From: Omar Polo <op@omarpolo.com>
Subject: Re: gotwebd fcgi/gotweb process split
To: Stefan Sperling <stsp@stsp.name>
Cc: gameoftrees@openbsd.org
Date: Mon, 14 Apr 2025 20:58:39 +0200

Stefan Sperling <stsp@stsp.name> wrote:
> On Fri, Apr 11, 2025 at 04:43:40PM +0200, Stefan Sperling wrote:
> > On Thu, Apr 10, 2025 at 10:52:59AM +0200, Stefan Sperling wrote:
> > > On Wed, Apr 09, 2025 at 06:31:42PM +0200, Stefan Sperling wrote:
> > > > This patch is still work-in-progress. It lacks documentation updates,
> > > > and one TODO is mentioned in the code.
> > > 
> > > updated patch which resolves remaining TODOs and updates docs, too.
> > 
> > Another update which fixes the two issues found by me so far:
> > 
> > 1) We need to do polling writes in fcgi_forward to avoid:
> >   fcgi_forward_response: wrote -1 of 1024 bytes: \
> >     Resource temporarily unavailable
> > 
> > 2) fix hang during '/etc/rc.d/gotwebd restart', causing rc to kill gotwebd
> > 
> 
> More issues fixed:
> 
>  1) Omar noticed a env->user NULL pointer deref, fixed here.
> 
>  2) When running as the new _gotwebd user the socket in /var/www/run must
>     be writable by the www user. Ensure it is writable and make the httpd
>     user's name configurable via gotwebd.conf.

modulo some stylistic nitpicks that we'll eventually sort out in-tree,
this looks fine to me!  That's quite the diff by the way, thank you!

one minor point is that now we run twice the prefork processes (gotwebd
and server), not sure if we want to actually point that out in the man
page.

ok op@

i'm running with this on my instance btw