Initialization

To use the library, an application must first call rad_auth_open(), rad_acct_open() or rad_server_open() to obtain a struct rad_handle *, which provides the context for subsequent operations. The former function is used for RADIUS authentication and the latter is used for RADIUS accounting. Calls to rad_auth_open(), rad_acct_open() and rad_server_open() always succeed unless insufficient virtual memory is available. If the necessary memory cannot be allocated, the functions return NULL. For compatibility with earlier versions of this library, rad_open() is provided as a synonym for rad_auth_open().

Before issuing any RADIUS requests, the library must be made aware of the servers it can contact. The easiest way to configure the library is to call rad_config(). rad_config() causes the library to read a configuration file whose format is described in radius.conf(5). The pathname of the configuration file is passed as the file argument to rad_config(). This argument may also be given as NULL, in which case the standard configuration file /etc/radius.conf is used. rad_config() returns 0 on success, or -1 if an error occurs.

The library can also be configured programmatically by calls to rad_add_server() or rad_add_server_ex(). rad_add_server() is a backward compatible function, implemented via rad_add_server_ex(). The host parameter specifies the server host, either as a fully qualified domain name or as a dotted-quad IP address in text form. The port parameter specifies the UDP port to contact on the server. If port is given as 0, the library looks up the ‘radius/udp’ or ‘radacct/udp’ service in the network services(5) database, and uses the port found there. If no entry is found, the library uses the standard RADIUS ports, 1812 for authentication and 1813 for accounting. The shared secret for the server host is passed to the secret parameter. It may be any NUL-terminated string of bytes. The RADIUS protocol ignores all but the leading 128 bytes of the shared secret. The timeout for receiving replies from the server is passed to the timeout parameter, in units of seconds. The maximum number of repeated requests to make before giving up is passed into the max_tries parameter. Time interval in seconds when the server will not be requested if it is marked as dead (did not answer on the last try) set with dead_time parameter. bindto parameter is an IP address on the multihomed host that is used as a source address for all requests. rad_add_server() returns 0 on success, or -1 if an error occurs.

rad_add_server() or rad_add_server_ex() may be called multiple times, and they may be used together with rad_config(). At most 10 servers may be specified. When multiple servers are given, they are tried in round-robin fashion until a valid response is received, or until each server's max_tries limit has been reached.

Creating a RADIUS Request

A RADIUS request consists of a code specifying the kind of request, and zero or more attributes which provide additional information. To begin constructing a new request, call rad_create_request(). In addition to the usual struct rad_handle *, this function takes a code parameter which specifies the type of the request. Most often this will be RAD_ACCESS_REQUEST. rad_create_request() returns 0 on success, or -1 on if an error occurs.

After the request has been created with rad_create_request(), attributes can be attached to it. This is done through calls to rad_put_addr(), rad_put_int(), and rad_put_string(). Each accepts a type parameter identifying the attribute, and a value which may be an Internet address, an integer, or a NUL-terminated string, respectively. Alternatively, rad_put_vendor_addr(), rad_put_vendor_int() or rad_put_vendor_string() may be used to specify vendor specific attributes. Vendor specific definitions may be found in <radlib_vs.h>

The library also provides a function rad_put_attr() which can be used to supply a raw, uninterpreted attribute. The data argument points to an array of bytes, and the len argument specifies its length.

It is possible adding the Message-Authenticator to the request. This is an HMAC-MD5 hash of the entire Access-Request packet (see RFC 3579). This attribute must be present in any packet that includes an EAP-Message attribute. It can be added by using the rad_put_message_authentic() function. The libradius library calculates the HMAC-MD5 hash implicitly before sending the request. If the Message-Authenticator was found inside the response packet, then the packet is silently dropped, if the validation failed. In order to get this feature, the library should be compiled with OpenSSL support.

The rad_put_X() functions return 0 on success, or -1 if an error occurs.

Sending the Request and Receiving the Response

After the RADIUS request has been constructed, it is sent either by means of rad_send_request() or by a combination of calls to rad_init_send_request() and rad_continue_send_request().

The rad_send_request() function sends the request and waits for a valid reply, retrying the defined servers in round-robin fashion as necessary. If a valid response is received, rad_send_request() returns the RADIUS code which specifies the type of the response. This will typically be RAD_ACCESS_ACCEPT, RAD_ACCESS_REJECT, or RAD_ACCESS_CHALLENGE. If no valid response is received, rad_send_request() returns -1.

As an alternative, if you do not wish to block waiting for a response, rad_init_send_request() and rad_continue_send_request() may be used instead. If a reply is received from the RADIUS server or a timeout occurs, these functions return a value as described for rad_send_request(). Otherwise, a value of zero is returned and the values pointed to by fd and tv are set to the descriptor and timeout that should be passed to select(2).

rad_init_send_request() must be called first, followed by repeated calls to rad_continue_send_request() as long as a return value of zero is given. Between each call, the application should call select(2), passing *fd as a read descriptor and timing out after the interval specified by tv. When select(2) returns, rad_continue_send_request() should be called with selected set to a non-zero value if select(2) indicated that the descriptor is readable.

Like RADIUS requests, each response may contain zero or more attributes. After a response has been received successfully by rad_send_request() or rad_continue_send_request(), its attributes can be extracted one by one using rad_get_attr(). Each time rad_get_attr() is called, it gets the next attribute from the current response, and stores a pointer to the data and the length of the data via the reference parameters data and len, respectively. Note that the data resides in the response itself, and must not be modified. A successful call to rad_get_attr() returns the RADIUS attribute type. If no more attributes remain in the current response, rad_get_attr() returns 0. If an error such as a malformed attribute is detected, -1 is returned.

If rad_get_attr() returns RAD_VENDOR_SPECIFIC, rad_get_vendor_attr() may be called to determine the vendor. The vendor specific RADIUS attribute type is returned. The reference parameters data and len (as returned from rad_get_attr()) are passed to rad_get_vendor_attr(), and are adjusted to point to the vendor specific attribute data.

The common types of attributes can be decoded using rad_cvt_addr(), rad_cvt_int(), and rad_cvt_string(). These functions accept a pointer to the attribute data, which should have been obtained using rad_get_attr() and optionally rad_get_vendor_attr(). In the case of rad_cvt_string(), the length len must also be given. These functions interpret the attribute as an Internet address, an integer, or a string, respectively, and return its value. rad_cvt_string() returns its value as a NUL-terminated string in dynamically allocated memory. The application should free the string using free(3) when it is no longer needed.

If insufficient virtual memory is available, rad_cvt_string() returns NULL. rad_cvt_addr() and rad_cvt_int() cannot fail.

The rad_request_authenticator() function may be used to obtain the Request-Authenticator attribute value associated with the current RADIUS server according to the supplied rad_handle. The target buffer buf of length len must be supplied and should be at least 16 bytes. The return value is the number of bytes written to buf or -1 to indicate that len was not large enough.

The rad_server_secret() returns the secret shared with the current RADIUS server according to the supplied rad_handle.

The rad_bind_to() assigns a source address for all requests to the current RADIUS server.

The rad_demangle() function demangles attributes containing passwords and MS-CHAPv1 MPPE-Keys. The return value is NULL on failure, or the plaintext attribute. This value should be freed using free(3) when it is no longer needed.

The rad_demangle_mppe_key() function demangles the send- and recv-keys when using MPPE (see RFC 2548). The return value is NULL on failure, or the plaintext attribute. This value should be freed using free(3) when it is no longer needed.

Obtaining Error Messages

Those functions which accept a struct rad_handle * argument record an error message if they fail. The error message can be retrieved by calling rad_strerror(). The message text is overwritten on each new error for the given struct rad_handle *. Thus the message must be copied if it is to be preserved through subsequent library calls using the same handle.

Cleanup

To free the resources used by the RADIUS library, call rad_close().

Server operation

Server mode operates much alike to client mode, except packet send and receive steps are swapped. To operate as server you should obtain server context with rad_server_open() function, passing opened and bound UDP socket file descriptor as argument. You should define allowed clients and their secrets using rad_add_server() function. port, timeout and max_tries arguments are ignored in server mode. You should call rad_receive_request() function to receive request from client. If you do not want to block on socket read, you are free to use any poll(), select() or non-blocking sockets for the socket. Received request can be parsed with same parsing functions as for client. To respond to the request you should call rad_create_response() and fill response content with same packet writing functions as for client. When packet is ready, it should be sent with rad_send_response().