Re: ping: [patch 1/6] PIE: Attach binary even after re-prelinked underneath

[Jan Kratochvil <jan dot kratochvil at redhat dot com>]

On Mon, 05 Jul 2010 19:06:02 +0200, Joel Brobecker wrote:
> > gdb/
> > 2010-07-02  Jan Kratochvil  <jan.kratochvil@redhat.com>
> > 	    Joel Brobecker  <brobecker@adacore.com>
> > 
> > 	Fix attaching to PIEs prelinked on the disk after the process was
> > 	started.
> > 	* solib-svr4.c (svr4_exec_displacement): New variable arch_size.
> > 	Verify it against bfd_get_arch_size.  Try to match arbitrary
> > 	displacement for the phdrs comparison.
> > 
> > gdb/testsuite/
> > 2010-07-02  Jan Kratochvil  <jan.kratochvil@redhat.com>
> > 	    Joel Brobecker  <brobecker@adacore.com>
> > 
> > 	* gdb.base/break-interp.exp: Run $binpie with new value "ATTACH", new
> > 	code for it.  New variable relink_args.
> > 	(prelinkYES): Call prelinkNO.
> > 	(test_attach): Accept new parameter relink_args.  Re-prelink the binary
> > 	in such case.  Move the core code to ...
> > 	(test_attach_gdb): ... a new function.  Send GDB command "file".
> > 	Extend expected "Attaching to " string.
> 
> This is OK, with one English error in one of my suggestions (mea culpa).

The "easier" -> "more easily" one?  My fault, I forgot to include this fix by
you, sorry.


> > +		  /* PT_GNU_STACK is an exception by being never relocated by
> > +		     prelink as its addresses are always zero.  */
> 
> I understand why you mean, now, about the PT_GNU_STACK entry not being
> changed during the prelink.  But I don't get the relationship between
> this comment and the code surrounding it. Can you explain that?

Code simplified for better readability in this mail:

		  /* PT_GNU_STACK is an exception by being never relocated by
		     prelink as its addresses are always zero.  */
		  if (memcmp (phdrp, phdr2p, sizeof (*phdrp)) == 0)
		    continue;

		  /* Check also other adjustment combinations - PR 11786.  */
		  *buf_vaddr_p -= displacement;
		  *buf_paddr_p -= displacement;
		  if (memcmp (phdrp, phdr2p, sizeof (*phdrp)) == 0)
		    continue;

For detected DISPLACEMENT value 0x3000000000 the latter test works:
  PHDR           0x000040 0x0000000000000040 0x0000000000000040 0x0001c0 0x0001c0 R E 0x8
->
  PHDR           0x000040 0x0000003000000040 0x0000003000000040 0x0001c0 0x0001c0 R E 0x8
as 0x0000003000000040 - 0x3000000000 == 0x0000000000000040.

But in the same executable there is also
  GNU_STACK      0x000000 0x0000000000000000 0x0000000000000000 0x000000 0x000000 RW  0x8
->
  GNU_STACK	 0x000000 0x0000000000000000 0x0000000000000000 0x000000 0x000000 RW  0x8
but 0x0000000000000000 - 0x3000000000 == 0xffffffd000000000 and thus
    0x0000000000000000 - 0x3000000000 != 0x0000000000000000
and we would fail the verification despite the executable perfectly matches.

I believe one should be looking at some two `readelf -Wa' dumps of an
executable with two prelink addresses while checking this code so it should be
apparent during real updates of that code.



Thanks,
Jan