[Bug 898829] Re: Segfault in dlerror.c 160 on call to dlopen

[Michael Owens]

Confirmed that this issue exists on 12.04 alpha as well.

-- 
You received this bug notification because you are a member of Ubuntu
Foundations Bugs, which is subscribed to eglibc in Ubuntu.
https://bugs.launchpad.net/bugs/898829

Title:
  Segfault in dlerror.c 160 on call to dlopen

Status in “eglibc” package in Ubuntu:
  New

Bug description:
  DistroRelease: Ubuntu 11.10
  Package: libc6 2.13-20ubuntu5
  Uname: 3.0.0-13-generic #22-Ubuntu SMP Wed Nov 2 13:27:26 UTC 2011 x86_64 x86_64 x86_64 GNU/Linux
  Architecture: amd64, i386 (this issue exists on both architectures)

  I've encountered a segfault in dlerrror.c on XUbuntu 11.10 64-bit
  desktop in an Apache module, specifically when one of the module's
  shared libraries calls dlopen(). I have searched launchpad for likely
  candidates and this problem may be related to the following bugs:

    https://bugs.launchpad.net/ubuntu/+source/eglibc/+bug/893605.
    https://bugs.launchpad.net/ubuntu/+source/nspluginwrapper/+bug/762387

  The Apache module (which I wrote) embeds a Ruby VM via libruby, which
  is dynamically linked. The segfault occurs on Ruby startup, where it
  attempts to load /usr/lib/ruby/1.9.1/x86_64-linux/etc.so via dlopen().
  This produces following:

  #0  __GI___libc_free (mem=0x20000) at malloc.c:3709
  #1  0x00007ffff68b2565 in _dlerror_run (operate=0x7ffff68b1ec0 <dlopen_doit>, args=0x7fffffffa530) at dlerror.c:160
  #2  0x00007ffff68b1fc1 in __dlopen (file=<optimized out>, mode=<optimized out>) at dlopen.c:88
  #3  0x00007ffff3f63c65 in dln_load (file=0x7ffff83e3100 "/usr/lib/ruby/1.9.1/x86_64-linux/etc.so") at dln.c:1276
  #4  0x00007ffff3f9d29a in load_ext (path=140737356478000) at load.c:554

  Ruby calls into dlopen.c as follows:

   /* Load file */
   if ((handle = (void*)dlopen(file, RTLD_LAZY|RTLD_GLOBAL)) == NULL) {
       error = dln_strerror();
       goto failed;
   }

  This is the first call that Ruby makes to dlopen(). dlopen() then
  proceeds into dlerror.c where on line 160 it attempts to free an error
  string in a result structure that has a bad memory address. The code
  is as follows:

    if (result->errstring != NULL)
      {
        /* Free the error string from the last failed command.  This can
    happen if `dlerror' was not run after an error was found.  */
        if (result->malloced)
   free ((char *) result->errstring); <-- dies here.
        result->errstring = NULL;
      }

  The value of the result structure is the following:

          $8 = {errcode = -131693408, returned = 32767, malloced = 192,
  objname = 0x7ffff2149010 "\004", errstring = 0x20000 <Address 0x20000
  out of bounds>}

  Since this appears to be the first DSO Ruby attempts to load, it lead
  me to suspect that that it may be related to bug 893605 above.

  To test the general call to dlopen() on the etc.so file in particular,
  I added a line in the Apache module to do so just before it enters
  Ruby initialization. This invocation of dlopen() worked without a
  problem. Yet when the Ruby library attempted it next, it caused the
  above sefault in dlerror.c.

  Guessing that that maybe it had something to do with visibility of
  symbols within Apache, I statically linked libruby to the module in
  the hopes that dlopen() would work then. But it again produced the
  same segfault.

  Wondering if it could be changes in either the Ruby or Apache code, I
  compiled and built the same versions of Ruby and Apache from Maverick
  (which work fine) on 11.10 to rule this out. They produced the exact
  same segfault in 11.10.

  With regard to the Apache module itself as the culprit, I have
  maintained it for over three years without incident from Hardy up to
  Maverick on both 32 bit and 64 bit systems. It also compiles and runs
  on FreeBSD without problems. Further, I have not made any substantial
  changes to it for the last eight months.

  To reproduce the bug, I put the compiled binaries on
  http://mikeowens.ws/bug/ along with a short README file explaining how
  to reproduce the problem. Also, the full source code for the project
  is at https://github.com/linterra/r4a. It has a debian folder so it
  can be built via pbuilder and dpkg-buildpackage.

  The full backtrace is attached.

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/eglibc/+bug/898829/+subscriptions