Re: Security considerations with REST calls

On Monday 14 September 2009 07:20:29 pm Sj Tib wrote:
> I have joined a development group where developers have created a "REST"
> interface that passes username and password as parameters (with every
> call) though the call is made over https e.g.
> https://<baseurl>?user=<username>&pwd=<password>.

Ew. But no real problem there.

> It is not truly a REST
> interface as they have designed it to make all calls as POST (instead of
> using the other HTTP verbs like PUT/DELETE/GET).

So they're using POST _and_ a querystring?

> I am not comfortable
> with the username and pwd being sent in with every HTTPS call - doesn't
> feel right

This is how HTTP Authentication often works.

> 1. Even if it is a POST with https can't someone see the username and
> password in the url going over the network?

No, unless you're using a proxy server.

> I am assuming the network
> sees the url in deciding where to forward it to.

Not at all.

Basic way this would work, at a low level:

 - You tell your HTTP library to fetch 
https://foo.com/bar/baz?user=smith&pwd=god.
 - The library asks your dns server what foo.com is.
 - The dns server returns an IP address, call it 1.2.3.4.
 - The library opens a secure (encrypted) connection to 1.2.3.4.
 - The library sends "POST /bar/baz?user=smith&pwd=god HTTP/1.1" over the 
secure channel.
 - The library sends a few newlines, then the body of the post.
 - 1.2.3.4 sends the response over the secure channel.

Keep in mind, by "http library", I mean "code running inside your program".

So, the only thing the network can see is:

 - You asked for foo.com
 - foo.com is 1.2.3.4
 - You connected to 1.2.3.4
 - You had some sort of encrypted communication with 1.2.3.4. (Or maybe it was 
garbage, who knows?)

That's it. Every part of the URL after the hostname is sent over the same 
encrypted channel as the POST body. (Again, unless you're using a proxy.)

> Is it any more/less
> vulnerable if GET/PUT/DELETE is used with https and the username/pwd is
> passed in url?

Completely irrelevant.

It's still ludicrously bad design -- you should be authenticating with a 
cookie or with HTTP authentication, not with the URL or the post body. But 
it's not particularly insecure inside a REST call, assuming that call is being 
made deep inside the app.

However, there are a few potential vulnerabilities, if you do this in a 
browser. The most obvious one is, the URL (complete with username and 
password) is now in your history. Some people won't care -- I click "save 
password" a lot.

But some people don't save passwords, or would like to be able to use a public 
computer without the password sticking around. And I certainly wouldn't want 
any shoulder-surfer to be able to read my password right out of the URL.

> If every REST
> call is an https call won't there be significant processing overheads
> not only on the server side but also on the PDA client side?

Depends. Do you value your clients' data? Yes? Use https.

On the server side, you can put an https proxy in front of it. You can even 
get machines which will do this in hardware. It's pretty much a "throw money 
at it" kind of problem. Again, ask yourself -- do you value your clients' 
data? Is your clients' data worth that expense? Probably.

On the client side, it'll drain slightly more battery on an iPhone, but 
nothing significant, especially compared to what you're already doing parsing 
the HTML, applying styles, doing animations, etc etc.

It's worth mentioning that, depending on the sensitivity of the data involved, 
you might be able to improve performance by switching to something as basic as 
HTTP challenge/response, and using straight HTTP. Anyone will be able to 
intercept/mangle the data itself, but they won't get your password, and you 
could always force HTTPS for any destructive operations, or for anything 
particularly sensitive.

But, if you use a session cookie, you'll want HTTPS again, because any 
eavesdropper can now hijack your session and use that to login.

> 3. What is the best practice security model to be used with REST API
> when the interface will be opened up to the world for all kinds of apps
> to be built on top of it? (Not planning on OAuth as the data is not
> expected to be consumed by other website apps)

In my opinion, abstract the authentication away, and never put it in the URL 
-- it's just bad taste. Once properly abstracted, you should be able to do 
what Rails does with restful_authentication -- you can transparently support 
http or https (whichever the client wants) or force one or the other, and you 
can transparently support HTML form + cookie authentication, HTTP 
authentication, CAS, or anything else you care to implement.

Anyone accessing this from a web browser probably wants to use HTML and 
cookies, while anyone accessing this as an API would probably find it easier to 
use HTTP auth. People in the outside world should use HTTPS (and you can force 
it), while anything inside your network (or on the same machine) could 
probably use HTTP.