(This is where you need to start copying from the end.) Move the blocks composing the source file system so that they start where the first partition would start.Calculate where the first partition would start (without actually creating the partition table).Resize the source file system as much as possible.With a tool such as ddrescue which can copy in reverse, you could simplify this slightly: Resize the file system again so it occupies all the partition.

You’ll need to calculate the offsets and sizes based on what you did in step 2, but the target is easy ( /dev/sdX1).

Copy the raw blocks from the second half of the disk to the newly-created partition, again using dd or something similar.

Use GPT for this you’ll be overwriting blocks at the start and end of the disk, so the copy of the file system at the start of the disk is now toast.

Partition the disk, creating one partition which is slightly larger than the file system (500G).

Copy the blocks to the second half of the disk, using dd or a similar tool.

Ideally you should reduce the file system as much as possible, that will reduce the amount of data you need to copy around in subsequent steps. fits within the first 499G’s worth of blocks on the disk. This ensures that all the data, file system structure etc.

Resize the ext4 file system to 499G (erring on the safe side).

Working with a terabyte disk, I would proceed as follows. Partition tables (at least in MBR or GPT style) live at the start and/or end of disks, so you can introduce them if you can free the required space.

Crystaldiskinfo safe to share info