On Sat, Jun 27, 2020 at 9:24 AM Richard Guy Briggs [off-list ref] wrote:

Provide a mechanism similar to CAP_AUDIT_CONTROL to explicitly give a
process in a non-init user namespace the capability to set audit
container identifiers of individual children.

Provide the /proc/$PID/audit_capcontid interface to capcontid.
Valid values are: 1==enabled, 0==disabled

Writing a "1" to this special file for the target process $PID will
enable the target process to set audit container identifiers of its
descendants.

A process must already have CAP_AUDIT_CONTROL in the initial user
namespace or have had audit_capcontid enabled by a previous use of this
feature by its parent on this process in order to be able to enable it
for another process.  The target process must be a descendant of the
calling process.

Report this action in new message type AUDIT_SET_CAPCONTID 1022 with
fields opid= capcontid= old-capcontid=

Signed-off-by: Richard Guy Briggs <redacted>
---
 fs/proc/base.c             | 57 +++++++++++++++++++++++++++++++++++++++++++++-
 include/linux/audit.h      | 14 ++++++++++++
 include/uapi/linux/audit.h |  1 +
 kernel/audit.c             | 38 ++++++++++++++++++++++++++++++-
 4 files changed, 108 insertions(+), 2 deletions(-)

This seems very similar to the capable/ns_capable combination I
mentioned in patch 11/13; any reasons why you feel that this might be
a better approach?  My current thinking is that the capable/ns_capable
approach is preferable as it leverages existing kernel mechanisms and
doesn't require us to reinvent the wheel in the audit subsystem.


--
paul moore
www.paul-moore.com