Re: [PATCH v7 17/18] conf, fwd: Allow switching to new rules received from pesto

[David Gibson <david@gibson.dropbear.id.au>]

[-- Attachment #1: Type: text/plain, Size: 5375 bytes --]

On Tue, May 05, 2026 at 12:04:09PM +0200, Stefano Brivio wrote:
> On Tue, 5 May 2026 11:08:27 +0200
> Laurent Vivier <lvivier@redhat.com> wrote:
> 
> > On 5/5/26 01:11, Stefano Brivio wrote:
> > > From: David Gibson <david@gibson.dropbear.id.au>
> > > 
> > > We can now receive updates to the forwarding rules from the pesto client
> > > and store them in a "pending" copy of the forwarding tables.  Implement
> > > switching to using the new rules.
> > > 
> > > The logic is in a new fwd_listen_switch().  For now this closes all
> > > listening sockets related to the old tables, swaps the active and pending
> > > tables, then listens based on the new tables.  In future we look to improve
> > > this so that we don't temporarily stop listening on ports that both the
> > > old and new tables specify.
> > > 
> > > Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
> > > Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
> > > ---
> > >   conf.c |  5 ++---
> > >   fwd.c  | 34 ++++++++++++++++++++++++++++++++++
> > >   fwd.h  |  1 +
> > >   3 files changed, 37 insertions(+), 3 deletions(-)
> > > 
> > > diff --git a/conf.c b/conf.c
> > > index f035fd3..75b8291 100644
> > > --- a/conf.c
> > > +++ b/conf.c
> > > @@ -2159,15 +2159,14 @@ void conf_handler(struct ctx *c, uint32_t events)
> > >   			fwd_rules_dump(info, fwd->rules, fwd->count,
> > >   				       "    ", "");
> > >   		}
> > > +
> > > +		fwd_listen_switch(c);
> > >   	}
> > >   
> > >   	if (events & EPOLLHUP) {
> > >   		debug("Configuration client hangup");
> > > -		goto close;
> > >   	}
> > >   
> > > -	return;
> > > -
> > >   close:
> > >   	conf_close(c);
> > >   
> > > diff --git a/fwd.c b/fwd.c
> > > index d93d2e5..35b9e2b 100644
> > > --- a/fwd.c
> > > +++ b/fwd.c
> > > @@ -534,6 +534,40 @@ int fwd_listen_init(const struct ctx *c)
> > >   	return 0;
> > >   }
> > >   
> > > +/**
> > > + * fwd_listen_switch() - Switch from current to pending rules table
> > > + * @c:		Execution context
> > > + */
> > > +void fwd_listen_switch(struct ctx *c)
> > > +{
> > > +	struct fwd_table *tmp[PIF_NUM_TYPES];
> > > +	unsigned i;
> > > +
> > > +	/* Stop listening on the old tables */
> > > +	for (i = 0; i < PIF_NUM_TYPES; i++) {
> > > +		struct fwd_table *fwd = c->fwd[i];
> > > +
> > > +		if (!fwd)
> > > +			continue;
> > > +
> > > +		debug("Flushing %u old %s rules", fwd->count, pif_name(i));
> > > +		fwd_listen_close(fwd);
> > > +		fwd->count = fwd->sock_count = 0;  
> > 
> > Perhaps we can reset fwd->count and fwd->sock_count in fwd_listen_close() as after 
> > fwd_listen_close() these values are wrong?
> 
> Right, while not strictly necessary it still looks like a good idea,
> I'll change that.

As noted elswhere this is correct as it is.

> > > +	}
> > > +
> > > +	/* Swap active and pending tables */
> > > +	static_assert(sizeof(tmp) == sizeof(c->fwd) &&
> > > +		      sizeof(tmp) == sizeof(c->fwd_pending),
> > > +		      "Temporary has wrong size");
> > > +	memcpy(&tmp, (void *)c->fwd, sizeof(tmp));
> > > +	memcpy((void *)c->fwd, (void *)c->fwd_pending, sizeof(tmp));
> > > +	memcpy((void *)c->fwd_pending, &tmp, sizeof(tmp));  
> > 
> > I know we have the static_assert(), but with memcpy() we usually use the sizeof() of the 
> > destination to avoid write overflow.
> 
> I'll change this as well.

Sounds good.


> > Why do we keep the old active table? Do we plan to have a "--restore" option?
> 
> It's just to add and delete rules using a temporary table so that we
> can abort cleanly and atomically on errors.
> 
> Are you asking why we don't wipe the old table afterwards? No
> particular reason for that, even though I'm not sure if it's useful.

Also looking ahead to listening socket continuity.  In that case we
need to keep the old table until we've stolen all the sockets we can
re-use for the new table.  *Then* we can fwd_listen_close() anything
left over, and wipe the table.

> Actually some kind of --restore option might be desirable, even though
> we would probably need to re-validate all the rules, or keep a "dirty"
> bit that's set on other types of changes and would tell us that the
> previous table can't be used as it is anymore.
> 
> -- 
> Stefano
> 
> 
> > > +
> > > +	/* Start listening on the new tables */
> > > +	if (fwd_listen_init(c) < 0)
> > > +		err("Error switching to new forwarding rules");
> > > +}
> > > +
> > >   /* See enum in kernel's include/net/tcp_states.h */
> > >   #define UDP_LISTEN	0x07
> > >   #define TCP_LISTEN	0x0a
> > > diff --git a/fwd.h b/fwd.h
> > > index ac24782..b60697d 100644
> > > --- a/fwd.h
> > > +++ b/fwd.h
> > > @@ -61,6 +61,7 @@ int fwd_listen_sync(const struct ctx *c, uint8_t pif,
> > >   		    const struct fwd_scan *tcp, const struct fwd_scan *udp);
> > >   void fwd_listen_close(const struct fwd_table *fwd);
> > >   int fwd_listen_init(const struct ctx *c);
> > > +void fwd_listen_switch(struct ctx *c);
> > >   
> > >   bool nat_inbound(const struct ctx *c, const union inany_addr *addr,
> > >   		 union inany_addr *translated);  
> > 
> 

-- 
David Gibson (he or they)	| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you, not the other way
				| around.
http://www.ozlabs.org/~dgibson

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]