openssl/util/shlib_wrap.sh.in
Richard Levitte 1828939974 Fix rpath-related Linux "test_shlibload" failure.
When libssl and libcrypto are compiled on Linux with "-rpath", but
not "--enable-new-dtags", the RPATH takes precedence over
LD_LIBRARY_PATH, and we end up running with the wrong libraries.
This is resolved by using full (or at least relative, rather than
just the filename to be found on LD_LIBRARY_PATH) paths to the
shared objects.

Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7626)
2018-11-14 00:41:57 +01:00

138 lines
5.1 KiB
Bash
Executable file

#!/bin/sh
{-
use lib '.';
use configdata;
sub shlib {
my $lib = shift;
return "" if $disabled{shared};
$lib = $unified_info{rename}->{$lib}
if defined $unified_info{rename}->{$lib};
$lib = $unified_info{sharednames}->{$lib}
. ($target{shlib_variant} || "")
. ($target{shared_extension} || ".so");
$lib =~ s|\.\$\(SHLIB_VERSION_NUMBER\)
|.$config{shlib_version_number}|x;
return $lib;
}
""; # Make sure no left over string sneaks its way into the script
-}
# To test this OpenSSL version's applications against another version's
# shared libraries, simply set
#
# OPENSSL_REGRESSION=/path/to/other/OpenSSL/build/tree
if [ -n "$OPENSSL_REGRESSION" ]; then
shlibwrap="$OPENSSL_REGRESSION/util/shlib_wrap.sh"
if [ -x "$shlibwrap" ]; then
# We clear OPENSSL_REGRESSION to avoid a loop, should the shlib_wrap.sh
# we exec also support that mechanism...
OPENSSL_REGRESSION= exec "$shlibwrap" "$@"
else
if [ -f "$shlibwrap" ]; then
echo "Not permitted to run $shlibwrap" >&2
else
echo "No $shlibwrap, perhaps OPENSSL_REGRESSION isn't properly set?" >&2
fi
exit 1
fi
fi
[ $# -ne 0 ] || set -x # debug mode without arguments:-)
THERE="`echo $0 | sed -e 's|[^/]*$||' 2>/dev/null`.."
[ -d "${THERE}" ] || exec "$@" # should never happen...
LIBCRYPTOSO="${THERE}/{- shlib('libcrypto') -}"
LIBSSLSO="${THERE}/{- shlib('libssl') -}"
SYSNAME=`(uname -s) 2>/dev/null`;
case "$SYSNAME" in
SunOS|IRIX*)
# SunOS and IRIX run-time linkers evaluate alternative
# variables depending on target ABI...
rld_var=LD_LIBRARY_PATH
case "`(/usr/bin/file "$LIBCRYPTOSO") 2>/dev/null`" in
*ELF\ 64*SPARC*|*ELF\ 64*AMD64*)
[ -n "$LD_LIBRARY_PATH_64" ] && rld_var=LD_LIBRARY_PATH_64
LD_PRELOAD_64="$LIBCRYPTOSO $LIBSSLSO"; export LD_PRELOAD_64
preload_var=LD_PRELOAD_64
;;
*ELF\ 32*SPARC*|*ELF\ 32*80386*)
# We only need to change LD_PRELOAD_32 and LD_LIBRARY_PATH_32
# on a multi-arch system. Otherwise, trust the fallbacks.
if [ -f /lib/64/ld.so.1 ]; then
[ -n "$LD_LIBRARY_PATH_32" ] && rld_var=LD_LIBRARY_PATH_32
LD_PRELOAD_32="$LIBCRYPTOSO $LIBSSLSO"; export LD_PRELOAD_32
preload_var=LD_PRELOAD_32
fi
;;
# Why are newly built .so's preloaded anyway? Because run-time
# .so lookup path embedded into application takes precedence
# over LD_LIBRARY_PATH and as result application ends up linking
# to previously installed .so's. On IRIX instead of preloading
# newly built .so's we trick run-time linker to fail to find
# the installed .so by setting _RLD_ROOT variable.
*ELF\ 32*MIPS*)
#_RLD_LIST="$LIBCRYPTOSO:$LIBSSLSO:DEFAULT"; export _RLD_LIST
_RLD_ROOT=/no/such/dir; export _RLD_ROOT
eval $rld_var=\"/usr/lib'${'$rld_var':+:$'$rld_var'}'\"
preload_var=_RLD_LIST
;;
*ELF\ N32*MIPS*)
[ -n "$LD_LIBRARYN32_PATH" ] && rld_var=LD_LIBRARYN32_PATH
#_RLDN32_LIST="$LIBCRYPTOSO:$LIBSSLSO:DEFAULT"; export _RLDN32_LIST
_RLDN32_ROOT=/no/such/dir; export _RLDN32_ROOT
eval $rld_var=\"/usr/lib32'${'$rld_var':+:$'$rld_var'}'\"
preload_var=_RLDN32_LIST
;;
*ELF\ 64*MIPS*)
[ -n "$LD_LIBRARY64_PATH" ] && rld_var=LD_LIBRARY64_PATH
#_RLD64_LIST="$LIBCRYPTOSO:$LIBSSLSO:DEFAULT"; export _RLD64_LIST
_RLD64_ROOT=/no/such/dir; export _RLD64_ROOT
eval $rld_var=\"/usr/lib64'${'$rld_var':+:$'$rld_var'}'\"
preload_var=_RLD64_LIST
;;
esac
eval $rld_var=\"${THERE}'${'$rld_var':+:$'$rld_var'}'\"; export $rld_var
unset rld_var
;;
*) LD_LIBRARY_PATH="${THERE}:$LD_LIBRARY_PATH" # Linux, ELF HP-UX
DYLD_LIBRARY_PATH="${THERE}:$DYLD_LIBRARY_PATH" # MacOS X
SHLIB_PATH="${THERE}:$SHLIB_PATH" # legacy HP-UX
LIBPATH="${THERE}:$LIBPATH" # AIX, OS/2
export LD_LIBRARY_PATH DYLD_LIBRARY_PATH SHLIB_PATH LIBPATH
# Even though $PATH is adjusted [for Windows sake], it doesn't
# necessarily does the trick. Trouble is that with introduction
# of SafeDllSearchMode in XP/2003 it's more appropriate to copy
# .DLLs in vicinity of executable, which is done elsewhere...
if [ "$OSTYPE" != msdosdjgpp ]; then
PATH="${THERE}:$PATH"; export PATH
fi
;;
esac
{- output_off() unless grep (/-rpath\b/, @{$config{LDFLAGS}}); ""; -}
if [ -f "$LIBCRYPTOSO" -a -z "$preload_var" ]; then
# Following three lines are major excuse for isolating them into
# this wrapper script. Original reason for setting LD_PRELOAD
# was to make it possible to pass 'make test' when user linked
# with -rpath pointing to previous version installation. Wrapping
# it into a script makes it possible to do so on multi-ABI
# platforms.
case "$SYSNAME" in
*BSD) LD_PRELOAD="$LIBCRYPTOSO:$LIBSSLSO" ;; # *BSD
*) LD_PRELOAD="$LIBCRYPTOSO $LIBSSLSO" ;; # SunOS, Linux, ELF HP-UX
esac
_RLD_LIST="$LIBCRYPTOSO:$LIBSSLSO:DEFAULT" # Tru64, o32 IRIX
DYLD_INSERT_LIBRARIES="$LIBCRYPTOSO:$LIBSSLSO" # MacOS X
export LD_PRELOAD _RLD_LIST DYLD_INSERT_LIBRARIES
fi
{- output_on() unless grep (/-rpath\b/, @{$config{LDFLAGS}}); ""; -}
cmd="$1"; [ -x "$cmd" ] || cmd="$cmd${EXE_EXT}"
shift
if [ $# -eq 0 ]; then
exec "$cmd" # old sh, such as Tru64 4.x, fails to expand empty "$@"
else
exec "$cmd" "$@"
fi