[PATCH] ld.so: add an --argv0 option

[PATCH] ld.so: add an --argv0 option

[Vincent Mihalkovic]

[-- Attachment #1: Type: text/plain, Size: 218 bytes --]

Hi,
a few years ago there was an effort for adding --argv0 option:
https://sourceware.org/legacy-ml/libc-alpha/2016-04/msg00576.html.
I made the old patch actual for version 2.31.9000, added a test case and
changelog.

[-- Attachment #2: changelog.txt --]
[-- Type: text/plain, Size: 240 bytes --]

2020-07-22  Vincent Mihalkovic  <vmihalko@redhat.com>

	* elf/Makefile: added argv0 test case, Modified.
	* elf/rtld.c: added --argv0 option, Modified.
	* elf/argv0test.c: test case, New file.
	* elf/tst-rtld-argv0.sh: test case, New file.

[-- Attachment #3: argv0.patch --]
[-- Type: text/x-patch, Size: 4788 bytes --]

diff --git a/elf/Makefile b/elf/Makefile
index 0b78721848..f38904d831 100644
--- a/elf/Makefile
+++ b/elf/Makefile
@@ -210,7 +210,8 @@ tests += restest1 preloadtest loadfail multiload origtest resolvfail \
 	 tst-filterobj tst-filterobj-dlopen tst-auxobj tst-auxobj-dlopen \
 	 tst-audit14 tst-audit15 tst-audit16 \
 	 tst-single_threaded tst-single_threaded-pthread \
-	 tst-tls-ie tst-tls-ie-dlmopen
+	 tst-tls-ie tst-tls-ie-dlmopen \
+	 argv0test
 #	 reldep9
 tests-internal += loadtest unload unload2 circleload1 \
 	 neededtest neededtest2 neededtest3 neededtest4 \
@@ -414,7 +415,7 @@ endif
 ifeq (yes,$(build-shared))
 ifeq ($(run-built-tests),yes)
 tests-special += $(objpfx)tst-pathopt.out $(objpfx)tst-rtld-load-self.out \
-		 $(objpfx)tst-rtld-preload.out
+		 $(objpfx)tst-rtld-preload.out $(objpfx)argv0test.out
 endif
 tests-special += $(objpfx)check-textrel.out $(objpfx)check-execstack.out \
 		 $(objpfx)check-wx-segment.out \
@@ -1796,3 +1797,11 @@ $(objpfx)tst-tls-ie-dlmopen.out: \
   $(objpfx)tst-tls-ie-mod6.so
 
 $(objpfx)tst-tls-surplus: $(libdl)
+
+ARGV0 = test-argv0
+$(objpfx)argv0test.out: tst-rtld-argv0.sh $(objpfx)ld.so \
+			$(objpfx)argv0test
+	$(SHELL) $< $(objpfx)ld.so $(objpfx)argv0test \
+            '$(test-wrapper-env)' '$(run_program_env)' \
+            '$(rpath-link)' '$(ARGV0)' > $@; \
+    $(evaluate-test)
diff --git a/elf/argv0test.c b/elf/argv0test.c
new file mode 100644
index 0000000000..4c79bebf23
--- /dev/null
+++ b/elf/argv0test.c
@@ -0,0 +1,13 @@
+#include <stdio.h>   // for printf
+#include <string.h>  // for strcmp
+
+
+int main( int argc, char *argv[] ) {
+	int result = strcmp( argv[0], "test-argv0");
+ 
+	printf ("argv[0] = %s, strcmp( argv[0], \"test-argv0\" ) = %d, %s\n", \
+					argv[0], result, !result ? "ok" : "wrong");
+
+	return result;
+
+}
diff --git a/elf/rtld.c b/elf/rtld.c
index 5b882163fa..cafa4f9bd3 100644
--- a/elf/rtld.c
+++ b/elf/rtld.c
@@ -1202,6 +1202,8 @@ dl_main (const ElfW(Phdr) *phdr,
 	 installing it.  */
       rtld_is_main = true;
 
+      char *argv0 = NULL;
+
       /* Note the place where the dynamic linker actually came from.  */
       GL(dl_rtld_map).l_name = rtld_progname;
 
@@ -1259,6 +1261,14 @@ dl_main (const ElfW(Phdr) *phdr,
 	else if (! strcmp (_dl_argv[1], "--preload") && _dl_argc > 2)
 	  {
 	    preloadarg = _dl_argv[2];
+	    _dl_skip_args += 2;
+	    _dl_argc -= 2;
+	    _dl_argv += 2;
+	  }
+	else if (! strcmp (_dl_argv[1], "--argv0") && _dl_argc > 2)
+	  {
+	    argv0 = _dl_argv[2];
+
 	    _dl_skip_args += 2;
 	    _dl_argc -= 2;
 	    _dl_argv += 2;
@@ -1292,7 +1302,8 @@ of this helper program; chances are you did not intend to run this program.\n\
   --inhibit-rpath LIST  ignore RUNPATH and RPATH information in object names\n\
 			in LIST\n\
   --audit LIST          use objects named in LIST as auditors\n\
-  --preload LIST        preload objects named in LIST\n");
+  --preload LIST        preload objects named in LIST\n\
+  --argv0 STRING        set argv[0] to STRING before running\n");
 
       ++_dl_skip_args;
       --_dl_argc;
@@ -1384,6 +1395,11 @@ of this helper program; chances are you did not intend to run this program.\n\
 	    break;
 	  }
 #endif
+
+      /* Set the argv[0] string now that we've processed the executable.  */
+      if (argv0 != NULL) {
+		_dl_argv[0] = argv0;
+	  }
     }
   else
     {
diff --git a/elf/tst-rtld-argv0.sh b/elf/tst-rtld-argv0.sh
new file mode 100755
index 0000000000..5f873b6c5c
--- /dev/null
+++ b/elf/tst-rtld-argv0.sh
@@ -0,0 +1,37 @@
+#!/bin/sh
+# Test --argv0 argument ld.so.
+# Copyright (C) 2019-2020 Free Software Foundation, Inc.
+# This file is part of the GNU C Library.
+#
+# The GNU C Library is free software; you can redistribute it and/or
+# modify it under the terms of the GNU Lesser General Public
+# License as published by the Free Software Foundation; either
+# version 2.1 of the License, or (at your option) any later version.
+#
+# The GNU C Library is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+# Lesser General Public License for more details.
+#
+# You should have received a copy of the GNU Lesser General Public
+# License along with the GNU C Library; if not, see
+# <https://www.gnu.org/licenses/>.
+
+set -e
+
+rtld=$1
+test_program=$2
+test_wrapper_env=$3
+run_program_env=$4
+library_path=$5
+argv0=$6
+
+echo "# [${test_wrapper_env}] [${run_program_env}] [$rtld] [--library-path]" \
+     "[$library_path] [--argv0] [$argv0] [$test_program]"
+${test_wrapper_env} \
+${run_program_env} \
+$rtld --library-path "$library_path" \
+  --argv0 "$argv0" $test_program 2>&1 && rc=0 || rc=$?
+echo "# exit status $rc"
+
+exit $rc

Re: [PATCH] ld.so: add an --argv0 option

[Florian Weimer]

* Vincent Mihalkovic via Libc-alpha:

> a few years ago there was an effort for adding --argv0 option:
> https://sourceware.org/legacy-ml/libc-alpha/2016-04/msg00576.html.
> I made the old patch actual for version 2.31.9000, added a test case and
> changelog.

Many shells support “exec -a” to get a similar effect.  The advantage is
that this does not perturb the argument vector layout.  Given that, I
think promoting --argv0 would merely introduce further compatibility
issues.

Thanks,
Florian

Re: [PATCH] ld.so: add an --argv0 option

[Vincent Mihalkovic]

Hi,

Sorry, I forgot to CC the libc-alpha mailing-list on my reply.

We do not need the option for shell scripting, where it is indeed not
needed,
as you say.  We are developing a tool for running dynamic analysis tools on
RPM packages fully automatically.  The tool appends custom linker flags
while
executing the %build section of RPMs to make the binaries use our custom ELF
interpreter.  While running the %check section, the interpreter takes care
of
running the binaries built in %build through the selected dynamic analyzer
in
a way that does not break the testing framework.  For this to work, we need
to run dynamic linker explicitly.  There is currently no way to preserve the
original argv[0], which some programs are sensitive to.  This unnecessarily
breaks the tests running in %check of some RPM packages.

An experimental implementation of the custom ELF interpreter is available
here:

    https://github.com/kdudka/cswrap/pull/2

thanks for considering this idea,
vincent mihalkovic

On Wed, Jul 22, 2020 at 11:00 AM Florian Weimer <fweimer@redhat.com> wrote:

> * Vincent Mihalkovic via Libc-alpha:
>
> > a few years ago there was an effort for adding --argv0 option:
> > https://sourceware.org/legacy-ml/libc-alpha/2016-04/msg00576.html.
> > I made the old patch actual for version 2.31.9000, added a test case and
> > changelog.
>
> Many shells support “exec -a” to get a similar effect.  The advantage is
> that this does not perturb the argument vector layout.  Given that, I
> think promoting --argv0 would merely introduce further compatibility
> issues.
>
> Thanks,
> Florian
>
>

Re: [PATCH] ld.so: add an --argv0 option

[Florian Weimer]

* Vincent Mihalkovic:

> Sorry, I forgot to CC the libc-alpha mailing-list on my reply.

And I think this won't make it to the list due to the HTML filters, so
here's a full quote:

> We do not need the option for shell scripting, where it is indeed not
> needed, as you say.  We are developing a tool for running dynamic
> analysis tools on RPM packages fully automatically.  The tool appends
> custom linker flags while executing the %build section of RPMs to make
> the binaries use our custom ELF interpreter.  While running the %check
> section, the interpreter takes care of running the binaries built in
> %build through the selected dynamic analyzer in a way that does not
> break the testing framework.  For this to work, we need to run dynamic
> linker explicitly.  There is currently no way to preserve the original
> argv[0], which some programs are sensitive to.  This unnecessarily
> breaks the tests running in %check of some RPM packages.
>
> An experimental implementation of the custom ELF interpreter is available
> here:
>
>     https://github.com/kdudka/cswrap/pull/2
>
> thanks for considering this idea,
> vincent mihalkovic

I was wrong about this, and it is not possible to get the desired
behavior using “exec -a”.  The question remains if just updating the
pointer is sufficient in this context, or if a more elaborate copying
operation is needed to preserve the expected semantics of the argument
vector.

I guest we could add --argv0 now (well, after the 2.32 release), and if
it's incompatible with some applications, we can perhaps tweak it later.

Thanks,
Florian

Re: [PATCH] ld.so: add an --argv0 option

[Adhemerval Zanella]

On 12/08/2020 08:06, Vincent Mihalkovic via Libc-alpha wrote:
> Hi,
> 
> I have seen that the development branch is open for glibc 2.33.
> In reaction to previous e-mail communication with Florian Weimer:
> 
>> "I guest we could add --argv0 now (well, after the 2.32 release), and if
>> it's incompatible with some applications, we can perhaps tweak it later."
>>
> I'm sending my argv0 patch again.
> 
> vincent mihalkovic
> 

No need to send ChangeLog.txt anymore.  As a side note I would like ask you send
the patch as inline instead of attachments.

> diff --git a/elf/Makefile b/elf/Makefile
> index 0b78721848..f38904d831 100644
> --- a/elf/Makefile
> +++ b/elf/Makefile
> @@ -210,7 +210,8 @@ tests += restest1 preloadtest loadfail multiload origtest resolvfail \
>  	 tst-filterobj tst-filterobj-dlopen tst-auxobj tst-auxobj-dlopen \
>  	 tst-audit14 tst-audit15 tst-audit16 \
>  	 tst-single_threaded tst-single_threaded-pthread \
> -	 tst-tls-ie tst-tls-ie-dlmopen
> +	 tst-tls-ie tst-tls-ie-dlmopen \
> +	 argv0test
>  #	 reldep9
>  tests-internal += loadtest unload unload2 circleload1 \
>  	 neededtest neededtest2 neededtest3 neededtest4 \

Ok.

> @@ -414,7 +415,7 @@ endif
>  ifeq (yes,$(build-shared))
>  ifeq ($(run-built-tests),yes)
>  tests-special += $(objpfx)tst-pathopt.out $(objpfx)tst-rtld-load-self.out \
> -		 $(objpfx)tst-rtld-preload.out
> +		 $(objpfx)tst-rtld-preload.out $(objpfx)argv0test.out
>  endif
>  tests-special += $(objpfx)check-textrel.out $(objpfx)check-execstack.out \
>  		 $(objpfx)check-wx-segment.out \
> @@ -1796,3 +1797,11 @@ $(objpfx)tst-tls-ie-dlmopen.out: \
>    $(objpfx)tst-tls-ie-mod6.so
>  
>  $(objpfx)tst-tls-surplus: $(libdl)
> +
> +ARGV0 = test-argv0
> +$(objpfx)argv0test.out: tst-rtld-argv0.sh $(objpfx)ld.so \
> +			$(objpfx)argv0test
> +	$(SHELL) $< $(objpfx)ld.so $(objpfx)argv0test \
> +            '$(test-wrapper-env)' '$(run_program_env)' \
> +            '$(rpath-link)' '$(ARGV0)' > $@; \
> +    $(evaluate-test)

Ok.

> diff --git a/elf/argv0test.c b/elf/argv0test.c
> new file mode 100644
> index 0000000000..4c79bebf23
> --- /dev/null
> +++ b/elf/argv0test.c
> @@ -0,0 +1,13 @@
> +#include <stdio.h>   // for printf
> +#include <string.h>  // for strcmp
> +
> +
> +int main( int argc, char *argv[] ) {
> +	int result = strcmp( argv[0], "test-argv0");
> + 
> +	printf ("argv[0] = %s, strcmp( argv[0], \"test-argv0\" ) = %d, %s\n", \
> +					argv[0], result, !result ? "ok" : "wrong");
> +
> +	return result;
> +
> +}
> diff --git a/elf/rtld.c b/elf/rtld.c

Please use libsupport even for such simple tests. Also use glibc code style [1].

> index 5b882163fa..cafa4f9bd3 100644
> --- a/elf/rtld.c
> +++ b/elf/rtld.c
> @@ -1202,6 +1202,8 @@ dl_main (const ElfW(Phdr) *phdr,
>  	 installing it.  */
>        rtld_is_main = true;
>  
> +      char *argv0 = NULL;
> +
>        /* Note the place where the dynamic linker actually came from.  */
>        GL(dl_rtld_map).l_name = rtld_progname;
>  
> @@ -1259,6 +1261,14 @@ dl_main (const ElfW(Phdr) *phdr,
>  	else if (! strcmp (_dl_argv[1], "--preload") && _dl_argc > 2)
>  	  {
>  	    preloadarg = _dl_argv[2];
> +	    _dl_skip_args += 2;
> +	    _dl_argc -= 2;
> +	    _dl_argv += 2;
> +	  }
> +	else if (! strcmp (_dl_argv[1], "--argv0") && _dl_argc > 2)
> +	  {
> +	    argv0 = _dl_argv[2];
> +
>  	    _dl_skip_args += 2;
>  	    _dl_argc -= 2;
>  	    _dl_argv += 2;

Ok.

> @@ -1292,7 +1302,8 @@ of this helper program; chances are you did not intend to run this program.\n\
>    --inhibit-rpath LIST  ignore RUNPATH and RPATH information in object names\n\
>  			in LIST\n\
>    --audit LIST          use objects named in LIST as auditors\n\
> -  --preload LIST        preload objects named in LIST\n");
> +  --preload LIST        preload objects named in LIST\n\
> +  --argv0 STRING        set argv[0] to STRING before running\n");
>  
>        ++_dl_skip_args;
>        --_dl_argc;
> @@ -1384,6 +1395,11 @@ of this helper program; chances are you did not intend to run this program.\n\
>  	    break;
>  	  }
>  #endif
> +
> +      /* Set the argv[0] string now that we've processed the executable.  */
> +      if (argv0 != NULL) {
> +		_dl_argv[0] = argv0;
> +	  }

Use the glibc code style and conventions [1]. In this case just drop the brackets.

>      }
>    else
>      {
> diff --git a/elf/tst-rtld-argv0.sh b/elf/tst-rtld-argv0.sh
> new file mode 100755
> index 0000000000..5f873b6c5c
> --- /dev/null
> +++ b/elf/tst-rtld-argv0.sh
> @@ -0,0 +1,37 @@
> +#!/bin/sh
> +# Test --argv0 argument ld.so.
> +# Copyright (C) 2019-2020 Free Software Foundation, Inc.
> +# This file is part of the GNU C Library.
> +#
> +# The GNU C Library is free software; you can redistribute it and/or
> +# modify it under the terms of the GNU Lesser General Public
> +# License as published by the Free Software Foundation; either
> +# version 2.1 of the License, or (at your option) any later version.
> +#
> +# The GNU C Library is distributed in the hope that it will be useful,
> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
> +# Lesser General Public License for more details.
> +#
> +# You should have received a copy of the GNU Lesser General Public
> +# License along with the GNU C Library; if not, see
> +# <https://www.gnu.org/licenses/>.
> +
> +set -e
> +
> +rtld=$1
> +test_program=$2
> +test_wrapper_env=$3
> +run_program_env=$4
> +library_path=$5
> +argv0=$6
> +
> +echo "# [${test_wrapper_env}] [${run_program_env}] [$rtld] [--library-path]" \
> +     "[$library_path] [--argv0] [$argv0] [$test_program]"
> +${test_wrapper_env} \
> +${run_program_env} \
> +$rtld --library-path "$library_path" \
> +  --argv0 "$argv0" $test_program 2>&1 && rc=0 || rc=$?
> +echo "# exit status $rc"
> +
> +exit $rc

Ok.

[1] https://sourceware.org/glibc/wiki/Style_and_Conventions

[PATCH] ld.so: add an --argv0 option

[Vincent Mihalkovic]

[-- Attachment #1: Type: text/plain, Size: 343 bytes --]

Hi,

I have seen that the development branch is open for glibc 2.33.
In reaction to previous e-mail communication with Florian Weimer:

> "I guest we could add --argv0 now (well, after the 2.32 release), and if
> it's incompatible with some applications, we can perhaps tweak it later."
>
I'm sending my argv0 patch again.

vincent mihalkovic

[-- Attachment #2: changelog.txt --]
[-- Type: text/plain, Size: 240 bytes --]

2020-07-22  Vincent Mihalkovic  <vmihalko@redhat.com>

	* elf/Makefile: added argv0 test case, Modified.
	* elf/rtld.c: added --argv0 option, Modified.
	* elf/argv0test.c: test case, New file.
	* elf/tst-rtld-argv0.sh: test case, New file.

[-- Attachment #3: argv0.patch --]
[-- Type: application/x-patch, Size: 4788 bytes --]

Re: [PATCH] ld.so: add an --argv0 option

[Florian Weimer]

* Mike Frysinger:

> On 23 Apr 2016 14:21, Florian Weimer wrote:
>> * Mike Frysinger:
>> > Sometimes when you run a program you want the argv[0] string passed to
>> > the app to be different than the actual path you used to load it.  We
>> > can't do this today with invoking via ld.so which can be limiting --
>> > some programs like to inspect their argv[0] and make decisions as to
>> > how it should (re)exec itself or helper tools.  For example, clang and
>> > gcc both do argv[0] inspection to support relocatable toolchains.
>> 
>> Interesting idea.  I think it's sufficiently to include this, but it
>> needs a ChangeLog entry and a test case.
>
> for the test, seems like i'll have to write a small C file that just
> displays its argv, and then another shell script which runs the ldso
> against that simple program and checks the output.  unless anyone has
> a simpler idea.

You'll need the shell script wrapper.

Supplying the expected value in another argument, with some prefix and
suffix character, and comparing argv[0] to the extracted orignal
argument, is likely simpler than writing argv[0] to standard output
and reading it back from the shell script.

Re: [PATCH] ld.so: add an --argv0 option

[Mike Frysinger]

[-- Attachment #1: Type: text/plain, Size: 869 bytes --]

On 23 Apr 2016 14:21, Florian Weimer wrote:
> * Mike Frysinger:
> > Sometimes when you run a program you want the argv[0] string passed to
> > the app to be different than the actual path you used to load it.  We
> > can't do this today with invoking via ld.so which can be limiting --
> > some programs like to inspect their argv[0] and make decisions as to
> > how it should (re)exec itself or helper tools.  For example, clang and
> > gcc both do argv[0] inspection to support relocatable toolchains.
> 
> Interesting idea.  I think it's sufficiently to include this, but it
> needs a ChangeLog entry and a test case.

for the test, seems like i'll have to write a small C file that just
displays its argv, and then another shell script which runs the ldso
against that simple program and checks the output.  unless anyone has
a simpler idea.
-mike

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH] ld.so: add an --argv0 option

[Florian Weimer]

* Mike Frysinger:

> Sometimes when you run a program you want the argv[0] string passed to
> the app to be different than the actual path you used to load it.  We
> can't do this today with invoking via ld.so which can be limiting --
> some programs like to inspect their argv[0] and make decisions as to
> how it should (re)exec itself or helper tools.  For example, clang and
> gcc both do argv[0] inspection to support relocatable toolchains.

Interesting idea.  I think it's sufficiently to include this, but it
needs a ChangeLog entry and a test case.

[PATCH] ld.so: add an --argv0 option

[Mike Frysinger]

Sometimes when you run a program you want the argv[0] string passed to
the app to be different than the actual path you used to load it.  We
can't do this today with invoking via ld.so which can be limiting --
some programs like to inspect their argv[0] and make decisions as to
how it should (re)exec itself or helper tools.  For example, clang and
gcc both do argv[0] inspection to support relocatable toolchains.
---
 elf/rtld.c | 17 ++++++++++++++++-
 1 file changed, 16 insertions(+), 1 deletion(-)

diff --git a/elf/rtld.c b/elf/rtld.c
index 647661c..a42b5f6 100644
--- a/elf/rtld.c
+++ b/elf/rtld.c
@@ -796,6 +796,8 @@ dl_main (const ElfW(Phdr) *phdr,
 	 installing it.  */
       rtld_is_main = true;
 
+      char *argv0 = NULL;
+
       /* Note the place where the dynamic linker actually came from.  */
       GL(dl_rtld_map).l_name = rtld_progname;
 
@@ -850,6 +852,14 @@ dl_main (const ElfW(Phdr) *phdr,
 	    _dl_argc -= 2;
 	    _dl_argv += 2;
 	  }
+	else if (! strcmp (_dl_argv[1], "--argv0") && _dl_argc > 2)
+	  {
+	    argv0 = _dl_argv[2];
+
+	    _dl_skip_args += 2;
+	    _dl_argc -= 2;
+	    _dl_argv += 2;
+	  }
 	else
 	  break;
 
@@ -878,7 +888,8 @@ of this helper program; chances are you did not intend to run this program.\n\
 			variable LD_LIBRARY_PATH\n\
   --inhibit-rpath LIST  ignore RUNPATH and RPATH information in object names\n\
 			in LIST\n\
-  --audit LIST          use objects named in LIST as auditors\n");
+  --audit LIST          use objects named in LIST as auditors\n\
+  --argv0 STRING        set argv[0] to STRING before running\n");
 
       ++_dl_skip_args;
       --_dl_argc;
@@ -971,6 +982,10 @@ of this helper program; chances are you did not intend to run this program.\n\
 	    break;
 	  }
 #endif
+
+      /* Set the argv[0] string now that we've processed the executable.  */
+      if (argv0 != NULL)
+	_dl_argv[0] = argv0;
     }
   else
     {
-- 
2.7.4