Hi,

quoted

quoted

I'm a bit concerned about things inadvertently breaking if we return an empty list for the well known LUNs.

According to SPC we shall report an empty list if there is no well-known LUNS.
FreeBSD has the same logic in REPORT LUNS handling. SCST does not support SELECT_WELLKNOWN case at all.

I don't know the history of the existing behaviour to send always LUN0 instead of empty list. Probably it was
for the SCSI_SELECT_ALL_ACCESSIBLE(0x02) case, where SPC allows LUN0. My patch keeps it for the 0x00, 0x02, 0x11 cases.
Thus, I believe it does not break the backward compatibility.

Will this change require users to update their LUN configuration? Some 
initiator operating systems require presence of a dummy LUN 0. Although 
I agree that it is cleaner not to provide a hardcoded LUN 0, I think 
Martin is concerned about this patch potentially breaking existing 
configurations and causing frustration among LIO users.

No reconfiguration on initiator side is required.
W-LUN is a specific LUN for the specific SCSI target device that is well known
for the Initiator. Generic Linux TCM does not have W-LUNs. Some storage
systems based on Linux TCM may have W-LUNs. But then they shall / already have
its own handling of REPORT LUNS command.
Basically, it is an error to report LUN0 as W-LUN for the Initiator that
expects some other numbers.

BR,
 Dmitry