Re: [PATCH] libtrace: Fix memory leak in tracefs_cpu
Steven Rostedt <[email protected]> Thu, 3 Apr 2025 15:14:04 -0400
| Newsgroups | org.kernel.vger.linux-trace-devel |
|---|---|
| Message-ID | <[email protected]> |
On Mon, 17 Mar 2025 11:24:36 +0000 "Kutzer, Philipp" <[email protected]> wrote: > From: Philipp Kutzer <[email protected]> > > Hi, > > with the introduction of the interface to allow the tracefs_cpu to handle buffer allocation on their > own a memory leak was built in. > (Commit 1e1cc549d7e7fa408081e03e4effd3591a131ae5 - libtracefs: Add API to read tracefs_cpu > and return a kbuffer) > > We cyclically call the tracefs_iterate_raw_events function from one of our applications and > noticed that our application was increasingly allocating memory. > > With the help of Address Sanitizer the root cause, allocating but not freeing memory > for "tcpu->buffer" was found. The fix will free memory during close and cleanup function > and has been successfully tested in our environment. The original problem (memory leak) > has been resolved. Note, I cleaned up the change log: libtrace: Fix memory leak in tracefs_cpu The introduction of the interface to allow the tracefs_cpu to handle buffer allocation on their own caused a memory leak. (Commit 1e1cc549d7 ("libtracefs: Add API to read tracefs_cpu and return a kbuffer")). The help of Address Sanitizer found the root cause. The "tcpu->buffer" was being allocated but not freed. The fix will free memory during close and cleanup functions. > > Signed-off-by: Philipp Kutzer mailto:[email protected] Also, this should be: Signed-off-by: Philipp Kutzer <[email protected]> > --- > > diff --git a/src/tracefs-record.c b/src/tracefs-record.c > index a2dcfc2..4cd0917 100644 > --- a/src/tracefs-record.c > +++ b/src/tracefs-record.c > @@ -293,6 +293,8 @@ void tracefs_cpu_close(struct tracefs_cpu *tcpu) > if (!tcpu) > return; > > + free(tcpu->buffer); > + tcpu->buffer = NULL; > close(tcpu->fd); > tracefs_cpu_free_fd(tcpu); > } This also had whitespace issues. I fixed it up and applied it. Thanks, -- Steve