Re: [PATCH cifs-utils] mount.cifs: handle multiple ip addresses per hostname

[Paulo Alcantara <pc@cjr.nz>]

Aurélien Aptel <aaptel@suse.com> writes:

> I would put in the commit msg that this requires recent kernel.

Agreed.

> We should probably merge kernel code first so we can reference it here
> in the commit msg, and say in the man page when did the kernel change.

Agreed.

> There can be cases where cifs-utils is more recent than kernel and
> mount.cifs will pass all the ip instead of trying them before passing
> the good one to the kernel but since it's an old kernel it won't try
> them all.

Good point!  Yes, we should handle both cases.

> We could add an option to enable new behavior or check the kernel
> version from within mount.cifs.. thoughts?

I don't like the idea of checking the version because the running kernel
might not have the expected patches.

Perhaps a new option would be better... I'll think more about it.

> Paulo Alcantara <pc@cjr.nz> writes:
>>  
>> +static void set_ip_params(char *options, size_t options_size, char *addrlist)
>> +{
>> +	char *s = addrlist + strlen(addrlist), *q = s;
>> +	char tmp;
>> +
>> +	do {
>> +		for (; s >= addrlist && *s != ','; s--);
>> +		tmp = *q;
>> +		*q = '\0';
>> +		strlcat(options, *options ? ",ip=" : "ip=", options_size);
>> +		strlcat(options, s + 1, options_size);
>> +		*q = tmp;
>> +	} while (q = s--, s >= addrlist);
>> +}
>
> I think you should write this in a clearer way and comment, this is hard
> to read.

That's horrible, indeed.  I'll definitely make it readable in next
version.

> I was going to say should we return errors if we truncate the ips, but
> none of the mount.cifs.c code checks for truncation so I guess we can
> ignore.

IIRC, yes.