[src/trunk]: src/sys/netinet Hold ip_off and ip_len in the fragment entry, in...

[maxv <maxv%NetBSD.org@localhost>]

details:   https://anonhg.NetBSD.org/src/rev/45c5a2d359f2
branches:  trunk
changeset: 433500:45c5a2d359f2
user:      maxv <maxv%NetBSD.org@localhost>
date:      Mon Sep 17 06:01:36 2018 +0000

description:
Hold ip_off and ip_len in the fragment entry, instead of always reading
the associated mbuf (and converting to host order). This reduces the
cache/TLB misses when processing long lists.

diffstat:

 sys/netinet/ip_reass.c |  63 +++++++++++++++++--------------------------------
 1 files changed, 22 insertions(+), 41 deletions(-)

diffs (156 lines):

diff -r 868b9fbfa0a7 -r 45c5a2d359f2 sys/netinet/ip_reass.c
--- a/sys/netinet/ip_reass.c    Mon Sep 17 01:49:54 2018 +0000
+++ b/sys/netinet/ip_reass.c    Mon Sep 17 06:01:36 2018 +0000
@@ -1,4 +1,4 @@
-/*     $NetBSD: ip_reass.c,v 1.18 2018/07/10 15:46:58 maxv Exp $       */
+/*     $NetBSD: ip_reass.c,v 1.19 2018/09/17 06:01:36 maxv Exp $       */
 
 /*
  * Copyright (c) 1982, 1986, 1988, 1993
@@ -46,7 +46,7 @@
  */
 
 #include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: ip_reass.c,v 1.18 2018/07/10 15:46:58 maxv Exp $");
+__KERNEL_RCSID(0, "$NetBSD: ip_reass.c,v 1.19 2018/09/17 06:01:36 maxv Exp $");
 
 #include <sys/param.h>
 #include <sys/types.h>
@@ -80,6 +80,8 @@
        struct ip *             ipqe_ip;
        struct mbuf *           ipqe_m;
        bool                    ipqe_mff;
+       uint16_t                ipqe_off;
+       uint16_t                ipqe_len;
 } ipfr_qent_t;
 
 TAILQ_HEAD(ipfr_qent_head, ipfr_qent);
@@ -215,7 +217,7 @@
 static struct mbuf *
 ip_reass(ipfr_qent_t *ipqe, ipfr_queue_t *fp, const u_int hash)
 {
-       struct ip *ip = ipqe->ipqe_ip, *qip;
+       struct ip *ip = ipqe->ipqe_ip;
        const int hlen = ip->ip_hl << 2;
        struct mbuf *m = ipqe->ipqe_m, *t;
        int ipsecflags = m->m_flags & (M_DECRYPTED|M_AUTHIPHDR);
@@ -230,16 +232,6 @@
        m->m_data += hlen;
        m->m_len -= hlen;
 
-#ifdef notyet
-       /* Make sure fragment limit is up-to-date. */
-       CHECK_NMBCLUSTER_PARAMS();
-
-       /* If we have too many fragments, drop the older half. */
-       if (ip_nfrags >= ip_maxfrags) {
-               ip_reass_drophalf(void);
-       }
-#endif
-
        /*
         * We are about to add a fragment; increment frag count.
         */
@@ -255,9 +247,9 @@
                 * never accept fragments  b) if maxfrag is -1, accept
                 * all fragments without limitation.
                 */
-               if (ip_maxfragpackets < 0)
-                       ;
-               else if (ip_nfragpackets >= ip_maxfragpackets) {
+               if (ip_maxfragpackets < 0) {
+                       /* no limit */
+               } else if (ip_nfragpackets >= ip_maxfragpackets) {
                        goto dropfrag;
                }
                fp = malloc(sizeof(ipfr_queue_t), M_FTABLE, M_NOWAIT);
@@ -285,7 +277,7 @@
         * Find a segment which begins after this one does.
         */
        TAILQ_FOREACH(q, &fp->ipq_fragq, ipqe_q) {
-               if (ntohs(q->ipqe_ip->ip_off) > ntohs(ip->ip_off))
+               if (q->ipqe_off > ipqe->ipqe_off)
                        break;
        }
        if (q != NULL) {
@@ -300,15 +292,14 @@
         * If it provides all of our data, drop us.
         */
        if (p != NULL) {
-               i = ntohs(p->ipqe_ip->ip_off) + ntohs(p->ipqe_ip->ip_len) -
-                   ntohs(ip->ip_off);
+               i = p->ipqe_off + p->ipqe_len - ipqe->ipqe_off;
                if (i > 0) {
-                       if (i >= ntohs(ip->ip_len)) {
+                       if (i >= ipqe->ipqe_len) {
                                goto dropfrag;
                        }
                        m_adj(ipqe->ipqe_m, i);
-                       ip->ip_off = htons(ntohs(ip->ip_off) + i);
-                       ip->ip_len = htons(ntohs(ip->ip_len) - i);
+                       ipqe->ipqe_off = ipqe->ipqe_off + i;
+                       ipqe->ipqe_len = ipqe->ipqe_len - i;
                }
        }
 
@@ -317,17 +308,13 @@
         * completely covered, dequeue them.
         */
        while (q != NULL) {
-               size_t end;
-
-               qip = q->ipqe_ip;
-               end = ntohs(ip->ip_off) + ntohs(ip->ip_len);
-               if (end <= ntohs(qip->ip_off)) {
+               i = ipqe->ipqe_off + ipqe->ipqe_len - q->ipqe_off;
+               if (i <= 0) {
                        break;
                }
-               i = end - ntohs(qip->ip_off);
-               if (i < ntohs(qip->ip_len)) {
-                       qip->ip_len = htons(ntohs(qip->ip_len) - i);
-                       qip->ip_off = htons(ntohs(qip->ip_off) + i);
+               if (i < q->ipqe_len) {
+                       q->ipqe_off = q->ipqe_off + i;
+                       q->ipqe_len = q->ipqe_len - i;
                        m_adj(q->ipqe_m, i);
                        break;
                }
@@ -351,12 +338,11 @@
        }
        next = 0;
        TAILQ_FOREACH(q, &fp->ipq_fragq, ipqe_q) {
-               qip = q->ipqe_ip;
-               if (ntohs(qip->ip_off) != next) {
+               if (q->ipqe_off != next) {
                        mutex_exit(&ipfr_lock);
                        return NULL;
                }
-               next += ntohs(qip->ip_len);
+               next += q->ipqe_len;
        }
        p = TAILQ_LAST(&fp->ipq_fragq, ipfr_qent_head);
        if (p->ipqe_mff) {
@@ -652,13 +638,6 @@
                return EINVAL;
        }
 
-       /*
-        * Adjust total IP length to not reflect header and convert
-        * offset of this to bytes.  XXX: clobbers struct ip.
-        */
-       ip->ip_len = htons(flen);
-       ip->ip_off = htons(off);
-
        /* Look for queue of fragments of this datagram. */
        mutex_enter(&ipfr_lock);
        hash = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
@@ -703,6 +682,8 @@
        ipqe->ipqe_mff = mff;
        ipqe->ipqe_m = m;
        ipqe->ipqe_ip = ip;
+       ipqe->ipqe_off = off;
+       ipqe->ipqe_len = flen;
 
        *m0 = ip_reass(ipqe, fp, hash);
        if (*m0) {