> > On Wed, Jan 10, 2018 at 09:17:22AM +0200, Leon Romanovsky wrote: > > On Tue, Jan 09, 2018 at 01:09:17PM -0700, Jason Gunthorpe wrote: > > > On Tue, Jan 02, 2018 at 10:18:32AM +0200, Leon Romanovsky wrote: > > > > > > > + /* PID == 0 means that this QP was created by kernel */ > > > > + if (qp->res.pid && nla_put_u32(msg, > > > > + RDMA_NLDEV_ATTR_RES_PID, qp- > >res.pid)) > > > > > > This is returning a pid in the init name space, obtained here: > > > > > > + res->pid = task_pid_nr(current); > > > > > > And since the netlink user is not running in the init name space this > > > will return the wrong pid #, and worse potentially pids the current > > > name space should not see. > > > > > > This API also needs to filter the results and only return pids > > > visible, and translate the pids as well.. > > > > Correct, PID namespace wasn't taken into account, exactly as it wasn't > > taken in CMA. So, right now, CMA netlink statistics is returning wrong > > and unfiltered PIDs. How do you want to progress with that part of the code? > > It should be fixed or deleted. > > > I personally have no plans to fix CMA netlink code and for my opinion it > > should be removed, instead of beating that dead horse. > > What is so wrong with it you can't use it anyhow? > We certainly need to be able to get detailed cm_id information. At least cm_id bound device, cm_id state, port num, qp-id associated with this cm_id, and src/dst ipaddrs/ports. However, it could be fetched and added to qp attributes. Ala iw_cxgb4's debugfs "qps" file: [root@stevo3 stevo]# cat /sys/kernel/debug/iw_cxgb4/0000\:04\:00.4/qps rc qp sq id 1136 rq id 1137 state 1 onchip 0 ep tid 12028 state 7 172.16.4.3:58725/58725->172.16.4.4:7174/7174 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Steve. -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
