During syncbarrier recovery, when JFS is waiting for all outstanding
transactions to complete before allowing new ones to start, there was a
window where a new transaction starts, but is not committed when JFS thought
it could stop flushing the journal to disk.  This caused the log_SYNCBARRIER
flag to not be reset, and no new transactions are allowed to start.

This is fixed by using the log->active field to determine when to reset the
log_FLUSH flag.