On Thu, 2024-12-19 at 21:02 -0500, Benjamin Marzinski wrote: > When multipathd handles link integrity FPIN events, it sets the rport > associated with the effected SCSI paths to Marginal. These patches > make > it do the same thing for NVMe paths. > > I actually have a stupid Fibre Channel question. Does the host number > and the target wwpn uniquely identfy an rport? I never see the > channel > set to anything other than 0, but I don't think that it MUST be 0. IMO, yes. As the WWPN is unique, it's by itself sufficient to identify the physical remote port. By adding the host number, which corresponds to a unique WWPN on the host, you uniquely identify the Linux fc_remote_port. This also fits the fc symlinks that udev creates under by-path. > But > if we knew that there was only one rport for a given host number and > target WWPN, we could stop checking as soon as we found one. These > patches don't do that. They check all the rports. I think we could stop if the WWPN is found. Martin > > Benjamin Marzinski (2): > libmultipath: export udev pthread cleanup functions > multipathd: set rport port_state to marginal for NVMe devices > > libmultipath/discovery.c | 4 +- > libmultipath/discovery.h | 2 + > libmultipath/libmultipath.version | 6 +++ > multipathd/fpin_handlers.c | 74 ++++++++++++++++++++++++++++- > -- > 4 files changed, 78 insertions(+), 8 deletions(-) >
