On Fri, Sep 09, 2022 at 12:24:51PM -0500, Mike Christie wrote:
>
> On 9/9/22 6:32 AM, Konstantin Shelekhin wrote:
> >> The patch solves the presence of non-deletable empty default_tg_pt_gp
> >> in RTPG.
> >> May be, a global attribute like target/core/alua/hide_emtpy_tpg would
> >> fit better than an attribute per each port group?
> >>
> >> I would always hide the empty default_lu_gp (not configurable) but I am
> >> afraid that it will be considered as not backward compatible change. 🙁
> > A module parameter perhaps? Or a CONFIG definition.
>
> For the ceph iscsi project we wanted this same behavior for a while and
> we had to use distro kernels. There are probably others that need the same
> thing so a kernel config option wouldn't work for them.
>
> Module param or a global attr in target/core/alua like Dimitry mentioned
> seem fine. If the new variable is set are you guys thinking that
> core_tpg_add_lun would just not call target_attach_tg_pt_gp? So the variable
> would be "make_default_tg_pt_gp"?
I thought it over one more time.
1. To not report empty port group is a completely backward compatible
change becasue there is no impact on userspace at all. The only change
is in the network response.
2. SPC-4 ("5.15.2.7 Target port asymmetric access state reporting")
tells that a target MAY not provide info about port groups that do not
contain the current port through that the RTPG is received.
So, according to SPC it is expected behaviour to not report the empty
port groups.
I will prepare new version of the patch with always skipping any empty
port group in RTPG response.
BR,
Dmitry
