Converting Applications from v1 to v2

This documents our conversion experience with Cyrus IMAPd, an application that uses almost every part of SASL, so it should give a good idea what caveats need to be looked for when one is converting an application which uses SASLv1 to use SASLv2.

The major changes in the SASLv2 API have to do with memory management. That is, the rule “If you allocate it, you free it” is now enforced. That means that if the application allocates something (for example, an interaction or callback response), it must free it. Likewise, the application does NOT free anything handed to it by the SASL library, such as responses given by sasl_client_step or sasl_decode.

Tips for both clients and servers

  • Change configure scripts to search for libsasl2 and include files prefixed with sasl/ (sasl/sasl.h, sasl/saslutil.h, etc)
  • sasl_decode64 now takes an additional parameter that is the size of the buffer it is passed.
  • External authentication properties are no longer handled by a sasl_external_properties_t. Instead you make 2 separate calls to sasl_setprop.
    • One with SASL_SSF_EXTERNAL to tell the SASL library what SSF is being provided by the external layer.
    • The other sets SASL_AUTH_EXTERNAL to indicate the authentication name.
  • sasl_getprop now returns its value in a const void \*\*
  • sasl_encode and sasl_decode now return a constant output buffer, which you do not need to free (it is only valid until the next call for this sasl_ conn_t, however)
  • The SASL_IP_REMOTE and SASL_IP_LOCAL properties are now SASL_IPLOCALPORT and SASL_IPREMOTEPORT and take strings instead of sockaddrs. These strings may also be passed to the sasl_[client/server]_new functions. They are in one of the following formats:
    • a.b.c.d;p (IPv4, with port)
    • e:f:g:h:i:j:k:l;p (IPv6, with port)
    • e:j:k:l;p (IPv6, abbreviated zero fields, with port)
  • Error handling and reporting is different. All of the functions that used to return a “reply” string no longer do. Now you should (always) check sasl_errdetail. Callbacks MUST likewise use sasl_seterror instead of setting their (now nonexistent) reply parameter.
  • Be very careful about your handling of maxoutbuf. If you claim that you can only read 4096 bytes at a time, be sure to only pass at most that much at a time to the SASL library!

Tips for clients

  • In sasl_client_new you can now pass ip address strings as parameters 3 and 4 instead of calling setprop later on sockaddr’s. This is preferred but not required (not passing them by either method disables mechs which require IP address information). You might find the iptostring() function in utils/smtptest.c to be useful for this. If the protocol supports the server sending data on success you should pass SASL_SUCCESS_DATA as a flag.
  • sasl_client_start loses the 3rd “secret” parameter. Also, NULL clientout and clientoutlen indicates that the protocol does not support client-send-first. A NULL return value indicates that there is no first client send. (as opposed to an empty string, which indicates that the first client send is the empty string).
  • Both sasl_client_start and sasl_client_step now take const clientout parameters that you are no longer responsible for freeing (it is only valid until the next call for this sasl_conn_t, however)
  • When interactions and callbacks happen you are responsible for freeing the results.

Tips for Servers

  • SASL_SECURITY_LAYER flag no longer exists, whether or not to use a security layer is solely determined by the security properties information, namely, the maxbufsize member of the sasl_security_properties_t
  • sasl_server_new now can take ip address strings.
  • sasl_checkpass no longer has a “reply” parameter. There are also considerably fewer possible values for the pwcheck_method option (now only auxprop, saslauthd, authdaemond, and pwcheck).
  • sasl_server_start / sasl_server_step have same output parameter deal as their equivalents on the client side
  • sasl_listmech has a constant output parameter
  • If you used to canonicalize the username in a SASL_CB_PROXY_POLICY callback you should now separate the functionality of authorization and canonicalization. That is, only do authorization in SASL_CB_PROXY_POLICY, and do canonicalization in the SASL_CB_CANON_USER callback