Transferring larger files to a qube with NFS or CIFS/SMB mounted share crashes the qube

[PeterLacknase]

How to file a helpful issue

Qubes OS release

Qubes Release 4.2.3 (R4.2) -- tested with kernel-default and kernel-latest

Brief summary

Any transfer to or from a qube with an NFS mount from a local NFSv3/NFSv4 server crashes/stalls the qube. After some time you'll have to kill the instance. No matter what template (tried it with Debian-12, Fedora-40 and Fedora-41) -- so SELinux is not the culprit.

Tried it with different server shares on different machines (Debian server with kernel NFS, OmniOS server with kernel NFS, TrueNAS Core/FreeBSD with NFS daemon).

On the command line after mounting the share within a qube everything works fine ... creating dirs, listing, touching files, deleting, renaming/moving, editing ... no matter the mount is NFSv3 or NFSv4. The same mounts work with "bare metal" linux clients (tried Debian and Arch) and MacOS. Even with larger files (500MB and above). But as soon as a qube tries to write to or read larger data from the share, the qube dies. The resources are fine (4GB/8GB/16GB -- you name it -- 40GB, 80GB, 120GB local space ... doesn't make any difference).

No suspicious status messages in dom0: /var/log/qubes/qrexec.$QUBENAME.log, or within the qube via dmesg or journalctl -x ... the server shows just open sessions. Everything is fine on server side. (With any of the tested servers.)

Steps to reproduce

In a qube(-template) install nfs-utils (Fedora-40/41) or nfs-common (Debian) deps included and mount the share. I did it via rc.local, but it's the same from the command line:

sudo vi /rw/config/rc.local ->

mkdir /home/user/test
echo '' >> /etc/fstab
echo '$SOMELOCALIP:/srv/SOMESHARE /home/user/test nfs rw,_netdev 0 0' >> /etc/fstab
mount /home/user/backup

Expected behavior

Should work like it did last time mid December. I dare to guess: Some updates to qubes core packages must have b0rked it, since it used to work with any template and now it doesn't ... again with any template I tried.

Actual behavior

The qube stalls/dies. No qubes backup via network :-(

[PeterLacknase]

Latest updates (kernel, qubes-utils etc. etc.) fixed the reading part.

But still ... when writing large(r) files, things get queued and the qube hangs immediately afterwards (again). Tested (again) with Fedora-[40|41]- and Debian-12-templates.

[PeterLacknase]

While searching for an alternative solution it seems to me that the same applies to cifs/smb mounts ...

mkdir /home/user/backup
echo '#The following line is added at boot by the /rw/config/rc.local script' >> /etc/fstab
echo '//$SERVERIP/$SHARENAME /home/user/backup cifs uid=user,credentials=/home/user/.smb,iocharset=utf8,vers=3,noperm 0 0' >> /etc/fstab
mount /home/user/backup

[PeterLacknase]

Reinstalled Qubes OS 4.2.3 from freshly downloaded ISO, updated everything and recovered my qubes from a full backup.
With Fedora-40/Debian-12 templates, standard kernel and the latest fixes the error is gone. I must have b0rked something myself while trying to compensate the memory ballooning.