Monday, November 4, 2013

Two Factor Authentication for SSH (with Google Authenticator)

Two factor authentication is a method of ensuring that a user has a physical device in addition to their password when logging in to some service. This works by using a time (or counter) based code which is generated by the device and checked by the host machine. Google provides a service which allows one to use their phone as the physical device using a simple app.

This service can be easily configured and greatly increases the security of your host.

Installing Dependencies

  1. There is only one: the Google-Authenticator software itself:
    # pkg install pam_google_authenticator
  2. On older FreeBSD intallations you may use:
    # pkg_add -r pam_google_authenticator
    On Debian derived systems use:
    # apt-get install libpam-google-authenticator

User configuration

Each user must run "google-authenticator" once prior to being able to login with ssh. This will be followed by a series of yes/no prompts which are fairly self-explanatory. Note that the alternate to time-based is to use a counter. It is easy to lose track of which number you are at so most people prefer time-based.
  1. $ google-authenticator
    Do you want authentication tokens to be time-based (y/n)
    ...
    
    Make sure to save the URL or secret key generated here as it will be required later.

Host Configuration

To enable use of Authenticator the host must be set up to use PAM which must be configured to prompt for Authenticator.
  1. Edit the file /etc/pam.d/sshd and add the following in the "auth" section prior to pam_unix:
    auth requisite pam_google_authenticator.so
  2. Edit /etc/ssh/sshd_config and uncomment
    ChallengeResponseAuthentication yes

Reload ssh config

  1. Finally, the ssh server needs to reload its configuration:
    # service sshd reload

Configure the device

  1. Follow the instructions provided by Google to install the authentication app and setup the phone.

That is it. Try logging into your machine from a remote machine now

Thanks bcallah for proof-reading this post.

Sunday, April 28, 2013

Pre-Interview NDAs Are Bad

I get quite a few emails from business folk asking me to interview with them or forward their request to other coders I know. Given the volume it isn't feasible to respond affirmatively to all these requests.

If you want to get a coder's attention there are a lot of things you could do, but there is one thing you shouldn't do: require them to sign an NDA before you interview them.

From the candidates point of view:

  1. There are a lot more ideas than qualified candidates.
  2. Its unlikely your idea is original. It doesn't mean anyone else is working on it, just that someone else probably thought of it.
  3. Lets say the candidate was working on a similar, if not identical project. If the candidate fails to continue with you now they have to consult a lawyer to make sure you can't sue them for a project they were working on before
  4. NDAs are hard legal documents and shouldn't be signed without consulting a lawyer. Does the candidate really want to find a lawyer before interviewing with you?
  5. An NDA puts the entire obligation on the candidate. What does the candidate get from you?
From a company founders point of view:
  1. Everyone talks about the companies they interview with to someone. Do you want to be that strange company which made them sign an NDA? It can harm your reputation easily.
  2. NDAs do not stop leaks. They serve to create liability when a leak occurs. Do you want to be the company that sues people that interview with them?

There are some exceptions; for example government and security jobs may require security clearance and an NDA. For more jobs it is possible to determine if a coder is qualified and a good fit without disclosing confidential company secrets.

Friday, December 21, 2012

Correctly Verifying an Email Address

Some services that accept email addresses want to ensure that these email addresses are valid.

There are multiple aspects to an email being valid:
  1. The address is syntactically valid.
  2. An SMTP server accepts mail for the address.
  3. A human being reads mail at the address.
  4. The address belongs to the person submitting it.

How does one verify an email address? I'll start with the wrong solutions and build up the correct one.

Possibility #0 - The Regular Expression

Discussions on a correct regular expression to parse email addresses are endless. They are almost always wrong. Even really basic pattern matching such as *@*.* is wrong: it will reject the valid email address n@ai.[5]

Even a fully correct regular expression does not tell you if the mailbox is valid or reachable.

This scores 0/4 on the validity checking scale.

Possibility #1 - The VRFY Command

The oldest mechanism for verifying an email address is the VRFY mechanism in RFC821 section 4.1.1:

VERIFY (VRFY) This command asks the receiver to confirm that the argument identifies a user. If it is a user name, the full name of the user (if known) and the fully specified mailbox are returned.

However this isn't sufficient. Most SMTP servers disable this feature for security and anti-spam reasons. This feature could be used to enumerate every username on the server to perform more targeted password guessing attacks:

Both SMTP VRFY and EXPN provide means for a potential spammer to test whether the addresses on his list are valid (VRFY)... Therefore, the MTA SHOULD control who is is allowed to issue these commands. This may be "on/off" or it may use access lists similar to those mentioned previously.

This feature wasn't guaranteed to be useful at the time the RFC was written:[1]

The VRFY and EXPN commands are not included in the minimum implementation (Section 4.5.1), and are not required to work across relays when they are implemented.

Finally, even if VRFY was fully implemented there is no guarantee that a human being reads the mail sent to that particular mailbox.

All of this makes VRFY useless as a validity checking mechanism so it scores 1/4 on the validity checking scale.

Possibility #2 - Sending a Probe Message

With this method you try to connect with a mail server and pretends to send a real mail message but cut off before sending the message content. This is wrong for a for the following reasons:

A system administrator that disabled VRFY has a policy of not allowing for the testing for email addresses. Therefore the ability to test the email address by sending a probe should be considered a bug and must not be used.

The system might be set up to detect signs up of a probe such as cutting off early may rate limit or block the sender.

In addition, the SMTP may be temporarily down or the mailbox temporarily unavailable but this method provides no resilience against failure. This is especially true if this mechanism is attempting to provide real-time feedback to the user after submitting a form.

This scores 1/4 on the validity checking scale.

Possibility #3 - Sending a Confirmation Mail

If one cares about if a human is reading the mailbox the simplest way to do so is send a confirmation mail. In the email include a link to a website (or set a special reply address) with some indication of what is being confirmed. For example, to confirm "user@example.com" is valid the link might be http://example.com/verify?email=user@example.com or http://example.com/verify?account=12345[2].

This method is resilient against temporary failures and forwarders. Temporary failures could be retried like a normal SMTP conversation.

This way it is unlikely that a non-human will trigger the verification email[3]. This approach solves some of the concerns, it suffers from a fatal flaw:

It isn't secure. It is usually trivial to guess the ID number, email account, other identifier. An attacker could sign up with someone else's email account and then go to the verification page for that user's account. It might be tempting to use a random ID but randomness implementations are usually not secure.

This scores 3/4 on the validity checking scale

Possibility #4 - Sending a Confirmation Mail + HMAC

The correct solution is to send a confirmation, but include a MAC of the identifier in the verification mechanism (reply, or url) as well. A MAC is a construction used to authenticate a message by combining a secret key and the message contents. One family of constructions, HMAC, is a particularly good choice. This way the url might become http://example.com/verify?email=user@example.com&mac=74e6f7298a9c2d168935f58c001bad88[4]

Remember that the HMAC is a specific construction, not a naive hash. It would be wise to use a framework native function such as PHP's hash_hmac. Failing to include a secret into the construction would make the MAC trivially defeated by brute force.

This scores 4/4 on the validity checking scale

Closing Notes

Getting email validation right is doable, but not as trivial as many of the existing solutions make it seem.

  1. Note that RFC1123 more specifically spells out that VRFY MUST be implemented but MAY be disabled.
  2. This is not my luggage password.
  3. It is still possible for a auto-reply bot to trigger reply based verification schemes. Bots that click every link in received email are uncommon.
  4. This is HMAC-MD5. It isn't insecure as collisions aren't important for HMAC. I chose it because it is short.
  5. n@ai is a in-use email address by a person named Ian:
    %dig +short ai MX
    10 mail.offshore.ai.
Thank you to bd for proofreading and reviewing this blog post.

Wednesday, November 21, 2012

Don't Use Timing Functions for Profiling

One common technique for profiling programs is to use the gettimeofday system call (with code that looks something like this):

Example (incorrect) code that uses gettimeofday - click to view
#include <time.h>
#include <stdlib.h>
#include <stdio.h>
void function(void)
{
  struct timeval before;
  struct timeval after;
  gettimeofday(&before, NULL);
  codetoprofile();
  gettimeofday(&after, NULL); 
  time_t delta = after.tv_sec - before.tv_sec;
  printf("%ld\n",delta);
}

However, using gettimeofday(2) or time(3) or any function designed to get a time of day to obtain profiling information is wrong for many reasons:

  1. Time can go backwards. In a virtualized environment this can happen quite often. In non-virtualized environments this can happen due to time zones. Even passing CLOCK_MONOTONIC to clock(3) doesn't help as it can go backwards during a leap second expansion.
  2. Time can change drastically for no reason. Systems with NTP enabled periodically sync their time with a time source. This can cause the system time to change by minutes, hours, or even days!
  3. These functions measure Wall Clock time. Time spent on entirely unrelated processes is going to be included in the profiling data!
  4. Even if you have disabled everything else on the system[1] the delta computed above includes both of User time and System Time. If your algorithm is very fast but the kernel has a slow implementation of some system call you won't learn much.
  5. gettimeofday relies on the cpu clock which may differ across cores resulting in time skew.

So what should be used instead?

There isn't a good, portable, function to obtain profiling information. However there are options for those not tied to a particular system (or those willing to maintain multiple implementations for different systems.

The getrusage(2) system call is one option for profiling data. This provides different fields for user time (ru_utime) and system time (ru_stime) at a relatively high level of precision and accuracy.

Using DTraces profiling provider also seems to be a decent choice although I limited experience with it.

Finally, using APIs meant to access hardware specific features such as FreeBSD's hwpmc is likely to provide the best results at the cost of being the least portable. Linux has similar features such as oprofile and perf. Using dedicated profilers such as Intel's vtunes[2] may also be worthwhile.

  1. Including networking, background process swapping, cron, etc.
  2. A FreeBSD version is available.
update 2012-11-26: Include note about clock skew across cores.
Update 2013-02-13: Update and fix a massive error I had w.r.t. clock(3)

Wednesday, October 31, 2012

Finding the majority element in a stream of numbers

Some time ago I came across the following question.

As input a finite stream stream of numbers is provided. Define an algorithm to find the majority element of the input. The algorithm need not provide a sensible result if no majority element exists. You may assume a transdichotomous memory model.

There are a few definitions which may not be immediately clear:
Stream
A possibly infinite set of data which may not be reused in either the forward or backward direction without explicitly storing it.
Majority element
An element in a set which occurs more than half the time.

Unfortunately this answer isn't of my own invention, but it is interesting and succinct.

The algorithm (click to view) Using 3 registers the accumulator, the guess and the current element (next):
  1. Initialize accumulator to 0
  2. Accept the next element of the stream and place it into next. If there are no more elements go to step #7.
  3. If accumulator is 0 place next into guess and increment accumulator.
  4. Else if guess matches next increment accumulator
  5. Else decrement accumulator
  6. Go to step 2
  7. Return the value in guess as the result

An interesting property of this algorithm is that it can be implemented in $O(n)$ time even on a single tape Turing Machine.

Tuesday, October 30, 2012

Cneonction: closed HTTP header

When you make a request to certain websites you may find an unusual header that looks a little strange:

[8000 eitan@radar ~ ]%curl -I http://www.imdb.com/ 2>/dev/null|grep close
Cneonction: close
[8001 eitan@radar ~ ]%curl -I http://maps.apple.com/ 2>/dev/null|grep close
Cneonction: close

This isn't a typo though. Some load balancers that sit between the web server and end user want to implement HTTP keep-alive without modifying the back end web server. The load balancer therefore has to add "Connection: Keep-Alive" to the HTTP header and also has to elide the "Connection: close" from the real webserver. However, if it completely removes the line the load balancer (acting as a TCP proxy) would have to stall before forwarding the complete text in order to recompute the TCP checksum. This increases latency on packet delivery.

Instead, the proxy uses a hack to keep the checksum unchanged. The TCP checksum of a packet is the 1s complement summation of all the 16 bit words (the final word might be right padded with zeros).[1] By manipulating the ordering, but not the content of the header the proxy can avoid changing the TCP checksum except by the fixed amount that the "Connection: Keep-Alive" adds (2061).

In particular:

>>>sum(ord(i) for i in "Connection") - sum(ord(i) for i in "Cneonction")

0

This reordering also keeps the packet size the same.

  1. RFC793
Edit 2012-10-31: Make the RFC a link and remove pointless "2>&1"
Thanks abbe for the inspiration! Thanks wxs for the proofreading.

Tuesday, October 9, 2012

Reduced Entropy in rand() and random()

TL;DR: Don't rely on undefined behavior, even when you think it should work.

I recently reported a minor issue to the FreeBSD security team.

The libc random functions had code1,2 designed to run when /dev/random is not available. This can easily occur in a chroot or jail environment.

if (!done) {
        struct timeval tv;
        unsigned long junk;

        gettimeofday(&tv, NULL);
        srandom((getpid() << 16) ^ tv.tv_sec ^ tv.tv_usec ^ junk);
        return;
}

This code is designed provide a minimal amount of entropy in the "failure" case. Unfortunately, it doesn't even provide the entropy it claims to. This is a minor issue because getpid, getimeday, and a single long variable don't provide a lot of entropy in the first place: (only $log_2{sizeof(long)}$ bits).

The point of the junk value is to add entropy by using uninitialized memory. This relies on the compiler being "stupid" enough not optimize it away.

Unfortunately clang and newer versions of gcc are smart enough to use the undefined behavior in undesired ways.

clang 3 removes any computation which relies on the undefined behavior and so produces the following object code:

 af0:   e8 5f fc ff ff          callq  754 <gettimeofday@plt>
 af5:   e8 7a fc ff ff          callq  774 <getpid@plt>
 afa:   e8 65 fc ff ff          callq  764 <srandom@plt>

Note that the junk variable is entirely unused and that the xor operation between gettimeofday and getpid is non-existent.

gcc 4.6 4 outputs:

 ce8:   e8 03 fa ff ff          callq  6f0 <gettimeofday@plt>
 ced:   e8 4e fa ff ff          callq  740 <getpid@plt>
 cf2:   48 8b 7c 24 08          mov    0x8(%rsp),%rdi
 cf7:   48 33 3c 24             xor    (%rsp),%rdi
 cfb:   c1 e0 10                shl    $0x10,%eax
 cfe:   48 98                   cltq
 d00:   48 31 c7                xor    %rax,%rdi
 d03:   e8 28 fa ff ff          callq  730 <srandom@plt>
Note that in this case the junk value appears to be (%rsp) which isn't all that random.

gcc 4.25 produces the following code

with the junk variable
 d9f:   e8 18 fa ff ff          callq  7bc <gettimeofday@plt>
 da4:   e8 43 fa ff ff          callq  7ec <getpid@plt>
 da9:   48 8b 3c 24             mov    (%rsp),%rdi
 dad:   48 33 7c 24 08          xor    0x8(%rsp),%rdi
 db2:   c1 e0 10                shl    $0x10,%eax
 db5:   48 98                   cltq
 db7:   48 31 c7                xor    %rax,%rdi
 dba:   48 31 df                xor    %rbx,%rdi
 dbd:   e8 1a fa ff ff          callq  <srandom@plt>
and without:
 d9f:   e8 18 fa ff ff          callq  7bc <gettimeofday@plt>
 da4:   e8 43 fa ff ff          callq  7ec <getpid@plt>
 da9:   48 8b 3c 24             mov    (%rsp),%rdi
 dad:   48 33 7c 24 08          xor    0x8(%rsp),%rdi
 db2:   c1 e0 10                shl    $0x10,%eax
 db5:   48 98                   cltq
 db7:   48 31 c7                xor    %rax,%rdi
 dba:   e8 1d fa ff ff          callq  7dc <srandom@plt>

The base version of gcc isn't vulnerable. However, with the upcoming switch to the clang compiler the FreeBSD does become vulnerable.

The first proposed fix was to add the volatile type qualifier to the junk variable. While this seemed to fix the code generation issue I didn't believe this to be a valid fix as the behavior is still undefined6 (I misread text of the standard). Additionally, the value is likely have be very predictable. A preference was expressed to remove the junk variable as using it may leak a small amount of stack data.

I proposed the simple and obvious fix of removing the use of the junk variable7

In a brief survey of other libraries I noticed similar issues. I will attempt to notify the vendors

It should be obvious, but undefined behavior is undefined and can't be relied on to ever to give a sensible result.

  1. random.c r165903 (line 316)
  2. rand.c r241046 (line 131)
  3. FreeBSD clang version 3.1 (branches/release_31 156863) 20120523 compiled with -O2 or -O3
  4. gcc46 (FreeBSD Ports Collection) 4.6.4 20120608 (prerelease) compiled with -O2
  5. gcc (GCC) 4.2.1 20070831 patched [FreeBSD] compiled with -O3
  6. sections 5.1.2.2.3 and 6.7.2.4 of ISO9899
  7. svn commit r241373
Edit 2010-10-10: update the paragraph referring to undefined behavior of volatile.