From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Huascar Tejeda" Subject: Re: [PATCH] net: reset network device counters on the fly Date: Tue, 6 Jan 2009 23:17:57 -0400 Message-ID: <3375b4020901061917h2d61e9bbpdcc608e1cd9cfd9a@mail.gmail.com> References: <3375b4020901061808g527cb872ia64b47147783b91c@mail.gmail.com> <49641399.80305@hp.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Cc: netdev@vger.kernel.org To: "Rick Jones" Return-path: Received: from mail-ew0-f17.google.com ([209.85.219.17]:55901 "EHLO mail-ew0-f17.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751477AbZAGDR7 (ORCPT ); Tue, 6 Jan 2009 22:17:59 -0500 Received: by ewy10 with SMTP id 10so8799000ewy.13 for ; Tue, 06 Jan 2009 19:17:57 -0800 (PST) In-Reply-To: <49641399.80305@hp.com> Content-Disposition: inline Sender: netdev-owner@vger.kernel.org List-ID: Thanks a lot for your reply and sorry for sending the patch as an attachment; Anyway, here is the patch in case someone finds it useful. --- net/core/dev.c.orig 2009-01-05 17:38:51.000000000 -0400 +++ net/core/dev.c 2009-01-06 16:36:35.000000000 -0400 @@ -161,6 +161,11 @@ #define PTYPE_HASH_SIZE (16) #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1) +/* + * Maximum length of the command sent by the user + */ +#define RSTDEV_COMMAND_MAX_SIZE (15) + static DEFINE_SPINLOCK(ptype_lock); static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; static struct list_head ptype_all __read_mostly; /* Taps */ @@ -2671,10 +2676,78 @@ sizeof(struct seq_net_private)); } + +/* + * Called when /proc/net/dev is written + */ +static ssize_t reset_device_counters(struct file *file, const char *buffer, size_t len, loff_t * off) +{ + struct net_device *dev; + struct net_device_stats *stats; + char cmd[RSTDEV_COMMAND_MAX_SIZE]; + unsigned long cmd_size = 0; + char *cmd_ptr = cmd; + char **argv = (char **) kmalloc(sizeof(char *), GFP_KERNEL); + char *tmp_token, *action, *ifname; + int argc = 0; + + cmd_size = ( len > RSTDEV_COMMAND_MAX_SIZE ) ? RSTDEV_COMMAND_MAX_SIZE : len; + if ( copy_from_user(cmd, buffer, cmd_size) ) { + return -EFAULT; + } + + //Parse user input + while ( (tmp_token = strsep(&cmd_ptr, " \n")) ) { + if ( *tmp_token ) { + argv[argc] = (char *) kmalloc((strlen(tmp_token) + 1) * sizeof(char), GFP_KERNEL); + strcpy(argv[argc++], tmp_token); + } + } + + if ( !argv[0] || !argv[1] ) { + printk(KERN_INFO "Please usage: reset \"[network interface]\" or \"reset all\"\n"); + return -EINVAL; + } + + //TODO: create a struct for this. + action = argv[0]; + ifname = argv[1]; + + kfree(argv); + + if ( strstr(action, "reset") ) { + if ( !strcmp(ifname, "all") ) { + for_each_netdev(&init_net, dev) { + stats = dev->get_stats(dev); + if ( stats ) { + memset(stats, 0, sizeof(struct net_device_stats)); + } + } + } else { + dev = dev_get_by_name(&init_net, ifname); + if ( dev ) { + stats = dev->get_stats(dev); + if ( stats ) { + memset(stats, 0, sizeof(struct net_device_stats)); + } + } else { + printk(KERN_INFO "Device %s not found!\n", ifname); + return -ENODEV; + } + } + } else { + printk(KERN_INFO "\"reset\" is the only supported command!\n"); + return -EINVAL; + } + + return cmd_size; +} + static const struct file_operations dev_seq_fops = { .owner = THIS_MODULE, .open = dev_seq_open, .read = seq_read, + .write = reset_device_counters, .llseek = seq_lseek, .release = seq_release_net, }; Huascar Tejeda On Tue, Jan 6, 2009 at 10:29 PM, Rick Jones wrote: > Huascar Tejeda wrote: >> >> Hello everyone, >> >> This is my first post to the mailing list and also my first kernel patch. >> >> There was an scenario where I needed to clear network device counters >> without shutting down the interface. >> With this patch I add this functionality to /proc/net/dev. >> >> Usage example: >> echo clear eth0 > /proc/net/dev >> echo clear all > /proc/net/dev >> >> Thanks for your comments and please feel free to reply with any >> suggestions you may have. > > Having been at least once bitten as a patch submittor, I'll point-out that > patches are generally requested to be inline, not attachments. Also, in any > kernel tree there should be a file describing how a patch should be put > together and what addtional bits of information need to be included. That > file is ./Documentation/SubmittingPatches. > > Finally, my recollection is this same sort of thing (zeroing counters) has > been proposed before, and shot-down. Doesn't a priori mean it will be > shot-down this time, but it suggests the chances are slim. You can probably > find the discussion in one or more of the various archives for the netdev > list. For those situations where you want to know the statistics for a > given interval, you can snap the stats to files at either end of the > interval and run them through "beforeafter" or something similar: > > http://ftp.cup.hp.com/dist/networking/tools/ > > rick jones > wondering what has become of his ethtool patch to recognize speeds other > than 10/100/1000/10000... perhaps it too tripped over some of the above :( >