[SSSD] [PATCH] krb5_child: Fix use after free

Sumit Bose sbose at redhat.com
Tue Apr 8 09:28:05 UTC 2014


On Tue, Apr 08, 2014 at 11:02:34AM +0200, Lukas Slebodnik wrote:
> On (07/04/14 22:21), Sumit Bose wrote:
> >On Mon, Apr 07, 2014 at 09:20:50PM +0200, Lukas Slebodnik wrote:
> >> On (07/04/14 21:01), Sumit Bose wrote:
> >> >On Mon, Apr 07, 2014 at 08:39:07PM +0200, Jakub Hrozek wrote:
> >> >> On Mon, Apr 07, 2014 at 08:35:20PM +0200, Lukas Slebodnik wrote:
> >> >> > On (07/04/14 20:30), Jakub Hrozek wrote:
> >> >> > >On Mon, Apr 07, 2014 at 08:03:32PM +0200, Lukas Slebodnik wrote:
> >> >> > >> On (07/04/14 18:53), Jakub Hrozek wrote:
> >> >> > >> >On Fri, Apr 04, 2014 at 03:41:38PM +0200, Lukas Slebodnik wrote:
> >> >> > >> >> On (04/04/14 15:18), Jakub Hrozek wrote:
> >> >> > >> >> >On Thu, Apr 03, 2014 at 07:11:37PM +0200, Jakub Hrozek wrote:
> >> >> > >> >> >> On Thu, Mar 20, 2014 at 05:53:31PM +0100, Lukas Slebodnik wrote:
> >> >> > >> >> >> > On (20/03/14 17:21), Jakub Hrozek wrote:
> >> >> > >> >> >> > >On Thu, Mar 20, 2014 at 05:00:00PM +0100, Sumit Bose wrote:
> >> >> > >> >> >> > >> On Thu, Mar 20, 2014 at 04:20:59PM +0100, Lukas Slebodnik wrote:
> >> >> > >> >> >> > >> > ehlo,
> >> >> > >> >> >> > >> > 
> >> >> > >> >> >> > >> > debug_prg_name is used in debug_fn and it was allocated under
> >> >> > >> >> >> > >> > talloc context "kr". The variable "kr" was removed before the last debug
> >> >> > >> >> >> > >> > messages in function main. It is very little change that it will be overridden.
> >> >> > >> >> >> > >> > 
> >> >> > >> >> >> > >> > It is possible to see this issue with exported environment variable
> >> >> > >> >> >> > >> > TALLOC_FREE_FILL=255
> >> >> > >> >> >> > >> 
> >> >> > >> >> >> > >> I'm pretty sure the patch works as expected and fixes the isssue. But I
> >> >> > >> >> >> > >> wonder if a different approach might be better? I think it does not make
> >> >> > >> >> >> > >> sense to allocate debug_prg_name on a given talloc context but that it
> >> >> > >> >> >> > >> would be better to just allocate it on NULL. This is e.g. done in the
> >> >> > >> >> >> > >> ldap_child (here a talloc_free() is missing on exit but this would be a
> >> >> > >> >> >> > >> different story).  Then debug_prg_name can even be allocate before kr
> >> >> > >> >> >> > >> and the debug messages for a failed allocation of kr can use the right
> >> >> > >> >> >> > >> program name and not "sssd".
> >> >> > >> >> >> > >
> >> >> > >> >> >> > >I agree, because also given that krb5_child is a short lived process,
> >> >> > >> >> >> > >we don't care too much about possible leaks.
> >> >> > >> >> >> > No problem.
> >> >> > >> >> >> > 
> >> >> > >> >> >> > New version attached.
> >> >> > >> >> >> > 
> >> >> > >> >> >> > LS
> >> >> > >> >> >> 
> >> >> > >> >> >> This version works for me, do you want to also add a talloc_free() on
> >> >> > >> >> >> exit to be clean or do you also consider this not needed for short-lived
> >> >> > >> >> >> processes?
> >> >> > >> >> >
> >> >> > >> >> >Thinking again, it would be nicer to explicitly free the string on child
> >> >> > >> >> >exit. Yes, the leak it's not a big deal for a short-lived process but it
> >> >> > >> >> >would read better.
> >> >> > >> >> 
> >> >> > >> >> With the first version of patch string was freed, because it was alocated
> >> >> > >> >> under "kr" context.
> >> >> > >> >> Now, you should decide which version do you want to push :-)
> >> >> > >> >> 
> >> >> > >> >> LS
> >> >> > >> >
> >> >> > >> >I think it's easier to explain with a patch :-)
> >> >> > >> 
> >> >> > >> You didn't test your patch :-)
> >> >> > >
> >> >> > >I left that to you :-)
> >> >> > >
> >> >> > >See another proposal below:
> >> >> > >
> >> >> > >> 
> >> >> > >> diff --git a/src/providers/krb5/krb5_child.c b/src/providers/krb5/krb5_child.c
> >> >> > >> index 764f6ac7bf57b1f7d882961b7c6fa518818aaf23..aec0d9389dd4f3ae005b73ff12ca8293cee7560f 100644
> >> >> > >> --- a/src/providers/krb5/krb5_child.c
> >> >> > >> +++ b/src/providers/krb5/krb5_child.c
> >> >> > >> @@ -2013,6 +2013,7 @@ int main(int argc, const char *argv[])
> >> >> > >>          DEBUG(SSSDBG_CRIT_FAILURE, "talloc failed.\n");
> >> >> > >>          exit(-1);
> >> >> > >>      }
> >> >> > >> +    talloc_steal(kr, debug_prg_name);
> >> >> > >> 
> >> >> > >>      if (debug_fd != -1) {
> >> >> > >>          ret = set_debug_file_from_fd(debug_fd);
> >> >> > >> 
> >> >> > >> // snip
> >> >> > >> done:
> >> >> > >>     krb5_cleanup(kr);
> >> >> > >>     talloc_free(kr);
> >> >> > >>     ^^^^^^^^^^^^^^^
> >> >> > >> //debug_prg_name is freed
> >> >> > >> 
> >> >> > >>     if (ret == EOK) {
> >> >> > >>         DEBUG(SSSDBG_TRACE_FUNC, "krb5_child completed successfully\n");
> >> >> > >>         ^^^^^^
> >> >> > >> // use after free
> >> >> > >          talloc_free(tmp_str);
> >> >> > >>         exit(0);
> >> >> > >>     } else {
> >> >> > >>         DEBUG(SSSDBG_CRIT_FAILURE, "krb5_child failed!\n");
> >> >> > >>         ^^^^^^
> >> >> > >> // use after free
> >> >> > >          talloc_free(tmp_str);
> >> >> > >>         exit(-1);
> >> >> > >>     }
> >> >> > >> 
> >> >> > >
> >> >> > >Or just:
> >> >> > > done:
> >> >> > >     krb5_cleanup(kr);
> >> >> > >     talloc_free(kr);
> >> >> > >     if (ret == EOK) {
> >> >> > >         DEBUG(SSSDBG_TRACE_FUNC, "krb5_child completed successfully\n");
> >> >> > >         talloc_free(tmp_str);
> >> >> > >         rv = 0;
> >> >> > >     } else {
> >> >> > >         DEBUG(SSSDBG_CRIT_FAILURE, "krb5_child failed!\n");
> >> >> > >         talloc_free(tmp_str);
> >> >> > >         rv = -1;
> >> >> > >     }
> >> >> > >     exit(rv);
> >> >> > 
> >> >> > And now, you can compare your solution with the first patch from this thread
> >> >> > and then try to read replies to the 1st mail from this thread :-)
> >> >> 
> >> >> Except first patch didn't allocate on NULL? I think that was the meat of
> >> >> Sumit's comments..
> >> >
> >> >yes, just allocate on NULL and free it explicitly in the end to make
> >> >valgrind happy. The reason for suggesting NULL is that debug_prg_name is
> >> >global and I think putting it under a specific talloc-hierarchy does not
> >> >make sense.
> >> >
> >> but there is no difference. You allocate debug_prg_name on NULL and then steal
> >> to another talloc context. (like in ldap_child)
> >
> >Why do you have to steal it? Just allocate on NULL and free it in the
> >end.
> >
> >bye,
> >Sumit
> >
> Attached patch use the same approach as in ldap_child.

To bring this thread to an end, I think it is ok for the time being to
do the same as ldap_child and I'll think the patch is good code-wise.
I'll run some test and will then give the final ACK.

bye,
Sumit

> 
> LS



More information about the sssd-devel mailing list