From: Rafael Sadowski <rafael@sizeofvoid.org>
Subject: Re: httpd: fix scan-build dead stores findings
To: Theo Buehler <tb@theobuehler.org>, tech@openbsd.org
Date: Sat, 3 Jan 2026 08:18:32 +0100

On Fri Jan 02, 2026 at 10:57:14PM +0100, Claudio Jeker wrote:
> On Fri, Jan 02, 2026 at 09:51:52PM +0100, Rafael Sadowski wrote:
> > On Fri Jan 02, 2026 at 08:48:09PM +0100, Theo Buehler wrote:
> > > On Fri, Jan 02, 2026 at 08:14:29PM +0100, Rafael Sadowski wrote:
> > > > On Thu Jan 01, 2026 at 10:12:57AM +0100, Stefan Sperling wrote:
> > > > > On Sun, Dec 28, 2025 at 06:55:28AM +0100, Rafael Sadowski wrote:
> > > > > > Hi,
> > > > > > 
> > > > > > scan-build flagged a few dead stores in httpd:
> > > > > > 
> > > > > > - config.c:636 – s incremented but never used after
> > > > > > - httpd.c:533 – q assigned but immediately overwritten in the loop
> > > > > > - server.c:891 – inrd/inwr assigned, then reassigned, never actually needed
> > > > > > - server_fcgi.c:678,690 – kv result unused
> > > > > > 
> > > > > > OK?
> > > > > > 
> > > > > > Diff removes all of them. OK?
> > > > > > 1. src/usr.sbin/httpd/config.c:636:3: warning: Value stored to 's' is never read [deadcode.DeadStores]
> > > > > >   636 |                 s += srv_conf->return_uri_len;
> > > > > > 
> > > > > > 2. src/usr.sbin/httpd/httpd.c:533:6: warning: Although the value stored to 'q' is used in the enclosing expression, the value is never actually read from 'q' [deadcode.DeadStores]
> > > > > >   533 |         p = q = label;
> > > > > >       |             ^   ~~~~~
> > > > > > 
> > > > > > 3. src/usr.sbin/httpd/server.c:891:15: warning: Value stored to 'inrd' during its initialization is never read [deadcode.DeadStores]
> > > > > >   891 |         evbuffercb               inrd = server_read;
> > > > > >       |                                  ^~~~   ~~~~~~~~~~~
> > > > > > 
> > > > > > 4. src/usr.sbin/httpd/server_fcgi.c:678:9: warning: Although the value stored to 'kv' is used in the enclosing expression, the value is never actually read from 'kv' [deadcode.DeadStores]
> > > > > >   678 |                         if ((kv = kv_find(&resp->http_headers, &key)) == NULL) {
> > > > > >       |                              ^    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > > > > > src/usr.sbin/httpd/server_fcgi.c:690:8: warning: Although the value stored to 'kv' is used in the enclosing expression, the value is never actually read from 'kv' [deadcode.DeadStores]
> > > > > >   690 |                 if ((kv = kv_find(&resp->http_headers, &key)) != NULL) {
> > > > > >       |                      ^    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > > > > > 
> > > > > > 
> > > > > > diff --git a/usr.sbin/httpd/config.c b/usr.sbin/httpd/config.c
> > > > > > index b45081129b7..aa2cd382ed3 100644
> > > > > > --- a/usr.sbin/httpd/config.c
> > > > > > +++ b/usr.sbin/httpd/config.c
> > > > > > @@ -531,7 +531,6 @@ config_getserver_config(struct httpd *env, struct server *srv,
> > > > > >  		if ((srv_conf->return_uri = get_data(p + s,
> > > > > >  		    srv_conf->return_uri_len)) == NULL)
> > > > > >  			goto fail;
> > > > > > -		s += srv_conf->return_uri_len;
> > > > > 
> > > > > I am not 100% sure about this one. On one hand scan-build is right that the
> > > > > value of s is never used in the remainder of this function. On the other
> > > > > hand, this code seems to be written with the assumption in mind that more
> > > > > optional chuks of data with variable length could be added in the future.
> > > > > If we ever grow the server config structure and process more data via the
> > > > > 'p + s' idiom then this line would need to be added back in to avoid a bug.
> > > > > So perhaps this line is worth keeping?
> > > > > 
> > > > > The rest is ok stsp@
> > > > 
> > > > Thanks Stefan,
> > > > 
> > > > I've committed the rest of the dead store fixes but left this one out for now.
> > > > 
> > > > I see your point about future expansion, but I'd lean towards removing
> > > > it. The comment already documents that more variable-length fields could
> > > > be added. If someone later adds TLS data or other fields, they will need
> > > > to modify this function anyway.
> > > 
> > > I think the more important thing to fix here is that there is no check
> > > whether the imsg contains enough data to copy reurn_uri_len bytes from
> > > p + s.
> > 
> > Thanks Theo! That's right, there is no check as in config_getserver.
> > Tested with regress/usr.sbin/httpd.
> > 
> > diff --git a/usr.sbin/httpd/config.c b/usr.sbin/httpd/config.c
> > index b45081129b7..7d41b57305f 100644
> > --- a/usr.sbin/httpd/config.c
> > +++ b/usr.sbin/httpd/config.c
> > @@ -513,6 +513,11 @@ config_getserver_config(struct httpd *env, struct server *srv,
> >  	/* Reset these variables to avoid free'ing invalid pointers */
> >  	serverconfig_reset(srv_conf);
> >  
> > +	if ((IMSG_DATA_SIZE(imsg) - s) < (size_t)srv_conf->return_uri_len) {
> > +		log_debug("%s: invalid message length", __func__);
> > +		goto fail;
> > +	}
> > +
> >  	TAILQ_FOREACH(parent, &srv->srv_hosts, entry) {
> >  		if (strcmp(parent->name, srv_conf->name) == 0)
> >  			break;
> > @@ -531,7 +536,6 @@ config_getserver_config(struct httpd *env, struct server *srv,
> >  		if ((srv_conf->return_uri = get_data(p + s,
> >  		    srv_conf->return_uri_len)) == NULL)
> >  			goto fail;
> > -		s += srv_conf->return_uri_len;
> >  	}
> >  
> >  	if (srv_conf->flags & SRVFLAG_LOCATION) {
> > 
> 
> It would be much nicer to use the real ibuf apis and not use direct imsg
> data access, then all those length checks can go. Also IMSG_DATA_SIZE is
> spelled imsg_get_len().

After reading a bit in bgpd and vmctl (In contrast, httpd/relayd is
really difficult to read), I completely agree with you, but I think this
will require a major design change. I would like to fix this here first.

OK?