HN Explorer

Tell HN: Github has dozens of public s3 passwords

57 points by kabuks ↗ HN
I almost pushed my s3 credentials to a public github repo for the third time in one day.

So, I got curious. Are there people out there who forget cover their tracks?

A quick search shows quite a few 'open' buckets out there. What's the best way to warn these folks? What other credentials are lurking out there?

Here's the search: http://github.com/search?langOverride=&language=&q=S3+Base.establish_connection&repo=&start_value=1&type=Code&x=0&y=0

And the first open bucket I found: http://github.com/prakashraman/jammmin/blob/a668672c69fafdb8317fec4fb19b7abb0b318e1a/app/scripts/s3_connect.rb

30 comments

[ 2.9 ms ] story [ 75.0 ms ] thread
relix
It's something that almost happened to me too. A git repository for a project that at first, in the prototype-stage, had credentials hard-coded, but inevitably moved on to be config-file-driven.

I concocted the following command to change all mentions of a specific word to another word, in a git repository:

  git filter-branch --tree-filter "find . -type f -exec sed -i -e 's/originalpassword/newpassword/g' {} \;"
Use this to check if there are any mentions of some word, e.g. your password, in the repository:

  git filter-branch --tree-filter "grep -r originalpassword * || true"
ritonlajoie
" What's the best way to warn these folks? What other credentials are lurking out there?"

Sure, posting that on HN is a good way...

pierrefar
How about leaving them some README-NOW.txt file in their S3 bucket?
tlrobinson
I rarely list the contents of my S3 buckets so I'd probably never notice this.

Send them a message on GitHub, perhaps?

jeebusroxors
How about something with github API? I'm pretty sure you can search/email?
igorgue
Yeah, this is a typical mistake, I've pushed Django's secret keys and even twitter credentials.

PS: I do it all the time on my private repos but I try to not do it on the public ones.

ulf
In case of django there is a nice workaround in using a non-git localsettings.py file.

Just put "from localsettings import *" in your settings-file, and keep all deployment-specific settings in the localsettings-file

[dead] megafotze
yup, there are a lot of credentials out there. you have to watch out what you commit as it will be public viewable in history views.
ck2
Protip: keep settings in a separate file

Mark the file as excluded from svn/git.

Make a settings.sample file for the project.

baxter
Rather than store passwords or sensitive config data in a file I'll usually try to store them in an environment variable. Heroku has a little guide on this: http://docs.heroku.com/config-vars
trevorturk
I use a similar strategy: http://trevorturk.com/2009/06/25/config-vars-and-heroku/

...this kind of thing doesn't have to be Heroku-specific, though.

ujeezy
Yup, symlinking config files after deploy is the approach I use. Thanks for the Cap task to set things up; at the rate I setup new projects, manually copying those files was getting old :)
jbeda
There is a business opportunity here. Managing credentials and secrets is (passwords, private certs, etc) is hard. Making that process turnkey, secure and easy would be useful.
thenduks
I'm not sure it's that hard.

You can use any number of very low-barrier strategies to solve this. One that was mentioned a few times in this thread is to use environment variables. Personally I just put a {whatever}.yml or similar in my project tree somewhere and throw it in .gitignore.

I can't imagine a process that starts with "go to westorecredentials.com and sign up..." being any easier. Willing to be surprised :)

fragmede
People are advocating not saving settings to source control. F-that, I want to be able to clone my repo and run the project without scp-ing an additional file around. A file, that by all rights should be part of the project.

Here's my solution - github 'extends' git somehow, so files can be marked private. Then, unless you have commit rights to that repo, you can't see the file's contents. So a public http checkouts lacks the secret password.

And because it's git, someone you give commit rights to is someone you trust. People you don't trust can just make their own repo.

thenduks
That sounds like too much software to me. What's with the aversion to scp? You only have to do it once per box you set up... There's something to be said for having GitHub do everything 1000% for you, but there's also something to the unix philosophy that would suggest git is good at the version control thing, and scp is good for the configuration-file-moving thing.
fragmede
so... which version of the config file are you using?
thenduks
The one listed first in a `git log config/whatever.yml`

Just because it isn't in a public repo on GitHub doesn't mean it isn't versioned properly. Quite the contrary, I use git locally and on a personal webserver for tons of stuff.

The point isn't that it shouldn't go into git. It's that it doesn't belong as part of your project. Different config options are often needed on each machine, especially development boxes. I don't have the same database.yml as my co-workers... I have my box in production mode or with memcache off for testing purposes maybe... There's tons of stuff like that.

I still keep those files (along with my ~/.git/config, ~/.ssh/config, .gemrc, ~/.profile, ~/.vim* and a hundred other things) in git, just not on GitHub in a public repo.

pskomoroch
There is a "message" button on each github user page.

I just sent this to a few people (including someone who had forked one of my projects and added their S3 keys to a config file):

Noticed you have your Amazon S3 keys out in the open on github. You might want to remove those config files from your repository as described in the thread here:

http://news.ycombinator.com/item?id=1574211

-Pete

gojomo
Remove and change at Amazon, as the values could still be floating around in fetches, caches, archives, indexes, etc. for a while.
kabuks
And still cached in their search cache
tlrobinson
Might be worth setting up a git hook to scan your commits for sensitive data.

Of course you'll need to include that sensitive data in the script, though the first few characters of AWS credentials should be unique enough.

I thought about setting up something similar for networking. If a packet contains my password in cleartext then pop up a warning allowing/denying (denying would have to force the connect to close, I guess). Might be too much overhead though.

prakashraman
My God, I can't believe I left it there. While I was developing a while ago I had put it there and completely forgot to remove it. What a costly mistake this could've been or even is.

Thank you, kabuks, so much for noticing this. I have changed my S3 key pair and am getting to cleaning up my git commits.

My God ! But thanks so much.