Module Name: src
Committed By: rin
Date: Wed Dec 12 01:40:21 UTC 2018
Modified Files:
src/sys/net: files.net
src/sys/netinet: in_offload.c in_offload.h
src/sys/netinet6: in6_offload.c in6_offload.h
src/sys/rump/net/lib/libnet: Makefile
Added Files:
src/sys/net: ether_sw_offload.c ether_sw_offload.h
Log Message:
PR kern/53562
Add ether_sw_offload_[tr]x: handle TX/RX offload options in software.
Since this violates separation b/w L2 and L3/L4, new files are added
rather than having the routines in sys/net/if_ethersubr.c.
OK msaitoh thorpej
To generate a diff of this commit:
cvs rdiff -u -r0 -r1.1 src/sys/net/ether_sw_offload.c \
src/sys/net/ether_sw_offload.h
cvs rdiff -u -r1.19 -r1.20 src/sys/net/files.net
cvs rdiff -u -r1.12 -r1.13 src/sys/netinet/in_offload.c
cvs rdiff -u -r1.11 -r1.12 src/sys/netinet/in_offload.h
cvs rdiff -u -r1.11 -r1.12 src/sys/netinet6/in6_offload.c
cvs rdiff -u -r1.9 -r1.10 src/sys/netinet6/in6_offload.h
cvs rdiff -u -r1.30 -r1.31 src/sys/rump/net/lib/libnet/Makefile
Please note that diffs are not public domain; they are subject to the
copyright notices on the relevant files.
Modified files:
Index: src/sys/net/files.net
diff -u src/sys/net/files.net:1.19 src/sys/net/files.net:1.20
--- src/sys/net/files.net:1.19 Sun Sep 23 09:21:03 2018
+++ src/sys/net/files.net Wed Dec 12 01:40:20 2018
@@ -1,4 +1,4 @@
-# $NetBSD: files.net,v 1.19 2018/09/23 09:21:03 maxv Exp $
+# $NetBSD: files.net,v 1.20 2018/12/12 01:40:20 rin Exp $
# XXX CLEANUP
define net
@@ -7,6 +7,7 @@ file net/bpf_filter.c bpf_filter
file net/bpf_stub.c net
file net/bsd-comp.c ppp & ppp_bsdcomp
file net/dl_print.c
+file net/ether_sw_offload.c bridge
file net/if.c net
file net/if_arcsubr.c arcnet needs-flag
file net/if_bridge.c bridge needs-flag
Index: src/sys/netinet/in_offload.c
diff -u src/sys/netinet/in_offload.c:1.12 src/sys/netinet/in_offload.c:1.13
--- src/sys/netinet/in_offload.c:1.12 Wed Sep 19 07:54:11 2018
+++ src/sys/netinet/in_offload.c Wed Dec 12 01:40:20 2018
@@ -1,4 +1,4 @@
-/* $NetBSD: in_offload.c,v 1.12 2018/09/19 07:54:11 rin Exp $ */
+/* $NetBSD: in_offload.c,v 1.13 2018/12/12 01:40:20 rin Exp $ */
/*
* Copyright (c)2005, 2006 YAMAMOTO Takashi,
@@ -27,7 +27,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: in_offload.c,v 1.12 2018/09/19 07:54:11 rin Exp $");
+__KERNEL_RCSID(0, "$NetBSD: in_offload.c,v 1.13 2018/12/12 01:40:20 rin Exp $");
#include <sys/param.h>
#include <sys/mbuf.h>
@@ -43,22 +43,23 @@ __KERNEL_RCSID(0, "$NetBSD: in_offload.c
/*
* Handle M_CSUM_TSOv4 in software. Split the TCP payload in chunks of
- * size MSS, and send them.
+ * size MSS, and return mbuf chain consists of them.
*/
-static int
-tcp4_segment(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *sa,
- struct rtentry *rt)
+struct mbuf *
+tcp4_segment(struct mbuf *m, int off)
{
int mss;
int iphlen, thlen;
int hlen, len;
struct ip *ip;
struct tcphdr *th;
- uint16_t ipid;
+ uint16_t ipid, phsum;
uint32_t tcpseq;
struct mbuf *hdr = NULL;
- struct mbuf *t;
- int error = 0;
+ struct mbuf *m0 = NULL;
+ struct mbuf *prev = NULL;
+ struct mbuf *n, *t;
+ int nsegs;
KASSERT((m->m_flags & M_PKTHDR) != 0);
KASSERT((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0);
@@ -66,107 +67,129 @@ tcp4_segment(struct ifnet *ifp, struct m
m->m_pkthdr.csum_flags = 0;
len = m->m_pkthdr.len;
- KASSERT(len >= sizeof(*ip) + sizeof(*th));
+ KASSERT(len >= off + sizeof(*ip) + sizeof(*th));
- if (m->m_len < sizeof(*ip)) {
- m = m_pullup(m, sizeof(*ip));
- if (m == NULL) {
- error = ENOMEM;
+ hlen = off + sizeof(*ip);
+ if (m->m_len < hlen) {
+ m = m_pullup(m, hlen);
+ if (m == NULL)
goto quit;
- }
}
- ip = mtod(m, struct ip *);
+ ip = (void *)(mtod(m, char *) + off);
iphlen = ip->ip_hl * 4;
KASSERT(ip->ip_v == IPVERSION);
KASSERT(iphlen >= sizeof(*ip));
KASSERT(ip->ip_p == IPPROTO_TCP);
ipid = ntohs(ip->ip_id);
- hlen = iphlen + sizeof(*th);
+ hlen = off + iphlen + sizeof(*th);
if (m->m_len < hlen) {
m = m_pullup(m, hlen);
- if (m == NULL) {
- error = ENOMEM;
+ if (m == NULL)
goto quit;
- }
}
- th = (void *)(mtod(m, char *) + iphlen);
+ th = (void *)(mtod(m, char *) + off + iphlen);
tcpseq = ntohl(th->th_seq);
thlen = th->th_off * 4;
- hlen = iphlen + thlen;
+ hlen = off + iphlen + thlen;
mss = m->m_pkthdr.segsz;
KASSERT(mss != 0);
KASSERT(len > hlen);
t = m_split(m, hlen, M_NOWAIT);
- if (t == NULL) {
- error = ENOMEM;
+ if (t == NULL)
goto quit;
- }
hdr = m;
m = t;
+
len -= hlen;
KASSERT(len % mss == 0);
- while (len > 0) {
- struct mbuf *n;
- n = m_dup(hdr, 0, hlen, M_NOWAIT);
- if (n == NULL) {
- error = ENOMEM;
- goto quit;
- }
+ ip = (void *)(mtod(hdr, char *) + off);
+ ip->ip_len = htons(iphlen + thlen + mss);
+ phsum = in_cksum_phdr(ip->ip_src.s_addr, ip->ip_dst.s_addr,
+ htons((uint16_t)(thlen + mss) + IPPROTO_TCP));
+
+ for (nsegs = len / mss; nsegs > 0; nsegs--) {
+ if (nsegs > 1) {
+ n = m_dup(hdr, 0, hlen, M_NOWAIT);
+ if (n == NULL)
+ goto quit;
+ } else
+ n = hdr;
KASSERT(n->m_len == hlen); /* XXX */
- t = m_split(m, mss, M_NOWAIT);
- if (t == NULL) {
- m_freem(n);
- error = ENOMEM;
- goto quit;
- }
+ if (nsegs > 1) {
+ t = m_split(m, mss, M_NOWAIT);
+ if (t == NULL) {
+ m_freem(n);
+ goto quit;
+ }
+ } else
+ t = m;
m_cat(n, m);
m = t;
KASSERT(n->m_len >= hlen); /* XXX */
+ if (m0 == NULL)
+ m0 = n;
+
+ if (prev != NULL)
+ prev->m_nextpkt = n;
+
n->m_pkthdr.len = hlen + mss;
- ip = mtod(n, struct ip *);
- KASSERT(ip->ip_v == IPVERSION);
- ip->ip_len = htons(n->m_pkthdr.len);
+ n->m_nextpkt = NULL; /* XXX */
+
+ ip = (void *)(mtod(n, char *) + off);
ip->ip_id = htons(ipid);
- th = (void *)(mtod(n, char *) + iphlen);
- th->th_seq = htonl(tcpseq);
ip->ip_sum = 0;
- ip->ip_sum = in_cksum(n, iphlen);
- th->th_sum = 0;
- th->th_sum = in4_cksum(n, IPPROTO_TCP, iphlen, thlen + mss);
+ ip->ip_sum = in4_cksum(n, 0, off, iphlen);
- error = ip_if_output(ifp, n, sa, rt);
- if (error) {
- goto quit;
- }
+ th = (void *)(mtod(n, char *) + off + iphlen);
+ th->th_seq = htonl(tcpseq);
+ th->th_sum = phsum;
+ th->th_sum = in4_cksum(n, 0, off + iphlen, thlen + mss);
tcpseq += mss;
ipid++;
- len -= mss;
+ prev = n;
}
+ return m0;
quit:
- if (hdr != NULL) {
+ if (hdr != NULL)
m_freem(hdr);
- }
- if (m != NULL) {
+ if (m != NULL)
+ m_freem(m);
+ for (m = m0; m != NULL; m = n) {
+ n = m->m_nextpkt;
m_freem(m);
}
- return error;
+ return NULL;
}
int
ip_tso_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *sa,
struct rtentry *rt)
{
- return tcp4_segment(ifp, m, sa, rt);
+ struct mbuf *n;
+ int error = 0;
+
+ m = tcp4_segment(m, 0);
+ if (m == NULL)
+ return ENOMEM;
+ do {
+ n = m->m_nextpkt;
+ if (error == 0)
+ error = ip_if_output(ifp, m, sa, rt);
+ else
+ m_freem(m);
+ m = n;
+ } while (m != NULL);
+ return error;
}
/*
Index: src/sys/netinet/in_offload.h
diff -u src/sys/netinet/in_offload.h:1.11 src/sys/netinet/in_offload.h:1.12
--- src/sys/netinet/in_offload.h:1.11 Wed Jul 11 06:00:34 2018
+++ src/sys/netinet/in_offload.h Wed Dec 12 01:40:20 2018
@@ -1,4 +1,4 @@
-/* $NetBSD: in_offload.h,v 1.11 2018/07/11 06:00:34 maxv Exp $ */
+/* $NetBSD: in_offload.h,v 1.12 2018/12/12 01:40:20 rin Exp $ */
/*
* Copyright (c)2005, 2006 YAMAMOTO Takashi,
@@ -32,6 +32,7 @@
/*
* Subroutines to do software-only equivalent of h/w offloading.
*/
+struct mbuf *tcp4_segment(struct mbuf *, int);
int ip_tso_output(struct ifnet *, struct mbuf *, const struct sockaddr *,
struct rtentry *);
void in_undefer_cksum(struct mbuf *, size_t, int);
Index: src/sys/netinet6/in6_offload.c
diff -u src/sys/netinet6/in6_offload.c:1.11 src/sys/netinet6/in6_offload.c:1.12
--- src/sys/netinet6/in6_offload.c:1.11 Wed Sep 19 07:54:11 2018
+++ src/sys/netinet6/in6_offload.c Wed Dec 12 01:40:20 2018
@@ -1,4 +1,4 @@
-/* $NetBSD: in6_offload.c,v 1.11 2018/09/19 07:54:11 rin Exp $ */
+/* $NetBSD: in6_offload.c,v 1.12 2018/12/12 01:40:20 rin Exp $ */
/*
* Copyright (c)2006 YAMAMOTO Takashi,
@@ -27,7 +27,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: in6_offload.c,v 1.11 2018/09/19 07:54:11 rin Exp $");
+__KERNEL_RCSID(0, "$NetBSD: in6_offload.c,v 1.12 2018/12/12 01:40:20 rin Exp $");
#include <sys/param.h>
#include <sys/mbuf.h>
@@ -45,11 +45,10 @@ __KERNEL_RCSID(0, "$NetBSD: in6_offload.
/*
* Handle M_CSUM_TSOv6 in software. Split the TCP payload in chunks of
- * size MSS, and send them.
+ * size MSS, and return mbuf chain consists of them.
*/
-static int
-tcp6_segment(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
- const struct sockaddr_in6 *dst, struct rtentry *rt)
+struct mbuf *
+tcp6_segment(struct mbuf *m, int off)
{
int mss;
int iphlen;
@@ -59,9 +58,12 @@ tcp6_segment(struct ifnet *ifp, struct i
struct ip6_hdr *iph;
struct tcphdr *th;
uint32_t tcpseq;
+ uint16_t phsum;
struct mbuf *hdr = NULL;
- struct mbuf *t;
- int error = 0;
+ struct mbuf *m0 = NULL;
+ struct mbuf *prev = NULL;
+ struct mbuf *n, *t;
+ int nsegs;
KASSERT((m->m_flags & M_PKTHDR) != 0);
KASSERT((m->m_pkthdr.csum_flags & M_CSUM_TSOv6) != 0);
@@ -69,101 +71,121 @@ tcp6_segment(struct ifnet *ifp, struct i
m->m_pkthdr.csum_flags = 0;
len = m->m_pkthdr.len;
- KASSERT(len >= sizeof(*iph) + sizeof(*th));
+ KASSERT(len >= off + sizeof(*iph) + sizeof(*th));
- if (m->m_len < sizeof(*iph)) {
- m = m_pullup(m, sizeof(*iph));
- if (m == NULL) {
- error = ENOMEM;
+ hlen = off + sizeof(*iph);
+ if (m->m_len < hlen) {
+ m = m_pullup(m, hlen);
+ if (m == NULL)
goto quit;
- }
}
- iph = mtod(m, struct ip6_hdr *);
+ iph = (void *)(mtod(m, char *) + off);
iphlen = sizeof(*iph);
KASSERT((iph->ip6_vfc & IPV6_VERSION_MASK) == IPV6_VERSION);
KASSERT(iph->ip6_nxt == IPPROTO_TCP);
- hlen = iphlen + sizeof(*th);
+ hlen = off + iphlen + sizeof(*th);
if (m->m_len < hlen) {
m = m_pullup(m, hlen);
- if (m == NULL) {
- error = ENOMEM;
+ if (m == NULL)
goto quit;
- }
}
- th = (void *)(mtod(m, char *) + iphlen);
+ th = (void *)(mtod(m, char *) + off + iphlen);
tcpseq = ntohl(th->th_seq);
thlen = th->th_off * 4;
- hlen = iphlen + thlen;
+ hlen = off + iphlen + thlen;
mss = m->m_pkthdr.segsz;
KASSERT(mss != 0);
KASSERT(len > hlen);
t = m_split(m, hlen, M_NOWAIT);
- if (t == NULL) {
- error = ENOMEM;
+ if (t == NULL)
goto quit;
- }
hdr = m;
m = t;
+
len -= hlen;
KASSERT(len % mss == 0);
- while (len > 0) {
- struct mbuf *n;
- n = m_dup(hdr, 0, hlen, M_NOWAIT);
- if (n == NULL) {
- error = ENOMEM;
- goto quit;
- }
+ iph = (void *)(mtod(hdr, char *) + off);
+ iph->ip6_plen = htons(thlen + mss);
+ phsum = in6_cksum_phdr(&iph->ip6_src, &iph->ip6_dst, htonl(thlen + mss),
+ htonl(IPPROTO_TCP));
+
+ for (nsegs = len / mss; nsegs > 0; nsegs--) {
+ if (nsegs > 1) {
+ n = m_dup(hdr, 0, hlen, M_NOWAIT);
+ if (n == NULL)
+ goto quit;
+ } else
+ n = hdr;
KASSERT(n->m_len == hlen); /* XXX */
- t = m_split(m, mss, M_NOWAIT);
- if (t == NULL) {
- m_freem(n);
- error = ENOMEM;
- goto quit;
- }
+ if (nsegs > 1) {
+ t = m_split(m, mss, M_NOWAIT);
+ if (t == NULL) {
+ m_freem(n);
+ goto quit;
+ }
+ } else
+ t = m;
m_cat(n, m);
m = t;
KASSERT(n->m_len >= hlen); /* XXX */
+ if (m0 == NULL)
+ m0 = n;
+
+ if (prev != NULL)
+ prev->m_nextpkt = n;
+
n->m_pkthdr.len = hlen + mss;
- iph = mtod(n, struct ip6_hdr *);
- KASSERT((iph->ip6_vfc & IPV6_VERSION_MASK) == IPV6_VERSION);
- iph->ip6_plen = htons(thlen + mss);
- th = (void *)(mtod(n, char *) + iphlen);
- th->th_seq = htonl(tcpseq);
- th->th_sum = 0;
- th->th_sum = in6_cksum(n, IPPROTO_TCP, iphlen, thlen + mss);
+ n->m_nextpkt = NULL; /* XXX */
- error = ip6_if_output(ifp, origifp, n, dst, rt);
- if (error) {
- goto quit;
- }
+ th = (void *)(mtod(n, char *) + off + iphlen);
+ th->th_seq = htonl(tcpseq);
+ th->th_sum = phsum;
+ th->th_sum = in6_cksum(n, 0, off + iphlen, thlen + mss);
tcpseq += mss;
- len -= mss;
+ prev = n;
}
+ return m0;
quit:
- if (hdr != NULL) {
+ if (hdr != NULL)
m_freem(hdr);
- }
- if (m != NULL) {
+ if (m != NULL)
+ m_freem(m);
+ for (m = m0; m != NULL; m = n) {
+ n = m->m_nextpkt;
m_freem(m);
}
- return error;
+ return NULL;
}
int
ip6_tso_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
const struct sockaddr_in6 *dst, struct rtentry *rt)
{
- return tcp6_segment(ifp, origifp, m, dst, rt);
+ struct mbuf *n;
+ int error = 0;
+
+ m = tcp6_segment(m, 0);
+ if (m == NULL)
+ return ENOMEM;
+ do {
+ n = m->m_nextpkt;
+ if (error == 0)
+ error = ip6_if_output(ifp, origifp, m, dst, rt);
+ else
+ m_freem(m);
+ m = n;
+ } while (m != NULL);
+ return error;
}
/*
Index: src/sys/netinet6/in6_offload.h
diff -u src/sys/netinet6/in6_offload.h:1.9 src/sys/netinet6/in6_offload.h:1.10
--- src/sys/netinet6/in6_offload.h:1.9 Fri Aug 10 06:55:04 2018
+++ src/sys/netinet6/in6_offload.h Wed Dec 12 01:40:20 2018
@@ -1,4 +1,4 @@
-/* $NetBSD: in6_offload.h,v 1.9 2018/08/10 06:55:04 maxv Exp $ */
+/* $NetBSD: in6_offload.h,v 1.10 2018/12/12 01:40:20 rin Exp $ */
/*
* Copyright (c)2005, 2006 YAMAMOTO Takashi,
@@ -32,6 +32,7 @@
/*
* Subroutines to do software-only equivalent of h/w offloading.
*/
+struct mbuf *tcp6_segment(struct mbuf *, int);
int ip6_tso_output(struct ifnet *, struct ifnet *, struct mbuf *,
const struct sockaddr_in6 *, struct rtentry *);
void in6_undefer_cksum(struct mbuf *, size_t, int);
Index: src/sys/rump/net/lib/libnet/Makefile
diff -u src/sys/rump/net/lib/libnet/Makefile:1.30 src/sys/rump/net/lib/libnet/Makefile:1.31
--- src/sys/rump/net/lib/libnet/Makefile:1.30 Tue Jan 17 08:10:37 2017
+++ src/sys/rump/net/lib/libnet/Makefile Wed Dec 12 01:40:20 2018
@@ -1,4 +1,4 @@
-# $NetBSD: Makefile,v 1.30 2017/01/17 08:10:37 ozaki-r Exp $
+# $NetBSD: Makefile,v 1.31 2018/12/12 01:40:20 rin Exp $
#
@@ -15,6 +15,7 @@ SRCS= if.c if_loop.c route.c rtsock.c ra
SRCS+= if_43.c
SRCS+= if_llatbl.c
SRCS+= net_component.c
+SRCS+= ether_sw_offload.c
.include <bsd.init.mk>
Added files:
Index: src/sys/net/ether_sw_offload.c
diff -u /dev/null src/sys/net/ether_sw_offload.c:1.1
--- /dev/null Wed Dec 12 01:40:21 2018
+++ src/sys/net/ether_sw_offload.c Wed Dec 12 01:40:20 2018
@@ -0,0 +1,312 @@
+/* $NetBSD: ether_sw_offload.c,v 1.1 2018/12/12 01:40:20 rin Exp $ */
+
+/*
+ * Copyright (c) 2018 The NetBSD Foundation, Inc.
+ * All rights reserved.
+ *
+ * This code is derived from software contributed to The NetBSD Foundation
+ * by Rin Okuyama.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ * notice, this list of conditions and the following disclaimer in the
+ * documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
+ * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+ * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <sys/cdefs.h>
+__KERNEL_RCSID(0, "$NetBSD: ether_sw_offload.c,v 1.1 2018/12/12 01:40:20 rin Exp $");
+
+#include <sys/param.h>
+#include <sys/types.h>
+#include <sys/mbuf.h>
+
+#include <net/if.h>
+#include <net/if_ether.h>
+#include <net/ether_sw_offload.h>
+
+#include <netinet/in.h>
+#include <netinet/in_offload.h>
+#include <netinet/ip.h>
+#include <netinet/ip6.h>
+#include <netinet/tcp.h>
+#include <netinet/udp.h>
+
+#include <netinet6/in6.h>
+#include <netinet6/in6_offload.h>
+
+/*
+ * Handle TX offload in software. For TSO, split the packet into
+ * chanks with payloads of size MSS. For chekcsum offload, update
+ * required checksum fields. The results are more than one packet
+ * in general. Return a mbuf chain consists of them.
+ */
+
+struct mbuf *
+ether_sw_offload_tx(struct ifnet *ifp, struct mbuf *m)
+{
+ struct ether_header *ep;
+ int flags, ehlen;
+ bool v4;
+
+ KASSERT(m->m_flags & M_PKTHDR);
+ flags = m->m_pkthdr.csum_flags;
+ if (flags == 0)
+ goto done;
+
+ /* Sanity check */
+ if (!TX_OFFLOAD_SUPPORTED(ifp->if_csum_flags_tx, flags))
+ goto quit;
+
+ KASSERT(m->m_pkthdr.len >= sizeof(*ep));
+ if (m->m_len < sizeof(*ep)) {
+ m = m_pullup(m, sizeof(*ep));
+ if (m == NULL)
+ return NULL;
+ }
+ ep = mtod(m, struct ether_header *);
+ switch (ntohs(ep->ether_type)) {
+ case ETHERTYPE_IP:
+ case ETHERTYPE_IPV6:
+ ehlen = ETHER_HDR_LEN;
+ break;
+ case ETHERTYPE_VLAN:
+ ehlen = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
+ break;
+ default:
+ goto quit;
+ }
+ KASSERT(m->m_pkthdr.len >= ehlen);
+
+ v4 = flags & (M_CSUM_TSOv4 | M_CSUM_IPv4 | M_CSUM_TCPv4 | M_CSUM_UDPv4);
+
+ if (flags & (M_CSUM_TSOv4 | M_CSUM_TSOv6)) {
+ /*
+ * tcp[46]_segment() assume that size of payloads is
+ * a multiple of MSS. Further, tcp6_segment() assumes
+ * no extention headers.
+ *
+ * XXX Do we need some KASSERT's?
+ */
+ if (v4)
+ return tcp4_segment(m, ehlen);
+ else
+ return tcp6_segment(m, ehlen);
+ }
+
+ if (v4)
+ in_undefer_cksum(m, ehlen, flags);
+ else
+ in6_undefer_cksum(m, ehlen, flags);
+done:
+ m->m_pkthdr.csum_flags = 0;
+ m->m_nextpkt = NULL;
+ return m;
+quit:
+ m_freem(m);
+ return NULL;
+}
+
+/*
+ * Handle RX offload in software.
+ *
+ * XXX Fragmented packets or packets with IPv6 extension headers
+ * are not currently supported.
+ */
+
+struct mbuf *
+ether_sw_offload_rx(struct ifnet *ifp, struct mbuf *m)
+{
+ struct ether_header *eh;
+ struct ip *ip;
+ struct ip6_hdr *ip6;
+ struct tcphdr *th;
+ struct udphdr *uh;
+ uint16_t sum, osum;
+ uint8_t proto;
+ int flags, enabled, len, ehlen, iphlen, l4offset;
+ bool v4;
+
+ flags = 0;
+
+ enabled = ifp->if_csum_flags_rx;
+ if (!(enabled & (M_CSUM_IPv4 | M_CSUM_TCPv4 | M_CSUM_UDPv4 |
+ M_CSUM_TCPv6 | M_CSUM_UDPv6)))
+ goto done;
+
+ KASSERT(m->m_flags & M_PKTHDR);
+ len = m->m_pkthdr.len;
+
+ KASSERT(len >= sizeof(*eh));
+ if (m->m_len < sizeof(*eh)) {
+ m = m_pullup(m, sizeof(*eh));
+ if (m == NULL)
+ return NULL;
+ }
+ eh = mtod(m, struct ether_header *);
+ switch (htons(eh->ether_type)) {
+ case ETHERTYPE_IP:
+ case ETHERTYPE_IPV6:
+ ehlen = ETHER_HDR_LEN;
+ break;
+ case ETHERTYPE_VLAN:
+ ehlen = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
+ break;
+ default:
+ goto done;
+ }
+
+ KASSERT(len >= ehlen);
+ len = m->m_pkthdr.len - ehlen;
+
+ KASSERT(len >= sizeof(*ip));
+ if (m->m_len < ehlen + sizeof(*ip)) {
+ m = m_pullup(m, ehlen + sizeof(*ip));
+ if (m == NULL)
+ return NULL;
+ }
+ ip = (void *)(mtod(m, char *) + ehlen);
+ v4 = (ip->ip_v == IPVERSION);
+
+ if (v4) {
+ if (enabled & M_CSUM_IPv4)
+ flags |= M_CSUM_IPv4;
+
+ iphlen = ip->ip_hl << 2;
+ KASSERT(iphlen >= sizeof(*ip));
+
+ len -= iphlen;
+ KASSERT(len >= 0);
+
+ if (in4_cksum(m, 0, ehlen, iphlen) != 0) {
+ if (enabled & M_CSUM_IPv4)
+ flags |= M_CSUM_IPv4_BAD;
+ /* Broken. Do not check further. */
+ goto done;
+ }
+
+ /* Check if fragmented. */
+ if (ntohs(ip->ip_off) & ~(IP_DF | IP_RF))
+ goto done;
+
+ proto = ip->ip_p;
+ switch (proto) {
+ case IPPROTO_TCP:
+ if (!(enabled & M_CSUM_TCPv4))
+ goto done;
+ break;
+ case IPPROTO_UDP:
+ if (!(enabled & M_CSUM_UDPv4))
+ goto done;
+ break;
+ default:
+ goto done;
+ }
+
+ sum = in_cksum_phdr(ip->ip_src.s_addr, ip->ip_dst.s_addr,
+ htons((uint16_t)len + proto));
+ } else {
+ KASSERT(len >= sizeof(*ip6));
+ if (m->m_len < ehlen + sizeof(*ip6)) {
+ m = m_pullup(m, ehlen + sizeof(*ip6));
+ if (m == NULL)
+ return NULL;
+ }
+ ip6 = (void *)(mtod(m, char *) + ehlen);
+ KASSERT((ip6->ip6_vfc & IPV6_VERSION_MASK) == IPV6_VERSION);
+
+ iphlen = sizeof(*ip6);
+
+ len -= iphlen;
+
+ proto = ip6->ip6_nxt;
+ switch (proto) {
+ case IPPROTO_TCP:
+ if (!(enabled & M_CSUM_TCPv6))
+ goto done;
+ break;
+ case IPPROTO_UDP:
+ if (!(enabled & M_CSUM_UDPv6))
+ goto done;
+ break;
+ default:
+ /* XXX Extension headers are not supported. */
+ goto done;
+ }
+
+ sum = in6_cksum_phdr(&ip6->ip6_src, &ip6->ip6_dst, htonl(len),
+ htonl(proto));
+ }
+
+ l4offset = ehlen + iphlen;
+ switch (proto) {
+ case IPPROTO_TCP:
+ KASSERT(len >= sizeof(*th));
+ if (m->m_len < l4offset + sizeof(*th)) {
+ m = m_pullup(m, l4offset + sizeof(*th));
+ if (m == NULL)
+ return NULL;
+ }
+ th = (void *)(mtod(m, char *) + l4offset);
+ osum = th->th_sum;
+ th->th_sum = sum;
+ if (v4) {
+ flags |= M_CSUM_TCPv4;
+ sum = in4_cksum(m, 0, l4offset, len);
+ } else {
+ flags |= M_CSUM_TCPv6;
+ sum = in6_cksum(m, 0, l4offset, len);
+ }
+ if (sum != osum)
+ flags |= M_CSUM_TCP_UDP_BAD;
+ th->th_sum = osum;
+ break;
+ case IPPROTO_UDP:
+ KASSERT(len >= sizeof(*uh));
+ if (m->m_len < l4offset + sizeof(*uh)) {
+ m = m_pullup(m, l4offset + sizeof(*uh));
+ if (m == NULL)
+ return NULL;
+ }
+ uh = (void *)(mtod(m, char *) + l4offset);
+ osum = uh->uh_sum;
+ if (osum == 0)
+ break;
+ uh->uh_sum = sum;
+ if (v4) {
+ flags |= M_CSUM_UDPv4;
+ sum = in4_cksum(m, 0, l4offset, len);
+ } else {
+ flags |= M_CSUM_UDPv6;
+ sum = in6_cksum(m, 0, l4offset, len);
+ }
+ if (sum == 0)
+ sum = 0xffff;
+ if (sum != osum)
+ flags |= M_CSUM_TCP_UDP_BAD;
+ uh->uh_sum = osum;
+ break;
+ default:
+ panic("%s: impossible", __func__);
+ break;
+ }
+
+done:
+ m->m_pkthdr.csum_flags = flags;
+ return m;
+}
Index: src/sys/net/ether_sw_offload.h
diff -u /dev/null src/sys/net/ether_sw_offload.h:1.1
--- /dev/null Wed Dec 12 01:40:21 2018
+++ src/sys/net/ether_sw_offload.h Wed Dec 12 01:40:20 2018
@@ -0,0 +1,55 @@
+/* $NetBSD: ether_sw_offload.h,v 1.1 2018/12/12 01:40:20 rin Exp $ */
+
+/*
+ * Copyright (c) 2018 The NetBSD Foundation, Inc.
+ * All rights reserved.
+ *
+ * This code is derived from software contributed to The NetBSD Foundation
+ * by Rin Okuyama.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ * notice, this list of conditions and the following disclaimer in the
+ * documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
+ * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+ * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef _NET_ETHER_SW_OFFLOAD_H_
+#define _NET_ETHER_SW_OFFLOAD_H_
+
+#ifdef _KERNEL
+#include <sys/mbuf.h>
+#include <net/if.h>
+
+/*
+ * Whether given TX offload options are supported by the interface.
+ * We need to check TSO bits first; M_CSUM_IPv4 is turned on even if
+ * TSOv4 is enabled (and some drivers depend on this behavior).
+ * Therefore we cannot test
+ * m->m_pkthdr.csum_flags & ~ifp->if_csum_flags_tx
+ * for interfaces that do not support hw IPv4 checksum.
+ */
+#define TX_OFFLOAD_SUPPORTED(if_flags, mbuf_flags) \
+ ((((mbuf_flags) & (M_CSUM_TSOv4 | M_CSUM_TSOv6)) & (if_flags)) || \
+ !((mbuf_flags) & ~(if_flags)))
+
+struct mbuf *ether_sw_offload_tx(struct ifnet *, struct mbuf *);
+struct mbuf *ether_sw_offload_rx(struct ifnet *, struct mbuf *);
+#endif /* _KERNEL */
+
+#endif /* !_NET_ETHER_SW_OFFLOAD_H_ */
