Seg fault occurs when running statically compiled binary from kernel using call_usermodehelper

[Will Deacon]

On Wed, Jul 10, 2013 at 07:52:00PM +0100, Russell King - ARM Linux wrote:
> On Wed, Jul 10, 2013 at 05:34:11PM +0100, Will Deacon wrote:
> > Ok, I've finally got to the bottom of this, but I'm not sure on the best way
> > to fix it.
> 
> I don't think you have!  You need to look back at the older ARM kernels
> to really get to the bottom of this...

It's hard enough looking at the current kernel with the internet connection
I'm currently on! Anyway, what I meant was that I can see why the
application is falling over.

> > The issue is that libc expects r0 to contain a function pointer
> > to be invoked at exit (rtld_fini), to clean up after a dynamic linker. If
> > this pointer is NULL, then it is ignored. We actually zero this pointer in
> > our ELF_PLAT_INIT macro.
> > 
> > At the same time, we have this strange code called next from the ARM ELF
> > loader:
> > 
> > 	regs->ARM_r2 = stack[2];	/* r2 (envp) */			\
> > 	regs->ARM_r1 = stack[1];	/* r1 (argv) */			\
> > 	regs->ARM_r0 = stack[0];	/* r0 (argc) */			\
> > 
> > which puts argc into r0.
> 
> You're sort of right.  It dates from the days when we had a.out binaries,
> those required argc, argv and envp in r0/r1/r2 - and ARM kernels carried
> this hack in  binfmt_aout.c to make it work in conjunction with the above:
> 
> static int load_aout_binary(struct linux_binprm * bprm)
> {
> ...
>         start_thread(regs, ex.a_entry, current->mm->start_stack);
> #ifndef __arm__
>         return 0;
> #else
>         return regs->ARM_r0;
> #endif
> }
> 
> ELF, on the other hand, never had that hack - ELF has always been zero
> in r0, and it's always retrieved the argc/argv/envp off the stack.

Aha, I was missing the a.out part, thanks. The ELF loader does call:

	ELF_PLAT_INIT(regs, reloc_func_desc);
	start_thread(regs, elf_entry, bprm->p);

back-to-back, which zeroes ARM_r0 and then promptly sticks argc there
straight away. libc uses the stack for everything (ARM_r0 gets clobbered
again with the return value of execve anyway), so we just have to assume
that nobody is using argv and/or envp from r1 and r2. I think that's a
sensible assumption.

> As the above hack got dropped from the kernel (I don't think it ever made
> it into mainline), I think we should be safe getting rid of this
> initialization of regs->ARM_r0 to r2, leaving them all as zeros.
> 
> We should probably also remove the selection of HAVE_AOUT from
> arch/arm/Kconfig too as this definitely won't work with any recent
> kernel (certainly not without binfmt_aout.c hacked in the above way.)

Sounds like a good idea, I'll write a patch...

Will