[mlnx-warmboot]: Add warmboot support on Mellanox platforms

files/scripts/syncd.sh

@@ -99,6 +99,10 @@ stop() {
         debug "Warm shutdown syncd process ..."
         /usr/bin/docker exec -i syncd /usr/bin/syncd_request_shutdown --warm
 
+        if [ x$sonic_asic_platform == x'mellanox' ]; then
+            /usr/bin/docker exec -i syncd /usr/bin/checkpoint.sh
+        fi
+
         # wait until syncd quits gracefully
         while docker top syncd | grep -q /usr/bin/syncd; do
             sleep 0.1

platform/mellanox/docker-syncd-mlnx.mk

@@ -13,4 +13,5 @@ $(DOCKER_SYNCD_MLNX)_CONTAINER_NAME = syncd
 $(DOCKER_SYNCD_MLNX)_RUN_OPT += --net=host --privileged -t
 $(DOCKER_SYNCD_MLNX)_RUN_OPT += -v /host/machine.conf:/etc/machine.conf
 $(DOCKER_SYNCD_MLNX)_RUN_OPT += -v /etc/sonic:/etc/sonic:ro
+$(DOCKER_SYNCD_MLNX)_RUN_OPT += -v /host/warmboot:/var/warmboot
 $(DOCKER_SYNCD_MLNX)_RUN_OPT += --tmpfs /run/criu

platform/mellanox/docker-syncd-mlnx/Dockerfile.j2

@@ -24,6 +24,9 @@ debs/{{ deb }}{{' '}}
 COPY ["start.sh", "/usr/bin/"]
 COPY ["supervisord.conf", "/etc/supervisor/conf.d/"]
 
+COPY ["restore.sh", "/usr/bin/"]
+COPY ["checkpoint.sh", "/usr/bin/"]
+
 ## Clean up
 RUN apt-get clean -y; apt-get autoclean -y; apt-get autoremove -y
 RUN rm -rf /debs

platform/mellanox/docker-syncd-mlnx/checkpoint.sh

@@ -0,0 +1,34 @@
+#!/usr/bin/env bash
+
+SYNCD_DUMP_DIR=/var/warmboot/syncd
+SYNCD_PID=`pidof syncd`
+SYNCD_EXTERNAL_ARGS=`ls -la /proc/$SYNCD_PID/fd | grep -ho "socket:\[[0-9]*\]" | sed "s/socket:/--external unix/g" | xargs`
+
+SDK_DUMP_DIR=/var/warmboot/sdk
+SDK_PID=`pidof sx_sdk`
+SDK_EXTERNAL_ARGS=`ls -la /proc/$SDK_PID/fd | grep -ho "socket:\[[0-9]*\]" | sed "s/socket:/--external unix/g" | xargs`
+
+SHM_DUMP_DIR=/var/warmboot/shm
+SHM_DIR=/dev/shm
+
+if [ -d $SYNCD_DUMP_DIR ]; then
+    rm -rf $SYNCD_DUMP_DIR
+fi
+
+if [ -d $SDK_DUMP_DIR ]; then
+    rm -rf $SDK_DUMP_DIR
+fi
+
+if [ -d $SHM_DUMP_DIR ]; then
+    rm -rf $SHM_DUMP_DIR
+fi
+
+mkdir $SYNCD_DUMP_DIR
+mkdir $SDK_DUMP_DIR
+mkdir $SHM_DUMP_DIR
+
+criu dump -D $SYNCD_DUMP_DIR -t $SYNCD_PID --shell-job $SYNCD_EXTERNAL_ARGS
+
+criu dump -D $SDK_DUMP_DIR -t $SDK_PID --shell-job $SDK_EXTERNAL_ARGS
+
+cp -rf $SHM_DIR/* $SHM_DUMP_DIR/

platform/mellanox/docker-syncd-mlnx/restore.sh

@@ -0,0 +1,21 @@
+#!/usr/bin/env bash
+
+SYNCD_DUMP_DIR=/var/warmboot/syncd
+SDK_DUMP_DIR=/var/warmboot/sdk
+SHM_DUMP_DIR=/var/warmboot/shm
+SHM_DIR=/dev/shm
+
+if [ -d $SHM_DUMP_DIR ]; then
+    cp -rf $SHM_DUMP_DIR/* $SHM_DIR/
+    rm -rf $SHM_DUMP_DIR
+fi
+
+if [ -d $SDK_DUMP_DIR ]; then
+    criu restore -d -D $SDK_DUMP_DIR --shell-job --ext-unix-sk
+    rm -rf $SDK_DUMP_DIR
+fi
+
+if [ -d $SYNCD_DUMP_DIR ]; then
+    criu restore -d -D $SYNCD_DUMP_DIR --shell-job --ext-unix-sk
+    rm -rf $SYNCD_DUMP_DIR
+fi

platform/mellanox/docker-syncd-mlnx/start.sh

@@ -1,5 +1,9 @@
 #!/usr/bin/env bash
 
+if [ -d /var/warmboot/syncd ]; then
+    echo 1000 > /proc/sys/kernel/ns_last_pid
+fi
+
 rm -f /var/run/rsyslogd.pid
 
 supervisorctl start rsyslogd

platform/mellanox/rules.mk

@@ -11,6 +11,7 @@ include $(PLATFORM_PATH)/libsaithrift-dev.mk
 include $(PLATFORM_PATH)/python-saithrift.mk
 include $(PLATFORM_PATH)/docker-ptf-mlnx.mk
 include $(PLATFORM_PATH)/mlnx-sfpd.mk
+include $(PLATFORM_PATH)/criu.mk
 
 SONIC_ALL += $(SONIC_ONE_IMAGE) \
              $(DOCKER_FPM)