how to get valid blocks in vmdk extent file

[jicheng_08]

In Esxi 6.0, I create a vm in NFS store, every virtual disk of the vm has disk_name.vmdk  and  disk_name-flat.vmdk.

disk_name.vmdk:

# Disk DescriptorFile

version=3

encoding="UTF-8"

CID=164e2cb0

parentCID=ffffffff

isNativeSnapshot="no"

createType="vmfs"

# Extent description

RW 8388608 VMFS "disk_name-flat.vmdk"

# The Disk Data Base

#DDB

ddb.adapterType = "lsilogic"

ddb.deletable = "true"

ddb.geometry.cylinders = "522"

ddb.geometry.heads = "255"

ddb.geometry.sectors = "63"

ddb.longContentID = "2cd42eda8ae3ff5540338b8f164e2cb0"

ddb.thinProvisioned = "1"

ddb.uuid = "d7 07 97 90 a3 6a 48 30-a0 fc d9 55 b9 6d ba 1f"

ddb.virtualHWVersion = "11"

I find https://www.vmware.com/support/developer/vddk/vmdk_50_technote.pdf , vmdk file format is discribled.

But I find my disk_name-flat.vmdk is inconsistent with pdf file.

The beginning of the disk_name-flat.vmdk is zero bit.

But in vmdk_50_technote.pdf , the beginning of the file may be ESXi Host Sparse Extent Header, why?

My real purpose is to get valid data on disk by parsing the contents of the vmdk file ,  I hope someone can help me, thanks very much!

[continuum]

Flat.vmdks are not virtual disks - they dont use a header like VMFSsparse, Sesparse or Workstation VMDK sparse types.

Treat them like dd images instead.

[jicheng_08]

Thank you very much.

I suppose the flat.vmdk can be prased accord to virtual disk format 5.0 specific, and I can get the valid disk blocks offset (really used by vms), so I can only backup this valid disk blocks. If the flat.vmdk can not be prased . I think I can use vddk 6.7 QueryAllocatedBlocks interface or vsphere api QueryChangedDiskAreas to get disk valid block offset.

Is there any other way to try ?

[continuum]

Forget parsing flat.vmdks.

To develope a backup tool look into ctk.vmdks.

[hwangwenchao]

How did you solve it? At present, I also encountered the same problem. Many data of VMDK in NFS are useless. I am currently parsing the vmdk file, and then filtering out all the invalid data, but it is too slow.