Centralized Login Management Comes to MySQL: Introducing PAM and Windows Authentication Support
By User13177919-Oracle on IX 16, 2011
This is why you need authentication plugins that fit the largest possible number of authentication setups.
I can only guess what these authentication solutions look like for your server. But chances are that your OS has a pretty good idea on how best to authenticate users in it. This is why we've decided to hook MySQL to the two most widely used OS authentication APIs : Pluggable Authentication Modules (a.k.a PAM) and the Windows Security Support Provider Interface
Let's see first how PAM works.
We'll first consider the simple scenario where each OS user holds a corresponding user account in MySQL's mysql.user table. Sounds pretty easy to setup ? Right : you just need to remember to create a MySQL user for each of the people that have access to the computer and grant them some access.
There's one inconvenience though : you'll need to maintain the correspondence between the MySQL user and the OS user. Ask the user to update his mysql password in addition to his OS password, remember to disable access in both mysql and the OS etc.
This is where the PAM authentication plugin in its simplest form can help you: if properly configured it will ask the OS if this user can login each time a user tries to login to the MySQL server. All you need to do is create your MySQL users in such a way that they use the PAM plugin to login.
Note that externally authenticated MySQL users don't need a password stored into the mysql.user table, as they are relying on the PAM library to check the login credentials. Good ! One less password to worry about and maintain.
So how do I set things up ?
First I need to check if I have the PAM library set up on my box. Usually PAM keeps its configuration in /etc/pam.d. Looking in my macbook's /etc/pam.d I see a number of configuration paths
mbpro:~$ ls -la /etc/pam.d/ total 0 drwxr-xr-x 17 root wheel 578 Aug 29 20:46 ./ drwxr-xr-x 103 root wheel 3502 Sep 12 19:22 ../ -rw-r--r-- 1 root wheel 264 Aug 29 20:42 authorization -rw-r--r-- 1 root wheel 174 Aug 29 20:46 checkpw -r--r--r-- 1 root wheel 200 Aug 29 20:46 chkpasswd -r--r--r-- 1 root wheel 203 Aug 29 20:46 cups -rw-r--r-- 1 root wheel 204 Aug 29 20:46 ftpd -r--r--r-- 1 root wheel 587 Aug 29 20:46 login -r--r--r-- 1 root wheel 152 Aug 29 20:46 login.term -r--r--r-- 1 root wheel 191 Aug 29 20:42 other -r--r--r-- 1 root wheel 197 Aug 29 20:46 passwd -r--r--r-- 1 root wheel 211 Aug 29 20:46 rshd -rw-r--r-- 1 root wheel 408 Aug 29 20:46 screensaver -rw-r--r-- 1 root wheel 140 Aug 29 20:46 smbd -r--r--r-- 1 root wheel 602 Aug 29 20:42 sshd -r--r--r-- 1 root wheel 356 Aug 29 20:46 su -r--r--r-- 1 root wheel 203 Aug 29 20:46 sudoGood, looks like the PAM library is already configured and under heavy use. Now all I need is to pick the right PAM path for my MySQL server. login sounds good. Let's see what's in it :
mbpro:~$ cat /etc/pam.d/login # login: auth account password session auth optional pam_krb5.so use_kcminit auth optional pam_ntlm.so try_first_pass auth optional pam_mount.so try_first_pass auth sufficient pam_serialnumber.so try_first_pass serverinstall auth required pam_opendirectory.so try_first_pass account required pam_nologin.so account required pam_opendirectory.so password required pam_opendirectory.so session required pam_launchd.so session required pam_uwtmp.so session optional pam_mount.so mbpro:~$I will not go into the details of all this, but will just highlight pam_opendirectory.so. This means that this PAM config will use Apple's OpenDirectory to verify logins. Nice ! But there's a number of bookkeeping modules that I don't really need for MySQL : e.g. pam_launchd.so.
I need something simpler. Hmm, let's try some of the existing daemons. Let's check cups.
$ cat /etc/pam.d/cups # cups: auth account password session auth required pam_opendirectory.so account required pam_permit.so password required pam_deny.so session required pam_permit.so
Good ! pam_opendirectory.so still there, there's auth and it's really short and sweet. This should do just fine. I could have created my own pam configuration file. And if I call it mysql the PAM plugin will pick it up without me needing to specify an authentication string. But I'll stick to what I have already.
Let's start the MySQL server and make sure the PAM plugin is available and installed.
mysql> select plugin_status from information_schema.plugins where plugin_name='authentication_pam'; -------------- select plugin_status from information_schema.plugins where plugin_name='authentication_pam' -------------- +---------------+ | plugin_status | +---------------+ | ACTIVE | +---------------+ 1 row in set (0.04 sec)
Now I can experiment. I need to create a MySQL user for my gkodinov OS account
mysql> create user gkodinov identified with 'authentication_pam' as 'cups'; -------------- create user gkodinov identified with 'authentication_pam' as 'cups' -------------- Query OK, 0 rows affected (0.00 sec)
One can never check too much, so let's see how is the user defined in the mysql.user table
mysql> select Password,plugin from mysql.user where user = 'gkodinov'; -------------- select Password,plugin from mysql.user where user = 'gkodinov' -------------- +----------+--------------------+ | Password | plugin | +----------+--------------------+ | | authentication_pam | +----------+--------------------+ 1 row in set (0.00 sec)
Goody ! No password to maintain and worry about.
I can now try logging in. Note that I'm not supplying my password as an argument on the command line : it's insecure. It's better to enter it on the prompt.
../client/mysql -u gkodinov -p test Enter password: Welcome to the MySQL monitor. Commands end with ; or \g. Your MySQL connection id is 6 Server version: 5.5.16-debug-log Source distribution Copyright (c) 2000, 2011, Oracle and/or its affiliates. All rights reserved. Oracle is a registered trademark of Oracle Corporation and/or its affiliates. Other names may be trademarks of their respective owners. Type 'help;' or '\h' for help. Type '\c' to clear the current input statement. mysql> select user(),current_user(); +--------------------+----------------+ | user() | current_user() | +--------------------+----------------+ | gkodinov@localhost | gkodinov@% | +--------------------+----------------+ 1 row in set (0.00 sec)
Wow ! I'm in. Let's see what happens if I mistype my password.
../client/mysql -u gkodinov -p test Enter password: ERROR 1045 (28000): Access denied for user 'gkodinov'@'localhost' (using password: YES)
Nothing unexpected. But there's nothing in the logs either (depending of course on you logging setup. On Linux you'd get a note in the secure syslog saying that the login failed). This can be quite annoying, specially if you're setting up PAM. So what can I do to diagnose PAM related problems ? I can start the server in a special debugging mode by setting the AUTHENTICATION_PAM_LOG environment variable to 1 before starting my server. Use it with caution : it prints a lot of data on the server's standard error. Thus it's not suitable for production use. You better set up your PAM modules or your external credentials storage to log authentication errors.
Among other things the debug log prints is the PAM error returned by the pam_authenticate() function. So for my failed attempt above I'd get:
... auth_pam_server:password haha received ... auth_pam_server:pam_authenticate rc=9 auth_pam_server: rc=9 PAM error: authentication error ...
9 is PAM_AUTH_ERR.
The above debugging log exposes one more possible problem that users should be aware of : When using PAM authentication the passwords are transferred from the client to the server in clear text. This may be dangerous if you're doing it over a network link that others can listen to. Unfortunately this is about the only way PAM can work : the library requires the password in clear text. So plan carefully and use SSL encryption of the client connection where needed !
Now that we seen how the PAM plugin can be used in simple scenarios it's time to think about more complex ones.
If you have more than a handful of users creating all these MySQL accounts for each such user can quickly become a problem. Each time you e.g. add somebody to your corporate directory you'll need to go to each and every MySQL server instance and add it as an externally authenticated MySQL user. And what's worse : grant him the privileges that he needs.
Luckily in such environments large groups of users share the same set of privileges : e.g. all DBAs have access to all the data in a MySQL server instance, all accountants have access to the tables holding the accounting data etc. While designing your application you have a pretty good idea what these typical privilege sets are. And you probably even have "template" users of different kinds : e.g. "accountant" with all the rights necessary to do his job. Or "dba" with a larger set of privileges etc. And those are pretty static user definitions. They change only when your application logic changes.
How cool would it be to only have these template users defined in each MySQL server and somehow map incoming logins to them ? And keep all logins and the mapping to the template users definable at OS level ? Of course the MySQL DBAs still need to be able to control the mapping. But other than this it's a good idea to let the OS user definitions be the only place where you need to add a new user account when you e.g. hire somebody.
The PAM plugin can answer this call. But since the PAM library API is so basic the PAM plugin needs to do the user mapping somehow otherwise.
In Unix the most obvious way of handling groups of users is through OS groups. Each user can belong to several such groups. E.g. on my Mac I have:
$ groups staff com.apple.sharepoint.group.2 com.apple.access_screensharing everyone _appstore localaccounts _appserverusr admin _appserveradm _lpadmin _lpoperator _developer com.apple.sharepoint.group.1This means my user id belongs to all the above groups. The information about the group membership is available to programs through the group information functions of the C standard library.
We have two questions to answer:
- How can we make the standard C library group information functions talk to an external directory (e.g. LDAP or OpenDirectory) ?
- How can we use OS groups to map OS logins to MySQL "template" user accounts that have the rights granted to them ?
The answer on the group mapping functions question is pretty standard. E.g. On linux you'll need to tweak your /etc/nsswitch.conf to allow groups to be stored and retrieved from a remote repository. In this way standard C library function like e.g. getgrnam() can connect to a remote repository and retrieve group and user information from it.
On my Mac I don't actually need to do anything : the C library functions just talk to OpenDirectory. And I can use the standard Mac admin tools to assign groups to my users.
This is why I will assume that the group mapping C standard library functions work in conjunction with the user verification PAM profile used and will concentrate on doing the mapping of OS groups to MySQL template users using the PAM plugin.
The second question has an easy answer too : the PAM plugin supports proxy users. So we can map the incoming logins using a default proxy user. To achieve this we use the full PAM authentication string syntax :
<pam service name>,<os_group_name>=<mysql user name>, <os_group_name>=<mysql user name>, ...
I assume my MySQL application has 2 "template" users : reader and writer. I will be mapping these to the everyone and admin OS groups. I do this by thoughtfully creating my MySQL default proxy user (as root) :
mysql> create user reader identified with 'authentication_pam' as 'cups'; -------------- create user reader identified with 'authentication_pam' as 'cups' -------------- Query OK, 0 rows affected (0.00 sec) mysql> create user writer identified with 'authentication_pam' as 'cups'; -------------- create user writer identified with 'authentication_pam' as 'cups' -------------- Query OK, 0 rows affected (0.00 sec) mysql> create user ''@'' identified with 'authentication_pam' as 'cups,admin=writer,everyone=reader'; -------------- create user ''@'' identified with 'authentication_pam' as 'cups,admin=writer,everyone=reader' -------------- Query OK, 0 rows affected (0.00 sec) mysql> grant proxy on reader to ''@''; -------------- grant proxy on reader to ''@'' -------------- Query OK, 0 rows affected (0.01 sec) mysql> grant proxy on writer to ''@''; -------------- grant proxy on writer to ''@'' -------------- Query OK, 0 rows affected (0.00 sec)Now when I login with my user account (gkodinov) I get :
mysql> select user(), current_user(), @@proxy_user; +--------------------+----------------+--------------+ | user() | current_user() | @@proxy_user | +--------------------+----------------+--------------+ | gkodinov@localhost | writer@% | ''@'' | +--------------------+----------------+--------------+ 1 row in set (0.00 sec)What does this mean ? I've logged in as gkodinov@localhost (as returned by user()). My effective user id (who's privileges set will I be enjoying during my session) is 'writer@%' (as returned by current_user()). And I was able to achieve this using the ''@'' proxy user. Note that the order in the authentication string is important : I've purposely specified the writer mapping before reader's. I did this because I assume that if a user is a member of both OS groups (as gkodinov is) I prefer it to map to the writer group.
Note also that I've defined my reader and writer users as authenticated through PAM. This is to cover "direct" login attempts for login ids reader and writer. In this case the initial MySQL user lookup will find the more specific names (reader and writer) instead of the catch-all user. But this will not matter, as both the template MySQL users and my proxy user are defined as authenticated using the PAM plugin, so the same PAM plugin will be used, this time without a proxy user.
As usual you can find more about the PAM plugin in its excellent online documentation.
Now let's see how Windows Authentication Plugin works
The plugin name is authentication_windows. Unlike the PAM plugin it is not forced to use the clear text password. The Windows API allows programs to use single sign on. Once the user is authenticated with the Windows OS any program can take a "ticket" and verify that this ticket was produced by a specific OS user. As a result the Windows authentication plugin doesn't even require the client to send a password. At connect time it will securely find out what OS user was used to run the MySQL client. In fact you don't even need a user name as well. If you don't supply a user name some clients will use a predefined user name ("auth_windows") and let the server windows authentication plugin to retrieve the OS user name and do the mapping to an SQL user based on it. As a result you also don't need to touch on ''@'' as you do with PAM.