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