87 comments

[ 3.4 ms ] story [ 172 ms ] thread
> A finger is around 1.5MΩ, but wet fingers are around 500kΩ. When the device boots, the drive will appear blank if the resistance between the electrode pair is higher than a threshold. It’s not the prettiest system, but I think it strikes a nice balance of ridiculousness and functionality. The germaphobes among us can run their fingers under the tap.

This may backfire for the hacker.

Authority: Give me the drive

Hacker: Ok. (hands drive)

Authority takes with dry hands and tries it. It doesn't work.

Authority gets upset and starts beating the person.

Authority works up a sweat during the beating

Authority tries drive again - it works

Authority concludes that Hacker's drive only works if Hacker is beaten.

$5 for a wrench really dates that joke
https://www.harborfreight.com/hand-tools/wrenches/adjustable...

Granted it's only 6" long, but hey it's a $5 wrench.

(comment deleted)
You know what they say about "it's not the size of the tool, it's the skill of the interrogator."
Everyone knows that the Snap-on wrench delivers a better beat-down.
Even the alt text on that comic agrees that $5 for a wrench is possible impossibly low
Would be better to have a false volume that contains boring but realistic files without the correct password or trick to unlock.
Unfortunately, it doesn't protect against Rubberhose cryptanalysis.
I think that is a wrong take.

Rubberhose FS does protect you from rubberhose cryptanalysis. That is the point.

You can argue, that it over-protects you, which might be just as or even more dangerous, and that is a fair criticism.

But I don't see how you can argue that it "doesn't protect" you.

It doesn't protect you from rubberhose cryptanalysis being performed on you. It might protect the data if you can resist revealing the ultimate secret until death.
I now see how you can argue that it doesn't protect you, in a very personal meaning of you.

Then, is there anything that can protect __you__ from rubber hose cryptanalysis in this sense?

If not, what does my parent comment even mean?

Well, rubberhose cryptanalysis isn't cryptanalysis. Once you start discussing it, you're not discussing cryptography anymore. There's no cryptosystem that can protect you from it.
If you want your life protected from the rubber hose, you need to be able to allow access to the data. But, that doesn't necessarily protect your life: if they don't like the data they see, then they'll keep trying to extract more keys.

If you want your data to be safe from the rubber hose, you need to have inconvenient access requirements. If your laptop can only be unlocked when 5 keys are turned 'simultaneously', and the keys are held by the ambassadors to the UN with security council vetos, that's going to be hard to accomplish for you, and likely difficult for any one else. Any one of the ambassadors could plausibly lose their key, and then the data is gone, etc.

Veracrypt supports this, called hidden volume.
I was thinking along the lines of setting the highest bit on the address line to either 1 or 0 based on the resistance, and programming the device to report half the capacity of the nand chip.
What if a banana republic journalist suddenly trying to stop the rogue official from deploying a suitcase nuke... has sweaty hands from the stress, while racing to access the USB flash drive with the documents that would locate the nuke in time?
I dunno man, but I can think of at least two whole scenarios more likely than your hypothetical
I don't need self-destructing USB drive, but a USB flash stick with a hard write-protect switch (that can't be overridden by the USB host) would be great for purposes like immutable installation and boot media.

(Also, like any USB device should, but even more here, you have to ensure that the device controller can't be modified by the USB host.)

That would be very useful. We sadly stopped innovating on optical media, but haven't really made the alternatives usable as a replacement. You used to have CDs / DVDs that were:

1. Completely immutable

2. Could not be maliciously altered by the reader

3. Could not masquerade as a keyboard or a mouse

I miss physical media, optical or magnetic. Or something new!! Flash drivers are just computers these days.
In what sense is a nand chip, a memory controller, and a usb plug, a "computer these days"?
The flash memory controller is most likely a embedded stored program computer. Likely an arm, maybe dual core.

A modern computer is really a distributed system, but not a lot of people frame the discussion that way.

You still do. I can't be the only person who burns install DVDs, finalized, verifies the hash on a few different systems, and then keeps them in a safe place for possible future use...

I've got some external USB readers/burners that excel for exactly this sort of thing, and just about anything that will run a modern OS will boot from them.

You're not the only one. I still do this, and use CDs/DVDs for all sorts of other longer-term storage as well. I'm unaware of any other media that works well for those sorts of use cases.
What is the shelf life of a modern CD/DVD? I might consider using them more
At room temperature, about 25 years -- but there are a lot of variables that can affect this in both directions.

https://www.clir.org/pubs/reports/pub121/sec4/

Maybe for commercial-grade optical media, but last time I burned a lot of consumer media , I found a decent amount of failures even just a few years out.

Admittedly this was almost 20 years ago at this point so maybe some things have improved since then but my understanding is that consumer media uses a different, ink-based technology to write data to disk, and that ink can degrade pretty quickly. This is in contrast to commercial optical media in which the data is stamped into the disk in a manner more akin to a vinyl record.

All CD-Rs use a dye layer, yes. The life expectancy the link I provided is talking about is for writeable media, not for read-only media that is stamped.

Some forms of writeable disc media don't use dyes and can last for over 100 years. Check out the link.

But, to your point, one of the variables that affect longevity is the quality of the manufacturing. If you're buying the least expensive media you can find, then the lifespan is indeed shorter. For my cheapo CD-Rs, I've found that they last 5-10 years. You can also make them last longer by storing them in better conditions (vertically, not laying flat, in an airtight container in the freezer).

I tend to go with cheap media, so my personal habit is to burn new copies of my storage discs every 5 years.

Yeah they sell archival-grade CD-Rs. CD-RWs are way more unstable; would not recommend.

One variable I'm recently able to validate (wow, 20 years later!) is write speed. Discs written (for archival) at slower speeds still work. Discs I wrote as fast as possible on my way out the door have mostly failed by now.

Ah interesting, I will check it out. Yea I was definitely buying relatively-cheap media back in the day -- I remember getting large spindles of Verbatim DVD-Rs for about $1/disc.
They make archival grade dvds, I just got some,

Verbatim DVD-R UltraLife Gold Archival Grade 95355

I still use BluRays, but the cost/GB has not kept up with everything else.
Really? What alternative is there for archival quality backups with better $/gb?
Tape.
Only the amortized cost is better for tape if you can fill up lots of tapes. Amazing if you are a data center. But if you don't archive a huge amount of data it's not even close to competitive.
> Completely immutable

Wouldn't the laser of a CD/DVD writer be powerful enough to alter the data after the fact? Not to overwrite it with completely arbitrary data, but flipping a few well-chosen bits should in principle be able to create a backdoor that wasn't there before (if it's a live or install disk, for example).

You can change things, yes, but not in a way that keeps the disc usable.

The issue here is that you can't "unburn" a bit that has been burned. Since you burn 0s, not 1s, all you can do is make more bits 0 -- which would then be corrected by the error correction mechanism. You could alter the EC codes, but that would render them useless and eliminate the ability to correct errors, causing the data to be unreadable by a normal drive due to EC failure.

So, in practice, they're pretty safe. The worst thing that an attacker could do is to render the media unusable.

That's not completely true either. I used to use some sketchy utility back in the day that could mark one recording session as inaccessible and let you add data to any remaining space on the disc. You would only see the new session. Hypothetically you can also manipulate the file table to make it look exactly like the real thing.

I don't remember if it required the disc to be un-finalized or not. It probably did.

But this is really an edge case

True, I was assuming the disc wasn't written as a multi-session one (most people never use multi-session). Multi-session discs are a bit different. But, afaik, if the disc has been finalized then the result is as "immutable" as a non-multisession one. I'm not completely certain about that, though.
Are the EC codes such that any alteration from 1 to 0 in the message results in at least one flip from 0 to 1 in the (correct) EC code?

If not, then you "just" have to find the right spots where such a simultaneous alteration of the payload and the EC code is possible, and which also happens to result in a desirable behavior change.

No, the Reed Solomon codes used in the CD have one EC byte for every 3 data bytes. But there's a interleaving system where the data is encoded with error correction (adding 4 bytes to every 24) interleaved so that burst errors (that affect nearby disk surface) are distributed in the data stream, then encoded again (adding 4 bytes to 24 + 4 to give 32, with 8 EC bytes).

So you'd have to find mutations that are big enough to not be automatically corrected (more than 2 bits, you can't change, say a single 0x30 to 0x31), and also happen have the right signature so they don't flag as erasures, so you can't freely change bytes to just anything. And then you have to also satisfy the outer code that your modified byte after interleaving is correct. And do it all by only bring able to bit-flip in one direction. The only upside is that you can also flip bits in the correction bytes if you need.

Also, if the data you're modifying has any coding or compression (ZIP, Word documents, videos, most images), or there are checksums, you have to not upset that. And if it's encrypted, you won't know what to change in the first place.

So it's theoretically possible, but it would be very tricky and highly unlikely that you could change a useful piece of data unless you were extremely lucky with what the data was, what you wanted it to be, and where it was on the disk.

If you're concerned about people modifying your CD-R: encrypt and SHA-256 the whole thing and you'll be fine. I assume this threat model mostly targets governments, so they should first do both of those things and maybe also stop leaving them on trains.

Cool. Any idea how well-secured that one is, for use with USB hosts that potentially have adversary tech active on them?

(Threats to modify data on the drive or the code of its controller; not concerned with threats that fry the device with power.)

It will work just fine for this. You can superglue the switch if you need to keep it read only.
But is the switch just an input that is read by the software on the device, or does it actually interrupt e.g. a data line, making it impossible for even hacked software on the USB device to write to memory?

I can't really find a conclusive answer.

It physically switches some circuit to disable the ability for data to be written onto the device. Why not contact them and get the details?
I was just attempting to restate neilv's question in a different way. If you don't know that's fine.
Thanks, I should've been more clear.

Maybe someone can put in an epic weekend, to determine technical and market and logistics feasibility for a trustworthy USB flash drive product, in time to apply for the YC W24 extended deadline. From there, maybe Kickstarter to enthusiasts to fund the Mk. I manufacturing run. With Mk. I for experience and credibility, do enterprise marketing&sales, and/or SBIR to develop solution for possible sale to Federal gov't and contractors. :)

Add AI and blockchain, so I know I can safely invest millions into it.
"It's Uber for bytes."
> making it impossible for even hacked software on the USB device to write to memory?

Signed firmware to prevent BadUSB-style attacks, price starts at $40 for 16GB, https://www.kanguru.com/products/kanguru-flashtrust-secure-f...

They have other models with signed firmware + FIPS 140 certified encryption + tamper resistance + physical write-protect switch.

They also sell an SSD enclosure with write-protect switch.

Signing firmware might help prevent replacing the firmware, but it doesn't do anything to protect against exploiting a security vulnerability in the current firmware.

The marketing is a little too breathless to assure me that they did anything to actually make the firmware trustworthy: "Digitally Signed, Trusted Secure Firmware (RSA-2048 Bit)".

Signing something doesn't make it secure against exploits.

Signing firmware is step 1 in raising the bar for attackers, since most flash drives have unsigned firmware which presents little barrier to attacks.

If you could recommend other commercial flash drives with signed firmware, they could be compared for security vulnerabilities.

Some models have FIPS-certified encryption, so they may have customers who are high-value targets, motivating attacks/testing for security vulnerabilities.

Most USB media control chips have hardware write protect pins that will physically prevent writing to memory, for example:

https://eu.mouser.com/datasheet/2/268/MCHP_S_A0001038949_1-2...

Obviously if someone has physical access they could take the unit apart and circumvent this (but then they could flip the switch too)

I have some of these usb keys, and I remember having remounted the device rw on linux, and the device failed with an io error when I attempted to write. So if I'm correct, the driver was configured as if the key was rw, but it failed because the hardware switch was ro.

So I believe that it is an actual hardware lock, in addition to the os having the ro/rw status info.

If someone is interested, Kanguru have faster models than the one linked (which I have too), which is a little slow by today's standard.

Not exactly what you describe, but for the installation and boot use cases you're describing, have you tried Ventoy? It's still all software, but it's an extra layer of protection against the disk images being modified.
It's exactly zero protection against anything being modified, including the chainloader itself.
(comment deleted)
SD Cards do have a write protect switch. Does anyone know if that's enforced on the card itself?
It's not enforced by the card. It's just a plastic tab that slides so as to tell the reader that the card should not be written. There is nothing to stop a malicious reader from writing to a write-protected card.
Wow really? That's annoying. Why go to all the trouble to add the switch if they won't implement it correctly
Because engineering consumer media to be robust to a Stuxnet-level attack vector is ridiculous?

The switch as-is prevents the file system from becoming corrupt when you don't "safely remove" the device, and that's plenty enough for the average photographer.

I think the purpose of the switch is the same as the write-protect notches on old floppy disks: it's not really a security measure, it's meant to prevent the owner from accidentally formatting/erasing/overwriting things.
Or to go even further, the write-protect tabs on cassette tapes.
No, it is electrically connected, but only to the connector. That's why one of the edge contacts is split in two. It's still up to the host to enforce.
I would even want that for a non-stick full size USB drive like one of those Sandisk 2 TB drives. I know that an FS can be made read-only, but it would be great if it were possible to protect from writes at the hardware level without making it too easy for someone (could be a family member or whomever) to turn off the protection.
I have 2 of them. An ancient 2 GB Sandisk and more recent and available Kanguru SS3 with Write Protect.

Great little tool.

Clever , but seems insufficient protection on which to bet one's life.

At a minimum, there should be an always-available partition to contain plausibly-deniable files. Just having an empty USB drive is already suspicious, especially in any stressful situation (why would anyone go to the trouble of carrying something empty?)

The decoy partition should report full capacity even tho it may be small (just like the bogus "large" Chinese knock-off drives that report huge available space but are actually only small memory chips that will fail only when you actually try to load it fully).

Should probably also ship them with a wide variety of bogus off-brand labels. If some authorities figure it out, we don't want bulletins to look for one particular marking set to work.

> why would anyone go to the trouble of carrying something empty?

1. It's new.

2. You're going to load data onto it at your destination.

3. You carry it in case you need to copy data.

I often carry an empty messenger-bag; a USB drive is a lot less trouble to carry. In fact you can easily lose one in the corner of a pocket, and carry it inadvertently.

Provided it had never been written, I would regard an empty USB drive as unremarkable, and not grounds for suspicion. But you can't buy a new USB drive with a hidden partition.

Using skin resistance is not a fool-proof method. The resistance of wet fingers can vary widely from person to person. Mine has always been lower than most others that I've measured, and my dry measurement is well below the 500KΩ threshold described.
How is “files not visible unless you lick your fingers” a “self-destructing USB drive”?
The magic smoke is in Part 3. Click "Next Post" twice.
Part 1 doesn’t even mention the destruction, that’s a bad way to write a series of blog posts.
(comment deleted)
Its nice to create a usb drive that protects people in danger and then publishing on the web how to undo all its benefits, i’d argue that having something with a common universal method of unlocking is almost useless imho, maybe even dangerous, but probably is a nice exercise

But also the cost for you would be the time to call law enforcement, for them is the time to verify the validity of a document, so its just nonsense

The last paragraph is a edit in the wrong place sorry
Plausibly deniable encryption is better.

Password protect the partition: if you give one password you get boring content, if you give another you get the good stuff.