Write barriers
Chapter 2, Database Hardware already mentioned that most hard drives have volatile write caches in them, and that the WAL writing scheme used by the database isn't compatible with those. The important requirement here is that when a file sync operation (fsync on UNIX) occurs, the filesystem must make sure all related data is written out to a non-volatile cache or disk itself. Filesystem journal metadata writes have a similar requirement. The writes for the metadata updates require that the journal updates are first written out to preserve proper ordering.
To support both of these situations, where something needs to be unconditionally written to the disk and to where an ordering is required, the Linux kernel implements what they call write barriers. A quote from the kernel documentation on barriers:
All requests queued before a barrier request must be finished (made it to the physical medium) before the barrier request is started, and all requests queued after...
