Re: [PATCH v3 1/9] tracing: Add support for SOFT_DISABLE to syscall events

[Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>]

(2013/07/20 0:09), Tom Zanussi wrote:
> The original SOFT_DISABLE patches didn't add support for soft disable
> of syscall events; this adds it and paves the way for future patches
> allowing triggers to be added to syscall events, since triggers are
> built on top of SOFT_DISABLE.
> 
> Add an array of ftrace_event_file pointers indexed by syscall number
> to the trace array alongside the existing enabled bitmaps.  The
> ftrace_event_file structs in turn contain the soft disable flags we
> need for per-syscall soft disable accounting; later patches add
> additional 'trigger' flags and per-syscall triggers and filters.
> 
> Signed-off-by: Tom Zanussi <tom.zanussi@linux.intel.com>
> ---
>  kernel/trace/trace.h          |  2 ++
>  kernel/trace/trace_syscalls.c | 14 ++++++++++++++
>  2 files changed, 16 insertions(+)
> 
> diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h
> index 4a4f6e1..af6eb2c 100644
> --- a/kernel/trace/trace.h
> +++ b/kernel/trace/trace.h
> @@ -202,6 +202,8 @@ struct trace_array {
>  	int			sys_refcount_exit;
>  	DECLARE_BITMAP(enabled_enter_syscalls, NR_syscalls);
>  	DECLARE_BITMAP(enabled_exit_syscalls, NR_syscalls);
> +	struct ftrace_event_file *enter_syscall_files[NR_syscalls];
> +	struct ftrace_event_file *exit_syscall_files[NR_syscalls];
>  #endif
>  	int			stop_count;
>  	int			clock_id;
> diff --git a/kernel/trace/trace_syscalls.c b/kernel/trace/trace_syscalls.c
> index 322e164..4915b69 100644
> --- a/kernel/trace/trace_syscalls.c
> +++ b/kernel/trace/trace_syscalls.c
> @@ -302,6 +302,7 @@ static int __init syscall_exit_define_fields(struct ftrace_event_call *call)
>  static void ftrace_syscall_enter(void *data, struct pt_regs *regs, long id)
>  {
>  	struct trace_array *tr = data;
> +	struct ftrace_event_file *ftrace_file;
>  	struct syscall_trace_enter *entry;
>  	struct syscall_metadata *sys_data;
>  	struct ring_buffer_event *event;
> @@ -317,6 +318,10 @@ static void ftrace_syscall_enter(void *data, struct pt_regs *regs, long id)
>  	if (!test_bit(syscall_nr, tr->enabled_enter_syscalls))
>  		return;
>  
> +	ftrace_file = rcu_dereference_raw(tr->enter_syscall_files[syscall_nr]);
> +	if (test_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &ftrace_file->flags))
> +		return;
> +

It seems here is no lock to protect ftrace_file. This could be racy with
unreg_event_syscall_enter(). For example,

<thread0>                          <thread1>
ftrace_syscall_enter()             unreg_event_syscall_enter()
 test_bit(enabled_enter_syscalls)    lock(syscall_trace_lock);
                                     clear_bit(enabled_enter_syscalls)
                                     tr->enter_syscall_files[num] = NULL
 ftrace_file = tr->enter_syscall_files[syscall_nr]

In this case, ftrace_file can be NULL.
And even if it is passed(checked there), unreg_event_syscall_enter() still
doesn't ensure that the other threads are using the ftrace_file
(Not completely disabled) as previous kprobe-tracer doesn't.

Please check the below patch I posted.

http://marc.info/?l=linux-kernel&m=137336270104492

AFAIK, tracepoint (on which syscall tracer depends) locks rcu_read_lock
too, so I think it is possible to use same approach in syscall tracer to
guarantee all running event handlers have done.

Thank you,

>  	sys_data = syscall_nr_to_meta(syscall_nr);
>  	if (!sys_data)
>  		return;
> @@ -345,6 +350,7 @@ static void ftrace_syscall_enter(void *data, struct pt_regs *regs, long id)
>  static void ftrace_syscall_exit(void *data, struct pt_regs *regs, long ret)
>  {
>  	struct trace_array *tr = data;
> +	struct ftrace_event_file *ftrace_file;
>  	struct syscall_trace_exit *entry;
>  	struct syscall_metadata *sys_data;
>  	struct ring_buffer_event *event;
> @@ -359,6 +365,10 @@ static void ftrace_syscall_exit(void *data, struct pt_regs *regs, long ret)
>  	if (!test_bit(syscall_nr, tr->enabled_exit_syscalls))
>  		return;
>  
> +	ftrace_file = rcu_dereference_raw(tr->exit_syscall_files[syscall_nr]);
> +	if (test_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &ftrace_file->flags))
> +		return;
> +
>  	sys_data = syscall_nr_to_meta(syscall_nr);
>  	if (!sys_data)
>  		return;
> @@ -397,6 +407,7 @@ static int reg_event_syscall_enter(struct ftrace_event_file *file,
>  	if (!tr->sys_refcount_enter)
>  		ret = register_trace_sys_enter(ftrace_syscall_enter, tr);
>  	if (!ret) {
> +		rcu_assign_pointer(tr->enter_syscall_files[num], file);
>  		set_bit(num, tr->enabled_enter_syscalls);
>  		tr->sys_refcount_enter++;
>  	}
> @@ -416,6 +427,7 @@ static void unreg_event_syscall_enter(struct ftrace_event_file *file,
>  	mutex_lock(&syscall_trace_lock);
>  	tr->sys_refcount_enter--;
>  	clear_bit(num, tr->enabled_enter_syscalls);
> +	rcu_assign_pointer(tr->enter_syscall_files[num], NULL);
>  	if (!tr->sys_refcount_enter)
>  		unregister_trace_sys_enter(ftrace_syscall_enter, tr);
>  	mutex_unlock(&syscall_trace_lock);
> @@ -435,6 +447,7 @@ static int reg_event_syscall_exit(struct ftrace_event_file *file,
>  	if (!tr->sys_refcount_exit)
>  		ret = register_trace_sys_exit(ftrace_syscall_exit, tr);
>  	if (!ret) {
> +		rcu_assign_pointer(tr->exit_syscall_files[num], file);
>  		set_bit(num, tr->enabled_exit_syscalls);
>  		tr->sys_refcount_exit++;
>  	}
> @@ -454,6 +467,7 @@ static void unreg_event_syscall_exit(struct ftrace_event_file *file,
>  	mutex_lock(&syscall_trace_lock);
>  	tr->sys_refcount_exit--;
>  	clear_bit(num, tr->enabled_exit_syscalls);
> +	rcu_assign_pointer(tr->exit_syscall_files[num], NULL);
>  	if (!tr->sys_refcount_exit)
>  		unregister_trace_sys_exit(ftrace_syscall_exit, tr);
>  	mutex_unlock(&syscall_trace_lock);
> 


-- 
Masami HIRAMATSU
IT Management Research Dept. Linux Technology Center
Hitachi, Ltd., Yokohama Research Laboratory
E-mail: masami.hiramatsu.pt@hitachi.com