RE: [PATCH v3 1/2] use HOST_NAME_MAX to size buffers for gethostname(2)

From: David Turner <david.turner@twosigma.com>
Date: Apr 19, 2017, 19:08 UTC
Message-ID: <b7a3844946934ecda1ba1ac5b972ff9d@exmbdft7.ad.twosigma.com>
In-Reply-To: <c0333c81-d3b2-ca2d-a553-75642d8fb949@web.de>

Show 11 quoted lines

> I had another look at this last night and cooked up the following patch.  Might
> have gone overboard with it..
> 
> -- >8 --
> Subject: [PATCH] gc: support arbitrary hostnames and pids in lock_repo_for_gc()
> 
> git gc writes its pid and hostname into a pidfile to prevent concurrent garbage
> collection.  Repositories may be shared between systems with different limits
> for host name length and different pid ranges.  Use a strbuf to store the file
> contents to allow for arbitrarily long hostnames and pids to be shown to the
> user on early abort.

This is pretty paranoid, but maybe the remote host has a longer pid_t than we do, so we should be using intmax_t when reading the pid, and only check its size before passing it to kill?

(Personally, I think this whole patch is kind of overkill, but some folks probably think the same about my original patches, so I'm happy to live and let live).