diff mbox series

[net-next] ethtool: ioctl: improve error checking for set_wol

Message ID 1685566429-2869-1-git-send-email-justin.chen@broadcom.com
State New
Headers
Series [net-next] ethtool: ioctl: improve error checking for set_wol |

Commit Message

Justin Chen May 31, 2023, 8:53 p.m. UTC 
  The netlink version of set_wol checks for not supported wolopts and avoids
setting wol when the correct wolopt is already set. If we do the same with
the ioctl version then we can remove these checks from the driver layer.

Signed-off-by: Justin Chen <justin.chen@broadcom.com>
---
 net/ethtool/ioctl.c | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)

Comments

Simon Horman June 1, 2023, 3:55 p.m. UTC | #1 
+ Daniil Tatianin <d-tatianin@yandex-team.ru>, Andrew Lunn <andrew@lunn.ch>
  as per ./scripts/get_maintainer.pl --git-min-percent 25 net/ethtool/ioctl.c

On Wed, May 31, 2023 at 01:53:49PM -0700, Justin Chen wrote:
> The netlink version of set_wol checks for not supported wolopts and avoids
> setting wol when the correct wolopt is already set. If we do the same with
> the ioctl version then we can remove these checks from the driver layer.

Hi Justin,

Are you planning follow-up patches for the driver layer?

> Signed-off-by: Justin Chen <justin.chen@broadcom.com>
> ---
>  net/ethtool/ioctl.c | 14 ++++++++++++--
>  1 file changed, 12 insertions(+), 2 deletions(-)
> 
> diff --git a/net/ethtool/ioctl.c b/net/ethtool/ioctl.c
> index 6bb778e10461..80f456f83db0 100644
> --- a/net/ethtool/ioctl.c
> +++ b/net/ethtool/ioctl.c
> @@ -1436,15 +1436,25 @@ static int ethtool_get_wol(struct net_device *dev, char __user *useraddr)
>  
>  static int ethtool_set_wol(struct net_device *dev, char __user *useraddr)
>  {
> -	struct ethtool_wolinfo wol;
> +	struct ethtool_wolinfo wol, cur_wol;
>  	int ret;
>  
> -	if (!dev->ethtool_ops->set_wol)
> +	if (!dev->ethtool_ops->get_wol || !dev->ethtool_ops->set_wol)
>  		return -EOPNOTSUPP;

Are there cases where (in-tree) drivers provide set_wol byt not get_wol?
If so, does this break their set_wol support?

>  
> +	memset(&cur_wol, 0, sizeof(struct ethtool_wolinfo));
> +	cur_wol.cmd = ETHTOOL_GWOL;
> +	dev->ethtool_ops->get_wol(dev, &cur_wol);
> +
>  	if (copy_from_user(&wol, useraddr, sizeof(wol)))
>  		return -EFAULT;
>  
> +	if (wol.wolopts & ~cur_wol.supported)
> +		return -EOPNOTSUPP;
> +
> +	if (wol.wolopts == cur_wol.wolopts)
> +		return 0;
> +
>  	ret = dev->ethtool_ops->set_wol(dev, &wol);
>  	if (ret)
>  		return ret;
Jakub Kicinski June 1, 2023, 4:03 p.m. UTC | #2 
On Thu, 1 Jun 2023 17:55:21 +0200 Simon Horman wrote:
> + Daniil Tatianin <d-tatianin@yandex-team.ru>, Andrew Lunn <andrew@lunn.ch>
>   as per ./scripts/get_maintainer.pl --git-min-percent 25 net/ethtool/ioctl.c

Sorry to chime in but always prefer running get_maintainer on the patch
rather than a file path. File path misses stuff like Fixes tags.
If it was up to me that option would have been removed :(
Justin Chen June 1, 2023, 4:22 p.m. UTC | #3 
On 6/1/2023 8:55 AM, Simon Horman wrote:
> + Daniil Tatianin <d-tatianin@yandex-team.ru>, Andrew Lunn <andrew@lunn.ch>
>    as per ./scripts/get_maintainer.pl --git-min-percent 25 net/ethtool/ioctl.c
> 
> On Wed, May 31, 2023 at 01:53:49PM -0700, Justin Chen wrote:
>> The netlink version of set_wol checks for not supported wolopts and avoids
>> setting wol when the correct wolopt is already set. If we do the same with
>> the ioctl version then we can remove these checks from the driver layer.
> 
> Hi Justin,
> 
> Are you planning follow-up patches for the driver layer?
> 

I was planning to for the Broadcom drivers since those I can test. But I 
could do it across the board if that is preferred.

>> Signed-off-by: Justin Chen <justin.chen@broadcom.com>
>> ---
>>   net/ethtool/ioctl.c | 14 ++++++++++++--
>>   1 file changed, 12 insertions(+), 2 deletions(-)
>>
>> diff --git a/net/ethtool/ioctl.c b/net/ethtool/ioctl.c
>> index 6bb778e10461..80f456f83db0 100644
>> --- a/net/ethtool/ioctl.c
>> +++ b/net/ethtool/ioctl.c
>> @@ -1436,15 +1436,25 @@ static int ethtool_get_wol(struct net_device *dev, char __user *useraddr)
>>   
>>   static int ethtool_set_wol(struct net_device *dev, char __user *useraddr)
>>   {
>> -	struct ethtool_wolinfo wol;
>> +	struct ethtool_wolinfo wol, cur_wol;
>>   	int ret;
>>   
>> -	if (!dev->ethtool_ops->set_wol)
>> +	if (!dev->ethtool_ops->get_wol || !dev->ethtool_ops->set_wol)
>>   		return -EOPNOTSUPP;
> 
> Are there cases where (in-tree) drivers provide set_wol byt not get_wol?
> If so, does this break their set_wol support?
> 

My original thought was to match netlink set wol behavior. So drivers 
that do that won't work with netlink set_wol right now. I'll skim around 
to see if any drivers do this. But I would reckon this should be a 
driver fix.

Thanks,
Justin

>>   
>> +	memset(&cur_wol, 0, sizeof(struct ethtool_wolinfo));
>> +	cur_wol.cmd = ETHTOOL_GWOL;
>> +	dev->ethtool_ops->get_wol(dev, &cur_wol);
>> +
>>   	if (copy_from_user(&wol, useraddr, sizeof(wol)))
>>   		return -EFAULT;
>>   
>> +	if (wol.wolopts & ~cur_wol.supported)
>> +		return -EOPNOTSUPP;
>> +
>> +	if (wol.wolopts == cur_wol.wolopts)
>> +		return 0;
>> +
>>   	ret = dev->ethtool_ops->set_wol(dev, &wol);
>>   	if (ret)
>>   		return ret;
Simon Horman June 1, 2023, 4:23 p.m. UTC | #4 
On Thu, Jun 01, 2023 at 09:03:10AM -0700, Jakub Kicinski wrote:
> On Thu, 1 Jun 2023 17:55:21 +0200 Simon Horman wrote:
> > + Daniil Tatianin <d-tatianin@yandex-team.ru>, Andrew Lunn <andrew@lunn.ch>
> >   as per ./scripts/get_maintainer.pl --git-min-percent 25 net/ethtool/ioctl.c
> Sorry to chime in but always prefer running get_maintainer on the patch
> rather than a file path. File path misses stuff like Fixes tags.
> If it was up to me that option would have been removed :(

Yes, sorry about that.
Justin Chen June 1, 2023, 4:23 p.m. UTC | #5 
On 6/1/2023 9:03 AM, Jakub Kicinski wrote:
> On Thu, 1 Jun 2023 17:55:21 +0200 Simon Horman wrote:
>> + Daniil Tatianin <d-tatianin@yandex-team.ru>, Andrew Lunn <andrew@lunn.ch>
>>    as per ./scripts/get_maintainer.pl --git-min-percent 25 net/ethtool/ioctl.c
> 
> Sorry to chime in but always prefer running get_maintainer on the patch
> rather than a file path. File path misses stuff like Fixes tags.
> If it was up to me that option would have been removed :(
Apologies, Will do next time.

Justin
Justin Chen June 1, 2023, 6:27 p.m. UTC | #6 
On 6/1/23 9:22 AM, Justin Chen wrote:
> 
> 
> On 6/1/2023 8:55 AM, Simon Horman wrote:
>> + Daniil Tatianin <d-tatianin@yandex-team.ru>, Andrew Lunn 
>> <andrew@lunn.ch>
>>    as per ./scripts/get_maintainer.pl --git-min-percent 25 
>> net/ethtool/ioctl.c
>>
>> On Wed, May 31, 2023 at 01:53:49PM -0700, Justin Chen wrote:
>>> The netlink version of set_wol checks for not supported wolopts and 
>>> avoids
>>> setting wol when the correct wolopt is already set. If we do the same 
>>> with
>>> the ioctl version then we can remove these checks from the driver layer.
>>
>> Hi Justin,
>>
>> Are you planning follow-up patches for the driver layer?
>>
> 
> I was planning to for the Broadcom drivers since those I can test. But I 
> could do it across the board if that is preferred.
> 
>>> Signed-off-by: Justin Chen <justin.chen@broadcom.com>
>>> ---
>>>   net/ethtool/ioctl.c | 14 ++++++++++++--
>>>   1 file changed, 12 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/net/ethtool/ioctl.c b/net/ethtool/ioctl.c
>>> index 6bb778e10461..80f456f83db0 100644
>>> --- a/net/ethtool/ioctl.c
>>> +++ b/net/ethtool/ioctl.c
>>> @@ -1436,15 +1436,25 @@ static int ethtool_get_wol(struct net_device 
>>> *dev, char __user *useraddr)
>>>   static int ethtool_set_wol(struct net_device *dev, char __user 
>>> *useraddr)
>>>   {
>>> -    struct ethtool_wolinfo wol;
>>> +    struct ethtool_wolinfo wol, cur_wol;
>>>       int ret;
>>> -    if (!dev->ethtool_ops->set_wol)
>>> +    if (!dev->ethtool_ops->get_wol || !dev->ethtool_ops->set_wol)
>>>           return -EOPNOTSUPP;
>>
>> Are there cases where (in-tree) drivers provide set_wol byt not get_wol?
>> If so, does this break their set_wol support?
>>
> 
> My original thought was to match netlink set wol behavior. So drivers 
> that do that won't work with netlink set_wol right now. I'll skim around 
> to see if any drivers do this. But I would reckon this should be a 
> driver fix.
> 
> Thanks,
> Justin
> 

I see a driver at drivers/net/phy/microchip.c. But this is a phy driver 
set_wol hook.

Justin

>>> +    memset(&cur_wol, 0, sizeof(struct ethtool_wolinfo));
>>> +    cur_wol.cmd = ETHTOOL_GWOL;
>>> +    dev->ethtool_ops->get_wol(dev, &cur_wol);
>>> +
>>>       if (copy_from_user(&wol, useraddr, sizeof(wol)))
>>>           return -EFAULT;
>>> +    if (wol.wolopts & ~cur_wol.supported)
>>> +        return -EOPNOTSUPP;
>>> +
>>> +    if (wol.wolopts == cur_wol.wolopts)
>>> +        return 0;
>>> +
>>>       ret = dev->ethtool_ops->set_wol(dev, &wol);
>>>       if (ret)
>>>           return ret;
Florian Fainelli June 1, 2023, 6:37 p.m. UTC | #7 
On 6/1/23 11:27, Justin Chen wrote:
> 
> 
> On 6/1/23 9:22 AM, Justin Chen wrote:
>>
>>
>> On 6/1/2023 8:55 AM, Simon Horman wrote:
>>> + Daniil Tatianin <d-tatianin@yandex-team.ru>, Andrew Lunn 
>>> <andrew@lunn.ch>
>>>    as per ./scripts/get_maintainer.pl --git-min-percent 25 
>>> net/ethtool/ioctl.c
>>>
>>> On Wed, May 31, 2023 at 01:53:49PM -0700, Justin Chen wrote:
>>>> The netlink version of set_wol checks for not supported wolopts and 
>>>> avoids
>>>> setting wol when the correct wolopt is already set. If we do the 
>>>> same with
>>>> the ioctl version then we can remove these checks from the driver 
>>>> layer.
>>>
>>> Hi Justin,
>>>
>>> Are you planning follow-up patches for the driver layer?
>>>
>>
>> I was planning to for the Broadcom drivers since those I can test. But 
>> I could do it across the board if that is preferred.
>>
>>>> Signed-off-by: Justin Chen <justin.chen@broadcom.com>
>>>> ---
>>>>   net/ethtool/ioctl.c | 14 ++++++++++++--
>>>>   1 file changed, 12 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/net/ethtool/ioctl.c b/net/ethtool/ioctl.c
>>>> index 6bb778e10461..80f456f83db0 100644
>>>> --- a/net/ethtool/ioctl.c
>>>> +++ b/net/ethtool/ioctl.c
>>>> @@ -1436,15 +1436,25 @@ static int ethtool_get_wol(struct net_device 
>>>> *dev, char __user *useraddr)
>>>>   static int ethtool_set_wol(struct net_device *dev, char __user 
>>>> *useraddr)
>>>>   {
>>>> -    struct ethtool_wolinfo wol;
>>>> +    struct ethtool_wolinfo wol, cur_wol;
>>>>       int ret;
>>>> -    if (!dev->ethtool_ops->set_wol)
>>>> +    if (!dev->ethtool_ops->get_wol || !dev->ethtool_ops->set_wol)
>>>>           return -EOPNOTSUPP;
>>>
>>> Are there cases where (in-tree) drivers provide set_wol byt not get_wol?
>>> If so, does this break their set_wol support?
>>>
>>
>> My original thought was to match netlink set wol behavior. So drivers 
>> that do that won't work with netlink set_wol right now. I'll skim 
>> around to see if any drivers do this. But I would reckon this should 
>> be a driver fix.
>>
>> Thanks,
>> Justin
>>
> 
> I see a driver at drivers/net/phy/microchip.c. But this is a phy driver 
> set_wol hook.

That part of the driver appears to be dead code. It attempts to pretend 
to support Wake-on-LAN, but it does not do any specific programming of 
wake-up filters, nor does it implement get_wol. It also does not make 
use of the recently introduced PHY_ALWAYS_CALL_SUSPEND flag.

When it is time to determine whether to suspend the PHY or not, 
eventually phy_suspend() will call phy_ethtool_get_wol(). Since no 
get_wol is implemented, the wol.wolopts will remain zero, therefore we 
will just suspend the PHY.

I suspect this was added to work around MAC drivers that may forcefully 
try to suspend the PHY, but that should not even be possible these days.

I would just remove that logic from microchip.c entirely.
Andrew Lunn June 1, 2023, 6:48 p.m. UTC | #8 
> > > I was planning to for the Broadcom drivers since those I can test.
> > > But I could do it across the board if that is preferred.
> > > 
> > > > > Signed-off-by: Justin Chen <justin.chen@broadcom.com>
> > > > > ---
> > > > >   net/ethtool/ioctl.c | 14 ++++++++++++--
> > > > >   1 file changed, 12 insertions(+), 2 deletions(-)
> > > > > 
> > > > > diff --git a/net/ethtool/ioctl.c b/net/ethtool/ioctl.c
> > > > > index 6bb778e10461..80f456f83db0 100644
> > > > > --- a/net/ethtool/ioctl.c
> > > > > +++ b/net/ethtool/ioctl.c
> > > > > @@ -1436,15 +1436,25 @@ static int ethtool_get_wol(struct
> > > > > net_device *dev, char __user *useraddr)
> > > > >   static int ethtool_set_wol(struct net_device *dev, char
> > > > > __user *useraddr)
> > > > >   {
> > > > > -    struct ethtool_wolinfo wol;
> > > > > +    struct ethtool_wolinfo wol, cur_wol;
> > > > >       int ret;
> > > > > -    if (!dev->ethtool_ops->set_wol)
> > > > > +    if (!dev->ethtool_ops->get_wol || !dev->ethtool_ops->set_wol)
> > > > >           return -EOPNOTSUPP;
> > > > 
> > > > Are there cases where (in-tree) drivers provide set_wol byt not get_wol?
> > > > If so, does this break their set_wol support?
> > > > 
> > > 
> > > My original thought was to match netlink set wol behavior. So
> > > drivers that do that won't work with netlink set_wol right now. I'll
> > > skim around to see if any drivers do this. But I would reckon this
> > > should be a driver fix.
> > > 
> > > Thanks,
> > > Justin
> > > 
> > 
> > I see a driver at drivers/net/phy/microchip.c. But this is a phy driver
> > set_wol hook.
> 
> That part of the driver appears to be dead code. It attempts to pretend to
> support Wake-on-LAN, but it does not do any specific programming of wake-up
> filters, nor does it implement get_wol. It also does not make use of the
> recently introduced PHY_ALWAYS_CALL_SUSPEND flag.
> 
> When it is time to determine whether to suspend the PHY or not, eventually
> phy_suspend() will call phy_ethtool_get_wol(). Since no get_wol is
> implemented, the wol.wolopts will remain zero, therefore we will just
> suspend the PHY.
> 
> I suspect this was added to work around MAC drivers that may forcefully try
> to suspend the PHY, but that should not even be possible these days.
> 
> I would just remove that logic from microchip.c entirely.

The Microchip developers are reasonably responsive. So we should Cc:
them.

	Andrew
Justin Chen June 1, 2023, 6:58 p.m. UTC | #9 
+ Woojung, UNGLinuxDriver

On 6/1/23 11:48 AM, Andrew Lunn wrote:
>>>> I was planning to for the Broadcom drivers since those I can test.
>>>> But I could do it across the board if that is preferred.
>>>>
>>>>>> Signed-off-by: Justin Chen <justin.chen@broadcom.com>
>>>>>> ---
>>>>>>    net/ethtool/ioctl.c | 14 ++++++++++++--
>>>>>>    1 file changed, 12 insertions(+), 2 deletions(-)
>>>>>>
>>>>>> diff --git a/net/ethtool/ioctl.c b/net/ethtool/ioctl.c
>>>>>> index 6bb778e10461..80f456f83db0 100644
>>>>>> --- a/net/ethtool/ioctl.c
>>>>>> +++ b/net/ethtool/ioctl.c
>>>>>> @@ -1436,15 +1436,25 @@ static int ethtool_get_wol(struct
>>>>>> net_device *dev, char __user *useraddr)
>>>>>>    static int ethtool_set_wol(struct net_device *dev, char
>>>>>> __user *useraddr)
>>>>>>    {
>>>>>> -    struct ethtool_wolinfo wol;
>>>>>> +    struct ethtool_wolinfo wol, cur_wol;
>>>>>>        int ret;
>>>>>> -    if (!dev->ethtool_ops->set_wol)
>>>>>> +    if (!dev->ethtool_ops->get_wol || !dev->ethtool_ops->set_wol)
>>>>>>            return -EOPNOTSUPP;
>>>>>
>>>>> Are there cases where (in-tree) drivers provide set_wol byt not get_wol?
>>>>> If so, does this break their set_wol support?
>>>>>
>>>>
>>>> My original thought was to match netlink set wol behavior. So
>>>> drivers that do that won't work with netlink set_wol right now. I'll
>>>> skim around to see if any drivers do this. But I would reckon this
>>>> should be a driver fix.
>>>>
>>>> Thanks,
>>>> Justin
>>>>
>>>
>>> I see a driver at drivers/net/phy/microchip.c. But this is a phy driver
>>> set_wol hook.
>>
>> That part of the driver appears to be dead code. It attempts to pretend to
>> support Wake-on-LAN, but it does not do any specific programming of wake-up
>> filters, nor does it implement get_wol. It also does not make use of the
>> recently introduced PHY_ALWAYS_CALL_SUSPEND flag.
>>
>> When it is time to determine whether to suspend the PHY or not, eventually
>> phy_suspend() will call phy_ethtool_get_wol(). Since no get_wol is
>> implemented, the wol.wolopts will remain zero, therefore we will just
>> suspend the PHY.
>>
>> I suspect this was added to work around MAC drivers that may forcefully try
>> to suspend the PHY, but that should not even be possible these days.
>>
>> I would just remove that logic from microchip.c entirely.
> 
> The Microchip developers are reasonably responsive. So we should Cc:
> them.
> 
> 	Andrew
Woojung.Huh@microchip.com June 1, 2023, 8:41 p.m. UTC | #10 
> On 6/1/23 11:48 AM, Andrew Lunn wrote:
> >>>> I was planning to for the Broadcom drivers since those I can test.
> >>>> But I could do it across the board if that is preferred.
> >>>>
> >>>>>> Signed-off-by: Justin Chen <justin.chen@broadcom.com>
> >>>>>> ---
> >>>>>>    net/ethtool/ioctl.c | 14 ++++++++++++--
> >>>>>>    1 file changed, 12 insertions(+), 2 deletions(-)
> >>>>>>
> >>>>>> diff --git a/net/ethtool/ioctl.c b/net/ethtool/ioctl.c
> >>>>>> index 6bb778e10461..80f456f83db0 100644
> >>>>>> --- a/net/ethtool/ioctl.c
> >>>>>> +++ b/net/ethtool/ioctl.c
> >>>>>> @@ -1436,15 +1436,25 @@ static int ethtool_get_wol(struct
> >>>>>> net_device *dev, char __user *useraddr)
> >>>>>>    static int ethtool_set_wol(struct net_device *dev, char
> >>>>>> __user *useraddr)
> >>>>>>    {
> >>>>>> -    struct ethtool_wolinfo wol;
> >>>>>> +    struct ethtool_wolinfo wol, cur_wol;
> >>>>>>        int ret;
> >>>>>> -    if (!dev->ethtool_ops->set_wol)
> >>>>>> +    if (!dev->ethtool_ops->get_wol || !dev->ethtool_ops->set_wol)
> >>>>>>            return -EOPNOTSUPP;
> >>>>>
> >>>>> Are there cases where (in-tree) drivers provide set_wol byt not
> get_wol?
> >>>>> If so, does this break their set_wol support?
> >>>>>
> >>>>
> >>>> My original thought was to match netlink set wol behavior. So
> >>>> drivers that do that won't work with netlink set_wol right now. I'll
> >>>> skim around to see if any drivers do this. But I would reckon this
> >>>> should be a driver fix.
> >>>>
> >>>> Thanks,
> >>>> Justin
> >>>>
> >>>
> >>> I see a driver at drivers/net/phy/microchip.c. But this is a phy driver
> >>> set_wol hook.
> >>
> >> That part of the driver appears to be dead code. It attempts to pretend to
> >> support Wake-on-LAN, but it does not do any specific programming of
> wake-up
> >> filters, nor does it implement get_wol. It also does not make use of the
> >> recently introduced PHY_ALWAYS_CALL_SUSPEND flag.
> >>
> >> When it is time to determine whether to suspend the PHY or not,
> eventually
> >> phy_suspend() will call phy_ethtool_get_wol(). Since no get_wol is
> >> implemented, the wol.wolopts will remain zero, therefore we will just
> >> suspend the PHY.
> >>
> >> I suspect this was added to work around MAC drivers that may forcefully
> try
> >> to suspend the PHY, but that should not even be possible these days.
> >>
> >> I would just remove that logic from microchip.c entirely.
> >
> > The Microchip developers are reasonably responsive. So we should Cc:
> > them.


set_wol in drivers/net/phy/microchip.c is used to set the flag
to avoid PHY power down at suspend time.
Looks it is old-fashioned now because frame work is not calling suspend
after calling get_wol. We will make a patch for it.

BTW, this patch is checking MAC driver set_wol and get_wol.
So I don't think it breaks drivers/net/phy/microchip.c suspend operation anyway.

Thanks.
Woojung
Simon Horman June 2, 2023, 8:54 a.m. UTC | #11 
On Thu, Jun 01, 2023 at 09:22:50AM -0700, Justin Chen wrote:
> 
> 
> On 6/1/2023 8:55 AM, Simon Horman wrote:
> > + Daniil Tatianin <d-tatianin@yandex-team.ru>, Andrew Lunn <andrew@lunn.ch>
> >    as per ./scripts/get_maintainer.pl --git-min-percent 25 net/ethtool/ioctl.c
> > 
> > On Wed, May 31, 2023 at 01:53:49PM -0700, Justin Chen wrote:
> > > The netlink version of set_wol checks for not supported wolopts and avoids
> > > setting wol when the correct wolopt is already set. If we do the same with
> > > the ioctl version then we can remove these checks from the driver layer.
> > 
> > Hi Justin,
> > 
> > Are you planning follow-up patches for the driver layer?
> > 
> 
> I was planning to for the Broadcom drivers since those I can test. But I
> could do it across the board if that is preferred.

I think that would be my suggestion.
But I'm unsure of the magnitude of change involved.
Or how risky it is in terms of introducing regressions.
In any case, perhaps it's best to start small.

> > > Signed-off-by: Justin Chen <justin.chen@broadcom.com>
> > > ---
> > >   net/ethtool/ioctl.c | 14 ++++++++++++--
> > >   1 file changed, 12 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/net/ethtool/ioctl.c b/net/ethtool/ioctl.c
> > > index 6bb778e10461..80f456f83db0 100644
> > > --- a/net/ethtool/ioctl.c
> > > +++ b/net/ethtool/ioctl.c
> > > @@ -1436,15 +1436,25 @@ static int ethtool_get_wol(struct net_device *dev, char __user *useraddr)
> > >   static int ethtool_set_wol(struct net_device *dev, char __user *useraddr)
> > >   {
> > > -	struct ethtool_wolinfo wol;
> > > +	struct ethtool_wolinfo wol, cur_wol;
> > >   	int ret;
> > > -	if (!dev->ethtool_ops->set_wol)
> > > +	if (!dev->ethtool_ops->get_wol || !dev->ethtool_ops->set_wol)
> > >   		return -EOPNOTSUPP;
> > 
> > Are there cases where (in-tree) drivers provide set_wol byt not get_wol?
> > If so, does this break their set_wol support?
> > 
> 
> My original thought was to match netlink set wol behavior. So drivers that
> do that won't work with netlink set_wol right now. I'll skim around to see
> if any drivers do this. But I would reckon this should be a driver fix.

It seems, from other discussion in a different sub-thread, that we are
likely clear there :)

As that was my only real reservation wrt this patch:

Reviewed-by: Simon Horman <simon.horman@corigine.com>
Jakub Kicinski June 5, 2023, 6:21 p.m. UTC | #12 
On Wed, 31 May 2023 13:53:49 -0700 Justin Chen wrote:
> The netlink version of set_wol checks for not supported wolopts and avoids
> setting wol when the correct wolopt is already set. If we do the same with
> the ioctl version then we can remove these checks from the driver layer.
> 
> Signed-off-by: Justin Chen <justin.chen@broadcom.com>

If I understand the discussion correctly the patch is ready to be
applied. Could you repost it? It has been marked as "changes requested"
in patchwork, I'm not 100% sure about the reason.
diff mbox series

Patch

diff --git a/net/ethtool/ioctl.c b/net/ethtool/ioctl.c
index 6bb778e10461..80f456f83db0 100644
--- a/net/ethtool/ioctl.c
+++ b/net/ethtool/ioctl.c
@@ -1436,15 +1436,25 @@  static int ethtool_get_wol(struct net_device *dev, char __user *useraddr)
 
 static int ethtool_set_wol(struct net_device *dev, char __user *useraddr)
 {
-	struct ethtool_wolinfo wol;
+	struct ethtool_wolinfo wol, cur_wol;
 	int ret;
 
-	if (!dev->ethtool_ops->set_wol)
+	if (!dev->ethtool_ops->get_wol || !dev->ethtool_ops->set_wol)
 		return -EOPNOTSUPP;
 
+	memset(&cur_wol, 0, sizeof(struct ethtool_wolinfo));
+	cur_wol.cmd = ETHTOOL_GWOL;
+	dev->ethtool_ops->get_wol(dev, &cur_wol);
+
 	if (copy_from_user(&wol, useraddr, sizeof(wol)))
 		return -EFAULT;
 
+	if (wol.wolopts & ~cur_wol.supported)
+		return -EOPNOTSUPP;
+
+	if (wol.wolopts == cur_wol.wolopts)
+		return 0;
+
 	ret = dev->ethtool_ops->set_wol(dev, &wol);
 	if (ret)
 		return ret;