Created attachment 18029 [details]
tests: Handle unknown syscall call errno.

Hi.

With the 1.3.0 release the test_syscall_swrap.c self-test was added.  It was not part of 1.2.9.  It can fail like this:

1/1 Test #16: test_syscall_swrap ...............***Failed    0.00 sec
[==========] Running 1 test(s).
[ RUN      ] test_uwrap_syscall_swrap
[  ERROR   ] --- 0x1 != 0x26
[   LINE   ] --- /uid_wrapper-1.3.0/tests/test_syscall_swrap.c:37: error: Failure!
[  FAILED  ] test_uwrap_syscall_swrap
[==========] 1 test(s) run.
[  PASSED  ] 0 test(s).
[  FAILED  ] 1 test(s), listed below:
[  FAILED  ] test_uwrap_syscall_swrap

Chaising this down was fun, and boils down to unknown syscalls resulting in different errno values in virtualized environments vs non-virtualized.  You can test it like this:

jas@kaka:~$ cat>foo.c
#include <stdio.h>
#include <unistd.h>
#include <sys/syscall.h>
#include <signal.h>
#include <errno.h>

void main (void)
{
	long rc;

	signal(SIGSYS, SIG_IGN);
	rc = syscall(123456789);
	signal(SIGSYS, SIG_DFL);
	printf ("rc %ld errno %d (EPERM %d ENOSYS %d)\n",
		rc, errno, EPERM, ENOSYS);
}
jas@kaka:~$ gcc foo.c
jas@kaka:~$ ./a.out 
rc -1 errno 38 (EPERM 1 ENOSYS 38)
jas@kaka:~$ 

If you run this in a VM it will output this instead:

rc -1 errno 1 (EPERM 1 ENOSYS 38)

Is the errno returned for unknown syscalls even documented anywhere?

I'm attaching a patch that we will use in Debian to work around this problem.  What do you think?

/Simon