program exit code, "at exit" handler and explicit close of shared objects

program exit code, "at exit" handler and explicit close of shared objects

[Roumen Petrov]

[-- Attachment #1: Type: text/plain, Size: 1372 bytes --]

Hello Cygwin developers,

This email is mainly for issue with exist code for a program.

It is inspired from openssl issue #10107 - "test_ssl_old freezes on 
cygwin". Actually after one or another work-around "unfreeze" test and 
this shows that one of subtests fail. This failure is main topic of this 
email.

More about failed test "testing connection with weak DH, expecting 
failure". So failure is expected and program code try to exit with 1. 
Unfortunately OS returns zero exit code.
Since version 1.1 OpenSSL uses "at exit handler" to clean-up allocated 
resources. Removing registered handler restores program exit code. So 
something in handler triggers issue.

After additional tests (research) I was able to isolate issues to simple 
test case. Please find attached "test-dlclose.c" and "Makefile".
First test is as is:

$ make
cc -g -Wall -Wextra    test-dlclose.c   -o test-dlclose
./test-dlclose
exit with code 33
make: [Makefile:4: all] Error 33 (ignored)

For next test change test-dlclose.c to define DLCLOSE_ATEXIT ( s/#if 0/#if 1/ ):
$ make
cc -g -Wall -Wextra    test-dlclose.c   -o test-dlclose
./test-dlclose
exit with code 33

As is visible make does not report error, i.e. program exit code is zero.

Is there a way to bypass issue?


For protocol:
$ uname -srvm -> CYGWIN_NT-10.0 3.0.7(0.338/5/3) 2019-04-30 18:08 x86_64

Regards,
Roumen Petrov

[-- Attachment #2: test-dlclose.c --]
[-- Type: text/x-csrc, Size: 701 bytes --]

#include <dlfcn.h>
#include <stdio.h>
#include <stdlib.h>

void *handle = NULL;

#if 0
# define DLCLOSE_ATEXIT	1
#endif

#ifdef DLCLOSE_ATEXIT
static void
dlclose_atexit(void) {
	if (dlclose(handle) != 0) {
		fprintf(stderr, "dlclose fail : %s\n", dlerror());
	}
}
#endif

int
main() {
	handle = dlopen("libz.so", RTLD_LAZY | RTLD_NOW);
	if (handle == NULL) {
		fprintf(stderr, "dlopen fail : %s\n", dlerror());
		return 1;
	}

#ifndef DLCLOSE_ATEXIT
	if (dlclose(handle) != 0) {
		fprintf(stderr, "dlclose fail : %s\n", dlerror());
		return 2;
	}
#else
    if (atexit(dlclose_atexit) != 0) {
		fprintf(stderr, "atexit fail\n");
		return 3;		
	}
#endif

	printf("exit with code 33\n");
	return 33;
}


[-- Attachment #3: Makefile --]
[-- Type: text/plain, Size: 62 bytes --]

CFLAGS = -g -Wall -Wextra

all: test-dlclose
	-./test-dlclose

[-- Attachment #4: Type: text/plain, Size: 219 bytes --]


--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple

Re: program exit code, "at exit" handler and explicit close of shared objects

[Ken Brown]

On 10/26/2019 11:07 AM, Roumen Petrov wrote:
> 
> After additional tests (research) I was able to isolate issues to simple test 
> case. Please find attached "test-dlclose.c" and "Makefile".
> First test is as is:
> 
> $ make
> cc -g -Wall -Wextra    test-dlclose.c   -o test-dlclose
> ./test-dlclose
> exit with code 33
> make: [Makefile:4: all] Error 33 (ignored)
> 
> For next test change test-dlclose.c to define DLCLOSE_ATEXIT ( s/#if 0/#if 1/ ):
> $ make
> cc -g -Wall -Wextra    test-dlclose.c   -o test-dlclose
> ./test-dlclose
> exit with code 33

I ran this second version of test-dlclose (with DLCLOSE_ATEXIT defined) under 
strace and got the following:

$ strace -o trace.out ./test-dlclose.exe 
exit with code 33
Segmentation fault

The strace output ends with

    30   30143 [main] test-dlclose 847 write: 18 = write(1, 0x600061E20, 18)
--- Process 33456 (pid: 847) unloaded DLL at 00000003e7930000
--- Process 33456 (pid: 847), exception c0000005 at 00000003e7931080
--- Process 33456 (pid: 847) thread 26088 exited with status 0xc0000005
--- Process 33456 (pid: 847) thread 33544 exited with status 0xc0000005
--- Process 33456 (pid: 847) thread 36020 exited with status 0xc0000005
--- Process 33456 exited with status 0xc0000005

The address 00000003e7930000 seems to be in /usr/bin/cygz.dll, which is the DLL 
that got unloaded.  After installing zlib-debuginfo, I ran addr2line to see 
where the crash occurred, and this too crashed:

$ addr2line -C -f -i -p -e /usr/bin/cygz.dll 0x3e7931080
Segmentation fault (core dumped)

That's as far as I've gotten.

Ken

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple

Re: program exit code, "at exit" handler and explicit close of shared objects

[Ken Brown]

On 10/29/2019 10:53 AM, Ken Brown wrote:
> On 10/26/2019 11:07 AM, Roumen Petrov wrote:
>>
>> After additional tests (research) I was able to isolate issues to simple test
>> case. Please find attached "test-dlclose.c" and "Makefile".
>> First test is as is:
>>
>> $ make
>> cc -g -Wall -Wextra    test-dlclose.c   -o test-dlclose
>> ./test-dlclose
>> exit with code 33
>> make: [Makefile:4: all] Error 33 (ignored)
>>
>> For next test change test-dlclose.c to define DLCLOSE_ATEXIT ( s/#if 0/#if 1/ ):
>> $ make
>> cc -g -Wall -Wextra    test-dlclose.c   -o test-dlclose
>> ./test-dlclose
>> exit with code 33
> 
> I ran this second version of test-dlclose (with DLCLOSE_ATEXIT defined) under
> strace and got the following:
> 
> $ strace -o trace.out ./test-dlclose.exe
> exit with code 33
> Segmentation fault
> 
> The strace output ends with
> 
>      30   30143 [main] test-dlclose 847 write: 18 = write(1, 0x600061E20, 18)
> --- Process 33456 (pid: 847) unloaded DLL at 00000003e7930000
> --- Process 33456 (pid: 847), exception c0000005 at 00000003e7931080
> --- Process 33456 (pid: 847) thread 26088 exited with status 0xc0000005
> --- Process 33456 (pid: 847) thread 33544 exited with status 0xc0000005
> --- Process 33456 (pid: 847) thread 36020 exited with status 0xc0000005
> --- Process 33456 exited with status 0xc0000005
> 
> The address 00000003e7930000 seems to be in /usr/bin/cygz.dll, which is the DLL

Sorry, I meant to say 00000003e7931080; 00000003e7930000 is the start of cygz.dll.

> that got unloaded.  After installing zlib-debuginfo, I ran addr2line to see
> where the crash occurred, and this too crashed:
> 
> $ addr2line -C -f -i -p -e /usr/bin/cygz.dll 0x3e7931080
> Segmentation fault (core dumped)
> 
> That's as far as I've gotten.
> 
> Ken
> 
> --
> Problem reports:       http://cygwin.com/problems.html
> FAQ:                   http://cygwin.com/faq/
> Documentation:         http://cygwin.com/docs.html
> Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
> 

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple

Re: program exit code, "at exit" handler and explicit close of shared objects

[Roumen Petrov]

Hello,

As non-regular user I did tests when I have access to environment.


Ken Brown wrote:

> On 10/29/2019 10:53 AM, Ken Brown wrote:
>> On 10/26/2019 11:07 AM, Roumen Petrov wrote:
>>> First test is as is:
>>> [SNIP]

>>> I ran this second version of test-dlclose (with DLCLOSE_ATEXIT defined) under
>>> strace and got the following:
>>>
>>> $ strace -o trace.out ./test-dlclose.exe
>>> exit with code 33
>>> Segmentation fault
>>>
>>> The strace output ends with
>>>
>>>       30   30143 [main] test-dlclose 847 write: 18 = write(1, 0x600061E20, 18)
>>> --- Process 33456 (pid: 847) unloaded DLL at 00000003e7930000
>>> --- Process 33456 (pid: 847), exception c0000005 at 00000003e7931080
>>> --- Process 33456 (pid: 847) thread 26088 exited with status 0xc0000005
>>> --- Process 33456 (pid: 847) thread 33544 exited with status 0xc0000005
>>> --- Process 33456 (pid: 847) thread 36020 exited with status 0xc0000005
>>> --- Process 33456 exited with status 0xc0000005
>>>
>>> The address 00000003e7930000 seems to be in /usr/bin/cygz.dll, which is the DLL
> Sorry, I meant to say 00000003e7931080; 00000003e7930000 is the start of cygz.dll.
>
>> that got unloaded.  After installing zlib-debuginfo, I ran addr2line to see
>> where the crash occurred, and this too crashed:
>>
>> $ addr2line -C -f -i -p -e /usr/bin/cygz.dll 0x3e7931080
>> Segmentation fault (core dumped)

The test was with compression library because this is default OpenSSL 
configuration - dynamic loading of compression library.

It seems to me issue could be reproduced with other libraries as well.
For instance I get the same with ssh ("libssh-4.so") or bz2 ("libbz2-1.so").



>> That's as far as I've gotten.
>>
>> Ken

Regards,
Roumen Petrov

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple