netifd: Link layer state awareness support on device level

Device layer is informed by netlink events regarding the link layer status. Link
layer status change results in a DEV_EVENT_LINK_UP/DEV_EVENT_LINK_DOWN broadcast
event for a given device.

Depends on uloop error callback patch.

Solves issue reported in https://dev.openwrt.org/ticket/14590

Signed-off-by: Hans Dedecker <dedeckeh@gmail.com>
Acked-by: Karl Vogel <karl.vogel@gmail.com>

3 files changed, 162 insertions, 25 deletions

diff --git a/device.c b/device.c
index feeb290..6a770ac 100644
--- a/device.c
+++ b/device.c
@@ -211,7 +211,7 @@ int device_claim(struct device_user *dep)
 		return 0;
 
 	dep->claimed = true;
-	D(DEVICE, "Claim %s %s, new refcount: %d\n", dev->type->name, dev->ifname, dev->active + 1);
+	D(DEVICE, "Claim %s %s, new active count: %d\n", dev->type->name, dev->ifname, dev->active + 1);
 	if (++dev->active != 1)
 		return 0;
 
@@ -237,7 +237,7 @@ void device_release(struct device_user *dep)
 
 	dep->claimed = false;
 	dev->active--;
-	D(DEVICE, "Release %s %s, new refcount: %d\n", dev->type->name, dev->ifname, dev->active);
+	D(DEVICE, "Release %s %s, new active count: %d\n", dev->type->name, dev->ifname, dev->active);
 	assert(dev->active >= 0);
 
 	if (dev->active)
@@ -394,6 +394,26 @@ void device_set_present(struct device *dev, bool state)
 	device_refresh_present(dev);
 }
 
+void device_set_link(struct device *dev, bool state)
+{
+	if (dev->link_active == state)
+		return;
+
+	netifd_log_message(L_NOTICE, "%s '%s' link is %s\n", dev->type->name, dev->ifname, state ? "up" : "down" );
+
+	dev->link_active = state;
+	device_broadcast_event(dev, state ? DEV_EVENT_LINK_UP : DEV_EVENT_LINK_DOWN);
+}
+
+void device_set_ifindex(struct device *dev, int ifindex)
+{
+	if (dev->ifindex == ifindex)
+		return;
+
+	dev->ifindex = ifindex;
+	device_broadcast_event(dev, DEV_EVENT_UPDATE_IFINDEX);
+}
+
 static int device_refcount(struct device *dev)
 {
 	struct list_head *list;
@@ -435,6 +455,9 @@ void device_add_user(struct device_user *dep, struct device *dev)
 		dep->cb(dep, DEV_EVENT_ADD);
 		if (dev->active)
 			dep->cb(dep, DEV_EVENT_UP);
+
+		if (dev->link_active)
+			dep->cb(dep, DEV_EVENT_LINK_UP);
 	}
 }
 
@@ -667,6 +690,8 @@ device_dump_status(struct blob_buf *b, struct device *dev)
 		return;
 
 	blobmsg_add_u8(b, "up", !!dev->active);
+	blobmsg_add_u8(b, "carrier", !!dev->link_active);
+
 	if (dev->type->dump_info)
 		dev->type->dump_info(dev, b);
 	else
diff --git a/device.h b/device.h
index d63ffe3..dd57927 100644
--- a/device.h
+++ b/device.h
@@ -68,6 +68,7 @@ enum device_event {
 	DEV_EVENT_REMOVE,
 
 	DEV_EVENT_UPDATE_IFNAME,
+	DEV_EVENT_UPDATE_IFINDEX,
 
 	DEV_EVENT_SETUP,
 	DEV_EVENT_TEARDOWN,
@@ -122,6 +123,7 @@ struct device {
 	bool sys_present;
 	bool present;
 	int active;
+	bool link_active;
 
 	bool external;
 	bool disabled;
@@ -178,6 +180,8 @@ void device_remove_user(struct device_user *dep);
 void device_broadcast_event(struct device *dev, enum device_event ev);
 
 void device_set_present(struct device *dev, bool state);
+void device_set_link(struct device *dev, bool state);
+void device_set_ifindex(struct device *dev, int ifindex);
 void device_refresh_present(struct device *dev);
 int device_claim(struct device_user *dep);
 void device_release(struct device_user *dep);
diff --git a/system-linux.c b/system-linux.c
index cbcc5bd..12682dc 100644
--- a/system-linux.c
+++ b/system-linux.c
@@ -58,7 +58,7 @@
 struct event_socket {
 	struct uloop_fd uloop;
 	struct nl_sock *sock;
-	struct nl_cb *cb;
+	int bufsize;
 };
 
 static int sock_ioctl = -1;
@@ -73,7 +73,38 @@ static void
 handler_nl_event(struct uloop_fd *u, unsigned int events)
 {
 	struct event_socket *ev = container_of(u, struct event_socket, uloop);
-	nl_recvmsgs(ev->sock, ev->cb);
+	int err;
+	socklen_t errlen = sizeof(err);
+
+	if (!u->error) {
+		nl_recvmsgs_default(ev->sock);
+		return;
+	}
+
+	if (getsockopt(u->fd, SOL_SOCKET, SO_ERROR, (void *)&err, &errlen))
+		goto abort;
+
+	switch(err) {
+	case ENOBUFS:
+		// Increase rx buffer size on netlink socket
+		ev->bufsize *= 2;
+		if (nl_socket_set_buffer_size(ev->sock, ev->bufsize, 0))
+			goto abort;
+
+		// Request full dump since some info got dropped
+		struct rtgenmsg msg = { .rtgen_family = AF_UNSPEC };
+		nl_send_simple(ev->sock, RTM_GETLINK, NLM_F_DUMP, &msg, sizeof(msg));
+		break;
+
+	default:
+		goto abort;
+	}
+	u->error = false;
+	return;
+
+abort:
+	uloop_fd_delete(&ev->uloop);
+	return;
 }
 
 static struct nl_sock *
@@ -96,7 +127,7 @@ create_socket(int protocol, int groups)
 
 static bool
 create_raw_event_socket(struct event_socket *ev, int protocol, int groups,
-			uloop_fd_handler cb)
+                        uloop_fd_handler cb, int flags)
 {
 	ev->sock = create_socket(protocol, groups);
 	if (!ev->sock)
@@ -104,7 +135,9 @@ create_raw_event_socket(struct event_socket *ev, int protocol, int groups,
 
 	ev->uloop.fd = nl_socket_get_fd(ev->sock);
 	ev->uloop.cb = cb;
-	uloop_fd_add(&ev->uloop, ULOOP_READ | ULOOP_EDGE_TRIGGER);
+	if (uloop_fd_add(&ev->uloop, ULOOP_READ|flags))
+		return false;
+
 	return true;
 }
 
@@ -112,14 +145,21 @@ static bool
 create_event_socket(struct event_socket *ev, int protocol,
 		    int (*cb)(struct nl_msg *msg, void *arg))
 {
-	// Prepare socket for link events
-	ev->cb = nl_cb_alloc(NL_CB_DEFAULT);
-	if (!ev->cb)
+	if (!create_raw_event_socket(ev, protocol, 0, handler_nl_event, ULOOP_ERROR_CB))
 		return false;
 
-	nl_cb_set(ev->cb, NL_CB_VALID, NL_CB_CUSTOM, cb, NULL);
+	// Install the valid custom callback handler
+	nl_socket_modify_cb(ev->sock, NL_CB_VALID, NL_CB_CUSTOM, cb, NULL);
+
+	// Disable sequence number checking on event sockets
+	nl_socket_disable_seq_check(ev->sock);
 
-	return create_raw_event_socket(ev, protocol, 0, handler_nl_event);
+	// Increase rx buffer size to 65K on event sockets
+	ev->bufsize = 65535;
+	if (nl_socket_set_buffer_size(ev->sock, ev->bufsize, 0))
+		return false;
+
+	return true;
 }
 
 int system_init(void)
@@ -128,7 +168,7 @@ int system_init(void)
 	static struct event_socket hotplug_event;
 
 	sock_ioctl = socket(AF_LOCAL, SOCK_DGRAM, 0);
-	fcntl(sock_ioctl, F_SETFD, fcntl(sock_ioctl, F_GETFD) | FD_CLOEXEC);
+	system_fd_set_cloexec(sock_ioctl);
 
 	// Prepare socket for routing / address control
 	sock_rtnl = create_socket(NETLINK_ROUTE, 0);
@@ -139,7 +179,7 @@ int system_init(void)
 		return -1;
 
 	if (!create_raw_event_socket(&hotplug_event, NETLINK_KOBJECT_UEVENT, 1,
-				     handle_hotplug_event))
+	                             handle_hotplug_event, 0))
 		return -1;
 
 	// Receive network link events form kernel
@@ -171,6 +211,10 @@ static void system_set_disable_ipv6(struct device *dev, const char *val)
 	system_set_dev_sysctl("/proc/sys/net/ipv6/conf/%s/disable_ipv6", dev->ifname, val);
 }
 
+#ifndef IFF_LOWER_UP
+#define IFF_LOWER_UP	0x10000
+#endif
+
 // Evaluate netlink messages
 static int cb_rtnl_event(struct nl_msg *msg, void *arg)
 {
@@ -178,19 +222,19 @@ static int cb_rtnl_event(struct nl_msg *msg, void *arg)
 	struct ifinfomsg *ifi = NLMSG_DATA(nh);
 	struct nlattr *nla[__IFLA_MAX];
 
-	if (nh->nlmsg_type != RTM_DELLINK && nh->nlmsg_type != RTM_NEWLINK)
+	if (nh->nlmsg_type != RTM_NEWLINK)
 		goto out;
 
 	nlmsg_parse(nh, sizeof(*ifi), nla, __IFLA_MAX - 1, NULL);
 	if (!nla[IFLA_IFNAME])
 		goto out;
 
-	struct device *dev = device_get(RTA_DATA(nla[IFLA_IFNAME]), false);
+	struct device *dev = device_get(nla_data(nla[IFLA_IFNAME]), false);
 	if (!dev)
 		goto out;
 
-	dev->ifindex = ifi->ifi_index;
-	/* TODO: parse link status */
+	device_set_ifindex(dev, ifi->ifi_index);
+	device_set_link(dev, ifi->ifi_flags & IFF_LOWER_UP ? true : false);
 
 out:
 	return 0;
@@ -538,7 +582,7 @@ void system_if_clear_state(struct device *dev)
 	if (dev->external)
 		return;
 
-	dev->ifindex = system_if_resolve(dev);
+	device_set_ifindex(dev, system_if_resolve(dev));
 	if (!dev->ifindex)
 		return;
 
@@ -782,7 +826,7 @@ int system_if_up(struct device *dev)
 {
 	system_if_get_settings(dev, &dev->orig_settings);
 	system_if_apply_settings(dev, &dev->settings);
-	dev->ifindex = system_if_resolve(dev);
+	device_set_ifindex(dev, system_if_resolve(dev));
 	return system_if_flags(dev->ifname, IFF_UP, 0);
 }
 
@@ -794,10 +838,78 @@ int system_if_down(struct device *dev)
 	return ret;
 }
 
+struct if_check_data {
+	struct device *dev;
+	int pending;
+	int ret;
+};
+
+static int cb_if_check_valid(struct nl_msg *msg, void *arg)
+{
+	struct nlmsghdr *nh = nlmsg_hdr(msg);
+	struct ifinfomsg *ifi = NLMSG_DATA(nh);
+	struct if_check_data *chk = (struct if_check_data *)arg;
+
+	if (nh->nlmsg_type != RTM_NEWLINK)
+		return NL_SKIP;
+
+	device_set_present(chk->dev, ifi->ifi_index > 0 ? true : false);
+	device_set_link(chk->dev, ifi->ifi_flags & IFF_LOWER_UP ? true : false);
+
+	return NL_OK;
+}
+
+static int cb_if_check_ack(struct nl_msg *msg, void *arg)
+{
+	struct if_check_data *chk = (struct if_check_data *)arg;
+	chk->pending = 0;
+	return NL_STOP;
+}
+
+static int cb_if_check_error(struct sockaddr_nl *nla, struct nlmsgerr *err, void *arg)
+{
+	struct if_check_data *chk = (struct if_check_data *)arg;
+
+	device_set_present(chk->dev, false);
+	device_set_link(chk->dev, false);
+	chk->pending = err->error;
+
+	return NL_STOP;
+}
+
 int system_if_check(struct device *dev)
 {
-	device_set_present(dev, (system_if_resolve(dev) > 0));
-	return 0;
+	struct nl_cb *cb = nl_cb_alloc(NL_CB_DEFAULT);
+	struct nl_msg *msg;
+	struct ifinfomsg ifi = {
+		.ifi_family = AF_UNSPEC,
+		.ifi_index = 0,
+	};
+	struct if_check_data chk = {
+		.dev = dev,
+		.pending = 1,
+	};
+	int ret = 1;
+
+	msg = nlmsg_alloc_simple(RTM_GETLINK, 0);
+	if (!msg || nlmsg_append(msg, &ifi, sizeof(ifi), 0) ||
+	    nla_put_string(msg, IFLA_IFNAME, dev->ifname))
+		goto out;
+
+	nl_cb_set(cb, NL_CB_VALID, NL_CB_CUSTOM, cb_if_check_valid, &chk);
+	nl_cb_set(cb, NL_CB_ACK, NL_CB_CUSTOM, cb_if_check_ack, &chk);
+	nl_cb_err(cb, NL_CB_CUSTOM, cb_if_check_error, &chk);
+
+	nl_send_auto_complete(sock_rtnl, msg);
+	while (chk.pending > 0)
+		nl_recvmsgs(sock_rtnl, cb);
+
+	nlmsg_free(msg);
+	ret = chk.pending;
+
+out:
+	nl_cb_put(cb);
+	return ret;
 }
 
 struct device *
@@ -915,14 +1027,10 @@ system_if_dump_info(struct device *dev, struct blob_buf *b)
 	char buf[64], *s;
 	void *c;
 	int dir_fd;
-	uint64_t val = 0;
 
 	snprintf(buf, sizeof(buf), "/sys/class/net/%s", dev->ifname);
 	dir_fd = open(buf, O_DIRECTORY);
 
-	if (read_uint64_file(dir_fd, "carrier", &val))
-		blobmsg_add_u8(b, "link", !!val);
-
 	memset(&ecmd, 0, sizeof(ecmd));
 	memset(&ifr, 0, sizeof(ifr));
 	strcpy(ifr.ifr_name, dev->ifname);