gpt: point alternate PartitionEntryLBA to alternate GPT table

[tobhe]

Currently both tables point to the primary GPT partition entry array which doesn't make a lot of sense. Pass "primary" to t.partitionArraySector() to get the right address for each of the tables instead.

I found this because I was building images for a Chrombook and noticed that cgpt always reported the secondary GPT header as corrupted.

For comparison here is the cgpt show output without this fix for a minimal ESP example:

       start        size    part  contents
           0           1          PMBR
           1           1          Pri GPT header
                                  Sig: [EFI PART]
                                  Rev: 0x00010000
                                  Size: 92 (blocks)
                                  Header CRC: 0x56d22730 
                                  My LBA: 1
                                  Alternate LBA: 212991
                                  First LBA: 34
                                  Last LBA: 212958
                                  Disk UUID: F4206CB2-52F5-4BEA-B903-62A6CC3DDAFB
                                  Entries LBA: 2
                                  Number of entries: 128
                                  Size of entry: 128
                                  Entries CRC: 0xc5a89a28 
           2          32          Pri GPT table
        2048      200706       1  Label: "EFI System"
                                  Type: EFI System Partition
                                  UUID: C2BDE81C-A933-421E-B854-A86FFD229D8E
           2          32 INVALID  Sec GPT table
           1           1 INVALID  Sec GPT header
                                  Sig: [EFI PART]
                                  Rev: 0x00010000
                                  Size: 92 (blocks)
                                  Header CRC: 0x995343f3 
                                  My LBA: 212991
                                  Alternate LBA: 1
                                  First LBA: 34
                                  Last LBA: 212958
                                  Disk UUID: F4206CB2-52F5-4BEA-B903-62A6CC3DDAFB
                                  Entries LBA: 2
                                  Number of entries: 128
                                  Size of entry: 128
                                  Entries CRC: 0xc5a89a28 INVALID

And in comparison the fixed version with the secondary table pointing to the secondary partition list:

       start        size    part  contents
           0           1          PMBR
           1           1          Pri GPT header
                                  Sig: [EFI PART]
                                  Rev: 0x00010000
                                  Size: 92 (blocks)
                                  Header CRC: 0x2dd2b955 
                                  My LBA: 1
                                  Alternate LBA: 212991
                                  First LBA: 34
                                  Last LBA: 212958
                                  Disk UUID: EF0BACC6-F30A-4455-9E3E-96AE05F78C6A
                                  Entries LBA: 2
                                  Number of entries: 128
                                  Size of entry: 128
                                  Entries CRC: 0x31d5ac19 
           2          32          Pri GPT table
        2048      200706       1  Label: "EFI System"
                                  Type: EFI System Partition
                                  UUID: 22712D7B-7398-4E20-8484-1FF7F73C3233
      212959          32          Sec GPT table
        2048      200706       1  Label: "EFI System"
                                  Type: EFI System Partition
                                  UUID: 22712D7B-7398-4E20-8484-1FF7F73C3233
      212991           1          Sec GPT header
                                  Sig: [EFI PART]
                                  Rev: 0x00010000
                                  Size: 92 (blocks)
                                  Header CRC: 0x76129696 
                                  My LBA: 212991
                                  Alternate LBA: 1
                                  First LBA: 34
                                  Last LBA: 212958
                                  Disk UUID: EF0BACC6-F30A-4455-9E3E-96AE05F78C6A
                                  Entries LBA: 212959
                                  Number of entries: 128
                                  Size of entry: 128
                                  Entries CRC: 0x31d5ac19 

[tobhe]

Just noticed that this was recently changes in https://github.com/diskfs/go-diskfs/pull/181. Unfortunately that PR doesn't have a lot of context so I don't really understand what motivated this.

[deitch]

This is a good catch, thank you. Unfortunately, with a year plus of time since #181, I do not recall either. 🤦‍♂️

The more I think about it, the more that change doesn't make sense to me. It should be the following:

• For primary header: starting LBA of partition entries = LBA(primary header) + 1
• For secondary header: starting LBA of partition entries = LBA(secondary header) - (size of entries table)

The above is what is provided by partitionArraySector(primary bool). So why was it changed? I have no idea.

Can you make the change locally to revert that apparently erroneous #181 and confirm that it works correctly? If so, I will then commit the revert.

[deitch]

Or does your comment "the fixed version with the secondary table pointing to the secondary partition list" mean, "I tested it with this commit and it now works correctly"?

[tobhe]

I tested with the commit included in this PR which should be equivalent to a revert of #181 and resulted in the second (correct) output of cpgt show I included above. I didn't test with an actual revert because I only found the responsible commit after I found the fix, but since the fix literally reverses the commit that broke it I think that should be good enough.

[deitch]

No need for a revert. I like a clear new commit better for things like this. You say it is good, all tests pass, let's merge it in.

[deitch]

Do you think cgpt could be used as a standalone test? Create a gpt filesystem in a file, run cgpt on it to validate it?

[tobhe]

Sure, that sounds like a good idea. I did write a small program that does that anyway. Let me see if I can clean it up and hook it up to the test infrastructure.