xref: /openbmc/linux/net/core/link_watch.c (revision 81de3bf3)
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_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 
48 static 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 
74 void 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 
82 static 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 
97 static 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 
110 static 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 
139 static 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 
164 static 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 
217 void 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 */
234 void linkwatch_run_queue(void)
235 {
236 	__linkwatch_run_queue(0);
237 }
238 
239 
240 static 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 
248 void 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 }
259 EXPORT_SYMBOL(linkwatch_fire_event);
260