Hi,
While running forwarding in parallel, I have introduced a hard
barrier for parallel local protocol processing. The packets are
requeued from shared to exclusive netlock.
Unless we unlock all protocol input routines at once, we need some
mechanism from to move from one queue to the other. The problem
is, that we have to remember next protocol field and offset of the
parsed header chain.
A simple example is IPv6 hop-by-hop options precessing. This code
is MP safe and can be moved from ip6_local() to ip6_ours() to run
in parallel. If there were any options, the offset and next protocol
are stored in a mbuf tag. Without tag we know that it is a regular
IPv6 header.
Of course mbuf tags kill performance, but who uses hop-by-hop options
anyway? pf drops such packets per default.
ok?
bluhm
Index: netinet/ip_input.c
===================================================================
RCS file: /data/mirror/openbsd/cvs/src/sys/netinet/ip_input.c,v
retrieving revision 1.378
diff -u -p -r1.378 ip_input.c
--- netinet/ip_input.c 12 Aug 2022 14:49:15 -0000 1.378
+++ netinet/ip_input.c 13 Aug 2022 13:36:17 -0000
@@ -560,11 +560,13 @@ ip_input_if(struct mbuf **mp, int *offp,
int
ip_local(struct mbuf **mp, int *offp, int nxt, int af)
{
- struct ip *ip;
+ if (*offp == 0) {
+ struct ip *ip;
- ip = mtod(*mp, struct ip *);
- *offp = ip->ip_hl << 2;
- nxt = ip->ip_p;
+ ip = mtod(*mp, struct ip *);
+ *offp = ip->ip_hl << 2;
+ nxt = ip->ip_p;
+ }
/* Check whether we are already in a IPv4/IPv6 local deliver loop. */
if (af == AF_UNSPEC)
Index: netinet6/ip6_input.c
===================================================================
RCS file: /data/mirror/openbsd/cvs/src/sys/netinet6/ip6_input.c,v
retrieving revision 1.252
diff -u -p -r1.252 ip6_input.c
--- netinet6/ip6_input.c 12 Aug 2022 14:49:15 -0000 1.252
+++ netinet6/ip6_input.c 13 Aug 2022 13:42:42 -0000
@@ -167,6 +167,11 @@ ip6_init(void)
#endif
}
+struct ip_offnxt {
+ int ion_off;
+ int ion_nxt;
+};
+
/*
* Enqueue packet for local delivery. Queuing is used as a boundary
* between the network layer (input/forward path) running with
@@ -175,10 +180,37 @@ ip6_init(void)
int
ip6_ours(struct mbuf **mp, int *offp, int nxt, int af)
{
+ /* ip6_hbhchcheck() may be run before, then off and nxt are set */
+ if (*offp == 0) {
+ nxt = ip6_hbhchcheck(mp, offp, NULL);
+ if (nxt == IPPROTO_DONE)
+ return IPPROTO_DONE;
+ }
+
/* We are already in a IPv4/IPv6 local deliver loop. */
if (af != AF_UNSPEC)
return ip6_local(mp, offp, nxt, af);
+ /* save values for later, use after dequeue */
+ if (*offp != sizeof(struct ip6_hdr)) {
+ struct m_tag *mtag;
+ struct ip_offnxt *ion;
+
+ /* mbuf tags are expensive, but only used for header options */
+ mtag = m_tag_get(PACKET_TAG_IP6_OFF_NXT, sizeof(*ion),
+ M_NOWAIT);
+ if (mtag == NULL) {
+ ip6stat_inc(ip6s_idropped);
+ m_freemp(mp);
+ return IPPROTO_DONE;
+ }
+ ion = (struct ip_offnxt *)(mtag + 1);
+ ion->ion_off = *offp;
+ ion->ion_nxt = nxt;
+
+ m_tag_prepend(*mp, mtag);
+ }
+
niq_enqueue(&ip6intrq, *mp);
*mp = NULL;
return IPPROTO_DONE;
@@ -584,9 +616,27 @@ ip6_input_if(struct mbuf **mp, int *offp
int
ip6_local(struct mbuf **mp, int *offp, int nxt, int af)
{
- nxt = ip6_hbhchcheck(mp, offp, NULL);
- if (nxt == IPPROTO_DONE)
- return IPPROTO_DONE;
+ if (*offp == 0) {
+ struct m_tag *mtag;
+
+ mtag = m_tag_find(*mp, PACKET_TAG_IP6_OFF_NXT, NULL);
+ if (mtag != NULL) {
+ struct ip_offnxt *ion;
+
+ ion = (struct ip_offnxt *)(mtag + 1);
+ *offp = ion->ion_off;
+ nxt = ion->ion_nxt;
+
+ m_tag_delete(*mp, mtag);
+ } else {
+ struct ip6_hdr *ip6;
+
+ ip6 = mtod(*mp, struct ip6_hdr *);
+ *offp = sizeof(struct ip6_hdr);
+ nxt = ip6->ip6_nxt;
+
+ }
+ }
/* Check whether we are already in a IPv4/IPv6 local deliver loop. */
if (af == AF_UNSPEC)
Index: sys/mbuf.h
===================================================================
RCS file: /data/mirror/openbsd/cvs/src/sys/sys/mbuf.h,v
retrieving revision 1.254
diff -u -p -r1.254 mbuf.h
--- sys/mbuf.h 14 Feb 2022 04:33:18 -0000 1.254
+++ sys/mbuf.h 13 Aug 2022 13:36:17 -0000
@@ -479,6 +479,7 @@ struct m_tag *m_tag_next(struct mbuf *,
#define PACKET_TAG_SRCROUTE 0x1000 /* IPv4 source routing options */
#define PACKET_TAG_TUNNEL 0x2000 /* Tunnel endpoint address */
#define PACKET_TAG_CARP_BAL_IP 0x4000 /* carp(4) ip balanced marker */
+#define PACKET_TAG_IP6_OFF_NXT 0x8000 /* IPv6 offset and next proto */
#define MTAG_BITS \
("\20\1IPSEC_IN_DONE\2IPSEC_OUT_DONE\3IPSEC_FLOWINFO" \
