Re: arm64: implement live patching

[Miroslav Benes]

On Mon, 11 Jul 2016, Jessica Yu wrote:

> +++ Miroslav Benes [11/07/16 16:03 +0200]:
> > On Mon, 27 Jun 2016, Torsten Duwe wrote:
> > 
> > > diff --git a/arch/arm64/include/asm/livepatch.h
> > > b/arch/arm64/include/asm/livepatch.h
> > > new file mode 100644
> > > index 0000000..6b9a3d1
> > > --- /dev/null
> > > +++ b/arch/arm64/include/asm/livepatch.h
> > > @@ -0,0 +1,37 @@
> > > +/*
> > > + * livepatch.h - arm64-specific Kernel Live Patching Core
> > > + *
> > > + * Copyright (C) 2016 SUSE
> > > + *
> > > + * This program is free software; you can redistribute it and/or
> > > + * modify it under the terms of the GNU General Public License
> > > + * as published by the Free Software Foundation; either version 2
> > > + * of the License, or (at your option) any later version.
> > > + *
> > > + * This program is distributed in the hope that it will be useful,
> > > + * but WITHOUT ANY WARRANTY; without even the implied warranty of
> > > + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
> > > + * GNU General Public License for more details.
> > > + *
> > > + * You should have received a copy of the GNU General Public License
> > > + * along with this program; if not, see <http://www.gnu.org/licenses/>.
> > > + */
> > > +#ifndef _ASM_ARM64_LIVEPATCH_H
> > > +#define _ASM_ARM64_LIVEPATCH_H
> > > +
> > > +#include <linux/module.h>
> > > +#include <linux/ftrace.h>
> > > +
> > > +#ifdef CONFIG_LIVEPATCH
> > 
> > A nit but we removed such guards in the other header files.
> > 
> > > +static inline int klp_check_compiler_support(void)
> > > +{
> > > +	return 0;
> > > +}
> > > +
> > > +static inline void klp_arch_set_pc(struct pt_regs *regs, unsigned long
> > > ip)
> > > +{
> > > +	regs->pc = ip;
> > > +}
> > > +#endif /* CONFIG_LIVEPATCH */
> > 
> > I also checked mod_arch_specific structure because of the way we deal
> > with relocations. It is defined only if CONFIG_ARM64_MODULE_PLTS is
> > enabled and there is a pointer to 'struct elf64_shdr' called plt. It is
> > used indirectly in apply_relocate_add() so we need it to stay. However it
> > points to an existing Elf section and SHF_ALLOC is added to its sh_flags
> > in module_frob_arch_sections() (arch/arm64/kernel/module-plts.c).
> > Therefore we should be ok.
> > 
> > Jessica, could you check it as well, please?
> 
> That sounds right, the plt will remain in module core memory, so we
> are fine there.
> 
> However I think the plt->sh_size calculation will be incorrect for
> livepatch modules. In calculating mod->arch.plt_max_entries (see:
> module-plts.c), count_plts() is called for every rela section.
> For livepatch modules, this means count_plts() will also be called for
> our .klp.rela sections, which is correct behavior. However,
> count_plts() only considers relas referring to SHN_UNDEF symbols, and
> since every rela in a klp rela section refers to a SHN_LIVEPATCH
> symbol, these are all ignored. So count_plts() may return an incorrect
> value for a klp rela section.

You're right. During the patch module creation we basically transform all 
SHN_UNDEF relas to SHN_LIVEPATCH, right? We must take it into account 
here.
 
> Miroslav, can you confirm the issue? I think the fix would be easy
> though; we can just add an additional check for SHN_LIVEPATCH in
> count_plts(). 

Yes, such a check should be sufficient.

Thanks for looking into it.
Miroslav