On Thu 24-03-22 18:11:48, Christoph Hellwig wrote: > On Thu, Mar 24, 2022 at 02:23:22PM +0100, Jan Kara wrote: > > Should't we call __invalidate_device() for the partition bdev here? Because > > if the NBD device has partitions, filesystem will be mounted on this > > partition and we want to invalidate it. Similarly the partition buffer > > cache is different from the buffer cache of the whole device and we should > > invalidate the partition one. In fact in cases like this I think we need > > to invalidate all the partitions and filesystems that are there on this > > disk so neither the old, nor the new code looks quite correct to me. Am I > > missing something? > > Well, that assumes just one partition is used, which kinda defeats the > purpose of partitions. I can exclude the __invalidate_device to not > change from one kind of broken to another, but I suspect the real > question is why we have this __invalidate_device call at all. I suppose it tries to cleanup after effectively hot-unplugging the device. But I think we don't need to untangle that in this patch set. I'd just prefer we would not change one questionable behavior for another similarly questionable one... Honza -- Jan Kara <jack@xxxxxxxx> SUSE Labs, CR
