On Thu, Dec 19, 2019 at 06:31:12PM -0500, Martin K. Petersen wrote: > > Michal, > > >> Please don't introduce a blist flag to work around deficiencies in the > >> matching interface. I suggest you tweak the matching functions so they > >> handle a NULL vendor string correctly. > > > > I don't think that will work with the interface for dynamically adding > > entries through sysfs. > > Please make it work :) > > There's nothing conceptually wrong with being able to do: > > echo ":Model:Flags" > /proc/scsi/device_info > > We keep running into issues where the same device needs to be listed > many times because it gets branded by different vendors. Actually the flag is really needed. The vendor string is a field of specific length, not a pointer. This makes sense because the vendor string is fixed length. The code adding the blacklist entries cannot handle NULL, and the matching code works with char array already. What can be done differently is stop space-padding the values and instead match the length of the string as provided. This will however cause API break. Currently short entries are space-padded to match exactly the provided vendor string. If we change the match to only the provided string length it will potentially match and blacklist additional devices. If a flag is required to trigger the prefix matching it can be used instead of the flag that disables vendor matching with empty vendor string. Thanks Michal
