[PATCH] gold: fix merging of non-section symbols

[PATCH] gold: fix merging of non-section symbols

[Xiao Jia]

os->output_address() returns a virtual address, while what we actually
want is an offset starting from the section.

Without this fix, programs linked with user-provided linker scripts may
segfault e.g. when trying to access string literals.  The linker script
that I was using to trigger the segfault was in a non-upstream Linux
kernel tree.  It uses RO_DATA_SECTION, which has merge sections such as
".rodata : { *(.rodata) *(.rodata.*) }", and strings in .rodata.* were
incorrectly relocated.

Signed-off-by: Xiao Jia <xiaoj@google.com>
---
 gold/object.cc | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/gold/object.cc b/gold/object.cc
index a631c99..be7cda9 100644
--- a/gold/object.cc
+++ b/gold/object.cc
@@ -2405,7 +2405,8 @@ Sized_relobj_file<size, big_endian>::compute_final_local_value_internal(
 	      // This is not a section symbol.  We can determine
 	      // the final value now.
 	      lv_out->set_output_value(
-		  os->output_address(this, shndx, lv_in->input_value()));
+		  os->output_address(this, shndx, lv_in->input_value())
+		  - os->address());
 	    }
 	  else if (!os->find_starting_output_address(this, shndx, &start))
 	    {
-- 
2.7.0.rc3.207.g0ac5344

Re: [PATCH] gold: fix merging of non-section symbols

[Cary Coutant]

Is there a PR for this bug?

> os->output_address() returns a virtual address, while what we actually
> want is an offset starting from the section.

I don't think that's right, unless we're doing a relocatable link. See
the use of relocatable_link_adjustment in the "else if" block
immediately following this one.

-cary

Re: [PATCH] gold: fix merging of non-section symbols

[Xiao Jia]

Sorry for the late reply, I spent time on something else.

On Thu, Mar 3, 2016 at 11:30 AM, Cary Coutant <ccoutant@gmail.com> wrote:
> Is there a PR for this bug?

I'm not familiar with the workflow.  What is a PR?

>
>> os->output_address() returns a virtual address, while what we actually
>> want is an offset starting from the section.
>
> I don't think that's right, unless we're doing a relocatable link. See
> the use of relocatable_link_adjustment in the "else if" block
> immediately following this one.

Yes in my case it's a relocatable symbol IIRC.  I'll take a look at
the code you mentioned.  Thanks!

Xiao

Re: [PATCH] gold: fix merging of non-section symbols

[Cary Coutant]

>> Is there a PR for this bug?
>
> I'm not familiar with the workflow.  What is a PR?

PR is short for Problem Report. Bugs should be filed in bugzilla, the
binutils bug-tracking database:

    https://sourceware.org/bugzilla/

-cary