cifs with dfs broken in pve-kernel-5.4.101-1

[dirks]

I have a cifs storage configured in storage.cfg that points to a dfs path. I just noticed that this stopped working. Going back in /var/log/messages* it correlates to the upgrade to pve-kernel-5.4.101-1-pve on 2021-03-02. The same dfs mount still works on my client machine which is still on 5.4.0-66. Some searching on the web shows some recent dfs related work on cifs https://lkml.org/lkml/2021/1/31/294 https://www.suse.com/support/update/announcement/2021/suse-su-20210108-1/ maybe related, did not had time yet to dig deeper.

Error messages in dmesg (only the first and last line seem relevant here, rest is for noserverino option, which to my knowledge is not configurable via storage.cfg):

CIFS VFS: \\example.com\foo error -2 on ioctl to get interface list
CIFS VFS: Autodisabling the use of server inode numbers on new server.
CIFS VFS: The server doesn't seem to support them properly or the files might be on different servers (DFS)
CIFS VFS: Hardlinks will not be recognized on this mount. Consider mounting with the "noserverino" option to silence this message.
CIFS VFS: cifs_read_super: get_root_inode_failed

Another cifs storage pointing directly (no dfs) to a share exported by a server works. For some reason, modifying the problematic mount entry from \\example.com\foo to the \\server\share\path does not help, so I created a fstab entry to the \\server\share\path (that works but not when pointing to dfs) on my nodes and use a directory entry in storage.cfg.

 

[dirks]

sorry just noticed wrong sub-forum should be in Proxmox VE: Installation and configuration

 

[dirks]

The problem was probably introduced somewhere between 5.4.0-67 and 5.4.0-71 as it arrived on my client (Ubuntu 20.04 based) today as well.

edit: with 5.4.0-70 it still works.

 

[dirks]

Will probably be fixed with 5.4.112

For reference:
- https://cdn.kernel.org/pub/linux/kernel/v5.x/ChangeLog-5.4.112
- https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1923670

Should already work with the pve 5.11 kernel but had no time to test on actual proxmox installation yet (only arch machine from colleague).

 

[Stoiko Ivanov]

Thanks for reporting the issue and digging into the archives to see what's causing it!
For the time being I'd suggest booting an older kernel (or trying the 5.11 based one) - We'll release an updated version once Ubuntu moves to a newer one (their next version should contain the linked fix, but is not tagged yet)

 

[dirks]

With 5.11.7-1-pve DFS works fine (pve-kernel-5.11: Installed: 7.0-0+3~bpo10).

 

[rjblake]

I noticed this error this morning

[Sun Oct 16 22:47:18 2022] CIFS: VFS: Autodisabling the use of server inode numbers on \\xxx.xxx.xxx.xxx\video (redacted)
[Sun Oct 16 22:47:18 2022] CIFS: VFS: The server doesn't seem to support them properly or the files might be on different servers (DFS)
[Sun Oct 16 22:47:18 2022] CIFS: VFS: Hardlinks will not be recognized on this mount. Consider mounting with the "noserverino" option to silence this message.

I am using Proxmox v7.2-11 (Linux prox-1 5.15.60-1-pve #1 SMP PVE 5.15.60-1 (Mon, 19 Sep 2022 17:53:17 +0200) x86_64 GNU/Linux). Is this still an issue? I seem to be able to access the folders/files on my NAS that I have mapped.

In my LXC I have added a mount point

mp0: /mnt/pve/lxc_shares/synods1821_video,mp=/volume1/video

and added the following to my /etc/fstab for the node

//xxx.xxx.xxx.xxx/video /mnt/pve/lxc_shares/synods1821_video cifs vers=3,dir_mode=0777,defaults,credentials=/root/.xxx.cred,file_mode=0766,_netdev,x-systemd.automount,noatime,uid=100000,gid=110000,nofail 0 0