Re: [PATCH v3 5/5] docs: Update info with grub.cfg netboot selection order

[Javier Martinez Canillas]

On 10/22/19 4:26 PM, Daniel Kiper wrote:
> On Tue, Oct 22, 2019 at 10:48:14AM +0200, Javier Martinez Canillas wrote:
>> From: Robert Marshall <address@hidden>
>>
>> Add documentation to the grub info page that specifies the order
>> netboot clients use to select a grub configuration file.
>>
>> Signed-off-by: Robert Marshall <address@hidden>
>> Signed-off-by: Javier Martinez Canillas <address@hidden>
> 
> Wow! We have a doc! Nice!
>

Yes, sorry for missing we had this and not including in the previous versions.

[snip]

>> +
>> +@example
>> +@group
>> +@samp{(FWPATH)}/grub.cfg-@samp{(UUID OF NIC)}
> 
> Hmmm... May each NIC have different UUID? I thought that UUID
> is assigned to the machine not to the NIC. However, there is
> a chance that I am missing something...
>

I think you are correct. The Client machine UUID option (97) isn't documented
in RFC 2132, but in RFC 4578 and it just says the following about the UUID:

2.3.  Client Machine Identifier Option Definition

   The format of the option is:

                Code  Len  Type  Machine Identifier
               +----+-----+----+-----+ . . . +-----+
               | 97 |  n  |  t |     | . . . |     |
               +----+-----+----+-----+ . . . +-----+

   Octet "t" describes the type of the machine identifier in the
   remaining octets in this option. 0 (zero) is the only value defined
   for this octet at the present time, and it describes the remaining
   octets as a 16-octet Globally Unique Identifier (GUID).  Octet "n" is
   17 for type 0.  (One definition of GUID can be found in Appendix A of
   the EFI specification [4].)

   This option MUST be present in all DHCP and PXE packets sent by PXE-
   compliant clients and servers.

Is the Client machine ID option (61) from RFC 2132 that says that a client ID
MAY consistent of a hardware type and address. But it says that a hardware
hardware type of 0 could also be used when the value field contains and ID
that is not a hardware address. And that is up to the the vendors and system
administrators to choose an ID that meets the requirement for uniqueness.

In other words, it MUST be unique and MAY contain a hardware address but that
may not necessarily be the case. I'm not the author of this patch but I guess
that's the source of the confusion.

I'll just change to mention that is the UUID of the machine and point to
the RFC 4578.

[snip]

>> +@example
>> +@group
>> +@samp{(FWPATH)}/grub.cfg-7726a678-7fc0-4853-a4f6-c85ac36a120a
>> +@samp{(FWPATH)}/grub.cfg-52-54-00-ec-33-81
>> +@samp{(FWPATH)}/grub.cfg-0A000082
>> +@samp{(FWPATH)}/grub.cfg-0A00008
>> +@samp{(FWPATH)}/grub.cfg-0A0000
>> +@samp{(FWPATH)}/grub.cfg-0A000
>> +@samp{(FWPATH)}/grub.cfg-0A00
>> +@samp{(FWPATH)}/grub.cfg-0A0
>> +@samp{(FWPATH)}/grub.cfg-0A
>> +@samp{(FWPATH)}/grub.cfg-0
>> +@samp{(FWPATH)}/grub.cfg
>> +@end group
>> +@end example
>> +
> 
> Is it possible to assign both IPv6 and IPv4 addresses at once to the NIC?

By looking at the networking code I believe that it's possible, although I
have only tested this patch-set with IPv4.

> What will happen if both are assigned? And first of all: does GRUB at

The grub_net_search_config_file() function iterates over all the interfaces
using the FOR_NET_NETWORK_LEVEL_INTERFACES() macro. So it will try to lookup
a grub.cfg with the suffix for the IPv4 or IPv6 address depending on the order
in which the interfaces were registered.

> this point support IPv6?
>

My understanding is that GRUB does support IPv6. But for example DHCPv6 isn't
supported. I've posted the patches we have to add DHCPv6 to bootp and efinet.

> Daniel
> 

Best regards,
-- 
Javier Martinez Canillas
Software Engineer - Desktop Hardware Enablement
Red Hat