strongSwan 4.2 - Configuration
strongSwan is an OpenSource IPsec solution for the Linux operating system. It currently supports the following major functions:
In the following examples we assume for reasons of clarity that left designates the local host and that right is the remote host. Certificates for users, hosts and gateways are issued by a ficticious strongSwan CA. How to generate private keys and certificates using OpenSSL will be explained in section 3. The CA certificate strongswanCert.pem must be present on all VPN end points in order to be able to authenticate the peers.
In this scenario two security gateways moon and sun will connect the two subnets moon-net and sun-net with each other through a VPN tunnel set up between the two gateways:
Configuration on gateway moon:
Configuration on gateway sun:
This is a setup between two single hosts which don't have a subnet behind them. Although IPsec transport mode would be sufficient for host-to-host connections we will use the default IPsec tunnel mode.
Configuration on host moon:
Configuration on host sun:
In a site-to-site setup a system administrator logged into the local gateway often would like to access the peer gateway or a server in the subnet behind the peer gateway over a secure IPsec tunnel.Since IP packets leaving a gateway via the outer network interface carry the IP address of this NIC, four IPsec Security Associations (SAs) must be set up to achieve full connectivity. The example below shows how this can be done without much additional typing work , using the "also" macro which includes connection definitions defined farther down in the ipsec.conf file.
Configuration on gateway moon:
Configuration on gateway sun:
As you certainly agree, the full four tunnel case described in the previous section becomes quite complex. If we could force the source address of the IP packets leaving the gateway through the outer interface to take on the IP address of the inner interface then we could use the single subnet-to-subnet tunnel from section 2.1. Such a setup becomes possible if we use the source routing capabilites of the ip route command that is already used by strongSwan 's updown scripts.
If we assume that the inner IP address of gateway moon is 10.1.0.1 and the inner IP address of gateway sun is 10.2.0.1, then the insertion of the parameter
in the connection definition of moon and
on sun, respectively, will install source routing on both gateways. As a result the command
executed on moon will leave the gateway with a source address of 10.1.0.1 and will therefore take the net-net IPsec tunnel.
Configuration on gateway moon:
Configuration on gateway sun:
This is a very common case where a strongSwan gateway serves an arbitrary number of remote VPN clients usually having dynamic IP addresses.
Configuration on gateway moon:
Configuration on roadwarrior carol:
Roadwarriors usually have dynamic IP addresses assigned by the ISP they are
currently attached to. In order to simplify the routing from
moon-net back to
the remote access client it would be desirable if the roadwarrior had
This virtual IP address can be assigned to a strongSwan roadwarrior by adding the parameter
to the roadwarrior's ipsec.conf. Of course the virtual IP of each roadwarrior must be distinct. In our example it is chosen from the address pool
which can be added to the gateway's ipsec.conf so that a single connection definition can handle multiple roadwarriors.
Configuration on gateway moon:
Configuration on roadwarrior carol:
This section is not a full-blown tutorial on how to use OpenSSL. It just lists a few points that are relevant if you want to generate your own certificates and CRLs for use with strongSwan.
The OpenSSL statement
creates a 2048 bit RSA private key strongswanKey.pem and a self-signed CA certificate strongswanCert.pem with a validity of 4 years (1460 days).
lists the properties of a X.509 certificate cert.pem. It allows you to verify whether the configuration defaults in openssl.cnf have been inserted correctly.
If you prefer the CA certificates to be in binary DER format then the following command achieves this transformation:
The directory /etc/ipsec.d/cacerts/ contains all required CA certificates either in binary DER or in base64 PEM format. Irrespective of the file suffix, Pluto "automagically" determines the correct format.
The OpenSSL statement
generates a 1024 bit RSA private key hostKey.pem and a certificate request hostReq.pem which has to be signed by the CA.
If you want to add a subjectAltName field to the host certificate you must edit the OpenSSL configuration file openssl.cnf and add the following line in the [ usr_cert ] section:
if you want to identify the host by its Fully Qualified Domain Name (FQDN ), or
if you want the ID to be of type IPV4_ADDR . Of course you could include both ID types with
but the use of an IP address for the identification of a host should be discouraged anyway.
For user certificates the appropriate ID type is USER_FQDN which can be specified as
or if the user's e-mail address is part of the subject's distinguished name
If you want to use the dynamic CRL fetching feature described in section 4.7 then you may include one or several crlDistributionPoints in your end certificates. This is also done in the [ usr_cert ] section of the openssl.cnf configuration file
If you have only a single http distribution point then the short form
also works. Due to a known bug in OpenSSL the latter notation fails with ldap URIs.
Now the certificate request can be signed by the CA with the command
If you omit the -days option then the default_days value (365 days) specified in openssl.cnf is used. The -notext option avoids that a human readable listing of the certificate is prepended to the base64 encoded certificate body.
Usually a Windows-based VPN client needs its private key, its host or user certificate, and the CA certificate. The most convenient way to load this information is to put everything into a PKCS#12 file:
An empty CRL that is signed by the CA can be generated with the command
If you omit the -crldays option then the default_crl_days value (30 days) specified in openssl.cnf is used.
If you prefer the CRL to be in binary DER format, then this conversion can be achieved with
The directory /etc/ipsec.d/crls/ contains all CRLs either in binary DER or in base64 PEM format. Irrespective of the file suffix, pluto "automagically" determines the correct format.
A specific host certificate stored in the file host.pem is revoked with the command
Next the CRL file must be updated
The content of the CRL file can be listed with the command
in the case of a base64 CRL, or alternatively for a CRL in DER format
Usually the local side is the same for all connections. Therefore it makes sense to put the definitions characterizing the strongSwan security gateway into the conn %default section of the configuration file /etc/ipsec.conf. If we assume throughout this document that the strongSwan security gateway is left and the peer is right (of course you could define the directions also the other way round) then we can write
The X.509 certificate by which the strongSwan security gateway will authenticate itself by sending it in binary form to its peers as part of the Internet Key Exchange (IKE) is specified by the line
The certificate can either be stored in base64 PEM-format or in the binary DER-format. Irrespective of the file suffix, Pluto "automagically" determines the correct format. Therefore
would also be valid alternatives.
When using relative pathnames as in the examples above, the certificate files must be stored in in the directory /etc/ipsec.d/certs. In order to distinguish strongSwan's own certificates from locally stored trusted peer certificates (see section 5.5 for details), they could also be stored in a subdirectory within the default directory /etc/ipsec.d/certs, as e.g. in
Absolute pathnames are also possible, as in
As an ID for the VPN gateway we recommend the use of a Fully Qualified Domain Name (FQDN) of the form
Important: When an FQDN identifier is used it must be explicitly included as a so called subjectAltName of type dnsName (DNS:) in the certificate indicated by leftcert. For details on how to generate certificates with subjectAltNames, please refer to section 3.2.
If you don't want to mess with subjectAltNames, you can use the certificate's Distinguished Name (DN) instead, which is an identifier of type DER_ASN1_DN and which can be written e.g. in the LDAP-type format
Since the subject's DN is part of the certificate, the leftid does not have to be declared explicitly. Thus the entry
Now we can proceed to define our connections. In many applications we might have dozens of mostly Windows-based road warriors connecting to a central strongSwan security gateway. The following most simple statement:
defines the general roadwarrior case. The line right=%any literally means that any IPSec peer is accepted, regardless of its current IP source address and its ID, as long as the peer presents a valid X.509 certificate signed by a CA the strongSwan security gateway puts explicit trust in. Additionally the signature during IKE main mode gives proof that the peer is in possession of the private RSA key matching the public key contained in the transmitted certificate.
The ID by which a peer is identifying itself during IKE main mode can by any of the ID types IPV4_ADDR, FQDN, USER_FQDN or DER_ASN1_DN. If one of the first three ID types is used, then the accompanying X.509 certificate of the peer must contain a matching subjectAltName field of the type ipAddress (IP:), dnsName (DNS:) or rfc822Name (email:), respectively. With the fourth type DER_ASN1_DN, the identifier must completely match the subject field of the peer's certificate. One of the two possible representations of a Distinguished Name (DN) is the LDAP-type format
rightid="C=CH,O=Linux strongSwan, CN=sun.strongswan.org"
Additional whitespace can be added everywhere as desired since it will be automatically eliminated by the X.509 parser. An exception is the single whitespace between individual words , like e.g. in Linux strongSwan, which is preserved by the parser.
The Relative Distinguished Names (RDNs) can alternatively be separated by a slash ( '/') instead of a comma (',')
This is the representation extracted from the certificate by the OpenSSL command line option
The following X.501 RDNs are supported by strongSwan
With the roadwarrior connection definition listed above, an IPsec SA for the strongSwan security gateway moon.strongswan.org itself can be established. If any roadwarrior should be able to reach e.g. the two subnets 10.0.1.0/24 and 10.0.3.0/24 behind the security gateway then the following connection definitions will make this possible
If not all peers in possession of a X.509 certificate signed by a specific certificate authority shall be given access to the Linux security gateway, then either a subset of them can be barred by listing the serial numbers of their certificates in a certificate revocation list (CRL) as specified in section 5.2 or as an alternative, access can be controlled by explicitly putting a roadwarrior entry for each eligible peer into ipsec.conf:
When the IP address of a peer is known to be stable, it can be specified as well. This entry is mandatory when the strongSwan host wants to act as the initiator of an IPSec connection.
In the last example the ID types FQDN, USER_FQDN, DER_ASN1_DN and IPV4_ADDR, respectively, were used. Of course all connection definitions presented so far have included the lines in the conn %defaults section, comprising among others a left and leftcert entry,.
Often roadwarriors are behind NAT-boxes with IPsec passthrough, which causes the inner IP source address of an IPsec tunnel to be different from the outer IP source address usually assigned dynamically by the ISP. Whereas the varying outer IP address can be handled by the right=%any construct, the inner IP address or subnet must always be declared in a connection definition. Therefore for the three roadwarriors rw1 to rw3 connecting to a strongSwan security gateway the following entries are required in /etc/ipsec.conf:
With the new wildcard parameter rightsubnetwithin these three entries can be reduced to the single connection definition
Any host will be accepted (of course after successful authentication based on the peer's X.509 certificate only) if it declares a client subnet lying totally within the brackets defined by the wildcard subnet definition (in our example 10.0.1.0/24). For each roadwarrior a connection instance tailored to the subnet of the particular client will be created, based on the generic rightsubnetwithin template.
This strongSwan feature can also be helpful with VPN clients getting a dynamically assigned inner IP from a DHCP server located on the NAT router box.
strongSwan offers the possibility to restrict the protocol and
optionally the ports in an IPsec SA using the rightprotoport and
Protocols and ports can be designated either by their numerical values or by their acronyms defined in /etc/services.
lists the following connection definitions:
Based on the protocol and port selectors, appropriate eroutes will be set up, so that only the specified payload types will pass through the IPsec tunnel.
Let's make a practical example:An organization has a sales department (OU=Sales) and a research group (OU=Research). In the company intranet there are separate subnets for Sales (10.0.0.0/24) and Research (10.0.1.0/24) but both groups share a common web server (10.0.2.100). The VPN clients use Virtual IP addresses that are either assigned statically or via DHCP-over-IPsec. The sales and research departments use IP addresses from separate DHCP address pools (10.1.0.0/24) and (10.1.1.0/24), respectively. An X.509 certificate is issued to each employee, containing in its subject distinguished name the country (C=CH), the company (O=ACME), the group membership (OU=Sales or OU=Research) and the common name (e.g. CN=Bart Simpson).
The IPsec policy defined above can now be enforced with the following three IPsec security associations:
Of course group specific tunneling could be implemented on the basis of the Virtual IP range specified by the rightsubnetwithin parameter alone, but the wildcard matching mechanism guarantees that only authorized user can access the corresponding subnets.The '*' character is used as a wildcard in relative distinguished names (RDNs). In order to match a wildcard template, the ID_DER_ASN1_DN of a peer must contain the same number of RDNs (selected from the list in section 4.3) appearing in the exact order defined by the template.
"C=CH, O=ACME, OU=Research, OU=Special Effects, CN=Bart Simpson"
matches the templates
but not the template
"C=CH, O=ACME, OU=*, CN=*"
which doesn't have the same number of RDNs.section 4.6,
access to specific client host and subnets can also be controlled on the basis of
the CA that issued the peer certificate.
In the example above, the connection "sales" can be used by peers presenting certificates issued by the Sales CA, only. In the same way, the use of the connection "research" is restricted to owners of certificates issued by the Research CA. The connection "web" is open to both "Sales" and "Research" peers because the required ACME Root CA is the issuer of the Research and Sales intermediate CAs. If no rightca parameter is present then any valid certificate issued by one of the trusted CAs in /etc/ipsec.d/cacerts can be used by the peer.
The leftca parameter usually doesn't have to be set explicitly because by default it is set to the issuer field of the certificate loaded vialeftcert. The statement
sets the CA requested from the peer to the CA used by the left side itself as e.g. in
RFC 3281) are the most powerful mechanism for implementing IPsec security policies. The rightgroups parameter in a connection definition restricts the access to members of the listed groups only. An IPsec peer must have a valid attribute certificate issued by a trusted Authorization Authority and listing one of the required group attributes in order to get admitted.
In the examples above membership of the group Sales is required for connection sales and membership of Research for connection research whereas connection web is accessible for both groups. Currently the attribute certificates of the peers must be loaded statically via the /etc/ipsec.d/acerts/ directory. In future releases of strongSwan it will be possible to fetch them from an LDAP directory server.
X.509 certificates received by strongSwan during the IKE protocol are automatically authenticated by going up the trust chain until a self-signed root CA certificate is reached. Usually host certificates are directly signed by a root CA, but strongSwan also supports multi-level hierarchies with intermediate CAs in between. All CA certificates belonging to a trust chain must be copied in either binary DER or base64 PEM format into the directory
Multiple CAs are supported, but presently they just create a large pool of valid user or host certificates and cannot be assigned to specific connection definitions in /etc/ipsec.conf.
By copying a CA certificate into /etc/ipsec.d/cacerts/, automatically all user or host certificates issued by this CA are declared valid. Unfortunately private keys might get compromised inadvertently or intentionally, personal certificates of users leaving a company have to be blocked immediately, etc. To this purpose certificate revocation lists (CRLs) have been created. CRLs contain the serial numbers of all user or host certificates that have been revoked due to various reasons.
After successful verification of the X.509 trust chain, Pluto searches its list of CRLs either obtained by loading them from the /etc/ipsec.d/crls/ directory or fetching them dynamically from a http or ldap server for the presence of a CRL issued by the CA that has signed the certificate.
If the serial number of the certificate is found in the CRL then the public key contained in the certificate is declared invalid and the IPSec SA will not be established. If no CRL is found or if the deadline defined in the nextUpdate field of the CRL has been reached, a warning is issued but the public key will nevertheless be accepted. CRLs must be stored in either binary DER or base64 PEM format in the crls directory. Section 3.3 will explain in detail how CRLs can be created using OpenSSL.
Pluto reads certificates and CRLs from their respective files during system startup and keeps them in memory in the form of chained lists. X.509 certificates have a finite life span defined by their validity field. Therefore it must be possible to replace CA or OCSP certificates kept in system memory without disturbing established ISAKMP SAs. Certificate revocation lists should also be updated in the regular intervals indicated by the nextUpdate field in the CRL body. The following interactive commands allow the dynamic replacement of the various files:
CRLs can also be automatically fetched from an http or ldap server using the CRL distribution points contained in X.509 certificates. The command
shows any pending fetch requests:
Oct 31 00:29:53 2002, trials: 2
In the example above, a http and a ldap URL were extracted from a received end certificate. An independent thread then tries to fetch a CRL from the designated distribution points. The same thread also periodically checks if any loaded CRLs are about to expire. The check interval can be defined in the config setup section of the ipsec.conf file:
In our example the thread wakes up every 600 seconds or 10 minutes in order to check the validity of the CRLs or to retry any pending fetch requests:
List of X.509 CRLs:
The first trial to update a CRL is started 2*crlcheckinterval before the nextUpdate time, i.e. when less than 20 minutes are left in our practical example. When crlcheckinterval is set to 0 (this is also the default value when the parameter is not set in ipsec.conf) then the CRL checking and updating thread is not started and dynamic CRL fetching is disabled.
The ipsec.conf option
activates the local caching of CRLs that were dynamically fetched from an HTTP or LDAP server. Cached copies are stored in /etc/ipsec.d/crls under a unique filename formed from the issuer's SubjectKeyIdentifier and the suffix .crl.
With the cached copy the CRL is immediately available after pluto's startup. When the local copy is about to expire it is automatically replaced with an updated CRL fetched from one of the defined CRL distribution points.
The Online Certificate Status Protocol is defined by RFC 2560. It can be used to query an OCSP server about the current status of an X.509 certificate and is often used as a more dynamic alternative to a static Certificate Revocation List (CRL). Both the OCSP request sent by the client and the OCSP response returned by the server are transported via a standard tcp/http connection. Therefore cURL support must be enabled in pluto/Makefile:
In the most simple OCSP setup, a default URI under which the OCSP server for a given CA can be accessed is defined in ipsec.conf:
The HTTP port can be freely chosen. In our example we have assumed tcp port 8880. The crlcheckinterval must be set to a value different from zero. Otherwise the OCSP fetching thread will not be started.
The well-known openssl-0.9.7 package from http://www.openssl.org implements an OCSP server that can be used in conjunction with an OpenSSL-based Public Key Infrastructure. The OCSP client integrated into Pluto does not contain any OpenSSL code though, but is based on the existing ASN.1 functionality of the X.509 patch.
The OpenSSL-based OCSP server is started with the following command:
The command consists of the parameters
How does Pluto get hold of the OCSP signer certificate? There are two possibilities: Either you put the OCSP certificate into the default directory
or alternatively Pluto can receive it as part of the OCSP response from the remote OCSP server. In the latter case, how can Pluto make sure that the server has indeed been authorized by the CA to deal out certificate status information? In order to ascertain the OCSP signer authority, an extended key usage attribute can be included in the OCSP server certificate. Just insert the parameter
in the [ usr_cert ] section of your openssl.cnf configuration file before the CA signs the OCSP server certificate.
For a given CA the corresponding ca section in ipsec.conf (see section 7) allows to define the URI of a single OCSP server. As an alternative an OCSP URI can be embedded into the VPN host and user certificates by putting the line
into the [ usr_cert ] section of your openssl.cnf configuration file. If an OCSP authorityInfoAccess extension is present in a certificate then this record overrides the default URI defined by the ca section.
By default Pluto is quite tolerant concerning the handling of CRLs. It is not mandatory for a CRL to be present in /etc/ipsec.d/crls/ and if the expiration date defined by the nextUpdate field of a CRL has been reached just a warning is issued, but a peer certificate will always be accepted if it has not been revoked.
If you want to enforce a stricter CRL policy then this can be done by setting the strictcrlpolicy option in the config setup section of the ipsec.conf file:
A certificate received from a peer will not be accepted if no corresponding CRL is present in /etc/ipsec.conf. And if an ISAKMP SA re-negotiation takes place after the nextUpdate deadline has been reached, the peer certificate will be declared invalid and the cached RSA public key will be deleted causing the connection in question to fail. Therefore if you are going to use the strictcrlpolicy=yes option, make sure that the CRLs will always be updated in time. Otherwise a total stand still will ensue.
As mentioned earlier the default setting is strictcrlpolicy=no
If you don't want to use trust chains based on CA certificates as proposed in section 4.3, you can alternatively import trusted peer certificates directly into Pluto. Thus you do not have to rely on the certificate to be transmitted by the peer as part of the IKE protocol.
the other side. In these cases it is recommended to add
to the connection definition[s] in order to avoid the sending of the host's own certificate. The default value is
If a peer certificate contains a subjectAltName extension, then this can be used to specify an alternative rightid or leftid type, respectively, as the example "conn sun" shows. If no rightid or leftid entry is present then the subject distinguished name contained in the certificate is taken as the ID.
Using the same rules concerning pathnames that apply to strongSwan's own certificates, the following two definitions are also valid for trusted peer certificates:
Besides the standard FreeS/WAN RSA private key format, Pluto has been enabled to load RSA private keys in the PKCS#1 file format. The key files can be optionally secured with a passphrase.
RSA private key files are declared in /etc/ipsec.secrets using the syntax
The key file can be either in base64 PEM-format or binary DER-format. The actual coding is detected "automagically" by Pluto. The example
uses a relative pathname. In that case Pluto will look for the private key file in the directory
As an alternative an absolute pathname can be given as in
In both cases make sure that the key files are root readable only.
Often a private key must be transported from the Certification Authority where it was generated to the target security gateway where it is going to be used. In order to protect the key it can be encrypted with 3DES using a symmetric transport key derived from a cryptographically strong passphrase.
Because of the weak security, key files protected by single DES will not be accepted by Pluto!!!
Once on the security gateway the private key can either be permanently unlocked so that it can be used by Pluto without having to know a passphrase
or as an option the key file can remain secured. In this case the passphrase unlocking the private key must be added after the pathname in /etc/ipsec.secrets
Some CAs distribute private keys embedded in a PKCS#12 file. Since Pluto is not able yet to read this format directly, the private key part must first be extracted using the command
if the key file moonKey.pem is to be secured again by a passphrase, or
if the private key is to be stored unlocked.
On a VPN gateway you would want to put the passphrase protecting the private
key file right into /etc/ipsec.secrets as described in the previous paragraph,
so that the gateway can be booted in unattended mode. The risk of keeping
unencrypted secrets on a server can be minimized by putting the box into a
locked room. As long as no one can get root access on the machine the private
keys are safe.
Since strongSwan is usually started during the boot process, no interactive console windows is available which can be used by Pluto to prompt for the passphrase. This must be initiated by the user by typing
which actually is an alias for the existing command
and which causes the prompt
to appear. If the passphrase was correct and the private key file could be successfully decrypted then
results. Otherwise the prompt
will give you another try. Entering a carriage return will abort the passphrase prompting.
strongSwan supports multiple private keys. Because the connections defined in ipsec.conf can find the correct private key on the basis of the public key contained in the certificate which assigned via the leftcert parameter, private key definitions without specific IDs can be used
Besides the definition of IPsec connections the ipsec.conf file can also be used to configure a few properties of the certification authorities needed to establish the X.509 trust chains. The following example shows the parameters that are currently available:
In a similar way as conn sections are used for connection definitions, an arbitrary number of optional ca sections define the basic properties of CAs.
Each ca section is named with a unique label
The only mandatory parameter is
which points to the CA certificate which usually resides in the default directory /etc/ipsec.d/cacerts/ but could also be retrieved via an absolute path name. If the CA certificate is stored on a smartcard then the notation
can be used. The selection of smartcard slots is described in more detail in section 8.1. From the certificate the CA's distinguished name and the serial number is extracted. If an optional subjectKeyAuthentifier is present then it can be used to uniquely identify consecutive generations of CA certificates carrying the same distinguished name. The OCSP URI
allows to define an individual OCSP server per CA. Also up to two additional CRL distribution points (CDPs) can be defined
which are added to any CDPs already present in the received certificates themselves. The last parameter
can be used to fill in the actual server name in LDAP CDPs where the host is missing as e.g. in the crluri2 above. In future releases this ldaphost parameter might be used to retrieve user, host and attribute certificates.
With the auto=add statement the ca definition is automatically loaded into Pluto during system startup. Setting auto=ignore will ignore the ca section. Additional ca definitions can be loaded from ipsec.conf during runtime with the command
deletes the labeled ca entry. And finally the command
first deletes the old definition in Pluto's memory and then loads the updated version from ipsec.conf. Any parameters which appear in several ca definitions can be put in a common ca %default section
Defining a smartcard-based connection in ipsec.conf is easy:
In most cases there is a single smartcard reader or cryptotoken and only one RSA private key safely stored on the crypto device. Thus usually the entry
which stands for the full notation
is sufficient where the first certificate/private key object enumerated by PKCS#11 module is used. If several certificate/private key objects are present then the nth object can be selected using
gives an overview over all certifcate objects made available by the PKCS#11 module. CA certificates are automatically available as trust anchors without the need to copy them into the /etc/ipsec.d/cacerts/ directory first.
As an alternative the certificate ID and/or the slot number defined by the PKCS#11 standard can be specified using the notation
will look in all available slots for ID 0x50 starting with the first slot (usually slot 0) whereas
will directly check slot 4 (which is usually the first slot on the second reader/token when using the OpenSC library) for a key with ID 0x50.
Since the smartcard signing operation needed to sign the hash with the
RSA private key during IKE Main Mode is protected by a PIN code,
the secret PIN must be made available to Pluto.
or with the general notation
On a personal notebook computer that could get stolen, you wouldn't want to store your PIN in ipsec.secrets.
Thus the alternative form
will prompt you for the PIN when you start up the first IPsec connection using the command
The ipsec up command calls the whack function which in turn communicates with Pluto over a socket. Since the whack function call is executed from a command window, Pluto can prompt you for the PIN over this socket connection. Unfortunately roadwarrior connections which just wait passively for peers cannot be initiated via the command window:
But if there is a corresponding entry
in ipsec.secrets, then the standard command
or the alias
can be used to enter the PIN code for this connection interactively. The command
can be executed at any time to check the current status of the PIN code[s].
Smartcard readers with an integrated PIN pad offer an increased security level because the PIN entry cannot be sniffed on the host computer e.g. by a surrepticiously installed key logger. In order to tell pluto not to prompt for the PIN on the host itself, the entry
can be used in ipsec.secrets. Because the key pad does not cache the PIN in the smartcard reader, it must be entered for every PKCS #11 session login. By default pluto does a session logout after every RSA signature. In order to avoid the repeated entry of the PIN code during the periodic IKE main mode rekeyings, the following parameter can be set in the config setup section of ipsec.conf:
The default setting is pkcs11keepstate=no.
strongSwan's smartcard solution is based on the PKCS#15 "Cryptographic Token Information Format Standard" fully supported by OpenSC library functions. Using the command
a fresh PKCS#15 file structure is created on a smartcard or cryptotoken. With the next command
a secret PIN code with auth-id 1 is stored in an unretrievable location on
the smart card. The PIN will protect the RSA signing operation. If the PIN
is entered incorrectly more than three times then the smartcard will be locked
and the PUK code can be used to unlock the card again.
By default the PKCS#15 smartcard record will be assigned the ID 45.
Using the --id option, multiple key records can be stored on a smartcard.
The pkcs15-tool can now be used to verify the contents of the smartcard.
If everything is ok then you are ready to use the generated PKCS#15 structure with strongSwan.
will enable the proxy mode that is disabled by default.
The default settings for inbase and outbase is hexadecimal. Thus the simplest call has the form
and the returned result might be a decrypted 128 bit AES key
The leading three characters represent the return code of the whack channel with 000 signifying that no error has occured. Here is another example showing the use of the inbase and outbase attributes
ipsec scdecrypt m/ewDnTs0k...woE= --inbase base64 --outbase text
where the result has the form
with --keyid %smartcard#1 being the default. If supported by the smartcard and PKCS#11 library RSA encryption can be used with the notation
with the example
ipsec scencrypt "This is a secret" --inbase ascii --outbase 64
returning the expected output
A strongSwan to strongSwan connection is symmetrical. Any of the four defined ID types can be used, even different types on either end of the connection, although this wouldn't make much sense.
Use the file peerCert.p12 to import PGPnet's X.509 certificate, the CA certificate, plus the encrypted private key in binary PKCS#12 format into the PGPkey tool. You will be prompted for the passphrase securing the private key.
Use the file myCert.pem to import the X.509 certificate of the strongSwan security gateway into the PGPkey tool. The PGPkeyTool does not accept X.509 certificates in binary DER format, so it must be imported in base64 format:
Make sure that there is no human-readable listing of the X.509 certificate in front of the line
otherwise PGPnet will refuse to load the *.PEM file. Any surplus lines can either be deleted by loading the certificate into a text editor or you can apply the command
to achieve the same effect.
With authentication based on X.509 certificates, PGPnet always sends the ID type DER_ASN1_DN, therefore rightid in the connection definition of the strongSwan security gateway must be an ASN.1 distinguished name.
In the receiving direction PGPnet accepts all four ID types from strongSwan.
SafeNet/Soft-PK and SafeNet/Soft-Remote can be configured to send their identity either as DER_ASN1_DN, IPV4_ADDR, FQDN, or USER_FQDN. In the receiving direction SafeNet/Soft-PK or SafeNet/Soft-Remote accept all four ID types coming from strongSwan.
SSH Sentinel sends its identity as DER_ASN1_DN if the subjectAltName field of its certificate is empty. If a subjectAltName field is present, then the corresponding type IPV4_ADDR, FQDN, or USER_FQDN is automatically chosen. With several subjectAltName entries, the precedence of the different ID types is not quite clear. In the receiving direction SSH Sentinel accepts all four ID types from strongSwan.
Windows 2000 and Windows XP always send the ID type DER_ASN1_DN, therefore rightid in the connection definition of the strongSwan security gateway must be an ASN.1 distinguished name.
In the receiving direction Windows 2000 or Windows XP accept all four ID types from strongSwan.
strongSwan offers the following monitoring functions:
lists all IKE and ESP cryptographic algorithms that are currently registered with strongSwan.
The listing has the following form:
List of registered IKE Encryption Algorithms:
#3 OAKLEY_BLOWFISH_CBC, blocksize: 64,
List of registered IKE Hash Algorithms:
#1 OAKLEY_MD5, hashsize: 128
List of registered IKE DH Groups:
#2 OAKLEY_GROUP_MODP1024, groupsize: 1024
List of registered ESP Encryption Algorithms:
#3 ESP_3DES, blocksize: 64, keylen: 168-168
List of registered ESP Authentication Algorithms:
#1 AUTH_ALGORITHM_HMAC_MD5, keylen: 128-128
lists all public keys currently installed in the chained list of public keys. These keys were statically loaded from ipsec.conf or aquired either from received certificates or retrieved from secure DNS servers using opportunistic mode.
The public key listing has the following form:
Feb 11 14:40:18 2005, 2048 RSA Key AwEAAa+uL, until Sep 09 13:17:25 2009 ok
It consists of
A public key can be associated with several IDs, e.g. using class=subjectAltNames in certificates and an ID can possess several public keys, e.g. retrieved from a secure DNS server.The command
ipsec listcerts [--utc]lists all local certificates, both strongSwan's own and those of trusted peer loaded via leftcert and rightcert, respectively. The output has the form
Feb 11 13:36:47 2005, count: 4
lists all CA certificates that have been loaded from /etc/ipsec.d/cacerts.
The output has the form
Feb 11 13:36:52 2005, count: 1
lists all Authorization Authority certificates that have been loaded from the directory /etc/ipsec.d/aacerts/.
The output has the form
Dec 20 13:29:55 2004, count: 1
lists all OCSP signer certificates that have either been loaded from /etc/ipsec.d/ocspcerts/ or have been received include in the OCSP server response.
The output has the form
Feb 09 22:56:17 2005, count: 1
lists all X.509 attribute certificates that have been loaded from the directory /etc/ipsec.d/acerts/.
The output has the form
Dec 20 13:29:56 2004
lists all group attributes either defined in right|leftgroups statements in ipsec.conf or contained in loaded X.509 attribute certificates.
The output has the form
Dec 20 13:29:55 2004, count: 4
lists the properties defined by the ca sections in ipsec.conf.
Jun 08 22:31:37 2004, "strongswan"
lists all CRLs that have been loaded from /etc/ipsec.d/crls.
The output has the form
Feb 11 13:37:00 2005, revoked certs: 1
lists all the contents of the OCSP response cache. The output has the form
issuer: 'C=CH, O=Linux strongSwan,
CN=strongSwan Root CA'
lists all smartcard records that are currently in use by pluto. The output has the form
Aug 17 16:47:59 2005, #1, count: 6
with pkcs11keepstate = no and
Aug 17 16:47:59 2005, #1, count: 6
with pkcs11keepstate = yes and shows
is equivalent to
strongSwan makes the following environment variables available in the updown script indicated by the leftupdown parameter:
(1) $PLUTO_PEER_ID/$PLUTO_MY_ID contain the IDs of the two ends of an established connection. In our examples these correspond to the strings defined by rightid and leftid, respectively.
(2) $PLUTO_PEER_PROTOCOL/$PLUTO_MY_PROTOCOL contain the protocol defined by the rightprotoport and leftprotoport options, respectively. Both variables contain the same protocol value. The variables take on the value '0' if no protocol has been defined.
(3) $PLUTO_PEER_PORT/$PLUTO_MY_PORT contain the ports defined by the rightprotoport and leftprotoport options, respectively. The variables take on the value '0' if no port has been defined.
(4) $PLUTO_PEER_CA contains the distinguished name of the CA that issued
the peer's certificate.
Starting with strongswan-2.7.0, the default _updown script automatically inserts and deletes dynamic iptables firewall rules upon the establishment or teardown, respectively, of an IPsec security association. This new feature is activated with the line
and can be used when the following prerequisites are fulfilled:
If you define a local client subnet with a netmask larger than /32 behind the gateway then the automatically inserted FORWARD iptables rules will not allow to access the internal IP address of the host although it is part of the client subnet definition. If you want additional INPUT and OUTPUT iptables rules to be inserted, so that the host itself can be accessed then add the following line:
The script also features a logging facility which will register the creation (+) and the expiration (-) of each successfully established VPN connection in a special syslog file in the following concise and easily readable format:
Jul 19 18:58:38 moon vpn:
If you are using a Linux 2.6 kernel older than 2.6.16 or an iptables version older than 1.3.5 then the IPsec policy matching rules will not be available. In order to make sure that only tunneled packets are accepted, a mark can be set on incoming ESP packets. This "ESP" mark will be retained on the decapsulated packet so that iptables rules inserted by the updown script can check on the presence of this mark. For this purpose the template located in
can be used Store a copy of _updown_espmark e.g. in /etc/ipsec.updown and load the script with the line
In addition for the dynamic updown script to work, the following static iptables rule must be applied:
if NAT traversal is employed. www.freeswan.org does public key authentication base on raw RSA public keys that are directly defined in /etc/ipsec.conf
When version 1.x of FreeS/WAN receives a certificate request (CR), it immediately drops the negotiation because it does not know how to answer the request. As a workaround strongSwan does not send a CR if the RSA key has been statically loaded using rightrsasigkey. A problem remains when roadwarriors initiate a connection. Since strongSwan does not know the identity of the initiating peer in advance, it will always send a CR, causing the rupture of the IKE negotiation if the peer is a standard FreeS/WAN host. To circumvent this problem the configuration parameter nocrsend can be set in the config setup section of /etc/ipsec.conf:
With this entry no certificate request is sent in any connection. The default setting is nocrsend=no.
strongSwan also supports RSA-based authentication using OpenPGP certificates and OpenPGP V3 fingerprints employed as KEY_ID identifiers.
OpenPGP certificates containing RSA public keys can now directly be loaded in ASCII armored PGP format using the leftcert and rightcert parameters in /etc/ipsec.conf:
The peer certificate must be stored locally (the default directory is /etc/ipsec.d/certs)/ since currently no trust can be established for PGP certificates received from a peer via the IKE protocol.
PGP private keys in unencrypted form can now directly be loaded in ASCII armored PGP format via an entry in /etc/ipsec.secrets:
: RSA gatewayKey.asc
Existing IDEA-encrypted legacy RSA private keys can be unlocked with the help of GnuPG and the IDEA extension (see http://www.gnupg.org/gph/en/pgp2x.html) using the following commands
The command ipsec listcerts shows all loaded PGP certificates in the following format:
Aug 28 09:51:55 2002, count: 1
The entries are
PGPnet configured to work with OpenPGP certificates aborts the IKE negotiation when it receives a X.509 certificate. Therefore it is recommended (mandatory for roadwarrior connections) to set
in the config setup section of /etc/ipsec.conf.
strongSwan supports the following suite of encryption and authentication algorithms for both IKE and ESP payloads.
NOTE: For IKE the SHA-1 algorithm is denoted by sha
The cryptographic IKE algorithms listed above are a fixed part of the strongSwan distribution. Particular algorithms can be added or removed in the programs/pluto/alg directory.
The cryptographic ESP algorithms listed above are a fixed part of the strongSwan distribution. If your Linux 2.4 or 2.6 kernel includes the CryptoAPI then additional ESP algorithms can be added or deleted as kernel modules.
The IKE and ESP cryptographic algorithms to be proposed to the peer as an initiator can be specified on a per connection basis in the form
or if you are more paranoid
If the ike and esp configuration parameters are missing in ipsec.conf, then the default settings
are implicitly assumed. The 3DES encryption algorithm and the MD5 and SHA-1 hash algorithms are hardcoded into strongSwan and cannot be removed.
If Perfect Forward Secrecy (PFS is desired), then a PFS group can beoptionally specified:
If the pfs parameter is missing, then pfs=yes is assumed by default. This means that in order to disable PFS, this must be done explicitly by setting pfs=no.
If the pfsgroup parameter is missing, then the default is
The ike and esp parameters are used to formulate one or several transform proposals that are sent to the peer if strongSwan is the initiator.
Attention! As a responder the first proposal received from the peer is accepted that is supported the by one of the registered algorithms listed by the command
If strongSwan as a responder wants to restrict the allowed cipher suites, then the '!' strict flag (exclamation mark) can be used to do so. The configuration
will only permit the listed algorithms defined above. All other methods are rejected even if strongSwan would be able to support them.
Currently please refer to README.NAT-Traversal document in the strongSwan distribution.
strongSwan implements the RFC 3706 Dead Peer Detection (DPD) keep-alive scheme. If an established IPsec SA has been idle (i.e. without any traffic) for N seconds (dpddelay=N) then strongSwan sends a "hello" message (R_U_THERE) and if the the peer supports DPD then it replies with an acknowledge message (R_U_THERE_ACK). If no response is received, then the R_U_THERE messages are repeated until a DPD timeout of M seconds (dpdtimeout=M) has elapsed. If still no traffic or R_U_THERE_ACK packets have been received, then the peer is declared to be dead and all SAs belonging to a common Phase 1 SA are deleted.
DPD support is tuneable on a per connection basis by using the dpdaction, dpddelay and dpdtimeout directives:
In the first example dpdaction=clear activates the DPD mechanism under the condition that the peer supports RFC 3706. The values dpddelay=30s and dpdtimeout=120s are assumed by default in the absence of these parameters, so that during idle periods an R_U_THERE packet is sent every 30 seconds. If no traffic or a no R_U_THERE_ACK packet is received from the peer within a 120 second time span, the peer will be declared dead and all SAs and associated eroutes will be cleared.
In the second example R_U_THERE packets are sent every 60 seconds and the
parameter setting dpdaction=hold will put the eroute of the ruptured connection
into a %trap state, so that when new outgoing traffic will occur, the
It is recommended to use dpdaction=hold for statically defined connections and dpdaction=clear for dynamic roadwarrior connections. The default value is dpdaction=none, which disables DPD.
The IKE Mode Config protocol <draft-ietf-ipsec-isakmp-mode-cfg-04.txt> allows the dynamic assignment of virtual IP addresses and optional DNS and WINS server information to IPsec clients. As a default the "Mode Config Pull Mode" is used where the client actively sends a Mode Config request to the server in order to obtain a virtual IP. The server answers with a Mode Config reply message containing the requested information.
Client side configuration (carol):
Server side configuration (moon):
The wildcard %modeconfig used in the leftsourceip parameter of the client will trigger a Mode Config request. Currently the server will return the virtual IP address defined by the rightsourceip parameter. In the future an LDAP-based lookup mechanism will be supported.
as part of the connection definition in ipsec.conf. The default value for the Mode Config mode is modeconfig=pull.
or with RSA signatures
in the connection definition, correspondingly. strongSwan can act either as an XAUTH client with
or as an XAUTH server with
with xauth=client being the default value. strongSwan integrates a default implementation where the XAUTH user credentials are stored on both the server and the client in the /etc/ipsec.secrets file, using the syntax
The client must not have more than one XAUTH entry whereas the server can contain an unlimited number of user credentials in ipsec.secrets. Either the prompting on the client side or the verification of the user credentials on the server side can be implemented as a customized XAUTH dynamic library module. The corresponding library interface is defined by the pluto/xauth.h header file.
FreeS/WAN base system:
Copyright © 1999-2004
NAT-Traversal, ipsec starter, delete SA and notification messagages:
Copyright © 2002-2003, Mathieu Lafon
Additional cryptoalgorithms (AES, etc.):
Copyright © 2002-2003, JuanJo Ciarlante
Dead Peer Detection:
Copyright (c) 2002-2004
Porting to Linux 2.6 kernel:
Copyright © 2003, Herbert XU
Dynamic CRL fetching:
IKE Mode Config and XAUTH protocols:
Virtual IP and source routing:
Copyright © 2003, Tuomo Soini
Port and Protocol selectors for outbound
PGPnet-RSA parts of patch:
Copyright © 2000, Kai Martius
X.509, OCSP and smartcard functionality:
Copyright © 2000, Andreas Hess, Patric Lichtsteiner, Roger Wegmann
This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version. See http://www.fsf.org/copyleft/gpl.txt.
This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.