On 07/31/2015 05:31 AM, Przemyslaw Marczak wrote:

Hello,

On 07/30/2015 01:19 PM, Palacios, Hector wrote:

...

Hello,

Commit d718ded056eefb6239bd2e0a57b7f6d99c6e9e4b introduced translation of UUID binary data to GUID string representation.

So, for example, if I use the 'gpt' command to create a partition table and pass a 'uuid' parameter like this: => gpt write mmc 0 "uuid_disk=${uuid_disk};start=2MiB,name=linux,size=64MiB,uuid=43f1961b-ce4c-4e6c-8f22-2230c5d532bd;"

As a result, when I print the partition table I get: => part list mmc 0

Partition Map for MMC device 0 -- Partition Type: EFI

Part Start LBA End LBA Name Attributes Type GUID Partition GUID 1 0x00001000 0x00020fff "linux" attrs: 0x0000000000000000 type: ebd0a0a2-b9e5-4433-87c0-68b6b72699c7 guid: 1b96f143-4cce-6c4e-8f22-2230c5d532bd

which prints the GUID string representation of my supplied UUID (with endian change on the first three parts).

The command 'part uuid' (despite its name) returns this same GUID representation: => part uuid mmc 0 1b96f143-4cce-6c4e-8f22-2230c5d532bd

I have some questions:

• Why is preferred the GUID representation when listing the partition

table?

Because it's defined by the specification and is the same as in the kernel. Please look at Apendix A of UEFI specification[1], which says about GUID text representation, which is simply shown in the table. The binary format of guid is consistent with RFC4122[2], which says that both guid and uuid are binary equal, but different represented by text.

You can check it by printing gpt with:

'mmc part' - let's store the guids somewhere

Then mount the same disk in linux, and next do:

'ls -lhgG /dev/disk/by-partuuid/'

You can see, that even if the name is 'by-partuuid', the ids are displayed as GUIDs. Probably there is a bug with the path name, since uuid is a reserved name for "universally unique identifier".

...

• Should the 'part uuid' return the UUID representation instead of the

GUID?

It returns a string representation of partition id, which is GUID text format.

...

• Should there be a 'part uuid' and a 'part guid' commands that return

the different representations?

I don't think so, because there is only one text format in the standard.

...

• Isn't it a bit inconsistent that the 'gpt' command reads the 'uuid'

parameter in UUID string representation and the 'part uuid' and 'part list' represent the number in GUID?

You're right. This is inconsistent for string comparison - we can't just compare the UUID string with GUID, even if the 16 bytes binary buffer is the same. I don't know why GPT support was introduced with uuid names, we have a functions to convert it in lib/uuid.c, so additional command could be added - but I think cleanup is better.

...

It may all sound as a futile discussion but in v2013.04 I had some variables to store the UUID numbers for my partitions that I used to generate the partition table, and then compared these variables with the values returned by 'part uuid' (as strings). Now on v2015.04 the strings do not match due to this endianness change on the representation.

Thanks

Hector

Thank you for pointing this issue.

After looking into the code, I found a bug in function gpt_fill_header(), which converts string to GUID, but it shouldn't, since we pass UUID's to gpt command.

So what can we do to fix this inconsistency:

• change the variable uuid_* to guid_* in the $partitions. This is a

string representation compatible with specification.

• update str_format argument to UUID_STR_GUID in gen_rand_uuid_str()

calls in gpt command.

Update uuid format argument in uuid_str_to_bin() calls:

• update UUID_STR_FORMAT_STD to FORMAT_GUID in gpt_fill_pte()

• leave UUID_STR_FORMAT_GUID in gpt_fill_header() - it is bug when UUID

is generated in gpt command.

Stephen, is that good for you?

I'm not completely sure what the implications of those changes are without tracking down all those functions. However, if it means zero changes to the text output of the "part list" and "part uuid" commands, then that seems fine.