Hacker News Reader

Comment by ajb | Hacker News Reader

So this always overwrites the device to test it, and I was wondering if that was necessary (say, you had started using it). Presumably you only need to write at most N+1 blocks, where N is the number of blocks the device actually has.l, in order to detect exaggeration. But at that point the fake device will have overwritten all of your files anyway, even if they were theoretically on different blocks (of the exaggerated inventory). So I guess the minimum harm to test a device is to hash all your files, then write at most N+1 (unused) blocks, stopping after each to check if any of your files got harmed. In theory that risks at most one block...

Of course, it's better to back up the suspect one.

View on HN

> hash all your files, then write at most N+1 (unused) blocks, stopping after each to check if any of your files got harmed

That strategy adds O(n^2) reads on top of O(n) writes, though.

Even reads don't come for free on modern multi-level cell NAND (due to read disturb), and for just a thousand blocks, you'd end up reading the first block a million times.

That's to say nothing of the time this would take.

You're right, this is a non-starter. I should stop posting late at night. The owner of a suspect device just needs to bite the bullet and use a destructive method.

This also has the option of f3probe: https://fight-flash-fraud.readthedocs.io/en/latest/usage.htm...

You are supposed to check an empty device. Some of the fake ones have firmware that will silently delete files or else fake writes. If you load it with data before confirming it is legit, you are likely to lose that data.

For a new device (ie no existing files on it), wouldn't the simplest approach be to full block 1 (whether 512 or 4k bytes) with a series of "1"'s, block 2 with a series of "2"'s, (etc). ie incrementing the number that gets written as the block number being written to is written.

Reading that back (either the full device or a random sample) should pretty quickly identify whether things are still in their expected location.

Well, with the remaining trust available at this point you might just as well use something cryptographically secure, like encrypted ones, twos, or simple HMACs of the block number.

A too-simple scheme is likely to be detected (and bypassed!) by the firmware a nearly no time.

Sure, that's a decent idea too. :)

Simpler: fill the drive with random data, hashing as you go, flush the kernel's buffer cache, hash the entire contents of the drive, and compare.

Conceptually,

  # tee /dev/DEVICE </dev/random | sha256sum
  # echo 1 > /proc/sys/vm/drop_caches
  # sha256sum /dev/DEVICE

though I wouldn't expect this exact command sequence to work unless tee's buffer size divides /dev/DEVICE's capacity and tee errors out writing past the end of /dev/DEVICE before writing to stdout.

I did exactly this earlier last week

the drive size divided by 4MB, so dd with bs=4M and fixed count

(with oflag=direct you don't even need to drop caches)

Oh, that's a clever way of doing things.

The "write the block #'s to the given block" would help identify where a fraudulent device goes wrong.

But for just checking if a device is storing data 100% correctly then your way would probably be more robust. :)

Reply on news.ycombinator.com