Missing security fix in elf/dl-open.c?

[Carlos O'Donell]

On 02/27/2015 04:19 AM, Florian Weimer wrote:
> On 02/27/2015 06:07 AM, Carlos O'Donell wrote:
>> On 02/24/2015 12:11 PM, Florian Weimer wrote:
>>> Some downstreams include this hunk in their patches related to
>>> CVE-2010-3847 and CVE-2011-0536:
>>>
>>> Index: glibc-2.12-2-gc4ccff1/elf/dl-object.c
>>> ===================================================================
>>> --- glibc-2.12-2-gc4ccff1.orig/elf/dl-object.c
>>> +++ glibc-2.12-2-gc4ccff1/elf/dl-object.c
>>> @@ -214,6 +214,9 @@ _dl_new_object (char *realname, const ch
>>>      out:
>>>        new->l_origin = origin;
>>>      }
>>> +  else if (INTUSE(__libc_enable_secure) && type == lt_executable)
>>> +    /* The origin of a privileged program cannot be trusted.  */
>>> +    new->l_origin = (char *) -1;
>>>
>>>    return new;
>>>  }
>>>
>>> I can't find this in glibc master.  Is the hunk above needed, or is it
>>> just hardening?
>>
>> Seems like additional hardening to me, and it could break real applications.
> 
> I don't understand much about ld.so, so here's what I guess the code
> does: It clears the origin of the main program if running as AT_SECURE.
> 
> However, something else already does this in Fedora 20
> (glibc-2.20-7.fc21.x86_64, which lacks this patch as well AFAICT).  I
> created a SUID binary with an $ORIGIN RPATH, and it is ignored, but only
> when actually running SUID.

Likely the `glibc-fedora-elf-ORIGIN.patch` patch in Fedora.

> It's also not clear to me why you would want to do this (whatever
> happens here) only for the main program, and not for other objects.
 
I don't know why either. It's just wrong. The Fedora patch fixes
_dl_dst_substitute which catches everything.

Security cargo-cult :-)

Cheers,
Carlos.