Re: Detecting user data on base types

On Thu, Jun 06, 2019 at 01:15:09PM +0300, Dan Carpenter wrote:
> On Wed, Jun 05, 2019 at 01:28:41PM +0100, Andrew Murray wrote:
> > Actually it looks there is logic associated to get_user in handle_get_user within
> > smatch_kernel_user_data.c - however this doesn't seem to trigger in the following (it looks
> > like it does something with the return value of the macro, but this is just an error flag):
> > 
> > do_pages_move -calls-> add_page_for_migration - the second argument of
> > add_page_for_migration 'addr' comes from get_user - but it isn't marked as USER_DATA in
> > the database. I don't understand why this is.
> > 
> 
> That's weird.  It's working for me.
> 
>         mm/migrate.c |        do_pages_move | add_page_for_migration |   USER_DATA |  1 | addr | 0-u64max
> 
> Btw, I pushed the fix for untagged_addr() and my DB was built with that
> fix.  You could try see if that makes a difference, I guess?
> 
> kchecker --info mm/migrate.c > info.txt
> ~/path/to/smatch/smatch_data/db/reload_partial.sh info.txt

I still don't see it. I'll investigate further.

(I'm running the tools on x86_64, cross building for ARM64 - I'll share some patches in
due course for cross-compilation).

Thanks,

Andrew Murray

> 
> regards,
> dan carpenter