Linux Audio

Check our new training course

Loading...
v3.1
 
  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#include <linux/moduleparam.h>
 
 13#include <linux/netdevice.h>
 14#include <linux/etherdevice.h>
 15#include <linux/string.h>
 16#include <linux/if_arp.h>
 17#include <linux/inetdevice.h>
 18#include <linux/inet.h>
 19#include <linux/interrupt.h>
 20#include <linux/netpoll.h>
 21#include <linux/sched.h>
 22#include <linux/delay.h>
 23#include <linux/rcupdate.h>
 24#include <linux/workqueue.h>
 25#include <linux/slab.h>
 
 
 26#include <net/tcp.h>
 27#include <net/udp.h>
 28#include <asm/unaligned.h>
 
 
 
 29#include <trace/events/napi.h>
 
 30
 31/*
 32 * We maintain a small pool of fully-sized skbs, to make sure the
 33 * message gets out even in extreme OOM situations.
 34 */
 35
 36#define MAX_UDP_CHUNK 1460
 37#define MAX_SKBS 32
 38
 39static struct sk_buff_head skb_pool;
 40
 41static atomic_t trapped;
 42
 43#define USEC_PER_POLL	50
 44#define NETPOLL_RX_ENABLED  1
 45#define NETPOLL_RX_DROP     2
 46
 47#define MAX_SKB_SIZE \
 48		(MAX_UDP_CHUNK + sizeof(struct udphdr) + \
 49				sizeof(struct iphdr) + sizeof(struct ethhdr))
 
 
 50
 51static void zap_completion_queue(void);
 52static void arp_reply(struct sk_buff *skb);
 53
 54static unsigned int carrier_timeout = 4;
 55module_param(carrier_timeout, uint, 0644);
 56
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 57static void queue_process(struct work_struct *work)
 58{
 59	struct netpoll_info *npinfo =
 60		container_of(work, struct netpoll_info, tx_work.work);
 61	struct sk_buff *skb;
 62	unsigned long flags;
 63
 64	while ((skb = skb_dequeue(&npinfo->txq))) {
 65		struct net_device *dev = skb->dev;
 66		const struct net_device_ops *ops = dev->netdev_ops;
 67		struct netdev_queue *txq;
 
 68
 69		if (!netif_device_present(dev) || !netif_running(dev)) {
 70			__kfree_skb(skb);
 71			continue;
 72		}
 73
 74		txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
 75
 76		local_irq_save(flags);
 77		__netif_tx_lock(txq, smp_processor_id());
 78		if (netif_tx_queue_frozen_or_stopped(txq) ||
 79		    ops->ndo_start_xmit(skb, dev) != NETDEV_TX_OK) {
 
 
 
 
 
 
 
 80			skb_queue_head(&npinfo->txq, skb);
 81			__netif_tx_unlock(txq);
 82			local_irq_restore(flags);
 83
 84			schedule_delayed_work(&npinfo->tx_work, HZ/10);
 85			return;
 86		}
 87		__netif_tx_unlock(txq);
 88		local_irq_restore(flags);
 89	}
 90}
 91
 92static __sum16 checksum_udp(struct sk_buff *skb, struct udphdr *uh,
 93			    unsigned short ulen, __be32 saddr, __be32 daddr)
 94{
 95	__wsum psum;
 96
 97	if (uh->check == 0 || skb_csum_unnecessary(skb))
 98		return 0;
 99
100	psum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0);
 
101
102	if (skb->ip_summed == CHECKSUM_COMPLETE &&
103	    !csum_fold(csum_add(psum, skb->csum)))
104		return 0;
105
106	skb->csum = psum;
107
108	return __skb_checksum_complete(skb);
109}
110
111/*
112 * Check whether delayed processing was scheduled for our NIC. If so,
113 * we attempt to grab the poll lock and use ->poll() to pump the card.
114 * If this fails, either we've recursed in ->poll() or it's already
115 * running on another CPU.
116 *
117 * Note: we don't mask interrupts with this lock because we're using
118 * trylock here and interrupts are already disabled in the softirq
119 * case. Further, we test the poll_owner to avoid recursion on UP
120 * systems where the lock doesn't exist.
121 *
122 * In cases where there is bi-directional communications, reading only
123 * one message at a time can lead to packets being dropped by the
124 * network adapter, forcing superfluous retries and possibly timeouts.
125 * Thus, we set our budget to greater than 1.
126 */
127static int poll_one_napi(struct netpoll_info *npinfo,
128			 struct napi_struct *napi, int budget)
129{
130	int work;
131
132	/* net_rx_action's ->poll() invocations and our's are
133	 * synchronized by this test which is only made while
134	 * holding the napi->poll_lock.
135	 */
136	if (!test_bit(NAPI_STATE_SCHED, &napi->state))
137		return budget;
138
139	npinfo->rx_flags |= NETPOLL_RX_DROP;
140	atomic_inc(&trapped);
141	set_bit(NAPI_STATE_NPSVC, &napi->state);
142
143	work = napi->poll(napi, budget);
144	trace_napi_poll(napi);
 
 
 
 
145
146	clear_bit(NAPI_STATE_NPSVC, &napi->state);
147	atomic_dec(&trapped);
148	npinfo->rx_flags &= ~NETPOLL_RX_DROP;
149
150	return budget - work;
151}
152
153static void poll_napi(struct net_device *dev)
154{
155	struct napi_struct *napi;
156	int budget = 16;
157
158	list_for_each_entry(napi, &dev->napi_list, dev_list) {
159		if (napi->poll_owner != smp_processor_id() &&
160		    spin_trylock(&napi->poll_lock)) {
161			budget = poll_one_napi(dev->npinfo, napi, budget);
162			spin_unlock(&napi->poll_lock);
163
164			if (!budget)
165				break;
 
 
166		}
167	}
168}
169
170static void service_arp_queue(struct netpoll_info *npi)
171{
172	if (npi) {
173		struct sk_buff *skb;
174
175		while ((skb = skb_dequeue(&npi->arp_tx)))
176			arp_reply(skb);
177	}
178}
179
180static void netpoll_poll_dev(struct net_device *dev)
181{
 
182	const struct net_device_ops *ops;
183
184	if (!dev || !netif_running(dev))
 
 
 
 
185		return;
186
187	ops = dev->netdev_ops;
188	if (!ops->ndo_poll_controller)
 
 
 
189		return;
 
190
191	/* Process pending work on NIC */
192	ops->ndo_poll_controller(dev);
 
193
194	poll_napi(dev);
195
196	if (dev->priv_flags & IFF_SLAVE) {
197		if (dev->npinfo) {
198			struct net_device *bond_dev = dev->master;
199			struct sk_buff *skb;
200			while ((skb = skb_dequeue(&dev->npinfo->arp_tx))) {
201				skb->dev = bond_dev;
202				skb_queue_tail(&bond_dev->npinfo->arp_tx, skb);
203			}
204		}
205	}
206
207	service_arp_queue(dev->npinfo);
208
209	zap_completion_queue();
210}
 
211
212static void refill_skbs(void)
213{
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
214	struct sk_buff *skb;
215	unsigned long flags;
216
217	spin_lock_irqsave(&skb_pool.lock, flags);
218	while (skb_pool.qlen < MAX_SKBS) {
 
 
219		skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
220		if (!skb)
221			break;
222
223		__skb_queue_tail(&skb_pool, skb);
224	}
225	spin_unlock_irqrestore(&skb_pool.lock, flags);
226}
227
228static void zap_completion_queue(void)
229{
230	unsigned long flags;
231	struct softnet_data *sd = &get_cpu_var(softnet_data);
232
233	if (sd->completion_queue) {
234		struct sk_buff *clist;
235
236		local_irq_save(flags);
237		clist = sd->completion_queue;
238		sd->completion_queue = NULL;
239		local_irq_restore(flags);
240
241		while (clist != NULL) {
242			struct sk_buff *skb = clist;
243			clist = clist->next;
244			if (skb->destructor) {
245				atomic_inc(&skb->users);
246				dev_kfree_skb_any(skb); /* put this one back */
247			} else {
248				__kfree_skb(skb);
249			}
250		}
251	}
252
253	put_cpu_var(softnet_data);
254}
255
256static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
257{
258	int count = 0;
259	struct sk_buff *skb;
260
261	zap_completion_queue();
262	refill_skbs();
263repeat:
264
265	skb = alloc_skb(len, GFP_ATOMIC);
266	if (!skb)
267		skb = skb_dequeue(&skb_pool);
268
269	if (!skb) {
270		if (++count < 10) {
271			netpoll_poll_dev(np->dev);
272			goto repeat;
273		}
274		return NULL;
275	}
276
277	atomic_set(&skb->users, 1);
278	skb_reserve(skb, reserve);
279	return skb;
280}
281
282static int netpoll_owner_active(struct net_device *dev)
283{
284	struct napi_struct *napi;
285
286	list_for_each_entry(napi, &dev->napi_list, dev_list) {
287		if (napi->poll_owner == smp_processor_id())
288			return 1;
289	}
290	return 0;
291}
292
293void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb,
294			     struct net_device *dev)
295{
296	int status = NETDEV_TX_BUSY;
 
297	unsigned long tries;
298	const struct net_device_ops *ops = dev->netdev_ops;
299	/* It is up to the caller to keep npinfo alive. */
300	struct netpoll_info *npinfo = np->dev->npinfo;
 
 
 
 
 
301
302	if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
303		__kfree_skb(skb);
304		return;
305	}
306
307	/* don't get messages out of order, and no recursion */
308	if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
309		struct netdev_queue *txq;
310		unsigned long flags;
311
312		txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
313
314		local_irq_save(flags);
315		/* try until next clock tick */
316		for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
317		     tries > 0; --tries) {
318			if (__netif_tx_trylock(txq)) {
319				if (!netif_tx_queue_stopped(txq)) {
320					status = ops->ndo_start_xmit(skb, dev);
321					if (status == NETDEV_TX_OK)
322						txq_trans_update(txq);
323				}
324				__netif_tx_unlock(txq);
325
326				if (status == NETDEV_TX_OK)
 
 
327					break;
328
329			}
330
331			/* tickle device maybe there is some cleanup */
332			netpoll_poll_dev(np->dev);
333
334			udelay(USEC_PER_POLL);
335		}
336
337		WARN_ONCE(!irqs_disabled(),
338			"netpoll_send_skb(): %s enabled interrupts in poll (%pF)\n",
339			dev->name, ops->ndo_start_xmit);
340
341		local_irq_restore(flags);
342	}
343
344	if (status != NETDEV_TX_OK) {
345		skb_queue_tail(&npinfo->txq, skb);
346		schedule_delayed_work(&npinfo->tx_work,0);
347	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
348}
349EXPORT_SYMBOL(netpoll_send_skb_on_dev);
350
351void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
352{
353	int total_len, eth_len, ip_len, udp_len;
354	struct sk_buff *skb;
355	struct udphdr *udph;
356	struct iphdr *iph;
357	struct ethhdr *eth;
 
 
 
 
 
358
359	udp_len = len + sizeof(*udph);
360	ip_len = eth_len = udp_len + sizeof(*iph);
361	total_len = eth_len + ETH_HLEN + NET_IP_ALIGN;
 
 
 
 
362
363	skb = find_skb(np, total_len, total_len - len);
 
364	if (!skb)
365		return;
366
367	skb_copy_to_linear_data(skb, msg, len);
368	skb->len += len;
369
370	skb_push(skb, sizeof(*udph));
371	skb_reset_transport_header(skb);
372	udph = udp_hdr(skb);
373	udph->source = htons(np->local_port);
374	udph->dest = htons(np->remote_port);
375	udph->len = htons(udp_len);
376	udph->check = 0;
377	udph->check = csum_tcpudp_magic(np->local_ip,
378					np->remote_ip,
379					udp_len, IPPROTO_UDP,
380					csum_partial(udph, udp_len, 0));
381	if (udph->check == 0)
382		udph->check = CSUM_MANGLED_0;
383
384	skb_push(skb, sizeof(*iph));
385	skb_reset_network_header(skb);
386	iph = ip_hdr(skb);
387
388	/* iph->version = 4; iph->ihl = 5; */
389	put_unaligned(0x45, (unsigned char *)iph);
390	iph->tos      = 0;
391	put_unaligned(htons(ip_len), &(iph->tot_len));
392	iph->id       = 0;
393	iph->frag_off = 0;
394	iph->ttl      = 64;
395	iph->protocol = IPPROTO_UDP;
396	iph->check    = 0;
397	put_unaligned(np->local_ip, &(iph->saddr));
398	put_unaligned(np->remote_ip, &(iph->daddr));
399	iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
400
401	eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
402	skb_reset_mac_header(skb);
403	skb->protocol = eth->h_proto = htons(ETH_P_IP);
404	memcpy(eth->h_source, np->dev->dev_addr, ETH_ALEN);
405	memcpy(eth->h_dest, np->remote_mac, ETH_ALEN);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
406
407	skb->dev = np->dev;
408
409	netpoll_send_skb(np, skb);
410}
411EXPORT_SYMBOL(netpoll_send_udp);
412
413static void arp_reply(struct sk_buff *skb)
414{
415	struct netpoll_info *npinfo = skb->dev->npinfo;
416	struct arphdr *arp;
417	unsigned char *arp_ptr;
418	int size, type = ARPOP_REPLY, ptype = ETH_P_ARP;
419	__be32 sip, tip;
420	unsigned char *sha;
421	struct sk_buff *send_skb;
422	struct netpoll *np, *tmp;
423	unsigned long flags;
424	int hits = 0;
425
426	if (list_empty(&npinfo->rx_np))
427		return;
428
429	/* Before checking the packet, we do some early
430	   inspection whether this is interesting at all */
431	spin_lock_irqsave(&npinfo->rx_lock, flags);
432	list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
433		if (np->dev == skb->dev)
434			hits++;
435	}
436	spin_unlock_irqrestore(&npinfo->rx_lock, flags);
437
438	/* No netpoll struct is using this dev */
439	if (!hits)
440		return;
441
442	/* No arp on this interface */
443	if (skb->dev->flags & IFF_NOARP)
444		return;
445
446	if (!pskb_may_pull(skb, arp_hdr_len(skb->dev)))
447		return;
448
449	skb_reset_network_header(skb);
450	skb_reset_transport_header(skb);
451	arp = arp_hdr(skb);
452
453	if ((arp->ar_hrd != htons(ARPHRD_ETHER) &&
454	     arp->ar_hrd != htons(ARPHRD_IEEE802)) ||
455	    arp->ar_pro != htons(ETH_P_IP) ||
456	    arp->ar_op != htons(ARPOP_REQUEST))
457		return;
458
459	arp_ptr = (unsigned char *)(arp+1);
460	/* save the location of the src hw addr */
461	sha = arp_ptr;
462	arp_ptr += skb->dev->addr_len;
463	memcpy(&sip, arp_ptr, 4);
464	arp_ptr += 4;
465	/* If we actually cared about dst hw addr,
466	   it would get copied here */
467	arp_ptr += skb->dev->addr_len;
468	memcpy(&tip, arp_ptr, 4);
469
470	/* Should we ignore arp? */
471	if (ipv4_is_loopback(tip) || ipv4_is_multicast(tip))
472		return;
473
474	size = arp_hdr_len(skb->dev);
475
476	spin_lock_irqsave(&npinfo->rx_lock, flags);
477	list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
478		if (tip != np->local_ip)
479			continue;
480
481		send_skb = find_skb(np, size + LL_ALLOCATED_SPACE(np->dev),
482				    LL_RESERVED_SPACE(np->dev));
483		if (!send_skb)
484			continue;
485
486		skb_reset_network_header(send_skb);
487		arp = (struct arphdr *) skb_put(send_skb, size);
488		send_skb->dev = skb->dev;
489		send_skb->protocol = htons(ETH_P_ARP);
490
491		/* Fill the device header for the ARP frame */
492		if (dev_hard_header(send_skb, skb->dev, ptype,
493				    sha, np->dev->dev_addr,
494				    send_skb->len) < 0) {
495			kfree_skb(send_skb);
496			continue;
497		}
498
499		/*
500		 * Fill out the arp protocol part.
501		 *
502		 * we only support ethernet device type,
503		 * which (according to RFC 1390) should
504		 * always equal 1 (Ethernet).
505		 */
506
507		arp->ar_hrd = htons(np->dev->type);
508		arp->ar_pro = htons(ETH_P_IP);
509		arp->ar_hln = np->dev->addr_len;
510		arp->ar_pln = 4;
511		arp->ar_op = htons(type);
512
513		arp_ptr = (unsigned char *)(arp + 1);
514		memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
515		arp_ptr += np->dev->addr_len;
516		memcpy(arp_ptr, &tip, 4);
517		arp_ptr += 4;
518		memcpy(arp_ptr, sha, np->dev->addr_len);
519		arp_ptr += np->dev->addr_len;
520		memcpy(arp_ptr, &sip, 4);
521
522		netpoll_send_skb(np, send_skb);
523
524		/* If there are several rx_hooks for the same address,
525		   we're fine by sending a single reply */
526		break;
527	}
528	spin_unlock_irqrestore(&npinfo->rx_lock, flags);
529}
530
531int __netpoll_rx(struct sk_buff *skb)
532{
533	int proto, len, ulen;
534	int hits = 0;
535	const struct iphdr *iph;
536	struct udphdr *uh;
537	struct netpoll_info *npinfo = skb->dev->npinfo;
538	struct netpoll *np, *tmp;
539
540	if (list_empty(&npinfo->rx_np))
541		goto out;
542
543	if (skb->dev->type != ARPHRD_ETHER)
544		goto out;
545
546	/* check if netpoll clients need ARP */
547	if (skb->protocol == htons(ETH_P_ARP) &&
548	    atomic_read(&trapped)) {
549		skb_queue_tail(&npinfo->arp_tx, skb);
550		return 1;
551	}
552
553	proto = ntohs(eth_hdr(skb)->h_proto);
554	if (proto != ETH_P_IP)
555		goto out;
556	if (skb->pkt_type == PACKET_OTHERHOST)
557		goto out;
558	if (skb_shared(skb))
559		goto out;
560
561	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
562		goto out;
563	iph = (struct iphdr *)skb->data;
564	if (iph->ihl < 5 || iph->version != 4)
565		goto out;
566	if (!pskb_may_pull(skb, iph->ihl*4))
567		goto out;
568	iph = (struct iphdr *)skb->data;
569	if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
570		goto out;
571
572	len = ntohs(iph->tot_len);
573	if (skb->len < len || len < iph->ihl*4)
574		goto out;
575
576	/*
577	 * Our transport medium may have padded the buffer out.
578	 * Now We trim to the true length of the frame.
579	 */
580	if (pskb_trim_rcsum(skb, len))
581		goto out;
582
583	iph = (struct iphdr *)skb->data;
584	if (iph->protocol != IPPROTO_UDP)
585		goto out;
586
587	len -= iph->ihl*4;
588	uh = (struct udphdr *)(((char *)iph) + iph->ihl*4);
589	ulen = ntohs(uh->len);
590
591	if (ulen != len)
592		goto out;
593	if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr))
594		goto out;
595
596	list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
597		if (np->local_ip && np->local_ip != iph->daddr)
598			continue;
599		if (np->remote_ip && np->remote_ip != iph->saddr)
600			continue;
601		if (np->local_port && np->local_port != ntohs(uh->dest))
602			continue;
603
604		np->rx_hook(np, ntohs(uh->source),
605			       (char *)(uh+1),
606			       ulen - sizeof(struct udphdr));
607		hits++;
608	}
609
610	if (!hits)
611		goto out;
612
613	kfree_skb(skb);
614	return 1;
 
615
616out:
617	if (atomic_read(&trapped)) {
618		kfree_skb(skb);
619		return 1;
 
 
 
 
 
 
 
 
620	}
621
622	return 0;
623}
624
625void netpoll_print_options(struct netpoll *np)
626{
627	printk(KERN_INFO "%s: local port %d\n",
628			 np->name, np->local_port);
629	printk(KERN_INFO "%s: local IP %pI4\n",
630			 np->name, &np->local_ip);
631	printk(KERN_INFO "%s: interface '%s'\n",
632			 np->name, np->dev_name);
633	printk(KERN_INFO "%s: remote port %d\n",
634			 np->name, np->remote_port);
635	printk(KERN_INFO "%s: remote IP %pI4\n",
636			 np->name, &np->remote_ip);
637	printk(KERN_INFO "%s: remote ethernet address %pM\n",
638	                 np->name, np->remote_mac);
639}
640EXPORT_SYMBOL(netpoll_print_options);
641
642int netpoll_parse_options(struct netpoll *np, char *opt)
643{
644	char *cur=opt, *delim;
 
 
645
646	if (*cur != '@') {
647		if ((delim = strchr(cur, '@')) == NULL)
648			goto parse_failed;
649		*delim = 0;
650		np->local_port = simple_strtol(cur, NULL, 10);
 
651		cur = delim;
652	}
653	cur++;
654
655	if (*cur != '/') {
 
656		if ((delim = strchr(cur, '/')) == NULL)
657			goto parse_failed;
658		*delim = 0;
659		np->local_ip = in_aton(cur);
 
 
 
 
660		cur = delim;
661	}
662	cur++;
663
664	if (*cur != ',') {
665		/* parse out dev name */
666		if ((delim = strchr(cur, ',')) == NULL)
667			goto parse_failed;
668		*delim = 0;
669		strlcpy(np->dev_name, cur, sizeof(np->dev_name));
670		cur = delim;
671	}
672	cur++;
673
674	if (*cur != '@') {
675		/* dst port */
676		if ((delim = strchr(cur, '@')) == NULL)
677			goto parse_failed;
678		*delim = 0;
679		if (*cur == ' ' || *cur == '\t')
680			printk(KERN_INFO "%s: warning: whitespace"
681					"is not allowed\n", np->name);
682		np->remote_port = simple_strtol(cur, NULL, 10);
683		cur = delim;
684	}
685	cur++;
686
687	/* dst ip */
688	if ((delim = strchr(cur, '/')) == NULL)
689		goto parse_failed;
690	*delim = 0;
691	np->remote_ip = in_aton(cur);
 
 
 
 
 
 
692	cur = delim + 1;
693
694	if (*cur != 0) {
695		/* MAC address */
696		if (!mac_pton(cur, np->remote_mac))
697			goto parse_failed;
698	}
699
700	netpoll_print_options(np);
701
702	return 0;
703
704 parse_failed:
705	printk(KERN_INFO "%s: couldn't parse config at '%s'!\n",
706	       np->name, cur);
707	return -1;
708}
709EXPORT_SYMBOL(netpoll_parse_options);
710
711int __netpoll_setup(struct netpoll *np)
712{
713	struct net_device *ndev = np->dev;
714	struct netpoll_info *npinfo;
715	const struct net_device_ops *ops;
716	unsigned long flags;
717	int err;
718
719	if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) ||
720	    !ndev->netdev_ops->ndo_poll_controller) {
721		printk(KERN_ERR "%s: %s doesn't support polling, aborting.\n",
722		       np->name, np->dev_name);
 
723		err = -ENOTSUPP;
724		goto out;
725	}
726
727	if (!ndev->npinfo) {
728		npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
729		if (!npinfo) {
730			err = -ENOMEM;
731			goto out;
732		}
733
734		npinfo->rx_flags = 0;
735		INIT_LIST_HEAD(&npinfo->rx_np);
736
737		spin_lock_init(&npinfo->rx_lock);
738		skb_queue_head_init(&npinfo->arp_tx);
739		skb_queue_head_init(&npinfo->txq);
740		INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
741
742		atomic_set(&npinfo->refcnt, 1);
743
744		ops = np->dev->netdev_ops;
745		if (ops->ndo_netpoll_setup) {
746			err = ops->ndo_netpoll_setup(ndev, npinfo);
747			if (err)
748				goto free_npinfo;
749		}
750	} else {
751		npinfo = ndev->npinfo;
752		atomic_inc(&npinfo->refcnt);
753	}
754
 
 
755	npinfo->netpoll = np;
756
757	if (np->rx_hook) {
758		spin_lock_irqsave(&npinfo->rx_lock, flags);
759		npinfo->rx_flags |= NETPOLL_RX_ENABLED;
760		list_add_tail(&np->rx, &npinfo->rx_np);
761		spin_unlock_irqrestore(&npinfo->rx_lock, flags);
762	}
763
764	/* last thing to do is link it to the net device structure */
765	rcu_assign_pointer(ndev->npinfo, npinfo);
766
767	return 0;
768
769free_npinfo:
770	kfree(npinfo);
771out:
772	return err;
773}
774EXPORT_SYMBOL_GPL(__netpoll_setup);
775
776int netpoll_setup(struct netpoll *np)
777{
778	struct net_device *ndev = NULL;
 
779	struct in_device *in_dev;
780	int err;
781
782	if (np->dev_name)
783		ndev = dev_get_by_name(&init_net, np->dev_name);
 
 
 
784	if (!ndev) {
785		printk(KERN_ERR "%s: %s doesn't exist, aborting.\n",
786		       np->name, np->dev_name);
787		return -ENODEV;
788	}
 
789
790	if (ndev->master) {
791		printk(KERN_ERR "%s: %s is a slave device, aborting.\n",
792		       np->name, np->dev_name);
793		err = -EBUSY;
794		goto put;
795	}
796
797	if (!netif_running(ndev)) {
798		unsigned long atmost, atleast;
799
800		printk(KERN_INFO "%s: device %s not up yet, forcing it\n",
801		       np->name, np->dev_name);
802
803		rtnl_lock();
804		err = dev_open(ndev);
805		rtnl_unlock();
806
807		if (err) {
808			printk(KERN_ERR "%s: failed to open %s\n",
809			       np->name, ndev->name);
810			goto put;
811		}
812
813		atleast = jiffies + HZ/10;
814		atmost = jiffies + carrier_timeout * HZ;
815		while (!netif_carrier_ok(ndev)) {
816			if (time_after(jiffies, atmost)) {
817				printk(KERN_NOTICE
818				       "%s: timeout waiting for carrier\n",
819				       np->name);
820				break;
821			}
822			msleep(1);
823		}
824
825		/* If carrier appears to come up instantly, we don't
826		 * trust it and pause so that we don't pump all our
827		 * queued console messages into the bitbucket.
828		 */
829
830		if (time_before(jiffies, atleast)) {
831			printk(KERN_NOTICE "%s: carrier detect appears"
832			       " untrustworthy, waiting 4 seconds\n",
833			       np->name);
834			msleep(4000);
835		}
836	}
837
838	if (!np->local_ip) {
839		rcu_read_lock();
840		in_dev = __in_dev_get_rcu(ndev);
841
842		if (!in_dev || !in_dev->ifa_list) {
843			rcu_read_unlock();
844			printk(KERN_ERR "%s: no IP address for %s, aborting\n",
845			       np->name, np->dev_name);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
846			err = -EDESTADDRREQ;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
847			goto put;
 
848		}
849
850		np->local_ip = in_dev->ifa_list->ifa_local;
851		rcu_read_unlock();
852		printk(KERN_INFO "%s: local IP %pI4\n", np->name, &np->local_ip);
853	}
854
855	np->dev = ndev;
856
857	/* fill up the skb queue */
858	refill_skbs();
859
860	rtnl_lock();
861	err = __netpoll_setup(np);
862	rtnl_unlock();
863
864	if (err)
865		goto put;
866
867	return 0;
868
 
 
869put:
870	dev_put(ndev);
 
 
 
 
 
871	return err;
872}
873EXPORT_SYMBOL(netpoll_setup);
874
875static int __init netpoll_init(void)
876{
877	skb_queue_head_init(&skb_pool);
878	return 0;
 
 
 
 
 
 
 
 
 
 
 
879}
880core_initcall(netpoll_init);
881
882void __netpoll_cleanup(struct netpoll *np)
883{
884	struct netpoll_info *npinfo;
885	unsigned long flags;
886
887	npinfo = np->dev->npinfo;
888	if (!npinfo)
889		return;
890
891	if (!list_empty(&npinfo->rx_np)) {
892		spin_lock_irqsave(&npinfo->rx_lock, flags);
893		list_del(&np->rx);
894		if (list_empty(&npinfo->rx_np))
895			npinfo->rx_flags &= ~NETPOLL_RX_ENABLED;
896		spin_unlock_irqrestore(&npinfo->rx_lock, flags);
897	}
898
899	if (atomic_dec_and_test(&npinfo->refcnt)) {
900		const struct net_device_ops *ops;
901
902		ops = np->dev->netdev_ops;
903		if (ops->ndo_netpoll_cleanup)
904			ops->ndo_netpoll_cleanup(np->dev);
905
906		rcu_assign_pointer(np->dev->npinfo, NULL);
 
 
 
907
908		/* avoid racing with NAPI reading npinfo */
909		synchronize_rcu_bh();
910
911		skb_queue_purge(&npinfo->arp_tx);
912		skb_queue_purge(&npinfo->txq);
913		cancel_delayed_work_sync(&npinfo->tx_work);
914
915		/* clean after last, unfinished work */
916		__skb_queue_purge(&npinfo->txq);
917		kfree(npinfo);
918	}
919}
920EXPORT_SYMBOL_GPL(__netpoll_cleanup);
921
922void netpoll_cleanup(struct netpoll *np)
923{
924	if (!np->dev)
925		return;
926
927	rtnl_lock();
 
928	__netpoll_cleanup(np);
929	rtnl_unlock();
930
931	dev_put(np->dev);
932	np->dev = NULL;
933}
934EXPORT_SYMBOL(netpoll_cleanup);
935
936int netpoll_trap(void)
937{
938	return atomic_read(&trapped);
 
 
939}
940EXPORT_SYMBOL(netpoll_trap);
941
942void netpoll_set_trap(int trap)
943{
944	if (trap)
945		atomic_inc(&trapped);
946	else
947		atomic_dec(&trapped);
 
 
948}
949EXPORT_SYMBOL(netpoll_set_trap);
v6.13.7
  1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 * Common framework for low-level network console, dump, and debugger code
  4 *
  5 * Sep 8 2003  Matt Mackall <mpm@selenic.com>
  6 *
  7 * based on the netconsole code from:
  8 *
  9 * Copyright (C) 2001  Ingo Molnar <mingo@redhat.com>
 10 * Copyright (C) 2002  Red Hat, Inc.
 11 */
 12
 13#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 14
 15#include <linux/moduleparam.h>
 16#include <linux/kernel.h>
 17#include <linux/netdevice.h>
 18#include <linux/etherdevice.h>
 19#include <linux/string.h>
 20#include <linux/if_arp.h>
 21#include <linux/inetdevice.h>
 22#include <linux/inet.h>
 23#include <linux/interrupt.h>
 24#include <linux/netpoll.h>
 25#include <linux/sched.h>
 26#include <linux/delay.h>
 27#include <linux/rcupdate.h>
 28#include <linux/workqueue.h>
 29#include <linux/slab.h>
 30#include <linux/export.h>
 31#include <linux/if_vlan.h>
 32#include <net/tcp.h>
 33#include <net/udp.h>
 34#include <net/addrconf.h>
 35#include <net/ndisc.h>
 36#include <net/ip6_checksum.h>
 37#include <linux/unaligned.h>
 38#include <trace/events/napi.h>
 39#include <linux/kconfig.h>
 40
 41/*
 42 * We maintain a small pool of fully-sized skbs, to make sure the
 43 * message gets out even in extreme OOM situations.
 44 */
 45
 46#define MAX_UDP_CHUNK 1460
 47#define MAX_SKBS 32
 
 
 
 
 
 48#define USEC_PER_POLL	50
 
 
 49
 50#define MAX_SKB_SIZE							\
 51	(sizeof(struct ethhdr) +					\
 52	 sizeof(struct iphdr) +						\
 53	 sizeof(struct udphdr) +					\
 54	 MAX_UDP_CHUNK)
 55
 56static void zap_completion_queue(void);
 
 57
 58static unsigned int carrier_timeout = 4;
 59module_param(carrier_timeout, uint, 0644);
 60
 61#define np_info(np, fmt, ...)				\
 62	pr_info("%s: " fmt, np->name, ##__VA_ARGS__)
 63#define np_err(np, fmt, ...)				\
 64	pr_err("%s: " fmt, np->name, ##__VA_ARGS__)
 65#define np_notice(np, fmt, ...)				\
 66	pr_notice("%s: " fmt, np->name, ##__VA_ARGS__)
 67
 68static netdev_tx_t netpoll_start_xmit(struct sk_buff *skb,
 69				      struct net_device *dev,
 70				      struct netdev_queue *txq)
 71{
 72	netdev_tx_t status = NETDEV_TX_OK;
 73	netdev_features_t features;
 74
 75	features = netif_skb_features(skb);
 76
 77	if (skb_vlan_tag_present(skb) &&
 78	    !vlan_hw_offload_capable(features, skb->vlan_proto)) {
 79		skb = __vlan_hwaccel_push_inside(skb);
 80		if (unlikely(!skb)) {
 81			/* This is actually a packet drop, but we
 82			 * don't want the code that calls this
 83			 * function to try and operate on a NULL skb.
 84			 */
 85			goto out;
 86		}
 87	}
 88
 89	status = netdev_start_xmit(skb, dev, txq, false);
 90
 91out:
 92	return status;
 93}
 94
 95static void queue_process(struct work_struct *work)
 96{
 97	struct netpoll_info *npinfo =
 98		container_of(work, struct netpoll_info, tx_work.work);
 99	struct sk_buff *skb;
100	unsigned long flags;
101
102	while ((skb = skb_dequeue(&npinfo->txq))) {
103		struct net_device *dev = skb->dev;
 
104		struct netdev_queue *txq;
105		unsigned int q_index;
106
107		if (!netif_device_present(dev) || !netif_running(dev)) {
108			kfree_skb(skb);
109			continue;
110		}
111
 
 
112		local_irq_save(flags);
113		/* check if skb->queue_mapping is still valid */
114		q_index = skb_get_queue_mapping(skb);
115		if (unlikely(q_index >= dev->real_num_tx_queues)) {
116			q_index = q_index % dev->real_num_tx_queues;
117			skb_set_queue_mapping(skb, q_index);
118		}
119		txq = netdev_get_tx_queue(dev, q_index);
120		HARD_TX_LOCK(dev, txq, smp_processor_id());
121		if (netif_xmit_frozen_or_stopped(txq) ||
122		    !dev_xmit_complete(netpoll_start_xmit(skb, dev, txq))) {
123			skb_queue_head(&npinfo->txq, skb);
124			HARD_TX_UNLOCK(dev, txq);
125			local_irq_restore(flags);
126
127			schedule_delayed_work(&npinfo->tx_work, HZ/10);
128			return;
129		}
130		HARD_TX_UNLOCK(dev, txq);
131		local_irq_restore(flags);
132	}
133}
134
135static int netif_local_xmit_active(struct net_device *dev)
 
136{
137	int i;
 
 
 
138
139	for (i = 0; i < dev->num_tx_queues; i++) {
140		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
141
142		if (READ_ONCE(txq->xmit_lock_owner) == smp_processor_id())
143			return 1;
144	}
 
 
145
146	return 0;
147}
148
149static void poll_one_napi(struct napi_struct *napi)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
150{
151	int work;
152
153	/* If we set this bit but see that it has already been set,
154	 * that indicates that napi has been disabled and we need
155	 * to abort this operation
156	 */
157	if (test_and_set_bit(NAPI_STATE_NPSVC, &napi->state))
158		return;
 
 
 
 
159
160	/* We explicitly pass the polling call a budget of 0 to
161	 * indicate that we are clearing the Tx path only.
162	 */
163	work = napi->poll(napi, 0);
164	WARN_ONCE(work, "%pS exceeded budget in poll\n", napi->poll);
165	trace_napi_poll(napi, work, 0);
166
167	clear_bit(NAPI_STATE_NPSVC, &napi->state);
 
 
 
 
168}
169
170static void poll_napi(struct net_device *dev)
171{
172	struct napi_struct *napi;
173	int cpu = smp_processor_id();
 
 
 
 
 
 
174
175	list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) {
176		if (cmpxchg(&napi->poll_owner, -1, cpu) == -1) {
177			poll_one_napi(napi);
178			smp_store_release(&napi->poll_owner, -1);
179		}
180	}
181}
182
183void netpoll_poll_dev(struct net_device *dev)
 
 
 
 
 
 
 
 
 
 
184{
185	struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
186	const struct net_device_ops *ops;
187
188	/* Don't do any rx activity if the dev_lock mutex is held
189	 * the dev_open/close paths use this to block netpoll activity
190	 * while changing device state
191	 */
192	if (!ni || down_trylock(&ni->dev_lock))
193		return;
194
195	/* Some drivers will take the same locks in poll and xmit,
196	 * we can't poll if local CPU is already in xmit.
197	 */
198	if (!netif_running(dev) || netif_local_xmit_active(dev)) {
199		up(&ni->dev_lock);
200		return;
201	}
202
203	ops = dev->netdev_ops;
204	if (ops->ndo_poll_controller)
205		ops->ndo_poll_controller(dev);
206
207	poll_napi(dev);
208
209	up(&ni->dev_lock);
 
 
 
 
 
 
 
 
 
 
 
210
211	zap_completion_queue();
212}
213EXPORT_SYMBOL(netpoll_poll_dev);
214
215void netpoll_poll_disable(struct net_device *dev)
216{
217	struct netpoll_info *ni;
218
219	might_sleep();
220	ni = rtnl_dereference(dev->npinfo);
221	if (ni)
222		down(&ni->dev_lock);
223}
224
225void netpoll_poll_enable(struct net_device *dev)
226{
227	struct netpoll_info *ni;
228
229	ni = rtnl_dereference(dev->npinfo);
230	if (ni)
231		up(&ni->dev_lock);
232}
233
234static void refill_skbs(struct netpoll *np)
235{
236	struct sk_buff_head *skb_pool;
237	struct sk_buff *skb;
238	unsigned long flags;
239
240	skb_pool = &np->skb_pool;
241
242	spin_lock_irqsave(&skb_pool->lock, flags);
243	while (skb_pool->qlen < MAX_SKBS) {
244		skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
245		if (!skb)
246			break;
247
248		__skb_queue_tail(skb_pool, skb);
249	}
250	spin_unlock_irqrestore(&skb_pool->lock, flags);
251}
252
253static void zap_completion_queue(void)
254{
255	unsigned long flags;
256	struct softnet_data *sd = &get_cpu_var(softnet_data);
257
258	if (sd->completion_queue) {
259		struct sk_buff *clist;
260
261		local_irq_save(flags);
262		clist = sd->completion_queue;
263		sd->completion_queue = NULL;
264		local_irq_restore(flags);
265
266		while (clist != NULL) {
267			struct sk_buff *skb = clist;
268			clist = clist->next;
269			if (!skb_irq_freeable(skb)) {
270				refcount_set(&skb->users, 1);
271				dev_kfree_skb_any(skb); /* put this one back */
272			} else {
273				__kfree_skb(skb);
274			}
275		}
276	}
277
278	put_cpu_var(softnet_data);
279}
280
281static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
282{
283	int count = 0;
284	struct sk_buff *skb;
285
286	zap_completion_queue();
287	refill_skbs(np);
288repeat:
289
290	skb = alloc_skb(len, GFP_ATOMIC);
291	if (!skb)
292		skb = skb_dequeue(&np->skb_pool);
293
294	if (!skb) {
295		if (++count < 10) {
296			netpoll_poll_dev(np->dev);
297			goto repeat;
298		}
299		return NULL;
300	}
301
302	refcount_set(&skb->users, 1);
303	skb_reserve(skb, reserve);
304	return skb;
305}
306
307static int netpoll_owner_active(struct net_device *dev)
308{
309	struct napi_struct *napi;
310
311	list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) {
312		if (READ_ONCE(napi->poll_owner) == smp_processor_id())
313			return 1;
314	}
315	return 0;
316}
317
318/* call with IRQ disabled */
319static netdev_tx_t __netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
320{
321	netdev_tx_t status = NETDEV_TX_BUSY;
322	struct net_device *dev;
323	unsigned long tries;
 
324	/* It is up to the caller to keep npinfo alive. */
325	struct netpoll_info *npinfo;
326
327	lockdep_assert_irqs_disabled();
328
329	dev = np->dev;
330	npinfo = rcu_dereference_bh(dev->npinfo);
331
332	if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
333		dev_kfree_skb_irq(skb);
334		return NET_XMIT_DROP;
335	}
336
337	/* don't get messages out of order, and no recursion */
338	if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
339		struct netdev_queue *txq;
 
340
341		txq = netdev_core_pick_tx(dev, skb, NULL);
342
 
343		/* try until next clock tick */
344		for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
345		     tries > 0; --tries) {
346			if (HARD_TX_TRYLOCK(dev, txq)) {
347				if (!netif_xmit_stopped(txq))
348					status = netpoll_start_xmit(skb, dev, txq);
 
 
 
 
349
350				HARD_TX_UNLOCK(dev, txq);
351
352				if (dev_xmit_complete(status))
353					break;
354
355			}
356
357			/* tickle device maybe there is some cleanup */
358			netpoll_poll_dev(np->dev);
359
360			udelay(USEC_PER_POLL);
361		}
362
363		WARN_ONCE(!irqs_disabled(),
364			"netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pS)\n",
365			dev->name, dev->netdev_ops->ndo_start_xmit);
366
 
367	}
368
369	if (!dev_xmit_complete(status)) {
370		skb_queue_tail(&npinfo->txq, skb);
371		schedule_delayed_work(&npinfo->tx_work,0);
372	}
373	return NETDEV_TX_OK;
374}
375
376netdev_tx_t netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
377{
378	unsigned long flags;
379	netdev_tx_t ret;
380
381	if (unlikely(!np)) {
382		dev_kfree_skb_irq(skb);
383		ret = NET_XMIT_DROP;
384	} else {
385		local_irq_save(flags);
386		ret = __netpoll_send_skb(np, skb);
387		local_irq_restore(flags);
388	}
389	return ret;
390}
391EXPORT_SYMBOL(netpoll_send_skb);
392
393void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
394{
395	int total_len, ip_len, udp_len;
396	struct sk_buff *skb;
397	struct udphdr *udph;
398	struct iphdr *iph;
399	struct ethhdr *eth;
400	static atomic_t ip_ident;
401	struct ipv6hdr *ip6h;
402
403	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
404		WARN_ON_ONCE(!irqs_disabled());
405
406	udp_len = len + sizeof(*udph);
407	if (np->ipv6)
408		ip_len = udp_len + sizeof(*ip6h);
409	else
410		ip_len = udp_len + sizeof(*iph);
411
412	total_len = ip_len + LL_RESERVED_SPACE(np->dev);
413
414	skb = find_skb(np, total_len + np->dev->needed_tailroom,
415		       total_len - len);
416	if (!skb)
417		return;
418
419	skb_copy_to_linear_data(skb, msg, len);
420	skb_put(skb, len);
421
422	skb_push(skb, sizeof(*udph));
423	skb_reset_transport_header(skb);
424	udph = udp_hdr(skb);
425	udph->source = htons(np->local_port);
426	udph->dest = htons(np->remote_port);
427	udph->len = htons(udp_len);
428
429	if (np->ipv6) {
430		udph->check = 0;
431		udph->check = csum_ipv6_magic(&np->local_ip.in6,
432					      &np->remote_ip.in6,
433					      udp_len, IPPROTO_UDP,
434					      csum_partial(udph, udp_len, 0));
435		if (udph->check == 0)
436			udph->check = CSUM_MANGLED_0;
437
438		skb_push(skb, sizeof(*ip6h));
439		skb_reset_network_header(skb);
440		ip6h = ipv6_hdr(skb);
441
442		/* ip6h->version = 6; ip6h->priority = 0; */
443		*(unsigned char *)ip6h = 0x60;
444		ip6h->flow_lbl[0] = 0;
445		ip6h->flow_lbl[1] = 0;
446		ip6h->flow_lbl[2] = 0;
447
448		ip6h->payload_len = htons(sizeof(struct udphdr) + len);
449		ip6h->nexthdr = IPPROTO_UDP;
450		ip6h->hop_limit = 32;
451		ip6h->saddr = np->local_ip.in6;
452		ip6h->daddr = np->remote_ip.in6;
453
454		eth = skb_push(skb, ETH_HLEN);
455		skb_reset_mac_header(skb);
456		skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
457	} else {
458		udph->check = 0;
459		udph->check = csum_tcpudp_magic(np->local_ip.ip,
460						np->remote_ip.ip,
461						udp_len, IPPROTO_UDP,
462						csum_partial(udph, udp_len, 0));
463		if (udph->check == 0)
464			udph->check = CSUM_MANGLED_0;
465
466		skb_push(skb, sizeof(*iph));
467		skb_reset_network_header(skb);
468		iph = ip_hdr(skb);
469
470		/* iph->version = 4; iph->ihl = 5; */
471		*(unsigned char *)iph = 0x45;
472		iph->tos      = 0;
473		put_unaligned(htons(ip_len), &(iph->tot_len));
474		iph->id       = htons(atomic_inc_return(&ip_ident));
475		iph->frag_off = 0;
476		iph->ttl      = 64;
477		iph->protocol = IPPROTO_UDP;
478		iph->check    = 0;
479		put_unaligned(np->local_ip.ip, &(iph->saddr));
480		put_unaligned(np->remote_ip.ip, &(iph->daddr));
481		iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
482
483		eth = skb_push(skb, ETH_HLEN);
484		skb_reset_mac_header(skb);
485		skb->protocol = eth->h_proto = htons(ETH_P_IP);
486	}
487
488	ether_addr_copy(eth->h_source, np->dev->dev_addr);
489	ether_addr_copy(eth->h_dest, np->remote_mac);
490
491	skb->dev = np->dev;
492
493	netpoll_send_skb(np, skb);
494}
495EXPORT_SYMBOL(netpoll_send_udp);
496
497void netpoll_print_options(struct netpoll *np)
498{
499	np_info(np, "local port %d\n", np->local_port);
500	if (np->ipv6)
501		np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
502	else
503		np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
504	np_info(np, "interface '%s'\n", np->dev_name);
505	np_info(np, "remote port %d\n", np->remote_port);
506	if (np->ipv6)
507		np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
508	else
509		np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
510	np_info(np, "remote ethernet address %pM\n", np->remote_mac);
511}
512EXPORT_SYMBOL(netpoll_print_options);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
513
514static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
515{
516	const char *end;
517
518	if (!strchr(str, ':') &&
519	    in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
520		if (!*end)
521			return 0;
522	}
523	if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
524#if IS_ENABLED(CONFIG_IPV6)
525		if (!*end)
526			return 1;
527#else
528		return -1;
529#endif
530	}
531	return -1;
 
532}
533
534static void skb_pool_flush(struct netpoll *np)
535{
536	struct sk_buff_head *skb_pool;
537
538	skb_pool = &np->skb_pool;
539	skb_queue_purge_reason(skb_pool, SKB_CONSUMED);
 
 
 
 
 
 
 
 
540}
 
541
542int netpoll_parse_options(struct netpoll *np, char *opt)
543{
544	char *cur=opt, *delim;
545	int ipv6;
546	bool ipversion_set = false;
547
548	if (*cur != '@') {
549		if ((delim = strchr(cur, '@')) == NULL)
550			goto parse_failed;
551		*delim = 0;
552		if (kstrtou16(cur, 10, &np->local_port))
553			goto parse_failed;
554		cur = delim;
555	}
556	cur++;
557
558	if (*cur != '/') {
559		ipversion_set = true;
560		if ((delim = strchr(cur, '/')) == NULL)
561			goto parse_failed;
562		*delim = 0;
563		ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
564		if (ipv6 < 0)
565			goto parse_failed;
566		else
567			np->ipv6 = (bool)ipv6;
568		cur = delim;
569	}
570	cur++;
571
572	if (*cur != ',') {
573		/* parse out dev name */
574		if ((delim = strchr(cur, ',')) == NULL)
575			goto parse_failed;
576		*delim = 0;
577		strscpy(np->dev_name, cur, sizeof(np->dev_name));
578		cur = delim;
579	}
580	cur++;
581
582	if (*cur != '@') {
583		/* dst port */
584		if ((delim = strchr(cur, '@')) == NULL)
585			goto parse_failed;
586		*delim = 0;
587		if (*cur == ' ' || *cur == '\t')
588			np_info(np, "warning: whitespace is not allowed\n");
589		if (kstrtou16(cur, 10, &np->remote_port))
590			goto parse_failed;
591		cur = delim;
592	}
593	cur++;
594
595	/* dst ip */
596	if ((delim = strchr(cur, '/')) == NULL)
597		goto parse_failed;
598	*delim = 0;
599	ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
600	if (ipv6 < 0)
601		goto parse_failed;
602	else if (ipversion_set && np->ipv6 != (bool)ipv6)
603		goto parse_failed;
604	else
605		np->ipv6 = (bool)ipv6;
606	cur = delim + 1;
607
608	if (*cur != 0) {
609		/* MAC address */
610		if (!mac_pton(cur, np->remote_mac))
611			goto parse_failed;
612	}
613
614	netpoll_print_options(np);
615
616	return 0;
617
618 parse_failed:
619	np_info(np, "couldn't parse config at '%s'!\n", cur);
 
620	return -1;
621}
622EXPORT_SYMBOL(netpoll_parse_options);
623
624int __netpoll_setup(struct netpoll *np, struct net_device *ndev)
625{
 
626	struct netpoll_info *npinfo;
627	const struct net_device_ops *ops;
 
628	int err;
629
630	skb_queue_head_init(&np->skb_pool);
631
632	if (ndev->priv_flags & IFF_DISABLE_NETPOLL) {
633		np_err(np, "%s doesn't support polling, aborting\n",
634		       ndev->name);
635		err = -ENOTSUPP;
636		goto out;
637	}
638
639	if (!rcu_access_pointer(ndev->npinfo)) {
640		npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
641		if (!npinfo) {
642			err = -ENOMEM;
643			goto out;
644		}
645
646		sema_init(&npinfo->dev_lock, 1);
 
 
 
 
647		skb_queue_head_init(&npinfo->txq);
648		INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
649
650		refcount_set(&npinfo->refcnt, 1);
651
652		ops = ndev->netdev_ops;
653		if (ops->ndo_netpoll_setup) {
654			err = ops->ndo_netpoll_setup(ndev);
655			if (err)
656				goto free_npinfo;
657		}
658	} else {
659		npinfo = rtnl_dereference(ndev->npinfo);
660		refcount_inc(&npinfo->refcnt);
661	}
662
663	np->dev = ndev;
664	strscpy(np->dev_name, ndev->name, IFNAMSIZ);
665	npinfo->netpoll = np;
666
667	/* fill up the skb queue */
668	refill_skbs(np);
 
 
 
 
669
670	/* last thing to do is link it to the net device structure */
671	rcu_assign_pointer(ndev->npinfo, npinfo);
672
673	return 0;
674
675free_npinfo:
676	kfree(npinfo);
677out:
678	return err;
679}
680EXPORT_SYMBOL_GPL(__netpoll_setup);
681
682int netpoll_setup(struct netpoll *np)
683{
684	struct net_device *ndev = NULL;
685	bool ip_overwritten = false;
686	struct in_device *in_dev;
687	int err;
688
689	rtnl_lock();
690	if (np->dev_name[0]) {
691		struct net *net = current->nsproxy->net_ns;
692		ndev = __dev_get_by_name(net, np->dev_name);
693	}
694	if (!ndev) {
695		np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
696		err = -ENODEV;
697		goto unlock;
698	}
699	netdev_hold(ndev, &np->dev_tracker, GFP_KERNEL);
700
701	if (netdev_master_upper_dev_get(ndev)) {
702		np_err(np, "%s is a slave device, aborting\n", np->dev_name);
 
703		err = -EBUSY;
704		goto put;
705	}
706
707	if (!netif_running(ndev)) {
708		unsigned long atmost;
709
710		np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
 
711
712		err = dev_open(ndev, NULL);
 
 
713
714		if (err) {
715			np_err(np, "failed to open %s\n", ndev->name);
 
716			goto put;
717		}
718
719		rtnl_unlock();
720		atmost = jiffies + carrier_timeout * HZ;
721		while (!netif_carrier_ok(ndev)) {
722			if (time_after(jiffies, atmost)) {
723				np_notice(np, "timeout waiting for carrier\n");
 
 
724				break;
725			}
726			msleep(1);
727		}
728
729		rtnl_lock();
 
 
 
 
 
 
 
 
 
 
730	}
731
732	if (!np->local_ip.ip) {
733		if (!np->ipv6) {
734			const struct in_ifaddr *ifa;
735
736			in_dev = __in_dev_get_rtnl(ndev);
737			if (!in_dev)
738				goto put_noaddr;
739
740			ifa = rtnl_dereference(in_dev->ifa_list);
741			if (!ifa) {
742put_noaddr:
743				np_err(np, "no IP address for %s, aborting\n",
744				       np->dev_name);
745				err = -EDESTADDRREQ;
746				goto put;
747			}
748
749			np->local_ip.ip = ifa->ifa_local;
750			ip_overwritten = true;
751			np_info(np, "local IP %pI4\n", &np->local_ip.ip);
752		} else {
753#if IS_ENABLED(CONFIG_IPV6)
754			struct inet6_dev *idev;
755
756			err = -EDESTADDRREQ;
757			idev = __in6_dev_get(ndev);
758			if (idev) {
759				struct inet6_ifaddr *ifp;
760
761				read_lock_bh(&idev->lock);
762				list_for_each_entry(ifp, &idev->addr_list, if_list) {
763					if (!!(ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL) !=
764					    !!(ipv6_addr_type(&np->remote_ip.in6) & IPV6_ADDR_LINKLOCAL))
765						continue;
766					np->local_ip.in6 = ifp->addr;
767					ip_overwritten = true;
768					err = 0;
769					break;
770				}
771				read_unlock_bh(&idev->lock);
772			}
773			if (err) {
774				np_err(np, "no IPv6 address for %s, aborting\n",
775				       np->dev_name);
776				goto put;
777			} else
778				np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
779#else
780			np_err(np, "IPv6 is not supported %s, aborting\n",
781			       np->dev_name);
782			err = -EINVAL;
783			goto put;
784#endif
785		}
 
 
 
 
786	}
787
788	err = __netpoll_setup(np, ndev);
 
 
 
 
 
 
 
 
789	if (err)
790		goto flush;
791	rtnl_unlock();
792	return 0;
793
794flush:
795	skb_pool_flush(np);
796put:
797	DEBUG_NET_WARN_ON_ONCE(np->dev);
798	if (ip_overwritten)
799		memset(&np->local_ip, 0, sizeof(np->local_ip));
800	netdev_put(ndev, &np->dev_tracker);
801unlock:
802	rtnl_unlock();
803	return err;
804}
805EXPORT_SYMBOL(netpoll_setup);
806
807static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
808{
809	struct netpoll_info *npinfo =
810			container_of(rcu_head, struct netpoll_info, rcu);
811
812	skb_queue_purge(&npinfo->txq);
813
814	/* we can't call cancel_delayed_work_sync here, as we are in softirq */
815	cancel_delayed_work(&npinfo->tx_work);
816
817	/* clean after last, unfinished work */
818	__skb_queue_purge(&npinfo->txq);
819	/* now cancel it again */
820	cancel_delayed_work(&npinfo->tx_work);
821	kfree(npinfo);
822}
 
823
824void __netpoll_cleanup(struct netpoll *np)
825{
826	struct netpoll_info *npinfo;
 
827
828	npinfo = rtnl_dereference(np->dev->npinfo);
829	if (!npinfo)
830		return;
831
832	if (refcount_dec_and_test(&npinfo->refcnt)) {
 
 
 
 
 
 
 
 
833		const struct net_device_ops *ops;
834
835		ops = np->dev->netdev_ops;
836		if (ops->ndo_netpoll_cleanup)
837			ops->ndo_netpoll_cleanup(np->dev);
838
839		RCU_INIT_POINTER(np->dev->npinfo, NULL);
840		call_rcu(&npinfo->rcu, rcu_cleanup_netpoll_info);
841	} else
842		RCU_INIT_POINTER(np->dev->npinfo, NULL);
843
844	skb_pool_flush(np);
 
 
 
 
 
 
 
 
 
 
845}
846EXPORT_SYMBOL_GPL(__netpoll_cleanup);
847
848void __netpoll_free(struct netpoll *np)
849{
850	ASSERT_RTNL();
 
851
852	/* Wait for transmitting packets to finish before freeing. */
853	synchronize_rcu();
854	__netpoll_cleanup(np);
855	kfree(np);
 
 
 
856}
857EXPORT_SYMBOL_GPL(__netpoll_free);
858
859void do_netpoll_cleanup(struct netpoll *np)
860{
861	__netpoll_cleanup(np);
862	netdev_put(np->dev, &np->dev_tracker);
863	np->dev = NULL;
864}
865EXPORT_SYMBOL(do_netpoll_cleanup);
866
867void netpoll_cleanup(struct netpoll *np)
868{
869	rtnl_lock();
870	if (!np->dev)
871		goto out;
872	do_netpoll_cleanup(np);
873out:
874	rtnl_unlock();
875}
876EXPORT_SYMBOL(netpoll_cleanup);