8 ms·

It concerns me that they were able to brute force a key for facebookcorewwwi.onion. If they can do that, what's to stop somebody else coming along and brute for

It concerns me that they were able to brute force a key for facebookcorewwwi.onion. If they can do that, what's to stop somebody else coming along and brute forcing a key for the same hostname. Looks like Tor hidden services are now broken to me... [edit] What's to stop Facebook from brute forcing a key for any of the existing hidden services? [edit2] If Facebook can brute force keys like this, so can the NSA and GCHQ. Tor hidden services are officially broken. [edit3] A colleague of mine suggested that this might be simply Facebooks way of making it public knowledge that Tor hidden services can no longer be relied upon. [edit4] Facebook are saying (on the Tor Talk list) that they generated a load of keys starting "facebook" and then just picked the one which looked most memorable, and were extremely lucky to get such a good one: http://archives.seul.org/tor/talk/Oct-2014/msg00433.html http://archives.seul.org/tor/talk/Oct-2014/msg00433.html

shawabawa3 12y ago
I'm hoping they just lucked out and came up with a backronym for it. I don't really know much about tor so I don't know how feasible bruteforcing is
- sgrenfro 12y ago
  FB engineer here. We got extremely lucky then went the backronym route as shawabawa3 guessed.
  - cesarb 12y ago
    You got so lucky that you'll probably want to keep using that name forever. The longer it's used, the greater the probability of compromise of its secret key. Which is why using TLS on top of the .onion address is brilliant: even if the secret key for the .onion address is compromised, the TLS certificate (which is rotated more often) will keep the connection safe. The worst that could happen would be someone hijacking the .onion address, but that would lead only to a DoS instead of the compromise that would happen without the redundant TLS layer. And the certificate also helps validate that the .onion address is really from facebook: as someone observed elsewhere in this discussion, the certificate is also valid for the non-.onion addresses, so just examining its alternate names extension is enough to prove that the certificate owner could also get a valid certificate for www.facebook.com (meaning the certificate owner is very probably facebook itself).
    - ultramancool 12y ago
      This is a good idea for other tor services too as that 1024 bit RSA is looking pretty shaky these days. Hopefully the tor devs will give it a bump to 2048 or move to Curve/Ed25519 soon, before this becomes a real problem. Progress: https://lists.torproject.org/pipermail/tor-dev/2013-August/005259.html https://lists.torproject.org/pipermail/tor-dev/2013-August/0...
    - 0x0 12y ago
      Someone else said getting SSL certs for .onion in the SNI doesn't require ANY kind of validation. So someone bruteforcing the .onion key could easily get their own valid SSL cert and have full access to the plaintext for anyone browsing the .onion site over SSL. The security of facebook over onion is now only protected by the hash power required to brute force the vanity address, instead of the integrity of the SSL CA system or the power required in factoring an SSL key. Even the requirement to spoof DNS or perform actual man-in-the-middle-of-the-wire hijacks has vanished.
      - akavel 12y ago
        And here's a link to a comment from a HNer claiming to have just got a certificate for the very same facebook's .onion address: https://news.ycombinator.com/item?id=8539066 https://news.ycombinator.com/item?id=8539066
  - pieter_mj 12y ago
    What we should expect is that facebook has enough computing power to bruteforce 12 to 13 characters in a reasonable timeframe. Did you ask NSA for a full 16-character bruteforce? :)
- im2w1l 12y ago
  Agreed. Likely they required facebook, and required that the rest of it would form words from some dictionary. I'd guess that the "i" at the end is a random alphanumeric (2⁵ possibilities). The positions of the words were likely arbitrary, since www comes after facebook. Lets arbitrarily say they have a word list of 128 words, that they want words from, all four letters long. This means we have 2⁷2⁷2⁵*4!/2 ~= 2²³ acceptable possibilities(the division because of the order of the random words).
mike_hearn 12y ago
Nothing, and it's interesting that they have enough computational power to do that for a relatively trivial project. Meanwhile, whilst I applaud Facebook going above and beyond here, this doesn't set a good precedent. Firstly, Onion service are very slow. There is no need to pay this cost for a service whose ownership is not actually hidden. If the Tor project made it easier to reliably identify traffic from Tor exit nodes, Facebook could apply whatever rules they wanted to Tor traffic without needing to slow things down for everyone. Secondly, by doing this, there's now a risk that other firms who want to be on the cutting edge of privacy will try to copycat this approach, even though it makes no sense and is very complex and expensive to set up. Worse, users might think it's some kind of "gold standard". Thirdly, it doesn't actually solve any of the reasons why Tor traffic is routinely discriminated against and harassed: Tor is effectively a "bulletproof ISP" that shields a lot of abuse and hacking. Merely making a Tor hidden service specifically for Facebook doesn't solve that, at all.
- justcommenting 12y ago
  > Firstly, Onion service are very slow. There is no need to pay this cost for a service whose ownership is not actually hidden. As long as you don't have to use it, I don't see why offering users a choice is a problem. >If the Tor project made it easier to reliably identify traffic from Tor exit nodes, Facebook could apply whatever rules they wanted to Tor traffic without needing to slow things down for everyone. ExoneraTor does this quite well in my opinion, and I don't follow how/why you think this will slow things down for everyone. Surely you're not referring to the entire Tor network? The main value in using this, in my opinion, is reducing the potential attack surface associated with MITM attacks--including CDNs--after your traffic exits Tor. Attacks on Facebook users involving Akamai have been documented by NSA, for example; Facebook is a PRISM partner, but this would arguably still stack the deck in favor of "going through the front door" to access Facebook user data.
  - mike_hearn 12y ago
    Accessing an .onion address via Tor is slower than accessing a normal website via Tor. If this onion service is now the official way to access Facebook via Tor and direct access doesn't work well, then this is slowing things down. Tor's current support for detecting usage is patchy. You can't query a random third party website for every login for a system like Facebook, so you need a list of IPs that can be refreshed quickly. But such lists tend to be incomplete or behind e.g. the "exit" flag doesn't mean what you'd intuitively expect, so it's sometimes possible for Tor traffic to turn up from an IP that is not identified as an exit. Re: MITM security. Even if Facebook got lucky here, we're talking about an 80 bit identifier and brute forcing these has been demonstrated before, I believe. I'm not sure this is much of an upgrade over just regular SSL CA + HSTS pinning. (edit: last paragraph)
untrothy 12y ago
I'm guessing they used something like scallion [0] or Shallot [1]. Bench marking Shallot on an Intel 3350P@3.10GHz: time ./shallot ^a -> 0.09 sec user time ./shallot ^aa -> 0.12 sec user time ./shallot ^aaa -> 0.12 sec user time ./shallot ^aaaa -> 0.47 sec user time ./shallot ^aaaaa -> 5.92 sec user time ./shallot ^aaaaaa -> 118 sec user Unfortunately OpenCL doesn't work with the nouveau drivers so I can't test scallion. Who knows how much they spent trying to brute force that onion address. EDIT: Ok looks like they went the backronym route [0] https://github.com/lachesis/scallion https://github.com/lachesis/scallion [1] https://github.com/katmagic/Shallot https://github.com/katmagic/Shallot
tyho 12y ago
There is a program called scallion [1] which can generate Vanity addresses like these. I am assuming that they searched for facebookwww?.onion and then just made up the acronym for the last character. They are therefore trying to brute force the first 11 characters of the address. The author of scallion estimates one can achieve 520 MH/s with a AMD Radeon HD5770 GPU [2] which retails for $190 [3], they then give the formula for calculating the time (in seconds) to have a 50% chance of finding a matching URL: 2^(5length-1) / hashspeed Which with a length of 11 and a hashspeed of 520M, would take about a year with the one GPU [4]. The total cost of the hardware to have a 50% chance of finding the vanity address "facebookwww?.onion" within a week would therefore be around $11 000 [5]. Imho, this is well within the realms of possibility for a company as large as Facebook and does not suggest a weakness in the .onion scheme. [1] https://github.com/lachesis/scallion https://github.com/lachesis/scallion [2] https://github.com/lachesis/scallion#speed--performance https://github.com/lachesis/scallion#speed--performance [3] https://www.amazon.com/dp/B0032F63TW https://www.amazon.com/dp/B0032F63TW [4] http://www.wolframalpha.com/input/?i=%282^%285*11-1%29+%2F+520*10^6%29+seconds http://www.wolframalpha.com/input/?i=%282^%285*11-1%29+%2F+5... [5] http://www.wolframalpha.com/input/?i=%24190+*+%282^%285*11-1%29+%2F+520*10^6%29+seconds+%2F+1+week http://www.wolframalpha.com/input/?i=%24190+*+%282^%285*11-1... [edit] Dustcore is very correct, correcting for my initial mistake, it would take 1.1 million years on a single GPU using scallion. Finding that sort of result in a month would require $2.6 billion worth of GPUs. Now I know facebook is known for spending billions on questionable purchases, but this would be a bit extreme even for them. How the hell have they managed this?
- dustcoin 12y ago
  The url isn't facebookwww? though, it is facebookcorewww?. The additional 4 characters didn't have to be "core", but the fact that it is an english word and relevant adds significant difficulty to your estimates.
  - tyho 12y ago
    I have appended my comment correcting for my mistake.
  - 12y ago
Llevel 12y ago
> If they can do that, what's to stop somebody else coming along and brute forcing a key for the same hostname. The .onion URL is created by hashing the public key (and possibly more information), and then it is stored in Tor's database of hidden service descriptors as noted by this[1]. This would indicate to me that if there's a hash conflict, such as the NSA trying to take over FB's .onion URL, the database of hidden service descriptors would reject the duplicate insertion to the database. [1] https://security.stackexchange.com/questions/23241/how-are-tor-hidden-nodes-services-found https://security.stackexchange.com/questions/23241/how-are-t...
- mike-cardwell 12y ago
  No, the way it works is the hidden service with the most recent announcement is the one that is currently used.
nly 12y ago
> If they can do that, what's to stop somebody else coming along and brute forcing a key for the same hostname. IIRC it's 80bit truncated SHA-1, so it's not even close to feasible unless there's a substantial preimage attack against the function (and none are known). It's clearly feasible to find something close enough to the human eye for a phishing/spoofing attack, but that's hardly a problem exclusive to Tor.
akavel 12y ago
Actually, to me the first thought after seeing the URL was totally: "wow, I wonder what backronym they invented for the ugly wwwi suffix." And the others posts in this thread seem to confirm that this was the case, and that this prettification mechanism is actually quite common on Tor already. Now, the other, more important for me observation is, that reportedly the TLS certificate is actually worth close to nothing, and giving false security, as a HNer claims to have got a valid cert issued for this very same facebook's .onion address already: https://news.ycombinator.com/item?id=8539066 https://news.ycombinator.com/item?id=8539066 -- if I understand correctly, cert issuers seem to happily accept any .onion URLs in "alternative addresses" in SSL certs without any verification. Anybody else could confirm/deny?
- wowaname 12y ago
  I cannot confirm nor deny, but if this is true, that just proves how flawed the CA system is.

shawabawa3 12y ago

I'm hoping they just lucked out and came up with a backronym for it. I don't really know much about tor so I don't know how feasible bruteforcing is

sgrenfro 12y ago
FB engineer here. We got extremely lucky then went the backronym route as shawabawa3 guessed.
- cesarb 12y ago
  You got so lucky that you'll probably want to keep using that name forever. The longer it's used, the greater the probability of compromise of its secret key. Which is why using TLS on top of the .onion address is brilliant: even if the secret key for the .onion address is compromised, the TLS certificate (which is rotated more often) will keep the connection safe. The worst that could happen would be someone hijacking the .onion address, but that would lead only to a DoS instead of the compromise that would happen without the redundant TLS layer. And the certificate also helps validate that the .onion address is really from facebook: as someone observed elsewhere in this discussion, the certificate is also valid for the non-.onion addresses, so just examining its alternate names extension is enough to prove that the certificate owner could also get a valid certificate for www.facebook.com (meaning the certificate owner is very probably facebook itself).
  - ultramancool 12y ago
    This is a good idea for other tor services too as that 1024 bit RSA is looking pretty shaky these days. Hopefully the tor devs will give it a bump to 2048 or move to Curve/Ed25519 soon, before this becomes a real problem. Progress: https://lists.torproject.org/pipermail/tor-dev/2013-August/005259.html https://lists.torproject.org/pipermail/tor-dev/2013-August/0...
  - 0x0 12y ago
    Someone else said getting SSL certs for .onion in the SNI doesn't require ANY kind of validation. So someone bruteforcing the .onion key could easily get their own valid SSL cert and have full access to the plaintext for anyone browsing the .onion site over SSL. The security of facebook over onion is now only protected by the hash power required to brute force the vanity address, instead of the integrity of the SSL CA system or the power required in factoring an SSL key. Even the requirement to spoof DNS or perform actual man-in-the-middle-of-the-wire hijacks has vanished.
    - akavel 12y ago
      And here's a link to a comment from a HNer claiming to have just got a certificate for the very same facebook's .onion address: https://news.ycombinator.com/item?id=8539066 https://news.ycombinator.com/item?id=8539066
- pieter_mj 12y ago
  What we should expect is that facebook has enough computing power to bruteforce 12 to 13 characters in a reasonable timeframe. Did you ask NSA for a full 16-character bruteforce? :)
im2w1l 12y ago
Agreed. Likely they required facebook, and required that the rest of it would form words from some dictionary. I'd guess that the "i" at the end is a random alphanumeric (2⁵ possibilities). The positions of the words were likely arbitrary, since www comes after facebook. Lets arbitrarily say they have a word list of 128 words, that they want words from, all four letters long. This means we have 2⁷2⁷2⁵*4!/2 ~= 2²³ acceptable possibilities(the division because of the order of the random words).

mike_hearn 12y ago

Nothing, and it's interesting that they have enough computational power to do that for a relatively trivial project. Meanwhile, whilst I applaud Facebook going above and beyond here, this doesn't set a good precedent. Firstly, Onion service are very slow. There is no need to pay this cost for a service whose ownership is not actually hidden. If the Tor project made it easier to reliably identify traffic from Tor exit nodes, Facebook could apply whatever rules they wanted to Tor traffic without needing to slow things down for everyone. Secondly, by doing this, there's now a risk that other firms who want to be on the cutting edge of privacy will try to copycat this approach, even though it makes no sense and is very complex and expensive to set up. Worse, users might think it's some kind of "gold standard". Thirdly, it doesn't actually solve any of the reasons why Tor traffic is routinely discriminated against and harassed: Tor is effectively a "bulletproof ISP" that shields a lot of abuse and hacking. Merely making a Tor hidden service specifically for Facebook doesn't solve that, at all.

justcommenting 12y ago
> Firstly, Onion service are very slow. There is no need to pay this cost for a service whose ownership is not actually hidden. As long as you don't have to use it, I don't see why offering users a choice is a problem. >If the Tor project made it easier to reliably identify traffic from Tor exit nodes, Facebook could apply whatever rules they wanted to Tor traffic without needing to slow things down for everyone. ExoneraTor does this quite well in my opinion, and I don't follow how/why you think this will slow things down for everyone. Surely you're not referring to the entire Tor network? The main value in using this, in my opinion, is reducing the potential attack surface associated with MITM attacks--including CDNs--after your traffic exits Tor. Attacks on Facebook users involving Akamai have been documented by NSA, for example; Facebook is a PRISM partner, but this would arguably still stack the deck in favor of "going through the front door" to access Facebook user data.
- mike_hearn 12y ago
  Accessing an .onion address via Tor is slower than accessing a normal website via Tor. If this onion service is now the official way to access Facebook via Tor and direct access doesn't work well, then this is slowing things down. Tor's current support for detecting usage is patchy. You can't query a random third party website for every login for a system like Facebook, so you need a list of IPs that can be refreshed quickly. But such lists tend to be incomplete or behind e.g. the "exit" flag doesn't mean what you'd intuitively expect, so it's sometimes possible for Tor traffic to turn up from an IP that is not identified as an exit. Re: MITM security. Even if Facebook got lucky here, we're talking about an 80 bit identifier and brute forcing these has been demonstrated before, I believe. I'm not sure this is much of an upgrade over just regular SSL CA + HSTS pinning. (edit: last paragraph)

untrothy 12y ago

I'm guessing they used something like scallion [0] or Shallot [1]. Bench marking Shallot on an Intel 3350P@3.10GHz: time ./shallot ^a -> 0.09 sec user time ./shallot ^aa -> 0.12 sec user time ./shallot ^aaa -> 0.12 sec user time ./shallot ^aaaa -> 0.47 sec user time ./shallot ^aaaaa -> 5.92 sec user time ./shallot ^aaaaaa -> 118 sec user Unfortunately OpenCL doesn't work with the nouveau drivers so I can't test scallion. Who knows how much they spent trying to brute force that onion address. EDIT: Ok looks like they went the backronym route [0] https://github.com/lachesis/scallion https://github.com/lachesis/scallion [1] https://github.com/katmagic/Shallot https://github.com/katmagic/Shallot

tyho 12y ago

There is a program called scallion [1] which can generate Vanity addresses like these. I am assuming that they searched for facebookwww?.onion and then just made up the acronym for the last character. They are therefore trying to brute force the first 11 characters of the address. The author of scallion estimates one can achieve 520 MH/s with a AMD Radeon HD5770 GPU [2] which retails for $190 [3], they then give the formula for calculating the time (in seconds) to have a 50% chance of finding a matching URL: 2^(5length-1) / hashspeed Which with a length of 11 and a hashspeed of 520M, would take about a year with the one GPU [4]. The total cost of the hardware to have a 50% chance of finding the vanity address "facebookwww?.onion" within a week would therefore be around $11 000 [5]. Imho, this is well within the realms of possibility for a company as large as Facebook and does not suggest a weakness in the .onion scheme. [1] https://github.com/lachesis/scallion https://github.com/lachesis/scallion [2] https://github.com/lachesis/scallion#speed--performance https://github.com/lachesis/scallion#speed--performance [3] https://www.amazon.com/dp/B0032F63TW https://www.amazon.com/dp/B0032F63TW [4] http://www.wolframalpha.com/input/?i=%282^%285*11-1%29+%2F+520*10^6%29+seconds http://www.wolframalpha.com/input/?i=%282^%285*11-1%29+%2F+5... [5] http://www.wolframalpha.com/input/?i=%24190+*+%282^%285*11-1%29+%2F+520*10^6%29+seconds+%2F+1+week http://www.wolframalpha.com/input/?i=%24190+*+%282^%285*11-1... [edit] Dustcore is very correct, correcting for my initial mistake, it would take 1.1 million years on a single GPU using scallion. Finding that sort of result in a month would require $2.6 billion worth of GPUs. Now I know facebook is known for spending billions on questionable purchases, but this would be a bit extreme even for them. How the hell have they managed this?

dustcoin 12y ago
The url isn't facebookwww? though, it is facebookcorewww?. The additional 4 characters didn't have to be "core", but the fact that it is an english word and relevant adds significant difficulty to your estimates.
- tyho 12y ago
  I have appended my comment correcting for my mistake.
- 12y ago

Llevel 12y ago

> If they can do that, what's to stop somebody else coming along and brute forcing a key for the same hostname. The .onion URL is created by hashing the public key (and possibly more information), and then it is stored in Tor's database of hidden service descriptors as noted by this[1]. This would indicate to me that if there's a hash conflict, such as the NSA trying to take over FB's .onion URL, the database of hidden service descriptors would reject the duplicate insertion to the database. [1] https://security.stackexchange.com/questions/23241/how-are-tor-hidden-nodes-services-found https://security.stackexchange.com/questions/23241/how-are-t...

mike-cardwell 12y ago
No, the way it works is the hidden service with the most recent announcement is the one that is currently used.

nly 12y ago

> If they can do that, what's to stop somebody else coming along and brute forcing a key for the same hostname. IIRC it's 80bit truncated SHA-1, so it's not even close to feasible unless there's a substantial preimage attack against the function (and none are known). It's clearly feasible to find something close enough to the human eye for a phishing/spoofing attack, but that's hardly a problem exclusive to Tor.

akavel 12y ago

Actually, to me the first thought after seeing the URL was totally: "wow, I wonder what backronym they invented for the ugly wwwi suffix." And the others posts in this thread seem to confirm that this was the case, and that this prettification mechanism is actually quite common on Tor already. Now, the other, more important for me observation is, that reportedly the TLS certificate is actually worth close to nothing, and giving false security, as a HNer claims to have got a valid cert issued for this very same facebook's .onion address already: https://news.ycombinator.com/item?id=8539066 https://news.ycombinator.com/item?id=8539066 -- if I understand correctly, cert issuers seem to happily accept any .onion URLs in "alternative addresses" in SSL certs without any verification. Anybody else could confirm/deny?

wowaname 12y ago
I cannot confirm nor deny, but if this is true, that just proves how flawed the CA system is.