
given file. The file will be periodically overwritten with new data.
*--statsd-server <server>*::
If set, periodically send statistics on current traffic and some important
events to the given statsd server (host:port).
Please see *STATSD SUPPORT* for more info.
*--statsd-prefix <prefix>*::
Sets the prefix to use for all statsd entries. [default: **vpncloud**]
Please see *STATSD SUPPORT* for more info.
*--daemon*::
Spawn a background process instead of running the process in the foreground.
If this flag is set, the process will first carry out all the
initialization, then drop permissions if *--user* or *--group* is used and
then spawn a background process and write its process id to a file if
*--pid-file* is set. Then, the main process will exit and the background
process continues to provide the VPN. At the time, when the main process
exits, the interface exists and is properly configured to be used.
*--no-port-forwarding*::
Disable automatic port forward. If this option is not set, VpnCloud tries to
detect a NAT router and automatically add a port forwarding to it.
*--hook <script>*::
Call the given script on an event. If the script is in the format *event:script*,
it will only be called for the specified event type, otherwise it will be called
for all events. This parameter can be given multiple times.
Please see the section *HOOK SCRIPTS* for more info.
*-v*, *--verbose*::
Print debug information, including information for data being received and
sent.
*-q*, *--quiet*::
Only print errors and warnings.
*-h*, *--help*::
Display the help.
== SUBCOMMANDS
The following subcommands can be given to run some special action instead of
running a VpnCloud instance. Any parameters must be given after the subcommand
name (except for -v -q and -h). Only the listed parameters are accepted for the
subcommands.
*genkey*::
Generate and print a random key pair and exit. The key pair is printed as
base62 and can be used as private-key, public-key and trusted-key options.
See *SECURITY* for more info.
*-p <password>*, *--password <password>*:::
Derive the key pair from the given password instead of creating randomly.
*ws-proxy*::
Run a websocket proxy instead of the normal VpnCloud instance.
See *WEBSOCKET PROXY* for more info.
*-l <addr>*, *--listen <addr>*:::
Listen on the given TCP address (IP:PORT or PORT). [default: **3210**]
*migrate-config*::
Migrate an old config to the current config format.
*--config-file*:::
The path of the config file to convert.
*completion*::
Output shell completions for the VpnCloud command.
*--shell*:::
The shell type to create completions for. [default: **bash**]
== DESCRIPTION
*VpnCloud* is a peer-to-peer VPN over UDP. It creates a virtual network
interface on the host and forwards all received data via UDP to the
destination. It can work in 3 different modes:
*Switch mode*:: In this mode, the VPN will dynamically learn addresses
as they are used as source addresses by peers and use them to forward data to
its destination. Addresses that have not been seen for some time
(option *switch_timeout*) will be forgotten. Data for unknown addresses will be
broadcast to all peers. This mode is the default mode for TAP devices that
process Ethernet frames but it can also be used with TUN devices and IP
packets.
*Hub mode*:: In this mode, all data will always be broadcast to all peers.
This mode uses lots of bandwidth and should only be used in special cases.
*Router mode*:: In this mode, data will be forwarded based on preconfigured
address ranges ("claims"). Data for unclaimed addresses will be silently
ignored. This mode is the default mode for TUN devices that work with IP
packets but it can also be used with TAP devices and Ethernet frames.
All connected VpnCloud nodes will form a peer-to-peer network and cross-connect
automatically until the network is fully connected. The nodes will periodically
exchange information with the other nodes to signal that they are still active
and to allow the automatic cross-connect behavior. There are some important
things to note:
. The cross-connect behavior can be able to connect nodes that are behind
firewalls or NATs as it can function as hole-punching.
. The management traffic will increase with the peer number quadratically.
It should still be reasonably small for high node numbers (below 10 KiB/s
for 10.000 nodes). A longer *peer_timeout* can be used to reduce the traffic
further. For high node numbers, router mode should be used as it never
broadcasts data.
VpnCloud does not implement any loop-avoidance. Since data received on the UDP
socket will only be sent to the local network interface and vice versa, VpnCloud
cannot produce loops on its own. On a TAP device, however STP data can be
transported to avoid loops caused by other network components.
For TAP devices, IEEE 802.1q frames (VLAN tagged) are detected and forwarded
based on separate MAC tables. Any nested tags (Q-in-Q) will be ignored.
== EXAMPLES
=== Simple multi-node connectivity
In the example scenario, a simple layer-3 network tunnel is established. Most
likely those commands need to be run as *root* using *sudo*.
First, VpnCloud need to be started on both nodes (the address after *-c* is the
address of the remote node and the the *X* in the interface address must be
unique among all nodes, e.g. 0, 1, 2, ...):
----
vpncloud -c REMOTE_HOST:PORT --ip 10.0.0.X/24 --password PASSWORD
----
Afterwards, the interface can be used to communicate.
=== Routed TUN example
In this example, 2 nodes and their subnets should communicate using IP.
First, VpnCloud need to be started on both nodes:
----
vpncloud -t tun -c REMOTE_HOST:PORT --ip 10.0.X.1 --claim 10.0.X.0/24 --password PASSWORD
----
It is important to configure the interface in a way that all addresses on the
VPN can be reached directly. E.g. if subnets 10.0.1.0/24, 10.0.2.0/24 and so on
are used, the interface needs to be configured as 10.0.1.1/16.
For TUN devices, this means that the prefix length of the subnets
(/24 in this example) must be different than the prefix length that the
interface is configured with (/16 in this example).
=== Important notes
. VpnCloud can be used to connect two separate networks. TAP networks can be
bridged using *brctl* and TUN networks must be routed. It is very important
to be careful when setting up such a scenario in order to avoid network loops,
security issues, DHCP issues and many more problems.
. TAP devices will forward DHCP data. If done intentionally, this can be used
to assign unique addresses to all participants. If this happens accidentally,
it can conflict with DHCP servers of the local network and can have severe
side effects.
== CONFIG FILES
The config file is a YAML file that contains configuration values. All entries
are optional and override the defaults. Please see the section *OPTIONS* for
detailed descriptions of the options.
*device*:: A key-value map with device settings
*type*::: Set the type of network. Same as *--type*
*name*::: Name of the virtual device. Same as *--device*
*path*::: Set the path of the base device. Same as *--device-path*
*fix-rp-filter*::: Fix the rp_filter settings on the host. Same as *--fix-rp-filter*
*ip*:: An IP address (plus optional prefix length) for the interface. Same as *--ip*
*ifup*:: A command to setup the network interface. Same as *--ifup*
*ifdown*:: A command to bring down the network interface. Same as *--ifdown*
*crypto*:: A key-value map with crypto settings
*algorithms*::: The encryption algorithms to support. See *--algorithm*
*password*::: The password to use for encryption. Same as *--password*
*private-key*::: The private key to use. Same as *--private-key*
*public-key*::: The public key to use. Same as *--public-key*
*trusted-keys*::: Other public keys to trust. See *--trusted-key*
*listen*:: The address on which to listen for data. Same as *--listen*
*peers*:: A list of addresses to connect to. See *--connect*
*peer_timeout*:: Peer timeout in seconds. Same as *--peer-timeout*
*keepalive*:: Periodically send message to keep connections alive. Same as *--keepalive*
*beacon*:: A key-value map with beacon settings
*store*::: Path or command to store beacons. Same as *--beacon-store*
*load*::: Path or command to load beacons. Same as *--beacon-load*
*interval*::: Interval for loading and storing beacons in seconds. Same as *--beacon-interval*
*password*::: Password to encrypt the beacon with. Same as *--beacon-password*
*mode*:: The mode of the VPN. Same as *--mode*
*switch_timeout*:: Switch table entry timeout in seconds. Same as *--switch-timeout*
*claims*:: A list of local subnets to claim. See *--claim*
*auto-claim*:: Whether to automatically claim the device ip. See *--no-auto-claim*
*port_forwarding*:: Whether to activate port forwardig. See *--no-port-forwarding*
*user*:: The name of a user to run the background process under. Same as *--user*
*group*:: The name of a group to run the background process under. Same as *--group*
*pid_file*:: The path of the pid file to create. Same as *--pid-file*
*stats_file*:: The path of the statistics file. Same as *--stats-file*
*statsd*:: A key-value map with statsd settings
*server*::: Server to report statistics to. Same as *--statsd-server*
*prefix*::: Prefix to use when reporting to statsd. Same as *--statsd-prefix*
*hook*:: A hook script to be called for every event type. See *HOOK SCRIPTS* for info.
*hooks*:: A map of event type to script for scripts that only fire for one event type. See *HOOK SCRIPTS* for info.
=== Example
device:
type: tun
name: vpncloud%d
ip: 10.0.1.1/16
crypto:
password: mysecret
listen: 3210
peers:
- remote.machine.foo:3210
- remote.machine.bar:3210
peer_timeout: 600
mode: normal
claims:
- 10.0.1.0/24
port_forwarding: true
user: nobody
group: nogroup
pid_file: /run/vpncloud.pid
== SECURITY
VpnCloud uses strong cryptography based on modern cryptographic primitives.
Before exchanging any payload data with peers a secure connection is
initialized based on key pairs. Each node has a key pair consisting of a
private and a public key (*--private-key* and *--public-key*). Those key pairs
can be generated via *genkey*.
To allow connections, nodes need to list the public keys of all other nodes as
trusted keys (*--trusted-key*). To simplify the key exchange, key pairs can be
derived from passwords (*--password*). If no trusted keys are configured, nodes
will only trust their own public key. Nodes configured with the same password
will therefore trust each others.
In the initialization phase of the connection, nodes agree on a temporary key
that is used to encrypt the next messages using a fast encryption algorithm.
VpnCloud automatically benchmarks all supported algorithms and negotiates to
use the fastest algorithm for each connection. Users can limit the supported
algorithms if they wish using *--algorithm*. Although highly discouraged, users
can opt out of encryption altogether by enabling the *plain* algorithm. (Note:
both nodes in a connection must support this, otherwise encryption will take
place.)
The temporary encryption keys are rotated periodically so they are never used
for a longer time.
Please refer to the security whitepaper for more details.
=== CVE-2019-14899
The Linux kernel contains a vulnerability that affects all VPNs disregarding of
the specific technology being used. Under some circumstances, the kernel accepts
packets for the address range configured on the vpn interface also on other
interfaces. This way, an attacker can test the presence of a VPN and find out
the IPs being used. Also the attacker can with some effort inject data and
manipulate connections that should be protected by the VPN.
To mitigate this, the rp_filter setting should be configured to strict mode,
which unfortunately a lot of distributions do not set as default.
VpnCloud will detect this misconfiguration and offers to fix it via
*--fix-rp-filter*.
Note: This vulnerability affects all VPN technologies as it is not located in
the VPN software but in the Linux kernel.
== BEACONS
Beacons are short character sequences that contain a timestamp and a list of
addresses. They can be published and retrieved by other nodes to find peers
without the need for static addresses.
The beacons are short (less than 100 characters), encrypted and encoded with
printable characters to allow publishing them in various places on the
internet, e.g.:
* On shared drives or synchronized folders (e.g. on Dropbox)
* Via a dedicated database
* Via a general purpose message board of message service (e.g. Twitter)
The beacons are very robust. They only consist of alphanumeric characters
and can be interleaved with non-alphanumeric characters (e.g. whitespace).
Also the beacons contain a prefix and suffix that depends on the configured
network magic and secret key (if set) so that all nodes can find beacons in
a long text.
When beacons are stored or loaded via a command (using the pipe character *|*),
the command is interpreted using the configured shell *sh*. This command has
access to the following environment variables:
*$begin*:: The prefix of the beacon.
*$end*:: The suffix of the beacon.
*$data* (only on store):: The middle part of the beacon. Do not use this
without prefix and suffix!
*$beacon* (only on store):: The full beacon consisting of prefix, data and
suffix.
The commands are called in separate threads, so even longer running commands
will not block the node.
== STATSD SUPPORT
When a statsd server is configured (either via **--statsd-server** or the
config option **statsd_server**), VpnCloud sends out the following statistics
every minute.
Gauge values:
*peer_count*:: Current number of peers
*table_entries*:: Number of routing table / switch table entries
The following statistics consist of two keys: *.bytes* and *.packets* that hold
the values in bytes and packets. All values refer to the traffic during the
last minute:
*traffic.protocol.inbound*:: Complete incoming traffic with all peers
*traffic.protocol.outbound*:: Complete outgoing traffic with all peers
*traffic.payload.inbound*:: Incoming payload traffic with all peers
*traffic.payload.outbound*:: Outgoing payload traffic with all peers
*invalid_protocol_traffic*:: Invalid incoming protocol traffic
*dropped_payload*:: Outgoing traffic that could not be routed
All keys are prefixed by a common prefix. The prefix defaults to *vpncloud* but
can be changed via **--statsd-prefix** or the config option **statsd_prefix**.
== WEBSOCKET PROXY
The websocket proxy mode replaces the local UDP port by a websocket proxy to allow
connectivity even in very restricted environments.
This means that instead of listening on a local port for incoming messages and
sending outgoing messages via this port, all UDP traffic will be forwarded to and
received from a remote proxy via the websocket protocol. This proxy opens a UDP
port for each VpnCloud instance that connects to it. The instance can use this port
remotely just like it would use a real local UDP port.
The proxy is transparent, it does not manipulate or even decrypt the messages it
forwards. Trust relations are still created between VpnCloud instances, not between
an instance and the proxy. The proxy only ever sees encrypted messages. Therefore,
the connection to it uses plain HTTP.
A websocket proxy can be stared by using the *ws-proxy* subcommand. A custom port
can be set using the *--listen* parameter. (Note that this port never conflicts
with a VpnCloud port on the same machine since VpnCloud uses UDP and the proxy uses
TCP.)
A VpnCloud instance can use a websocket proxy instead of opening a local port by
specifying the websocket proxy via its *--listen* parameter
(e.g. *--listen ws://example.com:3210*). Note that the websocket URL must start with
*ws:\/\/*, not *http:\/\/*.
== HOOK SCRIPTS
VpnCloud supports calling hook scripts on certain events. The scripts can either be
configured on the command line or in the config file. Hook scripts can either be
configured per event type or globally.
When an event occurs, the specified hook script is executed using
the current permissions of the user that started the instance. Note that this means
that if VpnCloud is configured to drop permissions, only the events *device_setup*
and *device_configured* will be executed using root permissions.
Hook scripts are executed using *sh -c*, so either binaries, shell scripts and even
shell commands can be used. The script will be executed in parallel to the VpnCloud
instance. Its output will be printed to the stdout of VpnCloud and the return code
is ignored.
The hook script will receive information on the event using environment variables.
The variable *EVENT* will contain the name of the event. Other variables depend on
the event type.
The following event types
*peer_connecting*::
A new peer connection is in the process of being established. The variable
*PEER* contains the address of the peer but no other information is known at
that point in time.
Variables: *IFNAME*, *PEER*
*peer_connected*::
A new peer successfully connected to this instance. Besides the peer address,
also a list of claims (*CLAIMS*, space separated) and the node id of the new
peer (*NODE_ID*) are given to the script.
Variables: *IFNAME*, *PEER*, *CLAIMS*, *NODE_ID*
*peer_disconnected*::
A peer connection has been closed. If the peer has been fully connected, the
node id is given (*NODE_ID*).
Variables: *IFNAME*, *PEER*, (*NODE_ID*)
*device_setup*::
This event is fired when the virtual device has been created but not yet
configured.
Variables: *IFNAME*
*device_configured*::
This event is fired when the virtual device is fully configured.
Variables: *IFNAME*
*vpn_started*::
This event is fired when the VPN is ready to be used.
Variables: *IFNAME*
*vpn_shutdown*::
This event is fired when the VPN s shutting down.
Variables: *IFNAME*
== DEVICE SETUP
The device is setup using the following steps:
. The device is created with the type and name given as *--type* and *--device*.
. Depending on the device type and the main network device of the systme, the
optimal MTU is determined and configured on the device.
. If and IP address (and optional prefix length) is given via *--ip*, the
interface is configured with the address and the given netmask (default:
255.255.255.0). Also the interface is set to be active.
. If a command is given as *--ifup*, the given command will be executed. The
name of the interface is stored in an environment variable as "IFNAME". Note
that VpnCloud waits for the command to exit before starting its normal
operation.
Note that most of the steps will need elevated permissions, so the vpncloud
command needs to be executed as root (e.g. via sudo). Beware that the ifup
command will also be executed using those permissions.
VpnCloud can drop the elevated permissions when *--user* and *--group* is
given.
== COPYRIGHT
Copyright (C) 2015-2020 Dennis Schwerdel
This software is licensed under GPL-3 or newer (see LICENSE.md)
Reference in New Issue View Git Blame Copy Permalink