Re: [PATCH v3 1/3] kprobes: Add text_alloc() and text_free()

[Mike Rapoport <rppt@linux.ibm.com>]

On Wed, Jul 15, 2020 at 12:36:02PM -0700, Kees Cook wrote:
> On Wed, Jul 15, 2020 at 01:32:27AM +0300, Jarkko Sakkinen wrote:
> > Introduce new API for allocating space for code generaed at run-time
> > leveraging from module_alloc() and module_memfree() code. Use this to
> > perform memory allocations in the kprobes code in order to loose the
> > bound between kprobes and modules subsystem.
> > 
> > Initially, use this API only with arch/x86 and define a new config
> > flag CONFIG_ARCH_HAS_TEXT_ALLOC to promote the availability of the
> > new API.
> > [...]
> > diff --git a/include/linux/text.h b/include/linux/text.h
> > new file mode 100644
> > index 000000000000..a27d4a42ecda
> > --- /dev/null
> > +++ b/include/linux/text.h
> > @@ -0,0 +1,17 @@
> > +/* SPDX-License-Identifier: GPL-2.0-only */
> > +
> > +#ifndef _LINUX_TEXT_H
> > +#define _LINUX_TEXT_H
> > +
> > +/*
> > + * An allocator used for allocating modules, core sections and init sections.
> > + * Returns NULL on failure.
> > + */
> > +void *text_alloc(unsigned long size);
> > +
> > +/*
> > + * Free memory returned from text_alloc().
> > + */
> > +void text_free(void *region);
> > +
> > +#endif /* _LINUX_TEXT_H */
> > diff --git a/kernel/kprobes.c b/kernel/kprobes.c
> > index 2e97febeef77..fa7687eb0c0e 100644
> > --- a/kernel/kprobes.c
> > +++ b/kernel/kprobes.c
> > @@ -35,6 +35,7 @@
> >  #include <linux/ftrace.h>
> >  #include <linux/cpu.h>
> >  #include <linux/jump_label.h>
> > +#include <linux/text.h>
> >  
> >  #include <asm/sections.h>
> >  #include <asm/cacheflush.h>
> > @@ -111,12 +112,20 @@ enum kprobe_slot_state {
> >  
> >  void __weak *alloc_insn_page(void)
> >  {
> > +#ifdef CONFIG_ARCH_HAS_TEXT_ALLOC
> > +	return text_alloc(PAGE_SIZE);
> > +#else
> >  	return module_alloc(PAGE_SIZE);
> > +#endif
> 
> This seems like it shouldn't be needed. I think text_alloc() should
> always be visible. In the ARCH_HAS... case it will call the arch
> implementation, and without it should just call module_alloc():
> 
> For example:
> void *text_alloc(unsigned long size)
> {
> #ifdef CONFIG_ARCH_HAS_TEXT_ALLOC
> 	return arch_text_alloc(size);
> #else
> 	return module_alloc(size);
> #endif
> }

This inverts the dependcy chain, IMHO, module_alloc() is a special case
of text_alloc() and not vice versa.

The addition of #ifdefs to alloc_insn_page() is not needed because it is
anyway overriden on x86, so I don't see a point in added #ifdefs until
another arch will enable ARCH_HAS_TEXT_ALLOC.

> -- 
> Kees Cook

-- 
Sincerely yours,
Mike.