[v1] ptp: Return -EINVAL on ptp_clock_register if required ops are NULL

[PATCH net] ptp: Return -EINVAL on ptp_clock_register if required ops are NULL

Posted by Tim Hostetler 3 months, 1 week ago

ptp_clock should never be registered unless it stubs one of gettimex64()
or gettime64() and settime64(). WARN_ON_ONCE and error out if either set
of function pointers is null.

Cc: stable@vger.kernel.org
Fixes: d7d38f5bd7be ("ptp: use the 64 bit get/set time methods for the posix clock.")
Suggested-by: Kuniyuki Iwashima <kuniyu@google.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@google.com>
Reviewed-by: Harshitha Ramamurthy <hramamurthy@google.com>
Signed-off-by: Tim Hostetler <thostet@google.com>
---
 drivers/ptp/ptp_clock.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/drivers/ptp/ptp_clock.c b/drivers/ptp/ptp_clock.c
index ef020599b771..0bc79076771b 100644
--- a/drivers/ptp/ptp_clock.c
+++ b/drivers/ptp/ptp_clock.c
@@ -325,6 +325,10 @@ struct ptp_clock *ptp_clock_register(struct ptp_clock_info *info,
 	if (info->n_alarm > PTP_MAX_ALARMS)
 		return ERR_PTR(-EINVAL);
 
+	if (WARN_ON_ONCE((!info->gettimex64 && !info->gettime64) ||
+			 !info->settime64))
+		return ERR_PTR(-EINVAL);
+
 	/* Initialize a clock structure. */
 	ptp = kzalloc(sizeof(struct ptp_clock), GFP_KERNEL);
 	if (!ptp) {
-- 
2.51.1.851.g4ebd6896fd-goog

Re: [PATCH net] ptp: Return -EINVAL on ptp_clock_register if required ops are NULL

Posted by Vadim Fedorenko 3 months ago

On 30/10/2025 18:08, Tim Hostetler wrote:
> ptp_clock should never be registered unless it stubs one of gettimex64()
> or gettime64() and settime64(). WARN_ON_ONCE and error out if either set
> of function pointers is null.
> 
> Cc: stable@vger.kernel.org
> Fixes: d7d38f5bd7be ("ptp: use the 64 bit get/set time methods for the posix clock.")
> Suggested-by: Kuniyuki Iwashima <kuniyu@google.com>
> Reviewed-by: Kuniyuki Iwashima <kuniyu@google.com>
> Reviewed-by: Harshitha Ramamurthy <hramamurthy@google.com>
> Signed-off-by: Tim Hostetler <thostet@google.com>
> ---
>   drivers/ptp/ptp_clock.c | 4 ++++
>   1 file changed, 4 insertions(+)
> 
> diff --git a/drivers/ptp/ptp_clock.c b/drivers/ptp/ptp_clock.c
> index ef020599b771..0bc79076771b 100644
> --- a/drivers/ptp/ptp_clock.c
> +++ b/drivers/ptp/ptp_clock.c
> @@ -325,6 +325,10 @@ struct ptp_clock *ptp_clock_register(struct ptp_clock_info *info,
>   	if (info->n_alarm > PTP_MAX_ALARMS)
>   		return ERR_PTR(-EINVAL);
>   
> +	if (WARN_ON_ONCE((!info->gettimex64 && !info->gettime64) ||
> +			 !info->settime64))
> +		return ERR_PTR(-EINVAL);
> +
>   	/* Initialize a clock structure. */
>   	ptp = kzalloc(sizeof(struct ptp_clock), GFP_KERNEL);
>   	if (!ptp) {


The patch itself LGTM, but I believe it should be targeted to net-next,
as it doesn't actually fix any problem with GVE patches landed net tree
already.

Reviewed-by: Vadim Fedorenko <vadim.fedorenko@linux.dev>

Re: [PATCH net] ptp: Return -EINVAL on ptp_clock_register if required ops are NULL

Posted by Tim Hostetler 3 months ago

On Tue, Nov 4, 2025 at 3:32 AM Vadim Fedorenko
<vadim.fedorenko@linux.dev> wrote:
>
> The patch itself LGTM, but I believe it should be targeted to net-next,
> as it doesn't actually fix any problem with GVE patches landed net tree
> already.
>
> Reviewed-by: Vadim Fedorenko <vadim.fedorenko@linux.dev>

Thanks Vadim and Jakub, I'll spin a v2 for net-next and fold all the
input validation into one WARN_ON_ONCE per Jakub's suggestion.

Re: [PATCH net] ptp: Return -EINVAL on ptp_clock_register if required ops are NULL

Posted by Jakub Kicinski 3 months, 1 week ago

On Thu, 30 Oct 2025 11:08:32 -0700 Tim Hostetler wrote:
> ptp_clock should never be registered unless it stubs one of gettimex64()
> or gettime64() and settime64(). WARN_ON_ONCE and error out if either set
> of function pointers is null.
> 
> Cc: stable@vger.kernel.org
> Fixes: d7d38f5bd7be ("ptp: use the 64 bit get/set time methods for the posix clock.")

This needs to go to net-next without the tags above.
The check can only help with new drivers, old ones _must_ be fixed 
like gve was. Not registering a driver is a regression.

> Suggested-by: Kuniyuki Iwashima <kuniyu@google.com>
> Reviewed-by: Kuniyuki Iwashima <kuniyu@google.com>
> Reviewed-by: Harshitha Ramamurthy <hramamurthy@google.com>
> Signed-off-by: Tim Hostetler <thostet@google.com>
> ---
>  drivers/ptp/ptp_clock.c | 4 ++++
>  1 file changed, 4 insertions(+)
> 
> diff --git a/drivers/ptp/ptp_clock.c b/drivers/ptp/ptp_clock.c
> index ef020599b771..0bc79076771b 100644
> --- a/drivers/ptp/ptp_clock.c
> +++ b/drivers/ptp/ptp_clock.c
> @@ -325,6 +325,10 @@ struct ptp_clock *ptp_clock_register(struct ptp_clock_info *info,
>  	if (info->n_alarm > PTP_MAX_ALARMS)

->n_alarm check is also input validation, you should probably fold it
into your new WARN_ON_ONCE(). Either that or remove the WARN_ON_ONCE()
annotation below. As is the checks are inconsistent.

>  		return ERR_PTR(-EINVAL);
>  
> +	if (WARN_ON_ONCE((!info->gettimex64 && !info->gettime64) ||
> +			 !info->settime64))
> +		return ERR_PTR(-EINVAL);
> +