Re: [PATCH 3/6] tracing: Change event_filter_read/write to verifyi_private != NULL

[Steven Rostedt]

On Tue, 2013-07-23 at 22:59 +0200, Oleg Nesterov wrote:
> event_filter_read/write() are racy, ftrace_event_call can be already
> freed by trace_remove_event_call() callers.
> 
> 1. Shift mutex_lock(event_mutex) from print/apply_event_filter to
>    the callers. Rename print/apply just in case.

I wouldn't do the rename unless there is a print_/apply_ version added
later. Just add a comment that event_mutex must be held for those
functions.

> 
> 2. Change the callers, event_filter_read() and event_filter_write()
>    to read i_private under this mutex and abort if it is NULL.
> 
> This fixes nothing, but now we can change debugfs_remove("filter")
> callers to nullify ->i_private and fix the the problem.
> 
> Signed-off-by: Oleg Nesterov <oleg@xxxxxxxxxx>
> ---
>  kernel/trace/trace.h               |    4 ++--
>  kernel/trace/trace_events.c        |   28 ++++++++++++++++++----------
>  kernel/trace/trace_events_filter.c |   21 ++++++++-------------
>  3 files changed, 28 insertions(+), 25 deletions(-)
> 
> diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h
> index afaae41..8b7c72b 100644
> --- a/kernel/trace/trace.h
> +++ b/kernel/trace/trace.h
> @@ -985,9 +985,9 @@ struct filter_pred {
>  
>  extern enum regex_type
>  filter_parse_regex(char *buff, int len, char **search, int *not);
> -extern void print_event_filter(struct ftrace_event_call *call,
> +extern void __print_event_filter(struct ftrace_event_call *call,
>  			       struct trace_seq *s);
> -extern int apply_event_filter(struct ftrace_event_call *call,
> +extern int __apply_event_filter(struct ftrace_event_call *call,
>  			      char *filter_string);
>  extern int apply_subsystem_event_filter(struct ftrace_subsystem_dir *dir,
>  					char *filter_string);
> diff --git a/kernel/trace/trace_events.c b/kernel/trace/trace_events.c
> index 821768e..0f081c0 100644
> --- a/kernel/trace/trace_events.c
> +++ b/kernel/trace/trace_events.c
> @@ -977,9 +977,9 @@ static ssize_t
>  event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
>  		  loff_t *ppos)
>  {
> -	struct ftrace_event_call *call = filp->private_data;
> +	struct ftrace_event_call *call;
>  	struct trace_seq *s;
> -	int r;
> +	int r = -ENODEV;
>  
>  	if (*ppos)
>  		return 0;
> @@ -987,14 +987,18 @@ event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
>  	s = kmalloc(sizeof(*s), GFP_KERNEL);
>  	if (!s)
>  		return -ENOMEM;
> -

Again, I prefer she space. Just my preference.

>  	trace_seq_init(s);
>  
> -	print_event_filter(call, s);
> -	r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
> +	mutex_lock(&event_mutex);
> +	call = event_file_data(filp);
> +	if (likely(call))

This isn't a fast path. Remove the "likely". It just makes the code
ugly.

> +		__print_event_filter(call, s);
> +	mutex_unlock(&event_mutex);
>  
> -	kfree(s);
> +	if (call)

Especially since we have no likely here.

> +		r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
>  
> +	kfree(s);
>  	return r;
>  }
>  
> @@ -1002,9 +1006,9 @@ static ssize_t
>  event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
>  		   loff_t *ppos)
>  {
> -	struct ftrace_event_call *call = filp->private_data;
> +	struct ftrace_event_call *call;
>  	char *buf;
> -	int err;
> +	int err = -ENODEV;
>  
>  	if (cnt >= PAGE_SIZE)
>  		return -EINVAL;
> @@ -1019,13 +1023,17 @@ event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
>  	}
>  	buf[cnt] = '\0';
>  
> -	err = apply_event_filter(call, buf);
> +	mutex_lock(&event_mutex);
> +	call = event_file_data(filp);
> +	if (likely(call))

Again, remove the likely.

> +		err = __apply_event_filter(call, buf);
> +	mutex_unlock(&event_mutex);
> +
>  	free_page((unsigned long) buf);
>  	if (err < 0)
>  		return err;
>  
>  	*ppos += cnt;
> -
>  	return cnt;
>  }
>  
> diff --git a/kernel/trace/trace_events_filter.c b/kernel/trace/trace_events_filter.c
> index 0c7b75a..71f76ee 100644
> --- a/kernel/trace/trace_events_filter.c
> +++ b/kernel/trace/trace_events_filter.c
> @@ -637,17 +637,15 @@ static void append_filter_err(struct filter_parse_state *ps,
>  	free_page((unsigned long) buf);
>  }
>  
> -void print_event_filter(struct ftrace_event_call *call, struct trace_seq *s)
> +/* caller must hold event_mutex */
> +void __print_event_filter(struct ftrace_event_call *call, struct trace_seq *s)
>  {
> -	struct event_filter *filter;
> +	struct event_filter *filter = call->filter;
>  
> -	mutex_lock(&event_mutex);
> -	filter = call->filter;
>  	if (filter && filter->filter_string)
>  		trace_seq_printf(s, "%s\n", filter->filter_string);
>  	else
>  		trace_seq_puts(s, "none\n");
> -	mutex_unlock(&event_mutex);
>  }
>  
>  void print_subsystem_event_filter(struct event_subsystem *system,
> @@ -1841,23 +1839,22 @@ static int create_system_filter(struct event_subsystem *system,
>  	return err;
>  }
>  
> -int apply_event_filter(struct ftrace_event_call *call, char *filter_string)
> +/* caller must hold event_mutex */
> +int __apply_event_filter(struct ftrace_event_call *call, char *filter_string)
>  {
>  	struct event_filter *filter;
>  	int err = 0;

We can remove the 0 init here.

>  
> -	mutex_lock(&event_mutex);
> -
>  	if (!strcmp(strstrip(filter_string), "0")) {
>  		filter_disable(call);
>  		filter = call->filter;
>  		if (!filter)
> -			goto out_unlock;
> +			goto out;

Remove the "out" label, and just return 0 here.

>  		RCU_INIT_POINTER(call->filter, NULL);
>  		/* Make sure the filter is not being used */
>  		synchronize_sched();
>  		__free_filter(filter);
> -		goto out_unlock;
> +		goto out;

Here too.

>  	}
>  
>  	err = create_filter(call, filter_string, true, &filter);
> @@ -1884,9 +1881,7 @@ int apply_event_filter(struct ftrace_event_call *call, char *filter_string)
>  			__free_filter(tmp);
>  		}
>  	}
> -out_unlock:
> -	mutex_unlock(&event_mutex);
> -
> +out:
>  	return err;
>  }
>  

-- Steve


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/