xref: /openbmc/linux/net/core/link_watch.c (revision 747f7a29)
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 
23 enum lw_bits {
24 	LW_URGENT = 0,
25 };
26 
27 static unsigned long linkwatch_flags;
28 static unsigned long linkwatch_nextevent;
29 
30 static void linkwatch_event(struct work_struct *dummy);
31 static DECLARE_DELAYED_WORK(linkwatch_work, linkwatch_event);
32 
33 static LIST_HEAD(lweventlist);
34 static DEFINE_SPINLOCK(lweventlist_lock);
35 
36 static unsigned char default_operstate(const struct net_device *dev)
37 {
38 	if (netif_testing(dev))
39 		return IF_OPER_TESTING;
40 
41 	if (!netif_carrier_ok(dev))
42 		return (dev->ifindex != dev_get_iflink(dev) ?
43 			IF_OPER_LOWERLAYERDOWN : IF_OPER_DOWN);
44 
45 	if (netif_dormant(dev))
46 		return IF_OPER_DORMANT;
47 
48 	return IF_OPER_UP;
49 }
50 
51 
52 static void rfc2863_policy(struct net_device *dev)
53 {
54 	unsigned char operstate = default_operstate(dev);
55 
56 	if (operstate == dev->operstate)
57 		return;
58 
59 	write_lock(&dev_base_lock);
60 
61 	switch(dev->link_mode) {
62 	case IF_LINK_MODE_TESTING:
63 		if (operstate == IF_OPER_UP)
64 			operstate = IF_OPER_TESTING;
65 		break;
66 
67 	case IF_LINK_MODE_DORMANT:
68 		if (operstate == IF_OPER_UP)
69 			operstate = IF_OPER_DORMANT;
70 		break;
71 	case IF_LINK_MODE_DEFAULT:
72 	default:
73 		break;
74 	}
75 
76 	dev->operstate = operstate;
77 
78 	write_unlock(&dev_base_lock);
79 }
80 
81 
82 void linkwatch_init_dev(struct net_device *dev)
83 {
84 	/* Handle pre-registration link state changes */
85 	if (!netif_carrier_ok(dev) || netif_dormant(dev) ||
86 	    netif_testing(dev))
87 		rfc2863_policy(dev);
88 }
89 
90 
91 static bool linkwatch_urgent_event(struct net_device *dev)
92 {
93 	if (!netif_running(dev))
94 		return false;
95 
96 	if (dev->ifindex != dev_get_iflink(dev))
97 		return true;
98 
99 	if (netif_is_lag_port(dev) || netif_is_lag_master(dev))
100 		return true;
101 
102 	return netif_carrier_ok(dev) &&	qdisc_tx_changing(dev);
103 }
104 
105 
106 static void linkwatch_add_event(struct net_device *dev)
107 {
108 	unsigned long flags;
109 
110 	spin_lock_irqsave(&lweventlist_lock, flags);
111 	if (list_empty(&dev->link_watch_list)) {
112 		list_add_tail(&dev->link_watch_list, &lweventlist);
113 		netdev_hold(dev, &dev->linkwatch_dev_tracker, GFP_ATOMIC);
114 	}
115 	spin_unlock_irqrestore(&lweventlist_lock, flags);
116 }
117 
118 
119 static void linkwatch_schedule_work(int urgent)
120 {
121 	unsigned long delay = linkwatch_nextevent - jiffies;
122 
123 	if (test_bit(LW_URGENT, &linkwatch_flags))
124 		return;
125 
126 	/* Minimise down-time: drop delay for up event. */
127 	if (urgent) {
128 		if (test_and_set_bit(LW_URGENT, &linkwatch_flags))
129 			return;
130 		delay = 0;
131 	}
132 
133 	/* If we wrap around we'll delay it by at most HZ. */
134 	if (delay > HZ)
135 		delay = 0;
136 
137 	/*
138 	 * If urgent, schedule immediate execution; otherwise, don't
139 	 * override the existing timer.
140 	 */
141 	if (test_bit(LW_URGENT, &linkwatch_flags))
142 		mod_delayed_work(system_wq, &linkwatch_work, 0);
143 	else
144 		schedule_delayed_work(&linkwatch_work, delay);
145 }
146 
147 
148 static void linkwatch_do_dev(struct net_device *dev)
149 {
150 	/*
151 	 * Make sure the above read is complete since it can be
152 	 * rewritten as soon as we clear the bit below.
153 	 */
154 	smp_mb__before_atomic();
155 
156 	/* We are about to handle this device,
157 	 * so new events can be accepted
158 	 */
159 	clear_bit(__LINK_STATE_LINKWATCH_PENDING, &dev->state);
160 
161 	rfc2863_policy(dev);
162 	if (dev->flags & IFF_UP) {
163 		if (netif_carrier_ok(dev))
164 			dev_activate(dev);
165 		else
166 			dev_deactivate(dev);
167 
168 		netdev_state_change(dev);
169 	}
170 	/* Note: our callers are responsible for calling netdev_tracker_free().
171 	 * This is the reason we use __dev_put() instead of dev_put().
172 	 */
173 	__dev_put(dev);
174 }
175 
176 static void __linkwatch_run_queue(int urgent_only)
177 {
178 #define MAX_DO_DEV_PER_LOOP	100
179 
180 	int do_dev = MAX_DO_DEV_PER_LOOP;
181 	struct net_device *dev;
182 	LIST_HEAD(wrk);
183 
184 	/* Give urgent case more budget */
185 	if (urgent_only)
186 		do_dev += MAX_DO_DEV_PER_LOOP;
187 
188 	/*
189 	 * Limit the number of linkwatch events to one
190 	 * per second so that a runaway driver does not
191 	 * cause a storm of messages on the netlink
192 	 * socket.  This limit does not apply to up events
193 	 * while the device qdisc is down.
194 	 */
195 	if (!urgent_only)
196 		linkwatch_nextevent = jiffies + HZ;
197 	/* Limit wrap-around effect on delay. */
198 	else if (time_after(linkwatch_nextevent, jiffies + HZ))
199 		linkwatch_nextevent = jiffies;
200 
201 	clear_bit(LW_URGENT, &linkwatch_flags);
202 
203 	spin_lock_irq(&lweventlist_lock);
204 	list_splice_init(&lweventlist, &wrk);
205 
206 	while (!list_empty(&wrk) && do_dev > 0) {
207 
208 		dev = list_first_entry(&wrk, struct net_device, link_watch_list);
209 		list_del_init(&dev->link_watch_list);
210 
211 		if (!netif_device_present(dev) ||
212 		    (urgent_only && !linkwatch_urgent_event(dev))) {
213 			list_add_tail(&dev->link_watch_list, &lweventlist);
214 			continue;
215 		}
216 		/* We must free netdev tracker under
217 		 * the spinlock protection.
218 		 */
219 		netdev_tracker_free(dev, &dev->linkwatch_dev_tracker);
220 		spin_unlock_irq(&lweventlist_lock);
221 		linkwatch_do_dev(dev);
222 		do_dev--;
223 		spin_lock_irq(&lweventlist_lock);
224 	}
225 
226 	/* Add the remaining work back to lweventlist */
227 	list_splice_init(&wrk, &lweventlist);
228 
229 	if (!list_empty(&lweventlist))
230 		linkwatch_schedule_work(0);
231 	spin_unlock_irq(&lweventlist_lock);
232 }
233 
234 void linkwatch_forget_dev(struct net_device *dev)
235 {
236 	unsigned long flags;
237 	int clean = 0;
238 
239 	spin_lock_irqsave(&lweventlist_lock, flags);
240 	if (!list_empty(&dev->link_watch_list)) {
241 		list_del_init(&dev->link_watch_list);
242 		clean = 1;
243 		/* We must release netdev tracker under
244 		 * the spinlock protection.
245 		 */
246 		netdev_tracker_free(dev, &dev->linkwatch_dev_tracker);
247 	}
248 	spin_unlock_irqrestore(&lweventlist_lock, flags);
249 	if (clean)
250 		linkwatch_do_dev(dev);
251 }
252 
253 
254 /* Must be called with the rtnl semaphore held */
255 void linkwatch_run_queue(void)
256 {
257 	__linkwatch_run_queue(0);
258 }
259 
260 
261 static void linkwatch_event(struct work_struct *dummy)
262 {
263 	rtnl_lock();
264 	__linkwatch_run_queue(time_after(linkwatch_nextevent, jiffies));
265 	rtnl_unlock();
266 }
267 
268 
269 void linkwatch_fire_event(struct net_device *dev)
270 {
271 	bool urgent = linkwatch_urgent_event(dev);
272 
273 	if (!test_and_set_bit(__LINK_STATE_LINKWATCH_PENDING, &dev->state)) {
274 		linkwatch_add_event(dev);
275 	} else if (!urgent)
276 		return;
277 
278 	linkwatch_schedule_work(urgent);
279 }
280 EXPORT_SYMBOL(linkwatch_fire_event);
281