This page describes a good general procedure:

http://docs.mongodb.org/manual/tutorial/backup-sharded-cluster-with-filesystem-snapshots/

However, regarding taking the snapshot itself, the more specific case where the user has the capability to do a filesystem snapshot which guarantees point in time consistency at the filesystem level (and includes the journal) makes the use of the fsyncLock/unLock an unnecessary step.

Adding a note something like this as part of that step would help avoid confusion here:

Note: Generally, locking/freezing the instance is required. However, if you have filesystem level point in time snapshot capability (Amazon's EBS, LVM, and other snapshot technologies provide this), and that snapshot includes both the MongoDB data and the journal, it is not necessary to fsync and lock the database to obtain a usable snapshot.