xref: /openbmc/linux/net/core/link_watch.c (revision fbb6b31a)
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 
22 enum lw_bits {
23 	LW_URGENT = 0,
24 };
25 
26 static unsigned long linkwatch_flags;
27 static unsigned long linkwatch_nextevent;
28 
29 static void linkwatch_event(struct work_struct *dummy);
30 static DECLARE_DELAYED_WORK(linkwatch_work, linkwatch_event);
31 
32 static LIST_HEAD(lweventlist);
33 static DEFINE_SPINLOCK(lweventlist_lock);
34 
35 static 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 
51 static 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(&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(&dev_base_lock);
78 }
79 
80 
81 void 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 
90 static 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 
105 static 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_track(dev, &dev->linkwatch_dev_tracker, GFP_ATOMIC);
113 	}
114 	spin_unlock_irqrestore(&lweventlist_lock, flags);
115 }
116 
117 
118 static 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 
147 static 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 	/* Note: our callers are responsible for calling netdev_tracker_free().
170 	 * This is the reason we use __dev_put() instead of dev_put().
171 	 */
172 	__dev_put(dev);
173 }
174 
175 static void __linkwatch_run_queue(int urgent_only)
176 {
177 #define MAX_DO_DEV_PER_LOOP	100
178 
179 	int do_dev = MAX_DO_DEV_PER_LOOP;
180 	struct net_device *dev;
181 	LIST_HEAD(wrk);
182 
183 	/* Give urgent case more budget */
184 	if (urgent_only)
185 		do_dev += MAX_DO_DEV_PER_LOOP;
186 
187 	/*
188 	 * Limit the number of linkwatch events to one
189 	 * per second so that a runaway driver does not
190 	 * cause a storm of messages on the netlink
191 	 * socket.  This limit does not apply to up events
192 	 * while the device qdisc is down.
193 	 */
194 	if (!urgent_only)
195 		linkwatch_nextevent = jiffies + HZ;
196 	/* Limit wrap-around effect on delay. */
197 	else if (time_after(linkwatch_nextevent, jiffies + HZ))
198 		linkwatch_nextevent = jiffies;
199 
200 	clear_bit(LW_URGENT, &linkwatch_flags);
201 
202 	spin_lock_irq(&lweventlist_lock);
203 	list_splice_init(&lweventlist, &wrk);
204 
205 	while (!list_empty(&wrk) && do_dev > 0) {
206 
207 		dev = list_first_entry(&wrk, struct net_device, link_watch_list);
208 		list_del_init(&dev->link_watch_list);
209 
210 		if (!netif_device_present(dev) ||
211 		    (urgent_only && !linkwatch_urgent_event(dev))) {
212 			list_add_tail(&dev->link_watch_list, &lweventlist);
213 			continue;
214 		}
215 		/* We must free netdev tracker under
216 		 * the spinlock protection.
217 		 */
218 		netdev_tracker_free(dev, &dev->linkwatch_dev_tracker);
219 		spin_unlock_irq(&lweventlist_lock);
220 		linkwatch_do_dev(dev);
221 		do_dev--;
222 		spin_lock_irq(&lweventlist_lock);
223 	}
224 
225 	/* Add the remaining work back to lweventlist */
226 	list_splice_init(&wrk, &lweventlist);
227 
228 	if (!list_empty(&lweventlist))
229 		linkwatch_schedule_work(0);
230 	spin_unlock_irq(&lweventlist_lock);
231 }
232 
233 void linkwatch_forget_dev(struct net_device *dev)
234 {
235 	unsigned long flags;
236 	int clean = 0;
237 
238 	spin_lock_irqsave(&lweventlist_lock, flags);
239 	if (!list_empty(&dev->link_watch_list)) {
240 		list_del_init(&dev->link_watch_list);
241 		clean = 1;
242 		/* We must release netdev tracker under
243 		 * the spinlock protection.
244 		 */
245 		netdev_tracker_free(dev, &dev->linkwatch_dev_tracker);
246 	}
247 	spin_unlock_irqrestore(&lweventlist_lock, flags);
248 	if (clean)
249 		linkwatch_do_dev(dev);
250 }
251 
252 
253 /* Must be called with the rtnl semaphore held */
254 void linkwatch_run_queue(void)
255 {
256 	__linkwatch_run_queue(0);
257 }
258 
259 
260 static void linkwatch_event(struct work_struct *dummy)
261 {
262 	rtnl_lock();
263 	__linkwatch_run_queue(time_after(linkwatch_nextevent, jiffies));
264 	rtnl_unlock();
265 }
266 
267 
268 void linkwatch_fire_event(struct net_device *dev)
269 {
270 	bool urgent = linkwatch_urgent_event(dev);
271 
272 	if (!test_and_set_bit(__LINK_STATE_LINKWATCH_PENDING, &dev->state)) {
273 		linkwatch_add_event(dev);
274 	} else if (!urgent)
275 		return;
276 
277 	linkwatch_schedule_work(urgent);
278 }
279 EXPORT_SYMBOL(linkwatch_fire_event);
280