Re: [RFC PATCH 0/2] improve vmalloc allocation

[Michal Hocko]

On Wed 24-10-18 19:34:18, Uladzislau Rezki wrote:
> Hi.
> 
> On Wed, Oct 24, 2018 at 08:22:52AM +0200, Michal Hocko wrote:
> > On Tue 23-10-18 12:30:44, Joel Fernandes wrote:
> > > On Tue, Oct 23, 2018 at 11:13:36AM -0600, Shuah Khan wrote:
> > > > On 10/23/2018 11:05 AM, Michal Hocko wrote:
> > > > > On Tue 23-10-18 08:26:40, Matthew Wilcox wrote:
> > > > >> On Tue, Oct 23, 2018 at 09:02:56AM -0600, Shuah Khan wrote:
> > > > > [...]
> > > > >>> The way it can be handled is by adding a test module under lib. test_kmod,
> > > > >>> test_sysctl, test_user_copy etc.
> > > > >>
> > > > >> The problem is that said module can only invoke functions which are
> > > > >> exported using EXPORT_SYMBOL.  And there's a cost to exporting them,
> > > > >> which I don't think we're willing to pay, purely to get test coverage.
> > > > > 
> > > > > Yes, I think we do not want to export internal functionality which might
> > > > > be still interesting for the testing coverage. Maybe we want something
> > > > > like EXPORT_SYMBOL_KSELFTEST which would allow to link within the
> > > > > kselftest machinery but it wouldn't allow the same for general modules
> > > > > and will not give any API promisses.
> > > > > 
> > > > 
> > > > I like this proposal. I think we will open up lot of test opportunities with
> > > > this approach.
> > > > 
> > > > Maybe we can use this stress test as a pilot and see where it takes us.
> > > 
> > > I am a bit worried that such an EXPORT_SYMBOL_KSELFTEST mechanism can be abused by
> > > out-of-tree module writers to call internal functionality.
> > > 
> > > How would you prevent that?
> > 
> > There is no way to prevent non-exported symbols abuse by 3rd party
> > AFAIK. EXPORT_SYMBOL_* is not there to prohibid abuse. It is a mere
> > signal of what is, well, an exported API.
> 
> Can we just use kallsyms_lookup_name()?

Heh, this is the abuse I've had in mind ;)
> <snip>
> static void *((*__my_vmalloc_node_range)(unsigned long size,
>     unsigned long align,unsigned long start, unsigned long end,
>     gfp_t gfp_mask,pgprot_t prot, unsigned long vm_flags,
>     int node, const void *caller));
> 
> __my_vmalloc_node_range = (void *) kallsyms_lookup_name("__vmalloc_node_range");
> <snip>

This is just too ugly to live. So I would go with it only if there is no
reasonable way to export what tests need with a sane interface.
-- 
Michal Hocko
SUSE Labs