On 04/06/20 23:05, Mike Christie wrote:
There is actually another isid fix needed for the core pr code: https://patchwork.kernel.org/patch/10525287/ Handling Bart's review comment for that patch ended up being crazy because of other issues in the PR code so I have not completed that fix.
Sorry for coming back to this so lately. I again had a look into your old patch. To me it seems to be clear that apart from architectural discussion a fix is needed here. The usage of get_unaligned_be64() is of course just wrong. Replacing with hex2bin I think is also not ideal, because hex2bin does not respect byte ordering of the ISID: in the protocol messages iSCSI receives 48 bit big endian ISID. On little endian machines hex2bin() does not do the byte swap necessary to have the same value in string format and in u64. BTW: I'm wondering if the double storage of ISID as u64 and as a string should be kept. I think I'd prefer to use u64 only. In that case lio_sess_get_initiator_sid() could just return s64 swapped to the right enianness and the isid values in session and pr could be s64 (with -1 meaning "no ISID" in session). BR, Bodo
