Re: [PATCHv2 net-next] dropwatch: Support monitoring of dropped frames

[Michal Schmidt]

Dne 04. 08. 20 v 18:09 izabela.bakollari@xxxxxxxxx napsala:
> From: Izabela Bakollari <izabela.bakollari@xxxxxxxxx>
>
> Dropwatch is a utility that monitors dropped frames by having userspace
> record them over the dropwatch protocol over a file. This augument
> allows live monitoring of dropped frames using tools like tcpdump.
>
> With this feature, dropwatch allows two additional commands (start and
> stop interface) which allows the assignment of a net_device to the
> dropwatch protocol. When assinged, dropwatch will clone dropped frames,
> and receive them on the assigned interface, allowing tools like tcpdump
> to monitor for them.
>
> With this feature, create a dummy ethernet interface (ip link add dev
> dummy0 type dummy), assign it to the dropwatch kernel subsystem, by using
> these new commands, and then monitor dropped frames in real time by
> running tcpdump -i dummy0.
>
> Signed-off-by: Izabela Bakollari <izabela.bakollari@xxxxxxxxx>
> ---
> Changes in v2:
> - protect the dummy ethernet interface from being changed by another
> thread/cpu
> ---
>   include/uapi/linux/net_dropmon.h |  3 ++
>   net/core/drop_monitor.c          | 84 ++++++++++++++++++++++++++++++++
>   2 files changed, 87 insertions(+)
[...]
> @@ -255,6 +259,21 @@ static void trace_drop_common(struct sk_buff *skb, void *location)
>   
>   out:
>   	spin_unlock_irqrestore(&data->lock, flags);
> +	spin_lock_irqsave(&interface_lock, flags);
> +	if (interface && interface != skb->dev) {
> +		skb = skb_clone(skb, GFP_ATOMIC);

I suggest naming the cloned skb "nskb". Less potential for confusion 
that way.

> +		if (skb) {
> +			skb->dev = interface;
> +			spin_unlock_irqrestore(&interface_lock, flags);
> +			netif_receive_skb(skb);
> +		} else {
> +			spin_unlock_irqrestore(&interface_lock, flags);
> +			pr_err("dropwatch: Not enough memory to clone dropped skb\n");

Maybe avoid logging the error here. In NET_DM_ALERT_MODE_PACKET mode, 
drop monitor does not log about the skb_clone() failure either.
We don't want to open the possibility to flood the logs in case this 
somehow gets triggered by every packet.

A coding style suggestion - can you rearrange it so that the error path 
code is spelled out first? Then the regular path does not have to be 
indented further:

      nskb = skb_clone(skb, GFP_ATOMIC);
      if (!nskb) {
              spin_unlock_irqrestore(&interface_lock, flags);
              return;
      }

      /* ... implicit else ... Proceed normally ... */

> +			return;
> +		}
> +	} else {
> +		spin_unlock_irqrestore(&interface_lock, flags);
> +	}
>   }
>   
>   static void trace_kfree_skb_hit(void *ignore, struct sk_buff *skb, void *location)
> @@ -1315,6 +1334,53 @@ static int net_dm_cmd_trace(struct sk_buff *skb,
>   	return -EOPNOTSUPP;
>   }
>   
> +static int net_dm_interface_start(struct net *net, const char *ifname)
> +{
> +	struct net_device *nd = dev_get_by_name(net, ifname);
> +
> +	if (nd)
> +		interface = nd;
> +	else
> +		return -ENODEV;
> +
> +	return 0;

Similarly here, consider:

  if (!nd)
          return -ENODEV;

  interface = nd;
  return 0;

But maybe I'm nitpicking ...

> +}
> +
> +static int net_dm_interface_stop(struct net *net, const char *ifname)
> +{
> +	dev_put(interface);
> +	interface = NULL;
> +
> +	return 0;
> +}
> +
> +static int net_dm_cmd_ifc_trace(struct sk_buff *skb, struct genl_info *info)
> +{
> +	struct net *net = sock_net(skb->sk);
> +	char ifname[IFNAMSIZ];
> +
> +	if (net_dm_is_monitoring())
> +		return -EBUSY;
> +
> +	memset(ifname, 0, IFNAMSIZ);
> +	nla_strlcpy(ifname, info->attrs[NET_DM_ATTR_IFNAME], IFNAMSIZ - 1);
> +
> +	switch (info->genlhdr->cmd) {
> +	case NET_DM_CMD_START_IFC:
> +		if (!interface)
> +			return net_dm_interface_start(net, ifname);
> +		else
> +			return -EBUSY;
> +	case NET_DM_CMD_STOP_IFC:
> +		if (interface)
> +			return net_dm_interface_stop(net, interface->name);
> +		else
> +			return -ENODEV;

... and here too.

Best regards,
Michal