Re: [patch] validate binary before use

[Jan Kratochvil <jan.kratochvil@redhat.com>]

On Tue, 29 Jan 2013 17:15:13 +0100, Aleksandar Ristovski wrote:
> --- gdb/solib.c	1 Jan 2013 06:32:51 -0000	1.169
> +++ gdb/solib.c	29 Jan 2013 15:46:39 -0000
> @@ -495,6 +495,17 @@ solib_map_sections (struct so_list *so)
>  	}
>      }
>  
> +  gdb_assert (ops->validate != NULL);
> +
> +  if (!ops->validate (so))

When thinking about it this approach regresses back again performance with
gdbserver over high latency links.  This is why the <library-list-svr4 /> XML
protocol has been put there.  With ops->validate there will be
a round-trip-time requirement on very every library listed from gdbserver.

Therefore with gdbserver the current
              p += sprintf (p, "<library name=\"%s\" lm=\"0x%lx\" "
                               "l_addr=\"0x%lx\" l_ld=\"0x%lx\"/>",
should be extended with hex-encoded build-id="..." parameter.  gdbserver does
not have the bfd but it already iterates program headers in get_dynamic so it
can find PT_NOTE there.

Therefore there should be new build_id field in struct so_list where
svr4_current_sos_via_xfer_libraries will put it from gdbserver.  In local mode
it should remain mostly as you wrote it / as suggested in the review as there
is currently no easy non-Linux way how to find PT_NOTE without bfd at hand.



Thanks,
Jan