kpatch: support eu-readelf as alternative to GNU readelf

[stephenchengCloud]

The kpatch script is used to load live patches on production systems that may be stripped down, with binutils (which provides readelf) not installed while elfutils (which provides eu-readelf) is available. Add a get_module_section_string() helper that auto-detects the available tool at startup and uses the appropriate invocation for each.

Other readelf usages in the build tooling (kpatch-build, lookup.c, test infrastructure) are left unchanged, as a compiler toolchain including binutils can reasonably be expected in a build environment.

[stephenchengCloud]

Tested:
get module name:

[root@eu1-dt021 ~]#  eu-readelf --string-dump=.gnu.linkonce.this_module lp_6_6_98__23_xs9__6_6_98__24_xs9.ko | awk 'NF>=3 && /^\s*\[/ {print $3; exit}'
lp_6_6_98__23_xs9__6_6_98__24_xs9
[root@eu1-dt021 ~]#
[root@eu1-dt021 ~]# readelf -p .gnu.linkonce.this_module lp_6_6_98__23_xs9__6_6_98__24_xs9.ko | grep '\[.*\]' | awk '{print $3}'
lp_6_6_98__23_xs9__6_6_98__24_xs9

get checksum:

[root@eu1-dt021 ~]# readelf -p .kpatch.checksum test_module_with_checksum.ko | grep '\[.*\]' | awk '{print $3}'
d41d8cd98f00b204e9800998ecf8427e
[root@eu1-dt021 ~]#  eu-readelf --string-dump=.kpatch.checksum test_module_with_checksum.ko | awk 'NF>=3 && /^\s*\[/ {print $3; exit}'
d41d8cd98f00b204e9800998ecf8427e
[root@eu1-dt021 ~]#

empty checksum:

stephenche@40 kpatch (v0.9.9) $ readelf -p .kpatch.checksum lp_6_6_98__23_xs9__6_6_98__24_xs9.ko 2>/dev/null | grep '\[.*\]' | awk '{print $3}'
stephenche@40 kpatch (v0.9.9) $ eu-readelf --string-dump=.kpatch.checksum lp_6_6_98__23_xs9__6_6_98__24_xs9.ko 2>/dev/nul

[joe-lawrence]

Hi @stephenchengCloud , sorry for the delay in reviewing. The change seems reasonable, I have a few small suggestions:

1- Copy your awk one-liner to extract the string to the readelf case so they're the same
2a- Check for eu-readelf in the path as well
3a- Add an else case to catch if neither tool is installed
3b- The else case returns 1, then
3c- The callers propogate/check for (3b) failure
4- Move get_module_section_string down to just above get_module_name (not shown here):

diff --git a/kpatch/kpatch b/kpatch/kpatch
index d75bd068356d..347f89230751 100755
--- a/kpatch/kpatch
+++ b/kpatch/kpatch
@@ -27,12 +27,16 @@ INSTALLDIR=/var/lib/kpatch
 SCRIPTDIR="$(readlink -f "$(dirname "$(type -p "$0")")")"
 if command -v readelf >/dev/null 2>&1; then
 	get_module_section_string() {
-		readelf -p "$2" "$1" | grep '\[.*\]' | awk '{print $3}'
+		readelf -p "$2" "$1" | awk 'NF>=3 && /^\s*\[/ {print $3; exit}'
 	}
-else
+elif command -v eu-readelf >/dev/null 2>&1; then
 	get_module_section_string() {
 		eu-readelf --string-dump="$2" "$1" | awk 'NF>=3 && /^\s*\[/ {print $3; exit}'
 	}
+else
+	get_module_section_string() {
+		return 1
+	}
 fi
 VERSION="0.9.11"
 POST_ENABLE_WAIT=15	# seconds
@@ -165,10 +169,10 @@ init_sysfs_var() {
 }
 
 verify_module_checksum () {
-	modname="$(get_module_name "$1")"
+	modname="$(get_module_name "$1")" || return 1
 	[[ -z "$modname" ]] && return 1
 
-	checksum="$(get_module_section_string "$1" .kpatch.checksum 2>/dev/null)"
+	checksum="$(get_module_section_string "$1" .kpatch.checksum 2>/dev/null)" || return 1
 
 	# Fail checksum match only if both exist and diverge
 	if [[ -n "$checksum" ]] && [[ -e "$SYSFS/${modname}/checksum" ]] ; then
@@ -325,7 +329,8 @@ load_module () {
 	fi
 
 	local modname
-	modname="$(get_module_name "$module")"
+	modname="$(get_module_name "$module")" || die "failed to get module name from $module"
+	[[ -z "$modname" ]] && die "failed to get module name from $module"
 	local moddir="$SYSFS/$modname"
 	if [[ -d "$moddir" ]] ; then
 		if [[ "$(cat "${moddir}/enabled")" -eq 0 ]]; then

^^ is untested, but hopefully clarifies the suggestions.

[stephenchengCloud]

Hi @joe-lawrence, thank you for your detailed suggestions.
I've updated the code and done a simple test on the awk change:

$ readelf -p .gnu.linkonce.this_module lp_6_6_98__23_xs9__6_6_98__24_xs9.ko | awk 'NF>=3 && /^\s*\[/ {print $3; exit}'
lp_6_6_98__23_xs9__6_6_98__24_xs9

[joe-lawrence]

LGTM, thanks @stephenchengCloud !