Date:   Fri, 14 Dec 2018 16:34:23 -0500
From:   Joe Lawrence <joe.lawrence@...hat.com>
To:     Nicholas Mc Guire <hofrat@...dl.org>,
        Josh Poimboeuf <jpoimboe@...hat.com>
Cc:     Jessica Yu <jeyu@...nel.org>, Jiri Kosina <jikos@...nel.org>,
        Miroslav Benes <mbenes@...e.cz>,
        Petr Mladek <pmladek@...e.com>, live-patching@...r.kernel.org,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH 1/2] livepatch: fix non-static warnings

On 12/14/2018 11:56 AM, Nicholas Mc Guire wrote:
> Sparse reported warnings about non-static symbols. For the variables a
> simple static attribute is fine - for those symbols referenced by
> livepatch via klp_func the symbol-names must be unmodified in the
> relocation table - to resolve this the __noclone attribute (as 
  ^^^^^^^^^^
nit: symbol table

> suggested by Joe Lawrence <joe.lawrence@...hat.com>) is used
> for the statically declared functions.
> 
> Signed-off-by: Nicholas Mc Guire <hofrat@...dl.org>
> Link: https://lkml.org/lkml/2018/12/13/827
> ---
> 
> sparse reported the following warnings:
> 
>   CHECK   samples/livepatch/livepatch-shadow-fix1.c
> samples/livepatch/livepatch-shadow-fix1.c:74:14: warning: symbol
>  'livepatch_fix1_dummy alloc' was not declared. Should it be static?
> samples/livepatch/livepatch-shadow-fix1.c:116:6: warning: symbol
>  'livepatch_fix1_dummy free' was not declared. Should it be static?
> 
>   CHECK   samples/livepatch/livepatch-shadow-mod.c
> samples/livepatch/livepatch-shadow-mod.c:99:1: warning: symbol
>  'dummy_list' was not declared. Should it be static?
> samples/livepatch/livepatch-shadow-mod.c:100:1: warning: symbol
>  'dummy_list_mutex' was not declared. Should it be static?
> samples/livepatch/livepatch-shadow-mod.c:107:23: warning: symbol
>  'dummy_alloc' was not declared. Should it be static?
> samples/livepatch/livepatch-shadow-mod.c:132:15: warning: symbol
>  'dummy_free' was not declared. Should it be static?
> samples/livepatch/livepatch-shadow-mod.c:140:15: warning: symbol
>  'dummy_check' was not declared. Should it be static?
> 
> Patch was compile tested with: x86_64_defconfig + FTRACE=y
> FUNCTION_TRACER=y, EXPERT=y, LATENCYTOP=y, SAMPLES=y,
> SAMPLE_LIVEPATCH=y
> 
> Patch was runtested on an Intel i3 with:
>    insmod samples/livepatch/livepatch-shadow-mod.ko
>    insmod samples/livepatch/livepatch-shadow-fix1.ko
>    insmod samples/livepatch/livepatch-shadow-fix2.ko
>    echo 0 > /sys/kernel/livepatch/livepatch_shadow_fix2/enabled
>    echo 0 > /sys/kernel/livepatch/livepatch_shadow_fix1/enabled
>    rmmod livepatch-shadow-fix2
>    rmmod livepatch-shadow-fix1
>    rmmod livepatch-shadow-mod
> and dmesg output checked.
> 
> Patch is against 4.20-rc6 (localversion-next is next-20181214)

Great testing notes, thanks for including these!

>  samples/livepatch/livepatch-shadow-fix1.c |  4 ++--
>  samples/livepatch/livepatch-shadow-mod.c  | 16 +++++++++++-----
>  2 files changed, 13 insertions(+), 7 deletions(-)

Almost.  We should only need to annotate with __noclone for those
function definitions which the samples will be patching.  As the commit
message says, these correlate to klp_func.old_name functions found in
klp_object.name objects (.ko modules or NULL for vmlinux).

For the functions defined in samples/livepatch/*.c those would be:

  livepatch-callbacks-busymod.c :: busymod_work_func()
  livepatch-shadow-mod.c :: dummy_alloc()
  livepatch-shadow-mod.c :: dummy_free()
  livepatch-shadow-mod.c :: dummy_check()

So even though livepatch-shadow-fix2 further refines
livepatch-shadow-fix1, the livepatch core is going to redirect the
original dummy_*() calls defined by livepatch-shadow-mod.c in both fix1
and fix2 cases.  Ie, the fixes modules aren't patched, only the original.

> 
> diff --git a/samples/livepatch/livepatch-shadow-fix1.c b/samples/livepatch/livepatch-shadow-fix1.c
> index 49b1355..eaab10f 100644
> --- a/samples/livepatch/livepatch-shadow-fix1.c
> +++ b/samples/livepatch/livepatch-shadow-fix1.c
> @@ -71,7 +71,7 @@ static int shadow_leak_ctor(void *obj, void *shadow_data, void *ctor_data)
>  	return 0;
>  }
>  
> -struct dummy *livepatch_fix1_dummy_alloc(void)
> +static __noclone struct dummy *livepatch_fix1_dummy_alloc(void)
>  {
>  	struct dummy *d;
>  	void *leak;
> @@ -108,7 +108,7 @@ static void livepatch_fix1_dummy_leak_dtor(void *obj, void *shadow_data)
>  			 __func__, d, *shadow_leak);
>  }
>  
> -void livepatch_fix1_dummy_free(struct dummy *d)
> +static __noclone void livepatch_fix1_dummy_free(struct dummy *d)
>  {
>  	void **shadow_leak;
>  
> diff --git a/samples/livepatch/livepatch-shadow-mod.c b/samples/livepatch/livepatch-shadow-mod.c
> index 4c54b25..0a72bc2 100644
> --- a/samples/livepatch/livepatch-shadow-mod.c
> +++ b/samples/livepatch/livepatch-shadow-mod.c
> @@ -30,6 +30,11 @@
>   * memory leak, please load these modules at your own risk -- some
>   * amount of memory may leaked before the bug is patched.
>   *
> + * NOTE - the __noclone attribute to those functions that are to be
> + * shared with other modules while being declared static. As livepatch
> + * needs the unmodified symbol names and the usual "static" would
> + * invoke gccs cloning mechanism that renames the functions this
> + * needs to be suppressed with the additional __noclone attribute.

I like the idea of providing the sample code reader this information,
but since the compiler might also optimize livepatch-callbacks-busymod.c
:: busymod_work_func(), it too should be annotated __noclone.  Would
that file deserve a similar comment?

I don't have a strong opinion, but would throw my vote at leaving this
in the commit message only.


BTW, Petr/Miroslav/Josh, should we be annotating the selftests in
similar fashion?

> [ ... snip ... ]

Thanks for working on this,

-- Joe

Powered by blists - more mailing lists