Re: [PATCH] objtool: Documentation: document UACCESS warnings

[Peter Zijlstra <peterz@infradead.org>]

On Mon, Mar 23, 2020 at 05:13:20PM -0700, Nick Desaulniers wrote:
> Compiling with Clang and CONFIG_KASAN=y was exposing a few warnings:
>   call to memset() with UACCESS enabled
> 
> Document how to fix these for future travelers.
> 
> Link: https://github.com/ClangBuiltLinux/linux/issues/876
> Suggested-by: Peter Zijlstra <peterz@infradead.org>
> Signed-off-by: Nick Desaulniers <ndesaulniers@google.com>
> ---
>  .../Documentation/stack-validation.txt        | 20 +++++++++++++++++++
>  1 file changed, 20 insertions(+)
> 
> diff --git a/tools/objtool/Documentation/stack-validation.txt b/tools/objtool/Documentation/stack-validation.txt
> index de094670050b..156fee13ba02 100644
> --- a/tools/objtool/Documentation/stack-validation.txt
> +++ b/tools/objtool/Documentation/stack-validation.txt
> @@ -289,6 +289,26 @@ they mean, and suggestions for how to fix them.
>        might be corrupt due to a gcc bug.  For more details, see:
>        https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70646
>  
> +9. file.o: warning: objtool: funcA() call to funcB() with UACCESS enabled
> +
> +   This means that an unexpected call to a non-whitelisted function exists
> +   outside of arch-specific guards.
> +   X86: SMAP (stac/clac): __uaccess_begin()/__uaccess_end()
> +   ARM: PAN: uaccess_enable()/uaccess_enable()
> +
> +   These functions should called to denote a minimal critical section around
> +   access to __user variables. See also: https://lwn.net/Articles/517475/
> +
> +   The intention of the warning is to prevent calls to funcB() from eventually
> +   calling schedule(), potentially leaking the AC flags state, and not
> +   restoring them correctly.
> +
> +   To fix, either:
> +   1) add the correct guards before and after calls to low level functions like
> +      __get_user_size()/__put_user_size().
> +   2) add funcB to uaccess_safe_builtin whitelist in tools/objtool/check.c, if
> +      funcB obviously does not call schedule().

There's sadly nothing obvious about 2); __fentry__ is enough to end up
in schedule() through preempt_enable().

So any function that has function tracing on (mostly everything) is
already disqualified.