[PATCH] kcov: check kcov_softirq in kcov_remote_stop()

[PATCH] kcov: check kcov_softirq in kcov_remote_stop()

[Andrey Konovalov]

kcov_remote_stop() should check that the corresponding kcov_remote_start()
actually found the specified remote handle and started collecting
coverage. This is done by checking the per thread kcov_softirq flag.

A particular failure scenario where this was observed involved a softirq
with a remote coverage collection section coming between check_kcov_mode()
and the access to t->kcov_area in __sanitizer_cov_trace_pc(). In that
softirq kcov_remote_start() bailed out after kcov_remote_find() check, but
the matching kcov_remote_stop() didn't check if kcov_remote_start()
succeeded, and overwrote per thread kcov parameters with invalid (zero)
values.

Fixes: 5ff3b30ab57d ("kcov: collect coverage from interrupts")
Signed-off-by: Andrey Konovalov <andreyknvl@google.com>
---
 kernel/kcov.c | 26 ++++++++++++++++++--------
 1 file changed, 18 insertions(+), 8 deletions(-)

diff --git a/kernel/kcov.c b/kernel/kcov.c
index 55c5d883a93e..6afae0bcbac4 100644
--- a/kernel/kcov.c
+++ b/kernel/kcov.c
@@ -427,7 +427,8 @@ void kcov_task_exit(struct task_struct *t)
 	 *        WARN_ON(!kcov->remote && kcov->t != t);
 	 *
 	 * For KCOV_REMOTE_ENABLE devices, the exiting task is either:
-	 * 2. A remote task between kcov_remote_start() and kcov_remote_stop().
+	 *
+	 * 1. A remote task between kcov_remote_start() and kcov_remote_stop().
 	 *    In this case we should print a warning right away, since a task
 	 *    shouldn't be exiting when it's in a kcov coverage collection
 	 *    section. Here t points to the task that is collecting remote
@@ -437,7 +438,7 @@ void kcov_task_exit(struct task_struct *t)
 	 *        WARN_ON(kcov->remote && kcov->t != t);
 	 *
 	 * 2. The task that created kcov exiting without calling KCOV_DISABLE,
-	 *    and then again we can make sure that t->kcov->t == t:
+	 *    and then again we make sure that t->kcov->t == t:
 	 *        WARN_ON(kcov->remote && kcov->t != t);
 	 *
 	 * By combining all three checks into one we get:
@@ -764,7 +765,7 @@ static const struct file_operations kcov_fops = {
  * Internally, kcov_remote_start() looks up the kcov device associated with the
  * provided handle, allocates an area for coverage collection, and saves the
  * pointers to kcov and area into the current task_struct to allow coverage to
- * be collected via __sanitizer_cov_trace_pc()
+ * be collected via __sanitizer_cov_trace_pc().
  * In turns kcov_remote_stop() clears those pointers from task_struct to stop
  * collecting coverage and copies all collected coverage into the kcov area.
  */
@@ -972,16 +973,25 @@ void kcov_remote_stop(void)
 		local_irq_restore(flags);
 		return;
 	}
-	kcov = t->kcov;
-	area = t->kcov_area;
-	size = t->kcov_size;
-	sequence = t->kcov_sequence;
-
+	/*
+	 * When in softirq, check if the corresponding kcov_remote_start()
+	 * actually found the remote handle and started collecting coverage.
+	 */
+	if (in_serving_softirq() && !t->kcov_softirq) {
+		local_irq_restore(flags);
+		return;
+	}
+	/* Make sure that kcov_softirq is only set when in softirq. */
 	if (WARN_ON(!in_serving_softirq() && t->kcov_softirq)) {
 		local_irq_restore(flags);
 		return;
 	}
 
+	kcov = t->kcov;
+	area = t->kcov_area;
+	size = t->kcov_size;
+	sequence = t->kcov_sequence;
+
 	kcov_stop(t);
 	if (in_serving_softirq()) {
 		t->kcov_softirq = 0;
-- 
2.27.0.278.ge193c7cf3a9-goog

Re: [PATCH] kcov: check kcov_softirq in kcov_remote_stop()

[Dmitry Vyukov]

On Mon, Jun 8, 2020 at 2:42 AM Andrey Konovalov <andreyknvl@google.com> wrote:
>
> kcov_remote_stop() should check that the corresponding kcov_remote_start()
> actually found the specified remote handle and started collecting
> coverage. This is done by checking the per thread kcov_softirq flag.
>
> A particular failure scenario where this was observed involved a softirq
> with a remote coverage collection section coming between check_kcov_mode()
> and the access to t->kcov_area in __sanitizer_cov_trace_pc(). In that
> softirq kcov_remote_start() bailed out after kcov_remote_find() check, but
> the matching kcov_remote_stop() didn't check if kcov_remote_start()
> succeeded, and overwrote per thread kcov parameters with invalid (zero)
> values.
>
> Fixes: 5ff3b30ab57d ("kcov: collect coverage from interrupts")
> Signed-off-by: Andrey Konovalov <andreyknvl@google.com>

Reviewed-by: Dmitry Vyukov <dvyukov@google.com>

> ---
>  kernel/kcov.c | 26 ++++++++++++++++++--------
>  1 file changed, 18 insertions(+), 8 deletions(-)
>
> diff --git a/kernel/kcov.c b/kernel/kcov.c
> index 55c5d883a93e..6afae0bcbac4 100644
> --- a/kernel/kcov.c
> +++ b/kernel/kcov.c
> @@ -427,7 +427,8 @@ void kcov_task_exit(struct task_struct *t)
>          *        WARN_ON(!kcov->remote && kcov->t != t);
>          *
>          * For KCOV_REMOTE_ENABLE devices, the exiting task is either:
> -        * 2. A remote task between kcov_remote_start() and kcov_remote_stop().
> +        *
> +        * 1. A remote task between kcov_remote_start() and kcov_remote_stop().
>          *    In this case we should print a warning right away, since a task
>          *    shouldn't be exiting when it's in a kcov coverage collection
>          *    section. Here t points to the task that is collecting remote
> @@ -437,7 +438,7 @@ void kcov_task_exit(struct task_struct *t)
>          *        WARN_ON(kcov->remote && kcov->t != t);
>          *
>          * 2. The task that created kcov exiting without calling KCOV_DISABLE,
> -        *    and then again we can make sure that t->kcov->t == t:
> +        *    and then again we make sure that t->kcov->t == t:
>          *        WARN_ON(kcov->remote && kcov->t != t);
>          *
>          * By combining all three checks into one we get:
> @@ -764,7 +765,7 @@ static const struct file_operations kcov_fops = {
>   * Internally, kcov_remote_start() looks up the kcov device associated with the
>   * provided handle, allocates an area for coverage collection, and saves the
>   * pointers to kcov and area into the current task_struct to allow coverage to
> - * be collected via __sanitizer_cov_trace_pc()
> + * be collected via __sanitizer_cov_trace_pc().
>   * In turns kcov_remote_stop() clears those pointers from task_struct to stop
>   * collecting coverage and copies all collected coverage into the kcov area.
>   */
> @@ -972,16 +973,25 @@ void kcov_remote_stop(void)
>                 local_irq_restore(flags);
>                 return;
>         }
> -       kcov = t->kcov;
> -       area = t->kcov_area;
> -       size = t->kcov_size;
> -       sequence = t->kcov_sequence;
> -
> +       /*
> +        * When in softirq, check if the corresponding kcov_remote_start()
> +        * actually found the remote handle and started collecting coverage.
> +        */
> +       if (in_serving_softirq() && !t->kcov_softirq) {
> +               local_irq_restore(flags);
> +               return;
> +       }
> +       /* Make sure that kcov_softirq is only set when in softirq. */
>         if (WARN_ON(!in_serving_softirq() && t->kcov_softirq)) {
>                 local_irq_restore(flags);
>                 return;
>         }
>
> +       kcov = t->kcov;
> +       area = t->kcov_area;
> +       size = t->kcov_size;
> +       sequence = t->kcov_sequence;
> +
>         kcov_stop(t);
>         if (in_serving_softirq()) {
>                 t->kcov_softirq = 0;
> --
> 2.27.0.278.ge193c7cf3a9-goog
>