Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6] / net / core / link_watch.c
1 /*
2  * Linux network device link state notification
3  *
4  * Author:
5  *     Stefan Rompf <sux@loplof.de>
6  *
7  * This program is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU General Public License
9  * as published by the Free Software Foundation; either version
10  * 2 of the License, or (at your option) any later version.
11  *
12  */
13
14 #include <linux/module.h>
15 #include <linux/netdevice.h>
16 #include <linux/if.h>
17 #include <net/sock.h>
18 #include <net/pkt_sched.h>
19 #include <linux/rtnetlink.h>
20 #include <linux/jiffies.h>
21 #include <linux/spinlock.h>
22 #include <linux/slab.h>
23 #include <linux/workqueue.h>
24 #include <linux/bitops.h>
25 #include <asm/types.h>
26
27
28 enum lw_bits {
29         LW_URGENT = 0,
30 };
31
32 static unsigned long linkwatch_flags;
33 static unsigned long linkwatch_nextevent;
34
35 static void linkwatch_event(struct work_struct *dummy);
36 static DECLARE_DELAYED_WORK(linkwatch_work, linkwatch_event);
37
38 static struct net_device *lweventlist;
39 static DEFINE_SPINLOCK(lweventlist_lock);
40
41 static unsigned char default_operstate(const struct net_device *dev)
42 {
43         if (!netif_carrier_ok(dev))
44                 return (dev->ifindex != dev->iflink ?
45                         IF_OPER_LOWERLAYERDOWN : IF_OPER_DOWN);
46
47         if (netif_dormant(dev))
48                 return IF_OPER_DORMANT;
49
50         return IF_OPER_UP;
51 }
52
53
54 static void rfc2863_policy(struct net_device *dev)
55 {
56         unsigned char operstate = default_operstate(dev);
57
58         if (operstate == dev->operstate)
59                 return;
60
61         write_lock_bh(&dev_base_lock);
62
63         switch(dev->link_mode) {
64         case IF_LINK_MODE_DORMANT:
65                 if (operstate == IF_OPER_UP)
66                         operstate = IF_OPER_DORMANT;
67                 break;
68
69         case IF_LINK_MODE_DEFAULT:
70         default:
71                 break;
72         }
73
74         dev->operstate = operstate;
75
76         write_unlock_bh(&dev_base_lock);
77 }
78
79
80 static int linkwatch_urgent_event(struct net_device *dev)
81 {
82         return netif_running(dev) && netif_carrier_ok(dev) &&
83                dev->qdisc != dev->qdisc_sleeping;
84 }
85
86
87 static void linkwatch_add_event(struct net_device *dev)
88 {
89         unsigned long flags;
90
91         spin_lock_irqsave(&lweventlist_lock, flags);
92         dev->link_watch_next = lweventlist;
93         lweventlist = dev;
94         spin_unlock_irqrestore(&lweventlist_lock, flags);
95 }
96
97
98 static void linkwatch_schedule_work(int urgent)
99 {
100         unsigned long delay = linkwatch_nextevent - jiffies;
101
102         if (test_bit(LW_URGENT, &linkwatch_flags))
103                 return;
104
105         /* Minimise down-time: drop delay for up event. */
106         if (urgent) {
107                 if (test_and_set_bit(LW_URGENT, &linkwatch_flags))
108                         return;
109                 delay = 0;
110         }
111
112         /* If we wrap around we'll delay it by at most HZ. */
113         if (delay > HZ)
114                 delay = 0;
115
116         /*
117          * This is true if we've scheduled it immeditately or if we don't
118          * need an immediate execution and it's already pending.
119          */
120         if (schedule_delayed_work(&linkwatch_work, delay) == !delay)
121                 return;
122
123         /* Don't bother if there is nothing urgent. */
124         if (!test_bit(LW_URGENT, &linkwatch_flags))
125                 return;
126
127         /* It's already running which is good enough. */
128         if (!cancel_delayed_work(&linkwatch_work))
129                 return;
130
131         /* Otherwise we reschedule it again for immediate exection. */
132         schedule_delayed_work(&linkwatch_work, 0);
133 }
134
135
136 static void __linkwatch_run_queue(int urgent_only)
137 {
138         struct net_device *next;
139
140         /*
141          * Limit the number of linkwatch events to one
142          * per second so that a runaway driver does not
143          * cause a storm of messages on the netlink
144          * socket.  This limit does not apply to up events
145          * while the device qdisc is down.
146          */
147         if (!urgent_only)
148                 linkwatch_nextevent = jiffies + HZ;
149         /* Limit wrap-around effect on delay. */
150         else if (time_after(linkwatch_nextevent, jiffies + HZ))
151                 linkwatch_nextevent = jiffies;
152
153         clear_bit(LW_URGENT, &linkwatch_flags);
154
155         spin_lock_irq(&lweventlist_lock);
156         next = lweventlist;
157         lweventlist = NULL;
158         spin_unlock_irq(&lweventlist_lock);
159
160         while (next) {
161                 struct net_device *dev = next;
162
163                 next = dev->link_watch_next;
164
165                 if (urgent_only && !linkwatch_urgent_event(dev)) {
166                         linkwatch_add_event(dev);
167                         continue;
168                 }
169
170                 /*
171                  * Make sure the above read is complete since it can be
172                  * rewritten as soon as we clear the bit below.
173                  */
174                 smp_mb__before_clear_bit();
175
176                 /* We are about to handle this device,
177                  * so new events can be accepted
178                  */
179                 clear_bit(__LINK_STATE_LINKWATCH_PENDING, &dev->state);
180
181                 rfc2863_policy(dev);
182                 if (dev->flags & IFF_UP) {
183                         if (netif_carrier_ok(dev)) {
184                                 WARN_ON(dev->qdisc_sleeping == &noop_qdisc);
185                                 dev_activate(dev);
186                         } else
187                                 dev_deactivate(dev);
188
189                         netdev_state_change(dev);
190                 }
191
192                 dev_put(dev);
193         }
194
195         if (lweventlist)
196                 linkwatch_schedule_work(0);
197 }
198
199
200 /* Must be called with the rtnl semaphore held */
201 void linkwatch_run_queue(void)
202 {
203         __linkwatch_run_queue(0);
204 }
205
206
207 static void linkwatch_event(struct work_struct *dummy)
208 {
209         rtnl_lock();
210         __linkwatch_run_queue(time_after(linkwatch_nextevent, jiffies));
211         rtnl_unlock();
212 }
213
214
215 void linkwatch_fire_event(struct net_device *dev)
216 {
217         int urgent = linkwatch_urgent_event(dev);
218
219         if (!test_and_set_bit(__LINK_STATE_LINKWATCH_PENDING, &dev->state)) {
220                 dev_hold(dev);
221
222                 linkwatch_add_event(dev);
223         } else if (!urgent)
224                 return;
225
226         linkwatch_schedule_work(urgent);
227 }
228
229 EXPORT_SYMBOL(linkwatch_fire_event);