xref: /linux/net/core/netpoll.c (revision 4003c9e78778e93188a09d6043a74f7154449d43)
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 
netpoll_start_xmit(struct sk_buff * skb,struct net_device * dev,struct netdev_queue * txq)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 
queue_process(struct work_struct * work)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 
netif_local_xmit_active(struct net_device * dev)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 
poll_one_napi(struct napi_struct * napi)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 
poll_napi(struct net_device * dev)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 
netpoll_poll_dev(struct net_device * dev)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 
netpoll_poll_disable(struct net_device * dev)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 
netpoll_poll_enable(struct net_device * dev)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 
refill_skbs(struct netpoll * np)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 
zap_completion_queue(void)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 
find_skb(struct netpoll * np,int len,int reserve)281 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
282 {
283 	int count = 0;
284 	struct sk_buff *skb;
285 
286 	zap_completion_queue();
287 	refill_skbs(np);
288 repeat:
289 
290 	skb = alloc_skb(len, GFP_ATOMIC);
291 	if (!skb)
292 		skb = skb_dequeue(&np->skb_pool);
293 
294 	if (!skb) {
295 		if (++count < 10) {
296 			netpoll_poll_dev(np->dev);
297 			goto repeat;
298 		}
299 		return NULL;
300 	}
301 
302 	refcount_set(&skb->users, 1);
303 	skb_reserve(skb, reserve);
304 	return skb;
305 }
306 
netpoll_owner_active(struct net_device * dev)307 static int netpoll_owner_active(struct net_device *dev)
308 {
309 	struct napi_struct *napi;
310 
311 	list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) {
312 		if (READ_ONCE(napi->poll_owner) == smp_processor_id())
313 			return 1;
314 	}
315 	return 0;
316 }
317 
318 /* call with IRQ disabled */
__netpoll_send_skb(struct netpoll * np,struct sk_buff * skb)319 static netdev_tx_t __netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
320 {
321 	netdev_tx_t status = NETDEV_TX_BUSY;
322 	netdev_tx_t ret = NET_XMIT_DROP;
323 	struct net_device *dev;
324 	unsigned long tries;
325 	/* It is up to the caller to keep npinfo alive. */
326 	struct netpoll_info *npinfo;
327 
328 	lockdep_assert_irqs_disabled();
329 
330 	dev = np->dev;
331 	rcu_read_lock();
332 	npinfo = rcu_dereference_bh(dev->npinfo);
333 
334 	if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
335 		dev_kfree_skb_irq(skb);
336 		goto out;
337 	}
338 
339 	/* don't get messages out of order, and no recursion */
340 	if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
341 		struct netdev_queue *txq;
342 
343 		txq = netdev_core_pick_tx(dev, skb, NULL);
344 
345 		/* try until next clock tick */
346 		for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
347 		     tries > 0; --tries) {
348 			if (HARD_TX_TRYLOCK(dev, txq)) {
349 				if (!netif_xmit_stopped(txq))
350 					status = netpoll_start_xmit(skb, dev, txq);
351 
352 				HARD_TX_UNLOCK(dev, txq);
353 
354 				if (dev_xmit_complete(status))
355 					break;
356 
357 			}
358 
359 			/* tickle device maybe there is some cleanup */
360 			netpoll_poll_dev(np->dev);
361 
362 			udelay(USEC_PER_POLL);
363 		}
364 
365 		WARN_ONCE(!irqs_disabled(),
366 			"netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pS)\n",
367 			dev->name, dev->netdev_ops->ndo_start_xmit);
368 
369 	}
370 
371 	if (!dev_xmit_complete(status)) {
372 		skb_queue_tail(&npinfo->txq, skb);
373 		schedule_delayed_work(&npinfo->tx_work,0);
374 	}
375 	ret = NETDEV_TX_OK;
376 out:
377 	rcu_read_unlock();
378 	return ret;
379 }
380 
netpoll_send_skb(struct netpoll * np,struct sk_buff * skb)381 netdev_tx_t netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
382 {
383 	unsigned long flags;
384 	netdev_tx_t ret;
385 
386 	if (unlikely(!np)) {
387 		dev_kfree_skb_irq(skb);
388 		ret = NET_XMIT_DROP;
389 	} else {
390 		local_irq_save(flags);
391 		ret = __netpoll_send_skb(np, skb);
392 		local_irq_restore(flags);
393 	}
394 	return ret;
395 }
396 EXPORT_SYMBOL(netpoll_send_skb);
397 
netpoll_send_udp(struct netpoll * np,const char * msg,int len)398 int netpoll_send_udp(struct netpoll *np, const char *msg, int len)
399 {
400 	int total_len, ip_len, udp_len;
401 	struct sk_buff *skb;
402 	struct udphdr *udph;
403 	struct iphdr *iph;
404 	struct ethhdr *eth;
405 	static atomic_t ip_ident;
406 	struct ipv6hdr *ip6h;
407 
408 	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
409 		WARN_ON_ONCE(!irqs_disabled());
410 
411 	udp_len = len + sizeof(*udph);
412 	if (np->ipv6)
413 		ip_len = udp_len + sizeof(*ip6h);
414 	else
415 		ip_len = udp_len + sizeof(*iph);
416 
417 	total_len = ip_len + LL_RESERVED_SPACE(np->dev);
418 
419 	skb = find_skb(np, total_len + np->dev->needed_tailroom,
420 		       total_len - len);
421 	if (!skb)
422 		return -ENOMEM;
423 
424 	skb_copy_to_linear_data(skb, msg, len);
425 	skb_put(skb, len);
426 
427 	skb_push(skb, sizeof(*udph));
428 	skb_reset_transport_header(skb);
429 	udph = udp_hdr(skb);
430 	udph->source = htons(np->local_port);
431 	udph->dest = htons(np->remote_port);
432 	udph->len = htons(udp_len);
433 
434 	if (np->ipv6) {
435 		udph->check = 0;
436 		udph->check = csum_ipv6_magic(&np->local_ip.in6,
437 					      &np->remote_ip.in6,
438 					      udp_len, IPPROTO_UDP,
439 					      csum_partial(udph, udp_len, 0));
440 		if (udph->check == 0)
441 			udph->check = CSUM_MANGLED_0;
442 
443 		skb_push(skb, sizeof(*ip6h));
444 		skb_reset_network_header(skb);
445 		ip6h = ipv6_hdr(skb);
446 
447 		/* ip6h->version = 6; ip6h->priority = 0; */
448 		*(unsigned char *)ip6h = 0x60;
449 		ip6h->flow_lbl[0] = 0;
450 		ip6h->flow_lbl[1] = 0;
451 		ip6h->flow_lbl[2] = 0;
452 
453 		ip6h->payload_len = htons(sizeof(struct udphdr) + len);
454 		ip6h->nexthdr = IPPROTO_UDP;
455 		ip6h->hop_limit = 32;
456 		ip6h->saddr = np->local_ip.in6;
457 		ip6h->daddr = np->remote_ip.in6;
458 
459 		eth = skb_push(skb, ETH_HLEN);
460 		skb_reset_mac_header(skb);
461 		skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
462 	} else {
463 		udph->check = 0;
464 		udph->check = csum_tcpudp_magic(np->local_ip.ip,
465 						np->remote_ip.ip,
466 						udp_len, IPPROTO_UDP,
467 						csum_partial(udph, udp_len, 0));
468 		if (udph->check == 0)
469 			udph->check = CSUM_MANGLED_0;
470 
471 		skb_push(skb, sizeof(*iph));
472 		skb_reset_network_header(skb);
473 		iph = ip_hdr(skb);
474 
475 		/* iph->version = 4; iph->ihl = 5; */
476 		*(unsigned char *)iph = 0x45;
477 		iph->tos      = 0;
478 		put_unaligned(htons(ip_len), &(iph->tot_len));
479 		iph->id       = htons(atomic_inc_return(&ip_ident));
480 		iph->frag_off = 0;
481 		iph->ttl      = 64;
482 		iph->protocol = IPPROTO_UDP;
483 		iph->check    = 0;
484 		put_unaligned(np->local_ip.ip, &(iph->saddr));
485 		put_unaligned(np->remote_ip.ip, &(iph->daddr));
486 		iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
487 
488 		eth = skb_push(skb, ETH_HLEN);
489 		skb_reset_mac_header(skb);
490 		skb->protocol = eth->h_proto = htons(ETH_P_IP);
491 	}
492 
493 	ether_addr_copy(eth->h_source, np->dev->dev_addr);
494 	ether_addr_copy(eth->h_dest, np->remote_mac);
495 
496 	skb->dev = np->dev;
497 
498 	return (int)netpoll_send_skb(np, skb);
499 }
500 EXPORT_SYMBOL(netpoll_send_udp);
501 
netpoll_print_options(struct netpoll * np)502 void netpoll_print_options(struct netpoll *np)
503 {
504 	np_info(np, "local port %d\n", np->local_port);
505 	if (np->ipv6)
506 		np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
507 	else
508 		np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
509 	np_info(np, "interface '%s'\n", np->dev_name);
510 	np_info(np, "remote port %d\n", np->remote_port);
511 	if (np->ipv6)
512 		np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
513 	else
514 		np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
515 	np_info(np, "remote ethernet address %pM\n", np->remote_mac);
516 }
517 EXPORT_SYMBOL(netpoll_print_options);
518 
netpoll_parse_ip_addr(const char * str,union inet_addr * addr)519 static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
520 {
521 	const char *end;
522 
523 	if (!strchr(str, ':') &&
524 	    in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
525 		if (!*end)
526 			return 0;
527 	}
528 	if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
529 #if IS_ENABLED(CONFIG_IPV6)
530 		if (!*end)
531 			return 1;
532 #else
533 		return -1;
534 #endif
535 	}
536 	return -1;
537 }
538 
skb_pool_flush(struct netpoll * np)539 static void skb_pool_flush(struct netpoll *np)
540 {
541 	struct sk_buff_head *skb_pool;
542 
543 	skb_pool = &np->skb_pool;
544 	skb_queue_purge_reason(skb_pool, SKB_CONSUMED);
545 }
546 
netpoll_parse_options(struct netpoll * np,char * opt)547 int netpoll_parse_options(struct netpoll *np, char *opt)
548 {
549 	char *cur=opt, *delim;
550 	int ipv6;
551 	bool ipversion_set = false;
552 
553 	if (*cur != '@') {
554 		if ((delim = strchr(cur, '@')) == NULL)
555 			goto parse_failed;
556 		*delim = 0;
557 		if (kstrtou16(cur, 10, &np->local_port))
558 			goto parse_failed;
559 		cur = delim;
560 	}
561 	cur++;
562 
563 	if (*cur != '/') {
564 		ipversion_set = true;
565 		if ((delim = strchr(cur, '/')) == NULL)
566 			goto parse_failed;
567 		*delim = 0;
568 		ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
569 		if (ipv6 < 0)
570 			goto parse_failed;
571 		else
572 			np->ipv6 = (bool)ipv6;
573 		cur = delim;
574 	}
575 	cur++;
576 
577 	if (*cur != ',') {
578 		/* parse out dev name */
579 		if ((delim = strchr(cur, ',')) == NULL)
580 			goto parse_failed;
581 		*delim = 0;
582 		strscpy(np->dev_name, cur, sizeof(np->dev_name));
583 		cur = delim;
584 	}
585 	cur++;
586 
587 	if (*cur != '@') {
588 		/* dst port */
589 		if ((delim = strchr(cur, '@')) == NULL)
590 			goto parse_failed;
591 		*delim = 0;
592 		if (*cur == ' ' || *cur == '\t')
593 			np_info(np, "warning: whitespace is not allowed\n");
594 		if (kstrtou16(cur, 10, &np->remote_port))
595 			goto parse_failed;
596 		cur = delim;
597 	}
598 	cur++;
599 
600 	/* dst ip */
601 	if ((delim = strchr(cur, '/')) == NULL)
602 		goto parse_failed;
603 	*delim = 0;
604 	ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
605 	if (ipv6 < 0)
606 		goto parse_failed;
607 	else if (ipversion_set && np->ipv6 != (bool)ipv6)
608 		goto parse_failed;
609 	else
610 		np->ipv6 = (bool)ipv6;
611 	cur = delim + 1;
612 
613 	if (*cur != 0) {
614 		/* MAC address */
615 		if (!mac_pton(cur, np->remote_mac))
616 			goto parse_failed;
617 	}
618 
619 	netpoll_print_options(np);
620 
621 	return 0;
622 
623  parse_failed:
624 	np_info(np, "couldn't parse config at '%s'!\n", cur);
625 	return -1;
626 }
627 EXPORT_SYMBOL(netpoll_parse_options);
628 
__netpoll_setup(struct netpoll * np,struct net_device * ndev)629 int __netpoll_setup(struct netpoll *np, struct net_device *ndev)
630 {
631 	struct netpoll_info *npinfo;
632 	const struct net_device_ops *ops;
633 	int err;
634 
635 	skb_queue_head_init(&np->skb_pool);
636 
637 	if (ndev->priv_flags & IFF_DISABLE_NETPOLL) {
638 		np_err(np, "%s doesn't support polling, aborting\n",
639 		       ndev->name);
640 		err = -ENOTSUPP;
641 		goto out;
642 	}
643 
644 	npinfo = rtnl_dereference(ndev->npinfo);
645 	if (!npinfo) {
646 		npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
647 		if (!npinfo) {
648 			err = -ENOMEM;
649 			goto out;
650 		}
651 
652 		sema_init(&npinfo->dev_lock, 1);
653 		skb_queue_head_init(&npinfo->txq);
654 		INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
655 
656 		refcount_set(&npinfo->refcnt, 1);
657 
658 		ops = ndev->netdev_ops;
659 		if (ops->ndo_netpoll_setup) {
660 			err = ops->ndo_netpoll_setup(ndev);
661 			if (err)
662 				goto free_npinfo;
663 		}
664 	} else {
665 		refcount_inc(&npinfo->refcnt);
666 	}
667 
668 	np->dev = ndev;
669 	strscpy(np->dev_name, ndev->name, IFNAMSIZ);
670 	npinfo->netpoll = np;
671 
672 	/* fill up the skb queue */
673 	refill_skbs(np);
674 
675 	/* last thing to do is link it to the net device structure */
676 	rcu_assign_pointer(ndev->npinfo, npinfo);
677 
678 	return 0;
679 
680 free_npinfo:
681 	kfree(npinfo);
682 out:
683 	return err;
684 }
685 EXPORT_SYMBOL_GPL(__netpoll_setup);
686 
netpoll_setup(struct netpoll * np)687 int netpoll_setup(struct netpoll *np)
688 {
689 	struct net_device *ndev = NULL;
690 	bool ip_overwritten = false;
691 	struct in_device *in_dev;
692 	int err;
693 
694 	rtnl_lock();
695 	if (np->dev_name[0]) {
696 		struct net *net = current->nsproxy->net_ns;
697 		ndev = __dev_get_by_name(net, np->dev_name);
698 	}
699 	if (!ndev) {
700 		np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
701 		err = -ENODEV;
702 		goto unlock;
703 	}
704 	netdev_hold(ndev, &np->dev_tracker, GFP_KERNEL);
705 
706 	if (netdev_master_upper_dev_get(ndev)) {
707 		np_err(np, "%s is a slave device, aborting\n", np->dev_name);
708 		err = -EBUSY;
709 		goto put;
710 	}
711 
712 	if (!netif_running(ndev)) {
713 		unsigned long atmost;
714 
715 		np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
716 
717 		err = dev_open(ndev, NULL);
718 
719 		if (err) {
720 			np_err(np, "failed to open %s\n", ndev->name);
721 			goto put;
722 		}
723 
724 		rtnl_unlock();
725 		atmost = jiffies + carrier_timeout * HZ;
726 		while (!netif_carrier_ok(ndev)) {
727 			if (time_after(jiffies, atmost)) {
728 				np_notice(np, "timeout waiting for carrier\n");
729 				break;
730 			}
731 			msleep(1);
732 		}
733 
734 		rtnl_lock();
735 	}
736 
737 	if (!np->local_ip.ip) {
738 		if (!np->ipv6) {
739 			const struct in_ifaddr *ifa;
740 
741 			in_dev = __in_dev_get_rtnl(ndev);
742 			if (!in_dev)
743 				goto put_noaddr;
744 
745 			ifa = rtnl_dereference(in_dev->ifa_list);
746 			if (!ifa) {
747 put_noaddr:
748 				np_err(np, "no IP address for %s, aborting\n",
749 				       np->dev_name);
750 				err = -EDESTADDRREQ;
751 				goto put;
752 			}
753 
754 			np->local_ip.ip = ifa->ifa_local;
755 			ip_overwritten = true;
756 			np_info(np, "local IP %pI4\n", &np->local_ip.ip);
757 		} else {
758 #if IS_ENABLED(CONFIG_IPV6)
759 			struct inet6_dev *idev;
760 
761 			err = -EDESTADDRREQ;
762 			idev = __in6_dev_get(ndev);
763 			if (idev) {
764 				struct inet6_ifaddr *ifp;
765 
766 				read_lock_bh(&idev->lock);
767 				list_for_each_entry(ifp, &idev->addr_list, if_list) {
768 					if (!!(ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL) !=
769 					    !!(ipv6_addr_type(&np->remote_ip.in6) & IPV6_ADDR_LINKLOCAL))
770 						continue;
771 					np->local_ip.in6 = ifp->addr;
772 					ip_overwritten = true;
773 					err = 0;
774 					break;
775 				}
776 				read_unlock_bh(&idev->lock);
777 			}
778 			if (err) {
779 				np_err(np, "no IPv6 address for %s, aborting\n",
780 				       np->dev_name);
781 				goto put;
782 			} else
783 				np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
784 #else
785 			np_err(np, "IPv6 is not supported %s, aborting\n",
786 			       np->dev_name);
787 			err = -EINVAL;
788 			goto put;
789 #endif
790 		}
791 	}
792 
793 	err = __netpoll_setup(np, ndev);
794 	if (err)
795 		goto flush;
796 	rtnl_unlock();
797 	return 0;
798 
799 flush:
800 	skb_pool_flush(np);
801 put:
802 	DEBUG_NET_WARN_ON_ONCE(np->dev);
803 	if (ip_overwritten)
804 		memset(&np->local_ip, 0, sizeof(np->local_ip));
805 	netdev_put(ndev, &np->dev_tracker);
806 unlock:
807 	rtnl_unlock();
808 	return err;
809 }
810 EXPORT_SYMBOL(netpoll_setup);
811 
rcu_cleanup_netpoll_info(struct rcu_head * rcu_head)812 static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
813 {
814 	struct netpoll_info *npinfo =
815 			container_of(rcu_head, struct netpoll_info, rcu);
816 
817 	skb_queue_purge(&npinfo->txq);
818 
819 	/* we can't call cancel_delayed_work_sync here, as we are in softirq */
820 	cancel_delayed_work(&npinfo->tx_work);
821 
822 	/* clean after last, unfinished work */
823 	__skb_queue_purge(&npinfo->txq);
824 	/* now cancel it again */
825 	cancel_delayed_work(&npinfo->tx_work);
826 	kfree(npinfo);
827 }
828 
__netpoll_cleanup(struct netpoll * np)829 void __netpoll_cleanup(struct netpoll *np)
830 {
831 	struct netpoll_info *npinfo;
832 
833 	npinfo = rtnl_dereference(np->dev->npinfo);
834 	if (!npinfo)
835 		return;
836 
837 	if (refcount_dec_and_test(&npinfo->refcnt)) {
838 		const struct net_device_ops *ops;
839 
840 		ops = np->dev->netdev_ops;
841 		if (ops->ndo_netpoll_cleanup)
842 			ops->ndo_netpoll_cleanup(np->dev);
843 
844 		RCU_INIT_POINTER(np->dev->npinfo, NULL);
845 		call_rcu(&npinfo->rcu, rcu_cleanup_netpoll_info);
846 	} else
847 		RCU_INIT_POINTER(np->dev->npinfo, NULL);
848 
849 	skb_pool_flush(np);
850 }
851 EXPORT_SYMBOL_GPL(__netpoll_cleanup);
852 
__netpoll_free(struct netpoll * np)853 void __netpoll_free(struct netpoll *np)
854 {
855 	ASSERT_RTNL();
856 
857 	/* Wait for transmitting packets to finish before freeing. */
858 	synchronize_rcu();
859 	__netpoll_cleanup(np);
860 	kfree(np);
861 }
862 EXPORT_SYMBOL_GPL(__netpoll_free);
863 
do_netpoll_cleanup(struct netpoll * np)864 void do_netpoll_cleanup(struct netpoll *np)
865 {
866 	__netpoll_cleanup(np);
867 	netdev_put(np->dev, &np->dev_tracker);
868 	np->dev = NULL;
869 }
870 EXPORT_SYMBOL(do_netpoll_cleanup);
871 
netpoll_cleanup(struct netpoll * np)872 void netpoll_cleanup(struct netpoll *np)
873 {
874 	rtnl_lock();
875 	if (!np->dev)
876 		goto out;
877 	do_netpoll_cleanup(np);
878 out:
879 	rtnl_unlock();
880 }
881 EXPORT_SYMBOL(netpoll_cleanup);
882