xref: /linux/net/core/netpoll.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
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 <linux/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 #define USEC_PER_POLL	50
49 
50 #define MAX_SKB_SIZE							\
51 	(sizeof(struct ethhdr) +					\
52 	 sizeof(struct iphdr) +						\
53 	 sizeof(struct udphdr) +					\
54 	 MAX_UDP_CHUNK)
55 
56 static void zap_completion_queue(void);
57 
58 static unsigned int carrier_timeout = 4;
59 module_param(carrier_timeout, uint, 0644);
60 
61 #define np_info(np, fmt, ...)				\
62 	pr_info("%s: " fmt, np->name, ##__VA_ARGS__)
63 #define np_err(np, fmt, ...)				\
64 	pr_err("%s: " fmt, np->name, ##__VA_ARGS__)
65 #define np_notice(np, fmt, ...)				\
66 	pr_notice("%s: " fmt, np->name, ##__VA_ARGS__)
67 
68 static netdev_tx_t netpoll_start_xmit(struct sk_buff *skb,
69 				      struct net_device *dev,
70 				      struct netdev_queue *txq)
71 {
72 	netdev_tx_t status = NETDEV_TX_OK;
73 	netdev_features_t features;
74 
75 	features = netif_skb_features(skb);
76 
77 	if (skb_vlan_tag_present(skb) &&
78 	    !vlan_hw_offload_capable(features, skb->vlan_proto)) {
79 		skb = __vlan_hwaccel_push_inside(skb);
80 		if (unlikely(!skb)) {
81 			/* This is actually a packet drop, but we
82 			 * don't want the code that calls this
83 			 * function to try and operate on a NULL skb.
84 			 */
85 			goto out;
86 		}
87 	}
88 
89 	status = netdev_start_xmit(skb, dev, txq, false);
90 
91 out:
92 	return status;
93 }
94 
95 static void queue_process(struct work_struct *work)
96 {
97 	struct netpoll_info *npinfo =
98 		container_of(work, struct netpoll_info, tx_work.work);
99 	struct sk_buff *skb;
100 	unsigned long flags;
101 
102 	while ((skb = skb_dequeue(&npinfo->txq))) {
103 		struct net_device *dev = skb->dev;
104 		struct netdev_queue *txq;
105 		unsigned int q_index;
106 
107 		if (!netif_device_present(dev) || !netif_running(dev)) {
108 			kfree_skb(skb);
109 			continue;
110 		}
111 
112 		local_irq_save(flags);
113 		/* check if skb->queue_mapping is still valid */
114 		q_index = skb_get_queue_mapping(skb);
115 		if (unlikely(q_index >= dev->real_num_tx_queues)) {
116 			q_index = q_index % dev->real_num_tx_queues;
117 			skb_set_queue_mapping(skb, q_index);
118 		}
119 		txq = netdev_get_tx_queue(dev, q_index);
120 		HARD_TX_LOCK(dev, txq, smp_processor_id());
121 		if (netif_xmit_frozen_or_stopped(txq) ||
122 		    !dev_xmit_complete(netpoll_start_xmit(skb, dev, txq))) {
123 			skb_queue_head(&npinfo->txq, skb);
124 			HARD_TX_UNLOCK(dev, txq);
125 			local_irq_restore(flags);
126 
127 			schedule_delayed_work(&npinfo->tx_work, HZ/10);
128 			return;
129 		}
130 		HARD_TX_UNLOCK(dev, txq);
131 		local_irq_restore(flags);
132 	}
133 }
134 
135 static int netif_local_xmit_active(struct net_device *dev)
136 {
137 	int i;
138 
139 	for (i = 0; i < dev->num_tx_queues; i++) {
140 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
141 
142 		if (READ_ONCE(txq->xmit_lock_owner) == smp_processor_id())
143 			return 1;
144 	}
145 
146 	return 0;
147 }
148 
149 static void poll_one_napi(struct napi_struct *napi)
150 {
151 	int work;
152 
153 	/* If we set this bit but see that it has already been set,
154 	 * that indicates that napi has been disabled and we need
155 	 * to abort this operation
156 	 */
157 	if (test_and_set_bit(NAPI_STATE_NPSVC, &napi->state))
158 		return;
159 
160 	/* We explicitly pass the polling call a budget of 0 to
161 	 * indicate that we are clearing the Tx path only.
162 	 */
163 	work = napi->poll(napi, 0);
164 	WARN_ONCE(work, "%pS exceeded budget in poll\n", napi->poll);
165 	trace_napi_poll(napi, work, 0);
166 
167 	clear_bit(NAPI_STATE_NPSVC, &napi->state);
168 }
169 
170 static void poll_napi(struct net_device *dev)
171 {
172 	struct napi_struct *napi;
173 	int cpu = smp_processor_id();
174 
175 	list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) {
176 		if (cmpxchg(&napi->poll_owner, -1, cpu) == -1) {
177 			poll_one_napi(napi);
178 			smp_store_release(&napi->poll_owner, -1);
179 		}
180 	}
181 }
182 
183 void netpoll_poll_dev(struct net_device *dev)
184 {
185 	struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
186 	const struct net_device_ops *ops;
187 
188 	/* Don't do any rx activity if the dev_lock mutex is held
189 	 * the dev_open/close paths use this to block netpoll activity
190 	 * while changing device state
191 	 */
192 	if (!ni || down_trylock(&ni->dev_lock))
193 		return;
194 
195 	/* Some drivers will take the same locks in poll and xmit,
196 	 * we can't poll if local CPU is already in xmit.
197 	 */
198 	if (!netif_running(dev) || netif_local_xmit_active(dev)) {
199 		up(&ni->dev_lock);
200 		return;
201 	}
202 
203 	ops = dev->netdev_ops;
204 	if (ops->ndo_poll_controller)
205 		ops->ndo_poll_controller(dev);
206 
207 	poll_napi(dev);
208 
209 	up(&ni->dev_lock);
210 
211 	zap_completion_queue();
212 }
213 EXPORT_SYMBOL(netpoll_poll_dev);
214 
215 void netpoll_poll_disable(struct net_device *dev)
216 {
217 	struct netpoll_info *ni;
218 
219 	might_sleep();
220 	ni = rtnl_dereference(dev->npinfo);
221 	if (ni)
222 		down(&ni->dev_lock);
223 }
224 
225 void netpoll_poll_enable(struct net_device *dev)
226 {
227 	struct netpoll_info *ni;
228 
229 	ni = rtnl_dereference(dev->npinfo);
230 	if (ni)
231 		up(&ni->dev_lock);
232 }
233 
234 static void refill_skbs(struct netpoll *np)
235 {
236 	struct sk_buff_head *skb_pool;
237 	struct sk_buff *skb;
238 	unsigned long flags;
239 
240 	skb_pool = &np->skb_pool;
241 
242 	spin_lock_irqsave(&skb_pool->lock, flags);
243 	while (skb_pool->qlen < MAX_SKBS) {
244 		skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
245 		if (!skb)
246 			break;
247 
248 		__skb_queue_tail(skb_pool, skb);
249 	}
250 	spin_unlock_irqrestore(&skb_pool->lock, flags);
251 }
252 
253 static void zap_completion_queue(void)
254 {
255 	unsigned long flags;
256 	struct softnet_data *sd = &get_cpu_var(softnet_data);
257 
258 	if (sd->completion_queue) {
259 		struct sk_buff *clist;
260 
261 		local_irq_save(flags);
262 		clist = sd->completion_queue;
263 		sd->completion_queue = NULL;
264 		local_irq_restore(flags);
265 
266 		while (clist != NULL) {
267 			struct sk_buff *skb = clist;
268 			clist = clist->next;
269 			if (!skb_irq_freeable(skb)) {
270 				refcount_set(&skb->users, 1);
271 				dev_kfree_skb_any(skb); /* put this one back */
272 			} else {
273 				__kfree_skb(skb);
274 			}
275 		}
276 	}
277 
278 	put_cpu_var(softnet_data);
279 }
280 
281 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
282 {
283 	int count = 0;
284 	struct sk_buff *skb;
285 
286 	zap_completion_queue();
287 	refill_skbs(np);
288 repeat:
289 
290 	skb = alloc_skb(len, GFP_ATOMIC);
291 	if (!skb)
292 		skb = skb_dequeue(&np->skb_pool);
293 
294 	if (!skb) {
295 		if (++count < 10) {
296 			netpoll_poll_dev(np->dev);
297 			goto repeat;
298 		}
299 		return NULL;
300 	}
301 
302 	refcount_set(&skb->users, 1);
303 	skb_reserve(skb, reserve);
304 	return skb;
305 }
306 
307 static int netpoll_owner_active(struct net_device *dev)
308 {
309 	struct napi_struct *napi;
310 
311 	list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) {
312 		if (READ_ONCE(napi->poll_owner) == smp_processor_id())
313 			return 1;
314 	}
315 	return 0;
316 }
317 
318 /* call with IRQ disabled */
319 static netdev_tx_t __netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
320 {
321 	netdev_tx_t status = NETDEV_TX_BUSY;
322 	struct net_device *dev;
323 	unsigned long tries;
324 	/* It is up to the caller to keep npinfo alive. */
325 	struct netpoll_info *npinfo;
326 
327 	lockdep_assert_irqs_disabled();
328 
329 	dev = np->dev;
330 	npinfo = rcu_dereference_bh(dev->npinfo);
331 
332 	if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
333 		dev_kfree_skb_irq(skb);
334 		return NET_XMIT_DROP;
335 	}
336 
337 	/* don't get messages out of order, and no recursion */
338 	if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
339 		struct netdev_queue *txq;
340 
341 		txq = netdev_core_pick_tx(dev, skb, NULL);
342 
343 		/* try until next clock tick */
344 		for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
345 		     tries > 0; --tries) {
346 			if (HARD_TX_TRYLOCK(dev, txq)) {
347 				if (!netif_xmit_stopped(txq))
348 					status = netpoll_start_xmit(skb, dev, txq);
349 
350 				HARD_TX_UNLOCK(dev, txq);
351 
352 				if (dev_xmit_complete(status))
353 					break;
354 
355 			}
356 
357 			/* tickle device maybe there is some cleanup */
358 			netpoll_poll_dev(np->dev);
359 
360 			udelay(USEC_PER_POLL);
361 		}
362 
363 		WARN_ONCE(!irqs_disabled(),
364 			"netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pS)\n",
365 			dev->name, dev->netdev_ops->ndo_start_xmit);
366 
367 	}
368 
369 	if (!dev_xmit_complete(status)) {
370 		skb_queue_tail(&npinfo->txq, skb);
371 		schedule_delayed_work(&npinfo->tx_work,0);
372 	}
373 	return NETDEV_TX_OK;
374 }
375 
376 netdev_tx_t netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
377 {
378 	unsigned long flags;
379 	netdev_tx_t ret;
380 
381 	if (unlikely(!np)) {
382 		dev_kfree_skb_irq(skb);
383 		ret = NET_XMIT_DROP;
384 	} else {
385 		local_irq_save(flags);
386 		ret = __netpoll_send_skb(np, skb);
387 		local_irq_restore(flags);
388 	}
389 	return ret;
390 }
391 EXPORT_SYMBOL(netpoll_send_skb);
392 
393 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
394 {
395 	int total_len, ip_len, udp_len;
396 	struct sk_buff *skb;
397 	struct udphdr *udph;
398 	struct iphdr *iph;
399 	struct ethhdr *eth;
400 	static atomic_t ip_ident;
401 	struct ipv6hdr *ip6h;
402 
403 	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
404 		WARN_ON_ONCE(!irqs_disabled());
405 
406 	udp_len = len + sizeof(*udph);
407 	if (np->ipv6)
408 		ip_len = udp_len + sizeof(*ip6h);
409 	else
410 		ip_len = udp_len + sizeof(*iph);
411 
412 	total_len = ip_len + LL_RESERVED_SPACE(np->dev);
413 
414 	skb = find_skb(np, total_len + np->dev->needed_tailroom,
415 		       total_len - len);
416 	if (!skb)
417 		return;
418 
419 	skb_copy_to_linear_data(skb, msg, len);
420 	skb_put(skb, len);
421 
422 	skb_push(skb, sizeof(*udph));
423 	skb_reset_transport_header(skb);
424 	udph = udp_hdr(skb);
425 	udph->source = htons(np->local_port);
426 	udph->dest = htons(np->remote_port);
427 	udph->len = htons(udp_len);
428 
429 	if (np->ipv6) {
430 		udph->check = 0;
431 		udph->check = csum_ipv6_magic(&np->local_ip.in6,
432 					      &np->remote_ip.in6,
433 					      udp_len, IPPROTO_UDP,
434 					      csum_partial(udph, udp_len, 0));
435 		if (udph->check == 0)
436 			udph->check = CSUM_MANGLED_0;
437 
438 		skb_push(skb, sizeof(*ip6h));
439 		skb_reset_network_header(skb);
440 		ip6h = ipv6_hdr(skb);
441 
442 		/* ip6h->version = 6; ip6h->priority = 0; */
443 		*(unsigned char *)ip6h = 0x60;
444 		ip6h->flow_lbl[0] = 0;
445 		ip6h->flow_lbl[1] = 0;
446 		ip6h->flow_lbl[2] = 0;
447 
448 		ip6h->payload_len = htons(sizeof(struct udphdr) + len);
449 		ip6h->nexthdr = IPPROTO_UDP;
450 		ip6h->hop_limit = 32;
451 		ip6h->saddr = np->local_ip.in6;
452 		ip6h->daddr = np->remote_ip.in6;
453 
454 		eth = skb_push(skb, ETH_HLEN);
455 		skb_reset_mac_header(skb);
456 		skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
457 	} else {
458 		udph->check = 0;
459 		udph->check = csum_tcpudp_magic(np->local_ip.ip,
460 						np->remote_ip.ip,
461 						udp_len, IPPROTO_UDP,
462 						csum_partial(udph, udp_len, 0));
463 		if (udph->check == 0)
464 			udph->check = CSUM_MANGLED_0;
465 
466 		skb_push(skb, sizeof(*iph));
467 		skb_reset_network_header(skb);
468 		iph = ip_hdr(skb);
469 
470 		/* iph->version = 4; iph->ihl = 5; */
471 		*(unsigned char *)iph = 0x45;
472 		iph->tos      = 0;
473 		put_unaligned(htons(ip_len), &(iph->tot_len));
474 		iph->id       = htons(atomic_inc_return(&ip_ident));
475 		iph->frag_off = 0;
476 		iph->ttl      = 64;
477 		iph->protocol = IPPROTO_UDP;
478 		iph->check    = 0;
479 		put_unaligned(np->local_ip.ip, &(iph->saddr));
480 		put_unaligned(np->remote_ip.ip, &(iph->daddr));
481 		iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
482 
483 		eth = skb_push(skb, ETH_HLEN);
484 		skb_reset_mac_header(skb);
485 		skb->protocol = eth->h_proto = htons(ETH_P_IP);
486 	}
487 
488 	ether_addr_copy(eth->h_source, np->dev->dev_addr);
489 	ether_addr_copy(eth->h_dest, np->remote_mac);
490 
491 	skb->dev = np->dev;
492 
493 	netpoll_send_skb(np, skb);
494 }
495 EXPORT_SYMBOL(netpoll_send_udp);
496 
497 void netpoll_print_options(struct netpoll *np)
498 {
499 	np_info(np, "local port %d\n", np->local_port);
500 	if (np->ipv6)
501 		np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
502 	else
503 		np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
504 	np_info(np, "interface '%s'\n", np->dev_name);
505 	np_info(np, "remote port %d\n", np->remote_port);
506 	if (np->ipv6)
507 		np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
508 	else
509 		np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
510 	np_info(np, "remote ethernet address %pM\n", np->remote_mac);
511 }
512 EXPORT_SYMBOL(netpoll_print_options);
513 
514 static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
515 {
516 	const char *end;
517 
518 	if (!strchr(str, ':') &&
519 	    in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
520 		if (!*end)
521 			return 0;
522 	}
523 	if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
524 #if IS_ENABLED(CONFIG_IPV6)
525 		if (!*end)
526 			return 1;
527 #else
528 		return -1;
529 #endif
530 	}
531 	return -1;
532 }
533 
534 static void skb_pool_flush(struct netpoll *np)
535 {
536 	struct sk_buff_head *skb_pool;
537 
538 	skb_pool = &np->skb_pool;
539 	skb_queue_purge_reason(skb_pool, SKB_CONSUMED);
540 }
541 
542 int netpoll_parse_options(struct netpoll *np, char *opt)
543 {
544 	char *cur=opt, *delim;
545 	int ipv6;
546 	bool ipversion_set = false;
547 
548 	if (*cur != '@') {
549 		if ((delim = strchr(cur, '@')) == NULL)
550 			goto parse_failed;
551 		*delim = 0;
552 		if (kstrtou16(cur, 10, &np->local_port))
553 			goto parse_failed;
554 		cur = delim;
555 	}
556 	cur++;
557 
558 	if (*cur != '/') {
559 		ipversion_set = true;
560 		if ((delim = strchr(cur, '/')) == NULL)
561 			goto parse_failed;
562 		*delim = 0;
563 		ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
564 		if (ipv6 < 0)
565 			goto parse_failed;
566 		else
567 			np->ipv6 = (bool)ipv6;
568 		cur = delim;
569 	}
570 	cur++;
571 
572 	if (*cur != ',') {
573 		/* parse out dev name */
574 		if ((delim = strchr(cur, ',')) == NULL)
575 			goto parse_failed;
576 		*delim = 0;
577 		strscpy(np->dev_name, cur, sizeof(np->dev_name));
578 		cur = delim;
579 	}
580 	cur++;
581 
582 	if (*cur != '@') {
583 		/* dst port */
584 		if ((delim = strchr(cur, '@')) == NULL)
585 			goto parse_failed;
586 		*delim = 0;
587 		if (*cur == ' ' || *cur == '\t')
588 			np_info(np, "warning: whitespace is not allowed\n");
589 		if (kstrtou16(cur, 10, &np->remote_port))
590 			goto parse_failed;
591 		cur = delim;
592 	}
593 	cur++;
594 
595 	/* dst ip */
596 	if ((delim = strchr(cur, '/')) == NULL)
597 		goto parse_failed;
598 	*delim = 0;
599 	ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
600 	if (ipv6 < 0)
601 		goto parse_failed;
602 	else if (ipversion_set && np->ipv6 != (bool)ipv6)
603 		goto parse_failed;
604 	else
605 		np->ipv6 = (bool)ipv6;
606 	cur = delim + 1;
607 
608 	if (*cur != 0) {
609 		/* MAC address */
610 		if (!mac_pton(cur, np->remote_mac))
611 			goto parse_failed;
612 	}
613 
614 	netpoll_print_options(np);
615 
616 	return 0;
617 
618  parse_failed:
619 	np_info(np, "couldn't parse config at '%s'!\n", cur);
620 	return -1;
621 }
622 EXPORT_SYMBOL(netpoll_parse_options);
623 
624 int __netpoll_setup(struct netpoll *np, struct net_device *ndev)
625 {
626 	struct netpoll_info *npinfo;
627 	const struct net_device_ops *ops;
628 	int err;
629 
630 	if (ndev->priv_flags & IFF_DISABLE_NETPOLL) {
631 		np_err(np, "%s doesn't support polling, aborting\n",
632 		       ndev->name);
633 		err = -ENOTSUPP;
634 		goto out;
635 	}
636 
637 	if (!rcu_access_pointer(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 = ndev->netdev_ops;
651 		if (ops->ndo_netpoll_setup) {
652 			err = ops->ndo_netpoll_setup(ndev);
653 			if (err)
654 				goto free_npinfo;
655 		}
656 	} else {
657 		npinfo = rtnl_dereference(ndev->npinfo);
658 		refcount_inc(&npinfo->refcnt);
659 	}
660 
661 	np->dev = ndev;
662 	strscpy(np->dev_name, ndev->name, IFNAMSIZ);
663 	npinfo->netpoll = np;
664 
665 	/* last thing to do is link it to the net device structure */
666 	rcu_assign_pointer(ndev->npinfo, npinfo);
667 
668 	return 0;
669 
670 free_npinfo:
671 	kfree(npinfo);
672 out:
673 	return err;
674 }
675 EXPORT_SYMBOL_GPL(__netpoll_setup);
676 
677 int netpoll_setup(struct netpoll *np)
678 {
679 	struct net_device *ndev = NULL;
680 	bool ip_overwritten = false;
681 	struct in_device *in_dev;
682 	int err;
683 
684 	skb_queue_head_init(&np->skb_pool);
685 
686 	rtnl_lock();
687 	if (np->dev_name[0]) {
688 		struct net *net = current->nsproxy->net_ns;
689 		ndev = __dev_get_by_name(net, np->dev_name);
690 	}
691 	if (!ndev) {
692 		np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
693 		err = -ENODEV;
694 		goto unlock;
695 	}
696 	netdev_hold(ndev, &np->dev_tracker, GFP_KERNEL);
697 
698 	if (netdev_master_upper_dev_get(ndev)) {
699 		np_err(np, "%s is a slave device, aborting\n", np->dev_name);
700 		err = -EBUSY;
701 		goto put;
702 	}
703 
704 	if (!netif_running(ndev)) {
705 		unsigned long atmost;
706 
707 		np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
708 
709 		err = dev_open(ndev, NULL);
710 
711 		if (err) {
712 			np_err(np, "failed to open %s\n", ndev->name);
713 			goto put;
714 		}
715 
716 		rtnl_unlock();
717 		atmost = jiffies + carrier_timeout * HZ;
718 		while (!netif_carrier_ok(ndev)) {
719 			if (time_after(jiffies, atmost)) {
720 				np_notice(np, "timeout waiting for carrier\n");
721 				break;
722 			}
723 			msleep(1);
724 		}
725 
726 		rtnl_lock();
727 	}
728 
729 	if (!np->local_ip.ip) {
730 		if (!np->ipv6) {
731 			const struct in_ifaddr *ifa;
732 
733 			in_dev = __in_dev_get_rtnl(ndev);
734 			if (!in_dev)
735 				goto put_noaddr;
736 
737 			ifa = rtnl_dereference(in_dev->ifa_list);
738 			if (!ifa) {
739 put_noaddr:
740 				np_err(np, "no IP address for %s, aborting\n",
741 				       np->dev_name);
742 				err = -EDESTADDRREQ;
743 				goto put;
744 			}
745 
746 			np->local_ip.ip = ifa->ifa_local;
747 			ip_overwritten = true;
748 			np_info(np, "local IP %pI4\n", &np->local_ip.ip);
749 		} else {
750 #if IS_ENABLED(CONFIG_IPV6)
751 			struct inet6_dev *idev;
752 
753 			err = -EDESTADDRREQ;
754 			idev = __in6_dev_get(ndev);
755 			if (idev) {
756 				struct inet6_ifaddr *ifp;
757 
758 				read_lock_bh(&idev->lock);
759 				list_for_each_entry(ifp, &idev->addr_list, if_list) {
760 					if (!!(ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL) !=
761 					    !!(ipv6_addr_type(&np->remote_ip.in6) & IPV6_ADDR_LINKLOCAL))
762 						continue;
763 					np->local_ip.in6 = ifp->addr;
764 					ip_overwritten = true;
765 					err = 0;
766 					break;
767 				}
768 				read_unlock_bh(&idev->lock);
769 			}
770 			if (err) {
771 				np_err(np, "no IPv6 address for %s, aborting\n",
772 				       np->dev_name);
773 				goto put;
774 			} else
775 				np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
776 #else
777 			np_err(np, "IPv6 is not supported %s, aborting\n",
778 			       np->dev_name);
779 			err = -EINVAL;
780 			goto put;
781 #endif
782 		}
783 	}
784 
785 	/* fill up the skb queue */
786 	refill_skbs(np);
787 
788 	err = __netpoll_setup(np, ndev);
789 	if (err)
790 		goto flush;
791 	rtnl_unlock();
792 	return 0;
793 
794 flush:
795 	skb_pool_flush(np);
796 put:
797 	DEBUG_NET_WARN_ON_ONCE(np->dev);
798 	if (ip_overwritten)
799 		memset(&np->local_ip, 0, sizeof(np->local_ip));
800 	netdev_put(ndev, &np->dev_tracker);
801 unlock:
802 	rtnl_unlock();
803 	return err;
804 }
805 EXPORT_SYMBOL(netpoll_setup);
806 
807 static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
808 {
809 	struct netpoll_info *npinfo =
810 			container_of(rcu_head, struct netpoll_info, rcu);
811 
812 	skb_queue_purge(&npinfo->txq);
813 
814 	/* we can't call cancel_delayed_work_sync here, as we are in softirq */
815 	cancel_delayed_work(&npinfo->tx_work);
816 
817 	/* clean after last, unfinished work */
818 	__skb_queue_purge(&npinfo->txq);
819 	/* now cancel it again */
820 	cancel_delayed_work(&npinfo->tx_work);
821 	kfree(npinfo);
822 }
823 
824 void __netpoll_cleanup(struct netpoll *np)
825 {
826 	struct netpoll_info *npinfo;
827 
828 	npinfo = rtnl_dereference(np->dev->npinfo);
829 	if (!npinfo)
830 		return;
831 
832 	if (refcount_dec_and_test(&npinfo->refcnt)) {
833 		const struct net_device_ops *ops;
834 
835 		ops = np->dev->netdev_ops;
836 		if (ops->ndo_netpoll_cleanup)
837 			ops->ndo_netpoll_cleanup(np->dev);
838 
839 		RCU_INIT_POINTER(np->dev->npinfo, NULL);
840 		call_rcu(&npinfo->rcu, rcu_cleanup_netpoll_info);
841 	} else
842 		RCU_INIT_POINTER(np->dev->npinfo, NULL);
843 
844 	skb_pool_flush(np);
845 }
846 EXPORT_SYMBOL_GPL(__netpoll_cleanup);
847 
848 void __netpoll_free(struct netpoll *np)
849 {
850 	ASSERT_RTNL();
851 
852 	/* Wait for transmitting packets to finish before freeing. */
853 	synchronize_rcu();
854 	__netpoll_cleanup(np);
855 	kfree(np);
856 }
857 EXPORT_SYMBOL_GPL(__netpoll_free);
858 
859 void do_netpoll_cleanup(struct netpoll *np)
860 {
861 	__netpoll_cleanup(np);
862 	netdev_put(np->dev, &np->dev_tracker);
863 	np->dev = NULL;
864 }
865 EXPORT_SYMBOL(do_netpoll_cleanup);
866 
867 void netpoll_cleanup(struct netpoll *np)
868 {
869 	rtnl_lock();
870 	if (!np->dev)
871 		goto out;
872 	do_netpoll_cleanup(np);
873 out:
874 	rtnl_unlock();
875 }
876 EXPORT_SYMBOL(netpoll_cleanup);
877