[7.0] PR/9723: gdb breakpoints silently fail on PIE binaries

[7.0] PR/9723: gdb breakpoints silently fail on PIE binaries

[Thiago Jung Bauermann]

Hi,

One painful aspect of GDB right now is that it doesn't handle PIE (Position 
Independent Executable) and doesn't even recognize a PIE binary when it sees 
one. I have an action item for 7.0 which is to make GDB at least complain that 
it is being given such a binary and warn the user appropriately. That would be 
enough to close the bugzilla in the subject and ship 7.0 (as we have discussed 
before in this mailing list).

Then for 7.1 somebody could push upstream one of the two circulating patches 
which implement PIE support, and close PR/9174 (gdb can't handle PIE).

All this to say: can somebody save me some research time and tell in high 
level how can I recognize from looking at an ELF file that I'm dealing with a 
PIE binary (NOT a PIC library, of course)? Something like "see if the frob bit 
in the bozo section is set") is enough, I can go from there.
-- 
[]'s
Thiago Jung Bauermann

Re: [7.0] PR/9723: gdb breakpoints silently fail on PIE binaries

[Paul Pluzhnikov]

On Sun, Jul 19, 2009 at 6:57 PM, Thiago Jung
Bauermann<thiago.bauermann@gmail.com> wrote:

> All this to say: can somebody save me some research time and tell in high
> level how can I recognize from looking at an ELF file that I'm dealing with a
> PIE binary (NOT a PIC library, of course)? Something like "see if the frob bit
> in the bozo section is set") is enough, I can go from there.

Elf*_Ehdr.e_type == ET_DYN for the main executable?

Elf*_Phdr.p_vaddr == 0 for the first PT_LOAD segment?
[This one is only true for non-prelinked -pie executable.]

Cheers,
-- 
Paul Pluzhnikov

Re: [7.0] PR/9723: gdb breakpoints silently fail on PIE binaries

[Thiago Jung Bauermann]

Em Segunda-feira 20 Julho 2009 01:13:00 Paul Pluzhnikov escreveu:
> On Sun, Jul 19, 2009 at 6:57 PM, Thiago Jung
>
> Bauermann<thiago.bauermann@gmail.com> wrote:
> > All this to say: can somebody save me some research time and tell in high
> > level how can I recognize from looking at an ELF file that I'm dealing
> > with a PIE binary (NOT a PIC library, of course)? Something like "see if
> > the frob bit in the bozo section is set") is enough, I can go from there.
>
> Elf*_Ehdr.e_type == ET_DYN for the main executable?
>
> Elf*_Phdr.p_vaddr == 0 for the first PT_LOAD segment?
> [This one is only true for non-prelinked -pie executable.]

Great, thanks for the tip. Since nobody jumped into discussion, I assume any 
of those ways will work and have no known side-effect or false 
positive/negative. We'll see. :-)

Now I have this issue, I'd like to ask people's opinion about it: by design 
(or defect?) the BFD library is a PITA to access the program header. I feel 
very tempted to use elf32.h and elf64.h directly in order to check that field. 
My only worry is that a cross-debugging session with a win32 gdb and a linux 
remote target would not be possible, unless windows includes such sysv elf 
headers too (perhaps it does?).

Do you have an opinion?
-- 
[]'s
Thiago Jung Bauermann

Re: [7.0] PR/9723: gdb breakpoints silently fail on PIE binaries

[Daniel Jacobowitz]

On Fri, Jul 24, 2009 at 11:12:41PM -0300, Thiago Jung Bauermann wrote:
> Now I have this issue, I'd like to ask people's opinion about it: by design 
> (or defect?) the BFD library is a PITA to access the program header. I feel 
> very tempted to use elf32.h and elf64.h directly in order to check that field. 
> My only worry is that a cross-debugging session with a win32 gdb and a linux 
> remote target would not be possible, unless windows includes such sysv elf 
> headers too (perhaps it does?).

elfread.c:elf_symfile_segments.  It's not much of a pain really?

-- 
Daniel Jacobowitz
CodeSourcery

Re: [7.0] PR/9723: gdb breakpoints silently fail on PIE binaries

[Mark Kettenis]

> From: Thiago Jung Bauermann <thiago.bauermann@gmail.com>
> Date: Fri, 24 Jul 2009 23:12:41 -0300
> 
> Em Segunda-feira 20 Julho 2009 01:13:00 Paul Pluzhnikov escreveu:
> > On Sun, Jul 19, 2009 at 6:57 PM, Thiago Jung
> >
> > Bauermann<thiago.bauermann@gmail.com> wrote:
> > > All this to say: can somebody save me some research time and tell in high
> > > level how can I recognize from looking at an ELF file that I'm dealing
> > > with a PIE binary (NOT a PIC library, of course)? Something like "see if
> > > the frob bit in the bozo section is set") is enough, I can go from there.
> >
> > Elf*_Ehdr.e_type == ET_DYN for the main executable?
> >
> > Elf*_Phdr.p_vaddr == 0 for the first PT_LOAD segment?
> > [This one is only true for non-prelinked -pie executable.]
> 
> Great, thanks for the tip. Since nobody jumped into discussion, I assume any 
> of those ways will work and have no known side-effect or false 
> positive/negative. We'll see. :-)

Well, I'm not sure the second method is something you can rely on.  I
think Elf*_Phdr.p_vaddr can be 0 for non-PIC executables as well on
some architectures.

> Now I have this issue, I'd like to ask people's opinion about it: by
> design (or defect?) the BFD library is a PITA to access the program
> header. I feel very tempted to use elf32.h and elf64.h directly in
> order to check that field.  My only worry is that a cross-debugging
> session with a win32 gdb and a linux remote target would not be
> possible, unless windows includes such sysv elf headers too (perhaps
> it does?).
> 
> Do you have an opinion?

Which elf32.h and elf64.h.  I don't think those are standardized
headers in any way.  Defenitely no files with those names under
/usr/include on OpenBSD.  Heck, even my Linux workstation at work
doesn't have any files named like that.

There is "elf/common.h" in the src tree though.

Anyway, if the BFD library doesn't provide some functionality you
need, I think the best thing to do would be to add it there.  There
seems to be an Elf_Internal_Phdr pointer in struct elf_obj_tdata, so
perhaps the only thing you need to add is a macro like elf_elfheader()
to access it.