Re: [PATCH rfc 2/2] fabrics: allow user to pass hostname instead of traddr

[Max Gurtovoy <maxg@mellanox.com>]

On 3/24/2020 9:11 PM, Sagi Grimberg wrote:
>
>>>>>   static int connect_ctrl(struct nvmf_disc_rsp_page_entry *e)
>>>>>   {
>>>>>       char argstr[BUF_SIZE], *p;
>>>>> @@ -1230,6 +1283,13 @@ static int discover_from_conf_file(const 
>>>>> char *desc, char *argstr,
>>>>>           if (cfg.persistent && !cfg.keep_alive_tmo)
>>>>>               cfg.keep_alive_tmo = NVMF_DEF_DISC_TMO;
>>>>> +        if (cfg.hostname) {
>>>>> +            ret = hostname2traddr(&cfg);
>>>>> +            if (ret)
>>>>> +                goto out;
>>>>> +            cfg.hostname = NULL;
>>>>> +        }
>>>>> +
>>>>
>>>> I guess you add the NULL to check mutual exclusion later ?
>>>
>>> The NULL assignment here is because we are looping on the conf
>>> file entries which can have a different hostname. The other
>>> locations are a one-shot so this assignment is not needed.
>>>
>>>> Can we check it in build_options ?
>>>
>>> When we get a hostname we resolve the traddr which is what
>>> build_options needs (hostname not an option that is passed to
>>> the driver).
>>>
>>> Does that answer your question?
>>
>> I meant to prevent a user to send them both in the command line (no 
>> to send it to the driver).
>
> build_options is just converting options to an argstr, not sure its a
> good idea to move that there...

You right, I was just thinking out loud.

We can check it before calling the resolution function hostname2traddr.

.


_______________________________________________
linux-nvme mailing list
linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme