[PATCH] perf bpf: Remove undefined behavior from bpf_perf_object__next

[PATCH] perf bpf: Remove undefined behavior from bpf_perf_object__next

[Ian Rogers]

bpf_perf_object__next folded the last element in the list test with the
empty list test. However, this meant that offsets were computed against
null and that a struct list_head was compared against a struct
bpf_perf_object. Working around this with clang's undefined behavior
sanitizer required -fno-sanitize=null and -fno-sanitize=object-size.

Remove the undefined behavior by using the regular Linux list APIs and
handling the starting case separately from the end testing case. Looking
at uses like bpf_perf_object__for_each, as the constant NULL or non-NULL
argument can be constant propagated the code is no less efficient.

Signed-off-by: Ian Rogers <irogers@google.com>
---
 tools/perf/util/bpf-loader.c | 18 +++++++-----------
 1 file changed, 7 insertions(+), 11 deletions(-)

diff --git a/tools/perf/util/bpf-loader.c b/tools/perf/util/bpf-loader.c
index f8ad581ea247..cdd6463a5b68 100644
--- a/tools/perf/util/bpf-loader.c
+++ b/tools/perf/util/bpf-loader.c
@@ -63,20 +63,16 @@ static struct hashmap *bpf_map_hash;
 static struct bpf_perf_object *
 bpf_perf_object__next(struct bpf_perf_object *prev)
 {
-	struct bpf_perf_object *next;
-
-	if (!prev)
-		next = list_first_entry(&bpf_objects_list,
-					struct bpf_perf_object,
-					list);
-	else
-		next = list_next_entry(prev, list);
+	if (!prev) {
+		if (list_empty(&bpf_objects_list))
+			return NULL;
 
-	/* Empty list is noticed here so don't need checking on entry. */
-	if (&next->list == &bpf_objects_list)
+		return list_first_entry(&bpf_objects_list, struct bpf_perf_object, list);
+	}
+	if (list_is_last(&prev->list, &bpf_objects_list))
 		return NULL;
 
-	return next;
+	return list_next_entry(prev, list);
 }
 
 #define bpf_perf_object__for_each(perf_obj, tmp)	\
-- 
2.37.1.359.gd136c6c3e2-goog

Re: [PATCH] perf bpf: Remove undefined behavior from bpf_perf_object__next

[Arnaldo Carvalho de Melo]

Em Tue, Jul 26, 2022 at 03:09:21PM -0700, Ian Rogers escreveu:
> bpf_perf_object__next folded the last element in the list test with the
> empty list test. However, this meant that offsets were computed against
> null and that a struct list_head was compared against a struct
> bpf_perf_object. Working around this with clang's undefined behavior
> sanitizer required -fno-sanitize=null and -fno-sanitize=object-size.
> in 
> Remove the undefined behav(ior by using the regular Linux list APIs and
> handling the starting case separately from the end testing case. Looking
> at uses like bpf_perf_object__for_each, as the constant NULL or non-NULL
> argument can be constant propagated the code is no less efficient.

Nicely spotted!

In some places people solve this with list_first_entry_or_null(), like
in cs_etm__queue_aux_records().

Applied.

- Arnado
 
> Signed-off-by: Ian Rogers <irogers@google.com>
> ---
>  tools/perf/util/bpf-loader.c | 18 +++++++-----------
>  1 file changed, 7 insertions(+), 11 deletions(-)
> 
> diff --git a/tools/perf/util/bpf-loader.c b/tools/perf/util/bpf-loader.c
> indelx f8ad581ea247..cdd6463a5b68 100644
> --- a/tools/perf/util/bpf-loader.c
> +++ b/tools/perf/util/bpf-loader.c
> @@ -63,20 +63,16 @@ static struct hashmap *bpf_map_hash;
>  static struct bpf_perf_object *
>  bpf_perf_object__next(struct bpf_perf_object *prev)
>  {
> -	struct bpf_perf_object *next;
> -
> -	if (!prev)
> -		next = list_first_entry(&bpf_objects_list,
> -					struct bpf_perf_object,
> -					list);
> -	else
> -		next = list_next_entry(prev, list);
> +	if (!prev) {
> +		if (list_empty(&bpf_objects_list))
> +			return NULL;
>  
> -	/* Empty list is noticed here so don't need checking on entry. */
> -	if (&next->list == &bpf_objects_list)
> +		return list_first_entry(&bpf_objects_list, struct bpf_perf_object, list);
> +	}
> +	if (list_is_last(&prev->list, &bpf_objects_list))
>  		return NULL;
>  
> -	return next;
> +	return list_next_entry(prev, list);
>  }
>  
>  #define bpf_perf_object__for_each(perf_obj, tmp)	\
> -- 
> 2.37.1.359.gd136c6c3e2-goog

-- 

- Arnaldo