Four days ago, a colleague of mine appropriated the old Linux box I wasn’t using anymore. He needed to run Windows-based instrumentation software, and I, basking in my smug confidence that all of my research code was mirrored on Github, because only an idiot wouldn’t back things up, put on my most magnaminous, cavalier smile and went “sure man, go ahead and wipe it, I got a copy of everything”.

Of course, I had no copy.

I also didn’t have a spare terabyte-sized hard drive lying around to pull an image onto, for later file carving attempts. So I had to work from a USB-booted Live Xubuntu and access the unmounted NTFS (previously ext4) partition directly. Here’s what I learned.

• SleuthKit/Autopsy are very popular. But unless I missed something, they require an image file, and that wasn’t going to happen.
• PhotoRec ran briefly, but didn’t produce anything useful, and then asked me whether I wanted to store the remaining (?) 950 GB as an image file. No thank you.
• extundelete refused to run at all, because the partition was now NTFS.
• TestDisk (i.e. attempting to restore the ext4) was out of the question, because I couldn’t risk ruining my colleague’s Windows system.

Having exhausted the popular open source options, I searched on and found R-Linux, an obscure freeware tool developed in Richmond Hill, Ontario, just a town over from here. It has a pleasant GUI and, after scanning my hard drive overnight, it recognized the old ext4 partition, starting exactly 986830848 sectors in.

A double-click on said partition reveals all of the old inodes found on the partition. You can open said inodes and look at their content—it’s binary junk, most of it; punctuated by the occasional image or text file. At any rate, it’s impossible to look through all of them by hand, and for some bizarre reason R-Linux, in all its beauty, offers no way of searching through the inodes’ contents (at least the “Find/Mark” function doesn’t appear to be doing that).

There is another, old-school way to search, though: grep. And it works suprisingly well:

sudo dd if=/dev/sda2 skip=986830848 bs=512 count=937498624 | grep --text -A 1000 -B 10 'import numpy as np' | tee results

First, we use dd to read out the raw bytes on /dev/sda2. We start at sector 986830848, we specify a sector (here: block) to have a size bs of 512 bytes each, and we read out 937498624 sectors (all of those were given to us by R-Linux). The output is piped to grep --text. The -A and -B arguments ask grep to spit out not only the lines matching import numpy as np (I was looking for Python code), but also 10 lines before and 1000 lines after. Finally, tee lets us see the output on the screen, while also saving it to a text file called results.

I recovered three of my five Python files this way. Some of them were found several times over, in various garbled and outdated incarnations. Two of them are lost, probably, forever. But that’s okay. I suppose I had to learn my lesson.