From f4cd9dec86fa8b3c2cef1902f0cb88011411f47e Mon Sep 17 00:00:00 2001
From: Vincent Bernat <vincent@bernat.ch>
Date: Sun, 9 Jun 2019 08:13:06 +0200
Subject: [PATCH] netlink: handle blocking read from netlink socket

It seems it is possible to run into a condition where the netlink
socket is not available for read. Set the MSG_DONTWAIT flag and fetch
an error if there is any.

Fix #333
---
 NEWS                          |  1 +
 src/daemon/event.c            | 26 ++++++++++++++++++++++++++
 src/daemon/interfaces-linux.c | 25 +------------------------
 src/daemon/lldpd.h            |  3 +++
 src/daemon/netlink.c          | 14 ++++++++------
 5 files changed, 39 insertions(+), 30 deletions(-)

diff --git a/NEWS b/NEWS
index 608188e8..2fb52626 100644
--- a/NEWS
+++ b/NEWS
@@ -6,6 +6,7 @@ lldpd (1.0.4)
   * Fix:
     + Better compliance for statsTLVsUnrecognizedTotal and
       statsAgeoutsTotal counters.
+    + On Linux, handle rare blocking case in Netlink code.
 
 lldpd (1.0.3)
   * Fix:
diff --git a/src/daemon/event.c b/src/daemon/event.c
index fc09e0c0..a82c19e8 100644
--- a/src/daemon/event.c
+++ b/src/daemon/event.c
@@ -914,3 +914,29 @@ levent_make_socket_blocking(int fd)
 	}
 	return 0;
 }
+
+#ifdef HOST_OS_LINUX
+/* Receive and log error from a socket when there is suspicion of an error. */
+void
+levent_recv_error(int fd, const char *source)
+{
+	do {
+		ssize_t n;
+		char buf[1024] = {};
+		struct msghdr msg = {
+			.msg_control = buf,
+			.msg_controllen = sizeof(buf)
+		};
+		if ((n = recvmsg(fd, &msg, MSG_ERRQUEUE | MSG_DONTWAIT)) <= 0) {
+			return;
+		}
+		struct cmsghdr *cmsg = CMSG_FIRSTHDR(&msg);
+		if (cmsg == NULL)
+			log_warnx("event", "received unknown error on %s",
+			    source);
+		else
+			log_warnx("event", "received error (level=%d/type=%d) on %s",
+			    cmsg->cmsg_level, cmsg->cmsg_type, source);
+	} while (1);
+}
+#endif
diff --git a/src/daemon/interfaces-linux.c b/src/daemon/interfaces-linux.c
index 9af41a58..a8280c85 100644
--- a/src/daemon/interfaces-linux.c
+++ b/src/daemon/interfaces-linux.c
@@ -71,29 +71,6 @@ iflinux_eth_send(struct lldpd *cfg, struct lldpd_hardware *hardware,
 	    buffer, size);
 }
 
-static void
-iflinux_error_recv(struct lldpd_hardware *hardware, int fd)
-{
-	do {
-		ssize_t n;
-		char buf[1024] = {};
-		struct msghdr msg = {
-			.msg_control = buf,
-			.msg_controllen = sizeof(buf)
-		};
-		if ((n = recvmsg(fd, &msg, MSG_ERRQUEUE)) <= 0) {
-			return;
-		}
-		struct cmsghdr *cmsg = CMSG_FIRSTHDR(&msg);
-		if (cmsg == NULL)
-			log_warnx("interfaces", "received unknown error on %s",
-			    hardware->h_ifname);
-		else
-			log_warnx("interfaces", "received error (level=%d/type=%d) on %s",
-			    cmsg->cmsg_level, cmsg->cmsg_type, hardware->h_ifname);
-	} while (1);
-}
-
 static int
 iflinux_generic_recv(struct lldpd_hardware *hardware,
     int fd, char *buffer, size_t size,
@@ -110,7 +87,7 @@ retry:
 		    &fromlen)) == -1) {
 		if (errno == EAGAIN && retry == 0) {
 			/* There may be an error queued in the socket. Clear it and retry. */
-			iflinux_error_recv(hardware, fd);
+			levent_recv_error(fd, hardware->h_ifname);
 			retry++;
 			goto retry;
 		}
diff --git a/src/daemon/lldpd.h b/src/daemon/lldpd.h
index 97306a2d..cf25dd67 100644
--- a/src/daemon/lldpd.h
+++ b/src/daemon/lldpd.h
@@ -127,6 +127,9 @@ void	 levent_schedule_pdu(struct lldpd_hardware *);
 void	 levent_schedule_cleanup(struct lldpd *);
 int	 levent_make_socket_nonblocking(int);
 int	 levent_make_socket_blocking(int);
+#ifdef HOST_OS_LINUX
+void	 levent_recv_error(int, const char*);
+#endif
 
 /* lldp.c */
 int	 lldp_send_shutdown(PROTO_SEND_SIG);
diff --git a/src/daemon/netlink.c b/src/daemon/netlink.c
index 4e5e34d6..d3e26dec 100644
--- a/src/daemon/netlink.c
+++ b/src/daemon/netlink.c
@@ -442,8 +442,7 @@ netlink_recv(struct lldpd *cfg,
     struct interfaces_device_list *ifs,
     struct interfaces_address_list *ifas)
 {
-	int end = 0, ret = 0;
-	int flags = MSG_PEEK | MSG_TRUNC;
+	int end = 0, ret = 0, flags, retry = 0;
 	struct iovec iov;
 	int link_update = 0;
 	int s = cfg->g_netlink->nl_socket;
@@ -471,12 +470,16 @@ netlink_recv(struct lldpd *cfg,
 			.msg_name = &peer,
 			.msg_namelen = sizeof(struct sockaddr_nl)
 		};
-
+		flags = MSG_PEEK | MSG_TRUNC;
 retry:
-		len = recvmsg(s, &rtnl_reply, flags);
+		len = recvmsg(s, &rtnl_reply, flags | MSG_DONTWAIT);
 		if (len == -1) {
 			if (errno == EAGAIN || errno == EWOULDBLOCK) {
-				log_debug("netlink", "should have received something, but didn't");
+				if (retry++ == 0) {
+					levent_recv_error(s, "netlink socket");
+					goto retry;
+				}
+				log_warnx("netlink", "should have received something, but didn't");
 				ret = 0;
 				goto out;
 			}
@@ -650,7 +653,6 @@ retry:
 				    msg->nlmsg_type, msg->nlmsg_len);
 			}
 		}
-		flags = MSG_PEEK | MSG_TRUNC;
 	}
 end:
 	if (link_update) {
-- 
2.39.5