netpoll: add generic support for bridge and bonding devices
[linux-2.6.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 #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 #define MAX_QUEUE_DEPTH (MAX_SKBS / 2)
39
40 static struct sk_buff_head skb_pool;
41
42 static atomic_t trapped;
43
44 #define USEC_PER_POLL   50
45 #define NETPOLL_RX_ENABLED  1
46 #define NETPOLL_RX_DROP     2
47
48 #define MAX_SKB_SIZE \
49                 (MAX_UDP_CHUNK + sizeof(struct udphdr) + \
50                                 sizeof(struct iphdr) + sizeof(struct ethhdr))
51
52 static void zap_completion_queue(void);
53 static void arp_reply(struct sk_buff *skb);
54
55 static unsigned int carrier_timeout = 4;
56 module_param(carrier_timeout, uint, 0644);
57
58 static void queue_process(struct work_struct *work)
59 {
60         struct netpoll_info *npinfo =
61                 container_of(work, struct netpoll_info, tx_work.work);
62         struct sk_buff *skb;
63         unsigned long flags;
64
65         while ((skb = skb_dequeue(&npinfo->txq))) {
66                 struct net_device *dev = skb->dev;
67                 const struct net_device_ops *ops = dev->netdev_ops;
68                 struct netdev_queue *txq;
69
70                 if (!netif_device_present(dev) || !netif_running(dev)) {
71                         __kfree_skb(skb);
72                         continue;
73                 }
74
75                 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
76
77                 local_irq_save(flags);
78                 __netif_tx_lock(txq, smp_processor_id());
79                 if (netif_tx_queue_stopped(txq) ||
80                     netif_tx_queue_frozen(txq) ||
81                     ops->ndo_start_xmit(skb, dev) != NETDEV_TX_OK) {
82                         skb_queue_head(&npinfo->txq, skb);
83                         __netif_tx_unlock(txq);
84                         local_irq_restore(flags);
85
86                         schedule_delayed_work(&npinfo->tx_work, HZ/10);
87                         return;
88                 }
89                 __netif_tx_unlock(txq);
90                 local_irq_restore(flags);
91         }
92 }
93
94 static __sum16 checksum_udp(struct sk_buff *skb, struct udphdr *uh,
95                             unsigned short ulen, __be32 saddr, __be32 daddr)
96 {
97         __wsum psum;
98
99         if (uh->check == 0 || skb_csum_unnecessary(skb))
100                 return 0;
101
102         psum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0);
103
104         if (skb->ip_summed == CHECKSUM_COMPLETE &&
105             !csum_fold(csum_add(psum, skb->csum)))
106                 return 0;
107
108         skb->csum = psum;
109
110         return __skb_checksum_complete(skb);
111 }
112
113 /*
114  * Check whether delayed processing was scheduled for our NIC. If so,
115  * we attempt to grab the poll lock and use ->poll() to pump the card.
116  * If this fails, either we've recursed in ->poll() or it's already
117  * running on another CPU.
118  *
119  * Note: we don't mask interrupts with this lock because we're using
120  * trylock here and interrupts are already disabled in the softirq
121  * case. Further, we test the poll_owner to avoid recursion on UP
122  * systems where the lock doesn't exist.
123  *
124  * In cases where there is bi-directional communications, reading only
125  * one message at a time can lead to packets being dropped by the
126  * network adapter, forcing superfluous retries and possibly timeouts.
127  * Thus, we set our budget to greater than 1.
128  */
129 static int poll_one_napi(struct netpoll_info *npinfo,
130                          struct napi_struct *napi, int budget)
131 {
132         int work;
133
134         /* net_rx_action's ->poll() invocations and our's are
135          * synchronized by this test which is only made while
136          * holding the napi->poll_lock.
137          */
138         if (!test_bit(NAPI_STATE_SCHED, &napi->state))
139                 return budget;
140
141         npinfo->rx_flags |= NETPOLL_RX_DROP;
142         atomic_inc(&trapped);
143         set_bit(NAPI_STATE_NPSVC, &napi->state);
144
145         work = napi->poll(napi, budget);
146         trace_napi_poll(napi);
147
148         clear_bit(NAPI_STATE_NPSVC, &napi->state);
149         atomic_dec(&trapped);
150         npinfo->rx_flags &= ~NETPOLL_RX_DROP;
151
152         return budget - work;
153 }
154
155 static void poll_napi(struct net_device *dev)
156 {
157         struct napi_struct *napi;
158         int budget = 16;
159
160         list_for_each_entry(napi, &dev->napi_list, dev_list) {
161                 if (napi->poll_owner != smp_processor_id() &&
162                     spin_trylock(&napi->poll_lock)) {
163                         budget = poll_one_napi(dev->npinfo, napi, budget);
164                         spin_unlock(&napi->poll_lock);
165
166                         if (!budget)
167                                 break;
168                 }
169         }
170 }
171
172 static void service_arp_queue(struct netpoll_info *npi)
173 {
174         if (npi) {
175                 struct sk_buff *skb;
176
177                 while ((skb = skb_dequeue(&npi->arp_tx)))
178                         arp_reply(skb);
179         }
180 }
181
182 void netpoll_poll_dev(struct net_device *dev)
183 {
184         const struct net_device_ops *ops;
185
186         if (!dev || !netif_running(dev))
187                 return;
188
189         ops = dev->netdev_ops;
190         if (!ops->ndo_poll_controller)
191                 return;
192
193         /* Process pending work on NIC */
194         ops->ndo_poll_controller(dev);
195
196         poll_napi(dev);
197
198         service_arp_queue(dev->npinfo);
199
200         zap_completion_queue();
201 }
202
203 void netpoll_poll(struct netpoll *np)
204 {
205         netpoll_poll_dev(np->dev);
206 }
207
208 static void refill_skbs(void)
209 {
210         struct sk_buff *skb;
211         unsigned long flags;
212
213         spin_lock_irqsave(&skb_pool.lock, flags);
214         while (skb_pool.qlen < MAX_SKBS) {
215                 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
216                 if (!skb)
217                         break;
218
219                 __skb_queue_tail(&skb_pool, skb);
220         }
221         spin_unlock_irqrestore(&skb_pool.lock, flags);
222 }
223
224 static void zap_completion_queue(void)
225 {
226         unsigned long flags;
227         struct softnet_data *sd = &get_cpu_var(softnet_data);
228
229         if (sd->completion_queue) {
230                 struct sk_buff *clist;
231
232                 local_irq_save(flags);
233                 clist = sd->completion_queue;
234                 sd->completion_queue = NULL;
235                 local_irq_restore(flags);
236
237                 while (clist != NULL) {
238                         struct sk_buff *skb = clist;
239                         clist = clist->next;
240                         if (skb->destructor) {
241                                 atomic_inc(&skb->users);
242                                 dev_kfree_skb_any(skb); /* put this one back */
243                         } else {
244                                 __kfree_skb(skb);
245                         }
246                 }
247         }
248
249         put_cpu_var(softnet_data);
250 }
251
252 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
253 {
254         int count = 0;
255         struct sk_buff *skb;
256
257         zap_completion_queue();
258         refill_skbs();
259 repeat:
260
261         skb = alloc_skb(len, GFP_ATOMIC);
262         if (!skb)
263                 skb = skb_dequeue(&skb_pool);
264
265         if (!skb) {
266                 if (++count < 10) {
267                         netpoll_poll(np);
268                         goto repeat;
269                 }
270                 return NULL;
271         }
272
273         atomic_set(&skb->users, 1);
274         skb_reserve(skb, reserve);
275         return skb;
276 }
277
278 static int netpoll_owner_active(struct net_device *dev)
279 {
280         struct napi_struct *napi;
281
282         list_for_each_entry(napi, &dev->napi_list, dev_list) {
283                 if (napi->poll_owner == smp_processor_id())
284                         return 1;
285         }
286         return 0;
287 }
288
289 void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
290 {
291         int status = NETDEV_TX_BUSY;
292         unsigned long tries;
293         struct net_device *dev = np->dev;
294         const struct net_device_ops *ops = dev->netdev_ops;
295         struct netpoll_info *npinfo = np->dev->npinfo;
296
297         if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
298                 __kfree_skb(skb);
299                 return;
300         }
301
302         /* don't get messages out of order, and no recursion */
303         if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
304                 struct netdev_queue *txq;
305                 unsigned long flags;
306
307                 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
308
309                 local_irq_save(flags);
310                 /* try until next clock tick */
311                 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
312                      tries > 0; --tries) {
313                         if (__netif_tx_trylock(txq)) {
314                                 if (!netif_tx_queue_stopped(txq)) {
315                                         dev->priv_flags |= IFF_IN_NETPOLL;
316                                         status = ops->ndo_start_xmit(skb, dev);
317                                         dev->priv_flags &= ~IFF_IN_NETPOLL;
318                                         if (status == NETDEV_TX_OK)
319                                                 txq_trans_update(txq);
320                                 }
321                                 __netif_tx_unlock(txq);
322
323                                 if (status == NETDEV_TX_OK)
324                                         break;
325
326                         }
327
328                         /* tickle device maybe there is some cleanup */
329                         netpoll_poll(np);
330
331                         udelay(USEC_PER_POLL);
332                 }
333
334                 WARN_ONCE(!irqs_disabled(),
335                         "netpoll_send_skb(): %s enabled interrupts in poll (%pF)\n",
336                         dev->name, ops->ndo_start_xmit);
337
338                 local_irq_restore(flags);
339         }
340
341         if (status != NETDEV_TX_OK) {
342                 skb_queue_tail(&npinfo->txq, skb);
343                 schedule_delayed_work(&npinfo->tx_work,0);
344         }
345 }
346
347 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
348 {
349         int total_len, eth_len, ip_len, udp_len;
350         struct sk_buff *skb;
351         struct udphdr *udph;
352         struct iphdr *iph;
353         struct ethhdr *eth;
354
355         udp_len = len + sizeof(*udph);
356         ip_len = eth_len = udp_len + sizeof(*iph);
357         total_len = eth_len + ETH_HLEN + NET_IP_ALIGN;
358
359         skb = find_skb(np, total_len, total_len - len);
360         if (!skb)
361                 return;
362
363         skb_copy_to_linear_data(skb, msg, len);
364         skb->len += len;
365
366         skb_push(skb, sizeof(*udph));
367         skb_reset_transport_header(skb);
368         udph = udp_hdr(skb);
369         udph->source = htons(np->local_port);
370         udph->dest = htons(np->remote_port);
371         udph->len = htons(udp_len);
372         udph->check = 0;
373         udph->check = csum_tcpudp_magic(np->local_ip,
374                                         np->remote_ip,
375                                         udp_len, IPPROTO_UDP,
376                                         csum_partial(udph, udp_len, 0));
377         if (udph->check == 0)
378                 udph->check = CSUM_MANGLED_0;
379
380         skb_push(skb, sizeof(*iph));
381         skb_reset_network_header(skb);
382         iph = ip_hdr(skb);
383
384         /* iph->version = 4; iph->ihl = 5; */
385         put_unaligned(0x45, (unsigned char *)iph);
386         iph->tos      = 0;
387         put_unaligned(htons(ip_len), &(iph->tot_len));
388         iph->id       = 0;
389         iph->frag_off = 0;
390         iph->ttl      = 64;
391         iph->protocol = IPPROTO_UDP;
392         iph->check    = 0;
393         put_unaligned(np->local_ip, &(iph->saddr));
394         put_unaligned(np->remote_ip, &(iph->daddr));
395         iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
396
397         eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
398         skb_reset_mac_header(skb);
399         skb->protocol = eth->h_proto = htons(ETH_P_IP);
400         memcpy(eth->h_source, np->dev->dev_addr, ETH_ALEN);
401         memcpy(eth->h_dest, np->remote_mac, ETH_ALEN);
402
403         skb->dev = np->dev;
404
405         netpoll_send_skb(np, skb);
406 }
407
408 static void arp_reply(struct sk_buff *skb)
409 {
410         struct netpoll_info *npinfo = skb->dev->npinfo;
411         struct arphdr *arp;
412         unsigned char *arp_ptr;
413         int size, type = ARPOP_REPLY, ptype = ETH_P_ARP;
414         __be32 sip, tip;
415         unsigned char *sha;
416         struct sk_buff *send_skb;
417         struct netpoll *np, *tmp;
418         unsigned long flags;
419         int hits = 0;
420
421         if (list_empty(&npinfo->rx_np))
422                 return;
423
424         /* Before checking the packet, we do some early
425            inspection whether this is interesting at all */
426         spin_lock_irqsave(&npinfo->rx_lock, flags);
427         list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
428                 if (np->dev == skb->dev)
429                         hits++;
430         }
431         spin_unlock_irqrestore(&npinfo->rx_lock, flags);
432
433         /* No netpoll struct is using this dev */
434         if (!hits)
435                 return;
436
437         /* No arp on this interface */
438         if (skb->dev->flags & IFF_NOARP)
439                 return;
440
441         if (!pskb_may_pull(skb, arp_hdr_len(skb->dev)))
442                 return;
443
444         skb_reset_network_header(skb);
445         skb_reset_transport_header(skb);
446         arp = arp_hdr(skb);
447
448         if ((arp->ar_hrd != htons(ARPHRD_ETHER) &&
449              arp->ar_hrd != htons(ARPHRD_IEEE802)) ||
450             arp->ar_pro != htons(ETH_P_IP) ||
451             arp->ar_op != htons(ARPOP_REQUEST))
452                 return;
453
454         arp_ptr = (unsigned char *)(arp+1);
455         /* save the location of the src hw addr */
456         sha = arp_ptr;
457         arp_ptr += skb->dev->addr_len;
458         memcpy(&sip, arp_ptr, 4);
459         arp_ptr += 4;
460         /* If we actually cared about dst hw addr,
461            it would get copied here */
462         arp_ptr += skb->dev->addr_len;
463         memcpy(&tip, arp_ptr, 4);
464
465         /* Should we ignore arp? */
466         if (ipv4_is_loopback(tip) || ipv4_is_multicast(tip))
467                 return;
468
469         size = arp_hdr_len(skb->dev);
470
471         spin_lock_irqsave(&npinfo->rx_lock, flags);
472         list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
473                 if (tip != np->local_ip)
474                         continue;
475
476                 send_skb = find_skb(np, size + LL_ALLOCATED_SPACE(np->dev),
477                                     LL_RESERVED_SPACE(np->dev));
478                 if (!send_skb)
479                         continue;
480
481                 skb_reset_network_header(send_skb);
482                 arp = (struct arphdr *) skb_put(send_skb, size);
483                 send_skb->dev = skb->dev;
484                 send_skb->protocol = htons(ETH_P_ARP);
485
486                 /* Fill the device header for the ARP frame */
487                 if (dev_hard_header(send_skb, skb->dev, ptype,
488                                     sha, np->dev->dev_addr,
489                                     send_skb->len) < 0) {
490                         kfree_skb(send_skb);
491                         continue;
492                 }
493
494                 /*
495                  * Fill out the arp protocol part.
496                  *
497                  * we only support ethernet device type,
498                  * which (according to RFC 1390) should
499                  * always equal 1 (Ethernet).
500                  */
501
502                 arp->ar_hrd = htons(np->dev->type);
503                 arp->ar_pro = htons(ETH_P_IP);
504                 arp->ar_hln = np->dev->addr_len;
505                 arp->ar_pln = 4;
506                 arp->ar_op = htons(type);
507
508                 arp_ptr = (unsigned char *)(arp + 1);
509                 memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
510                 arp_ptr += np->dev->addr_len;
511                 memcpy(arp_ptr, &tip, 4);
512                 arp_ptr += 4;
513                 memcpy(arp_ptr, sha, np->dev->addr_len);
514                 arp_ptr += np->dev->addr_len;
515                 memcpy(arp_ptr, &sip, 4);
516
517                 netpoll_send_skb(np, send_skb);
518
519                 /* If there are several rx_hooks for the same address,
520                    we're fine by sending a single reply */
521                 break;
522         }
523         spin_unlock_irqrestore(&npinfo->rx_lock, flags);
524 }
525
526 int __netpoll_rx(struct sk_buff *skb)
527 {
528         int proto, len, ulen;
529         int hits = 0;
530         struct iphdr *iph;
531         struct udphdr *uh;
532         struct netpoll_info *npinfo = skb->dev->npinfo;
533         struct netpoll *np, *tmp;
534
535         if (list_empty(&npinfo->rx_np))
536                 goto out;
537
538         if (skb->dev->type != ARPHRD_ETHER)
539                 goto out;
540
541         /* check if netpoll clients need ARP */
542         if (skb->protocol == htons(ETH_P_ARP) &&
543             atomic_read(&trapped)) {
544                 skb_queue_tail(&npinfo->arp_tx, skb);
545                 return 1;
546         }
547
548         proto = ntohs(eth_hdr(skb)->h_proto);
549         if (proto != ETH_P_IP)
550                 goto out;
551         if (skb->pkt_type == PACKET_OTHERHOST)
552                 goto out;
553         if (skb_shared(skb))
554                 goto out;
555
556         iph = (struct iphdr *)skb->data;
557         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
558                 goto out;
559         if (iph->ihl < 5 || iph->version != 4)
560                 goto out;
561         if (!pskb_may_pull(skb, iph->ihl*4))
562                 goto out;
563         if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
564                 goto out;
565
566         len = ntohs(iph->tot_len);
567         if (skb->len < len || len < iph->ihl*4)
568                 goto out;
569
570         /*
571          * Our transport medium may have padded the buffer out.
572          * Now We trim to the true length of the frame.
573          */
574         if (pskb_trim_rcsum(skb, len))
575                 goto out;
576
577         if (iph->protocol != IPPROTO_UDP)
578                 goto out;
579
580         len -= iph->ihl*4;
581         uh = (struct udphdr *)(((char *)iph) + iph->ihl*4);
582         ulen = ntohs(uh->len);
583
584         if (ulen != len)
585                 goto out;
586         if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr))
587                 goto out;
588
589         list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
590                 if (np->local_ip && np->local_ip != iph->daddr)
591                         continue;
592                 if (np->remote_ip && np->remote_ip != iph->saddr)
593                         continue;
594                 if (np->local_port && np->local_port != ntohs(uh->dest))
595                         continue;
596
597                 np->rx_hook(np, ntohs(uh->source),
598                                (char *)(uh+1),
599                                ulen - sizeof(struct udphdr));
600                 hits++;
601         }
602
603         if (!hits)
604                 goto out;
605
606         kfree_skb(skb);
607         return 1;
608
609 out:
610         if (atomic_read(&trapped)) {
611                 kfree_skb(skb);
612                 return 1;
613         }
614
615         return 0;
616 }
617
618 void netpoll_print_options(struct netpoll *np)
619 {
620         printk(KERN_INFO "%s: local port %d\n",
621                          np->name, np->local_port);
622         printk(KERN_INFO "%s: local IP %pI4\n",
623                          np->name, &np->local_ip);
624         printk(KERN_INFO "%s: interface '%s'\n",
625                          np->name, np->dev_name);
626         printk(KERN_INFO "%s: remote port %d\n",
627                          np->name, np->remote_port);
628         printk(KERN_INFO "%s: remote IP %pI4\n",
629                          np->name, &np->remote_ip);
630         printk(KERN_INFO "%s: remote ethernet address %pM\n",
631                          np->name, np->remote_mac);
632 }
633
634 int netpoll_parse_options(struct netpoll *np, char *opt)
635 {
636         char *cur=opt, *delim;
637
638         if (*cur != '@') {
639                 if ((delim = strchr(cur, '@')) == NULL)
640                         goto parse_failed;
641                 *delim = 0;
642                 np->local_port = simple_strtol(cur, NULL, 10);
643                 cur = delim;
644         }
645         cur++;
646
647         if (*cur != '/') {
648                 if ((delim = strchr(cur, '/')) == NULL)
649                         goto parse_failed;
650                 *delim = 0;
651                 np->local_ip = in_aton(cur);
652                 cur = delim;
653         }
654         cur++;
655
656         if (*cur != ',') {
657                 /* parse out dev name */
658                 if ((delim = strchr(cur, ',')) == NULL)
659                         goto parse_failed;
660                 *delim = 0;
661                 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
662                 cur = delim;
663         }
664         cur++;
665
666         if (*cur != '@') {
667                 /* dst port */
668                 if ((delim = strchr(cur, '@')) == NULL)
669                         goto parse_failed;
670                 *delim = 0;
671                 if (*cur == ' ' || *cur == '\t')
672                         printk(KERN_INFO "%s: warning: whitespace"
673                                         "is not allowed\n", np->name);
674                 np->remote_port = simple_strtol(cur, NULL, 10);
675                 cur = delim;
676         }
677         cur++;
678
679         /* dst ip */
680         if ((delim = strchr(cur, '/')) == NULL)
681                 goto parse_failed;
682         *delim = 0;
683         np->remote_ip = in_aton(cur);
684         cur = delim + 1;
685
686         if (*cur != 0) {
687                 /* MAC address */
688                 if ((delim = strchr(cur, ':')) == NULL)
689                         goto parse_failed;
690                 *delim = 0;
691                 np->remote_mac[0] = simple_strtol(cur, NULL, 16);
692                 cur = delim + 1;
693                 if ((delim = strchr(cur, ':')) == NULL)
694                         goto parse_failed;
695                 *delim = 0;
696                 np->remote_mac[1] = simple_strtol(cur, NULL, 16);
697                 cur = delim + 1;
698                 if ((delim = strchr(cur, ':')) == NULL)
699                         goto parse_failed;
700                 *delim = 0;
701                 np->remote_mac[2] = simple_strtol(cur, NULL, 16);
702                 cur = delim + 1;
703                 if ((delim = strchr(cur, ':')) == NULL)
704                         goto parse_failed;
705                 *delim = 0;
706                 np->remote_mac[3] = simple_strtol(cur, NULL, 16);
707                 cur = delim + 1;
708                 if ((delim = strchr(cur, ':')) == NULL)
709                         goto parse_failed;
710                 *delim = 0;
711                 np->remote_mac[4] = simple_strtol(cur, NULL, 16);
712                 cur = delim + 1;
713                 np->remote_mac[5] = simple_strtol(cur, NULL, 16);
714         }
715
716         netpoll_print_options(np);
717
718         return 0;
719
720  parse_failed:
721         printk(KERN_INFO "%s: couldn't parse config at '%s'!\n",
722                np->name, cur);
723         return -1;
724 }
725
726 int netpoll_setup(struct netpoll *np)
727 {
728         struct net_device *ndev = NULL;
729         struct in_device *in_dev;
730         struct netpoll_info *npinfo;
731         struct netpoll *npe, *tmp;
732         unsigned long flags;
733         int err;
734
735         if (np->dev_name)
736                 ndev = dev_get_by_name(&init_net, np->dev_name);
737         if (!ndev) {
738                 printk(KERN_ERR "%s: %s doesn't exist, aborting.\n",
739                        np->name, np->dev_name);
740                 return -ENODEV;
741         }
742
743         np->dev = ndev;
744         if (!ndev->npinfo) {
745                 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
746                 if (!npinfo) {
747                         err = -ENOMEM;
748                         goto put;
749                 }
750
751                 npinfo->rx_flags = 0;
752                 INIT_LIST_HEAD(&npinfo->rx_np);
753
754                 spin_lock_init(&npinfo->rx_lock);
755                 skb_queue_head_init(&npinfo->arp_tx);
756                 skb_queue_head_init(&npinfo->txq);
757                 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
758
759                 atomic_set(&npinfo->refcnt, 1);
760         } else {
761                 npinfo = ndev->npinfo;
762                 atomic_inc(&npinfo->refcnt);
763         }
764
765         npinfo->netpoll = np;
766
767         if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) ||
768             !ndev->netdev_ops->ndo_poll_controller) {
769                 printk(KERN_ERR "%s: %s doesn't support polling, aborting.\n",
770                        np->name, np->dev_name);
771                 err = -ENOTSUPP;
772                 goto release;
773         }
774
775         if (!netif_running(ndev)) {
776                 unsigned long atmost, atleast;
777
778                 printk(KERN_INFO "%s: device %s not up yet, forcing it\n",
779                        np->name, np->dev_name);
780
781                 rtnl_lock();
782                 err = dev_open(ndev);
783                 rtnl_unlock();
784
785                 if (err) {
786                         printk(KERN_ERR "%s: failed to open %s\n",
787                                np->name, ndev->name);
788                         goto release;
789                 }
790
791                 atleast = jiffies + HZ/10;
792                 atmost = jiffies + carrier_timeout * HZ;
793                 while (!netif_carrier_ok(ndev)) {
794                         if (time_after(jiffies, atmost)) {
795                                 printk(KERN_NOTICE
796                                        "%s: timeout waiting for carrier\n",
797                                        np->name);
798                                 break;
799                         }
800                         msleep(1);
801                 }
802
803                 /* If carrier appears to come up instantly, we don't
804                  * trust it and pause so that we don't pump all our
805                  * queued console messages into the bitbucket.
806                  */
807
808                 if (time_before(jiffies, atleast)) {
809                         printk(KERN_NOTICE "%s: carrier detect appears"
810                                " untrustworthy, waiting 4 seconds\n",
811                                np->name);
812                         msleep(4000);
813                 }
814         }
815
816         if (!np->local_ip) {
817                 rcu_read_lock();
818                 in_dev = __in_dev_get_rcu(ndev);
819
820                 if (!in_dev || !in_dev->ifa_list) {
821                         rcu_read_unlock();
822                         printk(KERN_ERR "%s: no IP address for %s, aborting\n",
823                                np->name, np->dev_name);
824                         err = -EDESTADDRREQ;
825                         goto release;
826                 }
827
828                 np->local_ip = in_dev->ifa_list->ifa_local;
829                 rcu_read_unlock();
830                 printk(KERN_INFO "%s: local IP %pI4\n", np->name, &np->local_ip);
831         }
832
833         if (np->rx_hook) {
834                 spin_lock_irqsave(&npinfo->rx_lock, flags);
835                 npinfo->rx_flags |= NETPOLL_RX_ENABLED;
836                 list_add_tail(&np->rx, &npinfo->rx_np);
837                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
838         }
839
840         /* fill up the skb queue */
841         refill_skbs();
842
843         /* last thing to do is link it to the net device structure */
844         ndev->npinfo = npinfo;
845
846         /* avoid racing with NAPI reading npinfo */
847         synchronize_rcu();
848
849         return 0;
850
851  release:
852         if (!ndev->npinfo) {
853                 spin_lock_irqsave(&npinfo->rx_lock, flags);
854                 list_for_each_entry_safe(npe, tmp, &npinfo->rx_np, rx) {
855                         npe->dev = NULL;
856                 }
857                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
858
859                 kfree(npinfo);
860         }
861 put:
862         dev_put(ndev);
863         return err;
864 }
865
866 static int __init netpoll_init(void)
867 {
868         skb_queue_head_init(&skb_pool);
869         return 0;
870 }
871 core_initcall(netpoll_init);
872
873 void netpoll_cleanup(struct netpoll *np)
874 {
875         struct netpoll_info *npinfo;
876         unsigned long flags;
877
878         if (np->dev) {
879                 npinfo = np->dev->npinfo;
880                 if (npinfo) {
881                         if (!list_empty(&npinfo->rx_np)) {
882                                 spin_lock_irqsave(&npinfo->rx_lock, flags);
883                                 list_del(&np->rx);
884                                 if (list_empty(&npinfo->rx_np))
885                                         npinfo->rx_flags &= ~NETPOLL_RX_ENABLED;
886                                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
887                         }
888
889                         if (atomic_dec_and_test(&npinfo->refcnt)) {
890                                 const struct net_device_ops *ops;
891                                 skb_queue_purge(&npinfo->arp_tx);
892                                 skb_queue_purge(&npinfo->txq);
893                                 cancel_rearming_delayed_work(&npinfo->tx_work);
894
895                                 /* clean after last, unfinished work */
896                                 __skb_queue_purge(&npinfo->txq);
897                                 kfree(npinfo);
898                                 ops = np->dev->netdev_ops;
899                                 if (ops->ndo_netpoll_cleanup)
900                                         ops->ndo_netpoll_cleanup(np->dev);
901                                 else
902                                         np->dev->npinfo = NULL;
903                         }
904                 }
905
906                 dev_put(np->dev);
907         }
908
909         np->dev = NULL;
910 }
911
912 int netpoll_trap(void)
913 {
914         return atomic_read(&trapped);
915 }
916
917 void netpoll_set_trap(int trap)
918 {
919         if (trap)
920                 atomic_inc(&trapped);
921         else
922                 atomic_dec(&trapped);
923 }
924
925 EXPORT_SYMBOL(netpoll_send_skb);
926 EXPORT_SYMBOL(netpoll_set_trap);
927 EXPORT_SYMBOL(netpoll_trap);
928 EXPORT_SYMBOL(netpoll_print_options);
929 EXPORT_SYMBOL(netpoll_parse_options);
930 EXPORT_SYMBOL(netpoll_setup);
931 EXPORT_SYMBOL(netpoll_cleanup);
932 EXPORT_SYMBOL(netpoll_send_udp);
933 EXPORT_SYMBOL(netpoll_poll_dev);
934 EXPORT_SYMBOL(netpoll_poll);