xref: /linux/net/core/netpoll.c (revision d0309c054362a235077327b46f727bc48878a3bc)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Common framework for low-level network console, dump, and debugger code
4  *
5  * Sep 8 2003  Matt Mackall <mpm@selenic.com>
6  *
7  * based on the netconsole code from:
8  *
9  * Copyright (C) 2001  Ingo Molnar <mingo@redhat.com>
10  * Copyright (C) 2002  Red Hat, Inc.
11  */
12 
13 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14 
15 #include <linux/moduleparam.h>
16 #include <linux/kernel.h>
17 #include <linux/netdevice.h>
18 #include <linux/etherdevice.h>
19 #include <linux/string.h>
20 #include <linux/if_arp.h>
21 #include <linux/inetdevice.h>
22 #include <linux/inet.h>
23 #include <linux/interrupt.h>
24 #include <linux/netpoll.h>
25 #include <linux/sched.h>
26 #include <linux/delay.h>
27 #include <linux/rcupdate.h>
28 #include <linux/workqueue.h>
29 #include <linux/slab.h>
30 #include <linux/export.h>
31 #include <linux/if_vlan.h>
32 #include <net/tcp.h>
33 #include <net/udp.h>
34 #include <net/addrconf.h>
35 #include <net/ndisc.h>
36 #include <net/ip6_checksum.h>
37 #include <linux/unaligned.h>
38 #include <trace/events/napi.h>
39 #include <linux/kconfig.h>
40 
41 /*
42  * We maintain a small pool of fully-sized skbs, to make sure the
43  * message gets out even in extreme OOM situations.
44  */
45 
46 #define MAX_UDP_CHUNK 1460
47 #define MAX_SKBS 32
48 #define USEC_PER_POLL	50
49 
50 #define MAX_SKB_SIZE							\
51 	(sizeof(struct ethhdr) +					\
52 	 sizeof(struct iphdr) +						\
53 	 sizeof(struct udphdr) +					\
54 	 MAX_UDP_CHUNK)
55 
56 static void zap_completion_queue(void);
57 
58 static unsigned int carrier_timeout = 4;
59 module_param(carrier_timeout, uint, 0644);
60 
netpoll_start_xmit(struct sk_buff * skb,struct net_device * dev,struct netdev_queue * txq)61 static netdev_tx_t netpoll_start_xmit(struct sk_buff *skb,
62 				      struct net_device *dev,
63 				      struct netdev_queue *txq)
64 {
65 	netdev_tx_t status = NETDEV_TX_OK;
66 	netdev_features_t features;
67 
68 	features = netif_skb_features(skb);
69 
70 	if (skb_vlan_tag_present(skb) &&
71 	    !vlan_hw_offload_capable(features, skb->vlan_proto)) {
72 		skb = __vlan_hwaccel_push_inside(skb);
73 		if (unlikely(!skb)) {
74 			/* This is actually a packet drop, but we
75 			 * don't want the code that calls this
76 			 * function to try and operate on a NULL skb.
77 			 */
78 			goto out;
79 		}
80 	}
81 
82 	status = netdev_start_xmit(skb, dev, txq, false);
83 
84 out:
85 	return status;
86 }
87 
queue_process(struct work_struct * work)88 static void queue_process(struct work_struct *work)
89 {
90 	struct netpoll_info *npinfo =
91 		container_of(work, struct netpoll_info, tx_work.work);
92 	struct sk_buff *skb;
93 	unsigned long flags;
94 
95 	while ((skb = skb_dequeue(&npinfo->txq))) {
96 		struct net_device *dev = skb->dev;
97 		struct netdev_queue *txq;
98 		unsigned int q_index;
99 
100 		if (!netif_device_present(dev) || !netif_running(dev)) {
101 			kfree_skb(skb);
102 			continue;
103 		}
104 
105 		local_irq_save(flags);
106 		/* check if skb->queue_mapping is still valid */
107 		q_index = skb_get_queue_mapping(skb);
108 		if (unlikely(q_index >= dev->real_num_tx_queues)) {
109 			q_index = q_index % dev->real_num_tx_queues;
110 			skb_set_queue_mapping(skb, q_index);
111 		}
112 		txq = netdev_get_tx_queue(dev, q_index);
113 		HARD_TX_LOCK(dev, txq, smp_processor_id());
114 		if (netif_xmit_frozen_or_stopped(txq) ||
115 		    !dev_xmit_complete(netpoll_start_xmit(skb, dev, txq))) {
116 			skb_queue_head(&npinfo->txq, skb);
117 			HARD_TX_UNLOCK(dev, txq);
118 			local_irq_restore(flags);
119 
120 			schedule_delayed_work(&npinfo->tx_work, HZ/10);
121 			return;
122 		}
123 		HARD_TX_UNLOCK(dev, txq);
124 		local_irq_restore(flags);
125 	}
126 }
127 
netif_local_xmit_active(struct net_device * dev)128 static int netif_local_xmit_active(struct net_device *dev)
129 {
130 	int i;
131 
132 	for (i = 0; i < dev->num_tx_queues; i++) {
133 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
134 
135 		if (READ_ONCE(txq->xmit_lock_owner) == smp_processor_id())
136 			return 1;
137 	}
138 
139 	return 0;
140 }
141 
poll_one_napi(struct napi_struct * napi)142 static void poll_one_napi(struct napi_struct *napi)
143 {
144 	int work;
145 
146 	/* If we set this bit but see that it has already been set,
147 	 * that indicates that napi has been disabled and we need
148 	 * to abort this operation
149 	 */
150 	if (test_and_set_bit(NAPI_STATE_NPSVC, &napi->state))
151 		return;
152 
153 	/* We explicitly pass the polling call a budget of 0 to
154 	 * indicate that we are clearing the Tx path only.
155 	 */
156 	work = napi->poll(napi, 0);
157 	WARN_ONCE(work, "%pS exceeded budget in poll\n", napi->poll);
158 	trace_napi_poll(napi, work, 0);
159 
160 	clear_bit(NAPI_STATE_NPSVC, &napi->state);
161 }
162 
poll_napi(struct net_device * dev)163 static void poll_napi(struct net_device *dev)
164 {
165 	struct napi_struct *napi;
166 	int cpu = smp_processor_id();
167 
168 	list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) {
169 		if (cmpxchg(&napi->poll_owner, -1, cpu) == -1) {
170 			poll_one_napi(napi);
171 			smp_store_release(&napi->poll_owner, -1);
172 		}
173 	}
174 }
175 
netpoll_poll_dev(struct net_device * dev)176 void netpoll_poll_dev(struct net_device *dev)
177 {
178 	struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
179 	const struct net_device_ops *ops;
180 
181 	/* Don't do any rx activity if the dev_lock mutex is held
182 	 * the dev_open/close paths use this to block netpoll activity
183 	 * while changing device state
184 	 */
185 	if (!ni || down_trylock(&ni->dev_lock))
186 		return;
187 
188 	/* Some drivers will take the same locks in poll and xmit,
189 	 * we can't poll if local CPU is already in xmit.
190 	 */
191 	if (!netif_running(dev) || netif_local_xmit_active(dev)) {
192 		up(&ni->dev_lock);
193 		return;
194 	}
195 
196 	ops = dev->netdev_ops;
197 	if (ops->ndo_poll_controller)
198 		ops->ndo_poll_controller(dev);
199 
200 	poll_napi(dev);
201 
202 	up(&ni->dev_lock);
203 
204 	zap_completion_queue();
205 }
206 EXPORT_SYMBOL(netpoll_poll_dev);
207 
netpoll_poll_disable(struct net_device * dev)208 void netpoll_poll_disable(struct net_device *dev)
209 {
210 	struct netpoll_info *ni;
211 
212 	might_sleep();
213 	ni = rtnl_dereference(dev->npinfo);
214 	if (ni)
215 		down(&ni->dev_lock);
216 }
217 
netpoll_poll_enable(struct net_device * dev)218 void netpoll_poll_enable(struct net_device *dev)
219 {
220 	struct netpoll_info *ni;
221 
222 	ni = rtnl_dereference(dev->npinfo);
223 	if (ni)
224 		up(&ni->dev_lock);
225 }
226 
refill_skbs(struct netpoll * np)227 static void refill_skbs(struct netpoll *np)
228 {
229 	struct sk_buff_head *skb_pool;
230 	struct sk_buff *skb;
231 
232 	skb_pool = &np->skb_pool;
233 
234 	while (READ_ONCE(skb_pool->qlen) < MAX_SKBS) {
235 		skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
236 		if (!skb)
237 			break;
238 
239 		skb_queue_tail(skb_pool, skb);
240 	}
241 }
242 
zap_completion_queue(void)243 static void zap_completion_queue(void)
244 {
245 	unsigned long flags;
246 	struct softnet_data *sd = &get_cpu_var(softnet_data);
247 
248 	if (sd->completion_queue) {
249 		struct sk_buff *clist;
250 
251 		local_irq_save(flags);
252 		clist = sd->completion_queue;
253 		sd->completion_queue = NULL;
254 		local_irq_restore(flags);
255 
256 		while (clist != NULL) {
257 			struct sk_buff *skb = clist;
258 			clist = clist->next;
259 			if (!skb_irq_freeable(skb)) {
260 				refcount_set(&skb->users, 1);
261 				dev_kfree_skb_any(skb); /* put this one back */
262 			} else {
263 				__kfree_skb(skb);
264 			}
265 		}
266 	}
267 
268 	put_cpu_var(softnet_data);
269 }
270 
find_skb(struct netpoll * np,int len,int reserve)271 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
272 {
273 	int count = 0;
274 	struct sk_buff *skb;
275 
276 	zap_completion_queue();
277 repeat:
278 
279 	skb = alloc_skb(len, GFP_ATOMIC);
280 	if (!skb) {
281 		skb = skb_dequeue(&np->skb_pool);
282 		schedule_work(&np->refill_wq);
283 	}
284 
285 	if (!skb) {
286 		if (++count < 10) {
287 			netpoll_poll_dev(np->dev);
288 			goto repeat;
289 		}
290 		return NULL;
291 	}
292 
293 	refcount_set(&skb->users, 1);
294 	skb_reserve(skb, reserve);
295 	return skb;
296 }
297 
netpoll_owner_active(struct net_device * dev)298 static int netpoll_owner_active(struct net_device *dev)
299 {
300 	struct napi_struct *napi;
301 
302 	list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) {
303 		if (READ_ONCE(napi->poll_owner) == smp_processor_id())
304 			return 1;
305 	}
306 	return 0;
307 }
308 
309 /* call with IRQ disabled */
__netpoll_send_skb(struct netpoll * np,struct sk_buff * skb)310 static netdev_tx_t __netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
311 {
312 	netdev_tx_t status = NETDEV_TX_BUSY;
313 	netdev_tx_t ret = NET_XMIT_DROP;
314 	struct net_device *dev;
315 	unsigned long tries;
316 	/* It is up to the caller to keep npinfo alive. */
317 	struct netpoll_info *npinfo;
318 
319 	lockdep_assert_irqs_disabled();
320 
321 	dev = np->dev;
322 	rcu_read_lock();
323 	npinfo = rcu_dereference_bh(dev->npinfo);
324 
325 	if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
326 		dev_kfree_skb_irq(skb);
327 		goto out;
328 	}
329 
330 	/* don't get messages out of order, and no recursion */
331 	if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
332 		struct netdev_queue *txq;
333 
334 		txq = netdev_core_pick_tx(dev, skb, NULL);
335 
336 		/* try until next clock tick */
337 		for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
338 		     tries > 0; --tries) {
339 			if (HARD_TX_TRYLOCK(dev, txq)) {
340 				if (!netif_xmit_stopped(txq))
341 					status = netpoll_start_xmit(skb, dev, txq);
342 
343 				HARD_TX_UNLOCK(dev, txq);
344 
345 				if (dev_xmit_complete(status))
346 					break;
347 
348 			}
349 
350 			/* tickle device maybe there is some cleanup */
351 			netpoll_poll_dev(np->dev);
352 
353 			udelay(USEC_PER_POLL);
354 		}
355 
356 		WARN_ONCE(!irqs_disabled(),
357 			"netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pS)\n",
358 			dev->name, dev->netdev_ops->ndo_start_xmit);
359 
360 	}
361 
362 	if (!dev_xmit_complete(status)) {
363 		skb_queue_tail(&npinfo->txq, skb);
364 		schedule_delayed_work(&npinfo->tx_work,0);
365 	}
366 	ret = NETDEV_TX_OK;
367 out:
368 	rcu_read_unlock();
369 	return ret;
370 }
371 
netpoll_udp_checksum(struct netpoll * np,struct sk_buff * skb,int len)372 static void netpoll_udp_checksum(struct netpoll *np, struct sk_buff *skb,
373 				 int len)
374 {
375 	struct udphdr *udph;
376 	int udp_len;
377 
378 	udp_len = len + sizeof(struct udphdr);
379 	udph = udp_hdr(skb);
380 
381 	/* check needs to be set, since it will be consumed in csum_partial */
382 	udph->check = 0;
383 	if (np->ipv6)
384 		udph->check = csum_ipv6_magic(&np->local_ip.in6,
385 					      &np->remote_ip.in6,
386 					      udp_len, IPPROTO_UDP,
387 					      csum_partial(udph, udp_len, 0));
388 	else
389 		udph->check = csum_tcpudp_magic(np->local_ip.ip,
390 						np->remote_ip.ip,
391 						udp_len, IPPROTO_UDP,
392 						csum_partial(udph, udp_len, 0));
393 	if (udph->check == 0)
394 		udph->check = CSUM_MANGLED_0;
395 }
396 
netpoll_send_skb(struct netpoll * np,struct sk_buff * skb)397 netdev_tx_t netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
398 {
399 	unsigned long flags;
400 	netdev_tx_t ret;
401 
402 	if (unlikely(!np)) {
403 		dev_kfree_skb_irq(skb);
404 		ret = NET_XMIT_DROP;
405 	} else {
406 		local_irq_save(flags);
407 		ret = __netpoll_send_skb(np, skb);
408 		local_irq_restore(flags);
409 	}
410 	return ret;
411 }
412 EXPORT_SYMBOL(netpoll_send_skb);
413 
push_ipv6(struct netpoll * np,struct sk_buff * skb,int len)414 static void push_ipv6(struct netpoll *np, struct sk_buff *skb, int len)
415 {
416 	struct ipv6hdr *ip6h;
417 
418 	skb_push(skb, sizeof(struct ipv6hdr));
419 	skb_reset_network_header(skb);
420 	ip6h = ipv6_hdr(skb);
421 
422 	/* ip6h->version = 6; ip6h->priority = 0; */
423 	*(unsigned char *)ip6h = 0x60;
424 	ip6h->flow_lbl[0] = 0;
425 	ip6h->flow_lbl[1] = 0;
426 	ip6h->flow_lbl[2] = 0;
427 
428 	ip6h->payload_len = htons(sizeof(struct udphdr) + len);
429 	ip6h->nexthdr = IPPROTO_UDP;
430 	ip6h->hop_limit = 32;
431 	ip6h->saddr = np->local_ip.in6;
432 	ip6h->daddr = np->remote_ip.in6;
433 
434 	skb->protocol = htons(ETH_P_IPV6);
435 }
436 
push_ipv4(struct netpoll * np,struct sk_buff * skb,int len)437 static void push_ipv4(struct netpoll *np, struct sk_buff *skb, int len)
438 {
439 	static atomic_t ip_ident;
440 	struct iphdr *iph;
441 	int ip_len;
442 
443 	ip_len = len + sizeof(struct udphdr) + sizeof(struct iphdr);
444 
445 	skb_push(skb, sizeof(struct iphdr));
446 	skb_reset_network_header(skb);
447 	iph = ip_hdr(skb);
448 
449 	/* iph->version = 4; iph->ihl = 5; */
450 	*(unsigned char *)iph = 0x45;
451 	iph->tos = 0;
452 	put_unaligned(htons(ip_len), &iph->tot_len);
453 	iph->id = htons(atomic_inc_return(&ip_ident));
454 	iph->frag_off = 0;
455 	iph->ttl = 64;
456 	iph->protocol = IPPROTO_UDP;
457 	iph->check = 0;
458 	put_unaligned(np->local_ip.ip, &iph->saddr);
459 	put_unaligned(np->remote_ip.ip, &iph->daddr);
460 	iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
461 	skb->protocol = htons(ETH_P_IP);
462 }
463 
push_udp(struct netpoll * np,struct sk_buff * skb,int len)464 static void push_udp(struct netpoll *np, struct sk_buff *skb, int len)
465 {
466 	struct udphdr *udph;
467 	int udp_len;
468 
469 	udp_len = len + sizeof(struct udphdr);
470 
471 	skb_push(skb, sizeof(struct udphdr));
472 	skb_reset_transport_header(skb);
473 
474 	udph = udp_hdr(skb);
475 	udph->source = htons(np->local_port);
476 	udph->dest = htons(np->remote_port);
477 	udph->len = htons(udp_len);
478 
479 	netpoll_udp_checksum(np, skb, len);
480 }
481 
push_eth(struct netpoll * np,struct sk_buff * skb)482 static void push_eth(struct netpoll *np, struct sk_buff *skb)
483 {
484 	struct ethhdr *eth;
485 
486 	eth = skb_push(skb, ETH_HLEN);
487 	skb_reset_mac_header(skb);
488 	ether_addr_copy(eth->h_source, np->dev->dev_addr);
489 	ether_addr_copy(eth->h_dest, np->remote_mac);
490 	if (np->ipv6)
491 		eth->h_proto = htons(ETH_P_IPV6);
492 	else
493 		eth->h_proto = htons(ETH_P_IP);
494 }
495 
netpoll_send_udp(struct netpoll * np,const char * msg,int len)496 int netpoll_send_udp(struct netpoll *np, const char *msg, int len)
497 {
498 	int total_len, ip_len, udp_len;
499 	struct sk_buff *skb;
500 
501 	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
502 		WARN_ON_ONCE(!irqs_disabled());
503 
504 	udp_len = len + sizeof(struct udphdr);
505 	if (np->ipv6)
506 		ip_len = udp_len + sizeof(struct ipv6hdr);
507 	else
508 		ip_len = udp_len + sizeof(struct iphdr);
509 
510 	total_len = ip_len + LL_RESERVED_SPACE(np->dev);
511 
512 	skb = find_skb(np, total_len + np->dev->needed_tailroom,
513 		       total_len - len);
514 	if (!skb)
515 		return -ENOMEM;
516 
517 	skb_copy_to_linear_data(skb, msg, len);
518 	skb_put(skb, len);
519 
520 	push_udp(np, skb, len);
521 	if (np->ipv6)
522 		push_ipv6(np, skb, len);
523 	else
524 		push_ipv4(np, skb, len);
525 	push_eth(np, skb);
526 	skb->dev = np->dev;
527 
528 	return (int)netpoll_send_skb(np, skb);
529 }
530 EXPORT_SYMBOL(netpoll_send_udp);
531 
532 
skb_pool_flush(struct netpoll * np)533 static void skb_pool_flush(struct netpoll *np)
534 {
535 	struct sk_buff_head *skb_pool;
536 
537 	cancel_work_sync(&np->refill_wq);
538 	skb_pool = &np->skb_pool;
539 	skb_queue_purge_reason(skb_pool, SKB_CONSUMED);
540 }
541 
refill_skbs_work_handler(struct work_struct * work)542 static void refill_skbs_work_handler(struct work_struct *work)
543 {
544 	struct netpoll *np =
545 		container_of(work, struct netpoll, refill_wq);
546 
547 	refill_skbs(np);
548 }
549 
__netpoll_setup(struct netpoll * np,struct net_device * ndev)550 int __netpoll_setup(struct netpoll *np, struct net_device *ndev)
551 {
552 	struct netpoll_info *npinfo;
553 	const struct net_device_ops *ops;
554 	int err;
555 
556 	skb_queue_head_init(&np->skb_pool);
557 
558 	if (ndev->priv_flags & IFF_DISABLE_NETPOLL) {
559 		np_err(np, "%s doesn't support polling, aborting\n",
560 		       ndev->name);
561 		err = -ENOTSUPP;
562 		goto out;
563 	}
564 
565 	npinfo = rtnl_dereference(ndev->npinfo);
566 	if (!npinfo) {
567 		npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
568 		if (!npinfo) {
569 			err = -ENOMEM;
570 			goto out;
571 		}
572 
573 		sema_init(&npinfo->dev_lock, 1);
574 		skb_queue_head_init(&npinfo->txq);
575 		INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
576 
577 		refcount_set(&npinfo->refcnt, 1);
578 
579 		ops = ndev->netdev_ops;
580 		if (ops->ndo_netpoll_setup) {
581 			err = ops->ndo_netpoll_setup(ndev);
582 			if (err)
583 				goto free_npinfo;
584 		}
585 	} else {
586 		refcount_inc(&npinfo->refcnt);
587 	}
588 
589 	np->dev = ndev;
590 	strscpy(np->dev_name, ndev->name, IFNAMSIZ);
591 
592 	/* fill up the skb queue */
593 	refill_skbs(np);
594 	INIT_WORK(&np->refill_wq, refill_skbs_work_handler);
595 
596 	/* last thing to do is link it to the net device structure */
597 	rcu_assign_pointer(ndev->npinfo, npinfo);
598 
599 	return 0;
600 
601 free_npinfo:
602 	kfree(npinfo);
603 out:
604 	return err;
605 }
606 EXPORT_SYMBOL_GPL(__netpoll_setup);
607 
608 /*
609  * Returns a pointer to a string representation of the identifier used
610  * to select the egress interface for the given netpoll instance. buf
611  * must be a buffer of length at least MAC_ADDR_STR_LEN + 1.
612  */
egress_dev(struct netpoll * np,char * buf)613 static char *egress_dev(struct netpoll *np, char *buf)
614 {
615 	if (np->dev_name[0])
616 		return np->dev_name;
617 
618 	snprintf(buf, MAC_ADDR_STR_LEN, "%pM", np->dev_mac);
619 	return buf;
620 }
621 
netpoll_wait_carrier(struct netpoll * np,struct net_device * ndev,unsigned int timeout)622 static void netpoll_wait_carrier(struct netpoll *np, struct net_device *ndev,
623 				 unsigned int timeout)
624 {
625 	unsigned long atmost;
626 
627 	atmost = jiffies + timeout * HZ;
628 	while (!netif_carrier_ok(ndev)) {
629 		if (time_after(jiffies, atmost)) {
630 			np_notice(np, "timeout waiting for carrier\n");
631 			break;
632 		}
633 		msleep(1);
634 	}
635 }
636 
637 /*
638  * Take the IPv6 from ndev and populate local_ip structure in netpoll
639  */
netpoll_take_ipv6(struct netpoll * np,struct net_device * ndev)640 static int netpoll_take_ipv6(struct netpoll *np, struct net_device *ndev)
641 {
642 	char buf[MAC_ADDR_STR_LEN + 1];
643 	int err = -EDESTADDRREQ;
644 	struct inet6_dev *idev;
645 
646 	if (!IS_ENABLED(CONFIG_IPV6)) {
647 		np_err(np, "IPv6 is not supported %s, aborting\n",
648 		       egress_dev(np, buf));
649 		return -EINVAL;
650 	}
651 
652 	idev = __in6_dev_get(ndev);
653 	if (idev) {
654 		struct inet6_ifaddr *ifp;
655 
656 		read_lock_bh(&idev->lock);
657 		list_for_each_entry(ifp, &idev->addr_list, if_list) {
658 			if (!!(ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL) !=
659 				!!(ipv6_addr_type(&np->remote_ip.in6) & IPV6_ADDR_LINKLOCAL))
660 				continue;
661 			/* Got the IP, let's return */
662 			np->local_ip.in6 = ifp->addr;
663 			err = 0;
664 			break;
665 		}
666 		read_unlock_bh(&idev->lock);
667 	}
668 	if (err) {
669 		np_err(np, "no IPv6 address for %s, aborting\n",
670 		       egress_dev(np, buf));
671 		return err;
672 	}
673 
674 	np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
675 	return 0;
676 }
677 
678 /*
679  * Take the IPv4 from ndev and populate local_ip structure in netpoll
680  */
netpoll_take_ipv4(struct netpoll * np,struct net_device * ndev)681 static int netpoll_take_ipv4(struct netpoll *np, struct net_device *ndev)
682 {
683 	char buf[MAC_ADDR_STR_LEN + 1];
684 	const struct in_ifaddr *ifa;
685 	struct in_device *in_dev;
686 
687 	in_dev = __in_dev_get_rtnl(ndev);
688 	if (!in_dev) {
689 		np_err(np, "no IP address for %s, aborting\n",
690 		       egress_dev(np, buf));
691 		return -EDESTADDRREQ;
692 	}
693 
694 	ifa = rtnl_dereference(in_dev->ifa_list);
695 	if (!ifa) {
696 		np_err(np, "no IP address for %s, aborting\n",
697 		       egress_dev(np, buf));
698 		return -EDESTADDRREQ;
699 	}
700 
701 	np->local_ip.ip = ifa->ifa_local;
702 	np_info(np, "local IP %pI4\n", &np->local_ip.ip);
703 
704 	return 0;
705 }
706 
netpoll_setup(struct netpoll * np)707 int netpoll_setup(struct netpoll *np)
708 {
709 	struct net *net = current->nsproxy->net_ns;
710 	char buf[MAC_ADDR_STR_LEN + 1];
711 	struct net_device *ndev = NULL;
712 	bool ip_overwritten = false;
713 	int err;
714 
715 	rtnl_lock();
716 	if (np->dev_name[0])
717 		ndev = __dev_get_by_name(net, np->dev_name);
718 	else if (is_valid_ether_addr(np->dev_mac))
719 		ndev = dev_getbyhwaddr(net, ARPHRD_ETHER, np->dev_mac);
720 
721 	if (!ndev) {
722 		np_err(np, "%s doesn't exist, aborting\n", egress_dev(np, buf));
723 		err = -ENODEV;
724 		goto unlock;
725 	}
726 	netdev_hold(ndev, &np->dev_tracker, GFP_KERNEL);
727 
728 	if (netdev_master_upper_dev_get(ndev)) {
729 		np_err(np, "%s is a slave device, aborting\n",
730 		       egress_dev(np, buf));
731 		err = -EBUSY;
732 		goto put;
733 	}
734 
735 	if (!netif_running(ndev)) {
736 		np_info(np, "device %s not up yet, forcing it\n",
737 			egress_dev(np, buf));
738 
739 		err = dev_open(ndev, NULL);
740 		if (err) {
741 			np_err(np, "failed to open %s\n", ndev->name);
742 			goto put;
743 		}
744 
745 		rtnl_unlock();
746 		netpoll_wait_carrier(np, ndev, carrier_timeout);
747 		rtnl_lock();
748 	}
749 
750 	if (!np->local_ip.ip) {
751 		if (!np->ipv6) {
752 			err = netpoll_take_ipv4(np, ndev);
753 			if (err)
754 				goto put;
755 		} else {
756 			err = netpoll_take_ipv6(np, ndev);
757 			if (err)
758 				goto put;
759 		}
760 		ip_overwritten = true;
761 	}
762 
763 	err = __netpoll_setup(np, ndev);
764 	if (err)
765 		goto flush;
766 	rtnl_unlock();
767 
768 	/* Make sure all NAPI polls which started before dev->npinfo
769 	 * was visible have exited before we start calling NAPI poll.
770 	 * NAPI skips locking if dev->npinfo is NULL.
771 	 */
772 	synchronize_rcu();
773 
774 	return 0;
775 
776 flush:
777 	skb_pool_flush(np);
778 put:
779 	DEBUG_NET_WARN_ON_ONCE(np->dev);
780 	if (ip_overwritten)
781 		memset(&np->local_ip, 0, sizeof(np->local_ip));
782 	netdev_put(ndev, &np->dev_tracker);
783 unlock:
784 	rtnl_unlock();
785 	return err;
786 }
787 EXPORT_SYMBOL(netpoll_setup);
788 
rcu_cleanup_netpoll_info(struct rcu_head * rcu_head)789 static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
790 {
791 	struct netpoll_info *npinfo =
792 			container_of(rcu_head, struct netpoll_info, rcu);
793 
794 	skb_queue_purge(&npinfo->txq);
795 
796 	/* we can't call cancel_delayed_work_sync here, as we are in softirq */
797 	cancel_delayed_work(&npinfo->tx_work);
798 
799 	/* clean after last, unfinished work */
800 	__skb_queue_purge(&npinfo->txq);
801 	/* now cancel it again */
802 	cancel_delayed_work(&npinfo->tx_work);
803 	kfree(npinfo);
804 }
805 
__netpoll_cleanup(struct netpoll * np)806 static void __netpoll_cleanup(struct netpoll *np)
807 {
808 	struct netpoll_info *npinfo;
809 
810 	npinfo = rtnl_dereference(np->dev->npinfo);
811 	if (!npinfo)
812 		return;
813 
814 	/* At this point, there is a single npinfo instance per netdevice, and
815 	 * its refcnt tracks how many netpoll structures are linked to it. We
816 	 * only perform npinfo cleanup when the refcnt decrements to zero.
817 	 */
818 	if (refcount_dec_and_test(&npinfo->refcnt)) {
819 		const struct net_device_ops *ops;
820 
821 		ops = np->dev->netdev_ops;
822 		if (ops->ndo_netpoll_cleanup)
823 			ops->ndo_netpoll_cleanup(np->dev);
824 
825 		RCU_INIT_POINTER(np->dev->npinfo, NULL);
826 		call_rcu(&npinfo->rcu, rcu_cleanup_netpoll_info);
827 	}
828 
829 	skb_pool_flush(np);
830 }
831 
__netpoll_free(struct netpoll * np)832 void __netpoll_free(struct netpoll *np)
833 {
834 	ASSERT_RTNL();
835 
836 	/* Wait for transmitting packets to finish before freeing. */
837 	synchronize_net();
838 	__netpoll_cleanup(np);
839 	kfree(np);
840 }
841 EXPORT_SYMBOL_GPL(__netpoll_free);
842 
do_netpoll_cleanup(struct netpoll * np)843 void do_netpoll_cleanup(struct netpoll *np)
844 {
845 	__netpoll_cleanup(np);
846 	netdev_put(np->dev, &np->dev_tracker);
847 	np->dev = NULL;
848 }
849 EXPORT_SYMBOL(do_netpoll_cleanup);
850 
netpoll_cleanup(struct netpoll * np)851 void netpoll_cleanup(struct netpoll *np)
852 {
853 	rtnl_lock();
854 	if (!np->dev)
855 		goto out;
856 	do_netpoll_cleanup(np);
857 out:
858 	rtnl_unlock();
859 }
860 EXPORT_SYMBOL(netpoll_cleanup);
861