Sorry, misread as idmap and when I replied with th...
# choria
b
Sorry, misread as idmap and when I replied with that, phone autocorrected to uidmap. With kerberos, the kernel does an upcall to get a session cred for a specific uid, and then the cred embeds the user name sent over the wire for the server to use for acls. If the container image is multi user, then run inside a user namespace it needs subuid/subgids allocating, which belong to the non-root host user. The namespace uid is mapped to the users allocated subuid range with the expecting of 0 which maps to the host user id. Nfs from uid then works because the kennel does an upcall for the existing host uid mapped to namespace uid0. But nfs for non root namespace users have no corresponding host id, so gssd/gssproxy can't obtain session creds. The 65k non-root namespace uids map to high numbers, not the host user Id, 0 is treated specially.