Re: [PATCH] Fix leak in credential_apply_config

From: Jeff King <peff@peff.net>
Date: Aug 20, 2021, 17:56 UTC
Message-ID: <YR/sxISzR0RebVMZ@coredump.intra.peff.net>
In-Reply-To: <ffa1786e-f69f-3d76-98dc-7fa5cdbd31c0@gmail.com>

On Fri, Aug 20, 2021 at 10:58:56AM -0400, Derrick Stolee wrote:

Show 22 quoted lines

> On 8/20/2021 4:44 AM, Mike Hommey wrote:
> >  	normalized_url = url_normalize(url.buf, &config.url);
> >  
> >  	git_config(urlmatch_config_entry, &config);
> > +	string_list_clear(&config.vars, 1);
> >  	free(normalized_url);
> >  	strbuf_release(&url);
> 
> A good find! This is obviously correct and a valuable change
> to make. If you are interested in doing a little extra work,
> then I think there is something more we could do here.
> 
> I took a look at the rest of "struct urlmatch_config" to see
> if anything else needed to be cleared, and it turns out that
> config.url.url is an allocated string, but happens to be
> equal to normalized_url, which is freed here.
> 
> Perhaps the optimal organization would be to have a
> clear_urlmatch_config() method that clears all allocated data
> within the config, and change things like url_normalize()
> return a 'const char *' to make it clear that the url should
> be freed somewhere else.

Yeah, I had the same thought; it feels like we're peeking into details of how url_config works (especially the knowledge that we we should be passing free_util).

> It would help unify the handling of code that is somewhat
> duplicated (but slightly different each time) across
> credential_apply_config(), http_init(), get_urlmatch(),
> and cmd__urlmatch_normalization().

Agreed. It looks like http_init() has the same leak that is fixed here.

-Peff