Linux Audio

Check our new training course

Loading...
v5.14.15
  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#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
 49static struct sk_buff_head skb_pool;
 50
 51DEFINE_STATIC_SRCU(netpoll_srcu);
 52
 53#define USEC_PER_POLL	50
 54
 55#define MAX_SKB_SIZE							\
 56	(sizeof(struct ethhdr) +					\
 57	 sizeof(struct iphdr) +						\
 58	 sizeof(struct udphdr) +					\
 59	 MAX_UDP_CHUNK)
 60
 61static void zap_completion_queue(void);
 62
 63static unsigned int carrier_timeout = 4;
 64module_param(carrier_timeout, uint, 0644);
 65
 66#define np_info(np, fmt, ...)				\
 67	pr_info("%s: " fmt, np->name, ##__VA_ARGS__)
 68#define np_err(np, fmt, ...)				\
 69	pr_err("%s: " fmt, np->name, ##__VA_ARGS__)
 70#define np_notice(np, fmt, ...)				\
 71	pr_notice("%s: " fmt, np->name, ##__VA_ARGS__)
 72
 73static netdev_tx_t netpoll_start_xmit(struct sk_buff *skb,
 74				      struct net_device *dev,
 75				      struct netdev_queue *txq)
 76{
 77	netdev_tx_t status = NETDEV_TX_OK;
 78	netdev_features_t features;
 79
 80	features = netif_skb_features(skb);
 81
 82	if (skb_vlan_tag_present(skb) &&
 83	    !vlan_hw_offload_capable(features, skb->vlan_proto)) {
 84		skb = __vlan_hwaccel_push_inside(skb);
 85		if (unlikely(!skb)) {
 86			/* This is actually a packet drop, but we
 87			 * don't want the code that calls this
 88			 * function to try and operate on a NULL skb.
 89			 */
 90			goto out;
 91		}
 92	}
 93
 94	status = netdev_start_xmit(skb, dev, txq, false);
 95
 96out:
 97	return status;
 98}
 99
100static void queue_process(struct work_struct *work)
101{
102	struct netpoll_info *npinfo =
103		container_of(work, struct netpoll_info, tx_work.work);
104	struct sk_buff *skb;
105	unsigned long flags;
106
107	while ((skb = skb_dequeue(&npinfo->txq))) {
108		struct net_device *dev = skb->dev;
109		struct netdev_queue *txq;
110		unsigned int q_index;
111
112		if (!netif_device_present(dev) || !netif_running(dev)) {
113			kfree_skb(skb);
114			continue;
115		}
116
117		local_irq_save(flags);
118		/* check if skb->queue_mapping is still valid */
119		q_index = skb_get_queue_mapping(skb);
120		if (unlikely(q_index >= dev->real_num_tx_queues)) {
121			q_index = q_index % dev->real_num_tx_queues;
122			skb_set_queue_mapping(skb, q_index);
123		}
124		txq = netdev_get_tx_queue(dev, q_index);
125		HARD_TX_LOCK(dev, txq, smp_processor_id());
126		if (netif_xmit_frozen_or_stopped(txq) ||
127		    !dev_xmit_complete(netpoll_start_xmit(skb, dev, txq))) {
128			skb_queue_head(&npinfo->txq, skb);
129			HARD_TX_UNLOCK(dev, txq);
130			local_irq_restore(flags);
131
132			schedule_delayed_work(&npinfo->tx_work, HZ/10);
133			return;
134		}
135		HARD_TX_UNLOCK(dev, txq);
136		local_irq_restore(flags);
137	}
138}
139
140static void poll_one_napi(struct napi_struct *napi)
141{
142	int work;
143
144	/* If we set this bit but see that it has already been set,
145	 * that indicates that napi has been disabled and we need
146	 * to abort this operation
147	 */
148	if (test_and_set_bit(NAPI_STATE_NPSVC, &napi->state))
149		return;
150
151	/* We explicilty pass the polling call a budget of 0 to
152	 * indicate that we are clearing the Tx path only.
153	 */
154	work = napi->poll(napi, 0);
155	WARN_ONCE(work, "%pS exceeded budget in poll\n", napi->poll);
156	trace_napi_poll(napi, work, 0);
157
158	clear_bit(NAPI_STATE_NPSVC, &napi->state);
159}
160
161static void poll_napi(struct net_device *dev)
162{
163	struct napi_struct *napi;
164	int cpu = smp_processor_id();
165
166	list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) {
167		if (cmpxchg(&napi->poll_owner, -1, cpu) == -1) {
168			poll_one_napi(napi);
169			smp_store_release(&napi->poll_owner, -1);
170		}
171	}
172}
173
174void netpoll_poll_dev(struct net_device *dev)
175{
176	struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
177	const struct net_device_ops *ops;
178
179	/* Don't do any rx activity if the dev_lock mutex is held
180	 * the dev_open/close paths use this to block netpoll activity
181	 * while changing device state
182	 */
183	if (!ni || down_trylock(&ni->dev_lock))
184		return;
185
186	if (!netif_running(dev)) {
187		up(&ni->dev_lock);
188		return;
189	}
190
191	ops = dev->netdev_ops;
192	if (ops->ndo_poll_controller)
193		ops->ndo_poll_controller(dev);
194
195	poll_napi(dev);
196
197	up(&ni->dev_lock);
198
199	zap_completion_queue();
200}
201EXPORT_SYMBOL(netpoll_poll_dev);
202
203void netpoll_poll_disable(struct net_device *dev)
204{
205	struct netpoll_info *ni;
206	int idx;
207	might_sleep();
208	idx = srcu_read_lock(&netpoll_srcu);
209	ni = srcu_dereference(dev->npinfo, &netpoll_srcu);
210	if (ni)
211		down(&ni->dev_lock);
212	srcu_read_unlock(&netpoll_srcu, idx);
213}
214EXPORT_SYMBOL(netpoll_poll_disable);
215
216void netpoll_poll_enable(struct net_device *dev)
217{
218	struct netpoll_info *ni;
219	rcu_read_lock();
220	ni = rcu_dereference(dev->npinfo);
221	if (ni)
222		up(&ni->dev_lock);
223	rcu_read_unlock();
224}
225EXPORT_SYMBOL(netpoll_poll_enable);
226
227static void refill_skbs(void)
228{
229	struct sk_buff *skb;
230	unsigned long flags;
231
232	spin_lock_irqsave(&skb_pool.lock, flags);
233	while (skb_pool.qlen < MAX_SKBS) {
234		skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
235		if (!skb)
236			break;
237
238		__skb_queue_tail(&skb_pool, skb);
239	}
240	spin_unlock_irqrestore(&skb_pool.lock, flags);
241}
242
243static void zap_completion_queue(void)
244{
245	unsigned long flags;
246	struct softnet_data *sd = &get_cpu_var(softnet_data);
247
248	if (sd->completion_queue) {
249		struct sk_buff *clist;
250
251		local_irq_save(flags);
252		clist = sd->completion_queue;
253		sd->completion_queue = NULL;
254		local_irq_restore(flags);
255
256		while (clist != NULL) {
257			struct sk_buff *skb = clist;
258			clist = clist->next;
259			if (!skb_irq_freeable(skb)) {
260				refcount_set(&skb->users, 1);
261				dev_kfree_skb_any(skb); /* put this one back */
262			} else {
263				__kfree_skb(skb);
264			}
265		}
266	}
267
268	put_cpu_var(softnet_data);
269}
270
271static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
272{
273	int count = 0;
274	struct sk_buff *skb;
275
276	zap_completion_queue();
277	refill_skbs();
278repeat:
279
280	skb = alloc_skb(len, GFP_ATOMIC);
281	if (!skb)
282		skb = skb_dequeue(&skb_pool);
283
284	if (!skb) {
285		if (++count < 10) {
286			netpoll_poll_dev(np->dev);
287			goto repeat;
288		}
289		return NULL;
290	}
291
292	refcount_set(&skb->users, 1);
293	skb_reserve(skb, reserve);
294	return skb;
295}
296
297static int netpoll_owner_active(struct net_device *dev)
298{
299	struct napi_struct *napi;
300
301	list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) {
302		if (napi->poll_owner == smp_processor_id())
303			return 1;
304	}
305	return 0;
306}
307
308/* call with IRQ disabled */
309static netdev_tx_t __netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
310{
311	netdev_tx_t status = NETDEV_TX_BUSY;
312	struct net_device *dev;
313	unsigned long tries;
314	/* It is up to the caller to keep npinfo alive. */
315	struct netpoll_info *npinfo;
316
317	lockdep_assert_irqs_disabled();
318
319	dev = np->dev;
320	npinfo = rcu_dereference_bh(dev->npinfo);
321
322	if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
323		dev_kfree_skb_irq(skb);
324		return NET_XMIT_DROP;
325	}
326
327	/* don't get messages out of order, and no recursion */
328	if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
329		struct netdev_queue *txq;
330
331		txq = netdev_core_pick_tx(dev, skb, NULL);
332
333		/* try until next clock tick */
334		for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
335		     tries > 0; --tries) {
336			if (HARD_TX_TRYLOCK(dev, txq)) {
337				if (!netif_xmit_stopped(txq))
338					status = netpoll_start_xmit(skb, dev, txq);
339
340				HARD_TX_UNLOCK(dev, txq);
341
342				if (dev_xmit_complete(status))
343					break;
344
345			}
346
347			/* tickle device maybe there is some cleanup */
348			netpoll_poll_dev(np->dev);
349
350			udelay(USEC_PER_POLL);
351		}
352
353		WARN_ONCE(!irqs_disabled(),
354			"netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pS)\n",
355			dev->name, dev->netdev_ops->ndo_start_xmit);
356
357	}
358
359	if (!dev_xmit_complete(status)) {
360		skb_queue_tail(&npinfo->txq, skb);
361		schedule_delayed_work(&npinfo->tx_work,0);
362	}
363	return NETDEV_TX_OK;
364}
365
366netdev_tx_t netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
367{
368	unsigned long flags;
369	netdev_tx_t ret;
370
371	if (unlikely(!np)) {
372		dev_kfree_skb_irq(skb);
373		ret = NET_XMIT_DROP;
374	} else {
375		local_irq_save(flags);
376		ret = __netpoll_send_skb(np, skb);
377		local_irq_restore(flags);
378	}
379	return ret;
380}
381EXPORT_SYMBOL(netpoll_send_skb);
382
383void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
384{
385	int total_len, ip_len, udp_len;
386	struct sk_buff *skb;
387	struct udphdr *udph;
388	struct iphdr *iph;
389	struct ethhdr *eth;
390	static atomic_t ip_ident;
391	struct ipv6hdr *ip6h;
392
393	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
394		WARN_ON_ONCE(!irqs_disabled());
395
396	udp_len = len + sizeof(*udph);
397	if (np->ipv6)
398		ip_len = udp_len + sizeof(*ip6h);
399	else
400		ip_len = udp_len + sizeof(*iph);
401
402	total_len = ip_len + LL_RESERVED_SPACE(np->dev);
403
404	skb = find_skb(np, total_len + np->dev->needed_tailroom,
405		       total_len - len);
406	if (!skb)
407		return;
408
409	skb_copy_to_linear_data(skb, msg, len);
410	skb_put(skb, len);
411
412	skb_push(skb, sizeof(*udph));
413	skb_reset_transport_header(skb);
414	udph = udp_hdr(skb);
415	udph->source = htons(np->local_port);
416	udph->dest = htons(np->remote_port);
417	udph->len = htons(udp_len);
418
419	if (np->ipv6) {
420		udph->check = 0;
421		udph->check = csum_ipv6_magic(&np->local_ip.in6,
422					      &np->remote_ip.in6,
423					      udp_len, IPPROTO_UDP,
424					      csum_partial(udph, udp_len, 0));
425		if (udph->check == 0)
426			udph->check = CSUM_MANGLED_0;
427
428		skb_push(skb, sizeof(*ip6h));
429		skb_reset_network_header(skb);
430		ip6h = ipv6_hdr(skb);
431
432		/* ip6h->version = 6; ip6h->priority = 0; */
433		*(unsigned char *)ip6h = 0x60;
434		ip6h->flow_lbl[0] = 0;
435		ip6h->flow_lbl[1] = 0;
436		ip6h->flow_lbl[2] = 0;
437
438		ip6h->payload_len = htons(sizeof(struct udphdr) + len);
439		ip6h->nexthdr = IPPROTO_UDP;
440		ip6h->hop_limit = 32;
441		ip6h->saddr = np->local_ip.in6;
442		ip6h->daddr = np->remote_ip.in6;
443
444		eth = skb_push(skb, ETH_HLEN);
445		skb_reset_mac_header(skb);
446		skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
447	} else {
448		udph->check = 0;
449		udph->check = csum_tcpudp_magic(np->local_ip.ip,
450						np->remote_ip.ip,
451						udp_len, IPPROTO_UDP,
452						csum_partial(udph, udp_len, 0));
453		if (udph->check == 0)
454			udph->check = CSUM_MANGLED_0;
455
456		skb_push(skb, sizeof(*iph));
457		skb_reset_network_header(skb);
458		iph = ip_hdr(skb);
459
460		/* iph->version = 4; iph->ihl = 5; */
461		*(unsigned char *)iph = 0x45;
462		iph->tos      = 0;
463		put_unaligned(htons(ip_len), &(iph->tot_len));
464		iph->id       = htons(atomic_inc_return(&ip_ident));
465		iph->frag_off = 0;
466		iph->ttl      = 64;
467		iph->protocol = IPPROTO_UDP;
468		iph->check    = 0;
469		put_unaligned(np->local_ip.ip, &(iph->saddr));
470		put_unaligned(np->remote_ip.ip, &(iph->daddr));
471		iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
472
473		eth = skb_push(skb, ETH_HLEN);
474		skb_reset_mac_header(skb);
475		skb->protocol = eth->h_proto = htons(ETH_P_IP);
476	}
477
478	ether_addr_copy(eth->h_source, np->dev->dev_addr);
479	ether_addr_copy(eth->h_dest, np->remote_mac);
480
481	skb->dev = np->dev;
482
483	netpoll_send_skb(np, skb);
484}
485EXPORT_SYMBOL(netpoll_send_udp);
486
487void netpoll_print_options(struct netpoll *np)
488{
489	np_info(np, "local port %d\n", np->local_port);
490	if (np->ipv6)
491		np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
492	else
493		np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
494	np_info(np, "interface '%s'\n", np->dev_name);
495	np_info(np, "remote port %d\n", np->remote_port);
496	if (np->ipv6)
497		np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
498	else
499		np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
500	np_info(np, "remote ethernet address %pM\n", np->remote_mac);
501}
502EXPORT_SYMBOL(netpoll_print_options);
503
504static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
505{
506	const char *end;
507
508	if (!strchr(str, ':') &&
509	    in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
510		if (!*end)
511			return 0;
512	}
513	if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
514#if IS_ENABLED(CONFIG_IPV6)
515		if (!*end)
516			return 1;
517#else
518		return -1;
519#endif
520	}
521	return -1;
522}
523
524int netpoll_parse_options(struct netpoll *np, char *opt)
525{
526	char *cur=opt, *delim;
527	int ipv6;
528	bool ipversion_set = false;
529
530	if (*cur != '@') {
531		if ((delim = strchr(cur, '@')) == NULL)
532			goto parse_failed;
533		*delim = 0;
534		if (kstrtou16(cur, 10, &np->local_port))
535			goto parse_failed;
536		cur = delim;
537	}
538	cur++;
539
540	if (*cur != '/') {
541		ipversion_set = true;
542		if ((delim = strchr(cur, '/')) == NULL)
543			goto parse_failed;
544		*delim = 0;
545		ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
546		if (ipv6 < 0)
547			goto parse_failed;
548		else
549			np->ipv6 = (bool)ipv6;
550		cur = delim;
551	}
552	cur++;
553
554	if (*cur != ',') {
555		/* parse out dev name */
556		if ((delim = strchr(cur, ',')) == NULL)
557			goto parse_failed;
558		*delim = 0;
559		strlcpy(np->dev_name, cur, sizeof(np->dev_name));
560		cur = delim;
561	}
562	cur++;
563
564	if (*cur != '@') {
565		/* dst port */
566		if ((delim = strchr(cur, '@')) == NULL)
567			goto parse_failed;
568		*delim = 0;
569		if (*cur == ' ' || *cur == '\t')
570			np_info(np, "warning: whitespace is not allowed\n");
571		if (kstrtou16(cur, 10, &np->remote_port))
572			goto parse_failed;
573		cur = delim;
574	}
575	cur++;
576
577	/* dst ip */
578	if ((delim = strchr(cur, '/')) == NULL)
579		goto parse_failed;
580	*delim = 0;
581	ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
582	if (ipv6 < 0)
583		goto parse_failed;
584	else if (ipversion_set && np->ipv6 != (bool)ipv6)
585		goto parse_failed;
586	else
587		np->ipv6 = (bool)ipv6;
588	cur = delim + 1;
589
590	if (*cur != 0) {
591		/* MAC address */
592		if (!mac_pton(cur, np->remote_mac))
593			goto parse_failed;
594	}
595
596	netpoll_print_options(np);
597
598	return 0;
599
600 parse_failed:
601	np_info(np, "couldn't parse config at '%s'!\n", cur);
602	return -1;
603}
604EXPORT_SYMBOL(netpoll_parse_options);
605
606int __netpoll_setup(struct netpoll *np, struct net_device *ndev)
607{
608	struct netpoll_info *npinfo;
609	const struct net_device_ops *ops;
610	int err;
611
612	np->dev = ndev;
613	strlcpy(np->dev_name, ndev->name, IFNAMSIZ);
614
615	if (ndev->priv_flags & IFF_DISABLE_NETPOLL) {
616		np_err(np, "%s doesn't support polling, aborting\n",
617		       np->dev_name);
618		err = -ENOTSUPP;
619		goto out;
620	}
621
622	if (!ndev->npinfo) {
623		npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
624		if (!npinfo) {
625			err = -ENOMEM;
626			goto out;
627		}
628
629		sema_init(&npinfo->dev_lock, 1);
630		skb_queue_head_init(&npinfo->txq);
631		INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
632
633		refcount_set(&npinfo->refcnt, 1);
634
635		ops = np->dev->netdev_ops;
636		if (ops->ndo_netpoll_setup) {
637			err = ops->ndo_netpoll_setup(ndev, npinfo);
638			if (err)
639				goto free_npinfo;
640		}
641	} else {
642		npinfo = rtnl_dereference(ndev->npinfo);
643		refcount_inc(&npinfo->refcnt);
644	}
645
646	npinfo->netpoll = np;
647
648	/* last thing to do is link it to the net device structure */
649	rcu_assign_pointer(ndev->npinfo, npinfo);
650
651	return 0;
652
653free_npinfo:
654	kfree(npinfo);
655out:
656	return err;
657}
658EXPORT_SYMBOL_GPL(__netpoll_setup);
659
660int netpoll_setup(struct netpoll *np)
661{
662	struct net_device *ndev = NULL;
663	struct in_device *in_dev;
664	int err;
665
666	rtnl_lock();
667	if (np->dev_name[0]) {
668		struct net *net = current->nsproxy->net_ns;
669		ndev = __dev_get_by_name(net, np->dev_name);
670	}
671	if (!ndev) {
672		np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
673		err = -ENODEV;
674		goto unlock;
675	}
676	dev_hold(ndev);
677
678	if (netdev_master_upper_dev_get(ndev)) {
679		np_err(np, "%s is a slave device, aborting\n", np->dev_name);
680		err = -EBUSY;
681		goto put;
682	}
683
684	if (!netif_running(ndev)) {
685		unsigned long atmost, atleast;
686
687		np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
688
689		err = dev_open(ndev, NULL);
690
691		if (err) {
692			np_err(np, "failed to open %s\n", ndev->name);
693			goto put;
694		}
695
696		rtnl_unlock();
697		atleast = jiffies + HZ/10;
698		atmost = jiffies + carrier_timeout * HZ;
699		while (!netif_carrier_ok(ndev)) {
700			if (time_after(jiffies, atmost)) {
701				np_notice(np, "timeout waiting for carrier\n");
702				break;
703			}
704			msleep(1);
705		}
706
707		/* If carrier appears to come up instantly, we don't
708		 * trust it and pause so that we don't pump all our
709		 * queued console messages into the bitbucket.
710		 */
711
712		if (time_before(jiffies, atleast)) {
713			np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
714			msleep(4000);
715		}
716		rtnl_lock();
717	}
718
719	if (!np->local_ip.ip) {
720		if (!np->ipv6) {
721			const struct in_ifaddr *ifa;
722
723			in_dev = __in_dev_get_rtnl(ndev);
724			if (!in_dev)
725				goto put_noaddr;
726
727			ifa = rtnl_dereference(in_dev->ifa_list);
728			if (!ifa) {
729put_noaddr:
730				np_err(np, "no IP address for %s, aborting\n",
731				       np->dev_name);
732				err = -EDESTADDRREQ;
733				goto put;
734			}
735
736			np->local_ip.ip = ifa->ifa_local;
737			np_info(np, "local IP %pI4\n", &np->local_ip.ip);
738		} else {
739#if IS_ENABLED(CONFIG_IPV6)
740			struct inet6_dev *idev;
741
742			err = -EDESTADDRREQ;
743			idev = __in6_dev_get(ndev);
744			if (idev) {
745				struct inet6_ifaddr *ifp;
746
747				read_lock_bh(&idev->lock);
748				list_for_each_entry(ifp, &idev->addr_list, if_list) {
749					if (!!(ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL) !=
750					    !!(ipv6_addr_type(&np->remote_ip.in6) & IPV6_ADDR_LINKLOCAL))
751						continue;
752					np->local_ip.in6 = ifp->addr;
753					err = 0;
754					break;
755				}
756				read_unlock_bh(&idev->lock);
757			}
758			if (err) {
759				np_err(np, "no IPv6 address for %s, aborting\n",
760				       np->dev_name);
761				goto put;
762			} else
763				np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
764#else
765			np_err(np, "IPv6 is not supported %s, aborting\n",
766			       np->dev_name);
767			err = -EINVAL;
768			goto put;
769#endif
770		}
771	}
772
773	/* fill up the skb queue */
774	refill_skbs();
775
776	err = __netpoll_setup(np, ndev);
777	if (err)
778		goto put;
779
780	rtnl_unlock();
781	return 0;
782
783put:
784	dev_put(ndev);
785unlock:
786	rtnl_unlock();
787	return err;
788}
789EXPORT_SYMBOL(netpoll_setup);
790
791static int __init netpoll_init(void)
792{
793	skb_queue_head_init(&skb_pool);
794	return 0;
795}
796core_initcall(netpoll_init);
797
798static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
799{
800	struct netpoll_info *npinfo =
801			container_of(rcu_head, struct netpoll_info, rcu);
802
803	skb_queue_purge(&npinfo->txq);
804
805	/* we can't call cancel_delayed_work_sync here, as we are in softirq */
806	cancel_delayed_work(&npinfo->tx_work);
807
808	/* clean after last, unfinished work */
809	__skb_queue_purge(&npinfo->txq);
810	/* now cancel it again */
811	cancel_delayed_work(&npinfo->tx_work);
812	kfree(npinfo);
813}
814
815void __netpoll_cleanup(struct netpoll *np)
816{
817	struct netpoll_info *npinfo;
818
819	npinfo = rtnl_dereference(np->dev->npinfo);
820	if (!npinfo)
821		return;
822
823	synchronize_srcu(&netpoll_srcu);
824
825	if (refcount_dec_and_test(&npinfo->refcnt)) {
826		const struct net_device_ops *ops;
827
828		ops = np->dev->netdev_ops;
829		if (ops->ndo_netpoll_cleanup)
830			ops->ndo_netpoll_cleanup(np->dev);
831
832		RCU_INIT_POINTER(np->dev->npinfo, NULL);
833		call_rcu(&npinfo->rcu, rcu_cleanup_netpoll_info);
834	} else
835		RCU_INIT_POINTER(np->dev->npinfo, NULL);
836}
837EXPORT_SYMBOL_GPL(__netpoll_cleanup);
838
839void __netpoll_free(struct netpoll *np)
840{
841	ASSERT_RTNL();
842
843	/* Wait for transmitting packets to finish before freeing. */
844	synchronize_rcu();
845	__netpoll_cleanup(np);
846	kfree(np);
847}
848EXPORT_SYMBOL_GPL(__netpoll_free);
849
850void netpoll_cleanup(struct netpoll *np)
851{
852	rtnl_lock();
853	if (!np->dev)
854		goto out;
855	__netpoll_cleanup(np);
856	dev_put(np->dev);
857	np->dev = NULL;
858out:
859	rtnl_unlock();
860}
861EXPORT_SYMBOL(netpoll_cleanup);
v6.2
  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#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
 49static struct sk_buff_head skb_pool;
 50
 51DEFINE_STATIC_SRCU(netpoll_srcu);
 52
 53#define USEC_PER_POLL	50
 54
 55#define MAX_SKB_SIZE							\
 56	(sizeof(struct ethhdr) +					\
 57	 sizeof(struct iphdr) +						\
 58	 sizeof(struct udphdr) +					\
 59	 MAX_UDP_CHUNK)
 60
 61static void zap_completion_queue(void);
 62
 63static unsigned int carrier_timeout = 4;
 64module_param(carrier_timeout, uint, 0644);
 65
 66#define np_info(np, fmt, ...)				\
 67	pr_info("%s: " fmt, np->name, ##__VA_ARGS__)
 68#define np_err(np, fmt, ...)				\
 69	pr_err("%s: " fmt, np->name, ##__VA_ARGS__)
 70#define np_notice(np, fmt, ...)				\
 71	pr_notice("%s: " fmt, np->name, ##__VA_ARGS__)
 72
 73static netdev_tx_t netpoll_start_xmit(struct sk_buff *skb,
 74				      struct net_device *dev,
 75				      struct netdev_queue *txq)
 76{
 77	netdev_tx_t status = NETDEV_TX_OK;
 78	netdev_features_t features;
 79
 80	features = netif_skb_features(skb);
 81
 82	if (skb_vlan_tag_present(skb) &&
 83	    !vlan_hw_offload_capable(features, skb->vlan_proto)) {
 84		skb = __vlan_hwaccel_push_inside(skb);
 85		if (unlikely(!skb)) {
 86			/* This is actually a packet drop, but we
 87			 * don't want the code that calls this
 88			 * function to try and operate on a NULL skb.
 89			 */
 90			goto out;
 91		}
 92	}
 93
 94	status = netdev_start_xmit(skb, dev, txq, false);
 95
 96out:
 97	return status;
 98}
 99
100static void queue_process(struct work_struct *work)
101{
102	struct netpoll_info *npinfo =
103		container_of(work, struct netpoll_info, tx_work.work);
104	struct sk_buff *skb;
105	unsigned long flags;
106
107	while ((skb = skb_dequeue(&npinfo->txq))) {
108		struct net_device *dev = skb->dev;
109		struct netdev_queue *txq;
110		unsigned int q_index;
111
112		if (!netif_device_present(dev) || !netif_running(dev)) {
113			kfree_skb(skb);
114			continue;
115		}
116
117		local_irq_save(flags);
118		/* check if skb->queue_mapping is still valid */
119		q_index = skb_get_queue_mapping(skb);
120		if (unlikely(q_index >= dev->real_num_tx_queues)) {
121			q_index = q_index % dev->real_num_tx_queues;
122			skb_set_queue_mapping(skb, q_index);
123		}
124		txq = netdev_get_tx_queue(dev, q_index);
125		HARD_TX_LOCK(dev, txq, smp_processor_id());
126		if (netif_xmit_frozen_or_stopped(txq) ||
127		    !dev_xmit_complete(netpoll_start_xmit(skb, dev, txq))) {
128			skb_queue_head(&npinfo->txq, skb);
129			HARD_TX_UNLOCK(dev, txq);
130			local_irq_restore(flags);
131
132			schedule_delayed_work(&npinfo->tx_work, HZ/10);
133			return;
134		}
135		HARD_TX_UNLOCK(dev, txq);
136		local_irq_restore(flags);
137	}
138}
139
140static void poll_one_napi(struct napi_struct *napi)
141{
142	int work;
143
144	/* If we set this bit but see that it has already been set,
145	 * that indicates that napi has been disabled and we need
146	 * to abort this operation
147	 */
148	if (test_and_set_bit(NAPI_STATE_NPSVC, &napi->state))
149		return;
150
151	/* We explicilty pass the polling call a budget of 0 to
152	 * indicate that we are clearing the Tx path only.
153	 */
154	work = napi->poll(napi, 0);
155	WARN_ONCE(work, "%pS exceeded budget in poll\n", napi->poll);
156	trace_napi_poll(napi, work, 0);
157
158	clear_bit(NAPI_STATE_NPSVC, &napi->state);
159}
160
161static void poll_napi(struct net_device *dev)
162{
163	struct napi_struct *napi;
164	int cpu = smp_processor_id();
165
166	list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) {
167		if (cmpxchg(&napi->poll_owner, -1, cpu) == -1) {
168			poll_one_napi(napi);
169			smp_store_release(&napi->poll_owner, -1);
170		}
171	}
172}
173
174void netpoll_poll_dev(struct net_device *dev)
175{
176	struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
177	const struct net_device_ops *ops;
178
179	/* Don't do any rx activity if the dev_lock mutex is held
180	 * the dev_open/close paths use this to block netpoll activity
181	 * while changing device state
182	 */
183	if (!ni || down_trylock(&ni->dev_lock))
184		return;
185
186	if (!netif_running(dev)) {
187		up(&ni->dev_lock);
188		return;
189	}
190
191	ops = dev->netdev_ops;
192	if (ops->ndo_poll_controller)
193		ops->ndo_poll_controller(dev);
194
195	poll_napi(dev);
196
197	up(&ni->dev_lock);
198
199	zap_completion_queue();
200}
201EXPORT_SYMBOL(netpoll_poll_dev);
202
203void netpoll_poll_disable(struct net_device *dev)
204{
205	struct netpoll_info *ni;
206	int idx;
207	might_sleep();
208	idx = srcu_read_lock(&netpoll_srcu);
209	ni = srcu_dereference(dev->npinfo, &netpoll_srcu);
210	if (ni)
211		down(&ni->dev_lock);
212	srcu_read_unlock(&netpoll_srcu, idx);
213}
214EXPORT_SYMBOL(netpoll_poll_disable);
215
216void netpoll_poll_enable(struct net_device *dev)
217{
218	struct netpoll_info *ni;
219	rcu_read_lock();
220	ni = rcu_dereference(dev->npinfo);
221	if (ni)
222		up(&ni->dev_lock);
223	rcu_read_unlock();
224}
225EXPORT_SYMBOL(netpoll_poll_enable);
226
227static void refill_skbs(void)
228{
229	struct sk_buff *skb;
230	unsigned long flags;
231
232	spin_lock_irqsave(&skb_pool.lock, flags);
233	while (skb_pool.qlen < MAX_SKBS) {
234		skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
235		if (!skb)
236			break;
237
238		__skb_queue_tail(&skb_pool, skb);
239	}
240	spin_unlock_irqrestore(&skb_pool.lock, flags);
241}
242
243static void zap_completion_queue(void)
244{
245	unsigned long flags;
246	struct softnet_data *sd = &get_cpu_var(softnet_data);
247
248	if (sd->completion_queue) {
249		struct sk_buff *clist;
250
251		local_irq_save(flags);
252		clist = sd->completion_queue;
253		sd->completion_queue = NULL;
254		local_irq_restore(flags);
255
256		while (clist != NULL) {
257			struct sk_buff *skb = clist;
258			clist = clist->next;
259			if (!skb_irq_freeable(skb)) {
260				refcount_set(&skb->users, 1);
261				dev_kfree_skb_any(skb); /* put this one back */
262			} else {
263				__kfree_skb(skb);
264			}
265		}
266	}
267
268	put_cpu_var(softnet_data);
269}
270
271static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
272{
273	int count = 0;
274	struct sk_buff *skb;
275
276	zap_completion_queue();
277	refill_skbs();
278repeat:
279
280	skb = alloc_skb(len, GFP_ATOMIC);
281	if (!skb)
282		skb = skb_dequeue(&skb_pool);
283
284	if (!skb) {
285		if (++count < 10) {
286			netpoll_poll_dev(np->dev);
287			goto repeat;
288		}
289		return NULL;
290	}
291
292	refcount_set(&skb->users, 1);
293	skb_reserve(skb, reserve);
294	return skb;
295}
296
297static int netpoll_owner_active(struct net_device *dev)
298{
299	struct napi_struct *napi;
300
301	list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) {
302		if (napi->poll_owner == smp_processor_id())
303			return 1;
304	}
305	return 0;
306}
307
308/* call with IRQ disabled */
309static netdev_tx_t __netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
310{
311	netdev_tx_t status = NETDEV_TX_BUSY;
312	struct net_device *dev;
313	unsigned long tries;
314	/* It is up to the caller to keep npinfo alive. */
315	struct netpoll_info *npinfo;
316
317	lockdep_assert_irqs_disabled();
318
319	dev = np->dev;
320	npinfo = rcu_dereference_bh(dev->npinfo);
321
322	if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
323		dev_kfree_skb_irq(skb);
324		return NET_XMIT_DROP;
325	}
326
327	/* don't get messages out of order, and no recursion */
328	if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
329		struct netdev_queue *txq;
330
331		txq = netdev_core_pick_tx(dev, skb, NULL);
332
333		/* try until next clock tick */
334		for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
335		     tries > 0; --tries) {
336			if (HARD_TX_TRYLOCK(dev, txq)) {
337				if (!netif_xmit_stopped(txq))
338					status = netpoll_start_xmit(skb, dev, txq);
339
340				HARD_TX_UNLOCK(dev, txq);
341
342				if (dev_xmit_complete(status))
343					break;
344
345			}
346
347			/* tickle device maybe there is some cleanup */
348			netpoll_poll_dev(np->dev);
349
350			udelay(USEC_PER_POLL);
351		}
352
353		WARN_ONCE(!irqs_disabled(),
354			"netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pS)\n",
355			dev->name, dev->netdev_ops->ndo_start_xmit);
356
357	}
358
359	if (!dev_xmit_complete(status)) {
360		skb_queue_tail(&npinfo->txq, skb);
361		schedule_delayed_work(&npinfo->tx_work,0);
362	}
363	return NETDEV_TX_OK;
364}
365
366netdev_tx_t netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
367{
368	unsigned long flags;
369	netdev_tx_t ret;
370
371	if (unlikely(!np)) {
372		dev_kfree_skb_irq(skb);
373		ret = NET_XMIT_DROP;
374	} else {
375		local_irq_save(flags);
376		ret = __netpoll_send_skb(np, skb);
377		local_irq_restore(flags);
378	}
379	return ret;
380}
381EXPORT_SYMBOL(netpoll_send_skb);
382
383void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
384{
385	int total_len, ip_len, udp_len;
386	struct sk_buff *skb;
387	struct udphdr *udph;
388	struct iphdr *iph;
389	struct ethhdr *eth;
390	static atomic_t ip_ident;
391	struct ipv6hdr *ip6h;
392
393	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
394		WARN_ON_ONCE(!irqs_disabled());
395
396	udp_len = len + sizeof(*udph);
397	if (np->ipv6)
398		ip_len = udp_len + sizeof(*ip6h);
399	else
400		ip_len = udp_len + sizeof(*iph);
401
402	total_len = ip_len + LL_RESERVED_SPACE(np->dev);
403
404	skb = find_skb(np, total_len + np->dev->needed_tailroom,
405		       total_len - len);
406	if (!skb)
407		return;
408
409	skb_copy_to_linear_data(skb, msg, len);
410	skb_put(skb, len);
411
412	skb_push(skb, sizeof(*udph));
413	skb_reset_transport_header(skb);
414	udph = udp_hdr(skb);
415	udph->source = htons(np->local_port);
416	udph->dest = htons(np->remote_port);
417	udph->len = htons(udp_len);
418
419	if (np->ipv6) {
420		udph->check = 0;
421		udph->check = csum_ipv6_magic(&np->local_ip.in6,
422					      &np->remote_ip.in6,
423					      udp_len, IPPROTO_UDP,
424					      csum_partial(udph, udp_len, 0));
425		if (udph->check == 0)
426			udph->check = CSUM_MANGLED_0;
427
428		skb_push(skb, sizeof(*ip6h));
429		skb_reset_network_header(skb);
430		ip6h = ipv6_hdr(skb);
431
432		/* ip6h->version = 6; ip6h->priority = 0; */
433		*(unsigned char *)ip6h = 0x60;
434		ip6h->flow_lbl[0] = 0;
435		ip6h->flow_lbl[1] = 0;
436		ip6h->flow_lbl[2] = 0;
437
438		ip6h->payload_len = htons(sizeof(struct udphdr) + len);
439		ip6h->nexthdr = IPPROTO_UDP;
440		ip6h->hop_limit = 32;
441		ip6h->saddr = np->local_ip.in6;
442		ip6h->daddr = np->remote_ip.in6;
443
444		eth = skb_push(skb, ETH_HLEN);
445		skb_reset_mac_header(skb);
446		skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
447	} else {
448		udph->check = 0;
449		udph->check = csum_tcpudp_magic(np->local_ip.ip,
450						np->remote_ip.ip,
451						udp_len, IPPROTO_UDP,
452						csum_partial(udph, udp_len, 0));
453		if (udph->check == 0)
454			udph->check = CSUM_MANGLED_0;
455
456		skb_push(skb, sizeof(*iph));
457		skb_reset_network_header(skb);
458		iph = ip_hdr(skb);
459
460		/* iph->version = 4; iph->ihl = 5; */
461		*(unsigned char *)iph = 0x45;
462		iph->tos      = 0;
463		put_unaligned(htons(ip_len), &(iph->tot_len));
464		iph->id       = htons(atomic_inc_return(&ip_ident));
465		iph->frag_off = 0;
466		iph->ttl      = 64;
467		iph->protocol = IPPROTO_UDP;
468		iph->check    = 0;
469		put_unaligned(np->local_ip.ip, &(iph->saddr));
470		put_unaligned(np->remote_ip.ip, &(iph->daddr));
471		iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
472
473		eth = skb_push(skb, ETH_HLEN);
474		skb_reset_mac_header(skb);
475		skb->protocol = eth->h_proto = htons(ETH_P_IP);
476	}
477
478	ether_addr_copy(eth->h_source, np->dev->dev_addr);
479	ether_addr_copy(eth->h_dest, np->remote_mac);
480
481	skb->dev = np->dev;
482
483	netpoll_send_skb(np, skb);
484}
485EXPORT_SYMBOL(netpoll_send_udp);
486
487void netpoll_print_options(struct netpoll *np)
488{
489	np_info(np, "local port %d\n", np->local_port);
490	if (np->ipv6)
491		np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
492	else
493		np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
494	np_info(np, "interface '%s'\n", np->dev_name);
495	np_info(np, "remote port %d\n", np->remote_port);
496	if (np->ipv6)
497		np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
498	else
499		np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
500	np_info(np, "remote ethernet address %pM\n", np->remote_mac);
501}
502EXPORT_SYMBOL(netpoll_print_options);
503
504static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
505{
506	const char *end;
507
508	if (!strchr(str, ':') &&
509	    in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
510		if (!*end)
511			return 0;
512	}
513	if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
514#if IS_ENABLED(CONFIG_IPV6)
515		if (!*end)
516			return 1;
517#else
518		return -1;
519#endif
520	}
521	return -1;
522}
523
524int netpoll_parse_options(struct netpoll *np, char *opt)
525{
526	char *cur=opt, *delim;
527	int ipv6;
528	bool ipversion_set = false;
529
530	if (*cur != '@') {
531		if ((delim = strchr(cur, '@')) == NULL)
532			goto parse_failed;
533		*delim = 0;
534		if (kstrtou16(cur, 10, &np->local_port))
535			goto parse_failed;
536		cur = delim;
537	}
538	cur++;
539
540	if (*cur != '/') {
541		ipversion_set = true;
542		if ((delim = strchr(cur, '/')) == NULL)
543			goto parse_failed;
544		*delim = 0;
545		ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
546		if (ipv6 < 0)
547			goto parse_failed;
548		else
549			np->ipv6 = (bool)ipv6;
550		cur = delim;
551	}
552	cur++;
553
554	if (*cur != ',') {
555		/* parse out dev name */
556		if ((delim = strchr(cur, ',')) == NULL)
557			goto parse_failed;
558		*delim = 0;
559		strscpy(np->dev_name, cur, sizeof(np->dev_name));
560		cur = delim;
561	}
562	cur++;
563
564	if (*cur != '@') {
565		/* dst port */
566		if ((delim = strchr(cur, '@')) == NULL)
567			goto parse_failed;
568		*delim = 0;
569		if (*cur == ' ' || *cur == '\t')
570			np_info(np, "warning: whitespace is not allowed\n");
571		if (kstrtou16(cur, 10, &np->remote_port))
572			goto parse_failed;
573		cur = delim;
574	}
575	cur++;
576
577	/* dst ip */
578	if ((delim = strchr(cur, '/')) == NULL)
579		goto parse_failed;
580	*delim = 0;
581	ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
582	if (ipv6 < 0)
583		goto parse_failed;
584	else if (ipversion_set && np->ipv6 != (bool)ipv6)
585		goto parse_failed;
586	else
587		np->ipv6 = (bool)ipv6;
588	cur = delim + 1;
589
590	if (*cur != 0) {
591		/* MAC address */
592		if (!mac_pton(cur, np->remote_mac))
593			goto parse_failed;
594	}
595
596	netpoll_print_options(np);
597
598	return 0;
599
600 parse_failed:
601	np_info(np, "couldn't parse config at '%s'!\n", cur);
602	return -1;
603}
604EXPORT_SYMBOL(netpoll_parse_options);
605
606int __netpoll_setup(struct netpoll *np, struct net_device *ndev)
607{
608	struct netpoll_info *npinfo;
609	const struct net_device_ops *ops;
610	int err;
611
612	np->dev = ndev;
613	strscpy(np->dev_name, ndev->name, IFNAMSIZ);
614
615	if (ndev->priv_flags & IFF_DISABLE_NETPOLL) {
616		np_err(np, "%s doesn't support polling, aborting\n",
617		       np->dev_name);
618		err = -ENOTSUPP;
619		goto out;
620	}
621
622	if (!ndev->npinfo) {
623		npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
624		if (!npinfo) {
625			err = -ENOMEM;
626			goto out;
627		}
628
629		sema_init(&npinfo->dev_lock, 1);
630		skb_queue_head_init(&npinfo->txq);
631		INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
632
633		refcount_set(&npinfo->refcnt, 1);
634
635		ops = np->dev->netdev_ops;
636		if (ops->ndo_netpoll_setup) {
637			err = ops->ndo_netpoll_setup(ndev, npinfo);
638			if (err)
639				goto free_npinfo;
640		}
641	} else {
642		npinfo = rtnl_dereference(ndev->npinfo);
643		refcount_inc(&npinfo->refcnt);
644	}
645
646	npinfo->netpoll = np;
647
648	/* last thing to do is link it to the net device structure */
649	rcu_assign_pointer(ndev->npinfo, npinfo);
650
651	return 0;
652
653free_npinfo:
654	kfree(npinfo);
655out:
656	return err;
657}
658EXPORT_SYMBOL_GPL(__netpoll_setup);
659
660int netpoll_setup(struct netpoll *np)
661{
662	struct net_device *ndev = NULL;
663	struct in_device *in_dev;
664	int err;
665
666	rtnl_lock();
667	if (np->dev_name[0]) {
668		struct net *net = current->nsproxy->net_ns;
669		ndev = __dev_get_by_name(net, np->dev_name);
670	}
671	if (!ndev) {
672		np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
673		err = -ENODEV;
674		goto unlock;
675	}
676	dev_hold(ndev);
677
678	if (netdev_master_upper_dev_get(ndev)) {
679		np_err(np, "%s is a slave device, aborting\n", np->dev_name);
680		err = -EBUSY;
681		goto put;
682	}
683
684	if (!netif_running(ndev)) {
685		unsigned long atmost, atleast;
686
687		np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
688
689		err = dev_open(ndev, NULL);
690
691		if (err) {
692			np_err(np, "failed to open %s\n", ndev->name);
693			goto put;
694		}
695
696		rtnl_unlock();
697		atleast = jiffies + HZ/10;
698		atmost = jiffies + carrier_timeout * HZ;
699		while (!netif_carrier_ok(ndev)) {
700			if (time_after(jiffies, atmost)) {
701				np_notice(np, "timeout waiting for carrier\n");
702				break;
703			}
704			msleep(1);
705		}
706
707		/* If carrier appears to come up instantly, we don't
708		 * trust it and pause so that we don't pump all our
709		 * queued console messages into the bitbucket.
710		 */
711
712		if (time_before(jiffies, atleast)) {
713			np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
714			msleep(4000);
715		}
716		rtnl_lock();
717	}
718
719	if (!np->local_ip.ip) {
720		if (!np->ipv6) {
721			const struct in_ifaddr *ifa;
722
723			in_dev = __in_dev_get_rtnl(ndev);
724			if (!in_dev)
725				goto put_noaddr;
726
727			ifa = rtnl_dereference(in_dev->ifa_list);
728			if (!ifa) {
729put_noaddr:
730				np_err(np, "no IP address for %s, aborting\n",
731				       np->dev_name);
732				err = -EDESTADDRREQ;
733				goto put;
734			}
735
736			np->local_ip.ip = ifa->ifa_local;
737			np_info(np, "local IP %pI4\n", &np->local_ip.ip);
738		} else {
739#if IS_ENABLED(CONFIG_IPV6)
740			struct inet6_dev *idev;
741
742			err = -EDESTADDRREQ;
743			idev = __in6_dev_get(ndev);
744			if (idev) {
745				struct inet6_ifaddr *ifp;
746
747				read_lock_bh(&idev->lock);
748				list_for_each_entry(ifp, &idev->addr_list, if_list) {
749					if (!!(ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL) !=
750					    !!(ipv6_addr_type(&np->remote_ip.in6) & IPV6_ADDR_LINKLOCAL))
751						continue;
752					np->local_ip.in6 = ifp->addr;
753					err = 0;
754					break;
755				}
756				read_unlock_bh(&idev->lock);
757			}
758			if (err) {
759				np_err(np, "no IPv6 address for %s, aborting\n",
760				       np->dev_name);
761				goto put;
762			} else
763				np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
764#else
765			np_err(np, "IPv6 is not supported %s, aborting\n",
766			       np->dev_name);
767			err = -EINVAL;
768			goto put;
769#endif
770		}
771	}
772
773	/* fill up the skb queue */
774	refill_skbs();
775
776	err = __netpoll_setup(np, ndev);
777	if (err)
778		goto put;
779	netdev_tracker_alloc(ndev, &np->dev_tracker, GFP_KERNEL);
780	rtnl_unlock();
781	return 0;
782
783put:
784	dev_put(ndev);
785unlock:
786	rtnl_unlock();
787	return err;
788}
789EXPORT_SYMBOL(netpoll_setup);
790
791static int __init netpoll_init(void)
792{
793	skb_queue_head_init(&skb_pool);
794	return 0;
795}
796core_initcall(netpoll_init);
797
798static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
799{
800	struct netpoll_info *npinfo =
801			container_of(rcu_head, struct netpoll_info, rcu);
802
803	skb_queue_purge(&npinfo->txq);
804
805	/* we can't call cancel_delayed_work_sync here, as we are in softirq */
806	cancel_delayed_work(&npinfo->tx_work);
807
808	/* clean after last, unfinished work */
809	__skb_queue_purge(&npinfo->txq);
810	/* now cancel it again */
811	cancel_delayed_work(&npinfo->tx_work);
812	kfree(npinfo);
813}
814
815void __netpoll_cleanup(struct netpoll *np)
816{
817	struct netpoll_info *npinfo;
818
819	npinfo = rtnl_dereference(np->dev->npinfo);
820	if (!npinfo)
821		return;
822
823	synchronize_srcu(&netpoll_srcu);
824
825	if (refcount_dec_and_test(&npinfo->refcnt)) {
826		const struct net_device_ops *ops;
827
828		ops = np->dev->netdev_ops;
829		if (ops->ndo_netpoll_cleanup)
830			ops->ndo_netpoll_cleanup(np->dev);
831
832		RCU_INIT_POINTER(np->dev->npinfo, NULL);
833		call_rcu(&npinfo->rcu, rcu_cleanup_netpoll_info);
834	} else
835		RCU_INIT_POINTER(np->dev->npinfo, NULL);
836}
837EXPORT_SYMBOL_GPL(__netpoll_cleanup);
838
839void __netpoll_free(struct netpoll *np)
840{
841	ASSERT_RTNL();
842
843	/* Wait for transmitting packets to finish before freeing. */
844	synchronize_rcu();
845	__netpoll_cleanup(np);
846	kfree(np);
847}
848EXPORT_SYMBOL_GPL(__netpoll_free);
849
850void netpoll_cleanup(struct netpoll *np)
851{
852	rtnl_lock();
853	if (!np->dev)
854		goto out;
855	__netpoll_cleanup(np);
856	netdev_put(np->dev, &np->dev_tracker);
857	np->dev = NULL;
858out:
859	rtnl_unlock();
860}
861EXPORT_SYMBOL(netpoll_cleanup);