Relax insmod spl failures, try to load zfs.ko

Instead of manually loading spl.ko every time, try to load it via
modprobe. Iff that fails, use modinfo to find the path to spl.ko and try
to load it with spl_hostid=0. Since udevd could have already loaded zfs
(and by extension, spl), insmod failures are not hard failures. Warn
that the module couldn't be loaded along with the error text and then
try to load zfs.ko anyway.

This moves all of the on-disk / /sys hostid manipulation to after the
modules have been loaded. This ensures that regardless of how spl was
loaded, /sys/module/spl/parameters/spl_hostid is set to 0 to force
/etc/hostid to take precedence.

Author: zdykstra

90zfsbootmenu/zfsbootmenu-init.sh

@@ -25,6 +25,38 @@ fi
 
 mkdir -p "${BASE}"
 
+# Attempt to load spl normally
+if ! _modload="$( modprobe spl 2>&1 )" ; then
+  zdebug "${_modload}"
+
+  # Capture the filename for spl.ko
+  _modfilename="$( modinfo -F filename spl )"
+
+  if [ -n "${_modfilename}" ] ; then
+    zinfo "loading ${_modfilename}"
+
+    # Load with a hostid of 0, so that /etc/hostid takes precedence and
+    # invalid spl.spl_hostid values are ignored
+
+    # There's a race condition between udev and insmod spl
+    # insmod failures are no longer a hard failure - they can be because
+    #  1. spl.ko is already loaded because of the race condition
+    #  2. there's an invalid parameter or value for spl.ko
+
+    if ! _modload="$( insmod "${_modfilename}" "spl_hostid=0" 2>&1 )" ; then
+      zwarn "${_modload}"
+      zwarn "unable to load SPL kernel module; attempting to load ZFS anyway"
+    fi
+  fi
+fi
+
+if ! _modload="$( modprobe zfs 2>&1 )" ; then
+  zerror "${_modload}"
+  emergency_shell "unable to load ZFS kernel modules"
+fi
+
+udevadm settle
+
 # Write out a default or overridden hostid
 if [ -n "${spl_hostid}" ] ; then
   if write_hostid "${spl_hostid}" ; then
@@ -42,26 +74,6 @@ elif [ ! -e /etc/hostid ]; then
   write_hostid "${default_hostid}"
 fi
 
-# only load spl.ko if it isn't already loaded
-if ! lsmod | grep -E -q "^spl" ; then
-  # Capture the filename for spl.ko
-  _modfilename="$( modinfo -F filename spl )"
-  zinfo "loading ${_modfilename}"
-
-  # Load with a hostid of 0, so that /etc/hostid takes precedence
-  if ! _modload="$( insmod "${_modfilename}" "spl_hostid=0" 2>&1 )" ; then
-    zdebug "${_modload}"
-    emergency_shell "unable to load SPL kernel module"
-  fi
-fi
-
-if ! _modload="$( modprobe zfs 2>&1 )" ; then
-  zdebug "${_modload}"
-  emergency_shell "unable to load ZFS kernel modules"
-fi
-
-udevadm settle
-
 # Prefer a specific pool when checking for a bootfs value
 # shellcheck disable=SC2154
 if [ "${root}" = "zfsbootmenu" ]; then