[slx-network] Retry bringing iface up in wait-loop

2 files changed, 22 insertions, 17 deletions

diff --git a/modules.d/slx-network/hooks/s3-parse-network-kcl.sh b/modules.d/slx-network/hooks/s3-parse-network-kcl.sh
index daf2e9b7..123d2d03 100755
--- a/modules.d/slx-network/hooks/s3-parse-network-kcl.sh
+++ b/modules.d/slx-network/hooks/s3-parse-network-kcl.sh
@@ -117,3 +117,4 @@ save_network_config &
 [ -n "$hostname" ] && echo "$hostname" > /proc/sys/kernel/hostname
 
 wait
+exit 0
diff --git a/modules.d/slx-network/hooks/s3-setup-bootif-network.sh b/modules.d/slx-network/hooks/s3-setup-bootif-network.sh
index 1ae58464..6b75170d 100755
--- a/modules.d/slx-network/hooks/s3-setup-bootif-network.sh
+++ b/modules.d/slx-network/hooks/s3-setup-bootif-network.sh
@@ -13,12 +13,12 @@ type emergency_shell >/dev/null 2>&1 || . /lib/dracut-lib.sh
 # do not run until the physical interface exists (driver loaded)
 _fails=0
 while ! [ -e "/sys/class/net/${SLX_PXE_NETIF}/device" ]; do
-	if (( ++_fails > 20 )); then
+	if (( ++_fails > 30 )); then
 		emergency_shell "Boot interface '${SLX_PXE_NETIF}' did not appear"
 		exit 1
 	fi
-	(( (_fails % 5) == 0 )) && udevadm trigger
-	usleep 500000
+	(( (_fails % 7) == 0 )) && udevadm trigger --subsystem-match=net
+	usleep 333333
 done
 
 # wrapper around splashtool to disable it if its not present
@@ -33,7 +33,7 @@ fi
 
 wait_for_iface() {
 	local _iface="$1"
-	local _timeout="${2:-50}"
+	local _timeout="$(( ${2:-50} * 3 ))"
 	local _fails=0
 	local _state
 	if [ "$_iface" = "$SLX_PXE_NETIF" ]; then
@@ -42,6 +42,7 @@ wait_for_iface() {
 		img="??-bridge"
 	fi
 	while (( _timeout-- > 0 )); do
+		ip link set dev "$_iface" up
 		_state="$(cat "/sys/class/net/${_iface}/operstate" 2> /dev/null)"
 		[ "$_state" = "up" ] && break
 		if (( ++_fails > 5 )) && [[ "$_state" = "unknown" || "$_state" = "" ]]; then
@@ -49,8 +50,8 @@ wait_for_iface() {
 				break
 			fi
 		fi
-		# every 500ms
-		usleep 500000
+		# every 333ms
+		usleep 333333
 		in=
 		(( (_timeout % 2) == 0 )) && in=in
 		_splashtool --icon "/opt/openslx/icons/${in}active/${img}.ppm" &
@@ -68,8 +69,7 @@ wait_for_iface() {
 {
 set -x
 
-ip link set dev "$SLX_PXE_NETIF" up
-if ! wait_for_iface "$SLX_PXE_NETIF" 60; then
+if ! wait_for_iface "$SLX_PXE_NETIF" 30; then
 	warn "'$SLX_PXE_NETIF' still not up after 30sec ... trying anyway."
 	# TODO handle case where we waited for 30sec and it is still not up
 fi
@@ -92,14 +92,14 @@ fi
 if [ -n "$SLX_BRIDGE" ]; then
 	for try in {1..10} ""; do
 		if (
-		set -e
-		brctl addbr "$SLX_BRIDGE"
-		brctl stp "$SLX_BRIDGE" 0
-		brctl setfd "$SLX_BRIDGE" 0.000000000001
-		ip link set addr "$SLX_PXE_MAC" "$SLX_BRIDGE"
-		brctl addif "$SLX_BRIDGE" "$MAIN_NETIF"
-		ip link set dev "$SLX_BRIDGE" up
-		wait_for_iface "$SLX_BRIDGE"
+				set -e
+				brctl addbr "$SLX_BRIDGE"
+				brctl stp "$SLX_BRIDGE" 0
+				brctl setfd "$SLX_BRIDGE" 0.000000000001
+				ip link set addr "$SLX_PXE_MAC" "$SLX_BRIDGE"
+				brctl addif "$SLX_BRIDGE" "$MAIN_NETIF"
+				ip link set dev "$SLX_BRIDGE" up
+				wait_for_iface "$SLX_BRIDGE"
 		); then
 			MAIN_NETIF="$SLX_BRIDGE"
 			break
@@ -112,7 +112,11 @@ if [ -n "$SLX_BRIDGE" ]; then
 		fi
 		warn "Failed to setup main network bridge on try $try. Retrying ..."
 		# delete bridge, inc try and sleep 100ms before trying again
-		[ -e "/sys/class/net/${SLX_BRIDGE}" ] && brctl delbr "$SLX_BRIDGE"
+		if [ -e "/sys/class/net/${SLX_BRIDGE}" ]; then
+			ip link set dev "$SLX_BRIDGE" down
+			brctl delif "$SLX_BRIDGE" "$MAIN_NETIF"
+			brctl delbr "$SLX_BRIDGE"
+		fi
 		usleep 100000
 	done
 fi