When and why to deprecate filesystems [LWN.net]

When and why to deprecate filesystems

Posted Mar 7, 2022 22:41 UTC (Mon) by jhoblitt (subscriber, #77733)
Parent article: When and why to deprecate filesystems

+1 to deprecate NFS v2/v3 in the kernel, even if it is only the server support. Ganesha has v3 server support for anyone unable to migrate to v4 and I have at least see fuse NFS clients. I have no doubt that NFS v2 will still be in use in the year 2222 but it is a shame for the kernel community to have to keep maintaining obsolete protocols when there are user-space alternatives.

to post comments

When and why to deprecate filesystems

Posted Mar 8, 2022 11:27 UTC (Tue) by jlayton (subscriber, #31672) [Link]

I too would like to see v2/3 go the way of the Dodo, but I don't think we'll probably be able to drop them that easily. One of the problems there is that at least some pNFS layouts (e.g. flexfiles) can use NFSv3 as a transport protocol, and some folks (Hammerspace, for instance) can use v3 servers as pNFS DS's. They may be able to eventually use v4.1 instead, but that sort of transition will take time.

This is probably a good discussion for the upcoming NFS Bakeathon.

When and why to deprecate filesystems

Posted Mar 12, 2022 19:29 UTC (Sat) by rwmj (subscriber, #5474) [Link] (1 responses)

I'd be happier with this is NFSv4 user mapping wasn't such a massive screw-up to the extent that on my home network I just mapped every local user to a single nobody user on NFS because getting anything else to work was impossible.

When and why to deprecate filesystems

Posted Mar 12, 2022 20:25 UTC (Sat) by jhoblitt (subscriber, #77733) [Link]

The documentation is definitely suboptimal. AFAIK, useful idmaping requires krb5. I haven't dug into the details -- this may be a server implementation decision instead of a limitation of the v4 protcol. If krb5 is in use, there is a good chance that UIDs/GIDs are coordinated and idmaping isn't nessicary. I use v4 heavily at my $day_job without remapping without much complaint other than there being no showmount equiv (but I think it so be possible to write a usespacer cli client that does ls on the root export).