Re: [RFC PATCH] http-backend: write error packet if backend command fails

From: Jeff King <peff@peff.net>
Date: Mar 28, 2020, 14:57 UTC
Message-ID: <20200328145741.GA1209174@coredump.intra.peff.net>
In-Reply-To: <b5f8b81498e1d152014acab92fa1b6e9701b3a0e.1585363771.git.liu.denton@gmail.com>

On Fri, Mar 27, 2020 at 10:50:32PM -0400, Denton Liu wrote:

Show 6 quoted lines

> If one tries to fetch packs with an incorrectly formatted parameter
> while using the smart HTTP protocol v2, the client will block forever.
> This is seen with the following command which does not terminate:
> 
> 	$ git -c protocol.version=2 clone https://github.com/git/git.git --shallow-since=20151012
> 	Cloning into 'git'...

That can't possibly be fixed by a change to http-backend.c, because github.com does not use git-http-backend. :)

We have a custom proxy layer that terminates the http connection, and speaks to "upload-pack --stateless-rpc" on the backend. There may be a bug in that proxy layer, but it's usually pretty robust to upload-pack hanging up the connection.

But since it works for v1 (which also dies!), and since you were able to reproduce the problem locally, I suspect it may be an issue in upload-pack itself.

Show 5 quoted lines

> This happens because when upload-pack detects invalid parameters, it
> will die(). When http-backend calls finish_command() and detects a
> non-zero exit code, it will simply call exit(1). Since there is no way
> in a CGI script to force a connection to terminate, the client keeps
> blocking on the read(), expecting more output.

When the CGI exits, that should close the descriptor it holds. If the webserver hands off the socket to the CGI, that would be sufficient. If it doesn't, then it should notice the CGI exiting and close the socket itself.

> Write an error packet if the backend command fails to start or finishes
> with an error so that the client can terminate the connection.

This is probably not a good idea; we don't know what state the protocol was in when the child died.

-Peff