Linux Audio

Check our new training course

Loading...
v5.9
  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 <asm/unaligned.h>
 38#include <trace/events/napi.h>
 39
 40/*
 41 * We maintain a small pool of fully-sized skbs, to make sure the
 42 * message gets out even in extreme OOM situations.
 43 */
 44
 45#define MAX_UDP_CHUNK 1460
 46#define MAX_SKBS 32
 47
 48static struct sk_buff_head skb_pool;
 49
 50DEFINE_STATIC_SRCU(netpoll_srcu);
 51
 52#define USEC_PER_POLL	50
 53
 54#define MAX_SKB_SIZE							\
 55	(sizeof(struct ethhdr) +					\
 56	 sizeof(struct iphdr) +						\
 57	 sizeof(struct udphdr) +					\
 58	 MAX_UDP_CHUNK)
 59
 60static void zap_completion_queue(void);
 61
 62static unsigned int carrier_timeout = 4;
 63module_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
 72static netdev_tx_t netpoll_start_xmit(struct sk_buff *skb,
 73				      struct net_device *dev,
 74				      struct netdev_queue *txq)
 75{
 76	netdev_tx_t status = NETDEV_TX_OK;
 77	netdev_features_t features;
 78
 79	features = netif_skb_features(skb);
 80
 81	if (skb_vlan_tag_present(skb) &&
 82	    !vlan_hw_offload_capable(features, skb->vlan_proto)) {
 83		skb = __vlan_hwaccel_push_inside(skb);
 84		if (unlikely(!skb)) {
 85			/* This is actually a packet drop, but we
 86			 * don't want the code that calls this
 87			 * function to try and operate on a NULL skb.
 88			 */
 89			goto out;
 90		}
 91	}
 92
 93	status = netdev_start_xmit(skb, dev, txq, false);
 94
 95out:
 96	return status;
 97}
 98
 99static void queue_process(struct work_struct *work)
100{
101	struct netpoll_info *npinfo =
102		container_of(work, struct netpoll_info, tx_work.work);
103	struct sk_buff *skb;
104	unsigned long flags;
105
106	while ((skb = skb_dequeue(&npinfo->txq))) {
107		struct net_device *dev = skb->dev;
108		struct netdev_queue *txq;
109		unsigned int q_index;
110
111		if (!netif_device_present(dev) || !netif_running(dev)) {
112			kfree_skb(skb);
113			continue;
114		}
115
116		local_irq_save(flags);
117		/* check if skb->queue_mapping is still valid */
118		q_index = skb_get_queue_mapping(skb);
119		if (unlikely(q_index >= dev->real_num_tx_queues)) {
120			q_index = q_index % dev->real_num_tx_queues;
121			skb_set_queue_mapping(skb, q_index);
122		}
123		txq = netdev_get_tx_queue(dev, q_index);
124		HARD_TX_LOCK(dev, txq, smp_processor_id());
125		if (netif_xmit_frozen_or_stopped(txq) ||
126		    !dev_xmit_complete(netpoll_start_xmit(skb, dev, txq))) {
127			skb_queue_head(&npinfo->txq, skb);
128			HARD_TX_UNLOCK(dev, txq);
129			local_irq_restore(flags);
130
131			schedule_delayed_work(&npinfo->tx_work, HZ/10);
132			return;
133		}
134		HARD_TX_UNLOCK(dev, txq);
135		local_irq_restore(flags);
136	}
137}
138
139static void poll_one_napi(struct napi_struct *napi)
140{
141	int work;
142
143	/* If we set this bit but see that it has already been set,
144	 * that indicates that napi has been disabled and we need
145	 * to abort this operation
146	 */
147	if (test_and_set_bit(NAPI_STATE_NPSVC, &napi->state))
148		return;
149
150	/* We explicilty pass the polling call a budget of 0 to
151	 * indicate that we are clearing the Tx path only.
152	 */
153	work = napi->poll(napi, 0);
154	WARN_ONCE(work, "%pS exceeded budget in poll\n", napi->poll);
155	trace_napi_poll(napi, work, 0);
156
157	clear_bit(NAPI_STATE_NPSVC, &napi->state);
158}
159
160static void poll_napi(struct net_device *dev)
161{
162	struct napi_struct *napi;
163	int cpu = smp_processor_id();
164
165	list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) {
166		if (cmpxchg(&napi->poll_owner, -1, cpu) == -1) {
167			poll_one_napi(napi);
168			smp_store_release(&napi->poll_owner, -1);
169		}
170	}
171}
172
173void netpoll_poll_dev(struct net_device *dev)
174{
175	struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
176	const struct net_device_ops *ops;
177
178	/* Don't do any rx activity if the dev_lock mutex is held
179	 * the dev_open/close paths use this to block netpoll activity
180	 * while changing device state
181	 */
182	if (!ni || down_trylock(&ni->dev_lock))
183		return;
184
185	if (!netif_running(dev)) {
186		up(&ni->dev_lock);
187		return;
188	}
189
190	ops = dev->netdev_ops;
191	if (ops->ndo_poll_controller)
192		ops->ndo_poll_controller(dev);
193
194	poll_napi(dev);
195
196	up(&ni->dev_lock);
197
198	zap_completion_queue();
199}
200EXPORT_SYMBOL(netpoll_poll_dev);
201
202void netpoll_poll_disable(struct net_device *dev)
203{
204	struct netpoll_info *ni;
205	int idx;
206	might_sleep();
207	idx = srcu_read_lock(&netpoll_srcu);
208	ni = srcu_dereference(dev->npinfo, &netpoll_srcu);
209	if (ni)
210		down(&ni->dev_lock);
211	srcu_read_unlock(&netpoll_srcu, idx);
212}
213EXPORT_SYMBOL(netpoll_poll_disable);
214
215void netpoll_poll_enable(struct net_device *dev)
216{
217	struct netpoll_info *ni;
218	rcu_read_lock();
219	ni = rcu_dereference(dev->npinfo);
220	if (ni)
221		up(&ni->dev_lock);
222	rcu_read_unlock();
223}
224EXPORT_SYMBOL(netpoll_poll_enable);
225
226static void refill_skbs(void)
227{
228	struct sk_buff *skb;
229	unsigned long flags;
230
231	spin_lock_irqsave(&skb_pool.lock, flags);
232	while (skb_pool.qlen < MAX_SKBS) {
233		skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
234		if (!skb)
235			break;
236
237		__skb_queue_tail(&skb_pool, skb);
238	}
239	spin_unlock_irqrestore(&skb_pool.lock, flags);
240}
241
242static void zap_completion_queue(void)
243{
244	unsigned long flags;
245	struct softnet_data *sd = &get_cpu_var(softnet_data);
246
247	if (sd->completion_queue) {
248		struct sk_buff *clist;
249
250		local_irq_save(flags);
251		clist = sd->completion_queue;
252		sd->completion_queue = NULL;
253		local_irq_restore(flags);
254
255		while (clist != NULL) {
256			struct sk_buff *skb = clist;
257			clist = clist->next;
258			if (!skb_irq_freeable(skb)) {
259				refcount_set(&skb->users, 1);
260				dev_kfree_skb_any(skb); /* put this one back */
261			} else {
262				__kfree_skb(skb);
263			}
264		}
265	}
266
267	put_cpu_var(softnet_data);
268}
269
270static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
271{
272	int count = 0;
273	struct sk_buff *skb;
274
275	zap_completion_queue();
276	refill_skbs();
277repeat:
278
279	skb = alloc_skb(len, GFP_ATOMIC);
280	if (!skb)
281		skb = skb_dequeue(&skb_pool);
282
283	if (!skb) {
284		if (++count < 10) {
285			netpoll_poll_dev(np->dev);
286			goto repeat;
287		}
288		return NULL;
289	}
290
291	refcount_set(&skb->users, 1);
292	skb_reserve(skb, reserve);
293	return skb;
294}
295
296static int netpoll_owner_active(struct net_device *dev)
297{
298	struct napi_struct *napi;
299
300	list_for_each_entry(napi, &dev->napi_list, dev_list) {
301		if (napi->poll_owner == smp_processor_id())
302			return 1;
303	}
304	return 0;
305}
306
307/* call with IRQ disabled */
308static netdev_tx_t __netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
 
309{
310	netdev_tx_t status = NETDEV_TX_BUSY;
311	struct net_device *dev;
312	unsigned long tries;
313	/* It is up to the caller to keep npinfo alive. */
314	struct netpoll_info *npinfo;
315
316	lockdep_assert_irqs_disabled();
317
318	dev = np->dev;
319	npinfo = rcu_dereference_bh(dev->npinfo);
320
321	if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
322		dev_kfree_skb_irq(skb);
323		return NET_XMIT_DROP;
324	}
325
326	/* don't get messages out of order, and no recursion */
327	if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
328		struct netdev_queue *txq;
329
330		txq = netdev_core_pick_tx(dev, skb, NULL);
331
332		/* try until next clock tick */
333		for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
334		     tries > 0; --tries) {
335			if (HARD_TX_TRYLOCK(dev, txq)) {
336				if (!netif_xmit_stopped(txq))
337					status = netpoll_start_xmit(skb, dev, txq);
338
339				HARD_TX_UNLOCK(dev, txq);
340
341				if (dev_xmit_complete(status))
342					break;
343
344			}
345
346			/* tickle device maybe there is some cleanup */
347			netpoll_poll_dev(np->dev);
348
349			udelay(USEC_PER_POLL);
350		}
351
352		WARN_ONCE(!irqs_disabled(),
353			"netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pS)\n",
354			dev->name, dev->netdev_ops->ndo_start_xmit);
355
356	}
357
358	if (!dev_xmit_complete(status)) {
359		skb_queue_tail(&npinfo->txq, skb);
360		schedule_delayed_work(&npinfo->tx_work,0);
361	}
362	return NETDEV_TX_OK;
363}
364
365netdev_tx_t netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
366{
367	unsigned long flags;
368	netdev_tx_t ret;
369
370	if (unlikely(!np)) {
371		dev_kfree_skb_irq(skb);
372		ret = NET_XMIT_DROP;
373	} else {
374		local_irq_save(flags);
375		ret = __netpoll_send_skb(np, skb);
376		local_irq_restore(flags);
377	}
378	return ret;
379}
380EXPORT_SYMBOL(netpoll_send_skb);
381
382void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
383{
384	int total_len, ip_len, udp_len;
385	struct sk_buff *skb;
386	struct udphdr *udph;
387	struct iphdr *iph;
388	struct ethhdr *eth;
389	static atomic_t ip_ident;
390	struct ipv6hdr *ip6h;
391
392	WARN_ON_ONCE(!irqs_disabled());
393
394	udp_len = len + sizeof(*udph);
395	if (np->ipv6)
396		ip_len = udp_len + sizeof(*ip6h);
397	else
398		ip_len = udp_len + sizeof(*iph);
399
400	total_len = ip_len + LL_RESERVED_SPACE(np->dev);
401
402	skb = find_skb(np, total_len + np->dev->needed_tailroom,
403		       total_len - len);
404	if (!skb)
405		return;
406
407	skb_copy_to_linear_data(skb, msg, len);
408	skb_put(skb, len);
409
410	skb_push(skb, sizeof(*udph));
411	skb_reset_transport_header(skb);
412	udph = udp_hdr(skb);
413	udph->source = htons(np->local_port);
414	udph->dest = htons(np->remote_port);
415	udph->len = htons(udp_len);
416
417	if (np->ipv6) {
418		udph->check = 0;
419		udph->check = csum_ipv6_magic(&np->local_ip.in6,
420					      &np->remote_ip.in6,
421					      udp_len, IPPROTO_UDP,
422					      csum_partial(udph, udp_len, 0));
423		if (udph->check == 0)
424			udph->check = CSUM_MANGLED_0;
425
426		skb_push(skb, sizeof(*ip6h));
427		skb_reset_network_header(skb);
428		ip6h = ipv6_hdr(skb);
429
430		/* ip6h->version = 6; ip6h->priority = 0; */
431		put_unaligned(0x60, (unsigned char *)ip6h);
432		ip6h->flow_lbl[0] = 0;
433		ip6h->flow_lbl[1] = 0;
434		ip6h->flow_lbl[2] = 0;
435
436		ip6h->payload_len = htons(sizeof(struct udphdr) + len);
437		ip6h->nexthdr = IPPROTO_UDP;
438		ip6h->hop_limit = 32;
439		ip6h->saddr = np->local_ip.in6;
440		ip6h->daddr = np->remote_ip.in6;
441
442		eth = skb_push(skb, ETH_HLEN);
443		skb_reset_mac_header(skb);
444		skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
445	} else {
446		udph->check = 0;
447		udph->check = csum_tcpudp_magic(np->local_ip.ip,
448						np->remote_ip.ip,
449						udp_len, IPPROTO_UDP,
450						csum_partial(udph, udp_len, 0));
451		if (udph->check == 0)
452			udph->check = CSUM_MANGLED_0;
453
454		skb_push(skb, sizeof(*iph));
455		skb_reset_network_header(skb);
456		iph = ip_hdr(skb);
457
458		/* iph->version = 4; iph->ihl = 5; */
459		put_unaligned(0x45, (unsigned char *)iph);
460		iph->tos      = 0;
461		put_unaligned(htons(ip_len), &(iph->tot_len));
462		iph->id       = htons(atomic_inc_return(&ip_ident));
463		iph->frag_off = 0;
464		iph->ttl      = 64;
465		iph->protocol = IPPROTO_UDP;
466		iph->check    = 0;
467		put_unaligned(np->local_ip.ip, &(iph->saddr));
468		put_unaligned(np->remote_ip.ip, &(iph->daddr));
469		iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
470
471		eth = skb_push(skb, ETH_HLEN);
472		skb_reset_mac_header(skb);
473		skb->protocol = eth->h_proto = htons(ETH_P_IP);
474	}
475
476	ether_addr_copy(eth->h_source, np->dev->dev_addr);
477	ether_addr_copy(eth->h_dest, np->remote_mac);
478
479	skb->dev = np->dev;
480
481	netpoll_send_skb(np, skb);
482}
483EXPORT_SYMBOL(netpoll_send_udp);
484
485void netpoll_print_options(struct netpoll *np)
486{
487	np_info(np, "local port %d\n", np->local_port);
488	if (np->ipv6)
489		np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
490	else
491		np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
492	np_info(np, "interface '%s'\n", np->dev_name);
493	np_info(np, "remote port %d\n", np->remote_port);
494	if (np->ipv6)
495		np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
496	else
497		np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
498	np_info(np, "remote ethernet address %pM\n", np->remote_mac);
499}
500EXPORT_SYMBOL(netpoll_print_options);
501
502static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
503{
504	const char *end;
505
506	if (!strchr(str, ':') &&
507	    in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
508		if (!*end)
509			return 0;
510	}
511	if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
512#if IS_ENABLED(CONFIG_IPV6)
513		if (!*end)
514			return 1;
515#else
516		return -1;
517#endif
518	}
519	return -1;
520}
521
522int netpoll_parse_options(struct netpoll *np, char *opt)
523{
524	char *cur=opt, *delim;
525	int ipv6;
526	bool ipversion_set = false;
527
528	if (*cur != '@') {
529		if ((delim = strchr(cur, '@')) == NULL)
530			goto parse_failed;
531		*delim = 0;
532		if (kstrtou16(cur, 10, &np->local_port))
533			goto parse_failed;
534		cur = delim;
535	}
536	cur++;
537
538	if (*cur != '/') {
539		ipversion_set = true;
540		if ((delim = strchr(cur, '/')) == NULL)
541			goto parse_failed;
542		*delim = 0;
543		ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
544		if (ipv6 < 0)
545			goto parse_failed;
546		else
547			np->ipv6 = (bool)ipv6;
548		cur = delim;
549	}
550	cur++;
551
552	if (*cur != ',') {
553		/* parse out dev name */
554		if ((delim = strchr(cur, ',')) == NULL)
555			goto parse_failed;
556		*delim = 0;
557		strlcpy(np->dev_name, cur, sizeof(np->dev_name));
558		cur = delim;
559	}
560	cur++;
561
562	if (*cur != '@') {
563		/* dst port */
564		if ((delim = strchr(cur, '@')) == NULL)
565			goto parse_failed;
566		*delim = 0;
567		if (*cur == ' ' || *cur == '\t')
568			np_info(np, "warning: whitespace is not allowed\n");
569		if (kstrtou16(cur, 10, &np->remote_port))
570			goto parse_failed;
571		cur = delim;
572	}
573	cur++;
574
575	/* dst ip */
576	if ((delim = strchr(cur, '/')) == NULL)
577		goto parse_failed;
578	*delim = 0;
579	ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
580	if (ipv6 < 0)
581		goto parse_failed;
582	else if (ipversion_set && np->ipv6 != (bool)ipv6)
583		goto parse_failed;
584	else
585		np->ipv6 = (bool)ipv6;
586	cur = delim + 1;
587
588	if (*cur != 0) {
589		/* MAC address */
590		if (!mac_pton(cur, np->remote_mac))
591			goto parse_failed;
592	}
593
594	netpoll_print_options(np);
595
596	return 0;
597
598 parse_failed:
599	np_info(np, "couldn't parse config at '%s'!\n", cur);
600	return -1;
601}
602EXPORT_SYMBOL(netpoll_parse_options);
603
604int __netpoll_setup(struct netpoll *np, struct net_device *ndev)
605{
606	struct netpoll_info *npinfo;
607	const struct net_device_ops *ops;
608	int err;
609
610	np->dev = ndev;
611	strlcpy(np->dev_name, ndev->name, IFNAMSIZ);
612
613	if (ndev->priv_flags & IFF_DISABLE_NETPOLL) {
614		np_err(np, "%s doesn't support polling, aborting\n",
615		       np->dev_name);
616		err = -ENOTSUPP;
617		goto out;
618	}
619
620	if (!ndev->npinfo) {
621		npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
622		if (!npinfo) {
623			err = -ENOMEM;
624			goto out;
625		}
626
627		sema_init(&npinfo->dev_lock, 1);
628		skb_queue_head_init(&npinfo->txq);
629		INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
630
631		refcount_set(&npinfo->refcnt, 1);
632
633		ops = np->dev->netdev_ops;
634		if (ops->ndo_netpoll_setup) {
635			err = ops->ndo_netpoll_setup(ndev, npinfo);
636			if (err)
637				goto free_npinfo;
638		}
639	} else {
640		npinfo = rtnl_dereference(ndev->npinfo);
641		refcount_inc(&npinfo->refcnt);
642	}
643
644	npinfo->netpoll = np;
645
646	/* last thing to do is link it to the net device structure */
647	rcu_assign_pointer(ndev->npinfo, npinfo);
648
649	return 0;
650
651free_npinfo:
652	kfree(npinfo);
653out:
654	return err;
655}
656EXPORT_SYMBOL_GPL(__netpoll_setup);
657
658int netpoll_setup(struct netpoll *np)
659{
660	struct net_device *ndev = NULL;
661	struct in_device *in_dev;
662	int err;
663
664	rtnl_lock();
665	if (np->dev_name[0]) {
666		struct net *net = current->nsproxy->net_ns;
667		ndev = __dev_get_by_name(net, np->dev_name);
668	}
669	if (!ndev) {
670		np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
671		err = -ENODEV;
672		goto unlock;
673	}
674	dev_hold(ndev);
675
676	if (netdev_master_upper_dev_get(ndev)) {
677		np_err(np, "%s is a slave device, aborting\n", np->dev_name);
678		err = -EBUSY;
679		goto put;
680	}
681
682	if (!netif_running(ndev)) {
683		unsigned long atmost, atleast;
684
685		np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
686
687		err = dev_open(ndev, NULL);
688
689		if (err) {
690			np_err(np, "failed to open %s\n", ndev->name);
691			goto put;
692		}
693
694		rtnl_unlock();
695		atleast = jiffies + HZ/10;
696		atmost = jiffies + carrier_timeout * HZ;
697		while (!netif_carrier_ok(ndev)) {
698			if (time_after(jiffies, atmost)) {
699				np_notice(np, "timeout waiting for carrier\n");
700				break;
701			}
702			msleep(1);
703		}
704
705		/* If carrier appears to come up instantly, we don't
706		 * trust it and pause so that we don't pump all our
707		 * queued console messages into the bitbucket.
708		 */
709
710		if (time_before(jiffies, atleast)) {
711			np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
712			msleep(4000);
713		}
714		rtnl_lock();
715	}
716
717	if (!np->local_ip.ip) {
718		if (!np->ipv6) {
719			const struct in_ifaddr *ifa;
720
721			in_dev = __in_dev_get_rtnl(ndev);
722			if (!in_dev)
723				goto put_noaddr;
724
725			ifa = rtnl_dereference(in_dev->ifa_list);
726			if (!ifa) {
727put_noaddr:
728				np_err(np, "no IP address for %s, aborting\n",
729				       np->dev_name);
730				err = -EDESTADDRREQ;
731				goto put;
732			}
733
734			np->local_ip.ip = ifa->ifa_local;
735			np_info(np, "local IP %pI4\n", &np->local_ip.ip);
736		} else {
737#if IS_ENABLED(CONFIG_IPV6)
738			struct inet6_dev *idev;
739
740			err = -EDESTADDRREQ;
741			idev = __in6_dev_get(ndev);
742			if (idev) {
743				struct inet6_ifaddr *ifp;
744
745				read_lock_bh(&idev->lock);
746				list_for_each_entry(ifp, &idev->addr_list, if_list) {
747					if (!!(ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL) !=
748					    !!(ipv6_addr_type(&np->remote_ip.in6) & IPV6_ADDR_LINKLOCAL))
749						continue;
750					np->local_ip.in6 = ifp->addr;
751					err = 0;
752					break;
753				}
754				read_unlock_bh(&idev->lock);
755			}
756			if (err) {
757				np_err(np, "no IPv6 address for %s, aborting\n",
758				       np->dev_name);
759				goto put;
760			} else
761				np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
762#else
763			np_err(np, "IPv6 is not supported %s, aborting\n",
764			       np->dev_name);
765			err = -EINVAL;
766			goto put;
767#endif
768		}
769	}
770
771	/* fill up the skb queue */
772	refill_skbs();
773
774	err = __netpoll_setup(np, ndev);
775	if (err)
776		goto put;
777
778	rtnl_unlock();
779	return 0;
780
781put:
782	dev_put(ndev);
783unlock:
784	rtnl_unlock();
785	return err;
786}
787EXPORT_SYMBOL(netpoll_setup);
788
789static int __init netpoll_init(void)
790{
791	skb_queue_head_init(&skb_pool);
792	return 0;
793}
794core_initcall(netpoll_init);
795
796static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
797{
798	struct netpoll_info *npinfo =
799			container_of(rcu_head, struct netpoll_info, rcu);
800
801	skb_queue_purge(&npinfo->txq);
802
803	/* we can't call cancel_delayed_work_sync here, as we are in softirq */
804	cancel_delayed_work(&npinfo->tx_work);
805
806	/* clean after last, unfinished work */
807	__skb_queue_purge(&npinfo->txq);
808	/* now cancel it again */
809	cancel_delayed_work(&npinfo->tx_work);
810	kfree(npinfo);
811}
812
813void __netpoll_cleanup(struct netpoll *np)
814{
815	struct netpoll_info *npinfo;
816
817	npinfo = rtnl_dereference(np->dev->npinfo);
818	if (!npinfo)
819		return;
820
821	synchronize_srcu(&netpoll_srcu);
822
823	if (refcount_dec_and_test(&npinfo->refcnt)) {
824		const struct net_device_ops *ops;
825
826		ops = np->dev->netdev_ops;
827		if (ops->ndo_netpoll_cleanup)
828			ops->ndo_netpoll_cleanup(np->dev);
829
830		RCU_INIT_POINTER(np->dev->npinfo, NULL);
831		call_rcu(&npinfo->rcu, rcu_cleanup_netpoll_info);
832	} else
833		RCU_INIT_POINTER(np->dev->npinfo, NULL);
834}
835EXPORT_SYMBOL_GPL(__netpoll_cleanup);
836
837void __netpoll_free(struct netpoll *np)
838{
839	ASSERT_RTNL();
840
841	/* Wait for transmitting packets to finish before freeing. */
842	synchronize_rcu();
843	__netpoll_cleanup(np);
844	kfree(np);
845}
846EXPORT_SYMBOL_GPL(__netpoll_free);
847
848void netpoll_cleanup(struct netpoll *np)
849{
850	rtnl_lock();
851	if (!np->dev)
852		goto out;
853	__netpoll_cleanup(np);
854	dev_put(np->dev);
855	np->dev = NULL;
856out:
857	rtnl_unlock();
858}
859EXPORT_SYMBOL(netpoll_cleanup);
v5.4
  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 <asm/unaligned.h>
 38#include <trace/events/napi.h>
 39
 40/*
 41 * We maintain a small pool of fully-sized skbs, to make sure the
 42 * message gets out even in extreme OOM situations.
 43 */
 44
 45#define MAX_UDP_CHUNK 1460
 46#define MAX_SKBS 32
 47
 48static struct sk_buff_head skb_pool;
 49
 50DEFINE_STATIC_SRCU(netpoll_srcu);
 51
 52#define USEC_PER_POLL	50
 53
 54#define MAX_SKB_SIZE							\
 55	(sizeof(struct ethhdr) +					\
 56	 sizeof(struct iphdr) +						\
 57	 sizeof(struct udphdr) +					\
 58	 MAX_UDP_CHUNK)
 59
 60static void zap_completion_queue(void);
 61
 62static unsigned int carrier_timeout = 4;
 63module_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
 72static 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
 94out:
 95	return status;
 96}
 97
 98static 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		    !dev_xmit_complete(netpoll_start_xmit(skb, dev, txq))) {
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
138static 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, "%pS 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
159static 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
172void 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}
199EXPORT_SYMBOL(netpoll_poll_dev);
200
201void 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}
212EXPORT_SYMBOL(netpoll_poll_disable);
213
214void 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}
223EXPORT_SYMBOL(netpoll_poll_enable);
224
225static 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
241static 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
269static 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();
276repeat:
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
295static 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 */
307void 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_core_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 (dev_xmit_complete(status))
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 (%pS)\n",
351			dev->name, dev->netdev_ops->ndo_start_xmit);
352
353	}
354
355	if (!dev_xmit_complete(status)) {
356		skb_queue_tail(&npinfo->txq, skb);
357		schedule_delayed_work(&npinfo->tx_work,0);
358	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
359}
360EXPORT_SYMBOL(netpoll_send_skb_on_dev);
361
362void 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}
463EXPORT_SYMBOL(netpoll_send_udp);
464
465void 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}
480EXPORT_SYMBOL(netpoll_print_options);
481
482static 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
502int 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}
582EXPORT_SYMBOL(netpoll_parse_options);
583
584int __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
593	if (ndev->priv_flags & IFF_DISABLE_NETPOLL) {
594		np_err(np, "%s doesn't support polling, aborting\n",
595		       np->dev_name);
596		err = -ENOTSUPP;
597		goto out;
598	}
599
600	if (!ndev->npinfo) {
601		npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
602		if (!npinfo) {
603			err = -ENOMEM;
604			goto out;
605		}
606
607		sema_init(&npinfo->dev_lock, 1);
608		skb_queue_head_init(&npinfo->txq);
609		INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
610
611		refcount_set(&npinfo->refcnt, 1);
612
613		ops = np->dev->netdev_ops;
614		if (ops->ndo_netpoll_setup) {
615			err = ops->ndo_netpoll_setup(ndev, npinfo);
616			if (err)
617				goto free_npinfo;
618		}
619	} else {
620		npinfo = rtnl_dereference(ndev->npinfo);
621		refcount_inc(&npinfo->refcnt);
622	}
623
624	npinfo->netpoll = np;
625
626	/* last thing to do is link it to the net device structure */
627	rcu_assign_pointer(ndev->npinfo, npinfo);
628
629	return 0;
630
631free_npinfo:
632	kfree(npinfo);
633out:
634	return err;
635}
636EXPORT_SYMBOL_GPL(__netpoll_setup);
637
638int netpoll_setup(struct netpoll *np)
639{
640	struct net_device *ndev = NULL;
641	struct in_device *in_dev;
642	int err;
643
644	rtnl_lock();
645	if (np->dev_name[0]) {
646		struct net *net = current->nsproxy->net_ns;
647		ndev = __dev_get_by_name(net, np->dev_name);
648	}
649	if (!ndev) {
650		np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
651		err = -ENODEV;
652		goto unlock;
653	}
654	dev_hold(ndev);
655
656	if (netdev_master_upper_dev_get(ndev)) {
657		np_err(np, "%s is a slave device, aborting\n", np->dev_name);
658		err = -EBUSY;
659		goto put;
660	}
661
662	if (!netif_running(ndev)) {
663		unsigned long atmost, atleast;
664
665		np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
666
667		err = dev_open(ndev, NULL);
668
669		if (err) {
670			np_err(np, "failed to open %s\n", ndev->name);
671			goto put;
672		}
673
674		rtnl_unlock();
675		atleast = jiffies + HZ/10;
676		atmost = jiffies + carrier_timeout * HZ;
677		while (!netif_carrier_ok(ndev)) {
678			if (time_after(jiffies, atmost)) {
679				np_notice(np, "timeout waiting for carrier\n");
680				break;
681			}
682			msleep(1);
683		}
684
685		/* If carrier appears to come up instantly, we don't
686		 * trust it and pause so that we don't pump all our
687		 * queued console messages into the bitbucket.
688		 */
689
690		if (time_before(jiffies, atleast)) {
691			np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
692			msleep(4000);
693		}
694		rtnl_lock();
695	}
696
697	if (!np->local_ip.ip) {
698		if (!np->ipv6) {
699			const struct in_ifaddr *ifa;
700
701			in_dev = __in_dev_get_rtnl(ndev);
702			if (!in_dev)
703				goto put_noaddr;
704
705			ifa = rtnl_dereference(in_dev->ifa_list);
706			if (!ifa) {
707put_noaddr:
708				np_err(np, "no IP address for %s, aborting\n",
709				       np->dev_name);
710				err = -EDESTADDRREQ;
711				goto put;
712			}
713
714			np->local_ip.ip = ifa->ifa_local;
715			np_info(np, "local IP %pI4\n", &np->local_ip.ip);
716		} else {
717#if IS_ENABLED(CONFIG_IPV6)
718			struct inet6_dev *idev;
719
720			err = -EDESTADDRREQ;
721			idev = __in6_dev_get(ndev);
722			if (idev) {
723				struct inet6_ifaddr *ifp;
724
725				read_lock_bh(&idev->lock);
726				list_for_each_entry(ifp, &idev->addr_list, if_list) {
727					if (!!(ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL) !=
728					    !!(ipv6_addr_type(&np->remote_ip.in6) & IPV6_ADDR_LINKLOCAL))
729						continue;
730					np->local_ip.in6 = ifp->addr;
731					err = 0;
732					break;
733				}
734				read_unlock_bh(&idev->lock);
735			}
736			if (err) {
737				np_err(np, "no IPv6 address for %s, aborting\n",
738				       np->dev_name);
739				goto put;
740			} else
741				np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
742#else
743			np_err(np, "IPv6 is not supported %s, aborting\n",
744			       np->dev_name);
745			err = -EINVAL;
746			goto put;
747#endif
748		}
749	}
750
751	/* fill up the skb queue */
752	refill_skbs();
753
754	err = __netpoll_setup(np, ndev);
755	if (err)
756		goto put;
757
758	rtnl_unlock();
759	return 0;
760
761put:
762	dev_put(ndev);
763unlock:
764	rtnl_unlock();
765	return err;
766}
767EXPORT_SYMBOL(netpoll_setup);
768
769static int __init netpoll_init(void)
770{
771	skb_queue_head_init(&skb_pool);
772	return 0;
773}
774core_initcall(netpoll_init);
775
776static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
777{
778	struct netpoll_info *npinfo =
779			container_of(rcu_head, struct netpoll_info, rcu);
780
781	skb_queue_purge(&npinfo->txq);
782
783	/* we can't call cancel_delayed_work_sync here, as we are in softirq */
784	cancel_delayed_work(&npinfo->tx_work);
785
786	/* clean after last, unfinished work */
787	__skb_queue_purge(&npinfo->txq);
788	/* now cancel it again */
789	cancel_delayed_work(&npinfo->tx_work);
790	kfree(npinfo);
791}
792
793void __netpoll_cleanup(struct netpoll *np)
794{
795	struct netpoll_info *npinfo;
796
797	npinfo = rtnl_dereference(np->dev->npinfo);
798	if (!npinfo)
799		return;
800
801	synchronize_srcu(&netpoll_srcu);
802
803	if (refcount_dec_and_test(&npinfo->refcnt)) {
804		const struct net_device_ops *ops;
805
806		ops = np->dev->netdev_ops;
807		if (ops->ndo_netpoll_cleanup)
808			ops->ndo_netpoll_cleanup(np->dev);
809
810		RCU_INIT_POINTER(np->dev->npinfo, NULL);
811		call_rcu(&npinfo->rcu, rcu_cleanup_netpoll_info);
812	} else
813		RCU_INIT_POINTER(np->dev->npinfo, NULL);
814}
815EXPORT_SYMBOL_GPL(__netpoll_cleanup);
816
817void __netpoll_free(struct netpoll *np)
818{
819	ASSERT_RTNL();
820
821	/* Wait for transmitting packets to finish before freeing. */
822	synchronize_rcu();
823	__netpoll_cleanup(np);
824	kfree(np);
825}
826EXPORT_SYMBOL_GPL(__netpoll_free);
827
828void netpoll_cleanup(struct netpoll *np)
829{
830	rtnl_lock();
831	if (!np->dev)
832		goto out;
833	__netpoll_cleanup(np);
834	dev_put(np->dev);
835	np->dev = NULL;
836out:
837	rtnl_unlock();
838}
839EXPORT_SYMBOL(netpoll_cleanup);