Re: [nfsv4] RFC 7530: Available attributes for READDIR and Absent File Systems

[Vitaliy Gusev <[e-mail]>]

> On 21 May 2017, at 22:26, David Noveck <[e-mail]> wrote:
> …
> The other relevant case for READDIR is one in which the operation is being done in a directory within a present file system and some of objects within the directory are the roots of absent file systems.  Given that context, the document could have been clearer if it had mentioned fsid and mounted_on_filled which need to be provided  (it is a SHOULD) for GETATTR.  At the time RFC7530 was done we could have made these "SHOULD"s as they are for GETATTR but we didn't do that since this is not a very important case and the client has the option to follow up with GETATTR.
> 
> As it is, the server MAY return these attributes.  If one is doing a server-side implementation, it is best to to provide them since they need to be available for GETATTR.  On the other hand, if one is doing a client-side implementation, you can’t assume they will be present and need to be prepared to follow up with an OTW GETATTR (which has to specify fs_locations to prevent it from being rejected).

So you specify case when server can return READDIR the same or lesser quantity of attributes as GETATTR
returns. This case is reasonable.

But what if list of attributes returned by READDIR for root of absent filesystem is more widely than GETATTR returns?
In other words, can other attributes than fs_locations, fsid, mounted_on_fileid, rdattr_error
be available for a root of absent filesystem in READDIR reply?

———
Thanks,
Vitaliy Gusev