fix(dmesg): detect when dmesg is truncated

To avoid false negatives when looking for a message
in dmesg, we were previously also grepping in known
on-disk archives of dmesg (dmesg.log, kern.log).
This in turn caused false positives because we have no
guarantee that we're grepping the dmesg of the current
running kernel. Hence we now only look in the live
`dmesg`, detect if it has been truncated, and report
it to the user.

spectre-meltdown-checker.sh

@@ -588,6 +588,22 @@ unload_cpuid()
 	fi
 }
 
+dmesg_grep()
+{
+	# grep for something in dmesg, ensuring that the dmesg buffer
+	# has not been truncated
+	dmesg_grepped=''
+	if ! dmesg | grep -qE '(^|\] )Linux version [0-9]'; then
+		# dmesg truncated
+		return 2
+	fi
+	dmesg_grepped=$(dmesg | grep -E "$1" | head -1)
+	# not found:
+	[ -z "$dmesg_grepped" ] && return 1
+	# found, output is in $dmesg_grepped
+	return 0
+}
+
 is_coreos()
 {
 	which coreos-install >/dev/null 2>&1 && which toolbox >/dev/null 2>&1 && return 0
@@ -1125,24 +1141,25 @@ check_variant3()
 				# RedHat Backport creates a dedicated file, see https://access.redhat.com/articles/3311301
 				kpti_enabled=$(cat /sys/kernel/debug/x86/pti_enabled 2>/dev/null)
 				_debug "kpti_enabled: file /sys/kernel/debug/x86/pti_enabled exists and says: $kpti_enabled"
-			elif dmesg | grep -Eq "$dmesg_grep"; then
+			fi
-				# if we can't find the flag, grep dmesg output
+			if [ -z "$kpti_enabled" ]; then
-				_debug "kpti_enabled: found hint in dmesg: "$(dmesg | grep -E "$dmesg_grep")
+				dmesg_grep "$dmesg_grep"; ret=$?
+				if [ $ret -eq 0 ]; then
+					_debug "kpti_enabled: found hint in dmesg: $dmesg_grepped"
 					kpti_enabled=1
-			elif [ -r /var/log/dmesg ] && grep -Eq "$dmesg_grep" /var/log/dmesg; then
+				elif [ $ret -eq 2 ]; then
-				# if we can't find the flag in dmesg output, grep in /var/log/dmesg when readable
+					_debug "kpti_enabled: dmesg truncated"
-				_debug "kpti_enabled: found hint in /var/log/dmesg: "$(grep -E "$dmesg_grep" /var/log/dmesg)
+					kpti_enabled=-1
-				kpti_enabled=1
+				fi
-			elif [ -r /var/log/kern.log ] && grep -Eq "$dmesg_grep" /var/log/kern.log; then
+			fi
-				# if we can't find the flag in dmesg output, grep in /var/log/kern.log when readable
+			if [ -z "$kpti_enabled" ]; then
-				_debug "kpti_enabled: found hint in /var/log/kern.log: "$(grep -E "$dmesg_grep" /var/log/kern.log)
-				kpti_enabled=1
-			else
 				_debug "kpti_enabled: couldn't find any hint that PTI is enabled"
 				kpti_enabled=0
 			fi
 			if [ "$kpti_enabled" = 1 ]; then
 				pstatus green YES
+			elif [ "$kpti_enabled" = -1 ]; then
+				pstatus yellow UNKNOWN "dmesg truncated, please reboot and relaunch this script"
 			else
 				pstatus red NO
 			fi
@@ -1177,15 +1194,12 @@ check_variant3()
 			_info_nol "* Checking if we're running under Xen PV (64 bits): "
 			if [ "$(uname -m)" = "x86_64" ]; then
 				# XXX do we have a better way that relying on dmesg?
-				if dmesg | grep -q 'Booting paravirtualized kernel on Xen$' ; then
+				dmesg_grep 'Booting paravirtualized kernel on Xen$'; ret=$?
-					pstatus green YES 'Xen PV is not vulnerable'
+				if [ $ret -eq 0 ]; then
-					xen_pv=1
-				elif [ -r /var/log/dmesg ] && grep -q 'Booting paravirtualized kernel on Xen$' /var/log/dmesg; then
-					pstatus green YES 'Xen PV is not vulnerable'
-					xen_pv=1
-				elif [ -r /var/log/kern.log ] && grep -q 'Booting paravirtualized kernel on Xen$' /var/log/kern.log; then
 					pstatus green YES 'Xen PV is not vulnerable'
 					xen_pv=1
+				elif [ $ret -eq 2 ]; then
+					pstatus yellow UNKNOWN "dmesg truncated, please reboot and relaunch this script"
 				else
 					pstatus blue NO
 				fi