[fast-reboot] Check if ASIC config has changed before warm reboot (#621)

[fast-reboot] Check if ASIC config has changed before fast reboot
* Adds script to compare checksums for config files between SONiC images
* Adds pre-check to fast reboot script to confirm config files match before performing fast reboot

Signed-off-by: Danny Allen <daall@microsoft.com>

* Fix formatting

* Incorporate feedback

* Add check for reboot type and override ability

* Clean up error checking

scripts/asic_config_check

@@ -0,0 +1,86 @@
+#!/bin/bash -e
+
+# Exit codes
+ASIC_CONFIG_UNCHANGED=0
+ASIC_CONFIG_CHANGED=1
+SRC_ASIC_CONFIG_NOT_FOUND=2
+DST_ASIC_CONFIG_NOT_FOUND=3
+
+# Logging utilities
+function error()
+{
+    logger -p user.err "$@"
+}
+
+function debug()
+{
+    logger -p user.info "$@"
+}
+
+# Retrieve the source ASIC config checksum from the source image
+# Exits with error SRC_ASIC_CONFIG_NOT_FOUND if the checksum is not found
+function GetSourceASICConfigChecksum()
+{
+    if [[ ! -f "/etc/sonic/asic_config_checksum" ]]; then
+            error "ASIC config not found in src image, can't verify changes"
+            exit "${SRC_ASIC_CONFIG_NOT_FOUND}"
+    fi
+}
+
+# Retrieve the destination ASIC config checksum from the destination image
+# Exits with error DST_ASIC_CONFIG_NOT_FOUND if the checksum is not found
+function GetDestinationASICConfigChecksum()
+{
+    DST_IMAGE_PATH="/host/image-${DST_SONIC_IMAGE#SONiC-OS-}"
+    FS_PATH="${DST_IMAGE_PATH}/fs.squashfs"
+    FS_MOUNTPOINT="/tmp/image-${DST_SONIC_IMAGE#SONiC-OS-}-fs"
+
+    # Verify that the destination image exists
+    if [[ ! -d ${DST_IMAGE_PATH} ]]; then
+        error "ASIC config not found in dst image, can't verify changes"
+        exit "${DST_ASIC_CONFIG_NOT_FOUND}"
+    fi
+
+    mkdir "${FS_MOUNTPOINT}"
+    mount -t squashfs "${FS_PATH}" "${FS_MOUNTPOINT}"
+
+    if [[ ! -f "${FS_MOUNTPOINT}/etc/sonic/asic_config_checksum" ]]; then
+        error "ASIC config not found in dst image, can't verify changes"
+        umount "${FS_MOUNTPOINT}"
+        rm -rf "${FS_MOUNTPOINT}"
+        exit "${DST_ASIC_CONFIG_NOT_FOUND}"
+    fi
+
+    cp "${FS_MOUNTPOINT}/etc/sonic/asic_config_checksum" /tmp/dst_asic_config_checksum
+    umount "${FS_MOUNTPOINT}"
+    rm -rf "${FS_MOUNTPOINT}"
+}
+
+# Confirm that the src and dst ASIC config checksums match
+# Exits with ASIC_CONFIG_CHANGED if the checksums differ
+function ConfirmASICConfigChecksumsMatch()
+{
+    SRC_CONFIG_CHECKSUM=$(cat /etc/sonic/asic_config_checksum)
+    DST_CONFIG_CHECKSUM=$(cat /tmp/dst_asic_config_checksum)
+    if [[ "${SRC_CONFIG_CHECKSUM}" != "${DST_CONFIG_CHECKSUM}" ]]; then
+        error "ASIC config may have changed, checksum failed"
+        exit "${ASIC_CONFIG_CHANGED}"
+    fi
+}
+
+# Main starts here
+debug "Checking that ASIC configuration has not changed"
+
+SRC_SONIC_IMAGE="$(sonic_installer list | grep "Current: " | cut -f2 -d' ')"
+DST_SONIC_IMAGE="$(sonic_installer list | grep "Next: " | cut -f2 -d' ')"
+if [[ "${SRC_SONIC_IMAGE}" == "${DST_SONIC_IMAGE}" ]]; then
+    debug "ASIC config unchanged, src and dst SONiC version are the same"
+    exit "${ASIC_CONFIG_UNCHANGED}"
+fi
+
+GetSourceASICConfigChecksum
+GetDestinationASICConfigChecksum
+ConfirmASICConfigChecksumsMatch
+
+debug "ASIC config unchanged, checksum passed"
+exit "${ASIC_CONFIG_UNCHANGED}"

scripts/fast-reboot

@@ -250,6 +250,23 @@ function reboot_pre_check()
         debug "Next image ${NEXT_SONIC_IMAGE} doesn't exist ..."
         exit ${EXIT_NEXT_IMAGE_NOT_EXISTS}
     fi
+    
+    # Make sure ASIC configuration has not changed between images
+    ASIC_CONFIG_CHECK_SCRIPT="/usr/bin/asic_config_check"
+    ASIC_CONFIG_CHECK_SUCCESS=0
+    if [[ "$REBOOT_TYPE" = "warm-reboot" || "$REBOOT_TYPE" = "fastfast-reboot" ]]; then
+        ASIC_CONFIG_CHECK_EXIT_CODE=0
+        ${ASIC_CONFIG_CHECK_SCRIPT} || ASIC_CONFIG_CHECK_EXIT_CODE=$?
+
+        if [[ "${ASIC_CONFIG_CHECK_EXIT_CODE}" != "${ASIC_CONFIG_CHECK_SUCCESS}" ]]; then
+            if [[ x"${FORCE}" == x"yes" ]]; then
+                debug "Ignoring ASIC config checksum failure..."
+            else
+                error "ASIC config may have changed: errno=${ASIC_CONFIG_CHECK_EXIT_CODE}"
+                exit "${EXIT_FAILURE}"
+            fi
+        fi
+    fi
 }
 
 function unload_kernel()

setup.py

@@ -52,6 +52,7 @@
     },
     scripts=[
         'scripts/aclshow',
+        'scripts/asic_config_check',
         'scripts/boot_part',
         'scripts/coredump-compress',
         'scripts/db_migrator.py',