LSMCD Secure User Data Using SASL

SASL (Simple Application and Security Layer) is the method used to secure data in LSMCD and Memcached. For details on the use of SASL in LSMCD see LSMCD Security Using SASL.

This wiki discusses a feature of LSMCD which is not available in traditional Memcached: separation of individual users' data. This means that data saved by one user is not visible to any other users. You must have LSMCD v1.2 or higher to use this feature. In Memcached and traditional LSMCD, any data stored is available to all users (all authorized users if you have SASL enabled), which allows fast population of the cache and high utilization. However, it is insecure and thus can't be used to cache any data which is deemed to be sensitive to a specific user.

This new option allows data to be available to only the user authorized to access it. Thus the advantages of Memcached performance becomes available to sensitive data.

Enabling SASL user protection is database wide. Once SASL user protection is enabled, all non-SASL user protected databases will need to be regenerated. You will also need to regenerate your databases (the files stored in the Cached.ShmDir parameter of your node.conf file) if you wish to remove SASL or SASL user protection.

There is a user interface for CloudLinux/cPanel. Installation and use is described at LSMCD Secure User Data CloudLinux/cPanel Interface.

To enable separating data by users, you need to specify in your node.conf file:

Cached.DataByUser=true

As mentioned above, once you have made this change you must delete your existing databases or LSMCD will refuse to come up, as it will notice the changed data condition.

The default is false so that data created by all users is visible to all users. Once it is set to true, each user's data can only be visible to that user.

Use

LSMCD can be used once configured and activated using the traditional Memcached protocols and user commands. However, any data visible will only be visible to the authenticated user that created it. This means that the same data may be stored multiple times for separate users, but each user will only see the data created by that user. Expiration and deletion will again by based on the criteria set when the user created the data or on the parameters for the system as a whole.

The default size for a user cache is by design set quite low to 1000 bytes to avoid the memory and disk overhead with a large number of users. You can specify the size in bytes with the parameter Cached.UserSize. You can also use the Cached.MemMaxSz parameter to have the cache begin aging out data when it reaches your specified size.

If you specify a realm qualified name (a name with a @hostname suffix) in your application, then that name will be used for storage. If you then specify a non-realm qualified name then the unqualified name will be resolved as a different name. This is so that names that appear different are handled differently.

Each user is designated a slice and all traffic will be sent to that slice. However, the size of the hash is determined individually for each user. So Cached.MemMaxSz can be specified to a non-zero value to limit the number of bytes a given user will be allowed to store before least recently used techniques will be used to reduce the data in the cache.

Anonymous User

To support the use of code that does not include authentication information (old code or code which wishes to use the facility but does not wish to deal with authentication), you can allow unauthenticated (anonymous) users to create and access data on your server that is distinct and separate from that created by authenticated users.

To enable separating data by users, you need to specify in your node.conf file:

Cached.Anonymous=true

This does not create a security hole as the data stored and accessed by unauthenticated users is totally separate from data stored for authenticated users when Cached.DataByUser is enabled.

The default is false so that you do not mistakenly allow unauthenticated users access to Memcached facilities (even though the data would be separated).

  • Admin
  • Last modified: 2018/08/14 12:22
  • by Robert Perper