A few NFS bugfix

[ston3lu]

Hi, I've been testing Vitastor-NFS for a while and have fixed a few bugs.

Looking forward to your review.

By submitting this pull request, I accept Vitastor CLA

[vitalif]

Hi, thanks!
Partially merged. At first glance I don't like two things:

1. add root inode to kv db when nfs start - what exactly does the missing root inode break, how to reproduce the problem?
2. return eof when read offset exceed file length - I know that it's a partial violation of the semantics but I didn't add this check on purpose - I wanted reads from large files to be data-only. So again, what does it really break?

[ston3lu]

1. add root inode to kv db when nfs start - what exactly does the missing root inode break, how to reproduce the problem?

The mount client typically determines whether to resend a readdir request to obtain the latest directory listing by comparing the directory's mtime. The most obvious case is that when client A creates a new file in the mounted root directory, client B cannot see the file.

2. return eof when read offset exceed file length - I know that it's a partial violation of the semantics but I didn't add this check on purpose - I wanted reads from large files to be data-only. So again, what does it really break?

This is because I'm using a go-nfs-client which reads file endings only by checking if the EOF flag is set. For ordinary NFS clients, I indeed haven't encountered any issues. However, this is still a protocol specification, and I hope to resolve it to prevent potential failures in other use cases in the future.