Re: [PATCH v8 2/7] powerpc/kprobes: Mark newly allocated probes as RO

[Russell Currey <ruscur@russell.cc>]

On Fri, 2020-04-03 at 00:18 +0530, Naveen N. Rao wrote:
> Naveen N. Rao wrote:
> > Russell Currey wrote:
> > > With CONFIG_STRICT_KERNEL_RWX=y and CONFIG_KPROBES=y, there will
> > > be one
> > > W+X page at boot by default.  This can be tested with
> > > CONFIG_PPC_PTDUMP=y and CONFIG_PPC_DEBUG_WX=y set, and checking
> > > the
> > > kernel log during boot.
> > > 
> > > powerpc doesn't implement its own alloc() for kprobes like other
> > > architectures do, but we couldn't immediately mark RO anyway
> > > since we do
> > > a memcpy to the page we allocate later.  After that, nothing
> > > should be
> > > allowed to modify the page, and write permissions are removed
> > > well
> > > before the kprobe is armed.
> > > 
> > > The memcpy() would fail if >1 probes were allocated, so use
> > > patch_instruction() instead which is safe for RO.
> > > 
> > > Reviewed-by: Daniel Axtens <dja@axtens.net>
> > > Signed-off-by: Russell Currey <ruscur@russell.cc>
> > > Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
> > > ---
> > >  arch/powerpc/kernel/kprobes.c | 17 +++++++++++++----
> > >  1 file changed, 13 insertions(+), 4 deletions(-)
> > > 
> > > diff --git a/arch/powerpc/kernel/kprobes.c
> > > b/arch/powerpc/kernel/kprobes.c
> > > index 81efb605113e..fa4502b4de35 100644
> > > --- a/arch/powerpc/kernel/kprobes.c
> > > +++ b/arch/powerpc/kernel/kprobes.c
> > > @@ -24,6 +24,8 @@
> > >  #include <asm/sstep.h>
> > >  #include <asm/sections.h>
> > >  #include <linux/uaccess.h>
> > > +#include <linux/set_memory.h>
> > > +#include <linux/vmalloc.h>
> > >  
> > >  DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL;
> > >  DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
> > > @@ -102,6 +104,16 @@ kprobe_opcode_t *kprobe_lookup_name(const
> > > char *name, unsigned int offset)
> > >  	return addr;
> > >  }
> > >  
> > > +void *alloc_insn_page(void)
> > > +{
> > > +	void *page = vmalloc_exec(PAGE_SIZE);
> > > +
> > > +	if (page)
> > > +		set_memory_ro((unsigned long)page, 1);
> > > +
> > > +	return page;
> > > +}
> > > +
> > 
> > This crashes for me with KPROBES_SANITY_TEST during the kretprobe
> > test.  
> 
> That isn't needed to reproduce this. After bootup, disabling
> optprobes 
> also shows the crash with kretprobes:
> 	sysctl debug.kprobes-optimization=0
> 
> The problem happens to be with patch_instruction() in 
> arch_prepare_kprobe(). During boot, on kprobe init, we register a
> probe 
> on kretprobe_trampoline for use with kretprobes (see 
> arch_init_kprobes()). This results in an instruction slot being 
> allocated, and arch_prepare_kprobe() to be called for copying the 
> instruction (nop) at kretprobe_trampoline. patch_instruction() is 
> failing resulting in corrupt instruction which we try to
> emulate/single 
> step causing the crash.

OK I think I've fixed it, KPROBES_SANITY_TEST passes too.  I'd
appreciate it if you could test v9, and thanks again for finding this -
very embarrassing bug on my side.

- Russell

> 
> 
> - Naveen
>