From fdb92706177c6e8f6bf61c59663d6d5de54d73b3 Mon Sep 17 00:00:00 2001
From: Tim Wilkinson <tim.j.wilkinson@gmail.com>
Date: Sun, 29 Jan 2023 19:21:03 -0800
Subject: [PATCH] Improve firmware failure error recovery (#683)

---
 files/usr/local/bin/upgrade_prepare.sh | 51 ++++++++++++++++----------
 files/usr/local/bin/uploadctlservices  | 10 ++---
 files/www/cgi-bin/admin                |  8 ++--
 3 files changed, 42 insertions(+), 27 deletions(-)

diff --git a/files/usr/local/bin/upgrade_prepare.sh b/files/usr/local/bin/upgrade_prepare.sh
index 7220674f..d6ff0bdb 100755
--- a/files/usr/local/bin/upgrade_prepare.sh
+++ b/files/usr/local/bin/upgrade_prepare.sh
@@ -39,23 +39,36 @@ LICENSE
 #
 SERVICES="dnsmasq:dnsmasq dropbear:dropbear urngd:urngd rpcd:rpcd telnet:telnetd manager:manager.lua log:logd"
 
-#
-# We unceremoniously kill services, and then stop them to prevent
-# procd restarting them again
-#
-for S in ${SERVICES}
-do
-    srv=$(echo ${S} | cut -d: -f1)
-    daemon=$(echo ${S} | cut -d: -f2 -s)
-    if [ "${daemon}" != "" ]; then
-        killall -KILL ${daemon}
-    fi
-    if [ -x /etc/init.d/${srv} ]; then
-        /etc/init.d/${srv} stop
-    fi
-done
+if [ "$1" = "restore" ]; then
+    #
+    # Restart everything
+    #
+    for S in ${SERVICES}
+    do
+        srv=$(echo ${S} | cut -d: -f1)
+        if [ -x /etc/init.d/${srv} ]; then
+            /etc/init.d/${srv} start
+        fi
+    done
+else
+    #
+    # We unceremoniously kill services, and then stop them to prevent
+    # procd restarting them again
+    #
+    for S in ${SERVICES}
+    do
+        srv=$(echo ${S} | cut -d: -f1)
+        daemon=$(echo ${S} | cut -d: -f2 -s)
+        if [ "${daemon}" != "" ]; then
+            killall -KILL ${daemon}
+        fi
+        if [ -x /etc/init.d/${srv} ]; then
+            /etc/init.d/${srv} stop
+        fi
+    done
 
-#
-# Drop page cache to take pressure of tmps
-#
-echo 3 > /proc/sys/vm/drop_caches
+    #
+    # Drop page cache to take pressure of tmps
+    #
+    echo 3 > /proc/sys/vm/drop_caches
+fi
diff --git a/files/usr/local/bin/uploadctlservices b/files/usr/local/bin/uploadctlservices
index 8508733b..9b135b58 100755
--- a/files/usr/local/bin/uploadctlservices
+++ b/files/usr/local/bin/uploadctlservices
@@ -35,19 +35,19 @@ remove_opkg_lists() {
 
 return_to_operating_mode() {
 
-
-    # If anything goes wrong after this point return the error code to the calling process
-    # as we could be in an unstable state.
-    set -e
-
     # Start up the services we stopped
+    err=0
     for name in $serviceslist
     do
         "/etc/init.d/${name}" start
+        if [ "$?" != "0" ]; then
+            err=-1
+        fi
     done
 
     rm -f /tmp/.upgrade_mode
 
+    exit $err
 }
 
 
diff --git a/files/www/cgi-bin/admin b/files/www/cgi-bin/admin
index 11fbf63d..f17caadc 100755
--- a/files/www/cgi-bin/admin
+++ b/files/www/cgi-bin/admin
@@ -188,7 +188,7 @@ if os.getenv("REQUEST_METHOD") == "POST" then
                     firmfile = meta.file
                     if firmfile:match("sysupgrade%.bin$") then
                         -- Uploading a system upgrade - clear out memory early
-                        os.execute("/usr/local/bin/upgrade_prepare.sh > /dev/null 2>&1")
+                        os.execute("/usr/local/bin/upgrade_prepare.sh stop > /dev/null 2>&1")
                     end
                 end
                 nixio.fs.mkdir("/tmp/web/upload")
@@ -398,7 +398,7 @@ if parms.button_apply_fw and nixio.fs.stat("/tmp/web/local_firmware.bin") then
     os.execute("mv -f /tmp/web/local_firmware.bin /tmp/web/upload/file")
     firmfile = "arednmesh-sideload-sysupgrade.bin"
     parms.button_ul_fw = "Upload"
-    os.execute("/usr/local/bin/upgrade_prepare.sh > /dev/null 2>&1")
+    os.execute("/usr/local/bin/upgrade_prepare.sh stop > /dev/null 2>&1")
 end
 
 -- upload fw
@@ -423,6 +423,7 @@ if parms.button_ul_fw and nixio.fs.stat("/tmp/web/upload/file") then
         fwout("Firmware CANNOT be updated")
         fwout("the uploaded file is not recognized")
         nixio.fs.remove(tmpdir .. "/firmware")
+        os.execute("/usr/local/bin/upgrade_prepare.sh restore > /dev/null 2>&1")
         if os.execute("/usr/local/bin/uploadctlservices restore > /dev/null 2>&1") ~= 0 then
             fwout("Failed to restart all services, please reboot this node.")
         end
@@ -434,7 +435,7 @@ if parms.button_dl_fw and parms.dl_fw ~= "default" then
     if get_default_gw() ~= "none" or uciserverpath:match("%.local%.mesh") then
         nixio.fs.remove(tmpdir .. "/firmware")
         os.execute("/usr/local/bin/uploadctlservices update > /dev/null 2>&1")
-        os.execute("/usr/local/bin/upgrade_prepare.sh > /dev/null 2>&1")
+        os.execute("/usr/local/bin/upgrade_prepare.sh stop > /dev/null 2>&1")
 
         fw_install = false
         local err
@@ -478,6 +479,7 @@ if parms.button_dl_fw and parms.dl_fw ~= "default" then
         if err then
             fwout("Firmware CANNOT be updated")
             fwout(err)
+            os.execute("/usr/local/bin/upgrade_prepare.sh restore > /dev/null 2>&1")
             if os.execute("/usr/local/bin/uploadctlservices restore > /dev/null 2>&1") ~= 0 then
                 fwout("Failed to restart all services, please reboot this node.")
             end