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 DEFINE_STATIC_SRCU(netpoll_srcu);
50 
51 #define USEC_PER_POLL	50
52 
53 #define MAX_SKB_SIZE							\
54 	(sizeof(struct ethhdr) +					\
55 	 sizeof(struct iphdr) +						\
56 	 sizeof(struct udphdr) +					\
57 	 MAX_UDP_CHUNK)
58 
59 static void zap_completion_queue(void);
60 static void netpoll_async_cleanup(struct work_struct *work);
61 
62 static unsigned int carrier_timeout = 4;
63 module_param(carrier_timeout, uint, 0644);
64 
65 #define np_info(np, fmt, ...)				\
66 	pr_info("%s: " fmt, np->name, ##__VA_ARGS__)
67 #define np_err(np, fmt, ...)				\
68 	pr_err("%s: " fmt, np->name, ##__VA_ARGS__)
69 #define np_notice(np, fmt, ...)				\
70 	pr_notice("%s: " fmt, np->name, ##__VA_ARGS__)
71 
netpoll_start_xmit(struct sk_buff * skb,struct net_device * dev,struct netdev_queue * txq)72 static int netpoll_start_xmit(struct sk_buff *skb, struct net_device *dev,
73 			      struct netdev_queue *txq)
74 {
75 	int status = NETDEV_TX_OK;
76 	netdev_features_t features;
77 
78 	features = netif_skb_features(skb);
79 
80 	if (skb_vlan_tag_present(skb) &&
81 	    !vlan_hw_offload_capable(features, skb->vlan_proto)) {
82 		skb = __vlan_hwaccel_push_inside(skb);
83 		if (unlikely(!skb)) {
84 			/* This is actually a packet drop, but we
85 			 * don't want the code that calls this
86 			 * function to try and operate on a NULL skb.
87 			 */
88 			goto out;
89 		}
90 	}
91 
92 	status = netdev_start_xmit(skb, dev, txq, false);
93 
94 out:
95 	return status;
96 }
97 
queue_process(struct work_struct * work)98 static void queue_process(struct work_struct *work)
99 {
100 	struct netpoll_info *npinfo =
101 		container_of(work, struct netpoll_info, tx_work.work);
102 	struct sk_buff *skb;
103 	unsigned long flags;
104 
105 	while ((skb = skb_dequeue(&npinfo->txq))) {
106 		struct net_device *dev = skb->dev;
107 		struct netdev_queue *txq;
108 		unsigned int q_index;
109 
110 		if (!netif_device_present(dev) || !netif_running(dev)) {
111 			kfree_skb(skb);
112 			continue;
113 		}
114 
115 		local_irq_save(flags);
116 		/* check if skb->queue_mapping is still valid */
117 		q_index = skb_get_queue_mapping(skb);
118 		if (unlikely(q_index >= dev->real_num_tx_queues)) {
119 			q_index = q_index % dev->real_num_tx_queues;
120 			skb_set_queue_mapping(skb, q_index);
121 		}
122 		txq = netdev_get_tx_queue(dev, q_index);
123 		HARD_TX_LOCK(dev, txq, smp_processor_id());
124 		if (netif_xmit_frozen_or_stopped(txq) ||
125 		    netpoll_start_xmit(skb, dev, txq) != NETDEV_TX_OK) {
126 			skb_queue_head(&npinfo->txq, skb);
127 			HARD_TX_UNLOCK(dev, txq);
128 			local_irq_restore(flags);
129 
130 			schedule_delayed_work(&npinfo->tx_work, HZ/10);
131 			return;
132 		}
133 		HARD_TX_UNLOCK(dev, txq);
134 		local_irq_restore(flags);
135 	}
136 }
137 
poll_one_napi(struct napi_struct * napi)138 static void poll_one_napi(struct napi_struct *napi)
139 {
140 	int work;
141 
142 	/* If we set this bit but see that it has already been set,
143 	 * that indicates that napi has been disabled and we need
144 	 * to abort this operation
145 	 */
146 	if (test_and_set_bit(NAPI_STATE_NPSVC, &napi->state))
147 		return;
148 
149 	/* We explicilty pass the polling call a budget of 0 to
150 	 * indicate that we are clearing the Tx path only.
151 	 */
152 	work = napi->poll(napi, 0);
153 	WARN_ONCE(work, "%pF exceeded budget in poll\n", napi->poll);
154 	trace_napi_poll(napi, work, 0);
155 
156 	clear_bit(NAPI_STATE_NPSVC, &napi->state);
157 }
158 
poll_napi(struct net_device * dev)159 static void poll_napi(struct net_device *dev)
160 {
161 	struct napi_struct *napi;
162 	int cpu = smp_processor_id();
163 
164 	list_for_each_entry(napi, &dev->napi_list, dev_list) {
165 		if (cmpxchg(&napi->poll_owner, -1, cpu) == -1) {
166 			poll_one_napi(napi);
167 			smp_store_release(&napi->poll_owner, -1);
168 		}
169 	}
170 }
171 
netpoll_poll_dev(struct net_device * dev)172 void netpoll_poll_dev(struct net_device *dev)
173 {
174 	struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
175 	const struct net_device_ops *ops;
176 
177 	/* Don't do any rx activity if the dev_lock mutex is held
178 	 * the dev_open/close paths use this to block netpoll activity
179 	 * while changing device state
180 	 */
181 	if (!ni || down_trylock(&ni->dev_lock))
182 		return;
183 
184 	if (!netif_running(dev)) {
185 		up(&ni->dev_lock);
186 		return;
187 	}
188 
189 	ops = dev->netdev_ops;
190 	if (ops->ndo_poll_controller)
191 		ops->ndo_poll_controller(dev);
192 
193 	poll_napi(dev);
194 
195 	up(&ni->dev_lock);
196 
197 	zap_completion_queue();
198 }
199 EXPORT_SYMBOL(netpoll_poll_dev);
200 
netpoll_poll_disable(struct net_device * dev)201 void netpoll_poll_disable(struct net_device *dev)
202 {
203 	struct netpoll_info *ni;
204 	int idx;
205 	might_sleep();
206 	idx = srcu_read_lock(&netpoll_srcu);
207 	ni = srcu_dereference(dev->npinfo, &netpoll_srcu);
208 	if (ni)
209 		down(&ni->dev_lock);
210 	srcu_read_unlock(&netpoll_srcu, idx);
211 }
212 EXPORT_SYMBOL(netpoll_poll_disable);
213 
netpoll_poll_enable(struct net_device * dev)214 void netpoll_poll_enable(struct net_device *dev)
215 {
216 	struct netpoll_info *ni;
217 	rcu_read_lock();
218 	ni = rcu_dereference(dev->npinfo);
219 	if (ni)
220 		up(&ni->dev_lock);
221 	rcu_read_unlock();
222 }
223 EXPORT_SYMBOL(netpoll_poll_enable);
224 
refill_skbs(void)225 static void refill_skbs(void)
226 {
227 	struct sk_buff *skb;
228 	unsigned long flags;
229 
230 	spin_lock_irqsave(&skb_pool.lock, flags);
231 	while (skb_pool.qlen < MAX_SKBS) {
232 		skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
233 		if (!skb)
234 			break;
235 
236 		__skb_queue_tail(&skb_pool, skb);
237 	}
238 	spin_unlock_irqrestore(&skb_pool.lock, flags);
239 }
240 
zap_completion_queue(void)241 static void zap_completion_queue(void)
242 {
243 	unsigned long flags;
244 	struct softnet_data *sd = &get_cpu_var(softnet_data);
245 
246 	if (sd->completion_queue) {
247 		struct sk_buff *clist;
248 
249 		local_irq_save(flags);
250 		clist = sd->completion_queue;
251 		sd->completion_queue = NULL;
252 		local_irq_restore(flags);
253 
254 		while (clist != NULL) {
255 			struct sk_buff *skb = clist;
256 			clist = clist->next;
257 			if (!skb_irq_freeable(skb)) {
258 				refcount_set(&skb->users, 1);
259 				dev_kfree_skb_any(skb); /* put this one back */
260 			} else {
261 				__kfree_skb(skb);
262 			}
263 		}
264 	}
265 
266 	put_cpu_var(softnet_data);
267 }
268 
find_skb(struct netpoll * np,int len,int reserve)269 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
270 {
271 	int count = 0;
272 	struct sk_buff *skb;
273 
274 	zap_completion_queue();
275 	refill_skbs();
276 repeat:
277 
278 	skb = alloc_skb(len, GFP_ATOMIC);
279 	if (!skb)
280 		skb = skb_dequeue(&skb_pool);
281 
282 	if (!skb) {
283 		if (++count < 10) {
284 			netpoll_poll_dev(np->dev);
285 			goto repeat;
286 		}
287 		return NULL;
288 	}
289 
290 	refcount_set(&skb->users, 1);
291 	skb_reserve(skb, reserve);
292 	return skb;
293 }
294 
netpoll_owner_active(struct net_device * dev)295 static int netpoll_owner_active(struct net_device *dev)
296 {
297 	struct napi_struct *napi;
298 
299 	list_for_each_entry(napi, &dev->napi_list, dev_list) {
300 		if (napi->poll_owner == smp_processor_id())
301 			return 1;
302 	}
303 	return 0;
304 }
305 
306 /* call with IRQ disabled */
netpoll_send_skb_on_dev(struct netpoll * np,struct sk_buff * skb,struct net_device * dev)307 void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb,
308 			     struct net_device *dev)
309 {
310 	int status = NETDEV_TX_BUSY;
311 	unsigned long tries;
312 	/* It is up to the caller to keep npinfo alive. */
313 	struct netpoll_info *npinfo;
314 
315 	lockdep_assert_irqs_disabled();
316 
317 	npinfo = rcu_dereference_bh(np->dev->npinfo);
318 	if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
319 		dev_kfree_skb_irq(skb);
320 		return;
321 	}
322 
323 	/* don't get messages out of order, and no recursion */
324 	if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
325 		struct netdev_queue *txq;
326 
327 		txq = netdev_pick_tx(dev, skb, NULL);
328 
329 		/* try until next clock tick */
330 		for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
331 		     tries > 0; --tries) {
332 			if (HARD_TX_TRYLOCK(dev, txq)) {
333 				if (!netif_xmit_stopped(txq))
334 					status = netpoll_start_xmit(skb, dev, txq);
335 
336 				HARD_TX_UNLOCK(dev, txq);
337 
338 				if (status == NETDEV_TX_OK)
339 					break;
340 
341 			}
342 
343 			/* tickle device maybe there is some cleanup */
344 			netpoll_poll_dev(np->dev);
345 
346 			udelay(USEC_PER_POLL);
347 		}
348 
349 		WARN_ONCE(!irqs_disabled(),
350 			"netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pF)\n",
351 			dev->name, dev->netdev_ops->ndo_start_xmit);
352 
353 	}
354 
355 	if (status != NETDEV_TX_OK) {
356 		skb_queue_tail(&npinfo->txq, skb);
357 		schedule_delayed_work(&npinfo->tx_work,0);
358 	}
359 }
360 EXPORT_SYMBOL(netpoll_send_skb_on_dev);
361 
netpoll_send_udp(struct netpoll * np,const char * msg,int len)362 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
363 {
364 	int total_len, ip_len, udp_len;
365 	struct sk_buff *skb;
366 	struct udphdr *udph;
367 	struct iphdr *iph;
368 	struct ethhdr *eth;
369 	static atomic_t ip_ident;
370 	struct ipv6hdr *ip6h;
371 
372 	WARN_ON_ONCE(!irqs_disabled());
373 
374 	udp_len = len + sizeof(*udph);
375 	if (np->ipv6)
376 		ip_len = udp_len + sizeof(*ip6h);
377 	else
378 		ip_len = udp_len + sizeof(*iph);
379 
380 	total_len = ip_len + LL_RESERVED_SPACE(np->dev);
381 
382 	skb = find_skb(np, total_len + np->dev->needed_tailroom,
383 		       total_len - len);
384 	if (!skb)
385 		return;
386 
387 	skb_copy_to_linear_data(skb, msg, len);
388 	skb_put(skb, len);
389 
390 	skb_push(skb, sizeof(*udph));
391 	skb_reset_transport_header(skb);
392 	udph = udp_hdr(skb);
393 	udph->source = htons(np->local_port);
394 	udph->dest = htons(np->remote_port);
395 	udph->len = htons(udp_len);
396 
397 	if (np->ipv6) {
398 		udph->check = 0;
399 		udph->check = csum_ipv6_magic(&np->local_ip.in6,
400 					      &np->remote_ip.in6,
401 					      udp_len, IPPROTO_UDP,
402 					      csum_partial(udph, udp_len, 0));
403 		if (udph->check == 0)
404 			udph->check = CSUM_MANGLED_0;
405 
406 		skb_push(skb, sizeof(*ip6h));
407 		skb_reset_network_header(skb);
408 		ip6h = ipv6_hdr(skb);
409 
410 		/* ip6h->version = 6; ip6h->priority = 0; */
411 		put_unaligned(0x60, (unsigned char *)ip6h);
412 		ip6h->flow_lbl[0] = 0;
413 		ip6h->flow_lbl[1] = 0;
414 		ip6h->flow_lbl[2] = 0;
415 
416 		ip6h->payload_len = htons(sizeof(struct udphdr) + len);
417 		ip6h->nexthdr = IPPROTO_UDP;
418 		ip6h->hop_limit = 32;
419 		ip6h->saddr = np->local_ip.in6;
420 		ip6h->daddr = np->remote_ip.in6;
421 
422 		eth = skb_push(skb, ETH_HLEN);
423 		skb_reset_mac_header(skb);
424 		skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
425 	} else {
426 		udph->check = 0;
427 		udph->check = csum_tcpudp_magic(np->local_ip.ip,
428 						np->remote_ip.ip,
429 						udp_len, IPPROTO_UDP,
430 						csum_partial(udph, udp_len, 0));
431 		if (udph->check == 0)
432 			udph->check = CSUM_MANGLED_0;
433 
434 		skb_push(skb, sizeof(*iph));
435 		skb_reset_network_header(skb);
436 		iph = ip_hdr(skb);
437 
438 		/* iph->version = 4; iph->ihl = 5; */
439 		put_unaligned(0x45, (unsigned char *)iph);
440 		iph->tos      = 0;
441 		put_unaligned(htons(ip_len), &(iph->tot_len));
442 		iph->id       = htons(atomic_inc_return(&ip_ident));
443 		iph->frag_off = 0;
444 		iph->ttl      = 64;
445 		iph->protocol = IPPROTO_UDP;
446 		iph->check    = 0;
447 		put_unaligned(np->local_ip.ip, &(iph->saddr));
448 		put_unaligned(np->remote_ip.ip, &(iph->daddr));
449 		iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
450 
451 		eth = skb_push(skb, ETH_HLEN);
452 		skb_reset_mac_header(skb);
453 		skb->protocol = eth->h_proto = htons(ETH_P_IP);
454 	}
455 
456 	ether_addr_copy(eth->h_source, np->dev->dev_addr);
457 	ether_addr_copy(eth->h_dest, np->remote_mac);
458 
459 	skb->dev = np->dev;
460 
461 	netpoll_send_skb(np, skb);
462 }
463 EXPORT_SYMBOL(netpoll_send_udp);
464 
netpoll_print_options(struct netpoll * np)465 void netpoll_print_options(struct netpoll *np)
466 {
467 	np_info(np, "local port %d\n", np->local_port);
468 	if (np->ipv6)
469 		np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
470 	else
471 		np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
472 	np_info(np, "interface '%s'\n", np->dev_name);
473 	np_info(np, "remote port %d\n", np->remote_port);
474 	if (np->ipv6)
475 		np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
476 	else
477 		np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
478 	np_info(np, "remote ethernet address %pM\n", np->remote_mac);
479 }
480 EXPORT_SYMBOL(netpoll_print_options);
481 
netpoll_parse_ip_addr(const char * str,union inet_addr * addr)482 static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
483 {
484 	const char *end;
485 
486 	if (!strchr(str, ':') &&
487 	    in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
488 		if (!*end)
489 			return 0;
490 	}
491 	if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
492 #if IS_ENABLED(CONFIG_IPV6)
493 		if (!*end)
494 			return 1;
495 #else
496 		return -1;
497 #endif
498 	}
499 	return -1;
500 }
501 
netpoll_parse_options(struct netpoll * np,char * opt)502 int netpoll_parse_options(struct netpoll *np, char *opt)
503 {
504 	char *cur=opt, *delim;
505 	int ipv6;
506 	bool ipversion_set = false;
507 
508 	if (*cur != '@') {
509 		if ((delim = strchr(cur, '@')) == NULL)
510 			goto parse_failed;
511 		*delim = 0;
512 		if (kstrtou16(cur, 10, &np->local_port))
513 			goto parse_failed;
514 		cur = delim;
515 	}
516 	cur++;
517 
518 	if (*cur != '/') {
519 		ipversion_set = true;
520 		if ((delim = strchr(cur, '/')) == NULL)
521 			goto parse_failed;
522 		*delim = 0;
523 		ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
524 		if (ipv6 < 0)
525 			goto parse_failed;
526 		else
527 			np->ipv6 = (bool)ipv6;
528 		cur = delim;
529 	}
530 	cur++;
531 
532 	if (*cur != ',') {
533 		/* parse out dev name */
534 		if ((delim = strchr(cur, ',')) == NULL)
535 			goto parse_failed;
536 		*delim = 0;
537 		strlcpy(np->dev_name, cur, sizeof(np->dev_name));
538 		cur = delim;
539 	}
540 	cur++;
541 
542 	if (*cur != '@') {
543 		/* dst port */
544 		if ((delim = strchr(cur, '@')) == NULL)
545 			goto parse_failed;
546 		*delim = 0;
547 		if (*cur == ' ' || *cur == '\t')
548 			np_info(np, "warning: whitespace is not allowed\n");
549 		if (kstrtou16(cur, 10, &np->remote_port))
550 			goto parse_failed;
551 		cur = delim;
552 	}
553 	cur++;
554 
555 	/* dst ip */
556 	if ((delim = strchr(cur, '/')) == NULL)
557 		goto parse_failed;
558 	*delim = 0;
559 	ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
560 	if (ipv6 < 0)
561 		goto parse_failed;
562 	else if (ipversion_set && np->ipv6 != (bool)ipv6)
563 		goto parse_failed;
564 	else
565 		np->ipv6 = (bool)ipv6;
566 	cur = delim + 1;
567 
568 	if (*cur != 0) {
569 		/* MAC address */
570 		if (!mac_pton(cur, np->remote_mac))
571 			goto parse_failed;
572 	}
573 
574 	netpoll_print_options(np);
575 
576 	return 0;
577 
578  parse_failed:
579 	np_info(np, "couldn't parse config at '%s'!\n", cur);
580 	return -1;
581 }
582 EXPORT_SYMBOL(netpoll_parse_options);
583 
__netpoll_setup(struct netpoll * np,struct net_device * ndev)584 int __netpoll_setup(struct netpoll *np, struct net_device *ndev)
585 {
586 	struct netpoll_info *npinfo;
587 	const struct net_device_ops *ops;
588 	int err;
589 
590 	np->dev = ndev;
591 	strlcpy(np->dev_name, ndev->name, IFNAMSIZ);
592 	INIT_WORK(&np->cleanup_work, netpoll_async_cleanup);
593 
594 	if (ndev->priv_flags & IFF_DISABLE_NETPOLL) {
595 		np_err(np, "%s doesn't support polling, aborting\n",
596 		       np->dev_name);
597 		err = -ENOTSUPP;
598 		goto out;
599 	}
600 
601 	if (!ndev->npinfo) {
602 		npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
603 		if (!npinfo) {
604 			err = -ENOMEM;
605 			goto out;
606 		}
607 
608 		sema_init(&npinfo->dev_lock, 1);
609 		skb_queue_head_init(&npinfo->txq);
610 		INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
611 
612 		refcount_set(&npinfo->refcnt, 1);
613 
614 		ops = np->dev->netdev_ops;
615 		if (ops->ndo_netpoll_setup) {
616 			err = ops->ndo_netpoll_setup(ndev, npinfo);
617 			if (err)
618 				goto free_npinfo;
619 		}
620 	} else {
621 		npinfo = rtnl_dereference(ndev->npinfo);
622 		refcount_inc(&npinfo->refcnt);
623 	}
624 
625 	npinfo->netpoll = np;
626 
627 	/* last thing to do is link it to the net device structure */
628 	rcu_assign_pointer(ndev->npinfo, npinfo);
629 
630 	return 0;
631 
632 free_npinfo:
633 	kfree(npinfo);
634 out:
635 	return err;
636 }
637 EXPORT_SYMBOL_GPL(__netpoll_setup);
638 
netpoll_setup(struct netpoll * np)639 int netpoll_setup(struct netpoll *np)
640 {
641 	struct net_device *ndev = NULL;
642 	struct in_device *in_dev;
643 	int err;
644 
645 	rtnl_lock();
646 	if (np->dev_name[0]) {
647 		struct net *net = current->nsproxy->net_ns;
648 		ndev = __dev_get_by_name(net, np->dev_name);
649 	}
650 	if (!ndev) {
651 		np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
652 		err = -ENODEV;
653 		goto unlock;
654 	}
655 	dev_hold(ndev);
656 
657 	if (netdev_master_upper_dev_get(ndev)) {
658 		np_err(np, "%s is a slave device, aborting\n", np->dev_name);
659 		err = -EBUSY;
660 		goto put;
661 	}
662 
663 	if (!netif_running(ndev)) {
664 		unsigned long atmost, atleast;
665 
666 		np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
667 
668 		err = dev_open(ndev);
669 
670 		if (err) {
671 			np_err(np, "failed to open %s\n", ndev->name);
672 			goto put;
673 		}
674 
675 		rtnl_unlock();
676 		atleast = jiffies + HZ/10;
677 		atmost = jiffies + carrier_timeout * HZ;
678 		while (!netif_carrier_ok(ndev)) {
679 			if (time_after(jiffies, atmost)) {
680 				np_notice(np, "timeout waiting for carrier\n");
681 				break;
682 			}
683 			msleep(1);
684 		}
685 
686 		/* If carrier appears to come up instantly, we don't
687 		 * trust it and pause so that we don't pump all our
688 		 * queued console messages into the bitbucket.
689 		 */
690 
691 		if (time_before(jiffies, atleast)) {
692 			np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
693 			msleep(4000);
694 		}
695 		rtnl_lock();
696 	}
697 
698 	if (!np->local_ip.ip) {
699 		if (!np->ipv6) {
700 			in_dev = __in_dev_get_rtnl(ndev);
701 
702 			if (!in_dev || !in_dev->ifa_list) {
703 				np_err(np, "no IP address for %s, aborting\n",
704 				       np->dev_name);
705 				err = -EDESTADDRREQ;
706 				goto put;
707 			}
708 
709 			np->local_ip.ip = in_dev->ifa_list->ifa_local;
710 			np_info(np, "local IP %pI4\n", &np->local_ip.ip);
711 		} else {
712 #if IS_ENABLED(CONFIG_IPV6)
713 			struct inet6_dev *idev;
714 
715 			err = -EDESTADDRREQ;
716 			idev = __in6_dev_get(ndev);
717 			if (idev) {
718 				struct inet6_ifaddr *ifp;
719 
720 				read_lock_bh(&idev->lock);
721 				list_for_each_entry(ifp, &idev->addr_list, if_list) {
722 					if (ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL)
723 						continue;
724 					np->local_ip.in6 = ifp->addr;
725 					err = 0;
726 					break;
727 				}
728 				read_unlock_bh(&idev->lock);
729 			}
730 			if (err) {
731 				np_err(np, "no IPv6 address for %s, aborting\n",
732 				       np->dev_name);
733 				goto put;
734 			} else
735 				np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
736 #else
737 			np_err(np, "IPv6 is not supported %s, aborting\n",
738 			       np->dev_name);
739 			err = -EINVAL;
740 			goto put;
741 #endif
742 		}
743 	}
744 
745 	/* fill up the skb queue */
746 	refill_skbs();
747 
748 	err = __netpoll_setup(np, ndev);
749 	if (err)
750 		goto put;
751 
752 	rtnl_unlock();
753 	return 0;
754 
755 put:
756 	dev_put(ndev);
757 unlock:
758 	rtnl_unlock();
759 	return err;
760 }
761 EXPORT_SYMBOL(netpoll_setup);
762 
netpoll_init(void)763 static int __init netpoll_init(void)
764 {
765 	skb_queue_head_init(&skb_pool);
766 	return 0;
767 }
768 core_initcall(netpoll_init);
769 
rcu_cleanup_netpoll_info(struct rcu_head * rcu_head)770 static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
771 {
772 	struct netpoll_info *npinfo =
773 			container_of(rcu_head, struct netpoll_info, rcu);
774 
775 	skb_queue_purge(&npinfo->txq);
776 
777 	/* we can't call cancel_delayed_work_sync here, as we are in softirq */
778 	cancel_delayed_work(&npinfo->tx_work);
779 
780 	/* clean after last, unfinished work */
781 	__skb_queue_purge(&npinfo->txq);
782 	/* now cancel it again */
783 	cancel_delayed_work(&npinfo->tx_work);
784 	kfree(npinfo);
785 }
786 
__netpoll_cleanup(struct netpoll * np)787 void __netpoll_cleanup(struct netpoll *np)
788 {
789 	struct netpoll_info *npinfo;
790 
791 	/* rtnl_dereference would be preferable here but
792 	 * rcu_cleanup_netpoll path can put us in here safely without
793 	 * holding the rtnl, so plain rcu_dereference it is
794 	 */
795 	npinfo = rtnl_dereference(np->dev->npinfo);
796 	if (!npinfo)
797 		return;
798 
799 	synchronize_srcu(&netpoll_srcu);
800 
801 	if (refcount_dec_and_test(&npinfo->refcnt)) {
802 		const struct net_device_ops *ops;
803 
804 		ops = np->dev->netdev_ops;
805 		if (ops->ndo_netpoll_cleanup)
806 			ops->ndo_netpoll_cleanup(np->dev);
807 
808 		RCU_INIT_POINTER(np->dev->npinfo, NULL);
809 		call_rcu_bh(&npinfo->rcu, rcu_cleanup_netpoll_info);
810 	} else
811 		RCU_INIT_POINTER(np->dev->npinfo, NULL);
812 }
813 EXPORT_SYMBOL_GPL(__netpoll_cleanup);
814 
netpoll_async_cleanup(struct work_struct * work)815 static void netpoll_async_cleanup(struct work_struct *work)
816 {
817 	struct netpoll *np = container_of(work, struct netpoll, cleanup_work);
818 
819 	rtnl_lock();
820 	__netpoll_cleanup(np);
821 	rtnl_unlock();
822 	kfree(np);
823 }
824 
__netpoll_free_async(struct netpoll * np)825 void __netpoll_free_async(struct netpoll *np)
826 {
827 	schedule_work(&np->cleanup_work);
828 }
829 EXPORT_SYMBOL_GPL(__netpoll_free_async);
830 
netpoll_cleanup(struct netpoll * np)831 void netpoll_cleanup(struct netpoll *np)
832 {
833 	rtnl_lock();
834 	if (!np->dev)
835 		goto out;
836 	__netpoll_cleanup(np);
837 	dev_put(np->dev);
838 	np->dev = NULL;
839 out:
840 	rtnl_unlock();
841 }
842 EXPORT_SYMBOL(netpoll_cleanup);
843