Hi all, I'm testing bcache device in ubuntu 16.04 with 4.4 and 4.15 (hwe-edge) kernels and noticed some strange fio behaviour. When i run fio with --ioengine=libaio and --sync=1 options - it only works as expected on 4.4 kernel. There're additional flush "write" requests, also to backing device of bcache. If i run it on 4.15 - then iops on caching device are same as fio shows and there're no any io on backing device. But, if i set ioengine to posixaio or sync or even if i set --fsync=1 option, then it works as expected on 4.15 - i can see flush io's. It looks like O_SYNC doesn't work with 4.15 when fio runs with libaio. Kernel was installed from ubuntu repos, but libaio is used same as for 4.4 kernel and I assume that it is the root cause. But i can't find any proof of this guess. Have any changes been done in kernel that affect O_SYNC or is it libaio(or fio) bug? Used fio-2.2.10 and fio-3.9. Hope for any help with understanding this behaviour. -- Regards, Aleksei Zakharov
