xref: /linux/net/core/netpoll.c (revision f4915933947c71f08ed1c5a6c9b4fdbe735e18cf)
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 repeat:
288 
289 	skb = alloc_skb(len, GFP_ATOMIC);
290 	if (!skb) {
291 		skb = skb_dequeue(&np->skb_pool);
292 		schedule_work(&np->refill_wq);
293 	}
294 
295 	if (!skb) {
296 		if (++count < 10) {
297 			netpoll_poll_dev(np->dev);
298 			goto repeat;
299 		}
300 		return NULL;
301 	}
302 
303 	refcount_set(&skb->users, 1);
304 	skb_reserve(skb, reserve);
305 	return skb;
306 }
307 
308 static int netpoll_owner_active(struct net_device *dev)
309 {
310 	struct napi_struct *napi;
311 
312 	list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) {
313 		if (READ_ONCE(napi->poll_owner) == smp_processor_id())
314 			return 1;
315 	}
316 	return 0;
317 }
318 
319 /* call with IRQ disabled */
320 static netdev_tx_t __netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
321 {
322 	netdev_tx_t status = NETDEV_TX_BUSY;
323 	netdev_tx_t ret = NET_XMIT_DROP;
324 	struct net_device *dev;
325 	unsigned long tries;
326 	/* It is up to the caller to keep npinfo alive. */
327 	struct netpoll_info *npinfo;
328 
329 	lockdep_assert_irqs_disabled();
330 
331 	dev = np->dev;
332 	rcu_read_lock();
333 	npinfo = rcu_dereference_bh(dev->npinfo);
334 
335 	if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
336 		dev_kfree_skb_irq(skb);
337 		goto out;
338 	}
339 
340 	/* don't get messages out of order, and no recursion */
341 	if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
342 		struct netdev_queue *txq;
343 
344 		txq = netdev_core_pick_tx(dev, skb, NULL);
345 
346 		/* try until next clock tick */
347 		for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
348 		     tries > 0; --tries) {
349 			if (HARD_TX_TRYLOCK(dev, txq)) {
350 				if (!netif_xmit_stopped(txq))
351 					status = netpoll_start_xmit(skb, dev, txq);
352 
353 				HARD_TX_UNLOCK(dev, txq);
354 
355 				if (dev_xmit_complete(status))
356 					break;
357 
358 			}
359 
360 			/* tickle device maybe there is some cleanup */
361 			netpoll_poll_dev(np->dev);
362 
363 			udelay(USEC_PER_POLL);
364 		}
365 
366 		WARN_ONCE(!irqs_disabled(),
367 			"netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pS)\n",
368 			dev->name, dev->netdev_ops->ndo_start_xmit);
369 
370 	}
371 
372 	if (!dev_xmit_complete(status)) {
373 		skb_queue_tail(&npinfo->txq, skb);
374 		schedule_delayed_work(&npinfo->tx_work,0);
375 	}
376 	ret = NETDEV_TX_OK;
377 out:
378 	rcu_read_unlock();
379 	return ret;
380 }
381 
382 netdev_tx_t netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
383 {
384 	unsigned long flags;
385 	netdev_tx_t ret;
386 
387 	if (unlikely(!np)) {
388 		dev_kfree_skb_irq(skb);
389 		ret = NET_XMIT_DROP;
390 	} else {
391 		local_irq_save(flags);
392 		ret = __netpoll_send_skb(np, skb);
393 		local_irq_restore(flags);
394 	}
395 	return ret;
396 }
397 EXPORT_SYMBOL(netpoll_send_skb);
398 
399 int netpoll_send_udp(struct netpoll *np, const char *msg, int len)
400 {
401 	int total_len, ip_len, udp_len;
402 	struct sk_buff *skb;
403 	struct udphdr *udph;
404 	struct iphdr *iph;
405 	struct ethhdr *eth;
406 	static atomic_t ip_ident;
407 	struct ipv6hdr *ip6h;
408 
409 	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
410 		WARN_ON_ONCE(!irqs_disabled());
411 
412 	udp_len = len + sizeof(*udph);
413 	if (np->ipv6)
414 		ip_len = udp_len + sizeof(*ip6h);
415 	else
416 		ip_len = udp_len + sizeof(*iph);
417 
418 	total_len = ip_len + LL_RESERVED_SPACE(np->dev);
419 
420 	skb = find_skb(np, total_len + np->dev->needed_tailroom,
421 		       total_len - len);
422 	if (!skb)
423 		return -ENOMEM;
424 
425 	skb_copy_to_linear_data(skb, msg, len);
426 	skb_put(skb, len);
427 
428 	skb_push(skb, sizeof(*udph));
429 	skb_reset_transport_header(skb);
430 	udph = udp_hdr(skb);
431 	udph->source = htons(np->local_port);
432 	udph->dest = htons(np->remote_port);
433 	udph->len = htons(udp_len);
434 
435 	if (np->ipv6) {
436 		udph->check = 0;
437 		udph->check = csum_ipv6_magic(&np->local_ip.in6,
438 					      &np->remote_ip.in6,
439 					      udp_len, IPPROTO_UDP,
440 					      csum_partial(udph, udp_len, 0));
441 		if (udph->check == 0)
442 			udph->check = CSUM_MANGLED_0;
443 
444 		skb_push(skb, sizeof(*ip6h));
445 		skb_reset_network_header(skb);
446 		ip6h = ipv6_hdr(skb);
447 
448 		/* ip6h->version = 6; ip6h->priority = 0; */
449 		*(unsigned char *)ip6h = 0x60;
450 		ip6h->flow_lbl[0] = 0;
451 		ip6h->flow_lbl[1] = 0;
452 		ip6h->flow_lbl[2] = 0;
453 
454 		ip6h->payload_len = htons(sizeof(struct udphdr) + len);
455 		ip6h->nexthdr = IPPROTO_UDP;
456 		ip6h->hop_limit = 32;
457 		ip6h->saddr = np->local_ip.in6;
458 		ip6h->daddr = np->remote_ip.in6;
459 
460 		eth = skb_push(skb, ETH_HLEN);
461 		skb_reset_mac_header(skb);
462 		skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
463 	} else {
464 		udph->check = 0;
465 		udph->check = csum_tcpudp_magic(np->local_ip.ip,
466 						np->remote_ip.ip,
467 						udp_len, IPPROTO_UDP,
468 						csum_partial(udph, udp_len, 0));
469 		if (udph->check == 0)
470 			udph->check = CSUM_MANGLED_0;
471 
472 		skb_push(skb, sizeof(*iph));
473 		skb_reset_network_header(skb);
474 		iph = ip_hdr(skb);
475 
476 		/* iph->version = 4; iph->ihl = 5; */
477 		*(unsigned char *)iph = 0x45;
478 		iph->tos      = 0;
479 		put_unaligned(htons(ip_len), &(iph->tot_len));
480 		iph->id       = htons(atomic_inc_return(&ip_ident));
481 		iph->frag_off = 0;
482 		iph->ttl      = 64;
483 		iph->protocol = IPPROTO_UDP;
484 		iph->check    = 0;
485 		put_unaligned(np->local_ip.ip, &(iph->saddr));
486 		put_unaligned(np->remote_ip.ip, &(iph->daddr));
487 		iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
488 
489 		eth = skb_push(skb, ETH_HLEN);
490 		skb_reset_mac_header(skb);
491 		skb->protocol = eth->h_proto = htons(ETH_P_IP);
492 	}
493 
494 	ether_addr_copy(eth->h_source, np->dev->dev_addr);
495 	ether_addr_copy(eth->h_dest, np->remote_mac);
496 
497 	skb->dev = np->dev;
498 
499 	return (int)netpoll_send_skb(np, skb);
500 }
501 EXPORT_SYMBOL(netpoll_send_udp);
502 
503 void netpoll_print_options(struct netpoll *np)
504 {
505 	np_info(np, "local port %d\n", np->local_port);
506 	if (np->ipv6)
507 		np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
508 	else
509 		np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
510 	np_info(np, "interface name '%s'\n", np->dev_name);
511 	np_info(np, "local ethernet address '%pM'\n", np->dev_mac);
512 	np_info(np, "remote port %d\n", np->remote_port);
513 	if (np->ipv6)
514 		np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
515 	else
516 		np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
517 	np_info(np, "remote ethernet address %pM\n", np->remote_mac);
518 }
519 EXPORT_SYMBOL(netpoll_print_options);
520 
521 static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
522 {
523 	const char *end;
524 
525 	if (!strchr(str, ':') &&
526 	    in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
527 		if (!*end)
528 			return 0;
529 	}
530 	if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
531 #if IS_ENABLED(CONFIG_IPV6)
532 		if (!*end)
533 			return 1;
534 #else
535 		return -1;
536 #endif
537 	}
538 	return -1;
539 }
540 
541 static void skb_pool_flush(struct netpoll *np)
542 {
543 	struct sk_buff_head *skb_pool;
544 
545 	cancel_work_sync(&np->refill_wq);
546 	skb_pool = &np->skb_pool;
547 	skb_queue_purge_reason(skb_pool, SKB_CONSUMED);
548 }
549 
550 int netpoll_parse_options(struct netpoll *np, char *opt)
551 {
552 	char *cur=opt, *delim;
553 	int ipv6;
554 	bool ipversion_set = false;
555 
556 	if (*cur != '@') {
557 		if ((delim = strchr(cur, '@')) == NULL)
558 			goto parse_failed;
559 		*delim = 0;
560 		if (kstrtou16(cur, 10, &np->local_port))
561 			goto parse_failed;
562 		cur = delim;
563 	}
564 	cur++;
565 
566 	if (*cur != '/') {
567 		ipversion_set = true;
568 		if ((delim = strchr(cur, '/')) == NULL)
569 			goto parse_failed;
570 		*delim = 0;
571 		ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
572 		if (ipv6 < 0)
573 			goto parse_failed;
574 		else
575 			np->ipv6 = (bool)ipv6;
576 		cur = delim;
577 	}
578 	cur++;
579 
580 	if (*cur != ',') {
581 		/* parse out dev_name or dev_mac */
582 		if ((delim = strchr(cur, ',')) == NULL)
583 			goto parse_failed;
584 		*delim = 0;
585 
586 		np->dev_name[0] = '\0';
587 		eth_broadcast_addr(np->dev_mac);
588 		if (!strchr(cur, ':'))
589 			strscpy(np->dev_name, cur, sizeof(np->dev_name));
590 		else if (!mac_pton(cur, np->dev_mac))
591 			goto parse_failed;
592 
593 		cur = delim;
594 	}
595 	cur++;
596 
597 	if (*cur != '@') {
598 		/* dst port */
599 		if ((delim = strchr(cur, '@')) == NULL)
600 			goto parse_failed;
601 		*delim = 0;
602 		if (*cur == ' ' || *cur == '\t')
603 			np_info(np, "warning: whitespace is not allowed\n");
604 		if (kstrtou16(cur, 10, &np->remote_port))
605 			goto parse_failed;
606 		cur = delim;
607 	}
608 	cur++;
609 
610 	/* dst ip */
611 	if ((delim = strchr(cur, '/')) == NULL)
612 		goto parse_failed;
613 	*delim = 0;
614 	ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
615 	if (ipv6 < 0)
616 		goto parse_failed;
617 	else if (ipversion_set && np->ipv6 != (bool)ipv6)
618 		goto parse_failed;
619 	else
620 		np->ipv6 = (bool)ipv6;
621 	cur = delim + 1;
622 
623 	if (*cur != 0) {
624 		/* MAC address */
625 		if (!mac_pton(cur, np->remote_mac))
626 			goto parse_failed;
627 	}
628 
629 	netpoll_print_options(np);
630 
631 	return 0;
632 
633  parse_failed:
634 	np_info(np, "couldn't parse config at '%s'!\n", cur);
635 	return -1;
636 }
637 EXPORT_SYMBOL(netpoll_parse_options);
638 
639 static void refill_skbs_work_handler(struct work_struct *work)
640 {
641 	struct netpoll *np =
642 		container_of(work, struct netpoll, refill_wq);
643 
644 	refill_skbs(np);
645 }
646 
647 int __netpoll_setup(struct netpoll *np, struct net_device *ndev)
648 {
649 	struct netpoll_info *npinfo;
650 	const struct net_device_ops *ops;
651 	int err;
652 
653 	skb_queue_head_init(&np->skb_pool);
654 
655 	if (ndev->priv_flags & IFF_DISABLE_NETPOLL) {
656 		np_err(np, "%s doesn't support polling, aborting\n",
657 		       ndev->name);
658 		err = -ENOTSUPP;
659 		goto out;
660 	}
661 
662 	npinfo = rtnl_dereference(ndev->npinfo);
663 	if (!npinfo) {
664 		npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
665 		if (!npinfo) {
666 			err = -ENOMEM;
667 			goto out;
668 		}
669 
670 		sema_init(&npinfo->dev_lock, 1);
671 		skb_queue_head_init(&npinfo->txq);
672 		INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
673 
674 		refcount_set(&npinfo->refcnt, 1);
675 
676 		ops = ndev->netdev_ops;
677 		if (ops->ndo_netpoll_setup) {
678 			err = ops->ndo_netpoll_setup(ndev);
679 			if (err)
680 				goto free_npinfo;
681 		}
682 	} else {
683 		refcount_inc(&npinfo->refcnt);
684 	}
685 
686 	np->dev = ndev;
687 	strscpy(np->dev_name, ndev->name, IFNAMSIZ);
688 	npinfo->netpoll = np;
689 
690 	/* fill up the skb queue */
691 	refill_skbs(np);
692 	INIT_WORK(&np->refill_wq, refill_skbs_work_handler);
693 
694 	/* last thing to do is link it to the net device structure */
695 	rcu_assign_pointer(ndev->npinfo, npinfo);
696 
697 	return 0;
698 
699 free_npinfo:
700 	kfree(npinfo);
701 out:
702 	return err;
703 }
704 EXPORT_SYMBOL_GPL(__netpoll_setup);
705 
706 /*
707  * Returns a pointer to a string representation of the identifier used
708  * to select the egress interface for the given netpoll instance. buf
709  * must be a buffer of length at least MAC_ADDR_STR_LEN + 1.
710  */
711 static char *egress_dev(struct netpoll *np, char *buf)
712 {
713 	if (np->dev_name[0])
714 		return np->dev_name;
715 
716 	snprintf(buf, MAC_ADDR_STR_LEN, "%pM", np->dev_mac);
717 	return buf;
718 }
719 
720 int netpoll_setup(struct netpoll *np)
721 {
722 	struct net *net = current->nsproxy->net_ns;
723 	char buf[MAC_ADDR_STR_LEN + 1];
724 	struct net_device *ndev = NULL;
725 	bool ip_overwritten = false;
726 	struct in_device *in_dev;
727 	int err;
728 
729 	rtnl_lock();
730 	if (np->dev_name[0])
731 		ndev = __dev_get_by_name(net, np->dev_name);
732 	else if (is_valid_ether_addr(np->dev_mac))
733 		ndev = dev_getbyhwaddr(net, ARPHRD_ETHER, np->dev_mac);
734 
735 	if (!ndev) {
736 		np_err(np, "%s doesn't exist, aborting\n", egress_dev(np, buf));
737 		err = -ENODEV;
738 		goto unlock;
739 	}
740 	netdev_hold(ndev, &np->dev_tracker, GFP_KERNEL);
741 
742 	if (netdev_master_upper_dev_get(ndev)) {
743 		np_err(np, "%s is a slave device, aborting\n",
744 		       egress_dev(np, buf));
745 		err = -EBUSY;
746 		goto put;
747 	}
748 
749 	if (!netif_running(ndev)) {
750 		unsigned long atmost;
751 
752 		np_info(np, "device %s not up yet, forcing it\n",
753 			egress_dev(np, buf));
754 
755 		err = dev_open(ndev, NULL);
756 
757 		if (err) {
758 			np_err(np, "failed to open %s\n", ndev->name);
759 			goto put;
760 		}
761 
762 		rtnl_unlock();
763 		atmost = jiffies + carrier_timeout * HZ;
764 		while (!netif_carrier_ok(ndev)) {
765 			if (time_after(jiffies, atmost)) {
766 				np_notice(np, "timeout waiting for carrier\n");
767 				break;
768 			}
769 			msleep(1);
770 		}
771 
772 		rtnl_lock();
773 	}
774 
775 	if (!np->local_ip.ip) {
776 		if (!np->ipv6) {
777 			const struct in_ifaddr *ifa;
778 
779 			in_dev = __in_dev_get_rtnl(ndev);
780 			if (!in_dev)
781 				goto put_noaddr;
782 
783 			ifa = rtnl_dereference(in_dev->ifa_list);
784 			if (!ifa) {
785 put_noaddr:
786 				np_err(np, "no IP address for %s, aborting\n",
787 				       egress_dev(np, buf));
788 				err = -EDESTADDRREQ;
789 				goto put;
790 			}
791 
792 			np->local_ip.ip = ifa->ifa_local;
793 			ip_overwritten = true;
794 			np_info(np, "local IP %pI4\n", &np->local_ip.ip);
795 		} else {
796 #if IS_ENABLED(CONFIG_IPV6)
797 			struct inet6_dev *idev;
798 
799 			err = -EDESTADDRREQ;
800 			idev = __in6_dev_get(ndev);
801 			if (idev) {
802 				struct inet6_ifaddr *ifp;
803 
804 				read_lock_bh(&idev->lock);
805 				list_for_each_entry(ifp, &idev->addr_list, if_list) {
806 					if (!!(ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL) !=
807 					    !!(ipv6_addr_type(&np->remote_ip.in6) & IPV6_ADDR_LINKLOCAL))
808 						continue;
809 					np->local_ip.in6 = ifp->addr;
810 					ip_overwritten = true;
811 					err = 0;
812 					break;
813 				}
814 				read_unlock_bh(&idev->lock);
815 			}
816 			if (err) {
817 				np_err(np, "no IPv6 address for %s, aborting\n",
818 				       egress_dev(np, buf));
819 				goto put;
820 			} else
821 				np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
822 #else
823 			np_err(np, "IPv6 is not supported %s, aborting\n",
824 			       egress_dev(np, buf));
825 			err = -EINVAL;
826 			goto put;
827 #endif
828 		}
829 	}
830 
831 	err = __netpoll_setup(np, ndev);
832 	if (err)
833 		goto flush;
834 	rtnl_unlock();
835 	return 0;
836 
837 flush:
838 	skb_pool_flush(np);
839 put:
840 	DEBUG_NET_WARN_ON_ONCE(np->dev);
841 	if (ip_overwritten)
842 		memset(&np->local_ip, 0, sizeof(np->local_ip));
843 	netdev_put(ndev, &np->dev_tracker);
844 unlock:
845 	rtnl_unlock();
846 	return err;
847 }
848 EXPORT_SYMBOL(netpoll_setup);
849 
850 static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
851 {
852 	struct netpoll_info *npinfo =
853 			container_of(rcu_head, struct netpoll_info, rcu);
854 
855 	skb_queue_purge(&npinfo->txq);
856 
857 	/* we can't call cancel_delayed_work_sync here, as we are in softirq */
858 	cancel_delayed_work(&npinfo->tx_work);
859 
860 	/* clean after last, unfinished work */
861 	__skb_queue_purge(&npinfo->txq);
862 	/* now cancel it again */
863 	cancel_delayed_work(&npinfo->tx_work);
864 	kfree(npinfo);
865 }
866 
867 void __netpoll_cleanup(struct netpoll *np)
868 {
869 	struct netpoll_info *npinfo;
870 
871 	npinfo = rtnl_dereference(np->dev->npinfo);
872 	if (!npinfo)
873 		return;
874 
875 	if (refcount_dec_and_test(&npinfo->refcnt)) {
876 		const struct net_device_ops *ops;
877 
878 		ops = np->dev->netdev_ops;
879 		if (ops->ndo_netpoll_cleanup)
880 			ops->ndo_netpoll_cleanup(np->dev);
881 
882 		RCU_INIT_POINTER(np->dev->npinfo, NULL);
883 		call_rcu(&npinfo->rcu, rcu_cleanup_netpoll_info);
884 	} else
885 		RCU_INIT_POINTER(np->dev->npinfo, NULL);
886 
887 	skb_pool_flush(np);
888 }
889 EXPORT_SYMBOL_GPL(__netpoll_cleanup);
890 
891 void __netpoll_free(struct netpoll *np)
892 {
893 	ASSERT_RTNL();
894 
895 	/* Wait for transmitting packets to finish before freeing. */
896 	synchronize_rcu();
897 	__netpoll_cleanup(np);
898 	kfree(np);
899 }
900 EXPORT_SYMBOL_GPL(__netpoll_free);
901 
902 void do_netpoll_cleanup(struct netpoll *np)
903 {
904 	__netpoll_cleanup(np);
905 	netdev_put(np->dev, &np->dev_tracker);
906 	np->dev = NULL;
907 }
908 EXPORT_SYMBOL(do_netpoll_cleanup);
909 
910 void netpoll_cleanup(struct netpoll *np)
911 {
912 	rtnl_lock();
913 	if (!np->dev)
914 		goto out;
915 	do_netpoll_cleanup(np);
916 out:
917 	rtnl_unlock();
918 }
919 EXPORT_SYMBOL(netpoll_cleanup);
920