Merge branch 'ping6-SOL_IPV6'

Jakub Kicinski says:

====================
net: ping6: support setting basic SOL_IPV6 options via cmsg

Support for IPV6_HOPLIMIT, IPV6_TCLASS, IPV6_DONTFRAG on ICMPv6
sockets and associated tests. I have no immediate plans to
implement IPV6_FLOWINFO and all the extension header stuff.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

net/ipv6/ip6_output.c

@@ -1487,6 +1487,7 @@ static int __ip6_append_data(struct sock *sk,
 
 	if (cork->length + length > mtu - headersize && ipc6->dontfrag &&
 	    (sk->sk_protocol == IPPROTO_UDP ||
+	     sk->sk_protocol == IPPROTO_ICMPV6 ||
 	     sk->sk_protocol == IPPROTO_RAW)) {
 		ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
 				sizeof(struct ipv6hdr));

net/ipv6/ping.c

@@ -101,11 +101,21 @@ static int ping_v6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
 	ipc6.sockc.tsflags = sk->sk_tsflags;
 	ipc6.sockc.mark = sk->sk_mark;
 
-	err = sock_cmsg_send(sk, msg, &ipc6.sockc);
-	if (err)
-		return err;
+	if (msg->msg_controllen) {
+		struct ipv6_txoptions opt = {};
+
+		opt.tot_len = sizeof(opt);
+		ipc6.opt = &opt;
 
-	/* TODO: use ip6_datagram_send_ctl to get options from cmsg */
+		err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, &ipc6);
+		if (err < 0)
+			return err;
+
+		/* Changes to txoptions and flow info are not implemented, yet.
+		 * Drop the options, fl6 is wiped below.
+		 */
+		ipc6.opt = NULL;
+	}
 
 	memset(&fl6, 0, sizeof(fl6));
 
@@ -140,7 +150,8 @@ static int ping_v6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
 	pfh.wcheck = 0;
 	pfh.family = AF_INET6;
 
-	ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
+	if (ipc6.hlimit < 0)
+		ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
 
 	lock_sock(sk);
 	err = ip6_append_data(sk, ping_getfrag, &pfh, len,

tools/testing/selftests/net/cmsg_ipv6.sh

+#!/bin/bash
+# SPDX-License-Identifier: GPL-2.0
+
+ksft_skip=4
+
+NS=ns
+IP6=2001:db8:1::1/64
+TGT6=2001:db8:1::2
+TMPF=`mktemp`
+
+cleanup()
+{
+    rm -f $TMPF
+    ip netns del $NS
+}
+
+trap cleanup EXIT
+
+NSEXE="ip netns exec $NS"
+
+tcpdump -h | grep immediate-mode >> /dev/null
+if [ $? -ne 0 ]; then
+    echo "SKIP - tcpdump with --immediate-mode option required"
+    exit $ksft_skip
+fi
+
+# Namespaces
+ip netns add $NS
+
+$NSEXE sysctl -w net.ipv4.ping_group_range='0 2147483647' > /dev/null
+
+# Connectivity
+ip -netns $NS link add type dummy
+ip -netns $NS link set dev dummy0 up
+ip -netns $NS addr add $IP6 dev dummy0
+
+# Test
+BAD=0
+TOTAL=0
+
+check_result() {
+    ((TOTAL++))
+    if [ $1 -ne $2 ]; then
+	echo "  Case $3 returned $1, expected $2"
+	((BAD++))
+    fi
+}
+
+# IPV6_DONTFRAG
+for ovr in setsock cmsg both diff; do
+    for df in 0 1; do
+	for p in u i r; do
+	    [ $p == "u" ] && prot=UDP
+	    [ $p == "i" ] && prot=ICMP
+	    [ $p == "r" ] && prot=RAW
+
+	    [ $ovr == "setsock" ] && m="-F $df"
+	    [ $ovr == "cmsg" ]    && m="-f $df"
+	    [ $ovr == "both" ]    && m="-F $df -f $df"
+	    [ $ovr == "diff" ]    && m="-F $((1 - df)) -f $df"
+
+	    $NSEXE ./cmsg_sender -s -S 2000 -6 -p $p $m $TGT6 1234
+	    check_result $? $df "DONTFRAG $prot $ovr"
+	done
+    done
+done
+
+# IPV6_TCLASS
+TOS=0x10
+TOS2=0x20
+
+ip -6 -netns $NS rule add tos $TOS lookup 300
+ip -6 -netns $NS route add table 300 prohibit any
+
+for ovr in setsock cmsg both diff; do
+    for p in u i r; do
+	[ $p == "u" ] && prot=UDP
+	[ $p == "i" ] && prot=ICMP
+	[ $p == "r" ] && prot=RAW
+
+	[ $ovr == "setsock" ] && m="-C"
+	[ $ovr == "cmsg" ]    && m="-c"
+	[ $ovr == "both" ]    && m="-C $((TOS2)) -c"
+	[ $ovr == "diff" ]    && m="-C $((TOS )) -c"
+
+	$NSEXE nohup tcpdump --immediate-mode -p -ni dummy0 -w $TMPF -c 4 2> /dev/null &
+	BG=$!
+	sleep 0.05
+
+	$NSEXE ./cmsg_sender -6 -p $p $m $((TOS2)) $TGT6 1234
+	check_result $? 0 "TCLASS $prot $ovr - pass"
+
+	while [ -d /proc/$BG ]; do
+	    $NSEXE ./cmsg_sender -6 -p u $TGT6 1234
+	done
+
+	tcpdump -r $TMPF -v 2>&1 | grep "class $TOS2" >> /dev/null
+	check_result $? 0 "TCLASS $prot $ovr - packet data"
+	rm $TMPF
+
+	[ $ovr == "both" ]    && m="-C $((TOS )) -c"
+	[ $ovr == "diff" ]    && m="-C $((TOS2)) -c"
+
+	$NSEXE ./cmsg_sender -6 -p $p $m $((TOS)) -s $TGT6 1234
+	check_result $? 1 "TCLASS $prot $ovr - rejection"
+    done
+done
+
+# IPV6_HOPLIMIT
+LIM=4
+
+for ovr in setsock cmsg both diff; do
+    for p in u i r; do
+	[ $p == "u" ] && prot=UDP
+	[ $p == "i" ] && prot=ICMP
+	[ $p == "r" ] && prot=RAW
+
+	[ $ovr == "setsock" ] && m="-L"
+	[ $ovr == "cmsg" ]    && m="-l"
+	[ $ovr == "both" ]    && m="-L $LIM -l"
+	[ $ovr == "diff" ]    && m="-L $((LIM + 1)) -l"
+
+	$NSEXE nohup tcpdump --immediate-mode -p -ni dummy0 -w $TMPF -c 4 2> /dev/null &
+	BG=$!
+	sleep 0.05
+
+	$NSEXE ./cmsg_sender -6 -p $p $m $LIM $TGT6 1234
+	check_result $? 0 "HOPLIMIT $prot $ovr - pass"
+
+	while [ -d /proc/$BG ]; do
+	    $NSEXE ./cmsg_sender -6 -p u $TGT6 1234
+	done
+
+	tcpdump -r $TMPF -v 2>&1 | grep "hlim $LIM[^0-9]" >> /dev/null
+	check_result $? 0 "HOPLIMIT $prot $ovr - packet data"
+	rm $TMPF
+    done
+done
+
+# IPV6 exthdr
+for p in u i r; do
+    # Very basic "does it crash" test
+    for h in h d r; do
+	$NSEXE ./cmsg_sender -p $p -6 -H $h $TGT6 1234
+	check_result $? 0 "ExtHdr $prot $ovr - pass"
+    done
+done
+
+# Summary
+if [ $BAD -ne 0 ]; then
+    echo "FAIL - $BAD/$TOTAL cases failed"
+    exit 1
+else
+    echo "OK"
+    exit 0
+fi

tools/testing/selftests/net/cmsg_sender.c

@@ -33,22 +33,28 @@ enum {
 	ERN_CMSG_RCV,
 };
 
+struct option_cmsg_u32 {
+	bool ena;
+	unsigned int val;
+};
+
 struct options {
 	bool silent_send;
 	const char *host;
 	const char *service;
+	unsigned int size;
 	struct {
 		unsigned int mark;
+		unsigned int dontfrag;
+		unsigned int tclass;
+		unsigned int hlimit;
 	} sockopt;
 	struct {
 		unsigned int family;
 		unsigned int type;
 		unsigned int proto;
 	} sock;
-	struct {
-		bool ena;
-		unsigned int val;
-	} mark;
+	struct option_cmsg_u32 mark;
 	struct {
 		bool ena;
 		unsigned int delay;
@@ -56,7 +62,14 @@ struct options {
 	struct {
 		bool ena;
 	} ts;
+	struct {
+		struct option_cmsg_u32 dontfrag;
+		struct option_cmsg_u32 tclass;
+		struct option_cmsg_u32 hlimit;
+		struct option_cmsg_u32 exthdr;
+	} v6;
 } opt = {
+	.size = 13,
 	.sock = {
 		.family	= AF_UNSPEC,
 		.type	= SOCK_DGRAM,
@@ -72,6 +85,7 @@ static void __attribute__((noreturn)) cs_usage(const char *bin)
 	printf("Usage: %s [opts] <dst host> <dst port / service>\n", bin);
 	printf("Options:\n"
 	       "\t\t-s      Silent send() failures\n"
+	       "\t\t-S      send() size\n"
 	       "\t\t-4/-6   Force IPv4 / IPv6 only\n"
 	       "\t\t-p prot Socket protocol\n"
 	       "\t\t        (u = UDP (default); i = ICMP; r = RAW)\n"
@@ -80,6 +94,14 @@ static void __attribute__((noreturn)) cs_usage(const char *bin)
 	       "\t\t-M val  Set SO_MARK via setsockopt\n"
 	       "\t\t-d val  Set SO_TXTIME with given delay (usec)\n"
 	       "\t\t-t      Enable time stamp reporting\n"
+	       "\t\t-f val  Set don't fragment via cmsg\n"
+	       "\t\t-F val  Set don't fragment via setsockopt\n"
+	       "\t\t-c val  Set TCLASS via cmsg\n"
+	       "\t\t-C val  Set TCLASS via setsockopt\n"
+	       "\t\t-l val  Set HOPLIMIT via cmsg\n"
+	       "\t\t-L val  Set HOPLIMIT via setsockopt\n"
+	       "\t\t-H type Add an IPv6 header option\n"
+	       "\t\t        (h = HOP; d = DST; r = RTDST)"
 	       "");
 	exit(ERN_HELP);
 }
@@ -88,11 +110,14 @@ static void cs_parse_args(int argc, char *argv[])
 {
 	char o;
 
-	while ((o = getopt(argc, argv, "46sp:m:M:d:t")) != -1) {
+	while ((o = getopt(argc, argv, "46sS:p:m:M:d:tf:F:c:C:l:L:H:")) != -1) {
 		switch (o) {
 		case 's':
 			opt.silent_send = true;
 			break;
+		case 'S':
+			opt.size = atoi(optarg);
+			break;
 		case '4':
 			opt.sock.family = AF_INET;
 			break;
@@ -126,6 +151,44 @@ static void cs_parse_args(int argc, char *argv[])
 		case 't':
 			opt.ts.ena = true;
 			break;
+		case 'f':
+			opt.v6.dontfrag.ena = true;
+			opt.v6.dontfrag.val = atoi(optarg);
+			break;
+		case 'F':
+			opt.sockopt.dontfrag = atoi(optarg);
+			break;
+		case 'c':
+			opt.v6.tclass.ena = true;
+			opt.v6.tclass.val = atoi(optarg);
+			break;
+		case 'C':
+			opt.sockopt.tclass = atoi(optarg);
+			break;
+		case 'l':
+			opt.v6.hlimit.ena = true;
+			opt.v6.hlimit.val = atoi(optarg);
+			break;
+		case 'L':
+			opt.sockopt.hlimit = atoi(optarg);
+			break;
+		case 'H':
+			opt.v6.exthdr.ena = true;
+			switch (optarg[0]) {
+			case 'h':
+				opt.v6.exthdr.val = IPV6_HOPOPTS;
+				break;
+			case 'd':
+				opt.v6.exthdr.val = IPV6_DSTOPTS;
+				break;
+			case 'r':
+				opt.v6.exthdr.val = IPV6_RTHDRDSTOPTS;
+				break;
+			default:
+				printf("Error: hdr type: %s\n", optarg);
+				break;
+			}
+			break;
 		}
 	}
 
@@ -136,6 +199,38 @@ static void cs_parse_args(int argc, char *argv[])
 	opt.service = argv[optind + 1];
 }
 
+static void memrnd(void *s, size_t n)
+{
+	int *dword = s;
+	char *byte;
+
+	for (; n >= 4; n -= 4)
+		*dword++ = rand();
+	byte = (void *)dword;
+	while (n--)
+		*byte++ = rand();
+}
+
+static void
+ca_write_cmsg_u32(char *cbuf, size_t cbuf_sz, size_t *cmsg_len,
+		  int level, int optname, struct option_cmsg_u32 *uopt)
+{
+	struct cmsghdr *cmsg;
+
+	if (!uopt->ena)
+		return;
+
+	cmsg = (struct cmsghdr *)(cbuf + *cmsg_len);
+	*cmsg_len += CMSG_SPACE(sizeof(__u32));
+	if (cbuf_sz < *cmsg_len)
+		error(ERN_CMSG_WR, EFAULT, "cmsg buffer too small");
+
+	cmsg->cmsg_level = level;
+	cmsg->cmsg_type = optname;
+	cmsg->cmsg_len = CMSG_LEN(sizeof(__u32));
+	*(__u32 *)CMSG_DATA(cmsg) = uopt->val;
+}
+
 static void
 cs_write_cmsg(int fd, struct msghdr *msg, char *cbuf, size_t cbuf_sz)
 {
@@ -145,17 +240,15 @@ cs_write_cmsg(int fd, struct msghdr *msg, char *cbuf, size_t cbuf_sz)
 	msg->msg_control = cbuf;
 	cmsg_len = 0;
 
-	if (opt.mark.ena) {
-		cmsg = (struct cmsghdr *)(cbuf + cmsg_len);
-		cmsg_len += CMSG_SPACE(sizeof(__u32));
-		if (cbuf_sz < cmsg_len)
-			error(ERN_CMSG_WR, EFAULT, "cmsg buffer too small");
+	ca_write_cmsg_u32(cbuf, cbuf_sz, &cmsg_len,
+			  SOL_SOCKET, SO_MARK, &opt.mark);
+	ca_write_cmsg_u32(cbuf, cbuf_sz, &cmsg_len,
+			  SOL_IPV6, IPV6_DONTFRAG, &opt.v6.dontfrag);
+	ca_write_cmsg_u32(cbuf, cbuf_sz, &cmsg_len,
+			  SOL_IPV6, IPV6_TCLASS, &opt.v6.tclass);
+	ca_write_cmsg_u32(cbuf, cbuf_sz, &cmsg_len,
+			  SOL_IPV6, IPV6_HOPLIMIT, &opt.v6.hlimit);
 
-		cmsg->cmsg_level = SOL_SOCKET;
-		cmsg->cmsg_type = SO_MARK;
-		cmsg->cmsg_len = CMSG_LEN(sizeof(__u32));
-		*(__u32 *)CMSG_DATA(cmsg) = opt.mark.val;
-	}
 	if (opt.txtime.ena) {
 		struct sock_txtime so_txtime = {
 			.clockid = CLOCK_MONOTONIC,
@@ -199,6 +292,17 @@ cs_write_cmsg(int fd, struct msghdr *msg, char *cbuf, size_t cbuf_sz)
 		*(__u32 *)CMSG_DATA(cmsg) = SOF_TIMESTAMPING_TX_SCHED |
 					    SOF_TIMESTAMPING_TX_SOFTWARE;
 	}
+	if (opt.v6.exthdr.ena) {
+		cmsg = (struct cmsghdr *)(cbuf + cmsg_len);
+		cmsg_len += CMSG_SPACE(8);
+		if (cbuf_sz < cmsg_len)
+			error(ERN_CMSG_WR, EFAULT, "cmsg buffer too small");
+
+		cmsg->cmsg_level = SOL_IPV6;
+		cmsg->cmsg_type = opt.v6.exthdr.val;
+		cmsg->cmsg_len = CMSG_LEN(8);
+		*(__u64 *)CMSG_DATA(cmsg) = 0;
+	}
 
 	if (cmsg_len)
 		msg->msg_controllen = cmsg_len;
@@ -286,18 +390,41 @@ cs_read_cmsg(int fd, struct msghdr *msg, char *cbuf, size_t cbuf_sz)
 	}
 }
 
+static void ca_set_sockopts(int fd)
+{
+	if (opt.sockopt.mark &&
+	    setsockopt(fd, SOL_SOCKET, SO_MARK,
+		       &opt.sockopt.mark, sizeof(opt.sockopt.mark)))
+		error(ERN_SOCKOPT, errno, "setsockopt SO_MARK");
+	if (opt.sockopt.dontfrag &&
+	    setsockopt(fd, SOL_IPV6, IPV6_DONTFRAG,
+		       &opt.sockopt.dontfrag, sizeof(opt.sockopt.dontfrag)))
+		error(ERN_SOCKOPT, errno, "setsockopt IPV6_DONTFRAG");
+	if (opt.sockopt.tclass &&
+	    setsockopt(fd, SOL_IPV6, IPV6_TCLASS,
+		       &opt.sockopt.tclass, sizeof(opt.sockopt.tclass)))
+		error(ERN_SOCKOPT, errno, "setsockopt IPV6_TCLASS");
+	if (opt.sockopt.hlimit &&
+	    setsockopt(fd, SOL_IPV6, IPV6_UNICAST_HOPS,
+		       &opt.sockopt.hlimit, sizeof(opt.sockopt.hlimit)))
+		error(ERN_SOCKOPT, errno, "setsockopt IPV6_HOPLIMIT");
+}
+
 int main(int argc, char *argv[])
 {
-	char buf[] = "blablablabla";
 	struct addrinfo hints, *ai;
 	struct iovec iov[1];
 	struct msghdr msg;
 	char cbuf[1024];
+	char *buf;
 	int err;
 	int fd;
 
 	cs_parse_args(argc, argv);
 
+	buf = malloc(opt.size);
+	memrnd(buf, opt.size);
+
 	memset(&hints, 0, sizeof(hints));
 	hints.ai_family = opt.sock.family;
 
@@ -326,17 +453,14 @@ int main(int argc, char *argv[])
 		buf[0] = ICMPV6_ECHO_REQUEST;
 		buf[1] = 0;
 	} else if (opt.sock.type == SOCK_RAW) {
-		struct udphdr hdr = { 1, 2, htons(sizeof(buf)), 0 };
+		struct udphdr hdr = { 1, 2, htons(opt.size), 0 };
 		struct sockaddr_in6 *sin6 = (void *)ai->ai_addr;;
 
 		memcpy(buf, &hdr, sizeof(hdr));
 		sin6->sin6_port = htons(opt.sock.proto);
 	}
 
-	if (opt.sockopt.mark &&
-	    setsockopt(fd, SOL_SOCKET, SO_MARK,
-		       &opt.sockopt.mark, sizeof(opt.sockopt.mark)))
-		error(ERN_SOCKOPT, errno, "setsockopt SO_MARK");
+	ca_set_sockopts(fd);
 
 	if (clock_gettime(CLOCK_REALTIME, &time_start_real))
 		error(ERN_GETTIME, errno, "gettime REALTIME");
@@ -344,7 +468,7 @@ int main(int argc, char *argv[])
 		error(ERN_GETTIME, errno, "gettime MONOTONIC");
 
 	iov[0].iov_base = buf;
-	iov[0].iov_len = sizeof(buf);
+	iov[0].iov_len = opt.size;
 
 	memset(&msg, 0, sizeof(msg));
 	msg.msg_name = ai->ai_addr;
@@ -360,7 +484,7 @@ int main(int argc, char *argv[])
 			fprintf(stderr, "send failed: %s\n", strerror(errno));
 		err = ERN_SEND;
 		goto err_out;
-	} else if (err != sizeof(buf)) {
+	} else if (err != (int)opt.size) {
 		fprintf(stderr, "short send\n");
 		err = ERN_SEND_SHORT;
 		goto err_out;