> I would assume that fsync() is all I need in order to know confidently > that my data and metadata will always survive a crash. I used to think that until Rusty pointed out the error of my ways with his "pull the plug" testing with and without ext3 barriers :)