Loading...
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/workqueue.h>
23#include <linux/bitops.h>
24#include <asm/types.h>
25
26
27enum lw_bits {
28 LW_URGENT = 0,
29};
30
31static unsigned long linkwatch_flags;
32static unsigned long linkwatch_nextevent;
33
34static void linkwatch_event(struct work_struct *dummy);
35static DECLARE_DELAYED_WORK(linkwatch_work, linkwatch_event);
36
37static LIST_HEAD(lweventlist);
38static DEFINE_SPINLOCK(lweventlist_lock);
39
40static unsigned char default_operstate(const struct net_device *dev)
41{
42 if (!netif_carrier_ok(dev))
43 return (dev->ifindex != dev->iflink ?
44 IF_OPER_LOWERLAYERDOWN : IF_OPER_DOWN);
45
46 if (netif_dormant(dev))
47 return IF_OPER_DORMANT;
48
49 return IF_OPER_UP;
50}
51
52
53static void rfc2863_policy(struct net_device *dev)
54{
55 unsigned char operstate = default_operstate(dev);
56
57 if (operstate == dev->operstate)
58 return;
59
60 write_lock_bh(&dev_base_lock);
61
62 switch(dev->link_mode) {
63 case IF_LINK_MODE_DORMANT:
64 if (operstate == IF_OPER_UP)
65 operstate = IF_OPER_DORMANT;
66 break;
67
68 case IF_LINK_MODE_DEFAULT:
69 default:
70 break;
71 }
72
73 dev->operstate = operstate;
74
75 write_unlock_bh(&dev_base_lock);
76}
77
78
79static bool linkwatch_urgent_event(struct net_device *dev)
80{
81 if (!netif_running(dev))
82 return false;
83
84 if (dev->ifindex != dev->iflink)
85 return true;
86
87 return netif_carrier_ok(dev) && qdisc_tx_changing(dev);
88}
89
90
91static void linkwatch_add_event(struct net_device *dev)
92{
93 unsigned long flags;
94
95 spin_lock_irqsave(&lweventlist_lock, flags);
96 if (list_empty(&dev->link_watch_list)) {
97 list_add_tail(&dev->link_watch_list, &lweventlist);
98 dev_hold(dev);
99 }
100 spin_unlock_irqrestore(&lweventlist_lock, flags);
101}
102
103
104static void linkwatch_schedule_work(int urgent)
105{
106 unsigned long delay = linkwatch_nextevent - jiffies;
107
108 if (test_bit(LW_URGENT, &linkwatch_flags))
109 return;
110
111 /* Minimise down-time: drop delay for up event. */
112 if (urgent) {
113 if (test_and_set_bit(LW_URGENT, &linkwatch_flags))
114 return;
115 delay = 0;
116 }
117
118 /* If we wrap around we'll delay it by at most HZ. */
119 if (delay > HZ)
120 delay = 0;
121
122 /*
123 * This is true if we've scheduled it immeditately or if we don't
124 * need an immediate execution and it's already pending.
125 */
126 if (schedule_delayed_work(&linkwatch_work, delay) == !delay)
127 return;
128
129 /* Don't bother if there is nothing urgent. */
130 if (!test_bit(LW_URGENT, &linkwatch_flags))
131 return;
132
133 /* It's already running which is good enough. */
134 if (!__cancel_delayed_work(&linkwatch_work))
135 return;
136
137 /* Otherwise we reschedule it again for immediate execution. */
138 schedule_delayed_work(&linkwatch_work, 0);
139}
140
141
142static void linkwatch_do_dev(struct net_device *dev)
143{
144 /*
145 * Make sure the above read is complete since it can be
146 * rewritten as soon as we clear the bit below.
147 */
148 smp_mb__before_clear_bit();
149
150 /* We are about to handle this device,
151 * so new events can be accepted
152 */
153 clear_bit(__LINK_STATE_LINKWATCH_PENDING, &dev->state);
154
155 rfc2863_policy(dev);
156 if (dev->flags & IFF_UP) {
157 if (netif_carrier_ok(dev))
158 dev_activate(dev);
159 else
160 dev_deactivate(dev);
161
162 netdev_state_change(dev);
163 }
164 dev_put(dev);
165}
166
167static void __linkwatch_run_queue(int urgent_only)
168{
169 struct net_device *dev;
170 LIST_HEAD(wrk);
171
172 /*
173 * Limit the number of linkwatch events to one
174 * per second so that a runaway driver does not
175 * cause a storm of messages on the netlink
176 * socket. This limit does not apply to up events
177 * while the device qdisc is down.
178 */
179 if (!urgent_only)
180 linkwatch_nextevent = jiffies + HZ;
181 /* Limit wrap-around effect on delay. */
182 else if (time_after(linkwatch_nextevent, jiffies + HZ))
183 linkwatch_nextevent = jiffies;
184
185 clear_bit(LW_URGENT, &linkwatch_flags);
186
187 spin_lock_irq(&lweventlist_lock);
188 list_splice_init(&lweventlist, &wrk);
189
190 while (!list_empty(&wrk)) {
191
192 dev = list_first_entry(&wrk, struct net_device, link_watch_list);
193 list_del_init(&dev->link_watch_list);
194
195 if (urgent_only && !linkwatch_urgent_event(dev)) {
196 list_add_tail(&dev->link_watch_list, &lweventlist);
197 continue;
198 }
199 spin_unlock_irq(&lweventlist_lock);
200 linkwatch_do_dev(dev);
201 spin_lock_irq(&lweventlist_lock);
202 }
203
204 if (!list_empty(&lweventlist))
205 linkwatch_schedule_work(0);
206 spin_unlock_irq(&lweventlist_lock);
207}
208
209void linkwatch_forget_dev(struct net_device *dev)
210{
211 unsigned long flags;
212 int clean = 0;
213
214 spin_lock_irqsave(&lweventlist_lock, flags);
215 if (!list_empty(&dev->link_watch_list)) {
216 list_del_init(&dev->link_watch_list);
217 clean = 1;
218 }
219 spin_unlock_irqrestore(&lweventlist_lock, flags);
220 if (clean)
221 linkwatch_do_dev(dev);
222}
223
224
225/* Must be called with the rtnl semaphore held */
226void linkwatch_run_queue(void)
227{
228 __linkwatch_run_queue(0);
229}
230
231
232static void linkwatch_event(struct work_struct *dummy)
233{
234 rtnl_lock();
235 __linkwatch_run_queue(time_after(linkwatch_nextevent, jiffies));
236 rtnl_unlock();
237}
238
239
240void linkwatch_fire_event(struct net_device *dev)
241{
242 bool urgent = linkwatch_urgent_event(dev);
243
244 if (!test_and_set_bit(__LINK_STATE_LINKWATCH_PENDING, &dev->state)) {
245 linkwatch_add_event(dev);
246 } else if (!urgent)
247 return;
248
249 linkwatch_schedule_work(urgent);
250}
251EXPORT_SYMBOL(linkwatch_fire_event);
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Linux network device link state notification
4 *
5 * Author:
6 * Stefan Rompf <sux@loplof.de>
7 */
8
9#include <linux/module.h>
10#include <linux/netdevice.h>
11#include <linux/if.h>
12#include <net/sock.h>
13#include <net/pkt_sched.h>
14#include <linux/rtnetlink.h>
15#include <linux/jiffies.h>
16#include <linux/spinlock.h>
17#include <linux/workqueue.h>
18#include <linux/bitops.h>
19#include <linux/types.h>
20
21
22enum lw_bits {
23 LW_URGENT = 0,
24};
25
26static unsigned long linkwatch_flags;
27static unsigned long linkwatch_nextevent;
28
29static void linkwatch_event(struct work_struct *dummy);
30static DECLARE_DELAYED_WORK(linkwatch_work, linkwatch_event);
31
32static LIST_HEAD(lweventlist);
33static DEFINE_SPINLOCK(lweventlist_lock);
34
35static unsigned char default_operstate(const struct net_device *dev)
36{
37 if (!netif_carrier_ok(dev))
38 return (dev->ifindex != dev_get_iflink(dev) ?
39 IF_OPER_LOWERLAYERDOWN : IF_OPER_DOWN);
40
41 if (netif_dormant(dev))
42 return IF_OPER_DORMANT;
43
44 return IF_OPER_UP;
45}
46
47
48static void rfc2863_policy(struct net_device *dev)
49{
50 unsigned char operstate = default_operstate(dev);
51
52 if (operstate == dev->operstate)
53 return;
54
55 write_lock_bh(&dev_base_lock);
56
57 switch(dev->link_mode) {
58 case IF_LINK_MODE_DORMANT:
59 if (operstate == IF_OPER_UP)
60 operstate = IF_OPER_DORMANT;
61 break;
62
63 case IF_LINK_MODE_DEFAULT:
64 default:
65 break;
66 }
67
68 dev->operstate = operstate;
69
70 write_unlock_bh(&dev_base_lock);
71}
72
73
74void linkwatch_init_dev(struct net_device *dev)
75{
76 /* Handle pre-registration link state changes */
77 if (!netif_carrier_ok(dev) || netif_dormant(dev))
78 rfc2863_policy(dev);
79}
80
81
82static bool linkwatch_urgent_event(struct net_device *dev)
83{
84 if (!netif_running(dev))
85 return false;
86
87 if (dev->ifindex != dev_get_iflink(dev))
88 return true;
89
90 if (netif_is_lag_port(dev) || netif_is_lag_master(dev))
91 return true;
92
93 return netif_carrier_ok(dev) && qdisc_tx_changing(dev);
94}
95
96
97static void linkwatch_add_event(struct net_device *dev)
98{
99 unsigned long flags;
100
101 spin_lock_irqsave(&lweventlist_lock, flags);
102 if (list_empty(&dev->link_watch_list)) {
103 list_add_tail(&dev->link_watch_list, &lweventlist);
104 dev_hold(dev);
105 }
106 spin_unlock_irqrestore(&lweventlist_lock, flags);
107}
108
109
110static void linkwatch_schedule_work(int urgent)
111{
112 unsigned long delay = linkwatch_nextevent - jiffies;
113
114 if (test_bit(LW_URGENT, &linkwatch_flags))
115 return;
116
117 /* Minimise down-time: drop delay for up event. */
118 if (urgent) {
119 if (test_and_set_bit(LW_URGENT, &linkwatch_flags))
120 return;
121 delay = 0;
122 }
123
124 /* If we wrap around we'll delay it by at most HZ. */
125 if (delay > HZ)
126 delay = 0;
127
128 /*
129 * If urgent, schedule immediate execution; otherwise, don't
130 * override the existing timer.
131 */
132 if (test_bit(LW_URGENT, &linkwatch_flags))
133 mod_delayed_work(system_wq, &linkwatch_work, 0);
134 else
135 schedule_delayed_work(&linkwatch_work, delay);
136}
137
138
139static void linkwatch_do_dev(struct net_device *dev)
140{
141 /*
142 * Make sure the above read is complete since it can be
143 * rewritten as soon as we clear the bit below.
144 */
145 smp_mb__before_atomic();
146
147 /* We are about to handle this device,
148 * so new events can be accepted
149 */
150 clear_bit(__LINK_STATE_LINKWATCH_PENDING, &dev->state);
151
152 rfc2863_policy(dev);
153 if (dev->flags & IFF_UP && netif_device_present(dev)) {
154 if (netif_carrier_ok(dev))
155 dev_activate(dev);
156 else
157 dev_deactivate(dev);
158
159 netdev_state_change(dev);
160 }
161 dev_put(dev);
162}
163
164static void __linkwatch_run_queue(int urgent_only)
165{
166#define MAX_DO_DEV_PER_LOOP 100
167
168 int do_dev = MAX_DO_DEV_PER_LOOP;
169 struct net_device *dev;
170 LIST_HEAD(wrk);
171
172 /* Give urgent case more budget */
173 if (urgent_only)
174 do_dev += MAX_DO_DEV_PER_LOOP;
175
176 /*
177 * Limit the number of linkwatch events to one
178 * per second so that a runaway driver does not
179 * cause a storm of messages on the netlink
180 * socket. This limit does not apply to up events
181 * while the device qdisc is down.
182 */
183 if (!urgent_only)
184 linkwatch_nextevent = jiffies + HZ;
185 /* Limit wrap-around effect on delay. */
186 else if (time_after(linkwatch_nextevent, jiffies + HZ))
187 linkwatch_nextevent = jiffies;
188
189 clear_bit(LW_URGENT, &linkwatch_flags);
190
191 spin_lock_irq(&lweventlist_lock);
192 list_splice_init(&lweventlist, &wrk);
193
194 while (!list_empty(&wrk) && do_dev > 0) {
195
196 dev = list_first_entry(&wrk, struct net_device, link_watch_list);
197 list_del_init(&dev->link_watch_list);
198
199 if (urgent_only && !linkwatch_urgent_event(dev)) {
200 list_add_tail(&dev->link_watch_list, &lweventlist);
201 continue;
202 }
203 spin_unlock_irq(&lweventlist_lock);
204 linkwatch_do_dev(dev);
205 do_dev--;
206 spin_lock_irq(&lweventlist_lock);
207 }
208
209 /* Add the remaining work back to lweventlist */
210 list_splice_init(&wrk, &lweventlist);
211
212 if (!list_empty(&lweventlist))
213 linkwatch_schedule_work(0);
214 spin_unlock_irq(&lweventlist_lock);
215}
216
217void linkwatch_forget_dev(struct net_device *dev)
218{
219 unsigned long flags;
220 int clean = 0;
221
222 spin_lock_irqsave(&lweventlist_lock, flags);
223 if (!list_empty(&dev->link_watch_list)) {
224 list_del_init(&dev->link_watch_list);
225 clean = 1;
226 }
227 spin_unlock_irqrestore(&lweventlist_lock, flags);
228 if (clean)
229 linkwatch_do_dev(dev);
230}
231
232
233/* Must be called with the rtnl semaphore held */
234void linkwatch_run_queue(void)
235{
236 __linkwatch_run_queue(0);
237}
238
239
240static void linkwatch_event(struct work_struct *dummy)
241{
242 rtnl_lock();
243 __linkwatch_run_queue(time_after(linkwatch_nextevent, jiffies));
244 rtnl_unlock();
245}
246
247
248void linkwatch_fire_event(struct net_device *dev)
249{
250 bool urgent = linkwatch_urgent_event(dev);
251
252 if (!test_and_set_bit(__LINK_STATE_LINKWATCH_PENDING, &dev->state)) {
253 linkwatch_add_event(dev);
254 } else if (!urgent)
255 return;
256
257 linkwatch_schedule_work(urgent);
258}
259EXPORT_SYMBOL(linkwatch_fire_event);