Re: [PATCH v2 1/3] tracing: kprobes: Prohibit probing on notrace function

[Masami Hiramatsu]

On Thu, 26 Jul 2018 14:53:27 +0900
Masami Hiramatsu <mhiramat@xxxxxxxxxx> wrote:

> Prohibit kprobe-events probing on notrace function.
> Since probing on the notrace function can cause recursive
> event call. In most case those are just skipped, but
> in some case it falls into infinit recursive call.
> 
> This protection can be disabled by the kconfig
> CONFIG_KPROBE_EVENTS_ON_NOTRACE=y, but it is highly
> recommended to keep it "n" for normal kernel.
> 
> Signed-off-by: Masami Hiramatsu <mhiramat@xxxxxxxxxx>
> Tested-by: Francis Deslauriers <francis.deslauriers@xxxxxxxxxxxx>
> ---
>   Changes from v1
>    - Add CONFIG_KPROBE_EVENTS_ON_NOTRACE kconfig for knocking down
>      the protection.
> ---
>  kernel/trace/Kconfig        |   18 ++++++++++++++++++
>  kernel/trace/trace_kprobe.c |   23 +++++++++++++++++++++++
>  2 files changed, 41 insertions(+)
> 
> diff --git a/kernel/trace/Kconfig b/kernel/trace/Kconfig
> index dcc0166d1997..24d5a58467a3 100644
> --- a/kernel/trace/Kconfig
> +++ b/kernel/trace/Kconfig
> @@ -456,6 +456,24 @@ config KPROBE_EVENTS
>  	  This option is also required by perf-probe subcommand of perf tools.
>  	  If you want to use perf tools, this option is strongly recommended.
>  
> +config KPROBE_EVENTS_ON_NOTRACE
> +	bool "Do NOT protect notrace function from kprobe events"
> +	depends on KPROBE_EVENTS
> +	default n
> +	help
> +	  This is only for the developers who want to debug ftrace itself
> +	  using kprobe events.
> +
> +	  Usually, ftrace related functions are protected from kprobe-events
> +	  to prevent an infinit recursion or any unexpected execution path
> +	  which leads to a kernel crash.
> +
> +	  This option disables such protection and allows you to put kprobe
> +	  events on ftrace functions for debugging ftrace by itself.
> +	  Note that this might let you shoot yourself in the foot.
> +
> +	  If unsure, say N.
> +
>  config UPROBE_EVENTS
>  	bool "Enable uprobes-based dynamic events"
>  	depends on ARCH_SUPPORTS_UPROBES
> diff --git a/kernel/trace/trace_kprobe.c b/kernel/trace/trace_kprobe.c
> index 27ace4513c43..1f1b4d712a7e 100644
> --- a/kernel/trace/trace_kprobe.c
> +++ b/kernel/trace/trace_kprobe.c
> @@ -496,6 +496,23 @@ disable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
>  	return ret;
>  }
>  
> +#ifdef CONFIG_KPROBE_EVENTS_ON_NOTRACE
> +#define within_notrace_func(tk)	(false)
> +#else
> +static bool within_notrace_func(struct trace_kprobe *tk)
> +{
> +	unsigned long offset, size, addr;
> +
> +	addr = kallsyms_lookup_name(trace_kprobe_symbol(tk));

Oops, I found this is wrong. (Thanks for old myself :P)
As ftracetest probepoint.tc said, kprobe-events can be
defined without symbol (direct address) for debugging.
In that case trace_kprobe_symbol will not return the symbol string.
I'll fix this series again.

Thank you,

> +	addr += trace_kprobe_offset(tk);
> +
> +	if (!kallsyms_lookup_size_offset(addr, &size, &offset))
> +		return true;	/* Out of range. */
> +
> +	return !ftrace_location_range(addr - offset, addr - offset + size);
> +}
> +#endif
> +
>  /* Internal register function - just handle k*probes and flags */
>  static int __register_trace_kprobe(struct trace_kprobe *tk)
>  {
> @@ -504,6 +521,12 @@ static int __register_trace_kprobe(struct trace_kprobe *tk)
>  	if (trace_probe_is_registered(&tk->tp))
>  		return -EINVAL;
>  
> +	if (within_notrace_func(tk)) {
> +		pr_warn("Could not probe notrace function %s\n",
> +			trace_kprobe_symbol(tk));
> +		return -EINVAL;
> +	}
> +
>  	for (i = 0; i < tk->tp.nr_args; i++)
>  		traceprobe_update_arg(&tk->tp.args[i]);
>  
> 


-- 
Masami Hiramatsu <mhiramat@xxxxxxxxxx>