On Wed, Sep 27, 2023 at 12:03 PM Andrii Nakryiko <andrii.nakryiko@xxxxxxxxx> wrote: > On Wed, Sep 27, 2023 at 2:52 AM Christian Brauner <brauner@xxxxxxxxxx> wrote: ... > > IOW, everything stays the same apart from the fact that bpf token fds > > are actually file descriptors referring to a detached bpffs file instead > > of an anonymous inode file. IOW, bpf tokens are actual bpffs objects > > tied to a bpffs instance. > > Ah, ok, this is a much smaller change than what I was about to make. > I'm glad I asked and thanks for elaborating! I'll use > alloc_file_pseudo() using bpffs mount in the next revision. Just a FYI, I'm still looking at v6 now, but moving from an anon_inode to a bpffs inode may mean we need to drop a LSM hook in bpf_token_create() to help mark the inode as a BPF token. Not a big deal either way, and I think it makes sense to use a bpffs inode as opposed to an anonymous inode, just wanted to let you know. -- paul-moore.com
