Hi Linus,
attached you find a patch against Linux 2.5.7 that revives the
IFF_RUNNING flag in the dev->flags element. Some time ago, Alexey has
replaced IFF_RUNNING by a bit in dev->state, AFAIK because of locking
reasons.
When enabled, a thread is compiled into the kernel that scans the
network device list on every netif_carrier_on()/_off() called by a
device driver and mirrors the !LINK_STATE_NOCARRIER bit of dev->state
into IFF_RUNNING. If these values differ for a device, a netlink message
is broadcasted. This allows user processes to be notified about
operative state changes. Actually, I've written some patches for the
zebra routing daemon so that it stops using an interface while the
driver thinks that it is down.
There are still drivers left that play with IFF_RUNNING directly, but
I think they can be updated during 2.5 development cycle. The patch
seems to be stable, I've also submitted a 2.4 version for the -ac tree.
Cheers, Stefan
diff -X dontdiff -urN linux-2.5.7/include/linux/netdevice.h linux-2.5.7-stefan/include/linux/netdevice.h
--- linux-2.5.7/include/linux/netdevice.h Mon Mar 18 21:37:18 2002
+++ linux-2.5.7-stefan/include/linux/netdevice.h Sat Mar 30 16:08:34 2002
@@ -210,6 +210,12 @@
__LINK_STATE_RX_SCHED
};
+/* This gets called by netif_carrier_on()/_off() whenever
+ * state of an interface changes
+ */
+#ifdef CONFIG_LINKWATCH
+extern void wake_linkwatch_thread(void);
+#endif
/*
* This structure holds at boot time configured netdevice settings. They
@@ -639,14 +645,24 @@
static inline void netif_carrier_on(struct net_device *dev)
{
+#ifdef CONFIG_LINKWATCH
+ if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state))
+ wake_linkwatch_thread();
+#else
clear_bit(__LINK_STATE_NOCARRIER, &dev->state);
+#endif
if (netif_running(dev))
__netdev_watchdog_up(dev);
}
static inline void netif_carrier_off(struct net_device *dev)
{
+#ifdef CONFIG_LINKWATCH
+ if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state))
+ wake_linkwatch_thread();
+#else
set_bit(__LINK_STATE_NOCARRIER, &dev->state);
+#endif
}
/* Hot-plugging. */
diff -X dontdiff -urN linux-2.5.7/net/Config.help linux-2.5.7-stefan/net/Config.help
--- linux-2.5.7/net/Config.help Mon Mar 18 21:37:07 2002
+++ linux-2.5.7-stefan/net/Config.help Sat Mar 30 15:46:28 2002
@@ -385,6 +385,14 @@
subnetwork boundaries. These shortcut connections bypass routers
enhancing overall network performance.
+CONFIG_LINKWATCH
+ When this option is enabled, the kernel will forward changes in the
+ operative ("RUNNING") state of an interface via the netlink socket.
+ This is most useful when running linux as a router. Note that currently
+ not many drivers maintain an administrative state, a few even break this
+ option. Compliant drivers change the RUNNING flag in ifconfig output
+ depending on operative state. If unsure, say 'N'.
+
CONFIG_ECONET
Econet is a fairly old and slow networking protocol mainly used by
Acorn computers to access file and print servers. It uses native
diff -X dontdiff -urN linux-2.5.7/net/Config.in linux-2.5.7-stefan/net/Config.in
--- linux-2.5.7/net/Config.in Mon Mar 18 21:37:08 2002
+++ linux-2.5.7-stefan/net/Config.in Sat Mar 30 15:45:38 2002
@@ -46,6 +46,9 @@
fi
fi
tristate '802.1Q VLAN Support' CONFIG_VLAN_8021Q
+if [ "$CONFIG_EXPERIMENTAL" = "y" ]; then
+ bool 'Device link state notification (EXPERIMENTAL)' CONFIG_LINKWATCH
+fi
comment ' '
tristate 'The IPX protocol' CONFIG_IPX
diff -X dontdiff -urN linux-2.5.7/net/core/Makefile linux-2.5.7-stefan/net/core/Makefile
--- linux-2.5.7/net/core/Makefile Mon Mar 18 21:37:13 2002
+++ linux-2.5.7-stefan/net/core/Makefile Sat Mar 30 15:42:35 2002
@@ -30,4 +30,6 @@
# Ugly. I wish all wireless drivers were moved in drivers/net/wireless
obj-$(CONFIG_NET_PCMCIA_RADIO) += wireless.o
+obj-$(CONFIG_LINKWATCH) += link_watch.o
+
include $(TOPDIR)/Rules.make
diff -X dontdiff -urN linux-2.5.7/net/core/dev.c linux-2.5.7-stefan/net/core/dev.c
--- linux-2.5.7/net/core/dev.c Mon Mar 18 21:37:15 2002
+++ linux-2.5.7-stefan/net/core/dev.c Sat Mar 30 15:42:35 2002
@@ -720,6 +720,12 @@
*/
dev->flags |= IFF_UP;
+ if (netif_carrier_ok(dev)) {
+ dev->flags |= IFF_RUNNING;
+ } else {
+ dev->flags &= ~IFF_RUNNING;
+ }
+
set_bit(__LINK_STATE_START, &dev->state);
/*
@@ -826,7 +832,7 @@
* Device is now down.
*/
- dev->flags &= ~IFF_UP;
+ dev->flags &= ~(IFF_UP | IFF_RUNNING);
#ifdef CONFIG_NET_FASTROUTE
dev_clear_fastroute(dev);
#endif
@@ -2668,7 +2674,9 @@
#ifdef CONFIG_NET_DIVERT
extern void dv_init(void);
#endif /* CONFIG_NET_DIVERT */
-
+#ifdef CONFIG_LINKWATCH
+extern void linkwatch_init(void);
+#endif /* CONFIG_LINKWATCH */
/*
* Callers must hold the rtnl semaphore. See the comment at the
@@ -2811,6 +2819,10 @@
*/
net_device_init();
+
+#ifdef CONFIG_LINKWATCH
+ linkwatch_init();
+#endif /* CONFIG_LINKWATCH */
return 0;
}
diff -X dontdiff -urN linux-2.5.7/net/core/link_watch.c linux-2.5.7-stefan/net/core/link_watch.c
--- linux-2.5.7/net/core/link_watch.c Thu Jan 1 01:00:00 1970
+++ linux-2.5.7-stefan/net/core/link_watch.c Sat Mar 30 16:58:51 2002
@@ -0,0 +1,106 @@
+/*
+ * Linux network device state notifaction
+ *
+ * Author:
+ * Stefan Rompf <[email protected]>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ *
+ */
+
+#include <linux/sched.h>
+#include <linux/wait.h>
+#include <linux/config.h>
+#include <linux/module.h>
+#include <linux/netdevice.h>
+#include <linux/if.h>
+#include <linux/rtnetlink.h>
+#include <asm/bitops.h>
+#include <asm/types.h>
+
+/* Used by wake_linkwatch_thread() */
+static wait_queue_head_t linkwatch_wqh;
+static unsigned long linkwatch_nowait;
+
+static int linkwatch_thread(void *dummy)
+{
+ struct net_device *dev;
+ DECLARE_WAITQUEUE(linkwatch_wq, current);
+
+ daemonize();
+ spin_lock_irq(¤t->sigmask_lock);
+ sigfillset(¤t->blocked);
+ recalc_sigpending();
+ spin_unlock_irq(¤t->sigmask_lock);
+
+ strcpy(current->comm, "linkwatchd");
+
+ while(1) {
+ clear_bit(0, &linkwatch_nowait);
+ rtnl_lock();
+ for (dev=dev_base; dev; dev = dev->next) {
+
+ /* State of netif_carrier_ok() is reflected
+ into dev_flags by this loop, and a netlink
+ message is omitted whenever the state
+ changes */
+
+ if (!(dev->flags & IFF_UP)) continue;
+
+ if (dev->flags & IFF_RUNNING) {
+ if (!netif_carrier_ok(dev)) {
+ write_lock(&dev_base_lock);
+ dev->flags &= ~IFF_RUNNING;
+ write_unlock(&dev_base_lock);
+ netdev_state_change(dev);
+ }
+ } else {
+ if (netif_carrier_ok(dev)) {
+ write_lock(&dev_base_lock);
+ dev->flags |= IFF_RUNNING;
+ write_unlock(&dev_base_lock);
+ netdev_state_change(dev);
+ }
+ }
+ }
+ rtnl_unlock();
+
+ /* Prevent runaway on rapid state changes */
+ set_current_state(TASK_INTERRUPTIBLE);
+ schedule_timeout(HZ);
+
+ set_current_state(TASK_INTERRUPTIBLE);
+ if (test_bit(0, &linkwatch_nowait)) {
+ set_current_state(TASK_RUNNING);
+ } else {
+ add_wait_queue(&linkwatch_wqh, &linkwatch_wq);
+ schedule();
+ remove_wait_queue(&linkwatch_wqh, &linkwatch_wq);
+ }
+ }
+
+ return 0; /* Not reached */
+}
+
+
+void wake_linkwatch_thread() {
+ set_bit(0, &linkwatch_nowait);
+ wake_up(&linkwatch_wqh);
+}
+
+
+void __init linkwatch_init(void) {
+ pid_t me;
+
+ init_waitqueue_head(&linkwatch_wqh);
+ linkwatch_nowait = 0;
+ me = kernel_thread(linkwatch_thread, NULL, CLONE_FS | CLONE_FILES | CLONE_SIGNAL);
+
+ if (!me) {
+ printk(KERN_INFO "Failed to init linkwatch thread, continuing without\n");
+ }
+}
+
diff -X dontdiff -urN linux-2.5.7/net/netsyms.c linux-2.5.7-stefan/net/netsyms.c
--- linux-2.5.7/net/netsyms.c Mon Mar 18 21:37:13 2002
+++ linux-2.5.7-stefan/net/netsyms.c Sat Mar 30 15:42:35 2002
@@ -596,4 +596,8 @@
EXPORT_SYMBOL(wireless_send_event);
#endif /* CONFIG_NET_RADIO || CONFIG_NET_PCMCIA_RADIO */
+#ifdef CONFIG_LINKWATCH
+EXPORT_SYMBOL(wake_linkwatch_thread);
+#endif
+
#endif /* CONFIG_NET */
Hi,
in the future can you please just post to netdev on networking
related issues? I responded to lk, but feel free to remove it
from the list.
-I am not sure the idea of using a kernel thread is the best. Maybe
move the checks to both the tx or rx softirqs instead of its own scheduling.
-In particular, it would be a better idea not to just go walking all the
devices; only walk devices that have raised an netif_carrier_.
-A shared devices bitmask across SMP should be enough (i.e no need for
per-CPU state)
-Another thing might be to double check that the condition that raised
the state change is still valid example -
in between the moment you say a link is down due to some bad hardware
fault to the moment some device timer recovers it;
-Also IFF_RUNNING seems to have inconsistent semantics in a lot of
drivers. It should really stand for "operational status" whereas
IFF_UP should stand for "admin status" -- anyone wanna shed historical
wisdom here?
cheers,
jamal