Loading...
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#include "dev.h"
22
23enum lw_bits {
24 LW_URGENT = 0,
25};
26
27static unsigned long linkwatch_flags;
28static unsigned long linkwatch_nextevent;
29
30static void linkwatch_event(struct work_struct *dummy);
31static DECLARE_DELAYED_WORK(linkwatch_work, linkwatch_event);
32
33static LIST_HEAD(lweventlist);
34static DEFINE_SPINLOCK(lweventlist_lock);
35
36static unsigned char default_operstate(const struct net_device *dev)
37{
38 if (netif_testing(dev))
39 return IF_OPER_TESTING;
40
41 /* Some uppers (DSA) have additional sources for being down, so
42 * first check whether lower is indeed the source of its down state.
43 */
44 if (!netif_carrier_ok(dev)) {
45 int iflink = dev_get_iflink(dev);
46 struct net_device *peer;
47
48 if (iflink == dev->ifindex)
49 return IF_OPER_DOWN;
50
51 peer = __dev_get_by_index(dev_net(dev), iflink);
52 if (!peer)
53 return IF_OPER_DOWN;
54
55 return netif_carrier_ok(peer) ? IF_OPER_DOWN :
56 IF_OPER_LOWERLAYERDOWN;
57 }
58
59 if (netif_dormant(dev))
60 return IF_OPER_DORMANT;
61
62 return IF_OPER_UP;
63}
64
65
66static void rfc2863_policy(struct net_device *dev)
67{
68 unsigned char operstate = default_operstate(dev);
69
70 if (operstate == dev->operstate)
71 return;
72
73 write_lock(&dev_base_lock);
74
75 switch(dev->link_mode) {
76 case IF_LINK_MODE_TESTING:
77 if (operstate == IF_OPER_UP)
78 operstate = IF_OPER_TESTING;
79 break;
80
81 case IF_LINK_MODE_DORMANT:
82 if (operstate == IF_OPER_UP)
83 operstate = IF_OPER_DORMANT;
84 break;
85 case IF_LINK_MODE_DEFAULT:
86 default:
87 break;
88 }
89
90 dev->operstate = operstate;
91
92 write_unlock(&dev_base_lock);
93}
94
95
96void linkwatch_init_dev(struct net_device *dev)
97{
98 /* Handle pre-registration link state changes */
99 if (!netif_carrier_ok(dev) || netif_dormant(dev) ||
100 netif_testing(dev))
101 rfc2863_policy(dev);
102}
103
104
105static bool linkwatch_urgent_event(struct net_device *dev)
106{
107 if (!netif_running(dev))
108 return false;
109
110 if (dev->ifindex != dev_get_iflink(dev))
111 return true;
112
113 if (netif_is_lag_port(dev) || netif_is_lag_master(dev))
114 return true;
115
116 return netif_carrier_ok(dev) && qdisc_tx_changing(dev);
117}
118
119
120static void linkwatch_add_event(struct net_device *dev)
121{
122 unsigned long flags;
123
124 spin_lock_irqsave(&lweventlist_lock, flags);
125 if (list_empty(&dev->link_watch_list)) {
126 list_add_tail(&dev->link_watch_list, &lweventlist);
127 netdev_hold(dev, &dev->linkwatch_dev_tracker, GFP_ATOMIC);
128 }
129 spin_unlock_irqrestore(&lweventlist_lock, flags);
130}
131
132
133static void linkwatch_schedule_work(int urgent)
134{
135 unsigned long delay = linkwatch_nextevent - jiffies;
136
137 if (test_bit(LW_URGENT, &linkwatch_flags))
138 return;
139
140 /* Minimise down-time: drop delay for up event. */
141 if (urgent) {
142 if (test_and_set_bit(LW_URGENT, &linkwatch_flags))
143 return;
144 delay = 0;
145 }
146
147 /* If we wrap around we'll delay it by at most HZ. */
148 if (delay > HZ)
149 delay = 0;
150
151 /*
152 * If urgent, schedule immediate execution; otherwise, don't
153 * override the existing timer.
154 */
155 if (test_bit(LW_URGENT, &linkwatch_flags))
156 mod_delayed_work(system_wq, &linkwatch_work, 0);
157 else
158 schedule_delayed_work(&linkwatch_work, delay);
159}
160
161
162static void linkwatch_do_dev(struct net_device *dev)
163{
164 /*
165 * Make sure the above read is complete since it can be
166 * rewritten as soon as we clear the bit below.
167 */
168 smp_mb__before_atomic();
169
170 /* We are about to handle this device,
171 * so new events can be accepted
172 */
173 clear_bit(__LINK_STATE_LINKWATCH_PENDING, &dev->state);
174
175 rfc2863_policy(dev);
176 if (dev->flags & IFF_UP) {
177 if (netif_carrier_ok(dev))
178 dev_activate(dev);
179 else
180 dev_deactivate(dev);
181
182 netdev_state_change(dev);
183 }
184 /* Note: our callers are responsible for calling netdev_tracker_free().
185 * This is the reason we use __dev_put() instead of dev_put().
186 */
187 __dev_put(dev);
188}
189
190static void __linkwatch_run_queue(int urgent_only)
191{
192#define MAX_DO_DEV_PER_LOOP 100
193
194 int do_dev = MAX_DO_DEV_PER_LOOP;
195 /* Use a local list here since we add non-urgent
196 * events back to the global one when called with
197 * urgent_only=1.
198 */
199 LIST_HEAD(wrk);
200
201 /* Give urgent case more budget */
202 if (urgent_only)
203 do_dev += MAX_DO_DEV_PER_LOOP;
204
205 /*
206 * Limit the number of linkwatch events to one
207 * per second so that a runaway driver does not
208 * cause a storm of messages on the netlink
209 * socket. This limit does not apply to up events
210 * while the device qdisc is down.
211 */
212 if (!urgent_only)
213 linkwatch_nextevent = jiffies + HZ;
214 /* Limit wrap-around effect on delay. */
215 else if (time_after(linkwatch_nextevent, jiffies + HZ))
216 linkwatch_nextevent = jiffies;
217
218 clear_bit(LW_URGENT, &linkwatch_flags);
219
220 spin_lock_irq(&lweventlist_lock);
221 list_splice_init(&lweventlist, &wrk);
222
223 while (!list_empty(&wrk) && do_dev > 0) {
224 struct net_device *dev;
225
226 dev = list_first_entry(&wrk, struct net_device, link_watch_list);
227 list_del_init(&dev->link_watch_list);
228
229 if (!netif_device_present(dev) ||
230 (urgent_only && !linkwatch_urgent_event(dev))) {
231 list_add_tail(&dev->link_watch_list, &lweventlist);
232 continue;
233 }
234 /* We must free netdev tracker under
235 * the spinlock protection.
236 */
237 netdev_tracker_free(dev, &dev->linkwatch_dev_tracker);
238 spin_unlock_irq(&lweventlist_lock);
239 linkwatch_do_dev(dev);
240 do_dev--;
241 spin_lock_irq(&lweventlist_lock);
242 }
243
244 /* Add the remaining work back to lweventlist */
245 list_splice_init(&wrk, &lweventlist);
246
247 if (!list_empty(&lweventlist))
248 linkwatch_schedule_work(0);
249 spin_unlock_irq(&lweventlist_lock);
250}
251
252void linkwatch_sync_dev(struct net_device *dev)
253{
254 unsigned long flags;
255 int clean = 0;
256
257 spin_lock_irqsave(&lweventlist_lock, flags);
258 if (!list_empty(&dev->link_watch_list)) {
259 list_del_init(&dev->link_watch_list);
260 clean = 1;
261 /* We must release netdev tracker under
262 * the spinlock protection.
263 */
264 netdev_tracker_free(dev, &dev->linkwatch_dev_tracker);
265 }
266 spin_unlock_irqrestore(&lweventlist_lock, flags);
267 if (clean)
268 linkwatch_do_dev(dev);
269}
270
271
272/* Must be called with the rtnl semaphore held */
273void linkwatch_run_queue(void)
274{
275 __linkwatch_run_queue(0);
276}
277
278
279static void linkwatch_event(struct work_struct *dummy)
280{
281 rtnl_lock();
282 __linkwatch_run_queue(time_after(linkwatch_nextevent, jiffies));
283 rtnl_unlock();
284}
285
286
287void linkwatch_fire_event(struct net_device *dev)
288{
289 bool urgent = linkwatch_urgent_event(dev);
290
291 if (!test_and_set_bit(__LINK_STATE_LINKWATCH_PENDING, &dev->state)) {
292 linkwatch_add_event(dev);
293 } else if (!urgent)
294 return;
295
296 linkwatch_schedule_work(urgent);
297}
298EXPORT_SYMBOL(linkwatch_fire_event);
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);