[v1] net: dev_ioctl: take ops lock in hwtstamp lower paths

[PATCH net] net: dev_ioctl: take ops lock in hwtstamp lower paths

Posted by Carolina Jubran 4 weeks ago

ndo hwtstamp callbacks are expected to run under the per-device ops
lock. Make the lower get/set paths consistent with the rest of ndo
invocations.

Kernel log:
WARNING: CPU: 13 PID: 51364 at ./include/net/netdev_lock.h:70 __netdev_update_features+0x4bd/0xe60
...
RIP: 0010:__netdev_update_features+0x4bd/0xe60
...
Call Trace:
<TASK>
netdev_update_features+0x1f/0x60
mlx5_hwtstamp_set+0x181/0x290 [mlx5_core]
mlx5e_hwtstamp_set+0x19/0x30 [mlx5_core]
dev_set_hwtstamp_phylib+0x9f/0x220
dev_set_hwtstamp_phylib+0x9f/0x220
dev_set_hwtstamp+0x13d/0x240
dev_ioctl+0x12f/0x4b0
sock_ioctl+0x171/0x370
__x64_sys_ioctl+0x3f7/0x900
? __sys_setsockopt+0x69/0xb0
do_syscall_64+0x6f/0x2e0
entry_SYSCALL_64_after_hwframe+0x4b/0x53
...
</TASK>
....
---[ end trace 0000000000000000 ]---

Fixes: ffb7ed19ac0a ("net: hold netdev instance lock during ioctl operations")
Signed-off-by: Carolina Jubran <cjubran@nvidia.com>
Reviewed-by: Cosmin Ratiu <cratiu@nvidia.com>
Reviewed-by: Dragos Tatulea <dtatulea@nvidia.com>

---
 net/core/dev_ioctl.c | 22 ++++++++++++++++++----
 1 file changed, 18 insertions(+), 4 deletions(-)

diff --git a/net/core/dev_ioctl.c b/net/core/dev_ioctl.c
index 9c0ad7f4b5d8..ad54b12d4b4c 100644
--- a/net/core/dev_ioctl.c
+++ b/net/core/dev_ioctl.c
@@ -464,8 +464,15 @@ int generic_hwtstamp_get_lower(struct net_device *dev,
 	if (!netif_device_present(dev))
 		return -ENODEV;
 
-	if (ops->ndo_hwtstamp_get)
-		return dev_get_hwtstamp_phylib(dev, kernel_cfg);
+	if (ops->ndo_hwtstamp_get) {
+		int err;
+
+		netdev_lock_ops(dev);
+		err = dev_get_hwtstamp_phylib(dev, kernel_cfg);
+		netdev_unlock_ops(dev);
+
+		return err;
+	}
 
 	/* Legacy path: unconverted lower driver */
 	return generic_hwtstamp_ioctl_lower(dev, SIOCGHWTSTAMP, kernel_cfg);
@@ -481,8 +488,15 @@ int generic_hwtstamp_set_lower(struct net_device *dev,
 	if (!netif_device_present(dev))
 		return -ENODEV;
 
-	if (ops->ndo_hwtstamp_set)
-		return dev_set_hwtstamp_phylib(dev, kernel_cfg, extack);
+	if (ops->ndo_hwtstamp_set) {
+		int err;
+
+		netdev_lock_ops(dev);
+		err = dev_set_hwtstamp_phylib(dev, kernel_cfg, extack);
+		netdev_unlock_ops(dev);
+
+		return err;
+	}
 
 	/* Legacy path: unconverted lower driver */
 	return generic_hwtstamp_ioctl_lower(dev, SIOCSHWTSTAMP, kernel_cfg);
-- 
2.38.1

Re: [PATCH net] net: dev_ioctl: take ops lock in hwtstamp lower paths

Posted by Kory Maincent 4 weeks ago

On Thu, 4 Sep 2025 21:28:06 +0300
Carolina Jubran <cjubran@nvidia.com> wrote:

> ndo hwtstamp callbacks are expected to run under the per-device ops
> lock. Make the lower get/set paths consistent with the rest of ndo
> invocations.
> 
> Kernel log:
> WARNING: CPU: 13 PID: 51364 at ./include/net/netdev_lock.h:70
> __netdev_update_features+0x4bd/0xe60 ...
> RIP: 0010:__netdev_update_features+0x4bd/0xe60
> ...
> Call Trace:
> <TASK>
> netdev_update_features+0x1f/0x60
> mlx5_hwtstamp_set+0x181/0x290 [mlx5_core]
> mlx5e_hwtstamp_set+0x19/0x30 [mlx5_core]

Where does these two functions come from? They are not mainline.
Else LGTM.

> dev_set_hwtstamp_phylib+0x9f/0x220
> dev_set_hwtstamp_phylib+0x9f/0x220
> dev_set_hwtstamp+0x13d/0x240
> dev_ioctl+0x12f/0x4b0
> sock_ioctl+0x171/0x370
> __x64_sys_ioctl+0x3f7/0x900
> ? __sys_setsockopt+0x69/0xb0
> do_syscall_64+0x6f/0x2e0
> entry_SYSCALL_64_after_hwframe+0x4b/0x53
> ...
> </TASK>
> ....
> ---[ end trace 0000000000000000 ]---
> 
> Fixes: ffb7ed19ac0a ("net: hold netdev instance lock during ioctl operations")
> Signed-off-by: Carolina Jubran <cjubran@nvidia.com>
> Reviewed-by: Cosmin Ratiu <cratiu@nvidia.com>
> Reviewed-by: Dragos Tatulea <dtatulea@nvidia.com>
> 
> ---
>  net/core/dev_ioctl.c | 22 ++++++++++++++++++----
>  1 file changed, 18 insertions(+), 4 deletions(-)
> 
> diff --git a/net/core/dev_ioctl.c b/net/core/dev_ioctl.c
> index 9c0ad7f4b5d8..ad54b12d4b4c 100644
> --- a/net/core/dev_ioctl.c
> +++ b/net/core/dev_ioctl.c
> @@ -464,8 +464,15 @@ int generic_hwtstamp_get_lower(struct net_device *dev,
>  	if (!netif_device_present(dev))
>  		return -ENODEV;
>  
> -	if (ops->ndo_hwtstamp_get)
> -		return dev_get_hwtstamp_phylib(dev, kernel_cfg);
> +	if (ops->ndo_hwtstamp_get) {
> +		int err;
> +
> +		netdev_lock_ops(dev);
> +		err = dev_get_hwtstamp_phylib(dev, kernel_cfg);
> +		netdev_unlock_ops(dev);
> +
> +		return err;
> +	}
>  
>  	/* Legacy path: unconverted lower driver */
>  	return generic_hwtstamp_ioctl_lower(dev, SIOCGHWTSTAMP, kernel_cfg);
> @@ -481,8 +488,15 @@ int generic_hwtstamp_set_lower(struct net_device *dev,
>  	if (!netif_device_present(dev))
>  		return -ENODEV;
>  
> -	if (ops->ndo_hwtstamp_set)
> -		return dev_set_hwtstamp_phylib(dev, kernel_cfg, extack);
> +	if (ops->ndo_hwtstamp_set) {
> +		int err;
> +
> +		netdev_lock_ops(dev);
> +		err = dev_set_hwtstamp_phylib(dev, kernel_cfg, extack);
> +		netdev_unlock_ops(dev);
> +
> +		return err;
> +	}
>  
>  	/* Legacy path: unconverted lower driver */
>  	return generic_hwtstamp_ioctl_lower(dev, SIOCSHWTSTAMP, kernel_cfg);



-- 
Köry Maincent, Bootlin
Embedded Linux and kernel engineering
https://bootlin.com

Re: [PATCH net] net: dev_ioctl: take ops lock in hwtstamp lower paths

Posted by Carolina Jubran 3 weeks, 6 days ago

On 05/09/2025 0:52, Kory Maincent wrote:
> On Thu, 4 Sep 2025 21:28:06 +0300
> Carolina Jubran <cjubran@nvidia.com> wrote:
>
>> ndo hwtstamp callbacks are expected to run under the per-device ops
>> lock. Make the lower get/set paths consistent with the rest of ndo
>> invocations.
>>
>> Kernel log:
>> WARNING: CPU: 13 PID: 51364 at ./include/net/netdev_lock.h:70
>> __netdev_update_features+0x4bd/0xe60 ...
>> RIP: 0010:__netdev_update_features+0x4bd/0xe60
>> ...
>> Call Trace:
>> <TASK>
>> netdev_update_features+0x1f/0x60
>> mlx5_hwtstamp_set+0x181/0x290 [mlx5_core]
>> mlx5e_hwtstamp_set+0x19/0x30 [mlx5_core]
> Where does these two functions come from? They are not mainline.
> Else LGTM.

You are right, I hit this when I was working on another patch to
convert the legacy ndo. I thought it would be nice to have the
kernel log in the commit message.

Thanks,
Carolina

Re: [PATCH net] net: dev_ioctl: take ops lock in hwtstamp lower paths

Posted by Simon Horman 3 weeks, 6 days ago

On Fri, Sep 05, 2025 at 08:35:37AM +0300, Carolina Jubran wrote:
> 
> On 05/09/2025 0:52, Kory Maincent wrote:
> > On Thu, 4 Sep 2025 21:28:06 +0300
> > Carolina Jubran <cjubran@nvidia.com> wrote:
> > 
> > > ndo hwtstamp callbacks are expected to run under the per-device ops
> > > lock. Make the lower get/set paths consistent with the rest of ndo
> > > invocations.
> > > 
> > > Kernel log:
> > > WARNING: CPU: 13 PID: 51364 at ./include/net/netdev_lock.h:70
> > > __netdev_update_features+0x4bd/0xe60 ...
> > > RIP: 0010:__netdev_update_features+0x4bd/0xe60
> > > ...
> > > Call Trace:
> > > <TASK>
> > > netdev_update_features+0x1f/0x60
> > > mlx5_hwtstamp_set+0x181/0x290 [mlx5_core]
> > > mlx5e_hwtstamp_set+0x19/0x30 [mlx5_core]
> > Where does these two functions come from? They are not mainline.
> > Else LGTM.
> 
> You are right, I hit this when I was working on another patch to
> convert the legacy ndo. I thought it would be nice to have the
> kernel log in the commit message.

Thanks Carolina,

I think it would be nice to note that in the commit message.
Kory's confusion seems reasonable as things stand.
And others may also be confused by it.

Re: [PATCH net] net: dev_ioctl: take ops lock in hwtstamp lower paths

Posted by Jakub Kicinski 3 weeks, 6 days ago

On Fri, 5 Sep 2025 14:07:16 +0100 Simon Horman wrote:
> I think it would be nice to note that in the commit message.

+1