DHCP Relay: add option -si to support using src intf ip in relay (son…

…ic-net#7052)

* add option si to support using src intf ip in relay

src/isc-dhcp/patch/0012-add-option-si-to-support-using-src-intf-ip-in-relay.patch

@@ -0,0 +1,192 @@
+From 16163f0693e30588216f0f280d5eba8bb53f7001 Mon Sep 17 00:00:00 2001
+From: Tianrong Zhang <trzhang@microsoft.com>
+Date: Fri, 12 Mar 2021 23:30:56 -0800
+Subject: [PATCH] add option -si to support using src intf ip in relay
+
+---
+ common/socket.c  | 119 ++++++++++++++++++++++++++++++++++++-----------
+ includes/dhcpd.h |   1 +
+ relay/dhcrelay.c |   8 ++++
+ 3 files changed, 100 insertions(+), 28 deletions(-)
+
+diff --git a/common/socket.c b/common/socket.c
+index 483eb9c..da9f501 100644
+--- a/common/socket.c
++++ b/common/socket.c
+@@ -83,6 +83,29 @@ static unsigned int global_v4_socket_references = 0;
+ static int global_v4_socket = -1;
+ #endif
+
++/*
++ * If set, uses "from" interface IP for packet Tx.
++ * If not set, kernel chooses appropriate src ip for tx pkts
++ */
++int use_src_intf_ip_for_tx;
++
++/*
++ * For both send_packet6() and receive_packet6() we need to allocate
++ * space for the cmsg header information.  We do this once and reuse
++ * the buffer.  We also need the control buf for send_packet() and
++ * receive_packet() when we use a single socket and IP_PKTINFO to
++ * send the packet out the correct interface.
++ */
++static void   *v4_control_buf = NULL;
++static size_t  v4_control_buf_len = 0;
++
++static void
++v4_allocate_cmsg_cbuf(void) {
++	v4_control_buf_len = CMSG_SPACE(sizeof(struct in_pktinfo));
++	v4_control_buf = dmalloc(v4_control_buf_len, MDL);
++	return;
++}
++
+ /*
+  * If we can't bind() to a specific interface, then we can only have
+  * a single socket. This variable insures that we don't try to listen
+@@ -712,37 +735,77 @@ ssize_t send_packet (interface, packet, raw, len, from, to, hto)
+ 	struct hardware *hto;
+ {
+ 	int result;
+-#ifdef IGNORE_HOSTUNREACH
+-	int retry = 0;
+-	do {
+-#endif
+-#if defined(IP_PKTINFO) && defined(IP_RECVPKTINFO) && defined(USE_V4_PKTINFO)
+-		struct in_pktinfo pktinfo;
+-
+-		if (interface->ifp != NULL) {
+-			memset(&pktinfo, 0, sizeof (pktinfo));
+-			pktinfo.ipi_ifindex = interface->ifp->ifr_index;
+-			if (setsockopt(interface->wfdesc, IPPROTO_IP,
+-				       IP_PKTINFO, (char *)&pktinfo,
+-				       sizeof(pktinfo)) < 0) 
+-				log_fatal("setsockopt: IP_PKTINFO: %m");
++	struct msghdr m;
++	struct iovec v;
++	struct sockaddr_in dst;
++	struct in_pktinfo *pktinfo;
++	struct cmsghdr *cmsg;
++	unsigned int ifindex;
++
++	/*
++	 * If necessary allocate space for the control message header.
++	 * The space is common between send and receive.
++	 */
++
++	if (v4_control_buf == NULL) {
++		v4_allocate_cmsg_cbuf();
++		if (v4_control_buf == NULL) {
++			log_error("send_packet: unable to allocate cmsg header");
++			return(ENOMEM);
+ 		}
+-#endif
+-		result = sendto (interface -> wfdesc, (char *)raw, len, 0,
+-				 (struct sockaddr *)to, sizeof *to);
+-#ifdef IGNORE_HOSTUNREACH
+-	} while (to -> sin_addr.s_addr == htonl (INADDR_BROADCAST) &&
+-		 result < 0 &&
+-		 (errno == EHOSTUNREACH ||
+-		  errno == ECONNREFUSED) &&
+-		 retry++ < 10);
+-#endif
++	}
++	memset(v4_control_buf, 0, v4_control_buf_len);
++
++	/*
++	 * Initialize our message header structure.
++	 */
++	memset(&m, 0, sizeof(m));
++
++	/*
++	 * Set the target address we're sending to.
++	 */
++	memcpy(&dst, to, sizeof(dst));
++	m.msg_name = &dst;
++	m.msg_namelen = sizeof(dst);
++	ifindex = if_nametoindex(interface->name);
++
++	/*
++	 * Set the data buffer we're sending. (Using this wacky
++	 * "scatter-gather" stuff... we only have a single chunk
++	 * of data to send, so we declare a single vector entry.)
++	 */
++	v.iov_base = (char *)raw;
++	v.iov_len = len;
++	m.msg_iov = &v;
++	m.msg_iovlen = 1;
++
++	/*
++	 * Setting the interface is a bit more involved.
++	 *
++	 * We have to create a "control message", and set that to
++	 * define the IP packet information. We let he kernel decide
++	 * the source IP address unless 'use_src_intf_ip_for_tx' is
++	 * set, in which case we use the IP address of the ingress
++	 * interface we received the request on as the source IP.
++	 */
++	m.msg_control = v4_control_buf;
++	m.msg_controllen = v4_control_buf_len;
++	cmsg = CMSG_FIRSTHDR(&m);
++	INSIST(cmsg != NULL);
++	cmsg->cmsg_level = IPPROTO_IP;
++	cmsg->cmsg_type = IP_PKTINFO;
++	cmsg->cmsg_len = CMSG_LEN(sizeof(*pktinfo));
++	pktinfo = (struct in_pktinfo *)CMSG_DATA(cmsg);
++	memset(pktinfo, 0, sizeof(*pktinfo));
++	pktinfo->ipi_ifindex = ifindex;
++	if (use_src_intf_ip_for_tx)
++		pktinfo->ipi_spec_dst = from;
++
++	result = sendmsg(interface->wfdesc, &m, 0);
+ 	if (result < 0) {
+-		log_error ("send_packet: %m");
+-		if (errno == ENETUNREACH)
+-			log_error ("send_packet: please consult README file%s",
+-				   " regarding broadcast address.");
++		log_error("send_packet: %m");
+ 	}
++
+ 	return result;
+ }
+
+diff --git a/includes/dhcpd.h b/includes/dhcpd.h
+index 36cd518..0c25582 100644
+--- a/includes/dhcpd.h
++++ b/includes/dhcpd.h
+@@ -2660,6 +2660,7 @@ ssize_t send_fallback6(struct interface_info *, struct packet *,
+ #endif
+
+ #ifdef USE_SOCKET_SEND
++extern int use_src_intf_ip_for_tx;
+ void if_reinitialize_send (struct interface_info *);
+ void if_register_send (struct interface_info *);
+ void if_deregister_send (struct interface_info *);
+diff --git a/relay/dhcrelay.c b/relay/dhcrelay.c
+index 221106a..c44a79d 100644
+--- a/relay/dhcrelay.c
++++ b/relay/dhcrelay.c
+@@ -97,6 +97,12 @@ struct downstream_intf_list {
+ isc_boolean_t use_if_id = ISC_FALSE;
+ #endif
+
++/*
++ * If not set, kernel chooses what the src ip is.
++ * If set, uses "from" interface IP for packet Tx.
++ */
++extern int use_src_intf_ip_for_tx = 0;
++
+ 	/* Maximum size of a packet with agent options added. */
+ int dhcp_max_agent_option_packet_length = DHCP_MTU_MIN;
+
+@@ -431,6 +437,8 @@ main(int argc, char **argv) {
+ #endif
+ 		} else if (!strcmp(argv[i], "-d")) {
+ 			/* no_daemon = 1; */
++		} else if (!strcmp(argv[i], "-si")) {
++			use_src_intf_ip_for_tx = 1;
+ 		} else if (!strcmp(argv[i], "-q")) {
+ 			quiet = 1;
+ 			quiet_interface_discovery = 1;
+-- 
+2.17.1
+

src/isc-dhcp/patch/series

@@ -10,3 +10,4 @@
 0009-Support-for-dual-tor-scenario.patch
 0010-Bugfix-correctly-set-interface-netmask.patch
 0011-dhcp-relay-Prevent-Buffer-Overrun.patch
+0012-add-option-si-to-support-using-src-intf-ip-in-relay.patch