Bug 1595191

Created attachment 1480775 [details]
Application compiled with GCC 2.95.3

Hello,

We have a C++ application that runs under CentOS 7.3.1611 with glibc-2.17-157, but under CentOS 7.5.1804 with glibc-2.17-222 the application "runs afoul" of the libio hardening. We are using g++ from GCC 2.95.3 to build this application.

Our test application is as follows:

```
#include <stdio.h>
#include <string>

int main(void)
{
    std::string str;
    printf("hello\n");
    return 1;
}
```

(so, we make no internal changes to the FILE* structures at all as part of this application). However, when we run this application, we obtain:

```
Fatal error: glibc detected an invalid stdio handle
Aborted
```

Our 2.95.3 build of GCC is built (unmodified) from the sources directly from the GNU website. To compile this version of GCC, we use a legacy build of GCC that we already had (that is, another version of 2.95.3). This version of 2.95.3 is able to compile applications that run under older versions of glibc (e.g., 7.3.1611), so I do not suspect the build of GCC here.

I have attached the compiled application to this issue.

Cheers,

Andrew

My apologies, when I wrote "glibc (e.g., 7.3.1611)", I meant "glibc (e.g., glibc-2.17-157)".

Further dependencies for the backport:

commit 2afece36f6006844e87d7cb2fcb1ad8b220b2623
Author: Florian Weimer <fweimer>
Date:   Wed May 16 17:00:35 2018 +0200

    support: Add TEST_COMPARE_BLOB, support_quote_blob

commit 5c0202af4b3d588c04bcec7baf05706b21cd7416
Author: Florian Weimer <fweimer>
Date:   Tue Jun 26 12:05:21 2018 +0200

    support: Add TEST_NO_SETVBUF
    
    This is sometimes needed for testing stdio streams, where the
    setvbuf call in the test driver could interfere with the test.

(In reply to Andrew V. Jones from comment #1)
> Created attachment 1480775 [details]
> Application compiled with GCC 2.95.3
> 
> Hello,
> 
> We have a C++ application that runs under CentOS 7.3.1611 with
> glibc-2.17-157, but under CentOS 7.5.1804 with glibc-2.17-222 the
> application "runs afoul" of the libio hardening. We are using g++ from GCC
> 2.95.3 to build this application.

Thank you.  I have put a glibc preview build with a potential fix here:

https://people.redhat.com/~fweimer/shcBVI8M85lD/

I would appreciate if you could check if it corrects behavior with your application, too.

Hi,

Yes, I can confirm that installing this new version of glibc corrects the old issue.

Steps:

    * Using GCC 2.95.3 on a 7.6 box

    * g++ test.cpp  # from above

    * ./a.out

    * Output: "Fatal error: glibc detected an invalid stdio handle\nAborted"

    * Created a repo.d for https://people.redhat.com/~fweimer/shcBVI8M85lD/

    * yum update

    * ./a.out

    * Output: "hello"

I also tested this with our "real" application (i.e., rather than `./a.out`) and it doesn't work on an up-to-date 7.6 box (as we knew), but it does work on a box with this new version of glibc installed.

Verified by upstream testsuite tests: libio/tst-vtables-interposed, libio/tst-vtables. Also verified with reproducer on x86_64 https://bugzilla.redhat.com/show_bug.cgi?id=1595191#c1

Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory, and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.

https://access.redhat.com/errata/RHSA-2019:2118

Unfortunately, the fix in this bug only covers programs linked against glibc 2.1 (which was released in 1999) and later versions. Even older binaries may need the fix for bug 1775816.