xref: /linux/net/core/netpoll.c (revision 47902f3611b392209e2a412bf7ec02dca95e666d)
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 #include <linux/moduleparam.h>
13 #include <linux/netdevice.h>
14 #include <linux/etherdevice.h>
15 #include <linux/string.h>
16 #include <linux/if_arp.h>
17 #include <linux/inetdevice.h>
18 #include <linux/inet.h>
19 #include <linux/interrupt.h>
20 #include <linux/netpoll.h>
21 #include <linux/sched.h>
22 #include <linux/delay.h>
23 #include <linux/rcupdate.h>
24 #include <linux/workqueue.h>
25 #include <linux/slab.h>
26 #include <net/tcp.h>
27 #include <net/udp.h>
28 #include <asm/unaligned.h>
29 #include <trace/events/napi.h>
30 
31 /*
32  * We maintain a small pool of fully-sized skbs, to make sure the
33  * message gets out even in extreme OOM situations.
34  */
35 
36 #define MAX_UDP_CHUNK 1460
37 #define MAX_SKBS 32
38 #define MAX_QUEUE_DEPTH (MAX_SKBS / 2)
39 
40 static struct sk_buff_head skb_pool;
41 
42 static atomic_t trapped;
43 
44 #define USEC_PER_POLL	50
45 #define NETPOLL_RX_ENABLED  1
46 #define NETPOLL_RX_DROP     2
47 
48 #define MAX_SKB_SIZE \
49 		(MAX_UDP_CHUNK + sizeof(struct udphdr) + \
50 				sizeof(struct iphdr) + sizeof(struct ethhdr))
51 
52 static void zap_completion_queue(void);
53 static void arp_reply(struct sk_buff *skb);
54 
55 static unsigned int carrier_timeout = 4;
56 module_param(carrier_timeout, uint, 0644);
57 
58 static void queue_process(struct work_struct *work)
59 {
60 	struct netpoll_info *npinfo =
61 		container_of(work, struct netpoll_info, tx_work.work);
62 	struct sk_buff *skb;
63 	unsigned long flags;
64 
65 	while ((skb = skb_dequeue(&npinfo->txq))) {
66 		struct net_device *dev = skb->dev;
67 		const struct net_device_ops *ops = dev->netdev_ops;
68 		struct netdev_queue *txq;
69 
70 		if (!netif_device_present(dev) || !netif_running(dev)) {
71 			__kfree_skb(skb);
72 			continue;
73 		}
74 
75 		txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
76 
77 		local_irq_save(flags);
78 		__netif_tx_lock(txq, smp_processor_id());
79 		if (netif_tx_queue_stopped(txq) ||
80 		    netif_tx_queue_frozen(txq) ||
81 		    ops->ndo_start_xmit(skb, dev) != NETDEV_TX_OK) {
82 			skb_queue_head(&npinfo->txq, skb);
83 			__netif_tx_unlock(txq);
84 			local_irq_restore(flags);
85 
86 			schedule_delayed_work(&npinfo->tx_work, HZ/10);
87 			return;
88 		}
89 		__netif_tx_unlock(txq);
90 		local_irq_restore(flags);
91 	}
92 }
93 
94 static __sum16 checksum_udp(struct sk_buff *skb, struct udphdr *uh,
95 			    unsigned short ulen, __be32 saddr, __be32 daddr)
96 {
97 	__wsum psum;
98 
99 	if (uh->check == 0 || skb_csum_unnecessary(skb))
100 		return 0;
101 
102 	psum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0);
103 
104 	if (skb->ip_summed == CHECKSUM_COMPLETE &&
105 	    !csum_fold(csum_add(psum, skb->csum)))
106 		return 0;
107 
108 	skb->csum = psum;
109 
110 	return __skb_checksum_complete(skb);
111 }
112 
113 /*
114  * Check whether delayed processing was scheduled for our NIC. If so,
115  * we attempt to grab the poll lock and use ->poll() to pump the card.
116  * If this fails, either we've recursed in ->poll() or it's already
117  * running on another CPU.
118  *
119  * Note: we don't mask interrupts with this lock because we're using
120  * trylock here and interrupts are already disabled in the softirq
121  * case. Further, we test the poll_owner to avoid recursion on UP
122  * systems where the lock doesn't exist.
123  *
124  * In cases where there is bi-directional communications, reading only
125  * one message at a time can lead to packets being dropped by the
126  * network adapter, forcing superfluous retries and possibly timeouts.
127  * Thus, we set our budget to greater than 1.
128  */
129 static int poll_one_napi(struct netpoll_info *npinfo,
130 			 struct napi_struct *napi, int budget)
131 {
132 	int work;
133 
134 	/* net_rx_action's ->poll() invocations and our's are
135 	 * synchronized by this test which is only made while
136 	 * holding the napi->poll_lock.
137 	 */
138 	if (!test_bit(NAPI_STATE_SCHED, &napi->state))
139 		return budget;
140 
141 	npinfo->rx_flags |= NETPOLL_RX_DROP;
142 	atomic_inc(&trapped);
143 	set_bit(NAPI_STATE_NPSVC, &napi->state);
144 
145 	work = napi->poll(napi, budget);
146 	trace_napi_poll(napi);
147 
148 	clear_bit(NAPI_STATE_NPSVC, &napi->state);
149 	atomic_dec(&trapped);
150 	npinfo->rx_flags &= ~NETPOLL_RX_DROP;
151 
152 	return budget - work;
153 }
154 
155 static void poll_napi(struct net_device *dev)
156 {
157 	struct napi_struct *napi;
158 	int budget = 16;
159 
160 	list_for_each_entry(napi, &dev->napi_list, dev_list) {
161 		if (napi->poll_owner != smp_processor_id() &&
162 		    spin_trylock(&napi->poll_lock)) {
163 			budget = poll_one_napi(dev->npinfo, napi, budget);
164 			spin_unlock(&napi->poll_lock);
165 
166 			if (!budget)
167 				break;
168 		}
169 	}
170 }
171 
172 static void service_arp_queue(struct netpoll_info *npi)
173 {
174 	if (npi) {
175 		struct sk_buff *skb;
176 
177 		while ((skb = skb_dequeue(&npi->arp_tx)))
178 			arp_reply(skb);
179 	}
180 }
181 
182 void netpoll_poll(struct netpoll *np)
183 {
184 	struct net_device *dev = np->dev;
185 	const struct net_device_ops *ops;
186 
187 	if (!dev || !netif_running(dev))
188 		return;
189 
190 	ops = dev->netdev_ops;
191 	if (!ops->ndo_poll_controller)
192 		return;
193 
194 	/* Process pending work on NIC */
195 	ops->ndo_poll_controller(dev);
196 
197 	poll_napi(dev);
198 
199 	service_arp_queue(dev->npinfo);
200 
201 	zap_completion_queue();
202 }
203 
204 static void refill_skbs(void)
205 {
206 	struct sk_buff *skb;
207 	unsigned long flags;
208 
209 	spin_lock_irqsave(&skb_pool.lock, flags);
210 	while (skb_pool.qlen < MAX_SKBS) {
211 		skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
212 		if (!skb)
213 			break;
214 
215 		__skb_queue_tail(&skb_pool, skb);
216 	}
217 	spin_unlock_irqrestore(&skb_pool.lock, flags);
218 }
219 
220 static void zap_completion_queue(void)
221 {
222 	unsigned long flags;
223 	struct softnet_data *sd = &get_cpu_var(softnet_data);
224 
225 	if (sd->completion_queue) {
226 		struct sk_buff *clist;
227 
228 		local_irq_save(flags);
229 		clist = sd->completion_queue;
230 		sd->completion_queue = NULL;
231 		local_irq_restore(flags);
232 
233 		while (clist != NULL) {
234 			struct sk_buff *skb = clist;
235 			clist = clist->next;
236 			if (skb->destructor) {
237 				atomic_inc(&skb->users);
238 				dev_kfree_skb_any(skb); /* put this one back */
239 			} else {
240 				__kfree_skb(skb);
241 			}
242 		}
243 	}
244 
245 	put_cpu_var(softnet_data);
246 }
247 
248 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
249 {
250 	int count = 0;
251 	struct sk_buff *skb;
252 
253 	zap_completion_queue();
254 	refill_skbs();
255 repeat:
256 
257 	skb = alloc_skb(len, GFP_ATOMIC);
258 	if (!skb)
259 		skb = skb_dequeue(&skb_pool);
260 
261 	if (!skb) {
262 		if (++count < 10) {
263 			netpoll_poll(np);
264 			goto repeat;
265 		}
266 		return NULL;
267 	}
268 
269 	atomic_set(&skb->users, 1);
270 	skb_reserve(skb, reserve);
271 	return skb;
272 }
273 
274 static int netpoll_owner_active(struct net_device *dev)
275 {
276 	struct napi_struct *napi;
277 
278 	list_for_each_entry(napi, &dev->napi_list, dev_list) {
279 		if (napi->poll_owner == smp_processor_id())
280 			return 1;
281 	}
282 	return 0;
283 }
284 
285 static void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
286 {
287 	int status = NETDEV_TX_BUSY;
288 	unsigned long tries;
289 	struct net_device *dev = np->dev;
290 	const struct net_device_ops *ops = dev->netdev_ops;
291 	struct netpoll_info *npinfo = np->dev->npinfo;
292 
293 	if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
294 		__kfree_skb(skb);
295 		return;
296 	}
297 
298 	/* don't get messages out of order, and no recursion */
299 	if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
300 		struct netdev_queue *txq;
301 		unsigned long flags;
302 
303 		txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
304 
305 		local_irq_save(flags);
306 		/* try until next clock tick */
307 		for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
308 		     tries > 0; --tries) {
309 			if (__netif_tx_trylock(txq)) {
310 				if (!netif_tx_queue_stopped(txq)) {
311 					status = ops->ndo_start_xmit(skb, dev);
312 					if (status == NETDEV_TX_OK)
313 						txq_trans_update(txq);
314 				}
315 				__netif_tx_unlock(txq);
316 
317 				if (status == NETDEV_TX_OK)
318 					break;
319 
320 			}
321 
322 			/* tickle device maybe there is some cleanup */
323 			netpoll_poll(np);
324 
325 			udelay(USEC_PER_POLL);
326 		}
327 
328 		WARN_ONCE(!irqs_disabled(),
329 			"netpoll_send_skb(): %s enabled interrupts in poll (%pF)\n",
330 			dev->name, ops->ndo_start_xmit);
331 
332 		local_irq_restore(flags);
333 	}
334 
335 	if (status != NETDEV_TX_OK) {
336 		skb_queue_tail(&npinfo->txq, skb);
337 		schedule_delayed_work(&npinfo->tx_work,0);
338 	}
339 }
340 
341 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
342 {
343 	int total_len, eth_len, ip_len, udp_len;
344 	struct sk_buff *skb;
345 	struct udphdr *udph;
346 	struct iphdr *iph;
347 	struct ethhdr *eth;
348 
349 	udp_len = len + sizeof(*udph);
350 	ip_len = eth_len = udp_len + sizeof(*iph);
351 	total_len = eth_len + ETH_HLEN + NET_IP_ALIGN;
352 
353 	skb = find_skb(np, total_len, total_len - len);
354 	if (!skb)
355 		return;
356 
357 	skb_copy_to_linear_data(skb, msg, len);
358 	skb->len += len;
359 
360 	skb_push(skb, sizeof(*udph));
361 	skb_reset_transport_header(skb);
362 	udph = udp_hdr(skb);
363 	udph->source = htons(np->local_port);
364 	udph->dest = htons(np->remote_port);
365 	udph->len = htons(udp_len);
366 	udph->check = 0;
367 	udph->check = csum_tcpudp_magic(np->local_ip,
368 					np->remote_ip,
369 					udp_len, IPPROTO_UDP,
370 					csum_partial(udph, udp_len, 0));
371 	if (udph->check == 0)
372 		udph->check = CSUM_MANGLED_0;
373 
374 	skb_push(skb, sizeof(*iph));
375 	skb_reset_network_header(skb);
376 	iph = ip_hdr(skb);
377 
378 	/* iph->version = 4; iph->ihl = 5; */
379 	put_unaligned(0x45, (unsigned char *)iph);
380 	iph->tos      = 0;
381 	put_unaligned(htons(ip_len), &(iph->tot_len));
382 	iph->id       = 0;
383 	iph->frag_off = 0;
384 	iph->ttl      = 64;
385 	iph->protocol = IPPROTO_UDP;
386 	iph->check    = 0;
387 	put_unaligned(np->local_ip, &(iph->saddr));
388 	put_unaligned(np->remote_ip, &(iph->daddr));
389 	iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
390 
391 	eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
392 	skb_reset_mac_header(skb);
393 	skb->protocol = eth->h_proto = htons(ETH_P_IP);
394 	memcpy(eth->h_source, np->dev->dev_addr, ETH_ALEN);
395 	memcpy(eth->h_dest, np->remote_mac, ETH_ALEN);
396 
397 	skb->dev = np->dev;
398 
399 	netpoll_send_skb(np, skb);
400 }
401 
402 static void arp_reply(struct sk_buff *skb)
403 {
404 	struct netpoll_info *npinfo = skb->dev->npinfo;
405 	struct arphdr *arp;
406 	unsigned char *arp_ptr;
407 	int size, type = ARPOP_REPLY, ptype = ETH_P_ARP;
408 	__be32 sip, tip;
409 	unsigned char *sha;
410 	struct sk_buff *send_skb;
411 	struct netpoll *np, *tmp;
412 	unsigned long flags;
413 	int hits = 0;
414 
415 	if (list_empty(&npinfo->rx_np))
416 		return;
417 
418 	/* Before checking the packet, we do some early
419 	   inspection whether this is interesting at all */
420 	spin_lock_irqsave(&npinfo->rx_lock, flags);
421 	list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
422 		if (np->dev == skb->dev)
423 			hits++;
424 	}
425 	spin_unlock_irqrestore(&npinfo->rx_lock, flags);
426 
427 	/* No netpoll struct is using this dev */
428 	if (!hits)
429 		return;
430 
431 	/* No arp on this interface */
432 	if (skb->dev->flags & IFF_NOARP)
433 		return;
434 
435 	if (!pskb_may_pull(skb, arp_hdr_len(skb->dev)))
436 		return;
437 
438 	skb_reset_network_header(skb);
439 	skb_reset_transport_header(skb);
440 	arp = arp_hdr(skb);
441 
442 	if ((arp->ar_hrd != htons(ARPHRD_ETHER) &&
443 	     arp->ar_hrd != htons(ARPHRD_IEEE802)) ||
444 	    arp->ar_pro != htons(ETH_P_IP) ||
445 	    arp->ar_op != htons(ARPOP_REQUEST))
446 		return;
447 
448 	arp_ptr = (unsigned char *)(arp+1);
449 	/* save the location of the src hw addr */
450 	sha = arp_ptr;
451 	arp_ptr += skb->dev->addr_len;
452 	memcpy(&sip, arp_ptr, 4);
453 	arp_ptr += 4;
454 	/* If we actually cared about dst hw addr,
455 	   it would get copied here */
456 	arp_ptr += skb->dev->addr_len;
457 	memcpy(&tip, arp_ptr, 4);
458 
459 	/* Should we ignore arp? */
460 	if (ipv4_is_loopback(tip) || ipv4_is_multicast(tip))
461 		return;
462 
463 	size = arp_hdr_len(skb->dev);
464 
465 	spin_lock_irqsave(&npinfo->rx_lock, flags);
466 	list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
467 		if (tip != np->local_ip)
468 			continue;
469 
470 		send_skb = find_skb(np, size + LL_ALLOCATED_SPACE(np->dev),
471 				    LL_RESERVED_SPACE(np->dev));
472 		if (!send_skb)
473 			continue;
474 
475 		skb_reset_network_header(send_skb);
476 		arp = (struct arphdr *) skb_put(send_skb, size);
477 		send_skb->dev = skb->dev;
478 		send_skb->protocol = htons(ETH_P_ARP);
479 
480 		/* Fill the device header for the ARP frame */
481 		if (dev_hard_header(send_skb, skb->dev, ptype,
482 				    sha, np->dev->dev_addr,
483 				    send_skb->len) < 0) {
484 			kfree_skb(send_skb);
485 			continue;
486 		}
487 
488 		/*
489 		 * Fill out the arp protocol part.
490 		 *
491 		 * we only support ethernet device type,
492 		 * which (according to RFC 1390) should
493 		 * always equal 1 (Ethernet).
494 		 */
495 
496 		arp->ar_hrd = htons(np->dev->type);
497 		arp->ar_pro = htons(ETH_P_IP);
498 		arp->ar_hln = np->dev->addr_len;
499 		arp->ar_pln = 4;
500 		arp->ar_op = htons(type);
501 
502 		arp_ptr = (unsigned char *)(arp + 1);
503 		memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
504 		arp_ptr += np->dev->addr_len;
505 		memcpy(arp_ptr, &tip, 4);
506 		arp_ptr += 4;
507 		memcpy(arp_ptr, sha, np->dev->addr_len);
508 		arp_ptr += np->dev->addr_len;
509 		memcpy(arp_ptr, &sip, 4);
510 
511 		netpoll_send_skb(np, send_skb);
512 
513 		/* If there are several rx_hooks for the same address,
514 		   we're fine by sending a single reply */
515 		break;
516 	}
517 	spin_unlock_irqrestore(&npinfo->rx_lock, flags);
518 }
519 
520 int __netpoll_rx(struct sk_buff *skb)
521 {
522 	int proto, len, ulen;
523 	int hits = 0;
524 	struct iphdr *iph;
525 	struct udphdr *uh;
526 	struct netpoll_info *npinfo = skb->dev->npinfo;
527 	struct netpoll *np, *tmp;
528 
529 	if (list_empty(&npinfo->rx_np))
530 		goto out;
531 
532 	if (skb->dev->type != ARPHRD_ETHER)
533 		goto out;
534 
535 	/* check if netpoll clients need ARP */
536 	if (skb->protocol == htons(ETH_P_ARP) &&
537 	    atomic_read(&trapped)) {
538 		skb_queue_tail(&npinfo->arp_tx, skb);
539 		return 1;
540 	}
541 
542 	proto = ntohs(eth_hdr(skb)->h_proto);
543 	if (proto != ETH_P_IP)
544 		goto out;
545 	if (skb->pkt_type == PACKET_OTHERHOST)
546 		goto out;
547 	if (skb_shared(skb))
548 		goto out;
549 
550 	iph = (struct iphdr *)skb->data;
551 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
552 		goto out;
553 	if (iph->ihl < 5 || iph->version != 4)
554 		goto out;
555 	if (!pskb_may_pull(skb, iph->ihl*4))
556 		goto out;
557 	if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
558 		goto out;
559 
560 	len = ntohs(iph->tot_len);
561 	if (skb->len < len || len < iph->ihl*4)
562 		goto out;
563 
564 	/*
565 	 * Our transport medium may have padded the buffer out.
566 	 * Now We trim to the true length of the frame.
567 	 */
568 	if (pskb_trim_rcsum(skb, len))
569 		goto out;
570 
571 	if (iph->protocol != IPPROTO_UDP)
572 		goto out;
573 
574 	len -= iph->ihl*4;
575 	uh = (struct udphdr *)(((char *)iph) + iph->ihl*4);
576 	ulen = ntohs(uh->len);
577 
578 	if (ulen != len)
579 		goto out;
580 	if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr))
581 		goto out;
582 
583 	list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
584 		if (np->local_ip && np->local_ip != iph->daddr)
585 			continue;
586 		if (np->remote_ip && np->remote_ip != iph->saddr)
587 			continue;
588 		if (np->local_port && np->local_port != ntohs(uh->dest))
589 			continue;
590 
591 		np->rx_hook(np, ntohs(uh->source),
592 			       (char *)(uh+1),
593 			       ulen - sizeof(struct udphdr));
594 		hits++;
595 	}
596 
597 	if (!hits)
598 		goto out;
599 
600 	kfree_skb(skb);
601 	return 1;
602 
603 out:
604 	if (atomic_read(&trapped)) {
605 		kfree_skb(skb);
606 		return 1;
607 	}
608 
609 	return 0;
610 }
611 
612 void netpoll_print_options(struct netpoll *np)
613 {
614 	printk(KERN_INFO "%s: local port %d\n",
615 			 np->name, np->local_port);
616 	printk(KERN_INFO "%s: local IP %pI4\n",
617 			 np->name, &np->local_ip);
618 	printk(KERN_INFO "%s: interface '%s'\n",
619 			 np->name, np->dev_name);
620 	printk(KERN_INFO "%s: remote port %d\n",
621 			 np->name, np->remote_port);
622 	printk(KERN_INFO "%s: remote IP %pI4\n",
623 			 np->name, &np->remote_ip);
624 	printk(KERN_INFO "%s: remote ethernet address %pM\n",
625 	                 np->name, np->remote_mac);
626 }
627 
628 int netpoll_parse_options(struct netpoll *np, char *opt)
629 {
630 	char *cur=opt, *delim;
631 
632 	if (*cur != '@') {
633 		if ((delim = strchr(cur, '@')) == NULL)
634 			goto parse_failed;
635 		*delim = 0;
636 		np->local_port = simple_strtol(cur, NULL, 10);
637 		cur = delim;
638 	}
639 	cur++;
640 
641 	if (*cur != '/') {
642 		if ((delim = strchr(cur, '/')) == NULL)
643 			goto parse_failed;
644 		*delim = 0;
645 		np->local_ip = in_aton(cur);
646 		cur = delim;
647 	}
648 	cur++;
649 
650 	if (*cur != ',') {
651 		/* parse out dev name */
652 		if ((delim = strchr(cur, ',')) == NULL)
653 			goto parse_failed;
654 		*delim = 0;
655 		strlcpy(np->dev_name, cur, sizeof(np->dev_name));
656 		cur = delim;
657 	}
658 	cur++;
659 
660 	if (*cur != '@') {
661 		/* dst port */
662 		if ((delim = strchr(cur, '@')) == NULL)
663 			goto parse_failed;
664 		*delim = 0;
665 		if (*cur == ' ' || *cur == '\t')
666 			printk(KERN_INFO "%s: warning: whitespace"
667 					"is not allowed\n", np->name);
668 		np->remote_port = simple_strtol(cur, NULL, 10);
669 		cur = delim;
670 	}
671 	cur++;
672 
673 	/* dst ip */
674 	if ((delim = strchr(cur, '/')) == NULL)
675 		goto parse_failed;
676 	*delim = 0;
677 	np->remote_ip = in_aton(cur);
678 	cur = delim + 1;
679 
680 	if (*cur != 0) {
681 		/* MAC address */
682 		if ((delim = strchr(cur, ':')) == NULL)
683 			goto parse_failed;
684 		*delim = 0;
685 		np->remote_mac[0] = simple_strtol(cur, NULL, 16);
686 		cur = delim + 1;
687 		if ((delim = strchr(cur, ':')) == NULL)
688 			goto parse_failed;
689 		*delim = 0;
690 		np->remote_mac[1] = simple_strtol(cur, NULL, 16);
691 		cur = delim + 1;
692 		if ((delim = strchr(cur, ':')) == NULL)
693 			goto parse_failed;
694 		*delim = 0;
695 		np->remote_mac[2] = simple_strtol(cur, NULL, 16);
696 		cur = delim + 1;
697 		if ((delim = strchr(cur, ':')) == NULL)
698 			goto parse_failed;
699 		*delim = 0;
700 		np->remote_mac[3] = simple_strtol(cur, NULL, 16);
701 		cur = delim + 1;
702 		if ((delim = strchr(cur, ':')) == NULL)
703 			goto parse_failed;
704 		*delim = 0;
705 		np->remote_mac[4] = simple_strtol(cur, NULL, 16);
706 		cur = delim + 1;
707 		np->remote_mac[5] = simple_strtol(cur, NULL, 16);
708 	}
709 
710 	netpoll_print_options(np);
711 
712 	return 0;
713 
714  parse_failed:
715 	printk(KERN_INFO "%s: couldn't parse config at '%s'!\n",
716 	       np->name, cur);
717 	return -1;
718 }
719 
720 int netpoll_setup(struct netpoll *np)
721 {
722 	struct net_device *ndev = NULL;
723 	struct in_device *in_dev;
724 	struct netpoll_info *npinfo;
725 	struct netpoll *npe, *tmp;
726 	unsigned long flags;
727 	int err;
728 
729 	if (np->dev_name)
730 		ndev = dev_get_by_name(&init_net, np->dev_name);
731 	if (!ndev) {
732 		printk(KERN_ERR "%s: %s doesn't exist, aborting.\n",
733 		       np->name, np->dev_name);
734 		return -ENODEV;
735 	}
736 
737 	np->dev = ndev;
738 	if (!ndev->npinfo) {
739 		npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
740 		if (!npinfo) {
741 			err = -ENOMEM;
742 			goto put;
743 		}
744 
745 		npinfo->rx_flags = 0;
746 		INIT_LIST_HEAD(&npinfo->rx_np);
747 
748 		spin_lock_init(&npinfo->rx_lock);
749 		skb_queue_head_init(&npinfo->arp_tx);
750 		skb_queue_head_init(&npinfo->txq);
751 		INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
752 
753 		atomic_set(&npinfo->refcnt, 1);
754 	} else {
755 		npinfo = ndev->npinfo;
756 		atomic_inc(&npinfo->refcnt);
757 	}
758 
759 	if (!ndev->netdev_ops->ndo_poll_controller) {
760 		printk(KERN_ERR "%s: %s doesn't support polling, aborting.\n",
761 		       np->name, np->dev_name);
762 		err = -ENOTSUPP;
763 		goto release;
764 	}
765 
766 	if (!netif_running(ndev)) {
767 		unsigned long atmost, atleast;
768 
769 		printk(KERN_INFO "%s: device %s not up yet, forcing it\n",
770 		       np->name, np->dev_name);
771 
772 		rtnl_lock();
773 		err = dev_open(ndev);
774 		rtnl_unlock();
775 
776 		if (err) {
777 			printk(KERN_ERR "%s: failed to open %s\n",
778 			       np->name, ndev->name);
779 			goto release;
780 		}
781 
782 		atleast = jiffies + HZ/10;
783 		atmost = jiffies + carrier_timeout * HZ;
784 		while (!netif_carrier_ok(ndev)) {
785 			if (time_after(jiffies, atmost)) {
786 				printk(KERN_NOTICE
787 				       "%s: timeout waiting for carrier\n",
788 				       np->name);
789 				break;
790 			}
791 			msleep(1);
792 		}
793 
794 		/* If carrier appears to come up instantly, we don't
795 		 * trust it and pause so that we don't pump all our
796 		 * queued console messages into the bitbucket.
797 		 */
798 
799 		if (time_before(jiffies, atleast)) {
800 			printk(KERN_NOTICE "%s: carrier detect appears"
801 			       " untrustworthy, waiting 4 seconds\n",
802 			       np->name);
803 			msleep(4000);
804 		}
805 	}
806 
807 	if (!np->local_ip) {
808 		rcu_read_lock();
809 		in_dev = __in_dev_get_rcu(ndev);
810 
811 		if (!in_dev || !in_dev->ifa_list) {
812 			rcu_read_unlock();
813 			printk(KERN_ERR "%s: no IP address for %s, aborting\n",
814 			       np->name, np->dev_name);
815 			err = -EDESTADDRREQ;
816 			goto release;
817 		}
818 
819 		np->local_ip = in_dev->ifa_list->ifa_local;
820 		rcu_read_unlock();
821 		printk(KERN_INFO "%s: local IP %pI4\n", np->name, &np->local_ip);
822 	}
823 
824 	if (np->rx_hook) {
825 		spin_lock_irqsave(&npinfo->rx_lock, flags);
826 		npinfo->rx_flags |= NETPOLL_RX_ENABLED;
827 		list_add_tail(&np->rx, &npinfo->rx_np);
828 		spin_unlock_irqrestore(&npinfo->rx_lock, flags);
829 	}
830 
831 	/* fill up the skb queue */
832 	refill_skbs();
833 
834 	/* last thing to do is link it to the net device structure */
835 	ndev->npinfo = npinfo;
836 
837 	/* avoid racing with NAPI reading npinfo */
838 	synchronize_rcu();
839 
840 	return 0;
841 
842  release:
843 	if (!ndev->npinfo) {
844 		spin_lock_irqsave(&npinfo->rx_lock, flags);
845 		list_for_each_entry_safe(npe, tmp, &npinfo->rx_np, rx) {
846 			npe->dev = NULL;
847 		}
848 		spin_unlock_irqrestore(&npinfo->rx_lock, flags);
849 
850 		kfree(npinfo);
851 	}
852 put:
853 	dev_put(ndev);
854 	return err;
855 }
856 
857 static int __init netpoll_init(void)
858 {
859 	skb_queue_head_init(&skb_pool);
860 	return 0;
861 }
862 core_initcall(netpoll_init);
863 
864 void netpoll_cleanup(struct netpoll *np)
865 {
866 	struct netpoll_info *npinfo;
867 	unsigned long flags;
868 
869 	if (np->dev) {
870 		npinfo = np->dev->npinfo;
871 		if (npinfo) {
872 			if (!list_empty(&npinfo->rx_np)) {
873 				spin_lock_irqsave(&npinfo->rx_lock, flags);
874 				list_del(&np->rx);
875 				if (list_empty(&npinfo->rx_np))
876 					npinfo->rx_flags &= ~NETPOLL_RX_ENABLED;
877 				spin_unlock_irqrestore(&npinfo->rx_lock, flags);
878 			}
879 
880 			if (atomic_dec_and_test(&npinfo->refcnt)) {
881 				skb_queue_purge(&npinfo->arp_tx);
882 				skb_queue_purge(&npinfo->txq);
883 				cancel_rearming_delayed_work(&npinfo->tx_work);
884 
885 				/* clean after last, unfinished work */
886 				__skb_queue_purge(&npinfo->txq);
887 				kfree(npinfo);
888 				np->dev->npinfo = NULL;
889 			}
890 		}
891 
892 		dev_put(np->dev);
893 	}
894 
895 	np->dev = NULL;
896 }
897 
898 int netpoll_trap(void)
899 {
900 	return atomic_read(&trapped);
901 }
902 
903 void netpoll_set_trap(int trap)
904 {
905 	if (trap)
906 		atomic_inc(&trapped);
907 	else
908 		atomic_dec(&trapped);
909 }
910 
911 EXPORT_SYMBOL(netpoll_set_trap);
912 EXPORT_SYMBOL(netpoll_trap);
913 EXPORT_SYMBOL(netpoll_print_options);
914 EXPORT_SYMBOL(netpoll_parse_options);
915 EXPORT_SYMBOL(netpoll_setup);
916 EXPORT_SYMBOL(netpoll_cleanup);
917 EXPORT_SYMBOL(netpoll_send_udp);
918 EXPORT_SYMBOL(netpoll_poll);
919