Linux Audio

Check our new training course

Loading...
v3.15
  1/*
  2 * Common framework for low-level network console, dump, and debugger code
  3 *
  4 * Sep 8 2003  Matt Mackall <mpm@selenic.com>
  5 *
  6 * based on the netconsole code from:
  7 *
  8 * Copyright (C) 2001  Ingo Molnar <mingo@redhat.com>
  9 * Copyright (C) 2002  Red Hat, Inc.
 10 */
 11
 12#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 13
 14#include <linux/moduleparam.h>
 15#include <linux/kernel.h>
 16#include <linux/netdevice.h>
 17#include <linux/etherdevice.h>
 18#include <linux/string.h>
 19#include <linux/if_arp.h>
 20#include <linux/inetdevice.h>
 21#include <linux/inet.h>
 22#include <linux/interrupt.h>
 23#include <linux/netpoll.h>
 24#include <linux/sched.h>
 25#include <linux/delay.h>
 26#include <linux/rcupdate.h>
 27#include <linux/workqueue.h>
 28#include <linux/slab.h>
 29#include <linux/export.h>
 30#include <linux/if_vlan.h>
 31#include <net/tcp.h>
 32#include <net/udp.h>
 33#include <net/addrconf.h>
 34#include <net/ndisc.h>
 35#include <net/ip6_checksum.h>
 36#include <asm/unaligned.h>
 37#include <trace/events/napi.h>
 38
 39/*
 40 * We maintain a small pool of fully-sized skbs, to make sure the
 41 * message gets out even in extreme OOM situations.
 42 */
 43
 44#define MAX_UDP_CHUNK 1460
 45#define MAX_SKBS 32
 46
 47static struct sk_buff_head skb_pool;
 48
 49DEFINE_STATIC_SRCU(netpoll_srcu);
 50
 51#define USEC_PER_POLL	50
 52
 53#define MAX_SKB_SIZE							\
 54	(sizeof(struct ethhdr) +					\
 55	 sizeof(struct iphdr) +						\
 56	 sizeof(struct udphdr) +					\
 57	 MAX_UDP_CHUNK)
 58
 59static void zap_completion_queue(void);
 60static void netpoll_async_cleanup(struct work_struct *work);
 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	const struct net_device_ops *ops = dev->netdev_ops;
 76	int status = NETDEV_TX_OK;
 77	netdev_features_t features;
 78
 79	features = netif_skb_features(skb);
 80
 81	if (vlan_tx_tag_present(skb) &&
 82	    !vlan_hw_offload_capable(features, skb->vlan_proto)) {
 83		skb = __vlan_put_tag(skb, skb->vlan_proto,
 84				     vlan_tx_tag_get(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		skb->vlan_tci = 0;
 93	}
 94
 95	status = ops->ndo_start_xmit(skb, dev);
 96	if (status == NETDEV_TX_OK)
 97		txq_trans_update(txq);
 98
 99out:
100	return status;
101}
102
103static void queue_process(struct work_struct *work)
104{
105	struct netpoll_info *npinfo =
106		container_of(work, struct netpoll_info, tx_work.work);
107	struct sk_buff *skb;
108	unsigned long flags;
109
110	while ((skb = skb_dequeue(&npinfo->txq))) {
111		struct net_device *dev = skb->dev;
112		struct netdev_queue *txq;
 
113
114		if (!netif_device_present(dev) || !netif_running(dev)) {
115			kfree_skb(skb);
116			continue;
117		}
118
119		txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
120
121		local_irq_save(flags);
 
 
 
 
 
 
 
122		HARD_TX_LOCK(dev, txq, smp_processor_id());
123		if (netif_xmit_frozen_or_stopped(txq) ||
124		    netpoll_start_xmit(skb, dev, txq) != NETDEV_TX_OK) {
125			skb_queue_head(&npinfo->txq, skb);
126			HARD_TX_UNLOCK(dev, txq);
127			local_irq_restore(flags);
128
129			schedule_delayed_work(&npinfo->tx_work, HZ/10);
130			return;
131		}
132		HARD_TX_UNLOCK(dev, txq);
133		local_irq_restore(flags);
134	}
135}
136
137/*
138 * Check whether delayed processing was scheduled for our NIC. If so,
139 * we attempt to grab the poll lock and use ->poll() to pump the card.
140 * If this fails, either we've recursed in ->poll() or it's already
141 * running on another CPU.
142 *
143 * Note: we don't mask interrupts with this lock because we're using
144 * trylock here and interrupts are already disabled in the softirq
145 * case. Further, we test the poll_owner to avoid recursion on UP
146 * systems where the lock doesn't exist.
147 */
148static int poll_one_napi(struct napi_struct *napi, int budget)
149{
150	int work;
151
152	/* net_rx_action's ->poll() invocations and our's are
153	 * synchronized by this test which is only made while
154	 * holding the napi->poll_lock.
155	 */
156	if (!test_bit(NAPI_STATE_SCHED, &napi->state))
157		return budget;
158
159	set_bit(NAPI_STATE_NPSVC, &napi->state);
 
 
 
 
 
160
161	work = napi->poll(napi, budget);
162	WARN_ONCE(work > budget, "%pF exceeded budget in poll\n", napi->poll);
163	trace_napi_poll(napi);
 
 
 
164
165	clear_bit(NAPI_STATE_NPSVC, &napi->state);
166
167	return budget - work;
168}
169
170static void poll_napi(struct net_device *dev, int budget)
171{
172	struct napi_struct *napi;
 
173
174	list_for_each_entry(napi, &dev->napi_list, dev_list) {
175		if (napi->poll_owner != smp_processor_id() &&
176		    spin_trylock(&napi->poll_lock)) {
177			budget = poll_one_napi(napi, budget);
178			spin_unlock(&napi->poll_lock);
179		}
180	}
181}
182
183static void netpoll_poll_dev(struct net_device *dev)
184{
185	const struct net_device_ops *ops;
186	struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
187	int budget = 0;
188
189	/* Don't do any rx activity if the dev_lock mutex is held
190	 * the dev_open/close paths use this to block netpoll activity
191	 * while changing device state
192	 */
193	if (down_trylock(&ni->dev_lock))
194		return;
195
196	if (!netif_running(dev)) {
197		up(&ni->dev_lock);
198		return;
199	}
200
201	ops = dev->netdev_ops;
202	if (!ops->ndo_poll_controller) {
203		up(&ni->dev_lock);
204		return;
205	}
206
207	/* Process pending work on NIC */
208	ops->ndo_poll_controller(dev);
209
210	poll_napi(dev, budget);
211
212	up(&ni->dev_lock);
213
214	zap_completion_queue();
215}
216
217void netpoll_poll_disable(struct net_device *dev)
218{
219	struct netpoll_info *ni;
220	int idx;
221	might_sleep();
222	idx = srcu_read_lock(&netpoll_srcu);
223	ni = srcu_dereference(dev->npinfo, &netpoll_srcu);
224	if (ni)
225		down(&ni->dev_lock);
226	srcu_read_unlock(&netpoll_srcu, idx);
227}
228EXPORT_SYMBOL(netpoll_poll_disable);
229
230void netpoll_poll_enable(struct net_device *dev)
231{
232	struct netpoll_info *ni;
233	rcu_read_lock();
234	ni = rcu_dereference(dev->npinfo);
235	if (ni)
236		up(&ni->dev_lock);
237	rcu_read_unlock();
238}
239EXPORT_SYMBOL(netpoll_poll_enable);
240
241static void refill_skbs(void)
242{
243	struct sk_buff *skb;
244	unsigned long flags;
245
246	spin_lock_irqsave(&skb_pool.lock, flags);
247	while (skb_pool.qlen < MAX_SKBS) {
248		skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
249		if (!skb)
250			break;
251
252		__skb_queue_tail(&skb_pool, skb);
253	}
254	spin_unlock_irqrestore(&skb_pool.lock, flags);
255}
256
257static void zap_completion_queue(void)
258{
259	unsigned long flags;
260	struct softnet_data *sd = &get_cpu_var(softnet_data);
261
262	if (sd->completion_queue) {
263		struct sk_buff *clist;
264
265		local_irq_save(flags);
266		clist = sd->completion_queue;
267		sd->completion_queue = NULL;
268		local_irq_restore(flags);
269
270		while (clist != NULL) {
271			struct sk_buff *skb = clist;
272			clist = clist->next;
273			if (!skb_irq_freeable(skb)) {
274				atomic_inc(&skb->users);
275				dev_kfree_skb_any(skb); /* put this one back */
276			} else {
277				__kfree_skb(skb);
278			}
279		}
280	}
281
282	put_cpu_var(softnet_data);
283}
284
285static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
286{
287	int count = 0;
288	struct sk_buff *skb;
289
290	zap_completion_queue();
291	refill_skbs();
292repeat:
293
294	skb = alloc_skb(len, GFP_ATOMIC);
295	if (!skb)
296		skb = skb_dequeue(&skb_pool);
297
298	if (!skb) {
299		if (++count < 10) {
300			netpoll_poll_dev(np->dev);
301			goto repeat;
302		}
303		return NULL;
304	}
305
306	atomic_set(&skb->users, 1);
307	skb_reserve(skb, reserve);
308	return skb;
309}
310
311static int netpoll_owner_active(struct net_device *dev)
312{
313	struct napi_struct *napi;
314
315	list_for_each_entry(napi, &dev->napi_list, dev_list) {
316		if (napi->poll_owner == smp_processor_id())
317			return 1;
318	}
319	return 0;
320}
321
322/* call with IRQ disabled */
323void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb,
324			     struct net_device *dev)
325{
326	int status = NETDEV_TX_BUSY;
327	unsigned long tries;
328	/* It is up to the caller to keep npinfo alive. */
329	struct netpoll_info *npinfo;
330
331	WARN_ON_ONCE(!irqs_disabled());
332
333	npinfo = rcu_dereference_bh(np->dev->npinfo);
334	if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
335		dev_kfree_skb_irq(skb);
336		return;
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_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 (status == NETDEV_TX_OK)
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 (%pF)\n",
367			dev->name, dev->netdev_ops->ndo_start_xmit);
368
369	}
370
371	if (status != NETDEV_TX_OK) {
372		skb_queue_tail(&npinfo->txq, skb);
373		schedule_delayed_work(&npinfo->tx_work,0);
374	}
375}
376EXPORT_SYMBOL(netpoll_send_skb_on_dev);
377
378void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
379{
380	int total_len, ip_len, udp_len;
381	struct sk_buff *skb;
382	struct udphdr *udph;
383	struct iphdr *iph;
384	struct ethhdr *eth;
385	static atomic_t ip_ident;
386	struct ipv6hdr *ip6h;
387
 
 
388	udp_len = len + sizeof(*udph);
389	if (np->ipv6)
390		ip_len = udp_len + sizeof(*ip6h);
391	else
392		ip_len = udp_len + sizeof(*iph);
393
394	total_len = ip_len + LL_RESERVED_SPACE(np->dev);
395
396	skb = find_skb(np, total_len + np->dev->needed_tailroom,
397		       total_len - len);
398	if (!skb)
399		return;
400
401	skb_copy_to_linear_data(skb, msg, len);
402	skb_put(skb, len);
403
404	skb_push(skb, sizeof(*udph));
405	skb_reset_transport_header(skb);
406	udph = udp_hdr(skb);
407	udph->source = htons(np->local_port);
408	udph->dest = htons(np->remote_port);
409	udph->len = htons(udp_len);
410
411	if (np->ipv6) {
412		udph->check = 0;
413		udph->check = csum_ipv6_magic(&np->local_ip.in6,
414					      &np->remote_ip.in6,
415					      udp_len, IPPROTO_UDP,
416					      csum_partial(udph, udp_len, 0));
417		if (udph->check == 0)
418			udph->check = CSUM_MANGLED_0;
419
420		skb_push(skb, sizeof(*ip6h));
421		skb_reset_network_header(skb);
422		ip6h = ipv6_hdr(skb);
423
424		/* ip6h->version = 6; ip6h->priority = 0; */
425		put_unaligned(0x60, (unsigned char *)ip6h);
426		ip6h->flow_lbl[0] = 0;
427		ip6h->flow_lbl[1] = 0;
428		ip6h->flow_lbl[2] = 0;
429
430		ip6h->payload_len = htons(sizeof(struct udphdr) + len);
431		ip6h->nexthdr = IPPROTO_UDP;
432		ip6h->hop_limit = 32;
433		ip6h->saddr = np->local_ip.in6;
434		ip6h->daddr = np->remote_ip.in6;
435
436		eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
437		skb_reset_mac_header(skb);
438		skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
439	} else {
440		udph->check = 0;
441		udph->check = csum_tcpudp_magic(np->local_ip.ip,
442						np->remote_ip.ip,
443						udp_len, IPPROTO_UDP,
444						csum_partial(udph, udp_len, 0));
445		if (udph->check == 0)
446			udph->check = CSUM_MANGLED_0;
447
448		skb_push(skb, sizeof(*iph));
449		skb_reset_network_header(skb);
450		iph = ip_hdr(skb);
451
452		/* iph->version = 4; iph->ihl = 5; */
453		put_unaligned(0x45, (unsigned char *)iph);
454		iph->tos      = 0;
455		put_unaligned(htons(ip_len), &(iph->tot_len));
456		iph->id       = htons(atomic_inc_return(&ip_ident));
457		iph->frag_off = 0;
458		iph->ttl      = 64;
459		iph->protocol = IPPROTO_UDP;
460		iph->check    = 0;
461		put_unaligned(np->local_ip.ip, &(iph->saddr));
462		put_unaligned(np->remote_ip.ip, &(iph->daddr));
463		iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
464
465		eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
466		skb_reset_mac_header(skb);
467		skb->protocol = eth->h_proto = htons(ETH_P_IP);
468	}
469
470	ether_addr_copy(eth->h_source, np->dev->dev_addr);
471	ether_addr_copy(eth->h_dest, np->remote_mac);
472
473	skb->dev = np->dev;
474
475	netpoll_send_skb(np, skb);
476}
477EXPORT_SYMBOL(netpoll_send_udp);
478
479void netpoll_print_options(struct netpoll *np)
480{
481	np_info(np, "local port %d\n", np->local_port);
482	if (np->ipv6)
483		np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
484	else
485		np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
486	np_info(np, "interface '%s'\n", np->dev_name);
487	np_info(np, "remote port %d\n", np->remote_port);
488	if (np->ipv6)
489		np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
490	else
491		np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
492	np_info(np, "remote ethernet address %pM\n", np->remote_mac);
493}
494EXPORT_SYMBOL(netpoll_print_options);
495
496static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
497{
498	const char *end;
499
500	if (!strchr(str, ':') &&
501	    in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
502		if (!*end)
503			return 0;
504	}
505	if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
506#if IS_ENABLED(CONFIG_IPV6)
507		if (!*end)
508			return 1;
509#else
510		return -1;
511#endif
512	}
513	return -1;
514}
515
516int netpoll_parse_options(struct netpoll *np, char *opt)
517{
518	char *cur=opt, *delim;
519	int ipv6;
520	bool ipversion_set = false;
521
522	if (*cur != '@') {
523		if ((delim = strchr(cur, '@')) == NULL)
524			goto parse_failed;
525		*delim = 0;
526		if (kstrtou16(cur, 10, &np->local_port))
527			goto parse_failed;
528		cur = delim;
529	}
530	cur++;
531
532	if (*cur != '/') {
533		ipversion_set = true;
534		if ((delim = strchr(cur, '/')) == NULL)
535			goto parse_failed;
536		*delim = 0;
537		ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
538		if (ipv6 < 0)
539			goto parse_failed;
540		else
541			np->ipv6 = (bool)ipv6;
542		cur = delim;
543	}
544	cur++;
545
546	if (*cur != ',') {
547		/* parse out dev name */
548		if ((delim = strchr(cur, ',')) == NULL)
549			goto parse_failed;
550		*delim = 0;
551		strlcpy(np->dev_name, cur, sizeof(np->dev_name));
552		cur = delim;
553	}
554	cur++;
555
556	if (*cur != '@') {
557		/* dst port */
558		if ((delim = strchr(cur, '@')) == NULL)
559			goto parse_failed;
560		*delim = 0;
561		if (*cur == ' ' || *cur == '\t')
562			np_info(np, "warning: whitespace is not allowed\n");
563		if (kstrtou16(cur, 10, &np->remote_port))
564			goto parse_failed;
565		cur = delim;
566	}
567	cur++;
568
569	/* dst ip */
570	if ((delim = strchr(cur, '/')) == NULL)
571		goto parse_failed;
572	*delim = 0;
573	ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
574	if (ipv6 < 0)
575		goto parse_failed;
576	else if (ipversion_set && np->ipv6 != (bool)ipv6)
577		goto parse_failed;
578	else
579		np->ipv6 = (bool)ipv6;
580	cur = delim + 1;
581
582	if (*cur != 0) {
583		/* MAC address */
584		if (!mac_pton(cur, np->remote_mac))
585			goto parse_failed;
586	}
587
588	netpoll_print_options(np);
589
590	return 0;
591
592 parse_failed:
593	np_info(np, "couldn't parse config at '%s'!\n", cur);
594	return -1;
595}
596EXPORT_SYMBOL(netpoll_parse_options);
597
598int __netpoll_setup(struct netpoll *np, struct net_device *ndev)
599{
600	struct netpoll_info *npinfo;
601	const struct net_device_ops *ops;
602	int err;
603
604	np->dev = ndev;
605	strlcpy(np->dev_name, ndev->name, IFNAMSIZ);
606	INIT_WORK(&np->cleanup_work, netpoll_async_cleanup);
607
608	if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) ||
609	    !ndev->netdev_ops->ndo_poll_controller) {
610		np_err(np, "%s doesn't support polling, aborting\n",
611		       np->dev_name);
612		err = -ENOTSUPP;
613		goto out;
614	}
615
616	if (!ndev->npinfo) {
617		npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
618		if (!npinfo) {
619			err = -ENOMEM;
620			goto out;
621		}
622
623		sema_init(&npinfo->dev_lock, 1);
624		skb_queue_head_init(&npinfo->txq);
625		INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
626
627		atomic_set(&npinfo->refcnt, 1);
628
629		ops = np->dev->netdev_ops;
630		if (ops->ndo_netpoll_setup) {
631			err = ops->ndo_netpoll_setup(ndev, npinfo);
632			if (err)
633				goto free_npinfo;
634		}
635	} else {
636		npinfo = rtnl_dereference(ndev->npinfo);
637		atomic_inc(&npinfo->refcnt);
638	}
639
640	npinfo->netpoll = np;
641
642	/* last thing to do is link it to the net device structure */
643	rcu_assign_pointer(ndev->npinfo, npinfo);
644
645	return 0;
646
647free_npinfo:
648	kfree(npinfo);
649out:
650	return err;
651}
652EXPORT_SYMBOL_GPL(__netpoll_setup);
653
654int netpoll_setup(struct netpoll *np)
655{
656	struct net_device *ndev = NULL;
657	struct in_device *in_dev;
658	int err;
659
660	rtnl_lock();
661	if (np->dev_name) {
662		struct net *net = current->nsproxy->net_ns;
663		ndev = __dev_get_by_name(net, np->dev_name);
664	}
665	if (!ndev) {
666		np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
667		err = -ENODEV;
668		goto unlock;
669	}
670	dev_hold(ndev);
671
672	if (netdev_master_upper_dev_get(ndev)) {
673		np_err(np, "%s is a slave device, aborting\n", np->dev_name);
674		err = -EBUSY;
675		goto put;
676	}
677
678	if (!netif_running(ndev)) {
679		unsigned long atmost, atleast;
680
681		np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
682
683		err = dev_open(ndev);
684
685		if (err) {
686			np_err(np, "failed to open %s\n", ndev->name);
687			goto put;
688		}
689
690		rtnl_unlock();
691		atleast = jiffies + HZ/10;
692		atmost = jiffies + carrier_timeout * HZ;
693		while (!netif_carrier_ok(ndev)) {
694			if (time_after(jiffies, atmost)) {
695				np_notice(np, "timeout waiting for carrier\n");
696				break;
697			}
698			msleep(1);
699		}
700
701		/* If carrier appears to come up instantly, we don't
702		 * trust it and pause so that we don't pump all our
703		 * queued console messages into the bitbucket.
704		 */
705
706		if (time_before(jiffies, atleast)) {
707			np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
708			msleep(4000);
709		}
710		rtnl_lock();
711	}
712
713	if (!np->local_ip.ip) {
714		if (!np->ipv6) {
715			in_dev = __in_dev_get_rtnl(ndev);
716
717			if (!in_dev || !in_dev->ifa_list) {
718				np_err(np, "no IP address for %s, aborting\n",
719				       np->dev_name);
720				err = -EDESTADDRREQ;
721				goto put;
722			}
723
724			np->local_ip.ip = in_dev->ifa_list->ifa_local;
725			np_info(np, "local IP %pI4\n", &np->local_ip.ip);
726		} else {
727#if IS_ENABLED(CONFIG_IPV6)
728			struct inet6_dev *idev;
729
730			err = -EDESTADDRREQ;
731			idev = __in6_dev_get(ndev);
732			if (idev) {
733				struct inet6_ifaddr *ifp;
734
735				read_lock_bh(&idev->lock);
736				list_for_each_entry(ifp, &idev->addr_list, if_list) {
737					if (ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL)
738						continue;
739					np->local_ip.in6 = ifp->addr;
740					err = 0;
741					break;
742				}
743				read_unlock_bh(&idev->lock);
744			}
745			if (err) {
746				np_err(np, "no IPv6 address for %s, aborting\n",
747				       np->dev_name);
748				goto put;
749			} else
750				np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
751#else
752			np_err(np, "IPv6 is not supported %s, aborting\n",
753			       np->dev_name);
754			err = -EINVAL;
755			goto put;
756#endif
757		}
758	}
759
760	/* fill up the skb queue */
761	refill_skbs();
762
763	err = __netpoll_setup(np, ndev);
764	if (err)
765		goto put;
766
767	rtnl_unlock();
768	return 0;
769
770put:
771	dev_put(ndev);
772unlock:
773	rtnl_unlock();
774	return err;
775}
776EXPORT_SYMBOL(netpoll_setup);
777
778static int __init netpoll_init(void)
779{
780	skb_queue_head_init(&skb_pool);
781	return 0;
782}
783core_initcall(netpoll_init);
784
785static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
786{
787	struct netpoll_info *npinfo =
788			container_of(rcu_head, struct netpoll_info, rcu);
789
790	skb_queue_purge(&npinfo->txq);
791
792	/* we can't call cancel_delayed_work_sync here, as we are in softirq */
793	cancel_delayed_work(&npinfo->tx_work);
794
795	/* clean after last, unfinished work */
796	__skb_queue_purge(&npinfo->txq);
797	/* now cancel it again */
798	cancel_delayed_work(&npinfo->tx_work);
799	kfree(npinfo);
800}
801
802void __netpoll_cleanup(struct netpoll *np)
803{
804	struct netpoll_info *npinfo;
805
806	/* rtnl_dereference would be preferable here but
807	 * rcu_cleanup_netpoll path can put us in here safely without
808	 * holding the rtnl, so plain rcu_dereference it is
809	 */
810	npinfo = rtnl_dereference(np->dev->npinfo);
811	if (!npinfo)
812		return;
813
814	synchronize_srcu(&netpoll_srcu);
815
816	if (atomic_dec_and_test(&npinfo->refcnt)) {
817		const struct net_device_ops *ops;
818
819		ops = np->dev->netdev_ops;
820		if (ops->ndo_netpoll_cleanup)
821			ops->ndo_netpoll_cleanup(np->dev);
822
823		RCU_INIT_POINTER(np->dev->npinfo, NULL);
824		call_rcu_bh(&npinfo->rcu, rcu_cleanup_netpoll_info);
825	}
 
826}
827EXPORT_SYMBOL_GPL(__netpoll_cleanup);
828
829static void netpoll_async_cleanup(struct work_struct *work)
830{
831	struct netpoll *np = container_of(work, struct netpoll, cleanup_work);
832
833	rtnl_lock();
834	__netpoll_cleanup(np);
835	rtnl_unlock();
836	kfree(np);
837}
838
839void __netpoll_free_async(struct netpoll *np)
840{
841	schedule_work(&np->cleanup_work);
842}
843EXPORT_SYMBOL_GPL(__netpoll_free_async);
844
845void netpoll_cleanup(struct netpoll *np)
846{
847	rtnl_lock();
848	if (!np->dev)
849		goto out;
850	__netpoll_cleanup(np);
851	dev_put(np->dev);
852	np->dev = NULL;
853out:
854	rtnl_unlock();
855}
856EXPORT_SYMBOL(netpoll_cleanup);
v4.17
  1/*
  2 * Common framework for low-level network console, dump, and debugger code
  3 *
  4 * Sep 8 2003  Matt Mackall <mpm@selenic.com>
  5 *
  6 * based on the netconsole code from:
  7 *
  8 * Copyright (C) 2001  Ingo Molnar <mingo@redhat.com>
  9 * Copyright (C) 2002  Red Hat, Inc.
 10 */
 11
 12#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 13
 14#include <linux/moduleparam.h>
 15#include <linux/kernel.h>
 16#include <linux/netdevice.h>
 17#include <linux/etherdevice.h>
 18#include <linux/string.h>
 19#include <linux/if_arp.h>
 20#include <linux/inetdevice.h>
 21#include <linux/inet.h>
 22#include <linux/interrupt.h>
 23#include <linux/netpoll.h>
 24#include <linux/sched.h>
 25#include <linux/delay.h>
 26#include <linux/rcupdate.h>
 27#include <linux/workqueue.h>
 28#include <linux/slab.h>
 29#include <linux/export.h>
 30#include <linux/if_vlan.h>
 31#include <net/tcp.h>
 32#include <net/udp.h>
 33#include <net/addrconf.h>
 34#include <net/ndisc.h>
 35#include <net/ip6_checksum.h>
 36#include <asm/unaligned.h>
 37#include <trace/events/napi.h>
 38
 39/*
 40 * We maintain a small pool of fully-sized skbs, to make sure the
 41 * message gets out even in extreme OOM situations.
 42 */
 43
 44#define MAX_UDP_CHUNK 1460
 45#define MAX_SKBS 32
 46
 47static struct sk_buff_head skb_pool;
 48
 49DEFINE_STATIC_SRCU(netpoll_srcu);
 50
 51#define USEC_PER_POLL	50
 52
 53#define MAX_SKB_SIZE							\
 54	(sizeof(struct ethhdr) +					\
 55	 sizeof(struct iphdr) +						\
 56	 sizeof(struct udphdr) +					\
 57	 MAX_UDP_CHUNK)
 58
 59static void zap_completion_queue(void);
 60static void netpoll_async_cleanup(struct work_struct *work);
 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		    netpoll_start_xmit(skb, dev, txq) != NETDEV_TX_OK) {
126			skb_queue_head(&npinfo->txq, skb);
127			HARD_TX_UNLOCK(dev, txq);
128			local_irq_restore(flags);
129
130			schedule_delayed_work(&npinfo->tx_work, HZ/10);
131			return;
132		}
133		HARD_TX_UNLOCK(dev, txq);
134		local_irq_restore(flags);
135	}
136}
137
138/*
139 * Check whether delayed processing was scheduled for our NIC. If so,
140 * we attempt to grab the poll lock and use ->poll() to pump the card.
141 * If this fails, either we've recursed in ->poll() or it's already
142 * running on another CPU.
143 *
144 * Note: we don't mask interrupts with this lock because we're using
145 * trylock here and interrupts are already disabled in the softirq
146 * case. Further, we test the poll_owner to avoid recursion on UP
147 * systems where the lock doesn't exist.
148 */
149static void poll_one_napi(struct napi_struct *napi)
150{
151	int work = 0;
152
153	/* net_rx_action's ->poll() invocations and our's are
154	 * synchronized by this test which is only made while
155	 * holding the napi->poll_lock.
156	 */
157	if (!test_bit(NAPI_STATE_SCHED, &napi->state))
158		return;
159
160	/* If we set this bit but see that it has already been set,
161	 * that indicates that napi has been disabled and we need
162	 * to abort this operation
163	 */
164	if (test_and_set_bit(NAPI_STATE_NPSVC, &napi->state))
165		return;
166
167	/* We explicilty pass the polling call a budget of 0 to
168	 * indicate that we are clearing the Tx path only.
169	 */
170	work = napi->poll(napi, 0);
171	WARN_ONCE(work, "%pF exceeded budget in poll\n", napi->poll);
172	trace_napi_poll(napi, work, 0);
173
174	clear_bit(NAPI_STATE_NPSVC, &napi->state);
 
 
175}
176
177static void poll_napi(struct net_device *dev)
178{
179	struct napi_struct *napi;
180	int cpu = smp_processor_id();
181
182	list_for_each_entry(napi, &dev->napi_list, dev_list) {
183		if (cmpxchg(&napi->poll_owner, -1, cpu) == -1) {
184			poll_one_napi(napi);
185			smp_store_release(&napi->poll_owner, -1);
 
186		}
187	}
188}
189
190static void netpoll_poll_dev(struct net_device *dev)
191{
192	const struct net_device_ops *ops;
193	struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
 
194
195	/* Don't do any rx activity if the dev_lock mutex is held
196	 * the dev_open/close paths use this to block netpoll activity
197	 * while changing device state
198	 */
199	if (down_trylock(&ni->dev_lock))
200		return;
201
202	if (!netif_running(dev)) {
203		up(&ni->dev_lock);
204		return;
205	}
206
207	ops = dev->netdev_ops;
208	if (!ops->ndo_poll_controller) {
209		up(&ni->dev_lock);
210		return;
211	}
212
213	/* Process pending work on NIC */
214	ops->ndo_poll_controller(dev);
215
216	poll_napi(dev);
217
218	up(&ni->dev_lock);
219
220	zap_completion_queue();
221}
222
223void netpoll_poll_disable(struct net_device *dev)
224{
225	struct netpoll_info *ni;
226	int idx;
227	might_sleep();
228	idx = srcu_read_lock(&netpoll_srcu);
229	ni = srcu_dereference(dev->npinfo, &netpoll_srcu);
230	if (ni)
231		down(&ni->dev_lock);
232	srcu_read_unlock(&netpoll_srcu, idx);
233}
234EXPORT_SYMBOL(netpoll_poll_disable);
235
236void netpoll_poll_enable(struct net_device *dev)
237{
238	struct netpoll_info *ni;
239	rcu_read_lock();
240	ni = rcu_dereference(dev->npinfo);
241	if (ni)
242		up(&ni->dev_lock);
243	rcu_read_unlock();
244}
245EXPORT_SYMBOL(netpoll_poll_enable);
246
247static void refill_skbs(void)
248{
249	struct sk_buff *skb;
250	unsigned long flags;
251
252	spin_lock_irqsave(&skb_pool.lock, flags);
253	while (skb_pool.qlen < MAX_SKBS) {
254		skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
255		if (!skb)
256			break;
257
258		__skb_queue_tail(&skb_pool, skb);
259	}
260	spin_unlock_irqrestore(&skb_pool.lock, flags);
261}
262
263static void zap_completion_queue(void)
264{
265	unsigned long flags;
266	struct softnet_data *sd = &get_cpu_var(softnet_data);
267
268	if (sd->completion_queue) {
269		struct sk_buff *clist;
270
271		local_irq_save(flags);
272		clist = sd->completion_queue;
273		sd->completion_queue = NULL;
274		local_irq_restore(flags);
275
276		while (clist != NULL) {
277			struct sk_buff *skb = clist;
278			clist = clist->next;
279			if (!skb_irq_freeable(skb)) {
280				refcount_set(&skb->users, 1);
281				dev_kfree_skb_any(skb); /* put this one back */
282			} else {
283				__kfree_skb(skb);
284			}
285		}
286	}
287
288	put_cpu_var(softnet_data);
289}
290
291static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
292{
293	int count = 0;
294	struct sk_buff *skb;
295
296	zap_completion_queue();
297	refill_skbs();
298repeat:
299
300	skb = alloc_skb(len, GFP_ATOMIC);
301	if (!skb)
302		skb = skb_dequeue(&skb_pool);
303
304	if (!skb) {
305		if (++count < 10) {
306			netpoll_poll_dev(np->dev);
307			goto repeat;
308		}
309		return NULL;
310	}
311
312	refcount_set(&skb->users, 1);
313	skb_reserve(skb, reserve);
314	return skb;
315}
316
317static int netpoll_owner_active(struct net_device *dev)
318{
319	struct napi_struct *napi;
320
321	list_for_each_entry(napi, &dev->napi_list, dev_list) {
322		if (napi->poll_owner == smp_processor_id())
323			return 1;
324	}
325	return 0;
326}
327
328/* call with IRQ disabled */
329void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb,
330			     struct net_device *dev)
331{
332	int status = NETDEV_TX_BUSY;
333	unsigned long tries;
334	/* It is up to the caller to keep npinfo alive. */
335	struct netpoll_info *npinfo;
336
337	lockdep_assert_irqs_disabled();
338
339	npinfo = rcu_dereference_bh(np->dev->npinfo);
340	if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
341		dev_kfree_skb_irq(skb);
342		return;
343	}
344
345	/* don't get messages out of order, and no recursion */
346	if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
347		struct netdev_queue *txq;
348
349		txq = netdev_pick_tx(dev, skb, NULL);
350
351		/* try until next clock tick */
352		for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
353		     tries > 0; --tries) {
354			if (HARD_TX_TRYLOCK(dev, txq)) {
355				if (!netif_xmit_stopped(txq))
356					status = netpoll_start_xmit(skb, dev, txq);
357
358				HARD_TX_UNLOCK(dev, txq);
359
360				if (status == NETDEV_TX_OK)
361					break;
362
363			}
364
365			/* tickle device maybe there is some cleanup */
366			netpoll_poll_dev(np->dev);
367
368			udelay(USEC_PER_POLL);
369		}
370
371		WARN_ONCE(!irqs_disabled(),
372			"netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pF)\n",
373			dev->name, dev->netdev_ops->ndo_start_xmit);
374
375	}
376
377	if (status != NETDEV_TX_OK) {
378		skb_queue_tail(&npinfo->txq, skb);
379		schedule_delayed_work(&npinfo->tx_work,0);
380	}
381}
382EXPORT_SYMBOL(netpoll_send_skb_on_dev);
383
384void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
385{
386	int total_len, ip_len, udp_len;
387	struct sk_buff *skb;
388	struct udphdr *udph;
389	struct iphdr *iph;
390	struct ethhdr *eth;
391	static atomic_t ip_ident;
392	struct ipv6hdr *ip6h;
393
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		put_unaligned(0x60, (unsigned char *)ip6h);
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		put_unaligned(0x45, (unsigned char *)iph);
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	INIT_WORK(&np->cleanup_work, netpoll_async_cleanup);
615
616	if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) ||
617	    !ndev->netdev_ops->ndo_poll_controller) {
618		np_err(np, "%s doesn't support polling, aborting\n",
619		       np->dev_name);
620		err = -ENOTSUPP;
621		goto out;
622	}
623
624	if (!ndev->npinfo) {
625		npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
626		if (!npinfo) {
627			err = -ENOMEM;
628			goto out;
629		}
630
631		sema_init(&npinfo->dev_lock, 1);
632		skb_queue_head_init(&npinfo->txq);
633		INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
634
635		refcount_set(&npinfo->refcnt, 1);
636
637		ops = np->dev->netdev_ops;
638		if (ops->ndo_netpoll_setup) {
639			err = ops->ndo_netpoll_setup(ndev, npinfo);
640			if (err)
641				goto free_npinfo;
642		}
643	} else {
644		npinfo = rtnl_dereference(ndev->npinfo);
645		refcount_inc(&npinfo->refcnt);
646	}
647
648	npinfo->netpoll = np;
649
650	/* last thing to do is link it to the net device structure */
651	rcu_assign_pointer(ndev->npinfo, npinfo);
652
653	return 0;
654
655free_npinfo:
656	kfree(npinfo);
657out:
658	return err;
659}
660EXPORT_SYMBOL_GPL(__netpoll_setup);
661
662int netpoll_setup(struct netpoll *np)
663{
664	struct net_device *ndev = NULL;
665	struct in_device *in_dev;
666	int err;
667
668	rtnl_lock();
669	if (np->dev_name[0]) {
670		struct net *net = current->nsproxy->net_ns;
671		ndev = __dev_get_by_name(net, np->dev_name);
672	}
673	if (!ndev) {
674		np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
675		err = -ENODEV;
676		goto unlock;
677	}
678	dev_hold(ndev);
679
680	if (netdev_master_upper_dev_get(ndev)) {
681		np_err(np, "%s is a slave device, aborting\n", np->dev_name);
682		err = -EBUSY;
683		goto put;
684	}
685
686	if (!netif_running(ndev)) {
687		unsigned long atmost, atleast;
688
689		np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
690
691		err = dev_open(ndev);
692
693		if (err) {
694			np_err(np, "failed to open %s\n", ndev->name);
695			goto put;
696		}
697
698		rtnl_unlock();
699		atleast = jiffies + HZ/10;
700		atmost = jiffies + carrier_timeout * HZ;
701		while (!netif_carrier_ok(ndev)) {
702			if (time_after(jiffies, atmost)) {
703				np_notice(np, "timeout waiting for carrier\n");
704				break;
705			}
706			msleep(1);
707		}
708
709		/* If carrier appears to come up instantly, we don't
710		 * trust it and pause so that we don't pump all our
711		 * queued console messages into the bitbucket.
712		 */
713
714		if (time_before(jiffies, atleast)) {
715			np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
716			msleep(4000);
717		}
718		rtnl_lock();
719	}
720
721	if (!np->local_ip.ip) {
722		if (!np->ipv6) {
723			in_dev = __in_dev_get_rtnl(ndev);
724
725			if (!in_dev || !in_dev->ifa_list) {
726				np_err(np, "no IP address for %s, aborting\n",
727				       np->dev_name);
728				err = -EDESTADDRREQ;
729				goto put;
730			}
731
732			np->local_ip.ip = in_dev->ifa_list->ifa_local;
733			np_info(np, "local IP %pI4\n", &np->local_ip.ip);
734		} else {
735#if IS_ENABLED(CONFIG_IPV6)
736			struct inet6_dev *idev;
737
738			err = -EDESTADDRREQ;
739			idev = __in6_dev_get(ndev);
740			if (idev) {
741				struct inet6_ifaddr *ifp;
742
743				read_lock_bh(&idev->lock);
744				list_for_each_entry(ifp, &idev->addr_list, if_list) {
745					if (ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL)
746						continue;
747					np->local_ip.in6 = ifp->addr;
748					err = 0;
749					break;
750				}
751				read_unlock_bh(&idev->lock);
752			}
753			if (err) {
754				np_err(np, "no IPv6 address for %s, aborting\n",
755				       np->dev_name);
756				goto put;
757			} else
758				np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
759#else
760			np_err(np, "IPv6 is not supported %s, aborting\n",
761			       np->dev_name);
762			err = -EINVAL;
763			goto put;
764#endif
765		}
766	}
767
768	/* fill up the skb queue */
769	refill_skbs();
770
771	err = __netpoll_setup(np, ndev);
772	if (err)
773		goto put;
774
775	rtnl_unlock();
776	return 0;
777
778put:
779	dev_put(ndev);
780unlock:
781	rtnl_unlock();
782	return err;
783}
784EXPORT_SYMBOL(netpoll_setup);
785
786static int __init netpoll_init(void)
787{
788	skb_queue_head_init(&skb_pool);
789	return 0;
790}
791core_initcall(netpoll_init);
792
793static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
794{
795	struct netpoll_info *npinfo =
796			container_of(rcu_head, struct netpoll_info, rcu);
797
798	skb_queue_purge(&npinfo->txq);
799
800	/* we can't call cancel_delayed_work_sync here, as we are in softirq */
801	cancel_delayed_work(&npinfo->tx_work);
802
803	/* clean after last, unfinished work */
804	__skb_queue_purge(&npinfo->txq);
805	/* now cancel it again */
806	cancel_delayed_work(&npinfo->tx_work);
807	kfree(npinfo);
808}
809
810void __netpoll_cleanup(struct netpoll *np)
811{
812	struct netpoll_info *npinfo;
813
814	/* rtnl_dereference would be preferable here but
815	 * rcu_cleanup_netpoll path can put us in here safely without
816	 * holding the rtnl, so plain rcu_dereference it is
817	 */
818	npinfo = rtnl_dereference(np->dev->npinfo);
819	if (!npinfo)
820		return;
821
822	synchronize_srcu(&netpoll_srcu);
823
824	if (refcount_dec_and_test(&npinfo->refcnt)) {
825		const struct net_device_ops *ops;
826
827		ops = np->dev->netdev_ops;
828		if (ops->ndo_netpoll_cleanup)
829			ops->ndo_netpoll_cleanup(np->dev);
830
831		RCU_INIT_POINTER(np->dev->npinfo, NULL);
832		call_rcu_bh(&npinfo->rcu, rcu_cleanup_netpoll_info);
833	} else
834		RCU_INIT_POINTER(np->dev->npinfo, NULL);
835}
836EXPORT_SYMBOL_GPL(__netpoll_cleanup);
837
838static void netpoll_async_cleanup(struct work_struct *work)
839{
840	struct netpoll *np = container_of(work, struct netpoll, cleanup_work);
841
842	rtnl_lock();
843	__netpoll_cleanup(np);
844	rtnl_unlock();
845	kfree(np);
846}
847
848void __netpoll_free_async(struct netpoll *np)
849{
850	schedule_work(&np->cleanup_work);
851}
852EXPORT_SYMBOL_GPL(__netpoll_free_async);
853
854void netpoll_cleanup(struct netpoll *np)
855{
856	rtnl_lock();
857	if (!np->dev)
858		goto out;
859	__netpoll_cleanup(np);
860	dev_put(np->dev);
861	np->dev = NULL;
862out:
863	rtnl_unlock();
864}
865EXPORT_SYMBOL(netpoll_cleanup);