Re: [PATCH updated] net: add ability to clear per-interface networkstatistics via procfs

[Eric Dumazet]

James Cammarata a écrit :
> > 1) Every call to get_proc_net() should be paired with a call to 
> > put_net(), or else you  leak  struct net.
>
> My mistake, I will add this and resubmit.
This version has a bug, please check again.
> > 2) I am not sure this is the right way to do this... Did you consider 
> > to extend ethtool instead ?
>
> Yes, I did consider that.  My reasoning for adding it to procfs is 
> that each
> driver defines their own ethtool interface, so it was much simpler to 
> add it here, with a lot less code changes and broader support.
Maybe much simpler but looks like a hack...
> ---
>
> --- linux-2.6.25.4/net/core/dev.c       2008-05-15 10:00:12.000000000 
> -0500
> +++ linux-2.6.25.4-jcammara/net/core/dev.c      2008-05-17 
> 09:50:22.000000000 -0500
> @@ -2455,2 +2455,80 @@
>
> +/**
> + * proc_net_dev_write - handle writes to /proc/net/dev
> + * @file: not used
> + * @buf: buffer to write
> + * @length: length of buf, at most PAGE_SIZE
> + * @ppos: not used
> + *
> + * Description: this provides a mechanism to clear statistics on a
> + * per-interface basis
> + * "echo 'net clear-stats ifdev' >/proc/net/dev"
> + * with  "ifdev" replaced by the device name you wish to clear.
> + *
> + */
> +static ssize_t proc_net_dev_write(struct file *file, const char 
> __user *buf,
> +                                  size_t length, loff_t *ppos)
> +{
> +        char *buffer, *p;
> +        char devname[IFNAMSIZ];
> +        struct net *net;
> +        struct net_device *dev;
> +        int err;
> +
> +        if (!buf || length > PAGE_SIZE)
> +                return -EINVAL;
> +
> +        buffer = (char *)__get_free_page(GFP_KERNEL);
> +        if (!buffer)
> +                return -ENOMEM;
> +
> +        err = -EFAULT;
> +        if (copy_from_user(buffer, buf, length))
net is not initialized so you will call put_net(garbage) and crash

My suggestion was better, since you dont have to test if (net) ,
just call put_net(net) at the right place (before out: label)
> +                goto out;
> +
> +        err = -EINVAL;
> +        if (length < PAGE_SIZE)
> +                buffer[length] = '\0';
> +        else if (buffer[PAGE_SIZE-1])
> +                goto out;
> +
> +        err = -ENXIO;
> +        net = get_proc_net(file->f_dentry->d_inode);
> +        if (!net)
> +                goto out;
> +
> +        /*
> +         * Usage: echo "net clear-stats ifdev" >/proc/net/dev
> +         * with  "ifdev" replaced by the device name you wish to clear.
> +         */
> +        if (!strncmp("net clear-stats",buffer,15)) {
> +                p = buffer + 16;
> +                if(sscanf(p,"%16s",devname)>0) {
> +                        dev = dev_get_by_name(net,devname);
> +                        if (dev) {
> +                                if (dev->get_stats) {
> +                                        struct net_device_stats *stats =
> +                                               dev->get_stats(dev);
> +                                        memset(stats,0,
> +                                               sizeof(struct 
> net_device_stats));
> +                                }
> +                                dev_put(dev);
> +                        }
> +                }
> +        }
> +
> +        /*
> +         * convert success returns so that we return the
> +         * number of bytes consumed.
> +         */
> +        if (!err)
> +                err = length;
> +
> + out:
> +        if(net)
> +                put_net(net);
> +        free_page((unsigned long)buffer);
> +        return err;
> +}
> +
> static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
> @@ -2498,2 +2576,3 @@
>        .read    = seq_read,
> +        .write   = proc_net_dev_write,
>        .llseek  = seq_lseek,




--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/