The Auth Module

See description of parameter auth_checks_in_dlg.

See description of parameter auth_checks_in_dlg.

These three module parameters control which optional integrity checks will be performed on the SIP message carrying digest response during digest authentication. auth_check_register controls integrity checks to be performed on REGISTER messages, auth_checks_no_dlg controls which optional integrity checks will be performed on SIP requests that have no To header field or no To tag (in other words the requests either establishing or outside dialogs). auth_checks_in_dlg controls which integrity checks will be performed on SIP requests within dialogs, such as BYE or re-INVITE. The default value for all three parameters is 0 (old behaviour, no extra checks). The set of integrity checks that can be performed on REGISTERs is typically different from sets of integrity checks that can be performed for other SIP request types, hence we have three independent module parameters.

Without the extra checks the nonce will protect only against expired values. Some reply attacks are still possible in the expire "window". A possible workaround is to always force qop authentication and always check the uri from the authorization header, but this would not work if an upstream proxy rewrites the uri and it will also not work with a lot of UA implementations.

In this case the nonce value will be used only to hold the expire time (see nonce_expire) and an MD5 over it and some secret (the MD5 is used to make sure that nobody tampers with the nonce expire time).

When the extra checks are enabled, the nonce will include and extra MD5 over the selected part/parts of the message (see below) and some other secret. This will be used to check if the selected part of the message is the same when an UA tries to reuse the nonce, thus protecting or severely limiting reply attacks.

The possible flag values for all three parameters are:

For example setting auth_checks_register to 3 would check if the callid or the request uri changed from the REGISTER message for which the original nonce was generated (this would allow nonce reuse only within the same UA and for the expire time). Note that enabling the extra checks will limit nonce caching by UAs, requiring extra challenges and roundtrips, but will provide much better protection.

When the secret parameter is set and the extra checks are enabled, the first half of the secret will be used for the expire time MD5 and the other half for the extra checks MD5, so make sure you have a long secret (32 chars or longer are recommended).

...
# For REGISTER requests we hash the Request-URI, Call-ID, and source IP of the
# request into the nonce string. This ensures that the generated credentials
# cannot be used with another registrar, user agent with another source IP
# address or Call-ID. Note that user agents that change Call-ID with every
# REGISTER message will not be able to register if you enable this.
modparam("auth", "auth_checks_register", 11)

# For dialog-establishing requests (such as the original INVITE, OPTIONS, etc)
# we hash the Request-URI and source IP. Hashing Call-ID and From tags takes
# some extra precaution, because these checks could render some UA unusable.
modparam("auth", "auth_checks_no_dlg", 9)

# For mid-dialog requests, such as re-INVITE, we can hash source IP and
# Request-URI just like in the previous case. In addition to that we can hash
# Call-ID and From tag because these are fixed within a dialog and are
# guaranteed not to change. This settings effectively restrict the usage of
# generated credentials to a single user agent within a single dialog.
modparam("auth", "auth_checks_in_dlg", 15)

...

If set, enable qop for challenges: each challenge will include a qop parameter. This is the recommended way, but some older non rfc3261 compliant UAs might get confused and might not authenticate properly if qop is enabled.

Enabling qop together with nonce_count will provide extra-security (protection against replay attacks) while still allowing credentials caching at the UA side and thus not compromising performance.

The possible values are: "auth", "auth-int" and "" (unset).

The default value is not-set ("").

See also: nonce_count.

...
modparam("auth", "qop", "auth")   # set qop=auth
...

If enabled the received nc value is remembered and checked against the older value (for a successful authentication the received nc must be greater then the previously received one, see rfc2617 for more details). This will provide protection against replay attacks while still allowing credentials caching at the UA side.

It depends on qop being enabled (if qop is not enabled, the challenges won't include qop and so the UA will probably not include the qop or nc parameters in its response).

If a response doesn't include qop or nc (for example obsolete UAs that don't support them) the response will be checked according to the other enabled nonce checks, in this order: one_time_nonce and auth_check_*. If a response includes nc only the normal nonce_expire checks and the nonce_count checks will be performed, all the other checks will be ignored.

The nonce_count checks work by tracking a limited number of nonces. The maximum number of tracked nonces is set using the nc_array_size or nc_array_order parameters. If this number is exceeded, older entries will be overwritten. As long as the maximum rate of challengeable messages per average response time is lower then nc_array_size, the nonce_count checks should work flawlessly. For optimum performance (maximum reuse of cache credentials) nc_array_size divided by nid_pool_no should be lower then the message rate multiplied by the desired nonce_expire.

The maximum accepted nc value is 255. If nc becomes greater then this, the nonce will be considered stale and the UA will be re-challenged.

Note: nonce_count should be enabled only in stateful mode (a transaction should be created prior to the authentication check to absorb possible retransmissions and all the replies should be sent statefuly, using t_reply()). If nonce_count and the authentication checks are used in the stateless mode then all retransmissions will be challenged.

The default value is 0 (off).

See also: qop, nc_array_size, nc_array_order, nid_pool_no, nonce_expire. one_time_nonce.

...
modparam("auth", "nonce_count", 1) # enable nonce_count support
modparam("auth", "qop", "auth")    # enable qop=auth

....
route{
...
	# go stateful and catch retransmissions
	if (!t_newtran()) {
	    xlog("L_NOTICE", "Failed to create new transaction\n");
 	    drop;
	};
	if (method=="REGISTER"){
		if (!www_authenticate("test", "credentials")){
			# reply must be sent with t_reply because the 
			# transaction is already created at this point 
			# (we are in "stateful" mode)
			if ($? == -2){
				t_reply("500", "Internal Server Error");
			}else if ($? == -3){
				t_reply("400", "Bad Request");
			}else{
				if ($digest_challenge) 
					append_to_reply("%$digest_challenge");
				t_reply("401", "Unauthorized");
			}
			drop;
		}
		if (!save_noreply("location")) {
			t_reply("400", "Invalid REGISTER Request");
			drop;
		}
		append_to_reply("%$contact");
		t_reply("$code", "$reason"); # no %, avps are used directly 
		drop;
	}else{
		if (!proxy_authenticate("my_realm", "credentials")){
			if ($? == -2){
				t_reply("500", "Internal Server Error");
			}else if ($? == -3){
				t_reply("400", "Bad Request");
			}else{
				if ($digest_challenge) 
					append_to_reply("%$digest_challenge");
				t_reply("401", "Unauthorized");
			}
			drop;
		}
	}
...

If set to 1 nonce reuse is disabled: each nonce is allowed only once, in the first reponse to a challenge. All the messages will be challenged, even retransmissions. Stateful mode should be used, to catch retransmissions before the authentication checks (using t_newtran() before the authentication checks and sending all the replies with t_reply()).

one_time_nonce provides enhanced replay protections at the cost of invalidating UA side credentials caching, challenging every message (and thus generating extra messages and extra round-trips) and requiring stateful mode. In general qop and nonce_count should be prefered (if possible) with fallback to auth_checks_*. Due to the disadvantages listed above, one_time_nonce should be used only if the extra checks provided by auth_checks_register, auth_checks_no_dlg and auth_checks_in_dlg are deemed insufficient for a specific setup.

Compared to nonce_count, one_time_nonce provides the same protection, but at a higher message cost. The only advantages are that it works with user agents that do not support qop and nc and that it uses less memory for the same supported number of maximum in-flight nonces (by a factor of 8). one_time_nonce can be used as fallback from nonce_count, when the UA doesn't support nc (it happens automatically when both of them are enabled).

Like nonce_count, one_time_nonce works by tracking a limited number of nonces. The maximum number of tracked nonces is set using the otn_in_flight_no or otn_in_flight_order parameters. If this number is exceeded, older entries will be overwritten. As long as the maximum rate of challengeable messages per average response time is lower then otn_in_flight_no, the one_time_nonce checks should work flawlessly.

The default value is 0 (off).

See also: otn_in_flight_no, otn_in_flight_order, nid_pool_no and nonce_count.

...
modparam("auth", "one_time_nonce", 1)
# Note: stateful mode should be used, see the nonce_count example
...

Controls the number of partitions for the nonce_count and one_time_nonce arrays (it's common to both of them to reduce the nonce size).

Instead of using single arrays for keeping nonce state, these arrays can be divided into more partitions. Each ser process is assigned to one of these partitions, allowing for higher concurrency on multi-CPU machines. Besides increasing performance, increasing nid_pool_no has also a negative effect: it could decrease the maximum supported in-flight nonces in certain conditions. In the worst case, when only one ser process receives most of the traffic (e.g. very busy tcp connection between two proxies), the in-flight nonces could be limited to the array size (nc_array_size for nonce_count or otn_in_flight_no for one_time_nonce) divided by the partitions number (nid_pool_no). However for normal traffic, when the process receiving a message is either random or chosen in a round-robin fashion the maximum in-flight nonces number will be very little influenced by nid_pool_no (the messages will be close to equally distributed to processes using different partitions).

nid_pool_no value should be one of: 1, 2, 4, 8, 16, 32 or 64 (the maximum value is 64 and all values should be of the form 2^k or else they will be rounded down to 2^k).

The default value is 1.

See also: nonce_count, one_time_nonce, nc_array_size and otn_in_flight_no.

...
modparam("auth", "nid_pool_no", 4)
...

Maximum number of in-flight nonces for nonce_count. It represents the maximum nonces for which state will be kept. When this number is exceeded, state for the older nonces will be discarded to make space for new ones (see nonce_count for more details).

The value should be of the form 2^k. If it's not it will be rounded down to 2^k (for example a value of 1000000 will be rounded down to 2^19=524288). nc_array_order can be used to directly specify the power of 2 (e.g. nc_array_order set to 20 is equivalent to nc_array_size set to 1048576).

The memory used to keep the nonce state will be nc_array_size in bytes.

The default value is 1048576 (1M in-flight nonces, using 1Mb memory).

See also: nonce_count and nid_pool_no.

...
modparam("auth", "nc_array_size", 4194304)   # 4Mb
...

Equivalent to nc_array_size, but instead of directly specifying the size, its value is the power at which 2 should be raised (log2(nc_array_size)).

nc_array_size = 2^nc_array_order. For more details see nc_array_size.

The default value is 20 (1M in-flight nonces, using 1Mb memory).

See also: nonce_count, nc_array_size and nid_pool_no.

...
modparam("auth", "nc_array_order", 22)   # 4Mb
...

Maximum number of in-flight nonces for one_time_nonce. It represents the maximum number of nonces remembered for the one-time-nonce check. When this number is exceeded, information about older nonces will be discarded and overwritten with information about the new generated ones (see one_time_nonce for more details).

The value should be of the form 2^k. If it's not it will be rounded down to 2^k (for example a value of 1000000 will be rounded down to 2^19=524288). otn_in_flight_no can be used to directly specify the power of 2 (e.g. otn_in_flight_order set to 19 is equivalent to otn_in_fligh_number set to 524288).

The memory used to keep the nonce information will be the otn_in_flight_no divided by 8 (only 1 bit of state is kept per nonce).

The default value is 1048576 (1M in-flight nonces, using 128Kb memory).

See also: one_time_nonce and nid_pool_no.

...
modparam("auth", "otn_in_flight_no", 8388608)   # 8 Mb (1Mb memory)
...

Equivalent to otn_in_flight_no, but instead of directly specifying the size, its value is the power at which 2 should be raised (log2(otn_in_flight_no)).

otn_in_flight_no = 2^otn_in_flight_order. For more details see otn_in_flight_order.

The default value is 20 (1M in-flight nonces, using 128Kb memory).

See also: one_time_nonce, otn_in_flight_no and nid_pool_no.

...
modparam("auth", "otn_in_flight_order", 23)   # 8 Mb (1Mb memory)
...

Secret phrase used to calculate the nonce value used to challenge the client for authentication.

If you use multiple servers in your installation, and would like to authenticate on the second server against the nonce generated at the first one its necessary to explicitly set the secret to the same value on all servers. However, as the use of a shared (and fixed) secret as nonce is insecure, it is much better is to stay with the default. Any clients should send the authenticated request to the server that issued the challenge.

Default value is randomly generated string.

...
modparam("auth", "secret", "johndoessecretphrase")
...

Nonces have limited lifetime. After a given period of time nonces will be considered invalid. This is to protect replay attacks. Credentials containing a stale nonce will be not authorized, but the user agent will be challenged again. This time the challenge will contain stale parameter which will indicate to the client that it doesn't have to disturb user by asking for username and password, it can recalculate credentials using existing username and password.

The value is in seconds and default value is 300 seconds.

...
modparam("auth", "nonce_expire", 600)   # Set nonce_expire to 600s
...

Maximum difference in seconds between a nonce creation time and the current time, if the nonce creation time appears to be in the future.

In some cases, like shortly after a system time backward adjustment or when the current proxy is part of a cluster which is not time-synchronized, it's possible to receive a nonce with creation time in the future. In this case if the difference is greater then nonce_auth_max_drift seconds, consider the nonce stale and re-challenge (otherwise after a dramatic time change backwards, it might happen that some previously generated nonces will be valid for too much time).

The default value is 3 seconds

See also: nonce_expire.

...
modparam("auth", "nonce_auth_max_drift", 1)   # set max drift to 1 s
...

If set to 1, www_challenge() and proxy_challenge() functions send reply statelessly no matter if transaction exists or not. If set to 0 (default), reply is sent statefully if transaction exists and stelelessly otherwise.

...
modparam("auth", "force_stateless_reply", 1)
...

Prefix to be automatically strip from realm. As an alternative to SRV records (not all SIP clients support SRV lookup), a subdomain of the master domain can be defined for SIP purposes (like sip.mydomain.net pointing to same IP address as the SRV record for mydomain.net). By ignoring the realm_prefix “sip.”, at authentication, sip.mydomain.net will be equivalent to mydomain.net .

Default value is empty string.

modparam("auth", "realm_prefix", "sip.")

If set to 1, pv_auth_check() uses domain parts of the URIs to check user identity.

...
modparam("auth", "use_domain", 1)
...

This function removes previously authorized credential headers from the message being processed by the server. That means that the downstream message will not contain credentials there were used by this server. This ensures that the proxy will not reveal information about credentials used to downstream elements and also the message will be a little bit shorter. The function must be called after www_authorize, proxy_authorize, www_authenticate or proxy_authenticate.

...
if (www_authenticate("realm", "subscriber")) {
    consume_credentials();
};
...

This function returns true of the request has Autorization or Proxy-Authorization header with provided realm. The parameter can be string with pseudo-variables.

...
if (has_credentials("myrealm")) {
    ...
}
...

The function challenges a user agent. It will generate a WWW-Authorize header field containing a digest challenge, it will put the header field into a response generated from the request the server is processing and send the reply. Upon reception of such a reply the user agent should compute credentials and retry the request. For more information regarding digest authentication see RFC2617. See module parameter force_stateless_reply regarding sending of the reply.

Meaning of the parameters is as follows:

This function can be used from REQUEST_ROUTE.

...
if (!www_authenticate("$td", "subscriber")) {
	www_challenge("$td", "1");
}
...

The function challenges a user agent. It will generate a Proxy-Authorize header field containing a digest challenge, it will put the header field into a response generated from the request the server is processing and send the reply. Upon reception of such a reply the user agent should compute credentials and retry the request. For more information regarding digest authentication see RFC2617. See module parameter force_stateless_reply regarding sending of the reply.

Meaning of the parameters is the same as for function www_challenge(realm, flags)

This function can be used from REQUEST_ROUTE.

...
if (!proxy_authenticate("$fd", "subscriber")) {
	proxy_challenge("$fd", "1");
};
...

The function challenges a user agent for authentication. It combines the functions www_challenge() and proxy_challenge(), by calling internally the first one for REGISTER requests and the second one for the rest of other request types.

Meaning of the parameters the same as for function www_challenge(realm, flags)

This function can be used from REQUEST_ROUTE.

...
if (!auth_check("$fd", "subscriber", "1")) {
	auth_challenge("$fd", "1");
};
...

The function verifies credentials according to RFC2617. If the credentials are verified successfully then the function will succeed and mark the credentials as authorized (marked credentials can be later used by some other functions). If the function was unable to verify the credentials for some reason then it will fail and the script should call www_challenge which will challenge the user again.

Negative codes may be interpreted as follows:

Meaning of the parameters is as follows:

When challenge header is built and stored in avp, append_to_reply() and the sl reply functions can be used to send appropriate SIP reply to challenge for authentication.

This function can be used from REQUEST_ROUTE.

...
if (!pv_www_authenticate("$td", "123abc", "0")) {
	www_challenge("$td", "1");
};
...

The function verifies credentials according to RFC2617. If the credentials are verified successfully then the function will succeed and mark the credentials as authorized (marked credentials can be later used by some other functions). If the function was unable to verify the credentials for some reason then it will fail and the script should call proxy_challenge which will challenge the user again. For more about the negative return codes, see the above function.

Meaning of the parameters is the same as for pv_www_authenticate(realm, passwd, flags)

This function can be used from REQUEST_ROUTE.

...
$avp(password)="xyz";
if (!pv_proxy_authenticate("$fd", "$avp(password)", "0")) {
	proxy_challenge("$fd", "1");
};
...

The function combines the functionalities of pv_www_authenticate and pv_proxy_authenticate, first being exectuted if the SIP request is a REGISTER, the second for the rest.

Meaning of the first three parameters is the same as for pv_www_authenticate(realm, passwd, flags).

Parameter checks can be used to control the behaviour of the function. If it is 1, then the function will check to see if the authentication username matches either To or From header username, a matter of whether it is for a REGISTER request or not. The parameter may be a pseudo variable.

This function can be used from REQUEST_ROUTE.

...
$avp(password)="xyz";
if (!pv_auth_check("$fd", "$avp(password)", "0", "1")) {
	proxy_challenge("$fd", "1");
};
...

Build WWW-Authentication header and set the resulting value in 'pvdest' pseudo-variable parameter.

Meaning of the realm and flags parameters is the same as for pv_www_authenticate(realm, passwd, flags)

This function can be used from ANY_ROUTE.

...
if (auth_get_www_authenticate("$fd", "0", "$var(wauth)")) {
	xlog("www authenticate header is [$var(wauth)]\n");
}
...