[dm-devel] [PATCH v2 10/17] libmultipath: add code to get vendor specific vpd data

[Martin Wilck]

On Mon, 2020-02-10 at 14:49 -0600, Benjamin Marzinski wrote:
> On Mon, Feb 10, 2020 at 02:49:38PM +0000, Martin Wilck wrote:
> > On Wed, 2020-02-05 at 12:58 -0600, Benjamin Marzinski wrote:
> > > This adds the wildcard 'g' for multipath and path formatted
> > > printing,
> > > which returns extra data from a device's vendor specific vpd
> > > page.  The
> > > specific vendor vpd page to use, and the vendor/product id to
> > > decode
> > > it
> > > can be set in the hwentry with vpd_vendor_pg and vpd_vendor_id.
> > > It
> > > can
> > > be configured in the devices section of multipath.conf with the
> > > vpd_vendor parameter. Currently, the only devices that use this
> > > are
> > > HPE
> > > 3PAR arrays, to return the Volume Name.
> > > 
> > > Signed-off-by: Benjamin Marzinski <bmarzins at redhat.com>
> > > ---
> > >  libmultipath/config.c      |  2 ++
> > >  libmultipath/config.h      |  1 +
> > >  libmultipath/dict.c        | 38
> > > ++++++++++++++++++++++++++++++++++++
> > >  libmultipath/discovery.c   | 40
> > > +++++++++++++++++++++++++++++++++++++-
> > >  libmultipath/hwtable.c     |  1 +
> > >  libmultipath/print.c       | 25 ++++++++++++++++++++++++
> > >  libmultipath/propsel.c     | 18 +++++++++++++++++
> > >  libmultipath/propsel.h     |  1 +
> > >  libmultipath/structs.h     | 15 ++++++++++++++
> > >  multipath/multipath.conf.5 |  8 ++++++++
> > >  10 files changed, 148 insertions(+), 1 deletion(-)
> > > 
> > >  struct sg_id {
> > >  	int host_no;
> > >  	int channel;
> > > @@ -255,6 +268,7 @@ struct path {
> > >  	char rev[PATH_REV_SIZE];
> > >  	char serial[SERIAL_SIZE];
> > >  	char tgt_node_name[NODE_NAME_SIZE];
> > > +	char vpd_data[VPD_DATA_SIZE];
> > 
> > Hm, 128 more bytes per path for a rarely-used property... do we
> > need to
> > store this in "struct path"? Can't we simply fetch that information
> > when someone requests it with the %g format wildcard? It doesn't
> > matter
> > much today as "struct path" is really thick already, but I have
> > hopes
> > to make it a bit leaner some day.
> 
> Well, the thought was that this information (which should help map
> the
> path to a physical array) would most often be wanted when things were
> going badly, and you might not be able to access the device.

Point taken. How about using a pointer? That would be only 8 extra
bytes for those cases where the field was unused.
Call me old-fashioned, I've grown up in a world where 1000x128 bytes
mattered.

Regards,
Martin