netdev - Re: [PATCH, net-next, v3, 1/2] Three virtual devices (ibmveth, virtio_net, and netvsc) all have similar code to set/get link settings and validate ethtool command. To eliminate duplication of code, it is factored out into core/ethtool.c.

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Thu, 19 Dec 2019 19:26:14 -0600
From:   Thomas Falcon <tlfalcon@...ux.ibm.com>
To:     Michal Kubecek <mkubecek@...e.cz>, netdev@...r.kernel.org
Cc:     Cris Forno <cforno12@...ux.vnet.ibm.com>, mst@...hat.com,
        jasowang@...hat.com, haiyangz@...rosoft.com,
        sthemmin@...rosoft.com, sashal@...nel.org,
        David Miller <davem@...emloft.net>,
        Jakub Kicinski <jakub.kicinski@...ronome.com>
Subject: Re: [PATCH, net-next, v3, 1/2] Three virtual devices (ibmveth,
 virtio_net, and netvsc) all have similar code to set/get link settings and
 validate ethtool command. To eliminate duplication of code, it is factored
 out into core/ethtool.c.

On 12/19/19 4:36 PM, Michal Kubecek wrote:
> On Thu, Dec 19, 2019 at 01:40:56PM -0600, Cris Forno wrote:
>> Signed-off-by: Cris Forno <cforno12@...ux.vnet.ibm.com>

Hi Michal, thanks for your comments.  I have a question on your 
suggestions for the ethtool_virtdev_validate_cmd below.

>> ---
>>   include/linux/ethtool.h |  2 ++
>>   net/core/ethtool.c      | 58 +++++++++++++++++++++++++++++++++++++++++++++++++
>>   2 files changed, 60 insertions(+)
>>
>> diff --git a/include/linux/ethtool.h b/include/linux/ethtool.h
>> index 95991e43..1b0417b 100644
>> --- a/include/linux/ethtool.h
>> +++ b/include/linux/ethtool.h
>> @@ -394,6 +394,8 @@ struct ethtool_ops {
>>   					  struct ethtool_coalesce *);
>>   	int	(*set_per_queue_coalesce)(struct net_device *, u32,
>>   					  struct ethtool_coalesce *);
>> +	bool    (*virtdev_validate_link_ksettings)(const struct
>> +						   ethtool_link_ksettings *);
>>   	int	(*get_link_ksettings)(struct net_device *,
>>   				      struct ethtool_link_ksettings *);
>>   	int	(*set_link_ksettings)(struct net_device *,
>> diff --git a/net/core/ethtool.c b/net/core/ethtool.c
>> index cd9bc67..4091a94 100644
>> --- a/net/core/ethtool.c
>> +++ b/net/core/ethtool.c
> You should probably rebase on top of current net-next; this file has
> been moved to net/ethtool/ioctl.c recently.
>
>> @@ -579,6 +579,32 @@ static int load_link_ksettings_from_user(struct ethtool_link_ksettings *to,
>>   	return 0;
>>   }
>>   
>> +/* Check if the user is trying to change anything besides speed/duplex */
>> +static bool
>> +ethtool_virtdev_validate_cmd(const struct ethtool_link_ksettings *cmd)
>> +{
>> +	struct ethtool_link_ksettings diff1 = *cmd;
>> +	struct ethtool_link_ksettings diff2 = {};
>> +
>> +	/* cmd is always set so we need to clear it, validate the port type
>> +	 * and also without autonegotiation we can ignore advertising
>> +	 */
>> +	diff1.base.speed = 0;
>> +	diff2.base.port = PORT_OTHER;
>> +	ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
>> +	diff1.base.duplex = 0;
>> +	diff1.base.cmd = 0;
>> +	diff1.base.link_mode_masks_nwords = 0;
>> +
>> +	return !memcmp(&diff1.base, &diff2.base, sizeof(diff1.base)) &&
>> +		bitmap_empty(diff1.link_modes.supported,
>> +			     __ETHTOOL_LINK_MODE_MASK_NBITS) &&
>> +		bitmap_empty(diff1.link_modes.advertising,
>> +			     __ETHTOOL_LINK_MODE_MASK_NBITS) &&
> Isn't this condition always true? You zeroed the advertising bitmap
> above. Could you just omit this part and clearing of advertising above?
>
>> +		bitmap_empty(diff1.link_modes.lp_advertising,
>> +			     __ETHTOOL_LINK_MODE_MASK_NBITS);
>> +}
> Another idea: instead of zeroing parts of diff1, you could copy these
> members from *cmd to diff2 and compare cmd->base with diff2.base. You
> could then drop diff1. And you wouldn't even need whole struct
> ethtool_link_ksettings for diff2 as you only compare embedded struct
> ethtool_link_settings (and check two bitmaps in cmd->link_modes).

If I understand your suggestion correctly, then the validate function 
might look something like this?

/* Check if the user is trying to change anything besides speed/duplex */
static bool
ethtool_virtdev_validate_cmd(const struct ethtool_link_ksettings *cmd)
{
     struct ethtool_link_settings base2 = {};

     base2.speed = cmd->base.speed;
     base2.port = PORT_OTHER;
     base2.duplex = cmd->base.duplex;
     base2.cmd = cmd->base.cmd;
     base2.link_mode_masks_nwords = cmd->base.link_mode_masks_nwords;

     return !memcmp(&base2, cmd->base, sizeof(base2)) &&
         bitmap_empty(cmd->link_modes.supported,
                  __ETHTOOL_LINK_MODE_MASK_NBITS) &&
         bitmap_empty(cmd->link_modes.lp_advertising,
                  __ETHTOOL_LINK_MODE_MASK_NBITS);
}

Thanks again,

Tom

>
>> +
>>   /* convert a kernel internal ethtool_link_ksettings to
>>    * ethtool_link_usettings in user space. return 0 on success, errno on
>>    * error.
>> @@ -660,6 +686,17 @@ static int ethtool_get_link_ksettings(struct net_device *dev,
>>   	return store_link_ksettings_for_user(useraddr, &link_ksettings);
>>   }
>>   
>> +static int
>> +ethtool_virtdev_get_link_ksettings(struct net_device *dev,
>> +				   struct ethtool_link_ksettings *cmd,
>> +				   u32 *speed, u8 *duplex)
>> +{
>> +	cmd->base.speed = *speed;
>> +	cmd->base.duplex = *duplex;
>> +	cmd->base.port = PORT_OTHER;
>> +	return 0;
>> +}
> speed and duplex can be passed by value here; if you prefer pointers,
> please make them const.
>
> Michal Kubecek