On Thu, Nov 28, 2024 at 08:21:23PM -0500, Jon Maloy wrote:On 2024-11-28 19:45, Jon Maloy wrote:Right. I think the confusion here is because splicing kind of does two things. First, it takes some cases that would work with tap, but optimises them. Second it makes some cases possible that aren't possible with just the tap interface: specifically redirecting guest side traffic with destination 0.0.0.0 or 127.0.0.1/8. I've been assuming that a --no-splice option would disable both cases. So traffic that *can* be redirected via tap instead would be, but things that are only possible with splice would just be disallowed. The latter has the arguable advantage that it eliminates the (small) network behavioural differences between pasta and passt mode.During testing it is sometimes useful to force traffic which would normally be forwarded by socket splicing through the tap interface. In this commit, we add a command switch making it possible to disable splicing for inbound local traffic. For outbound local traffic this seems to be much trickier, so I leave that for a possible later commit.I am looking for more input here. David suggested that I simply don't re-bind any sockets inwards towards the local namespace, so that all outbound traffic would use the default route and be forced to go via the tap interface. I tried this, and realized it won't work. Outgoing traffic using INADDR_ANY or loopback address will never be routed via the default route; if it doesn't find the destination port in the local name space it will simply return with 'connection refused'. There is no nice way to force such traffic via the default route, as far as I understand.I am even questioning if it is necessary: If the port is bound on the host, the client only needs to use some of the non-loopback addresses on the host to reach it via the tap interface.Right. In fact with both --no-splice as per your draft, and no -T and -U options, I don't think any use of splice is possible. We could maybe put a test in fwd_nat_from_splice() to check.///jon-- 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/~dgibsonSuggested-by: David Gibson <dgibson(a)redhat.com> Signed-off-by: Jon Maloy <jmaloy(a)redhat.com> --- conf.c | 5 +++++ fwd.c | 2 +- passt.h | 1 + 3 files changed, 7 insertions(+), 1 deletion(-) diff --git a/conf.c b/conf.c index eaa7d99..8d58652 100644 --- a/conf.c +++ b/conf.c @@ -890,6 +890,7 @@ static void usage(const char *name, FILE *f, int status) " --no-ndp Disable NDP responses\n" " --no-dhcpv6 Disable DHCPv6 server\n" " --no-ra Disable router advertisements\n" + " --no-splice Disable outbound socket splicing\n" " --freebind Bind to any address for forwarding\n" " --no-map-gw Don't map gateway address to host\n" " -4, --ipv4-only Enable IPv4 operation only\n" @@ -1319,6 +1320,7 @@ void conf(struct ctx *c, int argc, char **argv) {"no-dhcpv6", no_argument, &c->no_dhcpv6, 1 }, {"no-ndp", no_argument, &c->no_ndp, 1 }, {"no-ra", no_argument, &c->no_ra, 1 }, + {"no-splice", no_argument, &c->no_splice, 1 }, {"freebind", no_argument, &c->freebind, 1 }, {"no-map-gw", no_argument, &no_map_gw, 1 }, {"ipv4-only", no_argument, NULL, '4' }, @@ -1756,6 +1758,9 @@ void conf(struct ctx *c, int argc, char **argv) } } while (name != -1); + if (c->mode == MODE_PASST) + c->no_splice = 1; + if (c->mode == MODE_PASTA && !c->pasta_conf_ns) { if (copy_routes_opt) die("--no-copy-routes needs --config-net"); diff --git a/fwd.c b/fwd.c index 0b7f8b1..2829cd2 100644 --- a/fwd.c +++ b/fwd.c @@ -443,7 +443,7 @@ uint8_t fwd_nat_from_host(const struct ctx *c, uint8_t proto, else if (proto == IPPROTO_UDP) tgt->eport += c->udp.fwd_in.delta[tgt->eport]; - if (c->mode == MODE_PASTA && inany_is_loopback(&ini->eaddr) && + if (!c->no_splice && inany_is_loopback(&ini->eaddr) && (proto == IPPROTO_TCP || proto == IPPROTO_UDP)) { /* spliceable */ diff --git a/passt.h b/passt.h index c038630..0271e7c 100644 --- a/passt.h +++ b/passt.h @@ -291,6 +291,7 @@ struct ctx { int no_dhcpv6; int no_ndp; int no_ra; + int no_splice; int host_lo_to_ns_lo; int freebind;