Three speed-ups in the output path (two small, one substantial):

1) Require all callers to pass a valid route pointer to ip_output()
   so that we don't have to check and allocate one off the stack
   as was done before.  This eliminates one test and some stack
    bloat from the common (UDP and TCP) case.

2) Perform the IP header checksum in-line if it's of the usual length.
    This results in about a 5% speed-up in my packet-generation test.

3) Use ip_vhl field rather than ip_v and ip_hl bitfields.

1 files changed, 51 insertions, 39 deletions

diff --git a/sys/netinet/ip_output.c b/sys/netinet/ip_output.c
index 628921a..8400c46 100644
--- a/sys/netinet/ip_output.c
+++ b/sys/netinet/ip_output.c
@@ -31,9 +31,11 @@
  * SUCH DAMAGE.
  *
  *	@(#)ip_output.c	8.3 (Berkeley) 1/21/94
- *	$Id: ip_output.c,v 1.33 1996/03/26 18:56:51 fenner Exp $
+ *	$Id: ip_output.c,v 1.34 1996/04/03 13:52:20 phk Exp $
  */
 
+#define _IP_VHL
+
 #include <sys/param.h>
 #include <sys/queue.h>
 #include <sys/systm.h>
@@ -57,6 +59,7 @@
 #ifdef vax
 #include <machine/mtpr.h>
 #endif
+#include <machine/in_cksum.h>
 
 u_short ip_id;
 
@@ -89,30 +92,15 @@ ip_output(m0, opt, ro, flags, imo)
 	struct mbuf *m = m0;
 	int hlen = sizeof (struct ip);
 	int len, off, error = 0;
-	/*
-	 * It might seem obvious at first glance that one could easily
-	 * make a one-behind cache out of this by simply making `iproute'
-	 * static and eliminating the bzero() below.  However, this turns
-	 * out not to work, for two reasons:
-	 *
-	 * 1) This routine needs to be reentrant.  It can be called
-	 * recursively from encapsulating network interfaces, and it
-	 * is always called recursively from ip_mforward().
-	 *
-	 * 2) You turn out not to gain much.  There is already a one-
-	 * behind cache implemented for the specific case of forwarding,
-	 * and sends on a connected socket will use a route associated
-	 * with the PCB.  The only cases left are sends on unconnected
-	 * and raw sockets, and if these cases are really significant,
-	 * something is seriously wrong.
-	 */
-	struct route iproute;
 	struct sockaddr_in *dst;
 	struct in_ifaddr *ia;
 
 #ifdef	DIAGNOSTIC
 	if ((m->m_flags & M_PKTHDR) == 0)
 		panic("ip_output no HDR");
+	if (!ro)
+		panic("ip_output no route, proto = %d",
+		      mtod(m, struct ip *)->ip_p);
 #endif
 	if (opt) {
 		m = ip_insertoptions(m, opt, &len);
@@ -123,21 +111,14 @@ ip_output(m0, opt, ro, flags, imo)
 	 * Fill in IP header.
 	 */
 	if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) {
-		ip->ip_v = IPVERSION;
+		ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
 		ip->ip_off &= IP_DF;
 		ip->ip_id = htons(ip_id++);
-		ip->ip_hl = hlen >> 2;
 		ipstat.ips_localout++;
 	} else {
-		hlen = ip->ip_hl << 2;
-	}
-	/*
-	 * Route packet.
-	 */
-	if (ro == 0) {
-		ro = &iproute;
-		bzero((caddr_t)ro, sizeof (*ro));
+		hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 	}
+
 	dst = (struct sockaddr_in *)&ro->ro_dst;
 	/*
 	 * If there is a cached route,
@@ -349,7 +330,11 @@ sendit:
 		ip->ip_len = htons((u_short)ip->ip_len);
 		ip->ip_off = htons((u_short)ip->ip_off);
 		ip->ip_sum = 0;
-		ip->ip_sum = in_cksum(m, hlen);
+		if (ip->ip_vhl == IP_VHL_BORING) {
+			ip->ip_sum = in_cksum_hdr(ip);
+		} else {
+			ip->ip_sum = in_cksum(m, hlen);
+		}
 		error = (*ifp->if_output)(ifp, m,
 				(struct sockaddr *)dst, ro->ro_rt);
 		goto done;
@@ -405,7 +390,7 @@ sendit:
 		*mhip = *ip;
 		if (hlen > sizeof (struct ip)) {
 			mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip);
-			mhip->ip_hl = mhlen >> 2;
+			mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
 		}
 		m->m_len = mhlen;
 		mhip->ip_off = ((off - hlen) >> 3) + (ip->ip_off & ~IP_MF);
@@ -427,7 +412,11 @@ sendit:
 		m->m_pkthdr.rcvif = (struct ifnet *)0;
 		mhip->ip_off = htons((u_short)mhip->ip_off);
 		mhip->ip_sum = 0;
-		mhip->ip_sum = in_cksum(m, mhlen);
+		if (mhip->ip_vhl == IP_VHL_BORING) {
+			mhip->ip_sum = in_cksum_hdr(ip);
+		} else {
+			mhip->ip_sum = in_cksum(m, mhlen);
+		}
 		*mnext = m;
 		mnext = &m->m_nextpkt;
 		ipstat.ips_ofragments++;
@@ -442,7 +431,11 @@ sendit:
 	ip->ip_len = htons((u_short)m->m_pkthdr.len);
 	ip->ip_off = htons((u_short)(ip->ip_off | IP_MF));
 	ip->ip_sum = 0;
-	ip->ip_sum = in_cksum(m, hlen);
+	if (ip->ip_vhl == IP_VHL_BORING) {
+		ip->ip_sum = in_cksum_hdr(ip);
+	} else {
+		ip->ip_sum = in_cksum(m, hlen);
+	}
 sendorfree:
 	for (m = m0; m; m = m0) {
 		m0 = m->m_nextpkt;
@@ -458,9 +451,6 @@ sendorfree:
 		ipstat.ips_fragmented++;
     }
 done:
-	if (ro == &iproute && (flags & IP_ROUTETOIF) == 0 && ro->ro_rt)
-		RTFREE(ro->ro_rt);
-
 	return (error);
 bad:
 	m_freem(m0);
@@ -511,7 +501,7 @@ ip_insertoptions(m, opt, phlen)
 	ip = mtod(m, struct ip *);
 	(void)memcpy(ip + 1, p->ipopt_list, (unsigned)optlen);
 	*phlen = sizeof(struct ip) + optlen;
-	ip->ip_hl = *phlen >> 2;
+	ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
 	ip->ip_len += optlen;
 	return (m);
 }
@@ -529,7 +519,7 @@ ip_optcopy(ip, jp)
 
 	cp = (u_char *)(ip + 1);
 	dp = (u_char *)(jp + 1);
-	cnt = (ip->ip_hl << 2) - sizeof (struct ip);
+	cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof (struct ip);
 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
 		opt = cp[0];
 		if (opt == IPOPT_EOL)
@@ -1231,7 +1221,29 @@ ip_mloopback(ifp, m, dst)
 		ip->ip_len = htons((u_short)ip->ip_len);
 		ip->ip_off = htons((u_short)ip->ip_off);
 		ip->ip_sum = 0;
-		ip->ip_sum = in_cksum(copym, ip->ip_hl << 2);
+		if (ip->ip_vhl == IP_VHL_BORING) {
+			ip->ip_sum = in_cksum_hdr(ip);
+		} else {
+			ip->ip_sum = in_cksum(copym, 
+					      IP_VHL_HL(ip->ip_vhl) << 2);
+		}
+		/*
+		 * NB:
+		 * We can't simply call ip_input() directly because
+		 * the ip_mforward() depends on the `input interface'
+		 * being set to something unreasonable so that we don't
+		 * attempt to forward the looped-back copy.  
+		 * It's also not clear whether there are any lingering
+		 * reentrancy problems in other areas which might be
+		 * exposed by this code.  For the moment, we'll err
+		 * on the side of safety by continuing to abuse
+		 * loinput().
+		 */
+#ifdef notdef
+		copym->m_pkthdr.rcvif = &loif[0];
+		ip_input(copym)
+#else
 		(void) looutput(ifp, copym, (struct sockaddr *)dst, NULL);
+#endif
 	}
 }