illumos gate

So admin should do:

share -F nfs -o nohide  {SHARE}

Each resource must be shared.

Example:

zfs set -o sharenfs    pool/dataset

zfs set -o sharenfs=nohide  pool/dataset/subdataset

This change should be RTI'd: https://github.com/Nexenta/illumos-nexenta/commit/efbfae72dc9d4466fa85868000ef73bd4081263c

Andrew Stormont wrote:

This change should be RTI'd: https://github.com/Nexenta/illumos-nexenta/commit/efbfae72dc9d4466fa85868000ef73bd4081263c

Along with fixes for a lot of mem leaks this commit introduced.

Yuri Pankov wrote:

Along with fixes for a lot of mem leaks this commit introduced.

What leaks?

https://github.com/Nexenta/illumos-nexenta/commit/41bdbbaeaed5d6e456b8d42c37b12032cc9fa254

Yuri Pankov wrote:

https://github.com/Nexenta/illumos-nexenta/commit/41bdbbaeaed5d6e456b8d42c37b12032cc9fa254

It looks like it is modification of commit 2007c08f9799 "2986 nfs: exi refcounter leak at rfs3_lookup".

So using "nohide" feature requires to take the last sources version.

The fix for this is comprised of four fixes from illumos-nexenta:

rb11254 - nfs nohide - "nfssrv: need ability to go to submounts for v3 and v2 protocols"
closes #12112 rb3823 - nfs-nohide: lookup("..") for submount should be correct
NEX-3095 Issues related to NFS nohide
NEX-5196 share_nfs(1M): "nohide" description missing

It has been in production in Nexenta since 2016 or 2017, it seems.

Yes, seems it could be upstreamed.

With linux client and its automount it works fine like for nfs4, i.e. no need to specify each exported path manually.

One of setup was a VMware ESX server and bsros NFS server. Testing cases included
1. Create dataset 'data/b', all was shared and 'data/b' shared with nohide
1. Setup NFS datastore and specify data as NFS share .
2. Create VMs in 'data/b' and start VMs, stop VMs, susped VMs, start "fio".