Re: [PATCH 2/3] git-daemon: use getnameinfo to resolve hostname

From: Jeff King <peff@peff.net>
Date: Jan 14, 2009, 12:25 UTC
Message-ID: <20090114122536.GA5939@coredump.intra.peff.net>
In-Reply-To: <alpine.LSU.2.00.0901141148130.16109@fbirervta.pbzchgretzou.qr>

On Wed, Jan 14, 2009 at 11:48:38AM +0100, Jan Engelhardt wrote:

> This is much shorter than inet_ntop'ing, and also translated
> unresolvable addresses into a string.

Er, doesn't this totally change the meaning of REMOTE_ADDR from an IP address to a hostname? That seems bad because:

  - people already have hooks that compare REMOTE_ADDR against an
    address, so we are breaking their hooks

  - we are losing IP information in favor of hostname information; since
    (I assume) this is primarily intended for IP-based access control,
    we are adding an extra layer of indirection in the middle of our
    security model (i.e., I used to have to spoof an IP to fool your
    hook, but now I can do that _or_ spoof DNS).

So at the very least, you should be adding REMOTE_HOST in _addition_ to REMOTE_ADDR, not instead of. But that still leaves one final concern, which is that some git-daemon admins might not want to pay the cost for a reverse lookup for every request. It's extra network traffic, and adds extra latency to the process (but I don't personally run git-daemon, and I don't know whether big sites like kernel.org actually care about this).

-Peff