The Wikipedia article on Disk Encryption Theory mentions this
a powerful adversary can modify an unused sector on the disk and then request its decryption.
I understand that an adversary is capable of modifying an unused sector, but can anyone explain how they would be able to request its decryption?
I understand that an adversary is capable of modifying an unused sector, but can anyone explain how they would be able to request its decryption?
The first thing to understand here is that this is an adversary model. In cryptography we model our adversaries so we know exactly what capabilities they have and which they don't. In this particular case it seems whoever came up with this model decided that it would be a good idea to give the adversary the ability to change the ciphertext of and decrypt unused sectors.
Now we could model our adversary so that they don't have this capability. But in general we want to come up with models that are as powerful as possible so that if we prove a scheme secure in that model, it provides the maximum protection possible. The only limit to the powers of adversaries that we usually use is we don't give them powers that always break any scheme we could possible come up with. E.g. when we encrypt a challenge bit for the adversary we do not allow the adversary to ask for the decryption of that specific ciphertext.
So the default in cryptography is to always add powers right until we barely don't get trivial breaks of security. Now for this specific capability. It seems to be useful because it utterly breaks the ability of the disk encryption to use an unauthenticated deterministic stream cipher (e.g. initialized / keyed by the sector index). This is because an adversary can just look at the ciphertext, request a decryption and then compute any such static keystream as an XOR. The keystream could then later be used if the sector becomes used at some point to extract data and break security. Additionally if the scheme allows you to get some value out of knowing plaintext-ciphertext pairs (e.g. block-indexing parameters), getting them from an unused sector may prove useful, especially if the decryption capabilities of used sectors are somehow limited.
For why the model assumes adversaries can read the decrypted version of the sectors they can change (meaning they get access to a sector decryption oracle), see SEJPM's answer.
For how adversaries can get read access to data they should not: these things have happened, and history repeats itself. In my teenage, I got access¹ to a business school's timeshare system running ICL's MAXIMOP. When a file was deleted from Basic, that seemed for good. But when you used the Fortran compiler, you could open a file, set its size within quota, then read it from start, and you'd get whatever was on disc, often including your file. I wrote a program that could locate blocks containing a given string, and scavenge nearby ones. The same trick worked on the DEC PDP-11 timeshare system at my engineering school, and was a security issue².
¹ The business school was on the path from home to high school. I walked in, found a student login on the blackboard, and quickly it became my favorite place to compute decimal of π per Machin's method, and do my first implementation of RSA (mod was hard!). Students asked question about their assignment, I answered. I was the guy who knew where the paper rolls for the ASR-33 are, how to feed it, and all about the tape punch. Students, TAs and teachers where happy, and the few questions never went incisive.
² A teacher/sysop found it convenient to have a universal password hard-coded in the login program, allowing him to connect on student (not root) accounts to help them. When editing the source of that program, the previous version was deleted, and the password in clear there. At one stage someone with temporary root access replaced the login program with an executable including another permanent password granting root access. Three aspiring engineers interested in computers can keep such secret, when two are dead. A guy used that method to alter his low marks, that was noticed, and as the president of the computer club I was called in this mess. Some months after, a newspaper (France Soir IIRC) got word of the story, threw ink on paper complete with my first and last name within a 1-letter typo, stating I had obtained a fake diploma from my engineering school by altering my marks. I learned about the article from my grand-father in law, and had to deploy effort to clear up my name in my own family. In the internet age, that would have been much worse. I wear a white hat ever since.
