pam_tally2 does fsync() after writing to a tally file.
This causes hard drive cache flushes on every failed SSH login on many
(if not most) filesystems.
And an internet-exposed machine can have a lot of these failed logins.
This operation however doesn't seem to be necessary - the pam_tally2
module does not do any operation which would need explicit post-crash
ordering, it just does simple file reads and writes.
And doing a fsync() after them doesn't close any race if the system happens
to crash between a write being posted and its fsync() completion.
Let's remove this operation to get rid of all these extra cache flushes.