]> git.karo-electronics.de Git - karo-tx-linux.git/blob - net/core/netpoll.c
netpoll: Don't drop all received packets.
[karo-tx-linux.git] / net / core / netpoll.c
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
47 static struct sk_buff_head skb_pool;
48
49 static atomic_t trapped;
50
51 DEFINE_STATIC_SRCU(netpoll_srcu);
52
53 #define USEC_PER_POLL   50
54
55 #define MAX_SKB_SIZE                                                    \
56         (sizeof(struct ethhdr) +                                        \
57          sizeof(struct iphdr) +                                         \
58          sizeof(struct udphdr) +                                        \
59          MAX_UDP_CHUNK)
60
61 static void zap_completion_queue(void);
62 static void netpoll_neigh_reply(struct sk_buff *skb, struct netpoll_info *npinfo);
63 static void netpoll_async_cleanup(struct work_struct *work);
64
65 static unsigned int carrier_timeout = 4;
66 module_param(carrier_timeout, uint, 0644);
67
68 #define np_info(np, fmt, ...)                           \
69         pr_info("%s: " fmt, np->name, ##__VA_ARGS__)
70 #define np_err(np, fmt, ...)                            \
71         pr_err("%s: " fmt, np->name, ##__VA_ARGS__)
72 #define np_notice(np, fmt, ...)                         \
73         pr_notice("%s: " fmt, np->name, ##__VA_ARGS__)
74
75 static void queue_process(struct work_struct *work)
76 {
77         struct netpoll_info *npinfo =
78                 container_of(work, struct netpoll_info, tx_work.work);
79         struct sk_buff *skb;
80         unsigned long flags;
81
82         while ((skb = skb_dequeue(&npinfo->txq))) {
83                 struct net_device *dev = skb->dev;
84                 const struct net_device_ops *ops = dev->netdev_ops;
85                 struct netdev_queue *txq;
86
87                 if (!netif_device_present(dev) || !netif_running(dev)) {
88                         __kfree_skb(skb);
89                         continue;
90                 }
91
92                 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
93
94                 local_irq_save(flags);
95                 __netif_tx_lock(txq, smp_processor_id());
96                 if (netif_xmit_frozen_or_stopped(txq) ||
97                     ops->ndo_start_xmit(skb, dev) != NETDEV_TX_OK) {
98                         skb_queue_head(&npinfo->txq, skb);
99                         __netif_tx_unlock(txq);
100                         local_irq_restore(flags);
101
102                         schedule_delayed_work(&npinfo->tx_work, HZ/10);
103                         return;
104                 }
105                 __netif_tx_unlock(txq);
106                 local_irq_restore(flags);
107         }
108 }
109
110 static __sum16 checksum_udp(struct sk_buff *skb, struct udphdr *uh,
111                             unsigned short ulen, __be32 saddr, __be32 daddr)
112 {
113         __wsum psum;
114
115         if (uh->check == 0 || skb_csum_unnecessary(skb))
116                 return 0;
117
118         psum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0);
119
120         if (skb->ip_summed == CHECKSUM_COMPLETE &&
121             !csum_fold(csum_add(psum, skb->csum)))
122                 return 0;
123
124         skb->csum = psum;
125
126         return __skb_checksum_complete(skb);
127 }
128
129 /*
130  * Check whether delayed processing was scheduled for our NIC. If so,
131  * we attempt to grab the poll lock and use ->poll() to pump the card.
132  * If this fails, either we've recursed in ->poll() or it's already
133  * running on another CPU.
134  *
135  * Note: we don't mask interrupts with this lock because we're using
136  * trylock here and interrupts are already disabled in the softirq
137  * case. Further, we test the poll_owner to avoid recursion on UP
138  * systems where the lock doesn't exist.
139  *
140  * In cases where there is bi-directional communications, reading only
141  * one message at a time can lead to packets being dropped by the
142  * network adapter, forcing superfluous retries and possibly timeouts.
143  * Thus, we set our budget to greater than 1.
144  */
145 static int poll_one_napi(struct napi_struct *napi, int budget)
146 {
147         int work;
148
149         /* net_rx_action's ->poll() invocations and our's are
150          * synchronized by this test which is only made while
151          * holding the napi->poll_lock.
152          */
153         if (!test_bit(NAPI_STATE_SCHED, &napi->state))
154                 return budget;
155
156         set_bit(NAPI_STATE_NPSVC, &napi->state);
157
158         work = napi->poll(napi, budget);
159         WARN_ONCE(work > budget, "%pF exceeded budget in poll\n", napi->poll);
160         trace_napi_poll(napi);
161
162         clear_bit(NAPI_STATE_NPSVC, &napi->state);
163
164         return budget - work;
165 }
166
167 static void poll_napi(struct net_device *dev, int budget)
168 {
169         struct napi_struct *napi;
170
171         list_for_each_entry(napi, &dev->napi_list, dev_list) {
172                 if (napi->poll_owner != smp_processor_id() &&
173                     spin_trylock(&napi->poll_lock)) {
174                         budget = poll_one_napi(napi, budget);
175                         spin_unlock(&napi->poll_lock);
176                 }
177         }
178 }
179
180 static void service_neigh_queue(struct netpoll_info *npi)
181 {
182         if (npi) {
183                 struct sk_buff *skb;
184
185                 while ((skb = skb_dequeue(&npi->neigh_tx)))
186                         netpoll_neigh_reply(skb, npi);
187         }
188 }
189
190 static 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         bool rx_processing = netpoll_rx_processing(ni);
195         int budget = rx_processing? 16 : 0;
196
197         /* Don't do any rx activity if the dev_lock mutex is held
198          * the dev_open/close paths use this to block netpoll activity
199          * while changing device state
200          */
201         if (down_trylock(&ni->dev_lock))
202                 return;
203
204         if (!netif_running(dev)) {
205                 up(&ni->dev_lock);
206                 return;
207         }
208
209         if (rx_processing)
210                 atomic_inc(&trapped);
211
212         ops = dev->netdev_ops;
213         if (!ops->ndo_poll_controller) {
214                 up(&ni->dev_lock);
215                 return;
216         }
217
218         /* Process pending work on NIC */
219         ops->ndo_poll_controller(dev);
220
221         poll_napi(dev, budget);
222
223         if (rx_processing)
224                 atomic_dec(&trapped);
225
226         up(&ni->dev_lock);
227
228         if (dev->flags & IFF_SLAVE) {
229                 if (ni) {
230                         struct net_device *bond_dev;
231                         struct sk_buff *skb;
232                         struct netpoll_info *bond_ni;
233
234                         bond_dev = netdev_master_upper_dev_get_rcu(dev);
235                         bond_ni = rcu_dereference_bh(bond_dev->npinfo);
236                         while ((skb = skb_dequeue(&ni->neigh_tx))) {
237                                 skb->dev = bond_dev;
238                                 skb_queue_tail(&bond_ni->neigh_tx, skb);
239                         }
240                 }
241         }
242
243         service_neigh_queue(ni);
244
245         zap_completion_queue();
246 }
247
248 void netpoll_rx_disable(struct net_device *dev)
249 {
250         struct netpoll_info *ni;
251         int idx;
252         might_sleep();
253         idx = srcu_read_lock(&netpoll_srcu);
254         ni = srcu_dereference(dev->npinfo, &netpoll_srcu);
255         if (ni)
256                 down(&ni->dev_lock);
257         srcu_read_unlock(&netpoll_srcu, idx);
258 }
259 EXPORT_SYMBOL(netpoll_rx_disable);
260
261 void netpoll_rx_enable(struct net_device *dev)
262 {
263         struct netpoll_info *ni;
264         rcu_read_lock();
265         ni = rcu_dereference(dev->npinfo);
266         if (ni)
267                 up(&ni->dev_lock);
268         rcu_read_unlock();
269 }
270 EXPORT_SYMBOL(netpoll_rx_enable);
271
272 static void refill_skbs(void)
273 {
274         struct sk_buff *skb;
275         unsigned long flags;
276
277         spin_lock_irqsave(&skb_pool.lock, flags);
278         while (skb_pool.qlen < MAX_SKBS) {
279                 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
280                 if (!skb)
281                         break;
282
283                 __skb_queue_tail(&skb_pool, skb);
284         }
285         spin_unlock_irqrestore(&skb_pool.lock, flags);
286 }
287
288 static void zap_completion_queue(void)
289 {
290         unsigned long flags;
291         struct softnet_data *sd = &get_cpu_var(softnet_data);
292
293         if (sd->completion_queue) {
294                 struct sk_buff *clist;
295
296                 local_irq_save(flags);
297                 clist = sd->completion_queue;
298                 sd->completion_queue = NULL;
299                 local_irq_restore(flags);
300
301                 while (clist != NULL) {
302                         struct sk_buff *skb = clist;
303                         clist = clist->next;
304                         if (skb->destructor) {
305                                 atomic_inc(&skb->users);
306                                 dev_kfree_skb_any(skb); /* put this one back */
307                         } else {
308                                 __kfree_skb(skb);
309                         }
310                 }
311         }
312
313         put_cpu_var(softnet_data);
314 }
315
316 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
317 {
318         int count = 0;
319         struct sk_buff *skb;
320
321         zap_completion_queue();
322         refill_skbs();
323 repeat:
324
325         skb = alloc_skb(len, GFP_ATOMIC);
326         if (!skb)
327                 skb = skb_dequeue(&skb_pool);
328
329         if (!skb) {
330                 if (++count < 10) {
331                         netpoll_poll_dev(np->dev);
332                         goto repeat;
333                 }
334                 return NULL;
335         }
336
337         atomic_set(&skb->users, 1);
338         skb_reserve(skb, reserve);
339         return skb;
340 }
341
342 static int netpoll_owner_active(struct net_device *dev)
343 {
344         struct napi_struct *napi;
345
346         list_for_each_entry(napi, &dev->napi_list, dev_list) {
347                 if (napi->poll_owner == smp_processor_id())
348                         return 1;
349         }
350         return 0;
351 }
352
353 /* call with IRQ disabled */
354 void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb,
355                              struct net_device *dev)
356 {
357         int status = NETDEV_TX_BUSY;
358         unsigned long tries;
359         const struct net_device_ops *ops = dev->netdev_ops;
360         /* It is up to the caller to keep npinfo alive. */
361         struct netpoll_info *npinfo;
362
363         WARN_ON_ONCE(!irqs_disabled());
364
365         npinfo = rcu_dereference_bh(np->dev->npinfo);
366         if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
367                 __kfree_skb(skb);
368                 return;
369         }
370
371         /* don't get messages out of order, and no recursion */
372         if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
373                 struct netdev_queue *txq;
374
375                 txq = netdev_pick_tx(dev, skb, NULL);
376
377                 /* try until next clock tick */
378                 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
379                      tries > 0; --tries) {
380                         if (__netif_tx_trylock(txq)) {
381                                 if (!netif_xmit_stopped(txq)) {
382                                         if (vlan_tx_tag_present(skb) &&
383                                             !vlan_hw_offload_capable(netif_skb_features(skb),
384                                                                      skb->vlan_proto)) {
385                                                 skb = __vlan_put_tag(skb, skb->vlan_proto, vlan_tx_tag_get(skb));
386                                                 if (unlikely(!skb)) {
387                                                         /* This is actually a packet drop, but we
388                                                          * don't want the code at the end of this
389                                                          * function to try and re-queue a NULL skb.
390                                                          */
391                                                         status = NETDEV_TX_OK;
392                                                         goto unlock_txq;
393                                                 }
394                                                 skb->vlan_tci = 0;
395                                         }
396
397                                         status = ops->ndo_start_xmit(skb, dev);
398                                         if (status == NETDEV_TX_OK)
399                                                 txq_trans_update(txq);
400                                 }
401                         unlock_txq:
402                                 __netif_tx_unlock(txq);
403
404                                 if (status == NETDEV_TX_OK)
405                                         break;
406
407                         }
408
409                         /* tickle device maybe there is some cleanup */
410                         netpoll_poll_dev(np->dev);
411
412                         udelay(USEC_PER_POLL);
413                 }
414
415                 WARN_ONCE(!irqs_disabled(),
416                         "netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pF)\n",
417                         dev->name, ops->ndo_start_xmit);
418
419         }
420
421         if (status != NETDEV_TX_OK) {
422                 skb_queue_tail(&npinfo->txq, skb);
423                 schedule_delayed_work(&npinfo->tx_work,0);
424         }
425 }
426 EXPORT_SYMBOL(netpoll_send_skb_on_dev);
427
428 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
429 {
430         int total_len, ip_len, udp_len;
431         struct sk_buff *skb;
432         struct udphdr *udph;
433         struct iphdr *iph;
434         struct ethhdr *eth;
435         static atomic_t ip_ident;
436         struct ipv6hdr *ip6h;
437
438         udp_len = len + sizeof(*udph);
439         if (np->ipv6)
440                 ip_len = udp_len + sizeof(*ip6h);
441         else
442                 ip_len = udp_len + sizeof(*iph);
443
444         total_len = ip_len + LL_RESERVED_SPACE(np->dev);
445
446         skb = find_skb(np, total_len + np->dev->needed_tailroom,
447                        total_len - len);
448         if (!skb)
449                 return;
450
451         skb_copy_to_linear_data(skb, msg, len);
452         skb_put(skb, len);
453
454         skb_push(skb, sizeof(*udph));
455         skb_reset_transport_header(skb);
456         udph = udp_hdr(skb);
457         udph->source = htons(np->local_port);
458         udph->dest = htons(np->remote_port);
459         udph->len = htons(udp_len);
460
461         if (np->ipv6) {
462                 udph->check = 0;
463                 udph->check = csum_ipv6_magic(&np->local_ip.in6,
464                                               &np->remote_ip.in6,
465                                               udp_len, IPPROTO_UDP,
466                                               csum_partial(udph, udp_len, 0));
467                 if (udph->check == 0)
468                         udph->check = CSUM_MANGLED_0;
469
470                 skb_push(skb, sizeof(*ip6h));
471                 skb_reset_network_header(skb);
472                 ip6h = ipv6_hdr(skb);
473
474                 /* ip6h->version = 6; ip6h->priority = 0; */
475                 put_unaligned(0x60, (unsigned char *)ip6h);
476                 ip6h->flow_lbl[0] = 0;
477                 ip6h->flow_lbl[1] = 0;
478                 ip6h->flow_lbl[2] = 0;
479
480                 ip6h->payload_len = htons(sizeof(struct udphdr) + len);
481                 ip6h->nexthdr = IPPROTO_UDP;
482                 ip6h->hop_limit = 32;
483                 ip6h->saddr = np->local_ip.in6;
484                 ip6h->daddr = np->remote_ip.in6;
485
486                 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
487                 skb_reset_mac_header(skb);
488                 skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
489         } else {
490                 udph->check = 0;
491                 udph->check = csum_tcpudp_magic(np->local_ip.ip,
492                                                 np->remote_ip.ip,
493                                                 udp_len, IPPROTO_UDP,
494                                                 csum_partial(udph, udp_len, 0));
495                 if (udph->check == 0)
496                         udph->check = CSUM_MANGLED_0;
497
498                 skb_push(skb, sizeof(*iph));
499                 skb_reset_network_header(skb);
500                 iph = ip_hdr(skb);
501
502                 /* iph->version = 4; iph->ihl = 5; */
503                 put_unaligned(0x45, (unsigned char *)iph);
504                 iph->tos      = 0;
505                 put_unaligned(htons(ip_len), &(iph->tot_len));
506                 iph->id       = htons(atomic_inc_return(&ip_ident));
507                 iph->frag_off = 0;
508                 iph->ttl      = 64;
509                 iph->protocol = IPPROTO_UDP;
510                 iph->check    = 0;
511                 put_unaligned(np->local_ip.ip, &(iph->saddr));
512                 put_unaligned(np->remote_ip.ip, &(iph->daddr));
513                 iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
514
515                 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
516                 skb_reset_mac_header(skb);
517                 skb->protocol = eth->h_proto = htons(ETH_P_IP);
518         }
519
520         ether_addr_copy(eth->h_source, np->dev->dev_addr);
521         ether_addr_copy(eth->h_dest, np->remote_mac);
522
523         skb->dev = np->dev;
524
525         netpoll_send_skb(np, skb);
526 }
527 EXPORT_SYMBOL(netpoll_send_udp);
528
529 static void netpoll_neigh_reply(struct sk_buff *skb, struct netpoll_info *npinfo)
530 {
531         int size, type = ARPOP_REPLY;
532         __be32 sip, tip;
533         unsigned char *sha;
534         struct sk_buff *send_skb;
535         struct netpoll *np, *tmp;
536         unsigned long flags;
537         int hlen, tlen;
538         int hits = 0, proto;
539
540         if (!netpoll_rx_processing(npinfo))
541                 return;
542
543         /* Before checking the packet, we do some early
544            inspection whether this is interesting at all */
545         spin_lock_irqsave(&npinfo->rx_lock, flags);
546         list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
547                 if (np->dev == skb->dev)
548                         hits++;
549         }
550         spin_unlock_irqrestore(&npinfo->rx_lock, flags);
551
552         /* No netpoll struct is using this dev */
553         if (!hits)
554                 return;
555
556         proto = ntohs(eth_hdr(skb)->h_proto);
557         if (proto == ETH_P_ARP) {
558                 struct arphdr *arp;
559                 unsigned char *arp_ptr;
560                 /* No arp on this interface */
561                 if (skb->dev->flags & IFF_NOARP)
562                         return;
563
564                 if (!pskb_may_pull(skb, arp_hdr_len(skb->dev)))
565                         return;
566
567                 skb_reset_network_header(skb);
568                 skb_reset_transport_header(skb);
569                 arp = arp_hdr(skb);
570
571                 if ((arp->ar_hrd != htons(ARPHRD_ETHER) &&
572                      arp->ar_hrd != htons(ARPHRD_IEEE802)) ||
573                     arp->ar_pro != htons(ETH_P_IP) ||
574                     arp->ar_op != htons(ARPOP_REQUEST))
575                         return;
576
577                 arp_ptr = (unsigned char *)(arp+1);
578                 /* save the location of the src hw addr */
579                 sha = arp_ptr;
580                 arp_ptr += skb->dev->addr_len;
581                 memcpy(&sip, arp_ptr, 4);
582                 arp_ptr += 4;
583                 /* If we actually cared about dst hw addr,
584                    it would get copied here */
585                 arp_ptr += skb->dev->addr_len;
586                 memcpy(&tip, arp_ptr, 4);
587
588                 /* Should we ignore arp? */
589                 if (ipv4_is_loopback(tip) || ipv4_is_multicast(tip))
590                         return;
591
592                 size = arp_hdr_len(skb->dev);
593
594                 spin_lock_irqsave(&npinfo->rx_lock, flags);
595                 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
596                         if (tip != np->local_ip.ip)
597                                 continue;
598
599                         hlen = LL_RESERVED_SPACE(np->dev);
600                         tlen = np->dev->needed_tailroom;
601                         send_skb = find_skb(np, size + hlen + tlen, hlen);
602                         if (!send_skb)
603                                 continue;
604
605                         skb_reset_network_header(send_skb);
606                         arp = (struct arphdr *) skb_put(send_skb, size);
607                         send_skb->dev = skb->dev;
608                         send_skb->protocol = htons(ETH_P_ARP);
609
610                         /* Fill the device header for the ARP frame */
611                         if (dev_hard_header(send_skb, skb->dev, ETH_P_ARP,
612                                             sha, np->dev->dev_addr,
613                                             send_skb->len) < 0) {
614                                 kfree_skb(send_skb);
615                                 continue;
616                         }
617
618                         /*
619                          * Fill out the arp protocol part.
620                          *
621                          * we only support ethernet device type,
622                          * which (according to RFC 1390) should
623                          * always equal 1 (Ethernet).
624                          */
625
626                         arp->ar_hrd = htons(np->dev->type);
627                         arp->ar_pro = htons(ETH_P_IP);
628                         arp->ar_hln = np->dev->addr_len;
629                         arp->ar_pln = 4;
630                         arp->ar_op = htons(type);
631
632                         arp_ptr = (unsigned char *)(arp + 1);
633                         memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
634                         arp_ptr += np->dev->addr_len;
635                         memcpy(arp_ptr, &tip, 4);
636                         arp_ptr += 4;
637                         memcpy(arp_ptr, sha, np->dev->addr_len);
638                         arp_ptr += np->dev->addr_len;
639                         memcpy(arp_ptr, &sip, 4);
640
641                         netpoll_send_skb(np, send_skb);
642
643                         /* If there are several rx_skb_hooks for the same
644                          * address we're fine by sending a single reply
645                          */
646                         break;
647                 }
648                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
649         } else if( proto == ETH_P_IPV6) {
650 #if IS_ENABLED(CONFIG_IPV6)
651                 struct nd_msg *msg;
652                 u8 *lladdr = NULL;
653                 struct ipv6hdr *hdr;
654                 struct icmp6hdr *icmp6h;
655                 const struct in6_addr *saddr;
656                 const struct in6_addr *daddr;
657                 struct inet6_dev *in6_dev = NULL;
658                 struct in6_addr *target;
659
660                 in6_dev = in6_dev_get(skb->dev);
661                 if (!in6_dev || !in6_dev->cnf.accept_ra)
662                         return;
663
664                 if (!pskb_may_pull(skb, skb->len))
665                         return;
666
667                 msg = (struct nd_msg *)skb_transport_header(skb);
668
669                 __skb_push(skb, skb->data - skb_transport_header(skb));
670
671                 if (ipv6_hdr(skb)->hop_limit != 255)
672                         return;
673                 if (msg->icmph.icmp6_code != 0)
674                         return;
675                 if (msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
676                         return;
677
678                 saddr = &ipv6_hdr(skb)->saddr;
679                 daddr = &ipv6_hdr(skb)->daddr;
680
681                 size = sizeof(struct icmp6hdr) + sizeof(struct in6_addr);
682
683                 spin_lock_irqsave(&npinfo->rx_lock, flags);
684                 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
685                         if (!ipv6_addr_equal(daddr, &np->local_ip.in6))
686                                 continue;
687
688                         hlen = LL_RESERVED_SPACE(np->dev);
689                         tlen = np->dev->needed_tailroom;
690                         send_skb = find_skb(np, size + hlen + tlen, hlen);
691                         if (!send_skb)
692                                 continue;
693
694                         send_skb->protocol = htons(ETH_P_IPV6);
695                         send_skb->dev = skb->dev;
696
697                         skb_reset_network_header(send_skb);
698                         hdr = (struct ipv6hdr *) skb_put(send_skb, sizeof(struct ipv6hdr));
699                         *(__be32*)hdr = htonl(0x60000000);
700                         hdr->payload_len = htons(size);
701                         hdr->nexthdr = IPPROTO_ICMPV6;
702                         hdr->hop_limit = 255;
703                         hdr->saddr = *saddr;
704                         hdr->daddr = *daddr;
705
706                         icmp6h = (struct icmp6hdr *) skb_put(send_skb, sizeof(struct icmp6hdr));
707                         icmp6h->icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
708                         icmp6h->icmp6_router = 0;
709                         icmp6h->icmp6_solicited = 1;
710
711                         target = (struct in6_addr *) skb_put(send_skb, sizeof(struct in6_addr));
712                         *target = msg->target;
713                         icmp6h->icmp6_cksum = csum_ipv6_magic(saddr, daddr, size,
714                                                               IPPROTO_ICMPV6,
715                                                               csum_partial(icmp6h,
716                                                                            size, 0));
717
718                         if (dev_hard_header(send_skb, skb->dev, ETH_P_IPV6,
719                                             lladdr, np->dev->dev_addr,
720                                             send_skb->len) < 0) {
721                                 kfree_skb(send_skb);
722                                 continue;
723                         }
724
725                         netpoll_send_skb(np, send_skb);
726
727                         /* If there are several rx_skb_hooks for the same
728                          * address, we're fine by sending a single reply
729                          */
730                         break;
731                 }
732                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
733 #endif
734         }
735 }
736
737 static bool pkt_is_ns(struct sk_buff *skb)
738 {
739         struct nd_msg *msg;
740         struct ipv6hdr *hdr;
741
742         if (skb->protocol != htons(ETH_P_ARP))
743                 return false;
744         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr) + sizeof(struct nd_msg)))
745                 return false;
746
747         msg = (struct nd_msg *)skb_transport_header(skb);
748         __skb_push(skb, skb->data - skb_transport_header(skb));
749         hdr = ipv6_hdr(skb);
750
751         if (hdr->nexthdr != IPPROTO_ICMPV6)
752                 return false;
753         if (hdr->hop_limit != 255)
754                 return false;
755         if (msg->icmph.icmp6_code != 0)
756                 return false;
757         if (msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
758                 return false;
759
760         return true;
761 }
762
763 int __netpoll_rx(struct sk_buff *skb, struct netpoll_info *npinfo)
764 {
765         int proto, len, ulen, data_len;
766         int hits = 0, offset;
767         const struct iphdr *iph;
768         struct udphdr *uh;
769         struct netpoll *np, *tmp;
770         uint16_t source;
771
772         if (!netpoll_rx_processing(npinfo))
773                 goto out;
774
775         if (skb->dev->type != ARPHRD_ETHER)
776                 goto out;
777
778         /* check if netpoll clients need ARP */
779         if (skb->protocol == htons(ETH_P_ARP) && atomic_read(&trapped)) {
780                 skb_queue_tail(&npinfo->neigh_tx, skb);
781                 return 1;
782         } else if (pkt_is_ns(skb) && atomic_read(&trapped)) {
783                 skb_queue_tail(&npinfo->neigh_tx, skb);
784                 return 1;
785         }
786
787         if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
788                 skb = vlan_untag(skb);
789                 if (unlikely(!skb))
790                         goto out;
791         }
792
793         proto = ntohs(eth_hdr(skb)->h_proto);
794         if (proto != ETH_P_IP && proto != ETH_P_IPV6)
795                 goto out;
796         if (skb->pkt_type == PACKET_OTHERHOST)
797                 goto out;
798         if (skb_shared(skb))
799                 goto out;
800
801         if (proto == ETH_P_IP) {
802                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
803                         goto out;
804                 iph = (struct iphdr *)skb->data;
805                 if (iph->ihl < 5 || iph->version != 4)
806                         goto out;
807                 if (!pskb_may_pull(skb, iph->ihl*4))
808                         goto out;
809                 iph = (struct iphdr *)skb->data;
810                 if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
811                         goto out;
812
813                 len = ntohs(iph->tot_len);
814                 if (skb->len < len || len < iph->ihl*4)
815                         goto out;
816
817                 /*
818                  * Our transport medium may have padded the buffer out.
819                  * Now We trim to the true length of the frame.
820                  */
821                 if (pskb_trim_rcsum(skb, len))
822                         goto out;
823
824                 iph = (struct iphdr *)skb->data;
825                 if (iph->protocol != IPPROTO_UDP)
826                         goto out;
827
828                 len -= iph->ihl*4;
829                 uh = (struct udphdr *)(((char *)iph) + iph->ihl*4);
830                 offset = (unsigned char *)(uh + 1) - skb->data;
831                 ulen = ntohs(uh->len);
832                 data_len = skb->len - offset;
833                 source = ntohs(uh->source);
834
835                 if (ulen != len)
836                         goto out;
837                 if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr))
838                         goto out;
839                 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
840                         if (np->local_ip.ip && np->local_ip.ip != iph->daddr)
841                                 continue;
842                         if (np->remote_ip.ip && np->remote_ip.ip != iph->saddr)
843                                 continue;
844                         if (np->local_port && np->local_port != ntohs(uh->dest))
845                                 continue;
846
847                         np->rx_skb_hook(np, source, skb, offset, data_len);
848                         hits++;
849                 }
850         } else {
851 #if IS_ENABLED(CONFIG_IPV6)
852                 const struct ipv6hdr *ip6h;
853
854                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
855                         goto out;
856                 ip6h = (struct ipv6hdr *)skb->data;
857                 if (ip6h->version != 6)
858                         goto out;
859                 len = ntohs(ip6h->payload_len);
860                 if (!len)
861                         goto out;
862                 if (len + sizeof(struct ipv6hdr) > skb->len)
863                         goto out;
864                 if (pskb_trim_rcsum(skb, len + sizeof(struct ipv6hdr)))
865                         goto out;
866                 ip6h = ipv6_hdr(skb);
867                 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
868                         goto out;
869                 uh = udp_hdr(skb);
870                 offset = (unsigned char *)(uh + 1) - skb->data;
871                 ulen = ntohs(uh->len);
872                 data_len = skb->len - offset;
873                 source = ntohs(uh->source);
874                 if (ulen != skb->len)
875                         goto out;
876                 if (udp6_csum_init(skb, uh, IPPROTO_UDP))
877                         goto out;
878                 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
879                         if (!ipv6_addr_equal(&np->local_ip.in6, &ip6h->daddr))
880                                 continue;
881                         if (!ipv6_addr_equal(&np->remote_ip.in6, &ip6h->saddr))
882                                 continue;
883                         if (np->local_port && np->local_port != ntohs(uh->dest))
884                                 continue;
885
886                         np->rx_skb_hook(np, source, skb, offset, data_len);
887                         hits++;
888                 }
889 #endif
890         }
891
892         if (!hits)
893                 goto out;
894
895         kfree_skb(skb);
896         return 1;
897
898 out:
899         if (atomic_read(&trapped)) {
900                 kfree_skb(skb);
901                 return 1;
902         }
903
904         return 0;
905 }
906
907 void netpoll_print_options(struct netpoll *np)
908 {
909         np_info(np, "local port %d\n", np->local_port);
910         if (np->ipv6)
911                 np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
912         else
913                 np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
914         np_info(np, "interface '%s'\n", np->dev_name);
915         np_info(np, "remote port %d\n", np->remote_port);
916         if (np->ipv6)
917                 np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
918         else
919                 np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
920         np_info(np, "remote ethernet address %pM\n", np->remote_mac);
921 }
922 EXPORT_SYMBOL(netpoll_print_options);
923
924 static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
925 {
926         const char *end;
927
928         if (!strchr(str, ':') &&
929             in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
930                 if (!*end)
931                         return 0;
932         }
933         if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
934 #if IS_ENABLED(CONFIG_IPV6)
935                 if (!*end)
936                         return 1;
937 #else
938                 return -1;
939 #endif
940         }
941         return -1;
942 }
943
944 int netpoll_parse_options(struct netpoll *np, char *opt)
945 {
946         char *cur=opt, *delim;
947         int ipv6;
948         bool ipversion_set = false;
949
950         if (*cur != '@') {
951                 if ((delim = strchr(cur, '@')) == NULL)
952                         goto parse_failed;
953                 *delim = 0;
954                 if (kstrtou16(cur, 10, &np->local_port))
955                         goto parse_failed;
956                 cur = delim;
957         }
958         cur++;
959
960         if (*cur != '/') {
961                 ipversion_set = true;
962                 if ((delim = strchr(cur, '/')) == NULL)
963                         goto parse_failed;
964                 *delim = 0;
965                 ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
966                 if (ipv6 < 0)
967                         goto parse_failed;
968                 else
969                         np->ipv6 = (bool)ipv6;
970                 cur = delim;
971         }
972         cur++;
973
974         if (*cur != ',') {
975                 /* parse out dev name */
976                 if ((delim = strchr(cur, ',')) == NULL)
977                         goto parse_failed;
978                 *delim = 0;
979                 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
980                 cur = delim;
981         }
982         cur++;
983
984         if (*cur != '@') {
985                 /* dst port */
986                 if ((delim = strchr(cur, '@')) == NULL)
987                         goto parse_failed;
988                 *delim = 0;
989                 if (*cur == ' ' || *cur == '\t')
990                         np_info(np, "warning: whitespace is not allowed\n");
991                 if (kstrtou16(cur, 10, &np->remote_port))
992                         goto parse_failed;
993                 cur = delim;
994         }
995         cur++;
996
997         /* dst ip */
998         if ((delim = strchr(cur, '/')) == NULL)
999                 goto parse_failed;
1000         *delim = 0;
1001         ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
1002         if (ipv6 < 0)
1003                 goto parse_failed;
1004         else if (ipversion_set && np->ipv6 != (bool)ipv6)
1005                 goto parse_failed;
1006         else
1007                 np->ipv6 = (bool)ipv6;
1008         cur = delim + 1;
1009
1010         if (*cur != 0) {
1011                 /* MAC address */
1012                 if (!mac_pton(cur, np->remote_mac))
1013                         goto parse_failed;
1014         }
1015
1016         netpoll_print_options(np);
1017
1018         return 0;
1019
1020  parse_failed:
1021         np_info(np, "couldn't parse config at '%s'!\n", cur);
1022         return -1;
1023 }
1024 EXPORT_SYMBOL(netpoll_parse_options);
1025
1026 int __netpoll_setup(struct netpoll *np, struct net_device *ndev, gfp_t gfp)
1027 {
1028         struct netpoll_info *npinfo;
1029         const struct net_device_ops *ops;
1030         unsigned long flags;
1031         int err;
1032
1033         np->dev = ndev;
1034         strlcpy(np->dev_name, ndev->name, IFNAMSIZ);
1035         INIT_WORK(&np->cleanup_work, netpoll_async_cleanup);
1036
1037         if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) ||
1038             !ndev->netdev_ops->ndo_poll_controller) {
1039                 np_err(np, "%s doesn't support polling, aborting\n",
1040                        np->dev_name);
1041                 err = -ENOTSUPP;
1042                 goto out;
1043         }
1044
1045         if (!ndev->npinfo) {
1046                 npinfo = kmalloc(sizeof(*npinfo), gfp);
1047                 if (!npinfo) {
1048                         err = -ENOMEM;
1049                         goto out;
1050                 }
1051
1052                 INIT_LIST_HEAD(&npinfo->rx_np);
1053
1054                 spin_lock_init(&npinfo->rx_lock);
1055                 sema_init(&npinfo->dev_lock, 1);
1056                 skb_queue_head_init(&npinfo->neigh_tx);
1057                 skb_queue_head_init(&npinfo->txq);
1058                 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
1059
1060                 atomic_set(&npinfo->refcnt, 1);
1061
1062                 ops = np->dev->netdev_ops;
1063                 if (ops->ndo_netpoll_setup) {
1064                         err = ops->ndo_netpoll_setup(ndev, npinfo, gfp);
1065                         if (err)
1066                                 goto free_npinfo;
1067                 }
1068         } else {
1069                 npinfo = rtnl_dereference(ndev->npinfo);
1070                 atomic_inc(&npinfo->refcnt);
1071         }
1072
1073         npinfo->netpoll = np;
1074
1075         if (np->rx_skb_hook) {
1076                 spin_lock_irqsave(&npinfo->rx_lock, flags);
1077                 list_add_tail(&np->rx, &npinfo->rx_np);
1078                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
1079         }
1080
1081         /* last thing to do is link it to the net device structure */
1082         rcu_assign_pointer(ndev->npinfo, npinfo);
1083
1084         return 0;
1085
1086 free_npinfo:
1087         kfree(npinfo);
1088 out:
1089         return err;
1090 }
1091 EXPORT_SYMBOL_GPL(__netpoll_setup);
1092
1093 int netpoll_setup(struct netpoll *np)
1094 {
1095         struct net_device *ndev = NULL;
1096         struct in_device *in_dev;
1097         int err;
1098
1099         rtnl_lock();
1100         if (np->dev_name) {
1101                 struct net *net = current->nsproxy->net_ns;
1102                 ndev = __dev_get_by_name(net, np->dev_name);
1103         }
1104         if (!ndev) {
1105                 np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
1106                 err = -ENODEV;
1107                 goto unlock;
1108         }
1109         dev_hold(ndev);
1110
1111         if (netdev_master_upper_dev_get(ndev)) {
1112                 np_err(np, "%s is a slave device, aborting\n", np->dev_name);
1113                 err = -EBUSY;
1114                 goto put;
1115         }
1116
1117         if (!netif_running(ndev)) {
1118                 unsigned long atmost, atleast;
1119
1120                 np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
1121
1122                 err = dev_open(ndev);
1123
1124                 if (err) {
1125                         np_err(np, "failed to open %s\n", ndev->name);
1126                         goto put;
1127                 }
1128
1129                 rtnl_unlock();
1130                 atleast = jiffies + HZ/10;
1131                 atmost = jiffies + carrier_timeout * HZ;
1132                 while (!netif_carrier_ok(ndev)) {
1133                         if (time_after(jiffies, atmost)) {
1134                                 np_notice(np, "timeout waiting for carrier\n");
1135                                 break;
1136                         }
1137                         msleep(1);
1138                 }
1139
1140                 /* If carrier appears to come up instantly, we don't
1141                  * trust it and pause so that we don't pump all our
1142                  * queued console messages into the bitbucket.
1143                  */
1144
1145                 if (time_before(jiffies, atleast)) {
1146                         np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
1147                         msleep(4000);
1148                 }
1149                 rtnl_lock();
1150         }
1151
1152         if (!np->local_ip.ip) {
1153                 if (!np->ipv6) {
1154                         in_dev = __in_dev_get_rtnl(ndev);
1155
1156                         if (!in_dev || !in_dev->ifa_list) {
1157                                 np_err(np, "no IP address for %s, aborting\n",
1158                                        np->dev_name);
1159                                 err = -EDESTADDRREQ;
1160                                 goto put;
1161                         }
1162
1163                         np->local_ip.ip = in_dev->ifa_list->ifa_local;
1164                         np_info(np, "local IP %pI4\n", &np->local_ip.ip);
1165                 } else {
1166 #if IS_ENABLED(CONFIG_IPV6)
1167                         struct inet6_dev *idev;
1168
1169                         err = -EDESTADDRREQ;
1170                         idev = __in6_dev_get(ndev);
1171                         if (idev) {
1172                                 struct inet6_ifaddr *ifp;
1173
1174                                 read_lock_bh(&idev->lock);
1175                                 list_for_each_entry(ifp, &idev->addr_list, if_list) {
1176                                         if (ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL)
1177                                                 continue;
1178                                         np->local_ip.in6 = ifp->addr;
1179                                         err = 0;
1180                                         break;
1181                                 }
1182                                 read_unlock_bh(&idev->lock);
1183                         }
1184                         if (err) {
1185                                 np_err(np, "no IPv6 address for %s, aborting\n",
1186                                        np->dev_name);
1187                                 goto put;
1188                         } else
1189                                 np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
1190 #else
1191                         np_err(np, "IPv6 is not supported %s, aborting\n",
1192                                np->dev_name);
1193                         err = -EINVAL;
1194                         goto put;
1195 #endif
1196                 }
1197         }
1198
1199         /* fill up the skb queue */
1200         refill_skbs();
1201
1202         err = __netpoll_setup(np, ndev, GFP_KERNEL);
1203         if (err)
1204                 goto put;
1205
1206         rtnl_unlock();
1207         return 0;
1208
1209 put:
1210         dev_put(ndev);
1211 unlock:
1212         rtnl_unlock();
1213         return err;
1214 }
1215 EXPORT_SYMBOL(netpoll_setup);
1216
1217 static int __init netpoll_init(void)
1218 {
1219         skb_queue_head_init(&skb_pool);
1220         return 0;
1221 }
1222 core_initcall(netpoll_init);
1223
1224 static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
1225 {
1226         struct netpoll_info *npinfo =
1227                         container_of(rcu_head, struct netpoll_info, rcu);
1228
1229         skb_queue_purge(&npinfo->neigh_tx);
1230         skb_queue_purge(&npinfo->txq);
1231
1232         /* we can't call cancel_delayed_work_sync here, as we are in softirq */
1233         cancel_delayed_work(&npinfo->tx_work);
1234
1235         /* clean after last, unfinished work */
1236         __skb_queue_purge(&npinfo->txq);
1237         /* now cancel it again */
1238         cancel_delayed_work(&npinfo->tx_work);
1239         kfree(npinfo);
1240 }
1241
1242 void __netpoll_cleanup(struct netpoll *np)
1243 {
1244         struct netpoll_info *npinfo;
1245         unsigned long flags;
1246
1247         /* rtnl_dereference would be preferable here but
1248          * rcu_cleanup_netpoll path can put us in here safely without
1249          * holding the rtnl, so plain rcu_dereference it is
1250          */
1251         npinfo = rtnl_dereference(np->dev->npinfo);
1252         if (!npinfo)
1253                 return;
1254
1255         if (!list_empty(&npinfo->rx_np)) {
1256                 spin_lock_irqsave(&npinfo->rx_lock, flags);
1257                 list_del(&np->rx);
1258                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
1259         }
1260
1261         synchronize_srcu(&netpoll_srcu);
1262
1263         if (atomic_dec_and_test(&npinfo->refcnt)) {
1264                 const struct net_device_ops *ops;
1265
1266                 ops = np->dev->netdev_ops;
1267                 if (ops->ndo_netpoll_cleanup)
1268                         ops->ndo_netpoll_cleanup(np->dev);
1269
1270                 rcu_assign_pointer(np->dev->npinfo, NULL);
1271                 call_rcu_bh(&npinfo->rcu, rcu_cleanup_netpoll_info);
1272         }
1273 }
1274 EXPORT_SYMBOL_GPL(__netpoll_cleanup);
1275
1276 static void netpoll_async_cleanup(struct work_struct *work)
1277 {
1278         struct netpoll *np = container_of(work, struct netpoll, cleanup_work);
1279
1280         rtnl_lock();
1281         __netpoll_cleanup(np);
1282         rtnl_unlock();
1283         kfree(np);
1284 }
1285
1286 void __netpoll_free_async(struct netpoll *np)
1287 {
1288         schedule_work(&np->cleanup_work);
1289 }
1290 EXPORT_SYMBOL_GPL(__netpoll_free_async);
1291
1292 void netpoll_cleanup(struct netpoll *np)
1293 {
1294         rtnl_lock();
1295         if (!np->dev)
1296                 goto out;
1297         __netpoll_cleanup(np);
1298         dev_put(np->dev);
1299         np->dev = NULL;
1300 out:
1301         rtnl_unlock();
1302 }
1303 EXPORT_SYMBOL(netpoll_cleanup);
1304
1305 int netpoll_trap(void)
1306 {
1307         return atomic_read(&trapped);
1308 }
1309 EXPORT_SYMBOL(netpoll_trap);
1310
1311 void netpoll_set_trap(int trap)
1312 {
1313         if (trap)
1314                 atomic_inc(&trapped);
1315         else
1316                 atomic_dec(&trapped);
1317 }
1318 EXPORT_SYMBOL(netpoll_set_trap);