Hello Linus, Alan, David,
To make the API orthogonal, I have included a patch for SIOCGIFCOUNT,
which currently returns -EINVAL. The only reason I am providing this patch
is to make the API complete and make it easier to port applications from
other UNIX like OS'es.
I have verified and am sure that this patch works. The priority of this patch
is not very high, but it would be good to have a orthogonal/complete API.
The patch is against 2.4.2-2 (official rh 7.1). I can provide one against 2.4.10
or 2.4.10-pre16 if required.
--- dev.c.org Tue Oct 9 18:46:01 2001
+++ dev.c Tue Oct 9 18:36:39 2001
@@ -1651,6 +1651,40 @@
}
/*
+ * Implement the SIOCGIFCOUNT, ioctl to keep the API orthogonal.
+ * Basically taken from dev_ifconf - Balbir
+ */
+static int dev_ifcount(char *arg)
+{
+ struct net_device *dev;
+ int i;
+ unsigned int total;
+
+ total = 0;
+ /*
+ * May be introducing something like for_each_netdev(dev)
+ * on the lines of for_each_pci_dev would be useful here.
+ */
+ for (dev = dev_base; dev != NULL; dev = dev->next) {
+ for (i = 0; i < NPROTO; i++) {
+ if (gifconf_list[i]) {
+ int done;
+
+ done = gifconf_list[i](dev, NULL, 0);
+ if (done < 0)
+ return -EFAULT;
+ total++;
+ }
+ }
+ }
+
+ if (copy_to_user(arg, &total, sizeof(int)))
+ return -EFAULT;
+
+ return 0;
+}
+
+/*
* This is invoked by the /proc filesystem handler to display a device
* in detail.
*/
@@ -2218,6 +2252,18 @@
rtnl_shunlock();
return ret;
}
+
+ /*
+ * We do not need an exclusive lock for returning the count,
+ * shared lock is fine with us.
+ */
+ if (cmd == SIOCGIFCOUNT) {
+ rtnl_shlock();
+ ret = dev_ifcount((char *)arg);
+ rtnl_shunlock();
+ return ret;
+ }
+
if (cmd == SIOCGIFNAME) {
return dev_ifname((struct ifreq *)arg);
}
Comments, rejections, etc
Balbir
From: "BALBIR SINGH" <[email protected]>
Date: Tue, 09 Oct 2001 18:53:47 +0530
To make the API orthogonal, I have included a patch for SIOCGIFCOUNT,
which currently returns -EINVAL. The only reason I am providing this patch
is to make the API complete and make it easier to port applications from
other UNIX like OS'es.
There is no need for this change, and _EVEN_ if we put this
change in today every APP out there would _STILL_ need to deal with
all existing kernels which do not have SIOCGIFCOUNT implemented.
Furthermore, SIOCGIFCOUNT also gives no new functionality that does
not exist already. SIOCGIFCONF with a zero size with give the
behavior necessary to get the same answer as a SIOCGIFCOUNT would
provide. As far as I am aware, every system providing BSD sockets
provides this SIOCGIFCONF "feature".
Therefore, it is already quite easy to make applications portable
between Linux and other BSD socket based systems. Simply use the
SIOCGIFCONF method throughout.
Franks a lot,
David S. Miller
[email protected]
David S. Miller wrote:
> From: "BALBIR SINGH" <[email protected]>
> Date: Tue, 09 Oct 2001 18:53:47 +0530
>
> To make the API orthogonal, I have included a patch for SIOCGIFCOUNT,
> which currently returns -EINVAL. The only reason I am providing this patch
> is to make the API complete and make it easier to port applications from
> other UNIX like OS'es.
>
>There is no need for this change, and _EVEN_ if we put this
>change in today every APP out there would _STILL_ need to deal with
>all existing kernels which do not have SIOCGIFCOUNT implemented.
>
>Furthermore, SIOCGIFCOUNT also gives no new functionality that does
>not exist already. SIOCGIFCONF with a zero size with give the
>behavior necessary to get the same answer as a SIOCGIFCOUNT would
>provide. As far as I am aware, every system providing BSD sockets
>provides this SIOCGIFCONF "feature".
>
>Therefore, it is already quite easy to make applications portable
>between Linux and other BSD socket based systems. Simply use the
>SIOCGIFCONF method throughout.
>
I had mentioned even earlier, the same functionality can be obtained otherwise.
I was talking about porting from SUN, HP-UX, AIX, etc. I have had people coming
to me and telling me that SIOCGIFNUM (equivalent to SIOCGIFCOUNT) does not exist
and I have had to ask them to use SIOCGIFCONF, if ifc.ifc_len and ifc.ifc_buf set
to '0', to obtain the total size. Again, it is not a must it should be around.
But since the ioctl is there, it might be good to have it work the way some one
reading a man page would expect it to work, or if somebody wanted to port some
application from Sun, HP-UX or AIX would just do
#define SIOCGIFNUM SIOCGIFCOUNT
and get his application to compile without changing the code
Regards,
Balbir
>
>Franks a lot,
>David S. Miller
>[email protected]
>
>
Hello!
> application from Sun, HP-UX or AIX would just do
>
> #define SIOCGIFNUM SIOCGIFCOUNT
Grrr.... Listen what the hell did you call this ioctl with _another_ name,
absent both in BSD and in SVR4??? :-)
I remember the patch doing the same thing and I zapped this crap
exactly because it was another linuxish hack moving from bsd api
and not moving closer to svr4.
Alexey
PS. Well, and did you think about one thing: "linux" port of any application
which does not work with linuxes can smell a bit anti-social for many people,
who do not compile from sources. And for all the people, if the port is binary.