detect retpoline-compliant compiler from latest LKML patches

Remove superfluous 'YES' output when checking cpuinfo

README.md

@@ -28,7 +28,7 @@ Example of the output of the script:
 
 ```
 $ sudo ./spectre-meltdown-checker.sh
-Spectre and Meltdown mitigation detection tool v0.02
+Spectre and Meltdown mitigation detection tool v0.07
 
 CVE-2017-5753 [bounds check bypass] aka 'Spectre Variant 1'
 * Kernel compiled with LFENCE opcode inserted at the proper places:  NO  (only 38 opcodes found, should be >= 60)
@@ -41,8 +41,8 @@ CVE-2017-5715 [branch target injection] aka 'Spectre Variant 2'
 *   IBRS enabled for Kernel space:  NO
 *   IBRS enabled for User space:  NO
 * Mitigation 2
-*   Kernel recompiled with retpolines: UNKNOWN (check not yet implemented)
-> STATUS: VULNERABLE (IBRS hardware + kernel support OR retpolines-compiled kernel are needed to mitigate the vulnerability)
+*   Kernel compiled with retpolines:  NO
+> STATUS:  VULNERABLE  (IBRS hardware + kernel support OR kernel with retpolines are needed to mitigate the vulnerability)
 
 CVE-2017-5754 [rogue data cache load] aka 'Meltdown' aka 'Variant 3'
 * Kernel supports Page Table Isolation (PTI):  YES

spectre-meltdown-checker.sh

@@ -1,8 +1,9 @@
 #! /bin/sh
 # Spectre & Meltdown checker
 # Stephane Lesimple
-VERSION=0.04
+VERSION=0.15
 
+# print status function
 pstatus()
 {
 	case "$1" in
@@ -16,6 +17,11 @@ pstatus()
 	/bin/echo
 }
 
+# The 3 below functions are taken from the extract-linux script, available here:
+# https://github.com/torvalds/linux/blob/master/scripts/extract-vmlinux
+# The functions have been modified for better integration to this script
+# The original header of the file has been retained below
+
 # ----------------------------------------------------------------------
 # extract-vmlinux - Extract uncompressed vmlinux from a kernel image
 #
@@ -39,63 +45,90 @@ try_decompress()
         # "grep" that report the byte offset of the line instead of the pattern.
 
         # Try to find the header ($1) and decompress from here
-        for     pos in `tr "$1\n$2" "\n$2=" < "$img" | grep -abo "^$2"`
+        for     pos in `tr "$1\n$2" "\n$2=" < "$4" | grep -abo "^$2"`
         do
                 pos=${pos%%:*}
-                tail -c+$pos "$img" | $3 > $vmlinuxtmp 2> /dev/null
-                check_vmlinux $vmlinuxtmp && echo $vmlinuxtmp || rm -f $vmlinuxtmp
+                tail -c+$pos "$4" | $3 > $vmlinuxtmp 2> /dev/null
+                check_vmlinux "$vmlinuxtmp" && echo "$vmlinuxtmp" && return 0
         done
+	return 1
 }
 
 extract_vmlinux()
 {
-	img="$1"
-
+	[ -n "$1" ] || return 1
 	# Prepare temp files:
-	vmlinuxtmp=$(mktemp /tmp/vmlinux-XXX)
+	vmlinuxtmp="$(mktemp /tmp/vmlinux-XXX)"
 
 	# Initial attempt for uncompressed images or objects:
-	check_vmlinux $img
+	if check_vmlinux "$1"; then
+		cat "$1" > "$vmlinuxtmp"
+		echo "$vmlinuxtmp"
+		return 0
+	fi
 
 	# That didn't work, so retry after decompression.
-	try_decompress '\037\213\010' xy    gunzip     || \
-	try_decompress '\3757zXZ\000' abcde unxz       || \
-	try_decompress 'BZh'          xy    bunzip2    || \
-	try_decompress '\135\0\0\0'   xxx   unlzma     || \
-	try_decompress '\211\114\132' xy    'lzop -d'
+	try_decompress '\037\213\010' xy    gunzip     "$1" && return 0
+	try_decompress '\3757zXZ\000' abcde unxz       "$1" && return 0
+	try_decompress 'BZh'          xy    bunzip2    "$1" && return 0
+	try_decompress '\135\0\0\0'   xxx   unlzma     "$1" && return 0
+	try_decompress '\211\114\132' xy    'lzop -d'  "$1" && return 0
+	return 1
 }
 
+# end of extract-vmlinux functions
 
-/bin/echo "Spectre and Meltdown mitigation detection tool v$VERSION"
+/bin/echo -e "\033[1;34mSpectre and Meltdown mitigation detection tool v$VERSION\033[0m"
 /bin/echo
 
+# root check
+
+if [ "$(id -u)" -ne 0 ]; then
+	/bin/echo -e "\033[31mNote that you should launch this script with root privileges to get accurate information.\033[0m"
+	/bin/echo -e "\033[31mWe'll proceed but you might see permission denied errors.\033[0m"
+	/bin/echo -e "\033[31mTo run it as root, you can try the following command: sudo $0\033[0m"
+	/bin/echo
+fi
+
+/bin/echo -e "Checking vulnerabilities against \033[35m"$(uname -s) $(uname -r) $(uname -v) $(uname -m)"\033[0m"
+/bin/echo
+
+###########
 # SPECTRE 1
 /bin/echo -e "\033[1;34mCVE-2017-5753 [bounds check bypass] aka 'Spectre Variant 1'\033[0m"
 /bin/echo -n "* Kernel compiled with LFENCE opcode inserted at the proper places: "
 
 status=0
 img=''
+# try to find the image of the current running kernel
+[ -e /boot/vmlinuz-linux       ] && img=/boot/vmlinuz-linux
 [ -e /boot/vmlinuz-$(uname -r) ] && img=/boot/vmlinuz-$(uname -r)
-[ -e /boot/vmlinux-$(uname -r) ] && img=/boot/vmlinux-$(uname -r)
+[ -e /boot/kernel-$( uname -r) ] && img=/boot/kernel-$( uname -r)
 [ -e /boot/bzImage-$(uname -r) ] && img=/boot/bzImage-$(uname -r)
+[ -e /boot/kernel-genkernel-$(uname -m)-$(uname -r) ] && img=/boot/kernel-genkernel-$(uname -m)-$(uname -r)
 if [ -z "$img" ]; then
-	pstatus yellow UNKNOWN "couldn't find your kernel image in /boot"
+	pstatus yellow UNKNOWN "couldn't find your kernel image in /boot, if you used netboot, this is normal"
 else
 	vmlinux=$(extract_vmlinux $img)
 	if [ -z "$vmlinux" -o ! -r "$vmlinux" ]; then
-		pstatus yellow UNKNOWN "couldn't extract your kernel"
+		pstatus yellow UNKNOWN "couldn't extract your kernel from $img"
 	elif ! which objdump >/dev/null 2>&1; then
 		pstatus yellow UNKNOWN "missing 'objdump' tool, please install it, usually it's in the binutils package"
 	else
+		# here we disassemble the kernel and count the number of occurences of the LFENCE opcode
+		# in non-patched kernels, this has been empirically determined as being around 40-50
+		# in patched kernels, this is more around 70-80, sometimes way higher (100+)
+		# v0.13: 68 found in a 3.10.23-xxxx-std-ipv6-64 (with lots of modules compiled-in directly), which doesn't have the LFENCE patches,
+		# so let's push the threshold to 70.
+		# TODO LKML patch is starting to dump LFENCE in favor of the PAUSE opcode, we might need to check that (patch not stabilized yet)
 		nb_lfence=$(objdump -D "$vmlinux" | grep -wc lfence)
-		if [ "$nb_lfence" -lt 60 ]; then
-			pstatus red NO "only $nb_lfence opcodes found, should be >= 60"
+		if [ "$nb_lfence" -lt 70 ]; then
+			pstatus red NO "only $nb_lfence opcodes found, should be >= 70"
 			status=1
 		else
-			pstatus green YES "$nb_lfence opcodes found, which is >= 60"
+			pstatus green YES "$nb_lfence opcodes found, which is >= 70"
 			status=2
 		fi
-		rm -f $vmlinux
 	fi
 fi
 
@@ -104,45 +137,58 @@ fi
 [ "$status" = 1 ] && pstatus red VULNERABLE
 [ "$status" = 2 ] && pstatus green 'NOT VULNERABLE'
 
-
+###########
 # VARIANT 2
 /bin/echo
 /bin/echo -e "\033[1;34mCVE-2017-5715 [branch target injection] aka 'Spectre Variant 2'\033[0m"
 /bin/echo "* Mitigation 1"
 /bin/echo -n "*   Hardware (CPU microcode) support for mitigation: "
 if [ ! -e /dev/cpu/0/msr ]; then
+	# try to load the module ourselves (and remember it so we can rmmod it afterwards)
 	modprobe msr 2>/dev/null && insmod_msr=1
 fi
 if [ ! -e /dev/cpu/0/msr ]; then
 	pstatus yellow UNKNOWN "couldn't read /dev/cpu/0/msr, is msr support enabled in your kernel?"
 else
-	dd if=/dev/cpu/0/msr of=/dev/null bs=1 count=8 skip=72 2>/dev/null
+	# the new MSR 'SPEC_CTRL' is at offset 0x48
+	# here we use dd, it's the same as using 'rdmsr 0x48' but without needing the rdmsr tool
+	# if we get a read error, the MSR is not there
+	dd if=/dev/cpu/0/msr of=/dev/null bs=8 count=1 skip=9 2>/dev/null
 	if [ $? -eq 0 ]; then
 		pstatus green YES
 	else
 		pstatus red NO
 	fi
-	#dd if=/dev/cpu/0/msr of=/dev/null bs=1 count=8 skip=73 2>/dev/null
-	#/bin/echo $?
 fi
 
 if [ "$insmod_msr" = 1 ]; then
+	# if we used modprobe ourselves, rmmod the module
 	rmmod msr 2>/dev/null
 fi
 
 /bin/echo -n "*   Kernel support for IBRS: "
-if [ -e /sys/kernel/debug/sched_features ]; then
+if [ ! -e /sys/kernel/debug/sched_features ]; then
+	# try to mount the debugfs hierarchy ourselves and remember it to umount afterwards
 	mount -t debugfs debugfs /sys/kernel/debug 2>/dev/null && mounted_debugfs=1
 fi
 if [ -e /sys/kernel/debug/ibrs_enabled ]; then
+	# if the file is there, we have IBRS compiled-in
 	pstatus green YES
 	ibrs_supported=1
+	ibrs_enabled=$(cat /sys/kernel/debug/ibrs_enabled 2>/dev/null)
+elif [ -e /sys/kernel/debug/x86/ibrs_enabled ]; then
+	# RedHat uses a different path (see https://access.redhat.com/articles/3311301)
+	pstatus green YES
+	ibrs_supported=1
+	ibrs_enabled=$(cat /sys/kernel/debug/x86/ibrs_enabled 2>/dev/null)
 else
 	pstatus red NO
 fi
 
-ibrs_enabled=$(cat /sys/kernel/debug/ibrs_enabled 2>/dev/null)
 /bin/echo -n "*   IBRS enabled for Kernel space: "
+# 0 means disabled
+# 1 is enabled only for kernel space
+# 2 is enabled for kernel and user space
 case "$ibrs_enabled" in
 	"") [ "$ibrs_supported" = 1 ] && pstatus yellow UNKNOWN || pstatus red NO;;
 	0)     pstatus red NO;;
@@ -155,17 +201,14 @@ case "$ibrs_enabled" in
 	"") [ "$ibrs_supported" = 1 ] && pstatus yellow UNKNOWN || pstatus red NO;;
 	0 | 1) pstatus red NO;;
 	2) pstatus green YES;;
-	*) pstatus yellow unknown;;
+	*) pstatus yellow UNKNOWN;;
 esac
 
-if [ "$mounted_debugfs" = 1 ]; then
-	umount /sys/kernel/debug
-fi
-
 /bin/echo "* Mitigation 2"
-/bin/echo -n "*   Kernel recompiled with retpolines: "
-# XXX this doesn't mean the kernel has been compiled with a retpoline-aware gcc
+/bin/echo -n "*   Kernel compiled with retpoline option: "
+# We check the RETPOLINE kernel options
 if [ -e /proc/config.gz ]; then
+	# either the running kernel exports his own config
 	if zgrep -q '^CONFIG_RETPOLINE=y' /proc/config.gz; then
 		pstatus green YES
 		retpoline=1
@@ -173,65 +216,133 @@ if [ -e /proc/config.gz ]; then
 		pstatus red NO
 	fi
 elif [ -e /boot/config-$(uname -r) ]; then
+	# or we can find a config file in /root with the kernel release name
 	if grep  -q '^CONFIG_RETPOLINE=y' /boot/config-$(uname -r); then
 		pstatus green YES
 		retpoline=1
 	else
 		pstatus red NO
 	fi
+else
+	pstatus yellow UNKNOWN "couldn't read your kernel configuration"
 fi
 
+/bin/echo -n "*   Kernel compiled with a retpoline-aware compiler: "
+# Now check if the compiler used to compile the kernel knows how to insert retpolines in generated asm
+# For gcc, this is -mindirect-branch=thunk-extern (detected by the kernel makefiles)
+# See gcc commit https://github.com/hjl-tools/gcc/commit/23b517d4a67c02d3ef80b6109218f2aadad7bd79
+# In latest retpoline LKML patches, the noretpoline_setup symbol exists only if CONFIG_RETPOLINE is set
+# *AND* if the compiler is retpoline-compliant, so look for that symbol
+if [ -n "$vmlinux" ]; then
+	# look for the symbol
+	if [ -e /boot/System.map-$(uname -r) ]; then
+		if grep -qw noretpoline_setup /boot/System.map-$(uname -r); then
+			retpoline_compiler=1
+			pstatus green YES "noretpoline_setup symbol found in System.map"
+		fi
+	elif which nm >/dev/null 2>&1; then
+		# the proper way: use nm and look for the symbol
+		if nm "$vmlinux" 2>/dev/null | grep -qw 'noretpoline_setup'; then
+			retpoline_compiler=1
+			pstatus green YES "noretpoline_setup symbol found in vmlinux"
+		fi
+	elif grep -q noretpoline_setup "$vmlinux"; then
+		# if we don't have nm, nevermind, the symbol name is long enough to not have
+		# any false positive using good old grep directly on the binary
+		retpoline_compiler=1
+		pstatus green YES "noretpoline_setup symbol found in vmlinux"
+	fi
+	if [ "$retpoline_compiler" != 1 ]; then
+		pstatus red NO
+	fi
+else
+	pstatus yellow UNKNOWN "couldn't find your kernel image"
+fi
+
+
 /bin/echo -ne "> \033[46m\033[30mSTATUS:\033[0m "
 if grep -q AMD /proc/cpuinfo; then
 	pstatus green "NOT VULNERABLE" "your CPU is not vulnerable as per the vendor"
 elif [ "$ibrs_enabled" = 1 -o "$ibrs_enabled" = 2 ]; then
 	pstatus green "NOT VULNERABLE" "IBRS mitigates the vulnerability"
-elif [ "$retpoline" = 1 ]; then
-	pstatus green "NOT VULNERABLE" "retpolines mitigate the vulnerability"
+elif [ "$retpoline" = 1 -a "$retpoline_compiler" = 1 ]; then
+	pstatus green "NOT VULNERABLE" "retpoline mitigate the vulnerability"
 else
-	pstatus red VULNERABLE "IBRS hardware + kernel support OR kernel with retpolines kernel are needed to mitigate the vulnerability"
+	pstatus red VULNERABLE "IBRS hardware + kernel support OR kernel with retpoline are needed to mitigate the vulnerability"
 fi
 
+##########
 # MELTDOWN
 /bin/echo
 /bin/echo -e "\033[1;34mCVE-2017-5754 [rogue data cache load] aka 'Meltdown' aka 'Variant 3'\033[0m"
 /bin/echo -n "* Kernel supports Page Table Isolation (PTI): "
+kpti_support=0
+kpti_can_tell=0
 if [ -e /proc/config.gz ]; then
-	if zgrep -q '^CONFIG_PAGE_TABLE_ISOLATION=y' /proc/config.gz; then
-		pstatus green YES
+	# either the running kernel exports his own config
+	kpti_can_tell=1
+	if zgrep -q '^\(CONFIG_PAGE_TABLE_ISOLATION=y\|CONFIG_KAISER=y\)' /proc/config.gz; then
 		kpti_support=1
-	else
-		pstatus red NO
 	fi
 elif [ -e /boot/config-$(uname -r) ]; then
-	if grep  -q '^CONFIG_PAGE_TABLE_ISOLATION=y' /boot/config-$(uname -r); then
-		pstatus green YES
+	# or we can find a config file in /root with the kernel release name
+	kpti_can_tell=1
+	if grep  -q '^\(CONFIG_PAGE_TABLE_ISOLATION=y\|CONFIG_KAISER=y\)' /boot/config-$(uname -r); then
 		kpti_support=1
-	else
-		pstatus red NO
 	fi
-elif [ -e /boot/System.map-$(uname -r) ]; then
+fi
+if [ -e /boot/System.map-$(uname -r) ]; then
+	# it's not an elif: some backports don't have the PTI config but still include the patch
+	# so we try to find an exported symbol that is part of the PTI patch in System.map
+	kpti_can_tell=1
 	if grep -qw kpti_force_enabled /boot/System.map-$(uname -r); then
-		pstatus green YES
 		kpti_support=1
-	else
-		pstatus red NO
 	fi
+fi
+if [ -n "$vmlinux" ]; then
+	# same as above but in case we don't have System.map and only vmlinux, look for the
+	# nopti option that is part of the patch (kernel command line option)
+	kpti_can_tell=1
+	if strings "$vmlinux" | grep -qw nopti; then
+		kpti_support=1
+	fi
+fi
+
+if [ "$kpti_support" = 1 ]; then
+	pstatus green YES
+elif [ "$kpti_can_tell" = 1 ]; then
+	pstatus red NO
 else
-	pstatus yellow UNKNOWN
+	pstatus yellow UNKNOWN "couldn't read your kernel configuration"
 fi
 
 /bin/echo -n "* PTI enabled and active: "
 if grep ^flags /proc/cpuinfo | grep -qw pti; then
-	pstatus green YES
+	# vanilla PTI patch sets the 'pti' flag in cpuinfo
 	kpti_enabled=1
-elif dmesg | grep -q 'Kernel/User page tables isolation: enabled'; then
-	pstatus green YES
+elif grep ^flags /proc/cpuinfo | grep -qw kaiser; then
+	# kernel line 4.9 sets the 'kaiser' flag in cpuinfo
+	kpti_enabled=1
+elif [ -e /sys/kernel/debug/x86/pti_enabled ]; then
+	# RedHat Backport creates a dedicated file, see https://access.redhat.com/articles/3311301
+	kpti_enabled=$(cat /sys/kernel/debug/x86/pti_enabled 2>/dev/null)
+elif dmesg | grep -Eq 'Kernel/User page tables isolation: enabled|Kernel page table isolation enabled'; then
+	# if we can't find the flag, grep in dmesg
 	kpti_enabled=1
+else
+	kpti_enabled=0
+fi
+if [ "$kpti_enabled" = 1 ]; then
+	pstatus green YES
 else
 	pstatus red NO
 fi
 
+if [ "$mounted_debugfs" = 1 ]; then
+	# umount debugfs if we did mount it ourselves
+	umount /sys/kernel/debug
+fi
+
 /bin/echo -ne "> \033[46m\033[30mSTATUS:\033[0m "
 if grep -q AMD /proc/cpuinfo; then
 	pstatus green "NOT VULNERABLE" "your CPU is not vulnerable as per the vendor"
@@ -241,10 +352,6 @@ else
 	pstatus red "VULNERABLE" "PTI is needed to mitigate the vulnerability"
 fi
 
-
 /bin/echo
-if [ "$USER" != root ]; then
-	/bin/echo "Note that you should launch this script with root privileges to get accurate information"
-	/bin/echo "You can try the following command: sudo $0"
-fi
 
+[ -n "$vmlinux" -a -f "$vmlinux" ] && rm -f "$vmlinux"