[prev in list] [next in list] [prev in thread] [next in thread] 

List:       freeradius-users
Subject:    rlm_sql_freetds / rlm_sql_sybase
From:       Arran Cudbard-Bell <a.cudbardb () freeradius ! org>
Date:       2013-06-18 18:29:01
Message-ID: CBD3FB1F-FD0F-438B-8EFD-8088FBA9430E () freeradius ! org
[Download RAW message or body]

In 2.0.0 we had two sql drivers rlm_sql_freetds (db), and rlm_sql_sybase (ct-lib).

They used two different APIs but the same protocol, TDS. The reason why they were \
bundled with the server was to support Sybase and MSSQL servers natively, without an \
ODBC connector.

In 3.0.0 rlm_sql_sybase has been renamed to rlm_sql_freetds as it was more complete \
and used a newer API, and the old rlm_sql_freetds module has been removed.

The following changes were made:
* Integrate it into the build system (add configure.in and all.mk.in)
* Code cleanup.
* Don't leak connection handles or contexts or statements - the old code made no \
attempt to close connections, or free resources - it had the closure code copy/pasted \
                from rlm_sql_oracle and then commented out. Um...
* Pass userdata to message callbacks so they can write potential errors to the \
connection handle (which are later retrieved via sql_error(), if an actual error \
                occurred).
* Log informational messages from the client library and the client (I have no idea \
                why there's this distinction) using INFO() in real time.
* Select database on connection spawn. The API had no support for this, so we just \
issue a 'USE <db_name>;' statement every time a connection is created. The db context \
                change seems to persist across multiple queries, at least with MSSQL.
* Switch severity of server messages. During connection establishment they're INFO(), \
then, if they occur during a query, they're treated as ERROR()s (accessible from \
sql_error()). The reason for this is that there are no standard severity levels \
between TDS SQL servers, and so you have absolutely no idea whether the messages \
returned by the server are errors or not. The nice thing about sql_error() is it's \
only called when one of the API functions has returned an error code, so hopefully \
you won't get informational messages showing up as errors. It's the best hack I could \
think of. Blame Sybase (or whoever they bought the IP off) for coming up with such a \
                shitty protocol.
* Better and more consistent logging.

To get rlm_sql_freetds you need to have the freetds libraries available (apt-get \
install freetds-dev), or something else which provides a ct-lib API. The configure \
script isn't picky so long as it can find ctpublic.h and ct_command in libct it'll \
build the module.

The driver isn't marked as stable (once people test it and provide positive feedback \
it may be), but will build if you pass --with-experimental-modules or \
--enable-developer to configure, or build from .git (which automatically enables \
developer).

There's no additional driver options, though there may be some advantage to setting \
client version dynamically (let me know if you want/need this).

-Arran

Arran Cudbard-Bell <a.cudbardb@freeradius.org>
FreeRADIUS Development Team

-
List info/subscribe/unsubscribe? See http://www.freeradius.org/list/users.html


[prev in list] [next in list] [prev in thread] [next in thread]