[PATCH 3/3] usdt parser: make sure prbmap is cleaned up in case of error

[PATCH 3/3] usdt parser: make sure prbmap is cleaned up in case of error

[Kris Van Hees]

Signed-off-by: Kris Van Hees <kris.van.hees@oracle.com>
---
 libcommon/usdt_parser_notes.c | 13 +++++++------
 1 file changed, 7 insertions(+), 6 deletions(-)

diff --git a/libcommon/usdt_parser_notes.c b/libcommon/usdt_parser_notes.c
index d3d744fb..c98c9fb0 100644
--- a/libcommon/usdt_parser_notes.c
+++ b/libcommon/usdt_parser_notes.c
@@ -829,15 +829,16 @@ usdt_parse_notes(int out, dof_helper_t *dhp, usdt_data_t *data)
 	goto out;
 
 err:
-	rc = -1;
-
-out:
 	/*
-	 * All tracepoint probes in prbmap should have been removed during
-	 * proessing.
+	 * In case of an error, we don't know whether we can allocate iterators
+	 * so we have few options to provide cleanup.  Since cleanup is all we
+	 * do now, we plug the probe cleanup function into the ops for prbmap,
+	 * and let the hashtable destroy function take care of all cleanup.
 	 */
-	assert(dt_htab_entries(prbmap) == 0);
+	prb_htab_ops.del = (htab_del_fn)prb_del_probe;
+	rc = -1;
 
+out:
 	dt_htab_destroy(prvmap);
 	dt_htab_destroy(prbmap);
 
-- 
2.43.5

Re: [PATCH 3/3] usdt parser: make sure prbmap is cleaned up in case of error

[Eugene Loh]

Reviewed-by: Eugene Loh <eugene.loh@oracle.com>
but it'd be great if Nick could take a look, given that he reviewed the 
file earlier.
Out of curiosity, was there some case where the assertion failed? Could 
that be a test case?
(Minor nit:  I'd go for a sentence break after "allocate iterators.")

On 6/24/25 17:40, Kris Van Hees wrote:
> Signed-off-by: Kris Van Hees <kris.van.hees@oracle.com>
> ---
>   libcommon/usdt_parser_notes.c | 13 +++++++------
>   1 file changed, 7 insertions(+), 6 deletions(-)
>
> diff --git a/libcommon/usdt_parser_notes.c b/libcommon/usdt_parser_notes.c
> index d3d744fb..c98c9fb0 100644
> --- a/libcommon/usdt_parser_notes.c
> +++ b/libcommon/usdt_parser_notes.c
> @@ -829,15 +829,16 @@ usdt_parse_notes(int out, dof_helper_t *dhp, usdt_data_t *data)
>   	goto out;
>   
>   err:
> -	rc = -1;
> -
> -out:
>   	/*
> -	 * All tracepoint probes in prbmap should have been removed during
> -	 * proessing.
> +	 * In case of an error, we don't know whether we can allocate iterators
> +	 * so we have few options to provide cleanup.  Since cleanup is all we
> +	 * do now, we plug the probe cleanup function into the ops for prbmap,
> +	 * and let the hashtable destroy function take care of all cleanup.
>   	 */
> -	assert(dt_htab_entries(prbmap) == 0);
> +	prb_htab_ops.del = (htab_del_fn)prb_del_probe;
> +	rc = -1;
>   
> +out:
>   	dt_htab_destroy(prvmap);
>   	dt_htab_destroy(prbmap);
>   

Re: [PATCH 3/3] usdt parser: make sure prbmap is cleaned up in case of error

[Kris Van Hees]

On Mon, Jun 30, 2025 at 06:07:39PM -0400, Eugene Loh wrote:
> Reviewed-by: Eugene Loh <eugene.loh@oracle.com>
> but it'd be great if Nick could take a look, given that he reviewed the file
> earlier.
> Out of curiosity, was there some case where the assertion failed? Could that
> be a test case?

There was an assertion failure but not one that we can have a test for since
it is in dtprobed and the testsuite cannot capture those at this point.

> (Minor nit:  I'd go for a sentence break after "allocate iterators.")
> 
> On 6/24/25 17:40, Kris Van Hees wrote:
> > Signed-off-by: Kris Van Hees <kris.van.hees@oracle.com>
> > ---
> >   libcommon/usdt_parser_notes.c | 13 +++++++------
> >   1 file changed, 7 insertions(+), 6 deletions(-)
> > 
> > diff --git a/libcommon/usdt_parser_notes.c b/libcommon/usdt_parser_notes.c
> > index d3d744fb..c98c9fb0 100644
> > --- a/libcommon/usdt_parser_notes.c
> > +++ b/libcommon/usdt_parser_notes.c
> > @@ -829,15 +829,16 @@ usdt_parse_notes(int out, dof_helper_t *dhp, usdt_data_t *data)
> >   	goto out;
> >   err:
> > -	rc = -1;
> > -
> > -out:
> >   	/*
> > -	 * All tracepoint probes in prbmap should have been removed during
> > -	 * proessing.
> > +	 * In case of an error, we don't know whether we can allocate iterators
> > +	 * so we have few options to provide cleanup.  Since cleanup is all we
> > +	 * do now, we plug the probe cleanup function into the ops for prbmap,
> > +	 * and let the hashtable destroy function take care of all cleanup.
> >   	 */
> > -	assert(dt_htab_entries(prbmap) == 0);
> > +	prb_htab_ops.del = (htab_del_fn)prb_del_probe;
> > +	rc = -1;
> > +out:
> >   	dt_htab_destroy(prvmap);
> >   	dt_htab_destroy(prbmap);

Re: [PATCH 3/3] usdt parser: make sure prbmap is cleaned up in case of error

[Nick Alcock]

On 24 Jun 2025, Kris Van Hees said:

> Signed-off-by: Kris Van Hees <kris.van.hees@oracle.com>
> ---
>  libcommon/usdt_parser_notes.c | 13 +++++++------
>  1 file changed, 7 insertions(+), 6 deletions(-)
>
> diff --git a/libcommon/usdt_parser_notes.c b/libcommon/usdt_parser_notes.c
> index d3d744fb..c98c9fb0 100644
> --- a/libcommon/usdt_parser_notes.c
> +++ b/libcommon/usdt_parser_notes.c
> @@ -829,15 +829,16 @@ usdt_parse_notes(int out, dof_helper_t *dhp, usdt_data_t *data)
>  	goto out;
>  
>  err:
> -	rc = -1;
> -
> -out:
>  	/*
> -	 * All tracepoint probes in prbmap should have been removed during
> -	 * proessing.
> +	 * In case of an error, we don't know whether we can allocate iterators
> +	 * so we have few options to provide cleanup.  Since cleanup is all we
> +	 * do now, we plug the probe cleanup function into the ops for prbmap,
> +	 * and let the hashtable destroy function take care of all cleanup.
>  	 */
> -	assert(dt_htab_entries(prbmap) == 0);
> +	prb_htab_ops.del = (htab_del_fn)prb_del_probe;
> +	rc = -1;
>  
> +out:
>  	dt_htab_destroy(prvmap);
>  	dt_htab_destroy(prbmap);

That seems... better than we had before. I was seeing double-frees all
over the place with what you previously committed: I'll have to see if
this helps.

-- 
NULL && (void)