I have made a patch to fix the counter values in /proc/net/dev. The
problem was that the tx_bytes and rx_bytes will reset when ~4GB is
transferred. This patch has been tested to work with linux-2.5.0 but it
should work on all 2.4.* kernels. Also it should work with most of the
interface cards but not all yet.
This is the first version of my patch so please give me some comments
about it.
--
Miika
Patch to fix /proc/net/dev counters. Patch made by Miika Pekkarinen
<[email protected]> or <[email protected]>.
--- linux-2.5.0or/net/core/dev.c Thu Nov 8 00:39:36 2001
+++ linux/net/core/dev.c Tue Dec 11 21:30:49 2001
@@ -1688,7 +1688,7 @@
int size;
if (stats)
- size = sprintf(buffer, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu %8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
+ size = sprintf(buffer, "%6s:%14llu %7lu %4lu %4lu %4lu %5lu %10lu %9lu %14llu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
dev->name,
stats->rx_bytes,
stats->rx_packets, stats->rx_errors,
@@ -1724,8 +1724,8 @@
size = sprintf(buffer,
- "Inter-| Receive | Transmit\n"
- " face |bytes packets errs drop fifo frame compressed multicast|bytes packets errs drop fifo colls carrier compressed\n");
+ "Inter-| --------------------------- Receive -------------------------- | ------------------------- Transmit --------------------------\n"
+ " face |bytes packets errs drop fifo frame compressed multicast|bytes packets errs drop fifo colls carrier compressed\n");
pos += size;
len += size;
--- linux-2.5.0or/include/linux/netdevice.h Thu Nov 22 21:47:09 2001
+++ linux/include/linux/netdevice.h Tue Dec 11 21:24:54 2001
@@ -97,8 +97,9 @@
{
unsigned long rx_packets; /* total packets received */
unsigned long tx_packets; /* total packets transmitted */
- unsigned long rx_bytes; /* total bytes received */
- unsigned long tx_bytes; /* total bytes transmitted */
+ /* rx and tx counters fixed up to 64-bit */
+ long long rx_bytes; /* total bytes received */
+ long long tx_bytes; /* total bytes transmitted */
unsigned long rx_errors; /* bad packets received */
unsigned long tx_errors; /* packet transmit problems */
unsigned long rx_dropped; /* no space in linux buffers */
--- linux-2.5.0or/include/linux/netdevice.h Thu Nov 22 21:47:09 2001
+++ linux/include/linux/netdevice.h Tue Dec 11 21:24:54 2001
@@ -97,8 +97,9 @@
{
unsigned long rx_packets; /* total packets received */
unsigned long tx_packets; /* total packets transmitted */
- unsigned long rx_bytes; /* total bytes received */
- unsigned long tx_bytes; /* total bytes transmitted */
+ /* rx and tx counters fixed up to 64-bit */
+ long long rx_bytes; /* total bytes received
*/
+ long long tx_bytes; /* total bytes transmitted
*/
It is better to define these as unsigned long long, you are anyway
printing them as unsigned long long. Its about time this was fixed,
I have Gigabit cards and it is very easy to produce an overflow.
I think the MAINTAINER for ifconfig(8) should be made aware of these
changes as well.
Balbir
> I have made a patch to fix the counter values in /proc/net/dev. The
> problem was that the tx_bytes and rx_bytes will reset when ~4GB is
> transferred. This patch has been tested to work with linux-2.5.0 but it
> should work on all 2.4.* kernels. Also it should work with most of the
> interface cards but not all yet.
It's something we've always avoided doing. A long long is expensive to
update and not an atomic type on things like the x86. The firewalling
facilities let you collect 64bit accounting data if you need it