If I do
touch xxx
setfacl -m g::--- xxx
strace -f -ooutput chmod g+rw xxx
getfacl xxx
I see that the group acl for xxx is changed to include rw, and the file 'output' shows that chmod simply calls the fchmodat syscall, which uses notify_change to update the acl.
If I do
# make sure to be logged into console
sudo apt-get purge qemu-kvm
sudo modprobe kvm_intel
sudo apt-get install qemu-kvm
getfacl xxx
# see that group acl is still ---
sudo chmod g+rw /dev/kvm
# see that group acl is still ---
So the kernel is meant to be updating the acl in fchmodat, but is not doing it. I will try a xfs-backed VM to see if the bug is in ext4 itself, or in the generic fs code.
I'm now convinced this is a kernel bug.
If I do
touch xxx
setfacl -m g::--- xxx
strace -f -ooutput chmod g+rw xxx
getfacl xxx
I see that the group acl for xxx is changed to include rw, and the file 'output' shows that chmod simply calls the fchmodat syscall, which uses notify_change to update the acl.
If I do
# make sure to be logged into console
sudo apt-get purge qemu-kvm
sudo modprobe kvm_intel
sudo apt-get install qemu-kvm
getfacl xxx
# see that group acl is still ---
sudo chmod g+rw /dev/kvm
# see that group acl is still ---
So the kernel is meant to be updating the acl in fchmodat, but is not doing it. I will try a xfs-backed VM to see if the bug is in ext4 itself, or in the generic fs code.