Cuda_11.7.1_515.65.01_linux.run executing a chown root from /

Hello,
We’ve seen this with 11.7.1 and 11.7.0 (11.4 and earlier is ok, not sure about 11.5 or 11.6). The cuda*linux.run install is running a chown from what appears to be / , including network mounted storage.

This is a snippet of the messages file:
Aug 9 13:49:16 c4140c01-default S98cuda[1722]: chown: changing ownership of ‘./proc/sys/net/ipv6/route/gc_min_interval_ms’: Operation not permitted
Aug 9 13:49:16 c4140c01-default S98cuda[1722]: chown: changing ownership of ‘./proc/sys/net/ipv6/route/gc_thresh’: Operation not permitted
Aug 9 13:49:16 c4140c01-default S98cuda[1722]: chown: changing ownership of ‘./proc/sys/net/ipv6/route/gc_timeout’: Operation not permitted
Aug 9 13:49:16 c4140c01-default S98cuda[1722]: chown: changing ownership of ‘./proc/sys/net/ipv6/route/max_size’: Operation not permitted
Aug 9 13:49:16 c4140c01-default S98cuda[1722]: chown: changing ownership of ‘./proc/sys/net/ipv6/route/min_adv_mss’: Operation not permitted
Aug 9 13:49:16 c4140c01-default S98cuda[1722]: chown: changing ownership of ‘./proc/sys/net/ipv6/route/mtu_expires’: Operation not permitted
Aug 9 13:49:16 c4140c01-default S98cuda[1722]: chown: changing ownership of ‘./proc/sys/net/ipv6/route’: Operation not permitted

S98cuda is just an install script:
#!/bin/sh
rmmod nouveau
sh /share/src/cuda/11.7.1/cuda_11.7.1_515.65.01_linux.run --no-drm --silent
/usr/bin/nvidia-persistenced
/usr/bin/nvidia-smi -pm 1
/usr/bin/nvidia-smi -c 3

One thing to note, these nodes are stateless, the OS is running in memory (tmpfs).

Thanks,
Lewis