xref: /linux/net/core/netpoll.c (revision 026dadad6b44f0469a475efb4cae48269d8848bd)
1 /*
2  * Common framework for low-level network console, dump, and debugger code
3  *
4  * Sep 8 2003  Matt Mackall <mpm@selenic.com>
5  *
6  * based on the netconsole code from:
7  *
8  * Copyright (C) 2001  Ingo Molnar <mingo@redhat.com>
9  * Copyright (C) 2002  Red Hat, Inc.
10  */
11 
12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13 
14 #include <linux/moduleparam.h>
15 #include <linux/kernel.h>
16 #include <linux/netdevice.h>
17 #include <linux/etherdevice.h>
18 #include <linux/string.h>
19 #include <linux/if_arp.h>
20 #include <linux/inetdevice.h>
21 #include <linux/inet.h>
22 #include <linux/interrupt.h>
23 #include <linux/netpoll.h>
24 #include <linux/sched.h>
25 #include <linux/delay.h>
26 #include <linux/rcupdate.h>
27 #include <linux/workqueue.h>
28 #include <linux/slab.h>
29 #include <linux/export.h>
30 #include <linux/if_vlan.h>
31 #include <net/tcp.h>
32 #include <net/udp.h>
33 #include <net/addrconf.h>
34 #include <net/ndisc.h>
35 #include <net/ip6_checksum.h>
36 #include <asm/unaligned.h>
37 #include <trace/events/napi.h>
38 
39 /*
40  * We maintain a small pool of fully-sized skbs, to make sure the
41  * message gets out even in extreme OOM situations.
42  */
43 
44 #define MAX_UDP_CHUNK 1460
45 #define MAX_SKBS 32
46 
47 static struct sk_buff_head skb_pool;
48 
49 static atomic_t trapped;
50 
51 DEFINE_STATIC_SRCU(netpoll_srcu);
52 
53 #define USEC_PER_POLL	50
54 #define NETPOLL_RX_ENABLED  1
55 #define NETPOLL_RX_DROP     2
56 
57 #define MAX_SKB_SIZE							\
58 	(sizeof(struct ethhdr) +					\
59 	 sizeof(struct iphdr) +						\
60 	 sizeof(struct udphdr) +					\
61 	 MAX_UDP_CHUNK)
62 
63 static void zap_completion_queue(void);
64 static void netpoll_neigh_reply(struct sk_buff *skb, struct netpoll_info *npinfo);
65 static void netpoll_async_cleanup(struct work_struct *work);
66 
67 static unsigned int carrier_timeout = 4;
68 module_param(carrier_timeout, uint, 0644);
69 
70 #define np_info(np, fmt, ...)				\
71 	pr_info("%s: " fmt, np->name, ##__VA_ARGS__)
72 #define np_err(np, fmt, ...)				\
73 	pr_err("%s: " fmt, np->name, ##__VA_ARGS__)
74 #define np_notice(np, fmt, ...)				\
75 	pr_notice("%s: " fmt, np->name, ##__VA_ARGS__)
76 
77 static void queue_process(struct work_struct *work)
78 {
79 	struct netpoll_info *npinfo =
80 		container_of(work, struct netpoll_info, tx_work.work);
81 	struct sk_buff *skb;
82 	unsigned long flags;
83 
84 	while ((skb = skb_dequeue(&npinfo->txq))) {
85 		struct net_device *dev = skb->dev;
86 		const struct net_device_ops *ops = dev->netdev_ops;
87 		struct netdev_queue *txq;
88 
89 		if (!netif_device_present(dev) || !netif_running(dev)) {
90 			__kfree_skb(skb);
91 			continue;
92 		}
93 
94 		txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
95 
96 		local_irq_save(flags);
97 		__netif_tx_lock(txq, smp_processor_id());
98 		if (netif_xmit_frozen_or_stopped(txq) ||
99 		    ops->ndo_start_xmit(skb, dev) != NETDEV_TX_OK) {
100 			skb_queue_head(&npinfo->txq, skb);
101 			__netif_tx_unlock(txq);
102 			local_irq_restore(flags);
103 
104 			schedule_delayed_work(&npinfo->tx_work, HZ/10);
105 			return;
106 		}
107 		__netif_tx_unlock(txq);
108 		local_irq_restore(flags);
109 	}
110 }
111 
112 static __sum16 checksum_udp(struct sk_buff *skb, struct udphdr *uh,
113 			    unsigned short ulen, __be32 saddr, __be32 daddr)
114 {
115 	__wsum psum;
116 
117 	if (uh->check == 0 || skb_csum_unnecessary(skb))
118 		return 0;
119 
120 	psum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0);
121 
122 	if (skb->ip_summed == CHECKSUM_COMPLETE &&
123 	    !csum_fold(csum_add(psum, skb->csum)))
124 		return 0;
125 
126 	skb->csum = psum;
127 
128 	return __skb_checksum_complete(skb);
129 }
130 
131 /*
132  * Check whether delayed processing was scheduled for our NIC. If so,
133  * we attempt to grab the poll lock and use ->poll() to pump the card.
134  * If this fails, either we've recursed in ->poll() or it's already
135  * running on another CPU.
136  *
137  * Note: we don't mask interrupts with this lock because we're using
138  * trylock here and interrupts are already disabled in the softirq
139  * case. Further, we test the poll_owner to avoid recursion on UP
140  * systems where the lock doesn't exist.
141  *
142  * In cases where there is bi-directional communications, reading only
143  * one message at a time can lead to packets being dropped by the
144  * network adapter, forcing superfluous retries and possibly timeouts.
145  * Thus, we set our budget to greater than 1.
146  */
147 static int poll_one_napi(struct netpoll_info *npinfo,
148 			 struct napi_struct *napi, int budget)
149 {
150 	int work;
151 
152 	/* net_rx_action's ->poll() invocations and our's are
153 	 * synchronized by this test which is only made while
154 	 * holding the napi->poll_lock.
155 	 */
156 	if (!test_bit(NAPI_STATE_SCHED, &napi->state))
157 		return budget;
158 
159 	npinfo->rx_flags |= NETPOLL_RX_DROP;
160 	atomic_inc(&trapped);
161 	set_bit(NAPI_STATE_NPSVC, &napi->state);
162 
163 	work = napi->poll(napi, budget);
164 	trace_napi_poll(napi);
165 
166 	clear_bit(NAPI_STATE_NPSVC, &napi->state);
167 	atomic_dec(&trapped);
168 	npinfo->rx_flags &= ~NETPOLL_RX_DROP;
169 
170 	return budget - work;
171 }
172 
173 static void poll_napi(struct net_device *dev)
174 {
175 	struct napi_struct *napi;
176 	int budget = 16;
177 
178 	list_for_each_entry(napi, &dev->napi_list, dev_list) {
179 		if (napi->poll_owner != smp_processor_id() &&
180 		    spin_trylock(&napi->poll_lock)) {
181 			budget = poll_one_napi(rcu_dereference_bh(dev->npinfo),
182 					       napi, budget);
183 			spin_unlock(&napi->poll_lock);
184 
185 			if (!budget)
186 				break;
187 		}
188 	}
189 }
190 
191 static void service_neigh_queue(struct netpoll_info *npi)
192 {
193 	if (npi) {
194 		struct sk_buff *skb;
195 
196 		while ((skb = skb_dequeue(&npi->neigh_tx)))
197 			netpoll_neigh_reply(skb, npi);
198 	}
199 }
200 
201 static void netpoll_poll_dev(struct net_device *dev)
202 {
203 	const struct net_device_ops *ops;
204 	struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
205 
206 	/* Don't do any rx activity if the dev_lock mutex is held
207 	 * the dev_open/close paths use this to block netpoll activity
208 	 * while changing device state
209 	 */
210 	if (down_trylock(&ni->dev_lock))
211 		return;
212 
213 	if (!netif_running(dev)) {
214 		up(&ni->dev_lock);
215 		return;
216 	}
217 
218 	ops = dev->netdev_ops;
219 	if (!ops->ndo_poll_controller) {
220 		up(&ni->dev_lock);
221 		return;
222 	}
223 
224 	/* Process pending work on NIC */
225 	ops->ndo_poll_controller(dev);
226 
227 	poll_napi(dev);
228 
229 	up(&ni->dev_lock);
230 
231 	if (dev->flags & IFF_SLAVE) {
232 		if (ni) {
233 			struct net_device *bond_dev;
234 			struct sk_buff *skb;
235 			struct netpoll_info *bond_ni;
236 
237 			bond_dev = netdev_master_upper_dev_get_rcu(dev);
238 			bond_ni = rcu_dereference_bh(bond_dev->npinfo);
239 			while ((skb = skb_dequeue(&ni->neigh_tx))) {
240 				skb->dev = bond_dev;
241 				skb_queue_tail(&bond_ni->neigh_tx, skb);
242 			}
243 		}
244 	}
245 
246 	service_neigh_queue(ni);
247 
248 	zap_completion_queue();
249 }
250 
251 int netpoll_rx_disable(struct net_device *dev)
252 {
253 	struct netpoll_info *ni;
254 	int idx;
255 	might_sleep();
256 	idx = srcu_read_lock(&netpoll_srcu);
257 	ni = srcu_dereference(dev->npinfo, &netpoll_srcu);
258 	if (ni)
259 		down(&ni->dev_lock);
260 	srcu_read_unlock(&netpoll_srcu, idx);
261 	return 0;
262 }
263 EXPORT_SYMBOL(netpoll_rx_disable);
264 
265 void netpoll_rx_enable(struct net_device *dev)
266 {
267 	struct netpoll_info *ni;
268 	rcu_read_lock();
269 	ni = rcu_dereference(dev->npinfo);
270 	if (ni)
271 		up(&ni->dev_lock);
272 	rcu_read_unlock();
273 }
274 EXPORT_SYMBOL(netpoll_rx_enable);
275 
276 static void refill_skbs(void)
277 {
278 	struct sk_buff *skb;
279 	unsigned long flags;
280 
281 	spin_lock_irqsave(&skb_pool.lock, flags);
282 	while (skb_pool.qlen < MAX_SKBS) {
283 		skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
284 		if (!skb)
285 			break;
286 
287 		__skb_queue_tail(&skb_pool, skb);
288 	}
289 	spin_unlock_irqrestore(&skb_pool.lock, flags);
290 }
291 
292 static void zap_completion_queue(void)
293 {
294 	unsigned long flags;
295 	struct softnet_data *sd = &get_cpu_var(softnet_data);
296 
297 	if (sd->completion_queue) {
298 		struct sk_buff *clist;
299 
300 		local_irq_save(flags);
301 		clist = sd->completion_queue;
302 		sd->completion_queue = NULL;
303 		local_irq_restore(flags);
304 
305 		while (clist != NULL) {
306 			struct sk_buff *skb = clist;
307 			clist = clist->next;
308 			if (skb->destructor) {
309 				atomic_inc(&skb->users);
310 				dev_kfree_skb_any(skb); /* put this one back */
311 			} else {
312 				__kfree_skb(skb);
313 			}
314 		}
315 	}
316 
317 	put_cpu_var(softnet_data);
318 }
319 
320 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
321 {
322 	int count = 0;
323 	struct sk_buff *skb;
324 
325 	zap_completion_queue();
326 	refill_skbs();
327 repeat:
328 
329 	skb = alloc_skb(len, GFP_ATOMIC);
330 	if (!skb)
331 		skb = skb_dequeue(&skb_pool);
332 
333 	if (!skb) {
334 		if (++count < 10) {
335 			netpoll_poll_dev(np->dev);
336 			goto repeat;
337 		}
338 		return NULL;
339 	}
340 
341 	atomic_set(&skb->users, 1);
342 	skb_reserve(skb, reserve);
343 	return skb;
344 }
345 
346 static int netpoll_owner_active(struct net_device *dev)
347 {
348 	struct napi_struct *napi;
349 
350 	list_for_each_entry(napi, &dev->napi_list, dev_list) {
351 		if (napi->poll_owner == smp_processor_id())
352 			return 1;
353 	}
354 	return 0;
355 }
356 
357 /* call with IRQ disabled */
358 void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb,
359 			     struct net_device *dev)
360 {
361 	int status = NETDEV_TX_BUSY;
362 	unsigned long tries;
363 	const struct net_device_ops *ops = dev->netdev_ops;
364 	/* It is up to the caller to keep npinfo alive. */
365 	struct netpoll_info *npinfo;
366 
367 	WARN_ON_ONCE(!irqs_disabled());
368 
369 	npinfo = rcu_dereference_bh(np->dev->npinfo);
370 	if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
371 		__kfree_skb(skb);
372 		return;
373 	}
374 
375 	/* don't get messages out of order, and no recursion */
376 	if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
377 		struct netdev_queue *txq;
378 
379 		txq = netdev_pick_tx(dev, skb);
380 
381 		/* try until next clock tick */
382 		for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
383 		     tries > 0; --tries) {
384 			if (__netif_tx_trylock(txq)) {
385 				if (!netif_xmit_stopped(txq)) {
386 					if (vlan_tx_tag_present(skb) &&
387 					    !vlan_hw_offload_capable(netif_skb_features(skb),
388 								     skb->vlan_proto)) {
389 						skb = __vlan_put_tag(skb, skb->vlan_proto, vlan_tx_tag_get(skb));
390 						if (unlikely(!skb))
391 							break;
392 						skb->vlan_tci = 0;
393 					}
394 
395 					status = ops->ndo_start_xmit(skb, dev);
396 					if (status == NETDEV_TX_OK)
397 						txq_trans_update(txq);
398 				}
399 				__netif_tx_unlock(txq);
400 
401 				if (status == NETDEV_TX_OK)
402 					break;
403 
404 			}
405 
406 			/* tickle device maybe there is some cleanup */
407 			netpoll_poll_dev(np->dev);
408 
409 			udelay(USEC_PER_POLL);
410 		}
411 
412 		WARN_ONCE(!irqs_disabled(),
413 			"netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pF)\n",
414 			dev->name, ops->ndo_start_xmit);
415 
416 	}
417 
418 	if (status != NETDEV_TX_OK) {
419 		skb_queue_tail(&npinfo->txq, skb);
420 		schedule_delayed_work(&npinfo->tx_work,0);
421 	}
422 }
423 EXPORT_SYMBOL(netpoll_send_skb_on_dev);
424 
425 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
426 {
427 	int total_len, ip_len, udp_len;
428 	struct sk_buff *skb;
429 	struct udphdr *udph;
430 	struct iphdr *iph;
431 	struct ethhdr *eth;
432 	static atomic_t ip_ident;
433 	struct ipv6hdr *ip6h;
434 
435 	udp_len = len + sizeof(*udph);
436 	if (np->ipv6)
437 		ip_len = udp_len + sizeof(*ip6h);
438 	else
439 		ip_len = udp_len + sizeof(*iph);
440 
441 	total_len = ip_len + LL_RESERVED_SPACE(np->dev);
442 
443 	skb = find_skb(np, total_len + np->dev->needed_tailroom,
444 		       total_len - len);
445 	if (!skb)
446 		return;
447 
448 	skb_copy_to_linear_data(skb, msg, len);
449 	skb_put(skb, len);
450 
451 	skb_push(skb, sizeof(*udph));
452 	skb_reset_transport_header(skb);
453 	udph = udp_hdr(skb);
454 	udph->source = htons(np->local_port);
455 	udph->dest = htons(np->remote_port);
456 	udph->len = htons(udp_len);
457 
458 	if (np->ipv6) {
459 		udph->check = 0;
460 		udph->check = csum_ipv6_magic(&np->local_ip.in6,
461 					      &np->remote_ip.in6,
462 					      udp_len, IPPROTO_UDP,
463 					      csum_partial(udph, udp_len, 0));
464 		if (udph->check == 0)
465 			udph->check = CSUM_MANGLED_0;
466 
467 		skb_push(skb, sizeof(*ip6h));
468 		skb_reset_network_header(skb);
469 		ip6h = ipv6_hdr(skb);
470 
471 		/* ip6h->version = 6; ip6h->priority = 0; */
472 		put_unaligned(0x60, (unsigned char *)ip6h);
473 		ip6h->flow_lbl[0] = 0;
474 		ip6h->flow_lbl[1] = 0;
475 		ip6h->flow_lbl[2] = 0;
476 
477 		ip6h->payload_len = htons(sizeof(struct udphdr) + len);
478 		ip6h->nexthdr = IPPROTO_UDP;
479 		ip6h->hop_limit = 32;
480 		ip6h->saddr = np->local_ip.in6;
481 		ip6h->daddr = np->remote_ip.in6;
482 
483 		eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
484 		skb_reset_mac_header(skb);
485 		skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
486 	} else {
487 		udph->check = 0;
488 		udph->check = csum_tcpudp_magic(np->local_ip.ip,
489 						np->remote_ip.ip,
490 						udp_len, IPPROTO_UDP,
491 						csum_partial(udph, udp_len, 0));
492 		if (udph->check == 0)
493 			udph->check = CSUM_MANGLED_0;
494 
495 		skb_push(skb, sizeof(*iph));
496 		skb_reset_network_header(skb);
497 		iph = ip_hdr(skb);
498 
499 		/* iph->version = 4; iph->ihl = 5; */
500 		put_unaligned(0x45, (unsigned char *)iph);
501 		iph->tos      = 0;
502 		put_unaligned(htons(ip_len), &(iph->tot_len));
503 		iph->id       = htons(atomic_inc_return(&ip_ident));
504 		iph->frag_off = 0;
505 		iph->ttl      = 64;
506 		iph->protocol = IPPROTO_UDP;
507 		iph->check    = 0;
508 		put_unaligned(np->local_ip.ip, &(iph->saddr));
509 		put_unaligned(np->remote_ip.ip, &(iph->daddr));
510 		iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
511 
512 		eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
513 		skb_reset_mac_header(skb);
514 		skb->protocol = eth->h_proto = htons(ETH_P_IP);
515 	}
516 
517 	memcpy(eth->h_source, np->dev->dev_addr, ETH_ALEN);
518 	memcpy(eth->h_dest, np->remote_mac, ETH_ALEN);
519 
520 	skb->dev = np->dev;
521 
522 	netpoll_send_skb(np, skb);
523 }
524 EXPORT_SYMBOL(netpoll_send_udp);
525 
526 static void netpoll_neigh_reply(struct sk_buff *skb, struct netpoll_info *npinfo)
527 {
528 	int size, type = ARPOP_REPLY;
529 	__be32 sip, tip;
530 	unsigned char *sha;
531 	struct sk_buff *send_skb;
532 	struct netpoll *np, *tmp;
533 	unsigned long flags;
534 	int hlen, tlen;
535 	int hits = 0, proto;
536 
537 	if (list_empty(&npinfo->rx_np))
538 		return;
539 
540 	/* Before checking the packet, we do some early
541 	   inspection whether this is interesting at all */
542 	spin_lock_irqsave(&npinfo->rx_lock, flags);
543 	list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
544 		if (np->dev == skb->dev)
545 			hits++;
546 	}
547 	spin_unlock_irqrestore(&npinfo->rx_lock, flags);
548 
549 	/* No netpoll struct is using this dev */
550 	if (!hits)
551 		return;
552 
553 	proto = ntohs(eth_hdr(skb)->h_proto);
554 	if (proto == ETH_P_IP) {
555 		struct arphdr *arp;
556 		unsigned char *arp_ptr;
557 		/* No arp on this interface */
558 		if (skb->dev->flags & IFF_NOARP)
559 			return;
560 
561 		if (!pskb_may_pull(skb, arp_hdr_len(skb->dev)))
562 			return;
563 
564 		skb_reset_network_header(skb);
565 		skb_reset_transport_header(skb);
566 		arp = arp_hdr(skb);
567 
568 		if ((arp->ar_hrd != htons(ARPHRD_ETHER) &&
569 		     arp->ar_hrd != htons(ARPHRD_IEEE802)) ||
570 		    arp->ar_pro != htons(ETH_P_IP) ||
571 		    arp->ar_op != htons(ARPOP_REQUEST))
572 			return;
573 
574 		arp_ptr = (unsigned char *)(arp+1);
575 		/* save the location of the src hw addr */
576 		sha = arp_ptr;
577 		arp_ptr += skb->dev->addr_len;
578 		memcpy(&sip, arp_ptr, 4);
579 		arp_ptr += 4;
580 		/* If we actually cared about dst hw addr,
581 		   it would get copied here */
582 		arp_ptr += skb->dev->addr_len;
583 		memcpy(&tip, arp_ptr, 4);
584 
585 		/* Should we ignore arp? */
586 		if (ipv4_is_loopback(tip) || ipv4_is_multicast(tip))
587 			return;
588 
589 		size = arp_hdr_len(skb->dev);
590 
591 		spin_lock_irqsave(&npinfo->rx_lock, flags);
592 		list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
593 			if (tip != np->local_ip.ip)
594 				continue;
595 
596 			hlen = LL_RESERVED_SPACE(np->dev);
597 			tlen = np->dev->needed_tailroom;
598 			send_skb = find_skb(np, size + hlen + tlen, hlen);
599 			if (!send_skb)
600 				continue;
601 
602 			skb_reset_network_header(send_skb);
603 			arp = (struct arphdr *) skb_put(send_skb, size);
604 			send_skb->dev = skb->dev;
605 			send_skb->protocol = htons(ETH_P_ARP);
606 
607 			/* Fill the device header for the ARP frame */
608 			if (dev_hard_header(send_skb, skb->dev, ETH_P_ARP,
609 					    sha, np->dev->dev_addr,
610 					    send_skb->len) < 0) {
611 				kfree_skb(send_skb);
612 				continue;
613 			}
614 
615 			/*
616 			 * Fill out the arp protocol part.
617 			 *
618 			 * we only support ethernet device type,
619 			 * which (according to RFC 1390) should
620 			 * always equal 1 (Ethernet).
621 			 */
622 
623 			arp->ar_hrd = htons(np->dev->type);
624 			arp->ar_pro = htons(ETH_P_IP);
625 			arp->ar_hln = np->dev->addr_len;
626 			arp->ar_pln = 4;
627 			arp->ar_op = htons(type);
628 
629 			arp_ptr = (unsigned char *)(arp + 1);
630 			memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
631 			arp_ptr += np->dev->addr_len;
632 			memcpy(arp_ptr, &tip, 4);
633 			arp_ptr += 4;
634 			memcpy(arp_ptr, sha, np->dev->addr_len);
635 			arp_ptr += np->dev->addr_len;
636 			memcpy(arp_ptr, &sip, 4);
637 
638 			netpoll_send_skb(np, send_skb);
639 
640 			/* If there are several rx_hooks for the same address,
641 			   we're fine by sending a single reply */
642 			break;
643 		}
644 		spin_unlock_irqrestore(&npinfo->rx_lock, flags);
645 	} else if( proto == ETH_P_IPV6) {
646 #if IS_ENABLED(CONFIG_IPV6)
647 		struct nd_msg *msg;
648 		u8 *lladdr = NULL;
649 		struct ipv6hdr *hdr;
650 		struct icmp6hdr *icmp6h;
651 		const struct in6_addr *saddr;
652 		const struct in6_addr *daddr;
653 		struct inet6_dev *in6_dev = NULL;
654 		struct in6_addr *target;
655 
656 		in6_dev = in6_dev_get(skb->dev);
657 		if (!in6_dev || !in6_dev->cnf.accept_ra)
658 			return;
659 
660 		if (!pskb_may_pull(skb, skb->len))
661 			return;
662 
663 		msg = (struct nd_msg *)skb_transport_header(skb);
664 
665 		__skb_push(skb, skb->data - skb_transport_header(skb));
666 
667 		if (ipv6_hdr(skb)->hop_limit != 255)
668 			return;
669 		if (msg->icmph.icmp6_code != 0)
670 			return;
671 		if (msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
672 			return;
673 
674 		saddr = &ipv6_hdr(skb)->saddr;
675 		daddr = &ipv6_hdr(skb)->daddr;
676 
677 		size = sizeof(struct icmp6hdr) + sizeof(struct in6_addr);
678 
679 		spin_lock_irqsave(&npinfo->rx_lock, flags);
680 		list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
681 			if (!ipv6_addr_equal(daddr, &np->local_ip.in6))
682 				continue;
683 
684 			hlen = LL_RESERVED_SPACE(np->dev);
685 			tlen = np->dev->needed_tailroom;
686 			send_skb = find_skb(np, size + hlen + tlen, hlen);
687 			if (!send_skb)
688 				continue;
689 
690 			send_skb->protocol = htons(ETH_P_IPV6);
691 			send_skb->dev = skb->dev;
692 
693 			skb_reset_network_header(send_skb);
694 			skb_put(send_skb, sizeof(struct ipv6hdr));
695 			hdr = ipv6_hdr(send_skb);
696 
697 			*(__be32*)hdr = htonl(0x60000000);
698 
699 			hdr->payload_len = htons(size);
700 			hdr->nexthdr = IPPROTO_ICMPV6;
701 			hdr->hop_limit = 255;
702 			hdr->saddr = *saddr;
703 			hdr->daddr = *daddr;
704 
705 			send_skb->transport_header = send_skb->tail;
706 			skb_put(send_skb, size);
707 
708 			icmp6h = (struct icmp6hdr *)skb_transport_header(skb);
709 			icmp6h->icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
710 			icmp6h->icmp6_router = 0;
711 			icmp6h->icmp6_solicited = 1;
712 			target = (struct in6_addr *)(skb_transport_header(send_skb) + sizeof(struct icmp6hdr));
713 			*target = msg->target;
714 			icmp6h->icmp6_cksum = csum_ipv6_magic(saddr, daddr, size,
715 							      IPPROTO_ICMPV6,
716 							      csum_partial(icmp6h,
717 									   size, 0));
718 
719 			if (dev_hard_header(send_skb, skb->dev, ETH_P_IPV6,
720 					    lladdr, np->dev->dev_addr,
721 					    send_skb->len) < 0) {
722 				kfree_skb(send_skb);
723 				continue;
724 			}
725 
726 			netpoll_send_skb(np, send_skb);
727 
728 			/* If there are several rx_hooks for the same address,
729 			   we're fine by sending a single reply */
730 			break;
731 		}
732 		spin_unlock_irqrestore(&npinfo->rx_lock, flags);
733 #endif
734 	}
735 }
736 
737 static bool pkt_is_ns(struct sk_buff *skb)
738 {
739 	struct nd_msg *msg;
740 	struct ipv6hdr *hdr;
741 
742 	if (skb->protocol != htons(ETH_P_ARP))
743 		return false;
744 	if (!pskb_may_pull(skb, sizeof(struct ipv6hdr) + sizeof(struct nd_msg)))
745 		return false;
746 
747 	msg = (struct nd_msg *)skb_transport_header(skb);
748 	__skb_push(skb, skb->data - skb_transport_header(skb));
749 	hdr = ipv6_hdr(skb);
750 
751 	if (hdr->nexthdr != IPPROTO_ICMPV6)
752 		return false;
753 	if (hdr->hop_limit != 255)
754 		return false;
755 	if (msg->icmph.icmp6_code != 0)
756 		return false;
757 	if (msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
758 		return false;
759 
760 	return true;
761 }
762 
763 int __netpoll_rx(struct sk_buff *skb, struct netpoll_info *npinfo)
764 {
765 	int proto, len, ulen;
766 	int hits = 0;
767 	const struct iphdr *iph;
768 	struct udphdr *uh;
769 	struct netpoll *np, *tmp;
770 
771 	if (list_empty(&npinfo->rx_np))
772 		goto out;
773 
774 	if (skb->dev->type != ARPHRD_ETHER)
775 		goto out;
776 
777 	/* check if netpoll clients need ARP */
778 	if (skb->protocol == htons(ETH_P_ARP) && atomic_read(&trapped)) {
779 		skb_queue_tail(&npinfo->neigh_tx, skb);
780 		return 1;
781 	} else if (pkt_is_ns(skb) && atomic_read(&trapped)) {
782 		skb_queue_tail(&npinfo->neigh_tx, skb);
783 		return 1;
784 	}
785 
786 	if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
787 		skb = vlan_untag(skb);
788 		if (unlikely(!skb))
789 			goto out;
790 	}
791 
792 	proto = ntohs(eth_hdr(skb)->h_proto);
793 	if (proto != ETH_P_IP && proto != ETH_P_IPV6)
794 		goto out;
795 	if (skb->pkt_type == PACKET_OTHERHOST)
796 		goto out;
797 	if (skb_shared(skb))
798 		goto out;
799 
800 	if (proto == ETH_P_IP) {
801 		if (!pskb_may_pull(skb, sizeof(struct iphdr)))
802 			goto out;
803 		iph = (struct iphdr *)skb->data;
804 		if (iph->ihl < 5 || iph->version != 4)
805 			goto out;
806 		if (!pskb_may_pull(skb, iph->ihl*4))
807 			goto out;
808 		iph = (struct iphdr *)skb->data;
809 		if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
810 			goto out;
811 
812 		len = ntohs(iph->tot_len);
813 		if (skb->len < len || len < iph->ihl*4)
814 			goto out;
815 
816 		/*
817 		 * Our transport medium may have padded the buffer out.
818 		 * Now We trim to the true length of the frame.
819 		 */
820 		if (pskb_trim_rcsum(skb, len))
821 			goto out;
822 
823 		iph = (struct iphdr *)skb->data;
824 		if (iph->protocol != IPPROTO_UDP)
825 			goto out;
826 
827 		len -= iph->ihl*4;
828 		uh = (struct udphdr *)(((char *)iph) + iph->ihl*4);
829 		ulen = ntohs(uh->len);
830 
831 		if (ulen != len)
832 			goto out;
833 		if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr))
834 			goto out;
835 		list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
836 			if (np->local_ip.ip && np->local_ip.ip != iph->daddr)
837 				continue;
838 			if (np->remote_ip.ip && np->remote_ip.ip != iph->saddr)
839 				continue;
840 			if (np->local_port && np->local_port != ntohs(uh->dest))
841 				continue;
842 
843 			np->rx_hook(np, ntohs(uh->source),
844 				       (char *)(uh+1),
845 				       ulen - sizeof(struct udphdr));
846 			hits++;
847 		}
848 	} else {
849 #if IS_ENABLED(CONFIG_IPV6)
850 		const struct ipv6hdr *ip6h;
851 
852 		if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
853 			goto out;
854 		ip6h = (struct ipv6hdr *)skb->data;
855 		if (ip6h->version != 6)
856 			goto out;
857 		len = ntohs(ip6h->payload_len);
858 		if (!len)
859 			goto out;
860 		if (len + sizeof(struct ipv6hdr) > skb->len)
861 			goto out;
862 		if (pskb_trim_rcsum(skb, len + sizeof(struct ipv6hdr)))
863 			goto out;
864 		ip6h = ipv6_hdr(skb);
865 		if (!pskb_may_pull(skb, sizeof(struct udphdr)))
866 			goto out;
867 		uh = udp_hdr(skb);
868 		ulen = ntohs(uh->len);
869 		if (ulen != skb->len)
870 			goto out;
871 		if (udp6_csum_init(skb, uh, IPPROTO_UDP))
872 			goto out;
873 		list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
874 			if (!ipv6_addr_equal(&np->local_ip.in6, &ip6h->daddr))
875 				continue;
876 			if (!ipv6_addr_equal(&np->remote_ip.in6, &ip6h->saddr))
877 				continue;
878 			if (np->local_port && np->local_port != ntohs(uh->dest))
879 				continue;
880 
881 			np->rx_hook(np, ntohs(uh->source),
882 				       (char *)(uh+1),
883 				       ulen - sizeof(struct udphdr));
884 			hits++;
885 		}
886 #endif
887 	}
888 
889 	if (!hits)
890 		goto out;
891 
892 	kfree_skb(skb);
893 	return 1;
894 
895 out:
896 	if (atomic_read(&trapped)) {
897 		kfree_skb(skb);
898 		return 1;
899 	}
900 
901 	return 0;
902 }
903 
904 void netpoll_print_options(struct netpoll *np)
905 {
906 	np_info(np, "local port %d\n", np->local_port);
907 	if (np->ipv6)
908 		np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
909 	else
910 		np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
911 	np_info(np, "interface '%s'\n", np->dev_name);
912 	np_info(np, "remote port %d\n", np->remote_port);
913 	if (np->ipv6)
914 		np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
915 	else
916 		np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
917 	np_info(np, "remote ethernet address %pM\n", np->remote_mac);
918 }
919 EXPORT_SYMBOL(netpoll_print_options);
920 
921 static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
922 {
923 	const char *end;
924 
925 	if (!strchr(str, ':') &&
926 	    in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
927 		if (!*end)
928 			return 0;
929 	}
930 	if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
931 #if IS_ENABLED(CONFIG_IPV6)
932 		if (!*end)
933 			return 1;
934 #else
935 		return -1;
936 #endif
937 	}
938 	return -1;
939 }
940 
941 int netpoll_parse_options(struct netpoll *np, char *opt)
942 {
943 	char *cur=opt, *delim;
944 	int ipv6;
945 
946 	if (*cur != '@') {
947 		if ((delim = strchr(cur, '@')) == NULL)
948 			goto parse_failed;
949 		*delim = 0;
950 		if (kstrtou16(cur, 10, &np->local_port))
951 			goto parse_failed;
952 		cur = delim;
953 	}
954 	cur++;
955 
956 	if (*cur != '/') {
957 		if ((delim = strchr(cur, '/')) == NULL)
958 			goto parse_failed;
959 		*delim = 0;
960 		ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
961 		if (ipv6 < 0)
962 			goto parse_failed;
963 		else
964 			np->ipv6 = (bool)ipv6;
965 		cur = delim;
966 	}
967 	cur++;
968 
969 	if (*cur != ',') {
970 		/* parse out dev name */
971 		if ((delim = strchr(cur, ',')) == NULL)
972 			goto parse_failed;
973 		*delim = 0;
974 		strlcpy(np->dev_name, cur, sizeof(np->dev_name));
975 		cur = delim;
976 	}
977 	cur++;
978 
979 	if (*cur != '@') {
980 		/* dst port */
981 		if ((delim = strchr(cur, '@')) == NULL)
982 			goto parse_failed;
983 		*delim = 0;
984 		if (*cur == ' ' || *cur == '\t')
985 			np_info(np, "warning: whitespace is not allowed\n");
986 		if (kstrtou16(cur, 10, &np->remote_port))
987 			goto parse_failed;
988 		cur = delim;
989 	}
990 	cur++;
991 
992 	/* dst ip */
993 	if ((delim = strchr(cur, '/')) == NULL)
994 		goto parse_failed;
995 	*delim = 0;
996 	ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
997 	if (ipv6 < 0)
998 		goto parse_failed;
999 	else if (np->ipv6 != (bool)ipv6)
1000 		goto parse_failed;
1001 	else
1002 		np->ipv6 = (bool)ipv6;
1003 	cur = delim + 1;
1004 
1005 	if (*cur != 0) {
1006 		/* MAC address */
1007 		if (!mac_pton(cur, np->remote_mac))
1008 			goto parse_failed;
1009 	}
1010 
1011 	netpoll_print_options(np);
1012 
1013 	return 0;
1014 
1015  parse_failed:
1016 	np_info(np, "couldn't parse config at '%s'!\n", cur);
1017 	return -1;
1018 }
1019 EXPORT_SYMBOL(netpoll_parse_options);
1020 
1021 int __netpoll_setup(struct netpoll *np, struct net_device *ndev, gfp_t gfp)
1022 {
1023 	struct netpoll_info *npinfo;
1024 	const struct net_device_ops *ops;
1025 	unsigned long flags;
1026 	int err;
1027 
1028 	np->dev = ndev;
1029 	strlcpy(np->dev_name, ndev->name, IFNAMSIZ);
1030 	INIT_WORK(&np->cleanup_work, netpoll_async_cleanup);
1031 
1032 	if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) ||
1033 	    !ndev->netdev_ops->ndo_poll_controller) {
1034 		np_err(np, "%s doesn't support polling, aborting\n",
1035 		       np->dev_name);
1036 		err = -ENOTSUPP;
1037 		goto out;
1038 	}
1039 
1040 	if (!ndev->npinfo) {
1041 		npinfo = kmalloc(sizeof(*npinfo), gfp);
1042 		if (!npinfo) {
1043 			err = -ENOMEM;
1044 			goto out;
1045 		}
1046 
1047 		npinfo->rx_flags = 0;
1048 		INIT_LIST_HEAD(&npinfo->rx_np);
1049 
1050 		spin_lock_init(&npinfo->rx_lock);
1051 		sema_init(&npinfo->dev_lock, 1);
1052 		skb_queue_head_init(&npinfo->neigh_tx);
1053 		skb_queue_head_init(&npinfo->txq);
1054 		INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
1055 
1056 		atomic_set(&npinfo->refcnt, 1);
1057 
1058 		ops = np->dev->netdev_ops;
1059 		if (ops->ndo_netpoll_setup) {
1060 			err = ops->ndo_netpoll_setup(ndev, npinfo, gfp);
1061 			if (err)
1062 				goto free_npinfo;
1063 		}
1064 	} else {
1065 		npinfo = rtnl_dereference(ndev->npinfo);
1066 		atomic_inc(&npinfo->refcnt);
1067 	}
1068 
1069 	npinfo->netpoll = np;
1070 
1071 	if (np->rx_hook) {
1072 		spin_lock_irqsave(&npinfo->rx_lock, flags);
1073 		npinfo->rx_flags |= NETPOLL_RX_ENABLED;
1074 		list_add_tail(&np->rx, &npinfo->rx_np);
1075 		spin_unlock_irqrestore(&npinfo->rx_lock, flags);
1076 	}
1077 
1078 	/* last thing to do is link it to the net device structure */
1079 	rcu_assign_pointer(ndev->npinfo, npinfo);
1080 
1081 	return 0;
1082 
1083 free_npinfo:
1084 	kfree(npinfo);
1085 out:
1086 	return err;
1087 }
1088 EXPORT_SYMBOL_GPL(__netpoll_setup);
1089 
1090 int netpoll_setup(struct netpoll *np)
1091 {
1092 	struct net_device *ndev = NULL;
1093 	struct in_device *in_dev;
1094 	int err;
1095 
1096 	rtnl_lock();
1097 	if (np->dev_name) {
1098 		struct net *net = current->nsproxy->net_ns;
1099 		ndev = __dev_get_by_name(net, np->dev_name);
1100 	}
1101 	if (!ndev) {
1102 		np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
1103 		err = -ENODEV;
1104 		goto unlock;
1105 	}
1106 	dev_hold(ndev);
1107 
1108 	if (netdev_master_upper_dev_get(ndev)) {
1109 		np_err(np, "%s is a slave device, aborting\n", np->dev_name);
1110 		err = -EBUSY;
1111 		goto put;
1112 	}
1113 
1114 	if (!netif_running(ndev)) {
1115 		unsigned long atmost, atleast;
1116 
1117 		np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
1118 
1119 		err = dev_open(ndev);
1120 
1121 		if (err) {
1122 			np_err(np, "failed to open %s\n", ndev->name);
1123 			goto put;
1124 		}
1125 
1126 		rtnl_unlock();
1127 		atleast = jiffies + HZ/10;
1128 		atmost = jiffies + carrier_timeout * HZ;
1129 		while (!netif_carrier_ok(ndev)) {
1130 			if (time_after(jiffies, atmost)) {
1131 				np_notice(np, "timeout waiting for carrier\n");
1132 				break;
1133 			}
1134 			msleep(1);
1135 		}
1136 
1137 		/* If carrier appears to come up instantly, we don't
1138 		 * trust it and pause so that we don't pump all our
1139 		 * queued console messages into the bitbucket.
1140 		 */
1141 
1142 		if (time_before(jiffies, atleast)) {
1143 			np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
1144 			msleep(4000);
1145 		}
1146 		rtnl_lock();
1147 	}
1148 
1149 	if (!np->local_ip.ip) {
1150 		if (!np->ipv6) {
1151 			in_dev = __in_dev_get_rtnl(ndev);
1152 
1153 			if (!in_dev || !in_dev->ifa_list) {
1154 				np_err(np, "no IP address for %s, aborting\n",
1155 				       np->dev_name);
1156 				err = -EDESTADDRREQ;
1157 				goto put;
1158 			}
1159 
1160 			np->local_ip.ip = in_dev->ifa_list->ifa_local;
1161 			np_info(np, "local IP %pI4\n", &np->local_ip.ip);
1162 		} else {
1163 #if IS_ENABLED(CONFIG_IPV6)
1164 			struct inet6_dev *idev;
1165 
1166 			err = -EDESTADDRREQ;
1167 			idev = __in6_dev_get(ndev);
1168 			if (idev) {
1169 				struct inet6_ifaddr *ifp;
1170 
1171 				read_lock_bh(&idev->lock);
1172 				list_for_each_entry(ifp, &idev->addr_list, if_list) {
1173 					if (ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL)
1174 						continue;
1175 					np->local_ip.in6 = ifp->addr;
1176 					err = 0;
1177 					break;
1178 				}
1179 				read_unlock_bh(&idev->lock);
1180 			}
1181 			if (err) {
1182 				np_err(np, "no IPv6 address for %s, aborting\n",
1183 				       np->dev_name);
1184 				goto put;
1185 			} else
1186 				np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
1187 #else
1188 			np_err(np, "IPv6 is not supported %s, aborting\n",
1189 			       np->dev_name);
1190 			err = -EINVAL;
1191 			goto put;
1192 #endif
1193 		}
1194 	}
1195 
1196 	/* fill up the skb queue */
1197 	refill_skbs();
1198 
1199 	err = __netpoll_setup(np, ndev, GFP_KERNEL);
1200 	if (err)
1201 		goto put;
1202 
1203 	rtnl_unlock();
1204 	return 0;
1205 
1206 put:
1207 	dev_put(ndev);
1208 unlock:
1209 	rtnl_unlock();
1210 	return err;
1211 }
1212 EXPORT_SYMBOL(netpoll_setup);
1213 
1214 static int __init netpoll_init(void)
1215 {
1216 	skb_queue_head_init(&skb_pool);
1217 	return 0;
1218 }
1219 core_initcall(netpoll_init);
1220 
1221 static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
1222 {
1223 	struct netpoll_info *npinfo =
1224 			container_of(rcu_head, struct netpoll_info, rcu);
1225 
1226 	skb_queue_purge(&npinfo->neigh_tx);
1227 	skb_queue_purge(&npinfo->txq);
1228 
1229 	/* we can't call cancel_delayed_work_sync here, as we are in softirq */
1230 	cancel_delayed_work(&npinfo->tx_work);
1231 
1232 	/* clean after last, unfinished work */
1233 	__skb_queue_purge(&npinfo->txq);
1234 	/* now cancel it again */
1235 	cancel_delayed_work(&npinfo->tx_work);
1236 	kfree(npinfo);
1237 }
1238 
1239 void __netpoll_cleanup(struct netpoll *np)
1240 {
1241 	struct netpoll_info *npinfo;
1242 	unsigned long flags;
1243 
1244 	/* rtnl_dereference would be preferable here but
1245 	 * rcu_cleanup_netpoll path can put us in here safely without
1246 	 * holding the rtnl, so plain rcu_dereference it is
1247 	 */
1248 	npinfo = rtnl_dereference(np->dev->npinfo);
1249 	if (!npinfo)
1250 		return;
1251 
1252 	if (!list_empty(&npinfo->rx_np)) {
1253 		spin_lock_irqsave(&npinfo->rx_lock, flags);
1254 		list_del(&np->rx);
1255 		if (list_empty(&npinfo->rx_np))
1256 			npinfo->rx_flags &= ~NETPOLL_RX_ENABLED;
1257 		spin_unlock_irqrestore(&npinfo->rx_lock, flags);
1258 	}
1259 
1260 	synchronize_srcu(&netpoll_srcu);
1261 
1262 	if (atomic_dec_and_test(&npinfo->refcnt)) {
1263 		const struct net_device_ops *ops;
1264 
1265 		ops = np->dev->netdev_ops;
1266 		if (ops->ndo_netpoll_cleanup)
1267 			ops->ndo_netpoll_cleanup(np->dev);
1268 
1269 		rcu_assign_pointer(np->dev->npinfo, NULL);
1270 		call_rcu_bh(&npinfo->rcu, rcu_cleanup_netpoll_info);
1271 	}
1272 }
1273 EXPORT_SYMBOL_GPL(__netpoll_cleanup);
1274 
1275 static void netpoll_async_cleanup(struct work_struct *work)
1276 {
1277 	struct netpoll *np = container_of(work, struct netpoll, cleanup_work);
1278 
1279 	rtnl_lock();
1280 	__netpoll_cleanup(np);
1281 	rtnl_unlock();
1282 	kfree(np);
1283 }
1284 
1285 void __netpoll_free_async(struct netpoll *np)
1286 {
1287 	schedule_work(&np->cleanup_work);
1288 }
1289 EXPORT_SYMBOL_GPL(__netpoll_free_async);
1290 
1291 void netpoll_cleanup(struct netpoll *np)
1292 {
1293 	if (!np->dev)
1294 		return;
1295 
1296 	rtnl_lock();
1297 	__netpoll_cleanup(np);
1298 	rtnl_unlock();
1299 
1300 	dev_put(np->dev);
1301 	np->dev = NULL;
1302 }
1303 EXPORT_SYMBOL(netpoll_cleanup);
1304 
1305 int netpoll_trap(void)
1306 {
1307 	return atomic_read(&trapped);
1308 }
1309 EXPORT_SYMBOL(netpoll_trap);
1310 
1311 void netpoll_set_trap(int trap)
1312 {
1313 	if (trap)
1314 		atomic_inc(&trapped);
1315 	else
1316 		atomic_dec(&trapped);
1317 }
1318 EXPORT_SYMBOL(netpoll_set_trap);
1319