Hello.
On Wed, Oct 16, 2024 at 09:36:33AM GMT, Chen Ridong <chenridong@xxxxxxxxxxxxxxx> wrote:
> As mentioned above, when cgroup_bpf_inherit returns an error in
> cgroup_setup_root, cgrp->bpf.refcnt has been exited. If cgrp->bpf.refcnt is
> killed again in the cgroup_kill_sb function, the data of cgrp->bpf.refcnt
> may have become NULL, leading to NULL pointer dereference.
>
> To fix this issue, goto err when cgroup_bpf_inherit returns an error.
> Additionally, if cgroup_bpf_inherit returns an error after rebinding
> subsystems, the root_cgrp->self.refcnt is exited, which leads to
> cgroup1_root_to_use return 1 (restart) when subsystems is mounted next.
> This is due to a failure trying to get the refcnt(the root is root_cgrp,
> without rebinding back to cgrp_dfl_root). So move the call to
> cgroup_bpf_inherit above rebind_subsystems in the cgroup_setup_root.
>
> Fixes: 04f8ef5643bc ("cgroup: Fix memory leak caused by missing cgroup_bpf_offline")
> Signed-off-by: Chen Ridong <chenridong@xxxxxxxxxx>
Hm, I always thought that BPF progs can only be attached to the default
hierarchy (cgroup_bpf_prog_attach/cgroup_get_from_fd should prevent
that).
Thus I wonder whether cgroup_bpf_inherit (which is more like
cgroup_bpf_init in this case) needs to be called no v1 roots at all (and
with such a change, 04f8ef5643bc could be effectively reverted too).
Or can bpf data be used on v1 hierarchies somehow?
Thanks,
Michal
Attachment:
signature.asc
Description: PGP signature
