xref: /linux/net/core/netpoll.c (revision c6df6213a95fa9674cc48d77042141942dd0809b)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Common framework for low-level network console, dump, and debugger code
4  *
5  * Sep 8 2003  Matt Mackall <mpm@selenic.com>
6  *
7  * based on the netconsole code from:
8  *
9  * Copyright (C) 2001  Ingo Molnar <mingo@redhat.com>
10  * Copyright (C) 2002  Red Hat, Inc.
11  */
12 
13 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14 
15 #include <linux/moduleparam.h>
16 #include <linux/kernel.h>
17 #include <linux/netdevice.h>
18 #include <linux/etherdevice.h>
19 #include <linux/string.h>
20 #include <linux/if_arp.h>
21 #include <linux/inetdevice.h>
22 #include <linux/inet.h>
23 #include <linux/interrupt.h>
24 #include <linux/netpoll.h>
25 #include <linux/sched.h>
26 #include <linux/delay.h>
27 #include <linux/rcupdate.h>
28 #include <linux/workqueue.h>
29 #include <linux/slab.h>
30 #include <linux/export.h>
31 #include <linux/if_vlan.h>
32 #include <net/tcp.h>
33 #include <net/udp.h>
34 #include <net/addrconf.h>
35 #include <net/ndisc.h>
36 #include <net/ip6_checksum.h>
37 #include <asm/unaligned.h>
38 #include <trace/events/napi.h>
39 #include <linux/kconfig.h>
40 
41 /*
42  * We maintain a small pool of fully-sized skbs, to make sure the
43  * message gets out even in extreme OOM situations.
44  */
45 
46 #define MAX_UDP_CHUNK 1460
47 #define MAX_SKBS 32
48 
49 static struct sk_buff_head skb_pool;
50 
51 DEFINE_STATIC_SRCU(netpoll_srcu);
52 
53 #define USEC_PER_POLL	50
54 
55 #define MAX_SKB_SIZE							\
56 	(sizeof(struct ethhdr) +					\
57 	 sizeof(struct iphdr) +						\
58 	 sizeof(struct udphdr) +					\
59 	 MAX_UDP_CHUNK)
60 
61 static void zap_completion_queue(void);
62 
63 static unsigned int carrier_timeout = 4;
64 module_param(carrier_timeout, uint, 0644);
65 
66 #define np_info(np, fmt, ...)				\
67 	pr_info("%s: " fmt, np->name, ##__VA_ARGS__)
68 #define np_err(np, fmt, ...)				\
69 	pr_err("%s: " fmt, np->name, ##__VA_ARGS__)
70 #define np_notice(np, fmt, ...)				\
71 	pr_notice("%s: " fmt, np->name, ##__VA_ARGS__)
72 
73 static netdev_tx_t netpoll_start_xmit(struct sk_buff *skb,
74 				      struct net_device *dev,
75 				      struct netdev_queue *txq)
76 {
77 	netdev_tx_t status = NETDEV_TX_OK;
78 	netdev_features_t features;
79 
80 	features = netif_skb_features(skb);
81 
82 	if (skb_vlan_tag_present(skb) &&
83 	    !vlan_hw_offload_capable(features, skb->vlan_proto)) {
84 		skb = __vlan_hwaccel_push_inside(skb);
85 		if (unlikely(!skb)) {
86 			/* This is actually a packet drop, but we
87 			 * don't want the code that calls this
88 			 * function to try and operate on a NULL skb.
89 			 */
90 			goto out;
91 		}
92 	}
93 
94 	status = netdev_start_xmit(skb, dev, txq, false);
95 
96 out:
97 	return status;
98 }
99 
100 static void queue_process(struct work_struct *work)
101 {
102 	struct netpoll_info *npinfo =
103 		container_of(work, struct netpoll_info, tx_work.work);
104 	struct sk_buff *skb;
105 	unsigned long flags;
106 
107 	while ((skb = skb_dequeue(&npinfo->txq))) {
108 		struct net_device *dev = skb->dev;
109 		struct netdev_queue *txq;
110 		unsigned int q_index;
111 
112 		if (!netif_device_present(dev) || !netif_running(dev)) {
113 			kfree_skb(skb);
114 			continue;
115 		}
116 
117 		local_irq_save(flags);
118 		/* check if skb->queue_mapping is still valid */
119 		q_index = skb_get_queue_mapping(skb);
120 		if (unlikely(q_index >= dev->real_num_tx_queues)) {
121 			q_index = q_index % dev->real_num_tx_queues;
122 			skb_set_queue_mapping(skb, q_index);
123 		}
124 		txq = netdev_get_tx_queue(dev, q_index);
125 		HARD_TX_LOCK(dev, txq, smp_processor_id());
126 		if (netif_xmit_frozen_or_stopped(txq) ||
127 		    !dev_xmit_complete(netpoll_start_xmit(skb, dev, txq))) {
128 			skb_queue_head(&npinfo->txq, skb);
129 			HARD_TX_UNLOCK(dev, txq);
130 			local_irq_restore(flags);
131 
132 			schedule_delayed_work(&npinfo->tx_work, HZ/10);
133 			return;
134 		}
135 		HARD_TX_UNLOCK(dev, txq);
136 		local_irq_restore(flags);
137 	}
138 }
139 
140 static int netif_local_xmit_active(struct net_device *dev)
141 {
142 	int i;
143 
144 	for (i = 0; i < dev->num_tx_queues; i++) {
145 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
146 
147 		if (READ_ONCE(txq->xmit_lock_owner) == smp_processor_id())
148 			return 1;
149 	}
150 
151 	return 0;
152 }
153 
154 static void poll_one_napi(struct napi_struct *napi)
155 {
156 	int work;
157 
158 	/* If we set this bit but see that it has already been set,
159 	 * that indicates that napi has been disabled and we need
160 	 * to abort this operation
161 	 */
162 	if (test_and_set_bit(NAPI_STATE_NPSVC, &napi->state))
163 		return;
164 
165 	/* We explicilty pass the polling call a budget of 0 to
166 	 * indicate that we are clearing the Tx path only.
167 	 */
168 	work = napi->poll(napi, 0);
169 	WARN_ONCE(work, "%pS exceeded budget in poll\n", napi->poll);
170 	trace_napi_poll(napi, work, 0);
171 
172 	clear_bit(NAPI_STATE_NPSVC, &napi->state);
173 }
174 
175 static void poll_napi(struct net_device *dev)
176 {
177 	struct napi_struct *napi;
178 	int cpu = smp_processor_id();
179 
180 	list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) {
181 		if (cmpxchg(&napi->poll_owner, -1, cpu) == -1) {
182 			poll_one_napi(napi);
183 			smp_store_release(&napi->poll_owner, -1);
184 		}
185 	}
186 }
187 
188 void netpoll_poll_dev(struct net_device *dev)
189 {
190 	struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
191 	const struct net_device_ops *ops;
192 
193 	/* Don't do any rx activity if the dev_lock mutex is held
194 	 * the dev_open/close paths use this to block netpoll activity
195 	 * while changing device state
196 	 */
197 	if (!ni || down_trylock(&ni->dev_lock))
198 		return;
199 
200 	/* Some drivers will take the same locks in poll and xmit,
201 	 * we can't poll if local CPU is already in xmit.
202 	 */
203 	if (!netif_running(dev) || netif_local_xmit_active(dev)) {
204 		up(&ni->dev_lock);
205 		return;
206 	}
207 
208 	ops = dev->netdev_ops;
209 	if (ops->ndo_poll_controller)
210 		ops->ndo_poll_controller(dev);
211 
212 	poll_napi(dev);
213 
214 	up(&ni->dev_lock);
215 
216 	zap_completion_queue();
217 }
218 EXPORT_SYMBOL(netpoll_poll_dev);
219 
220 void netpoll_poll_disable(struct net_device *dev)
221 {
222 	struct netpoll_info *ni;
223 	int idx;
224 	might_sleep();
225 	idx = srcu_read_lock(&netpoll_srcu);
226 	ni = srcu_dereference(dev->npinfo, &netpoll_srcu);
227 	if (ni)
228 		down(&ni->dev_lock);
229 	srcu_read_unlock(&netpoll_srcu, idx);
230 }
231 
232 void netpoll_poll_enable(struct net_device *dev)
233 {
234 	struct netpoll_info *ni;
235 	rcu_read_lock();
236 	ni = rcu_dereference(dev->npinfo);
237 	if (ni)
238 		up(&ni->dev_lock);
239 	rcu_read_unlock();
240 }
241 
242 static void refill_skbs(void)
243 {
244 	struct sk_buff *skb;
245 	unsigned long flags;
246 
247 	spin_lock_irqsave(&skb_pool.lock, flags);
248 	while (skb_pool.qlen < MAX_SKBS) {
249 		skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
250 		if (!skb)
251 			break;
252 
253 		__skb_queue_tail(&skb_pool, skb);
254 	}
255 	spin_unlock_irqrestore(&skb_pool.lock, flags);
256 }
257 
258 static void zap_completion_queue(void)
259 {
260 	unsigned long flags;
261 	struct softnet_data *sd = &get_cpu_var(softnet_data);
262 
263 	if (sd->completion_queue) {
264 		struct sk_buff *clist;
265 
266 		local_irq_save(flags);
267 		clist = sd->completion_queue;
268 		sd->completion_queue = NULL;
269 		local_irq_restore(flags);
270 
271 		while (clist != NULL) {
272 			struct sk_buff *skb = clist;
273 			clist = clist->next;
274 			if (!skb_irq_freeable(skb)) {
275 				refcount_set(&skb->users, 1);
276 				dev_kfree_skb_any(skb); /* put this one back */
277 			} else {
278 				__kfree_skb(skb);
279 			}
280 		}
281 	}
282 
283 	put_cpu_var(softnet_data);
284 }
285 
286 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
287 {
288 	int count = 0;
289 	struct sk_buff *skb;
290 
291 	zap_completion_queue();
292 	refill_skbs();
293 repeat:
294 
295 	skb = alloc_skb(len, GFP_ATOMIC);
296 	if (!skb)
297 		skb = skb_dequeue(&skb_pool);
298 
299 	if (!skb) {
300 		if (++count < 10) {
301 			netpoll_poll_dev(np->dev);
302 			goto repeat;
303 		}
304 		return NULL;
305 	}
306 
307 	refcount_set(&skb->users, 1);
308 	skb_reserve(skb, reserve);
309 	return skb;
310 }
311 
312 static int netpoll_owner_active(struct net_device *dev)
313 {
314 	struct napi_struct *napi;
315 
316 	list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) {
317 		if (READ_ONCE(napi->poll_owner) == smp_processor_id())
318 			return 1;
319 	}
320 	return 0;
321 }
322 
323 /* call with IRQ disabled */
324 static netdev_tx_t __netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
325 {
326 	netdev_tx_t status = NETDEV_TX_BUSY;
327 	struct net_device *dev;
328 	unsigned long tries;
329 	/* It is up to the caller to keep npinfo alive. */
330 	struct netpoll_info *npinfo;
331 
332 	lockdep_assert_irqs_disabled();
333 
334 	dev = np->dev;
335 	npinfo = rcu_dereference_bh(dev->npinfo);
336 
337 	if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
338 		dev_kfree_skb_irq(skb);
339 		return NET_XMIT_DROP;
340 	}
341 
342 	/* don't get messages out of order, and no recursion */
343 	if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
344 		struct netdev_queue *txq;
345 
346 		txq = netdev_core_pick_tx(dev, skb, NULL);
347 
348 		/* try until next clock tick */
349 		for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
350 		     tries > 0; --tries) {
351 			if (HARD_TX_TRYLOCK(dev, txq)) {
352 				if (!netif_xmit_stopped(txq))
353 					status = netpoll_start_xmit(skb, dev, txq);
354 
355 				HARD_TX_UNLOCK(dev, txq);
356 
357 				if (dev_xmit_complete(status))
358 					break;
359 
360 			}
361 
362 			/* tickle device maybe there is some cleanup */
363 			netpoll_poll_dev(np->dev);
364 
365 			udelay(USEC_PER_POLL);
366 		}
367 
368 		WARN_ONCE(!irqs_disabled(),
369 			"netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pS)\n",
370 			dev->name, dev->netdev_ops->ndo_start_xmit);
371 
372 	}
373 
374 	if (!dev_xmit_complete(status)) {
375 		skb_queue_tail(&npinfo->txq, skb);
376 		schedule_delayed_work(&npinfo->tx_work,0);
377 	}
378 	return NETDEV_TX_OK;
379 }
380 
381 netdev_tx_t netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
382 {
383 	unsigned long flags;
384 	netdev_tx_t ret;
385 
386 	if (unlikely(!np)) {
387 		dev_kfree_skb_irq(skb);
388 		ret = NET_XMIT_DROP;
389 	} else {
390 		local_irq_save(flags);
391 		ret = __netpoll_send_skb(np, skb);
392 		local_irq_restore(flags);
393 	}
394 	return ret;
395 }
396 EXPORT_SYMBOL(netpoll_send_skb);
397 
398 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
399 {
400 	int total_len, ip_len, udp_len;
401 	struct sk_buff *skb;
402 	struct udphdr *udph;
403 	struct iphdr *iph;
404 	struct ethhdr *eth;
405 	static atomic_t ip_ident;
406 	struct ipv6hdr *ip6h;
407 
408 	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
409 		WARN_ON_ONCE(!irqs_disabled());
410 
411 	udp_len = len + sizeof(*udph);
412 	if (np->ipv6)
413 		ip_len = udp_len + sizeof(*ip6h);
414 	else
415 		ip_len = udp_len + sizeof(*iph);
416 
417 	total_len = ip_len + LL_RESERVED_SPACE(np->dev);
418 
419 	skb = find_skb(np, total_len + np->dev->needed_tailroom,
420 		       total_len - len);
421 	if (!skb)
422 		return;
423 
424 	skb_copy_to_linear_data(skb, msg, len);
425 	skb_put(skb, len);
426 
427 	skb_push(skb, sizeof(*udph));
428 	skb_reset_transport_header(skb);
429 	udph = udp_hdr(skb);
430 	udph->source = htons(np->local_port);
431 	udph->dest = htons(np->remote_port);
432 	udph->len = htons(udp_len);
433 
434 	if (np->ipv6) {
435 		udph->check = 0;
436 		udph->check = csum_ipv6_magic(&np->local_ip.in6,
437 					      &np->remote_ip.in6,
438 					      udp_len, IPPROTO_UDP,
439 					      csum_partial(udph, udp_len, 0));
440 		if (udph->check == 0)
441 			udph->check = CSUM_MANGLED_0;
442 
443 		skb_push(skb, sizeof(*ip6h));
444 		skb_reset_network_header(skb);
445 		ip6h = ipv6_hdr(skb);
446 
447 		/* ip6h->version = 6; ip6h->priority = 0; */
448 		*(unsigned char *)ip6h = 0x60;
449 		ip6h->flow_lbl[0] = 0;
450 		ip6h->flow_lbl[1] = 0;
451 		ip6h->flow_lbl[2] = 0;
452 
453 		ip6h->payload_len = htons(sizeof(struct udphdr) + len);
454 		ip6h->nexthdr = IPPROTO_UDP;
455 		ip6h->hop_limit = 32;
456 		ip6h->saddr = np->local_ip.in6;
457 		ip6h->daddr = np->remote_ip.in6;
458 
459 		eth = skb_push(skb, ETH_HLEN);
460 		skb_reset_mac_header(skb);
461 		skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
462 	} else {
463 		udph->check = 0;
464 		udph->check = csum_tcpudp_magic(np->local_ip.ip,
465 						np->remote_ip.ip,
466 						udp_len, IPPROTO_UDP,
467 						csum_partial(udph, udp_len, 0));
468 		if (udph->check == 0)
469 			udph->check = CSUM_MANGLED_0;
470 
471 		skb_push(skb, sizeof(*iph));
472 		skb_reset_network_header(skb);
473 		iph = ip_hdr(skb);
474 
475 		/* iph->version = 4; iph->ihl = 5; */
476 		*(unsigned char *)iph = 0x45;
477 		iph->tos      = 0;
478 		put_unaligned(htons(ip_len), &(iph->tot_len));
479 		iph->id       = htons(atomic_inc_return(&ip_ident));
480 		iph->frag_off = 0;
481 		iph->ttl      = 64;
482 		iph->protocol = IPPROTO_UDP;
483 		iph->check    = 0;
484 		put_unaligned(np->local_ip.ip, &(iph->saddr));
485 		put_unaligned(np->remote_ip.ip, &(iph->daddr));
486 		iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
487 
488 		eth = skb_push(skb, ETH_HLEN);
489 		skb_reset_mac_header(skb);
490 		skb->protocol = eth->h_proto = htons(ETH_P_IP);
491 	}
492 
493 	ether_addr_copy(eth->h_source, np->dev->dev_addr);
494 	ether_addr_copy(eth->h_dest, np->remote_mac);
495 
496 	skb->dev = np->dev;
497 
498 	netpoll_send_skb(np, skb);
499 }
500 EXPORT_SYMBOL(netpoll_send_udp);
501 
502 void netpoll_print_options(struct netpoll *np)
503 {
504 	np_info(np, "local port %d\n", np->local_port);
505 	if (np->ipv6)
506 		np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
507 	else
508 		np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
509 	np_info(np, "interface '%s'\n", np->dev_name);
510 	np_info(np, "remote port %d\n", np->remote_port);
511 	if (np->ipv6)
512 		np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
513 	else
514 		np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
515 	np_info(np, "remote ethernet address %pM\n", np->remote_mac);
516 }
517 EXPORT_SYMBOL(netpoll_print_options);
518 
519 static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
520 {
521 	const char *end;
522 
523 	if (!strchr(str, ':') &&
524 	    in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
525 		if (!*end)
526 			return 0;
527 	}
528 	if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
529 #if IS_ENABLED(CONFIG_IPV6)
530 		if (!*end)
531 			return 1;
532 #else
533 		return -1;
534 #endif
535 	}
536 	return -1;
537 }
538 
539 int netpoll_parse_options(struct netpoll *np, char *opt)
540 {
541 	char *cur=opt, *delim;
542 	int ipv6;
543 	bool ipversion_set = false;
544 
545 	if (*cur != '@') {
546 		if ((delim = strchr(cur, '@')) == NULL)
547 			goto parse_failed;
548 		*delim = 0;
549 		if (kstrtou16(cur, 10, &np->local_port))
550 			goto parse_failed;
551 		cur = delim;
552 	}
553 	cur++;
554 
555 	if (*cur != '/') {
556 		ipversion_set = true;
557 		if ((delim = strchr(cur, '/')) == NULL)
558 			goto parse_failed;
559 		*delim = 0;
560 		ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
561 		if (ipv6 < 0)
562 			goto parse_failed;
563 		else
564 			np->ipv6 = (bool)ipv6;
565 		cur = delim;
566 	}
567 	cur++;
568 
569 	if (*cur != ',') {
570 		/* parse out dev name */
571 		if ((delim = strchr(cur, ',')) == NULL)
572 			goto parse_failed;
573 		*delim = 0;
574 		strscpy(np->dev_name, cur, sizeof(np->dev_name));
575 		cur = delim;
576 	}
577 	cur++;
578 
579 	if (*cur != '@') {
580 		/* dst port */
581 		if ((delim = strchr(cur, '@')) == NULL)
582 			goto parse_failed;
583 		*delim = 0;
584 		if (*cur == ' ' || *cur == '\t')
585 			np_info(np, "warning: whitespace is not allowed\n");
586 		if (kstrtou16(cur, 10, &np->remote_port))
587 			goto parse_failed;
588 		cur = delim;
589 	}
590 	cur++;
591 
592 	/* dst ip */
593 	if ((delim = strchr(cur, '/')) == NULL)
594 		goto parse_failed;
595 	*delim = 0;
596 	ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
597 	if (ipv6 < 0)
598 		goto parse_failed;
599 	else if (ipversion_set && np->ipv6 != (bool)ipv6)
600 		goto parse_failed;
601 	else
602 		np->ipv6 = (bool)ipv6;
603 	cur = delim + 1;
604 
605 	if (*cur != 0) {
606 		/* MAC address */
607 		if (!mac_pton(cur, np->remote_mac))
608 			goto parse_failed;
609 	}
610 
611 	netpoll_print_options(np);
612 
613 	return 0;
614 
615  parse_failed:
616 	np_info(np, "couldn't parse config at '%s'!\n", cur);
617 	return -1;
618 }
619 EXPORT_SYMBOL(netpoll_parse_options);
620 
621 int __netpoll_setup(struct netpoll *np, struct net_device *ndev)
622 {
623 	struct netpoll_info *npinfo;
624 	const struct net_device_ops *ops;
625 	int err;
626 
627 	np->dev = ndev;
628 	strscpy(np->dev_name, ndev->name, IFNAMSIZ);
629 
630 	if (ndev->priv_flags & IFF_DISABLE_NETPOLL) {
631 		np_err(np, "%s doesn't support polling, aborting\n",
632 		       np->dev_name);
633 		err = -ENOTSUPP;
634 		goto out;
635 	}
636 
637 	if (!ndev->npinfo) {
638 		npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
639 		if (!npinfo) {
640 			err = -ENOMEM;
641 			goto out;
642 		}
643 
644 		sema_init(&npinfo->dev_lock, 1);
645 		skb_queue_head_init(&npinfo->txq);
646 		INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
647 
648 		refcount_set(&npinfo->refcnt, 1);
649 
650 		ops = np->dev->netdev_ops;
651 		if (ops->ndo_netpoll_setup) {
652 			err = ops->ndo_netpoll_setup(ndev, npinfo);
653 			if (err)
654 				goto free_npinfo;
655 		}
656 	} else {
657 		npinfo = rtnl_dereference(ndev->npinfo);
658 		refcount_inc(&npinfo->refcnt);
659 	}
660 
661 	npinfo->netpoll = np;
662 
663 	/* last thing to do is link it to the net device structure */
664 	rcu_assign_pointer(ndev->npinfo, npinfo);
665 
666 	return 0;
667 
668 free_npinfo:
669 	kfree(npinfo);
670 out:
671 	return err;
672 }
673 EXPORT_SYMBOL_GPL(__netpoll_setup);
674 
675 int netpoll_setup(struct netpoll *np)
676 {
677 	struct net_device *ndev = NULL;
678 	struct in_device *in_dev;
679 	int err;
680 
681 	rtnl_lock();
682 	if (np->dev_name[0]) {
683 		struct net *net = current->nsproxy->net_ns;
684 		ndev = __dev_get_by_name(net, np->dev_name);
685 	}
686 	if (!ndev) {
687 		np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
688 		err = -ENODEV;
689 		goto unlock;
690 	}
691 	netdev_hold(ndev, &np->dev_tracker, GFP_KERNEL);
692 
693 	if (netdev_master_upper_dev_get(ndev)) {
694 		np_err(np, "%s is a slave device, aborting\n", np->dev_name);
695 		err = -EBUSY;
696 		goto put;
697 	}
698 
699 	if (!netif_running(ndev)) {
700 		unsigned long atmost;
701 
702 		np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
703 
704 		err = dev_open(ndev, NULL);
705 
706 		if (err) {
707 			np_err(np, "failed to open %s\n", ndev->name);
708 			goto put;
709 		}
710 
711 		rtnl_unlock();
712 		atmost = jiffies + carrier_timeout * HZ;
713 		while (!netif_carrier_ok(ndev)) {
714 			if (time_after(jiffies, atmost)) {
715 				np_notice(np, "timeout waiting for carrier\n");
716 				break;
717 			}
718 			msleep(1);
719 		}
720 
721 		rtnl_lock();
722 	}
723 
724 	if (!np->local_ip.ip) {
725 		if (!np->ipv6) {
726 			const struct in_ifaddr *ifa;
727 
728 			in_dev = __in_dev_get_rtnl(ndev);
729 			if (!in_dev)
730 				goto put_noaddr;
731 
732 			ifa = rtnl_dereference(in_dev->ifa_list);
733 			if (!ifa) {
734 put_noaddr:
735 				np_err(np, "no IP address for %s, aborting\n",
736 				       np->dev_name);
737 				err = -EDESTADDRREQ;
738 				goto put;
739 			}
740 
741 			np->local_ip.ip = ifa->ifa_local;
742 			np_info(np, "local IP %pI4\n", &np->local_ip.ip);
743 		} else {
744 #if IS_ENABLED(CONFIG_IPV6)
745 			struct inet6_dev *idev;
746 
747 			err = -EDESTADDRREQ;
748 			idev = __in6_dev_get(ndev);
749 			if (idev) {
750 				struct inet6_ifaddr *ifp;
751 
752 				read_lock_bh(&idev->lock);
753 				list_for_each_entry(ifp, &idev->addr_list, if_list) {
754 					if (!!(ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL) !=
755 					    !!(ipv6_addr_type(&np->remote_ip.in6) & IPV6_ADDR_LINKLOCAL))
756 						continue;
757 					np->local_ip.in6 = ifp->addr;
758 					err = 0;
759 					break;
760 				}
761 				read_unlock_bh(&idev->lock);
762 			}
763 			if (err) {
764 				np_err(np, "no IPv6 address for %s, aborting\n",
765 				       np->dev_name);
766 				goto put;
767 			} else
768 				np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
769 #else
770 			np_err(np, "IPv6 is not supported %s, aborting\n",
771 			       np->dev_name);
772 			err = -EINVAL;
773 			goto put;
774 #endif
775 		}
776 	}
777 
778 	/* fill up the skb queue */
779 	refill_skbs();
780 
781 	err = __netpoll_setup(np, ndev);
782 	if (err)
783 		goto put;
784 	rtnl_unlock();
785 	return 0;
786 
787 put:
788 	netdev_put(ndev, &np->dev_tracker);
789 unlock:
790 	rtnl_unlock();
791 	return err;
792 }
793 EXPORT_SYMBOL(netpoll_setup);
794 
795 static int __init netpoll_init(void)
796 {
797 	skb_queue_head_init(&skb_pool);
798 	return 0;
799 }
800 core_initcall(netpoll_init);
801 
802 static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
803 {
804 	struct netpoll_info *npinfo =
805 			container_of(rcu_head, struct netpoll_info, rcu);
806 
807 	skb_queue_purge(&npinfo->txq);
808 
809 	/* we can't call cancel_delayed_work_sync here, as we are in softirq */
810 	cancel_delayed_work(&npinfo->tx_work);
811 
812 	/* clean after last, unfinished work */
813 	__skb_queue_purge(&npinfo->txq);
814 	/* now cancel it again */
815 	cancel_delayed_work(&npinfo->tx_work);
816 	kfree(npinfo);
817 }
818 
819 void __netpoll_cleanup(struct netpoll *np)
820 {
821 	struct netpoll_info *npinfo;
822 
823 	npinfo = rtnl_dereference(np->dev->npinfo);
824 	if (!npinfo)
825 		return;
826 
827 	synchronize_srcu(&netpoll_srcu);
828 
829 	if (refcount_dec_and_test(&npinfo->refcnt)) {
830 		const struct net_device_ops *ops;
831 
832 		ops = np->dev->netdev_ops;
833 		if (ops->ndo_netpoll_cleanup)
834 			ops->ndo_netpoll_cleanup(np->dev);
835 
836 		RCU_INIT_POINTER(np->dev->npinfo, NULL);
837 		call_rcu(&npinfo->rcu, rcu_cleanup_netpoll_info);
838 	} else
839 		RCU_INIT_POINTER(np->dev->npinfo, NULL);
840 }
841 EXPORT_SYMBOL_GPL(__netpoll_cleanup);
842 
843 void __netpoll_free(struct netpoll *np)
844 {
845 	ASSERT_RTNL();
846 
847 	/* Wait for transmitting packets to finish before freeing. */
848 	synchronize_rcu();
849 	__netpoll_cleanup(np);
850 	kfree(np);
851 }
852 EXPORT_SYMBOL_GPL(__netpoll_free);
853 
854 void netpoll_cleanup(struct netpoll *np)
855 {
856 	rtnl_lock();
857 	if (!np->dev)
858 		goto out;
859 	__netpoll_cleanup(np);
860 	netdev_put(np->dev, &np->dev_tracker);
861 	np->dev = NULL;
862 out:
863 	rtnl_unlock();
864 }
865 EXPORT_SYMBOL(netpoll_cleanup);
866