Hi all, after long and painful debugging session I finally managed to get FCoE VN2VN running on a point-to-point connection. There wasn't so much a single issue, but rather an entire slew of rather annoying problems: - rport kref handling is bonkers. Taking a kref once and releasing it when attempting to free the structure is just not what krefs are designed for. - The FCoE stack reacts allergic to out-of-order frames. While in theory this shouldn't happen, it's hard to avoid with modern HW. Plus I even had been seeing this when using virtio. And to make matters worse, it would occasionally just drop any out of order frames, while handling them correctly in other cases. - Failed FC REC/ABTS will cause an I/O error. There are perfectly legit cases where a REC might fail (eg if the exchange was already sent and it just got stuck in the RX queue), so it should not considered an error. A simple retry will be sufficient. - FLOGI are handled ... interestingly. If both ends attempt an FLOGI the state machine is advanved on the incoming FLOGI request, so the stack gets confused when the (original) FLOGI ACC is being returned. Additionally I've added lots of debugging statements and some general cleanup. As usual, comments and reviews are welcome. The entire patchset can be found at git://git.kernel.org/hare/scsi-devel/h/vn2vn Hannes Reinecke (22): libfc: Revisit kref handling libfc: additional debugging messages fcoe: FIP debugging libfc: spurious I/O error under high load libfc: Do not attempt to login if the port is already started libfc: Debug PRLI failures fcoe: filter out frames from invalid vlans fcoe: make R_A_TOV and E_D_TOV configurable libfc: use configured lport R_A_TOV when sending exchange libfc: use configured e_d_tov for remote port state retries fcoe: inhibit writing invalid values into the 'enabled' attribute libfc: don't fail sequence abort for completed exchanges libfc: do not overwrite DID_TIME_OUT status libfc: use error code for fc_rport_error() libfc: frame alloc failure messages fc: add missing ELS explanation values libfc: Send LS_RJT responses on frame allocation failure libfc: don't advance state machine for incoming FLOGI libfc: Implement RTV responder libfc: Do not drop out-of-order frames libfc: reset timeout on queue full fcoe: set default TC priority drivers/scsi/fcoe/fcoe.c | 21 ++- drivers/scsi/fcoe/fcoe_ctlr.c | 63 +++++++- drivers/scsi/fcoe/fcoe_sysfs.c | 83 +++++++++- drivers/scsi/libfc/fc_exch.c | 64 ++++++-- drivers/scsi/libfc/fc_fcp.c | 131 ++++++++++++---- drivers/scsi/libfc/fc_rport.c | 349 ++++++++++++++++++++++++++++++----------- include/scsi/libfc.h | 8 +- include/uapi/scsi/fc/fc_els.h | 17 +- 8 files changed, 584 insertions(+), 152 deletions(-) -- 1.8.5.6 -- To unsubscribe from this list: send the line "unsubscribe linux-scsi" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
