Re: [PATCH RFC 3/7] tracing: Stop modifying the input buffer in ftrace_set_clr_event()

[Steven Rostedt <rostedt@goodmis.org>]

On Thu, 13 Aug 2026 14:15:00 +0000
sashiko-bot@kernel.org wrote:

> Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
> - [High] Invalid free (causing kernel panic) and memory leak due to `strsep()` modifying a pointer managed by `__free(kfree)`.
> --
> 
> --- Patch [3]: [PATCH RFC 3/7] tracing: Stop modifying the input buffer in ftrace_set_clr_event() ---
> commit c968a27c39638474537de841b7e2191407463142
> Author: Thomas Weißschuh <thomas.weissschuh@linutronix.de>
> 
> tracing: Stop modifying the input buffer in ftrace_set_clr_event()
> 
> This commit changes ftrace_set_clr_event() to copy its input buffer using
> kstrdup() and manage it with __free(kfree). This removes the need to
> restore the original buffer modifications before returning.
> 
> > diff --git a/kernel/trace/trace_events.c b/kernel/trace/trace_events.c
> > index 6aa32c4922807..c75a7f5b55def 100644
> > --- a/kernel/trace/trace_events.c
> > +++ b/kernel/trace/trace_events.c
> > @@ -1411,14 +1411,17 @@ static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
> >  	return ret;
> >  }
> >  
> > -int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
> > +int ftrace_set_clr_event(struct trace_array *tr, const char *_buf, int set)
> >  {
> >  	char *event = NULL, *sub = NULL, *match, *mod;
> >  
> >  	if (!tr)
> >  		return -ENOENT;
> >  
> > +	char *buf __free(kfree) = kstrdup(_buf, GFP_KERNEL);
> > +	if (!buf)
> > +		return -ENOMEM;
> > +  
> 
> [Severity: High]
> Will this lead to an invalid free or memory leak when buf is later modified? 
> 
> Further down in ftrace_set_clr_event(), buf is passed to strsep(), which
> modifies the pointer in place:
> 
>     match = strsep(&buf, ":");
> 
> If strsep() advances the pointer to the next token, the cleanup function
> will call kfree() on an interior pointer, causing a slab corruption bug or
> kernel panic upon scope exit. If buf is advanced to NULL (when no colon is
> found), it will result in a memory leak of the duplicated string since
> kfree(NULL) is a no-op.
> 
> Could a separate pointer be used to track the allocation for cleanup,
> leaving buf free to be safely advanced by strsep()?

Yes, this is a legitimate bug.

I guess you need to have:

int ftrace_set_clr_event(struct trace_array *tr, const char *buffer, int set)
{
	char *event = NULL, *sub = NULL, *match, *mod;
	char *buf;
	int ret;

	if (!tr)
		return -ENOENT;

	char *_buf __free(kfree) = kstrdup(_buf, GFP_KERNEL);
	if (!_buf)
		return -ENOMEM;

	/* Need to keep _buf pointing to the start of the buffer */
	buf = _buf;

-- Steve


> 
> >  	/* Modules events can be appended with :mod:<module> */
> >  	mod = strstr(buf, ":mod:");
> >  	if (mod) {  
> 
> [ ... ]
> 
> > @@ -1455,15 +1458,7 @@ int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
> >  			match = NULL;
> >  	}
> >  
> > -	ret = __ftrace_set_clr_event(tr, match, sub, event, set, mod);
> > -
> > -	/* Put back the colon to allow this to be called again */
> > -	if (buf)
> > -		*(buf - 1) = ':';
> > -	if (mod)
> > -		*(mod - 5) = ':';
> > -
> > -	return ret;
> > +	return __ftrace_set_clr_event(tr, match, sub, event, set, mod);
> >  }  
>