Rob 💚
@robn@social.lol replied  ·  activity timestamp 3 days ago
@dexter I think I need a lot more info/experience. I don't actually know much about NFS or SMB, so I can only guess the problems you're running in to.

OpenZFS definitely has enough information to uniquely identify an object within a pool (dataset id/object id pair). Skimming the NFSv4 spec, it seems like it should be nbd to put all datasets into a single virtual namespace. But thats me imagining NFS as a first-class interface to ZFS, rather that a standalone server using POSIX interfaces. Its quite possible we don't expose enough info for a server to use.

I'd be happy to chat to an NFS server person if they'd like to work with me on plumvong useful info our org ZFS for a server to use!
Michael Dexter
@dexter@bsd.network replied  ·  activity timestamp 2 days ago
@robn I don’t know what the original motivation was but different datasets a.k.a. “file systems” behave just like that on the POSIX level: as if they are on different partitions. I suspect this is to keep from exhausting inode numbers, in practice it makes for #weird behavior if you nest datasets in a single share. There might be opportunities for OpenZFS to be smarter, but a blunt instrument that shares/pools/spans inode numbers might be adequate.

💚