Re: [PATCH 1/2] tracing: Prevent unloadable modules from using trace_bprintk()

[Ingo Molnar <mingo@elte.hu>]

* Steven Rostedt <rostedt@goodmis.org> wrote:

> From: Steven Rostedt <srostedt@redhat.com>
> 
> While debugging a module, I found that unloading the module and
> then reading the ring buffer can cause strange side effects, including
> a kernel crash.
> 
> This is due to the trace_bprintk(). The trace_bprintk() is a faster
> version of trace_printk(). The difference is that trace_bprintk()
> only copies the arguments and a pointer to the format string into
> the ring buffer.
> 
> If a module uses this function and is unloaded, the pointer back to
> the format string in the module is still around. If the trace file
> is read, then the pointer is referenced and this can cause a kernel
> oops.
> 
> The simple solution is to not let modules use trace_bprintk() and
> instead it will use the slower version of this.
> 
> When talking with Frederic Weisbecker about it, he suggested not to
> punish modules that can not be unloaded since they do not have
> this side effect. Modules that can not be unloaded can still use
> trace_bprintk(). We added a check for MODVERSIONS to be set to make
> sure that the module and kernel have the same options. If you
> run without MODVERSIONS set, and you load a module that was compiled
> differently, then that's just your tough luck.
> 
> Cc: Frederic Weisbecker <fweisbec@gmail.com>
> Cc: Thomas Gleixner <tglx@linutronix.de>
> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
> ---
>  include/linux/kernel.h      |   21 +++++++++++++++++++--
>  kernel/trace/trace_printk.c |    2 ++
>  2 files changed, 21 insertions(+), 2 deletions(-)
> 
> diff --git a/include/linux/kernel.h b/include/linux/kernel.h
> index 2b0a35e..1003476 100644
> --- a/include/linux/kernel.h
> +++ b/include/linux/kernel.h
> @@ -538,6 +538,23 @@ do {									\
>  		____trace_printk_check_format(fmt, ##args);		\
>  } while (0)
>  
> +/*
> + * Module code must not use trace_bprintk, because if it is unloaded
> + * then we leave a pointer back to the module code inside
> + * the ring buffer, and then reading the ring buffer may cause a bug.
> + *
> + * We do allow for modules to use it if the kernel does not allow
> + * unloading of modules, and MODVERSIONS is set (to make sure kernel
> + * and module are the same). If you load modules without MODVERSIONS
> + * set, then you deserve what you get.
> + */
> +#if defined(MODULE) &&							\
> +	(defined(CONFIG_MODULE_UNLOAD) || !defined(CONFIG_MODVERSIONS))

Erm. Ignore checkpatch when the solution is to mess up the code ...

> +# define FORCE_TRACEPRINTK 1
> +#else
> +# define FORCE_TRACEPRINTK 0
> +#endif
> +
>  /**
>   * trace_printk - printf formatting in the ftrace buffer
>   * @fmt: the printf format for printing
> @@ -558,14 +575,14 @@ do {									\
>  #define trace_printk(fmt, args...)					\
>  do {									\
>  	__trace_printk_check_format(fmt, ##args);			\
> -	if (__builtin_constant_p(fmt)) {				\
> +	if (__builtin_constant_p(fmt) && !FORCE_TRACEPRINTK) {		\
>  		static const char *trace_printk_fmt			\
>  		  __attribute__((section("__trace_printk_fmt"))) =	\
>  			__builtin_constant_p(fmt) ? fmt : NULL;		\
>  									\
>  		__trace_bprintk(_THIS_IP_, trace_printk_fmt, ##args);	\
>  	} else								\
> -		__trace_printk(_THIS_IP_, fmt, ##args);		\
> +		__trace_printk(_THIS_IP_, fmt, ##args);			\
>  } while (0)
>  
>  extern int
> diff --git a/kernel/trace/trace_printk.c b/kernel/trace/trace_printk.c
> index 2547d88..c4a5db6 100644
> --- a/kernel/trace/trace_printk.c
> +++ b/kernel/trace/trace_printk.c
> @@ -115,7 +115,9 @@ int __trace_bprintk(unsigned long ip, const char *fmt, ...)
>  	va_end(ap);
>  	return ret;
>  }
> +#if !FORCE_TRACEPRINTK
>  EXPORT_SYMBOL_GPL(__trace_bprintk);
> +#endif

Looks quite ugly all around. Cannot suggest anything better though straight away - 
so please Cc: it more widely and get an ack from the module folks: Rusty, Linus, 
akpm.

Thanks,

	Ingo