Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc-2.6
[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 <net/tcp.h>
26 #include <net/udp.h>
27 #include <asm/unaligned.h>
28 #include <trace/events/napi.h>
29
30 /*
31  * We maintain a small pool of fully-sized skbs, to make sure the
32  * message gets out even in extreme OOM situations.
33  */
34
35 #define MAX_UDP_CHUNK 1460
36 #define MAX_SKBS 32
37 #define MAX_QUEUE_DEPTH (MAX_SKBS / 2)
38
39 static struct sk_buff_head skb_pool;
40
41 static atomic_t trapped;
42
43 #define USEC_PER_POLL   50
44 #define NETPOLL_RX_ENABLED  1
45 #define NETPOLL_RX_DROP     2
46
47 #define MAX_SKB_SIZE \
48                 (MAX_UDP_CHUNK + sizeof(struct udphdr) + \
49                                 sizeof(struct iphdr) + sizeof(struct ethhdr))
50
51 static void zap_completion_queue(void);
52 static void arp_reply(struct sk_buff *skb);
53
54 static unsigned int carrier_timeout = 4;
55 module_param(carrier_timeout, uint, 0644);
56
57 static void queue_process(struct work_struct *work)
58 {
59         struct netpoll_info *npinfo =
60                 container_of(work, struct netpoll_info, tx_work.work);
61         struct sk_buff *skb;
62         unsigned long flags;
63
64         while ((skb = skb_dequeue(&npinfo->txq))) {
65                 struct net_device *dev = skb->dev;
66                 const struct net_device_ops *ops = dev->netdev_ops;
67                 struct netdev_queue *txq;
68
69                 if (!netif_device_present(dev) || !netif_running(dev)) {
70                         __kfree_skb(skb);
71                         continue;
72                 }
73
74                 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
75
76                 local_irq_save(flags);
77                 __netif_tx_lock(txq, smp_processor_id());
78                 if (netif_tx_queue_stopped(txq) ||
79                     netif_tx_queue_frozen(txq) ||
80                     ops->ndo_start_xmit(skb, dev) != NETDEV_TX_OK) {
81                         skb_queue_head(&npinfo->txq, skb);
82                         __netif_tx_unlock(txq);
83                         local_irq_restore(flags);
84
85                         schedule_delayed_work(&npinfo->tx_work, HZ/10);
86                         return;
87                 }
88                 __netif_tx_unlock(txq);
89                 local_irq_restore(flags);
90         }
91 }
92
93 static __sum16 checksum_udp(struct sk_buff *skb, struct udphdr *uh,
94                             unsigned short ulen, __be32 saddr, __be32 daddr)
95 {
96         __wsum psum;
97
98         if (uh->check == 0 || skb_csum_unnecessary(skb))
99                 return 0;
100
101         psum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0);
102
103         if (skb->ip_summed == CHECKSUM_COMPLETE &&
104             !csum_fold(csum_add(psum, skb->csum)))
105                 return 0;
106
107         skb->csum = psum;
108
109         return __skb_checksum_complete(skb);
110 }
111
112 /*
113  * Check whether delayed processing was scheduled for our NIC. If so,
114  * we attempt to grab the poll lock and use ->poll() to pump the card.
115  * If this fails, either we've recursed in ->poll() or it's already
116  * running on another CPU.
117  *
118  * Note: we don't mask interrupts with this lock because we're using
119  * trylock here and interrupts are already disabled in the softirq
120  * case. Further, we test the poll_owner to avoid recursion on UP
121  * systems where the lock doesn't exist.
122  *
123  * In cases where there is bi-directional communications, reading only
124  * one message at a time can lead to packets being dropped by the
125  * network adapter, forcing superfluous retries and possibly timeouts.
126  * Thus, we set our budget to greater than 1.
127  */
128 static int poll_one_napi(struct netpoll_info *npinfo,
129                          struct napi_struct *napi, int budget)
130 {
131         int work;
132
133         /* net_rx_action's ->poll() invocations and our's are
134          * synchronized by this test which is only made while
135          * holding the napi->poll_lock.
136          */
137         if (!test_bit(NAPI_STATE_SCHED, &napi->state))
138                 return budget;
139
140         npinfo->rx_flags |= NETPOLL_RX_DROP;
141         atomic_inc(&trapped);
142         set_bit(NAPI_STATE_NPSVC, &napi->state);
143
144         work = napi->poll(napi, budget);
145         trace_napi_poll(napi);
146
147         clear_bit(NAPI_STATE_NPSVC, &napi->state);
148         atomic_dec(&trapped);
149         npinfo->rx_flags &= ~NETPOLL_RX_DROP;
150
151         return budget - work;
152 }
153
154 static void poll_napi(struct net_device *dev)
155 {
156         struct napi_struct *napi;
157         int budget = 16;
158
159         list_for_each_entry(napi, &dev->napi_list, dev_list) {
160                 if (napi->poll_owner != smp_processor_id() &&
161                     spin_trylock(&napi->poll_lock)) {
162                         budget = poll_one_napi(dev->npinfo, napi, budget);
163                         spin_unlock(&napi->poll_lock);
164
165                         if (!budget)
166                                 break;
167                 }
168         }
169 }
170
171 static void service_arp_queue(struct netpoll_info *npi)
172 {
173         if (npi) {
174                 struct sk_buff *skb;
175
176                 while ((skb = skb_dequeue(&npi->arp_tx)))
177                         arp_reply(skb);
178         }
179 }
180
181 void netpoll_poll(struct netpoll *np)
182 {
183         struct net_device *dev = np->dev;
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 static void refill_skbs(void)
204 {
205         struct sk_buff *skb;
206         unsigned long flags;
207
208         spin_lock_irqsave(&skb_pool.lock, flags);
209         while (skb_pool.qlen < MAX_SKBS) {
210                 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
211                 if (!skb)
212                         break;
213
214                 __skb_queue_tail(&skb_pool, skb);
215         }
216         spin_unlock_irqrestore(&skb_pool.lock, flags);
217 }
218
219 static void zap_completion_queue(void)
220 {
221         unsigned long flags;
222         struct softnet_data *sd = &get_cpu_var(softnet_data);
223
224         if (sd->completion_queue) {
225                 struct sk_buff *clist;
226
227                 local_irq_save(flags);
228                 clist = sd->completion_queue;
229                 sd->completion_queue = NULL;
230                 local_irq_restore(flags);
231
232                 while (clist != NULL) {
233                         struct sk_buff *skb = clist;
234                         clist = clist->next;
235                         if (skb->destructor) {
236                                 atomic_inc(&skb->users);
237                                 dev_kfree_skb_any(skb); /* put this one back */
238                         } else {
239                                 __kfree_skb(skb);
240                         }
241                 }
242         }
243
244         put_cpu_var(softnet_data);
245 }
246
247 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
248 {
249         int count = 0;
250         struct sk_buff *skb;
251
252         zap_completion_queue();
253         refill_skbs();
254 repeat:
255
256         skb = alloc_skb(len, GFP_ATOMIC);
257         if (!skb)
258                 skb = skb_dequeue(&skb_pool);
259
260         if (!skb) {
261                 if (++count < 10) {
262                         netpoll_poll(np);
263                         goto repeat;
264                 }
265                 return NULL;
266         }
267
268         atomic_set(&skb->users, 1);
269         skb_reserve(skb, reserve);
270         return skb;
271 }
272
273 static int netpoll_owner_active(struct net_device *dev)
274 {
275         struct napi_struct *napi;
276
277         list_for_each_entry(napi, &dev->napi_list, dev_list) {
278                 if (napi->poll_owner == smp_processor_id())
279                         return 1;
280         }
281         return 0;
282 }
283
284 static void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
285 {
286         int status = NETDEV_TX_BUSY;
287         unsigned long tries;
288         struct net_device *dev = np->dev;
289         const struct net_device_ops *ops = dev->netdev_ops;
290         struct netpoll_info *npinfo = np->dev->npinfo;
291
292         if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
293                 __kfree_skb(skb);
294                 return;
295         }
296
297         /* don't get messages out of order, and no recursion */
298         if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
299                 struct netdev_queue *txq;
300                 unsigned long flags;
301
302                 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
303
304                 local_irq_save(flags);
305                 /* try until next clock tick */
306                 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
307                      tries > 0; --tries) {
308                         if (__netif_tx_trylock(txq)) {
309                                 if (!netif_tx_queue_stopped(txq)) {
310                                         status = ops->ndo_start_xmit(skb, dev);
311                                         if (status == NETDEV_TX_OK)
312                                                 txq_trans_update(txq);
313                                 }
314                                 __netif_tx_unlock(txq);
315
316                                 if (status == NETDEV_TX_OK)
317                                         break;
318
319                         }
320
321                         /* tickle device maybe there is some cleanup */
322                         netpoll_poll(np);
323
324                         udelay(USEC_PER_POLL);
325                 }
326
327                 WARN_ONCE(!irqs_disabled(),
328                         "netpoll_send_skb(): %s enabled interrupts in poll (%pF)\n",
329                         dev->name, ops->ndo_start_xmit);
330
331                 local_irq_restore(flags);
332         }
333
334         if (status != NETDEV_TX_OK) {
335                 skb_queue_tail(&npinfo->txq, skb);
336                 schedule_delayed_work(&npinfo->tx_work,0);
337         }
338 }
339
340 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
341 {
342         int total_len, eth_len, ip_len, udp_len;
343         struct sk_buff *skb;
344         struct udphdr *udph;
345         struct iphdr *iph;
346         struct ethhdr *eth;
347
348         udp_len = len + sizeof(*udph);
349         ip_len = eth_len = udp_len + sizeof(*iph);
350         total_len = eth_len + ETH_HLEN + NET_IP_ALIGN;
351
352         skb = find_skb(np, total_len, total_len - len);
353         if (!skb)
354                 return;
355
356         skb_copy_to_linear_data(skb, msg, len);
357         skb->len += len;
358
359         skb_push(skb, sizeof(*udph));
360         skb_reset_transport_header(skb);
361         udph = udp_hdr(skb);
362         udph->source = htons(np->local_port);
363         udph->dest = htons(np->remote_port);
364         udph->len = htons(udp_len);
365         udph->check = 0;
366         udph->check = csum_tcpudp_magic(np->local_ip,
367                                         np->remote_ip,
368                                         udp_len, IPPROTO_UDP,
369                                         csum_partial(udph, udp_len, 0));
370         if (udph->check == 0)
371                 udph->check = CSUM_MANGLED_0;
372
373         skb_push(skb, sizeof(*iph));
374         skb_reset_network_header(skb);
375         iph = ip_hdr(skb);
376
377         /* iph->version = 4; iph->ihl = 5; */
378         put_unaligned(0x45, (unsigned char *)iph);
379         iph->tos      = 0;
380         put_unaligned(htons(ip_len), &(iph->tot_len));
381         iph->id       = 0;
382         iph->frag_off = 0;
383         iph->ttl      = 64;
384         iph->protocol = IPPROTO_UDP;
385         iph->check    = 0;
386         put_unaligned(np->local_ip, &(iph->saddr));
387         put_unaligned(np->remote_ip, &(iph->daddr));
388         iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
389
390         eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
391         skb_reset_mac_header(skb);
392         skb->protocol = eth->h_proto = htons(ETH_P_IP);
393         memcpy(eth->h_source, np->dev->dev_addr, ETH_ALEN);
394         memcpy(eth->h_dest, np->remote_mac, ETH_ALEN);
395
396         skb->dev = np->dev;
397
398         netpoll_send_skb(np, skb);
399 }
400
401 static void arp_reply(struct sk_buff *skb)
402 {
403         struct netpoll_info *npinfo = skb->dev->npinfo;
404         struct arphdr *arp;
405         unsigned char *arp_ptr;
406         int size, type = ARPOP_REPLY, ptype = ETH_P_ARP;
407         __be32 sip, tip;
408         unsigned char *sha;
409         struct sk_buff *send_skb;
410         struct netpoll *np, *tmp;
411         unsigned long flags;
412         int hits = 0;
413
414         if (list_empty(&npinfo->rx_np))
415                 return;
416
417         /* Before checking the packet, we do some early
418            inspection whether this is interesting at all */
419         spin_lock_irqsave(&npinfo->rx_lock, flags);
420         list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
421                 if (np->dev == skb->dev)
422                         hits++;
423         }
424         spin_unlock_irqrestore(&npinfo->rx_lock, flags);
425
426         /* No netpoll struct is using this dev */
427         if (!hits)
428                 return;
429
430         /* No arp on this interface */
431         if (skb->dev->flags & IFF_NOARP)
432                 return;
433
434         if (!pskb_may_pull(skb, arp_hdr_len(skb->dev)))
435                 return;
436
437         skb_reset_network_header(skb);
438         skb_reset_transport_header(skb);
439         arp = arp_hdr(skb);
440
441         if ((arp->ar_hrd != htons(ARPHRD_ETHER) &&
442              arp->ar_hrd != htons(ARPHRD_IEEE802)) ||
443             arp->ar_pro != htons(ETH_P_IP) ||
444             arp->ar_op != htons(ARPOP_REQUEST))
445                 return;
446
447         arp_ptr = (unsigned char *)(arp+1);
448         /* save the location of the src hw addr */
449         sha = arp_ptr;
450         arp_ptr += skb->dev->addr_len;
451         memcpy(&sip, arp_ptr, 4);
452         arp_ptr += 4;
453         /* If we actually cared about dst hw addr,
454            it would get copied here */
455         arp_ptr += skb->dev->addr_len;
456         memcpy(&tip, arp_ptr, 4);
457
458         /* Should we ignore arp? */
459         if (ipv4_is_loopback(tip) || ipv4_is_multicast(tip))
460                 return;
461
462         size = arp_hdr_len(skb->dev);
463
464         spin_lock_irqsave(&npinfo->rx_lock, flags);
465         list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
466                 if (tip != np->local_ip)
467                         continue;
468
469                 send_skb = find_skb(np, size + LL_ALLOCATED_SPACE(np->dev),
470                                     LL_RESERVED_SPACE(np->dev));
471                 if (!send_skb)
472                         continue;
473
474                 skb_reset_network_header(send_skb);
475                 arp = (struct arphdr *) skb_put(send_skb, size);
476                 send_skb->dev = skb->dev;
477                 send_skb->protocol = htons(ETH_P_ARP);
478
479                 /* Fill the device header for the ARP frame */
480                 if (dev_hard_header(send_skb, skb->dev, ptype,
481                                     sha, np->dev->dev_addr,
482                                     send_skb->len) < 0) {
483                         kfree_skb(send_skb);
484                         continue;
485                 }
486
487                 /*
488                  * Fill out the arp protocol part.
489                  *
490                  * we only support ethernet device type,
491                  * which (according to RFC 1390) should
492                  * always equal 1 (Ethernet).
493                  */
494
495                 arp->ar_hrd = htons(np->dev->type);
496                 arp->ar_pro = htons(ETH_P_IP);
497                 arp->ar_hln = np->dev->addr_len;
498                 arp->ar_pln = 4;
499                 arp->ar_op = htons(type);
500
501                 arp_ptr = (unsigned char *)(arp + 1);
502                 memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
503                 arp_ptr += np->dev->addr_len;
504                 memcpy(arp_ptr, &tip, 4);
505                 arp_ptr += 4;
506                 memcpy(arp_ptr, sha, np->dev->addr_len);
507                 arp_ptr += np->dev->addr_len;
508                 memcpy(arp_ptr, &sip, 4);
509
510                 netpoll_send_skb(np, send_skb);
511
512                 /* If there are several rx_hooks for the same address,
513                    we're fine by sending a single reply */
514                 break;
515         }
516         spin_unlock_irqrestore(&npinfo->rx_lock, flags);
517 }
518
519 int __netpoll_rx(struct sk_buff *skb)
520 {
521         int proto, len, ulen;
522         int hits = 0;
523         struct iphdr *iph;
524         struct udphdr *uh;
525         struct netpoll_info *npinfo = skb->dev->npinfo;
526         struct netpoll *np, *tmp;
527
528         if (list_empty(&npinfo->rx_np))
529                 goto out;
530
531         if (skb->dev->type != ARPHRD_ETHER)
532                 goto out;
533
534         /* check if netpoll clients need ARP */
535         if (skb->protocol == htons(ETH_P_ARP) &&
536             atomic_read(&trapped)) {
537                 skb_queue_tail(&npinfo->arp_tx, skb);
538                 return 1;
539         }
540
541         proto = ntohs(eth_hdr(skb)->h_proto);
542         if (proto != ETH_P_IP)
543                 goto out;
544         if (skb->pkt_type == PACKET_OTHERHOST)
545                 goto out;
546         if (skb_shared(skb))
547                 goto out;
548
549         iph = (struct iphdr *)skb->data;
550         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
551                 goto out;
552         if (iph->ihl < 5 || iph->version != 4)
553                 goto out;
554         if (!pskb_may_pull(skb, iph->ihl*4))
555                 goto out;
556         if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
557                 goto out;
558
559         len = ntohs(iph->tot_len);
560         if (skb->len < len || len < iph->ihl*4)
561                 goto out;
562
563         /*
564          * Our transport medium may have padded the buffer out.
565          * Now We trim to the true length of the frame.
566          */
567         if (pskb_trim_rcsum(skb, len))
568                 goto out;
569
570         if (iph->protocol != IPPROTO_UDP)
571                 goto out;
572
573         len -= iph->ihl*4;
574         uh = (struct udphdr *)(((char *)iph) + iph->ihl*4);
575         ulen = ntohs(uh->len);
576
577         if (ulen != len)
578                 goto out;
579         if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr))
580                 goto out;
581
582         list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
583                 if (np->local_ip && np->local_ip != iph->daddr)
584                         continue;
585                 if (np->remote_ip && np->remote_ip != iph->saddr)
586                         continue;
587                 if (np->local_port && np->local_port != ntohs(uh->dest))
588                         continue;
589
590                 np->rx_hook(np, ntohs(uh->source),
591                                (char *)(uh+1),
592                                ulen - sizeof(struct udphdr));
593                 hits++;
594         }
595
596         if (!hits)
597                 goto out;
598
599         kfree_skb(skb);
600         return 1;
601
602 out:
603         if (atomic_read(&trapped)) {
604                 kfree_skb(skb);
605                 return 1;
606         }
607
608         return 0;
609 }
610
611 void netpoll_print_options(struct netpoll *np)
612 {
613         printk(KERN_INFO "%s: local port %d\n",
614                          np->name, np->local_port);
615         printk(KERN_INFO "%s: local IP %pI4\n",
616                          np->name, &np->local_ip);
617         printk(KERN_INFO "%s: interface '%s'\n",
618                          np->name, np->dev_name);
619         printk(KERN_INFO "%s: remote port %d\n",
620                          np->name, np->remote_port);
621         printk(KERN_INFO "%s: remote IP %pI4\n",
622                          np->name, &np->remote_ip);
623         printk(KERN_INFO "%s: remote ethernet address %pM\n",
624                          np->name, np->remote_mac);
625 }
626
627 int netpoll_parse_options(struct netpoll *np, char *opt)
628 {
629         char *cur=opt, *delim;
630
631         if (*cur != '@') {
632                 if ((delim = strchr(cur, '@')) == NULL)
633                         goto parse_failed;
634                 *delim = 0;
635                 np->local_port = simple_strtol(cur, NULL, 10);
636                 cur = delim;
637         }
638         cur++;
639
640         if (*cur != '/') {
641                 if ((delim = strchr(cur, '/')) == NULL)
642                         goto parse_failed;
643                 *delim = 0;
644                 np->local_ip = in_aton(cur);
645                 cur = delim;
646         }
647         cur++;
648
649         if (*cur != ',') {
650                 /* parse out dev name */
651                 if ((delim = strchr(cur, ',')) == NULL)
652                         goto parse_failed;
653                 *delim = 0;
654                 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
655                 cur = delim;
656         }
657         cur++;
658
659         if (*cur != '@') {
660                 /* dst port */
661                 if ((delim = strchr(cur, '@')) == NULL)
662                         goto parse_failed;
663                 *delim = 0;
664                 if (*cur == ' ' || *cur == '\t')
665                         printk(KERN_INFO "%s: warning: whitespace"
666                                         "is not allowed\n", np->name);
667                 np->remote_port = simple_strtol(cur, NULL, 10);
668                 cur = delim;
669         }
670         cur++;
671
672         /* dst ip */
673         if ((delim = strchr(cur, '/')) == NULL)
674                 goto parse_failed;
675         *delim = 0;
676         np->remote_ip = in_aton(cur);
677         cur = delim + 1;
678
679         if (*cur != 0) {
680                 /* MAC address */
681                 if ((delim = strchr(cur, ':')) == NULL)
682                         goto parse_failed;
683                 *delim = 0;
684                 np->remote_mac[0] = simple_strtol(cur, NULL, 16);
685                 cur = delim + 1;
686                 if ((delim = strchr(cur, ':')) == NULL)
687                         goto parse_failed;
688                 *delim = 0;
689                 np->remote_mac[1] = simple_strtol(cur, NULL, 16);
690                 cur = delim + 1;
691                 if ((delim = strchr(cur, ':')) == NULL)
692                         goto parse_failed;
693                 *delim = 0;
694                 np->remote_mac[2] = simple_strtol(cur, NULL, 16);
695                 cur = delim + 1;
696                 if ((delim = strchr(cur, ':')) == NULL)
697                         goto parse_failed;
698                 *delim = 0;
699                 np->remote_mac[3] = simple_strtol(cur, NULL, 16);
700                 cur = delim + 1;
701                 if ((delim = strchr(cur, ':')) == NULL)
702                         goto parse_failed;
703                 *delim = 0;
704                 np->remote_mac[4] = simple_strtol(cur, NULL, 16);
705                 cur = delim + 1;
706                 np->remote_mac[5] = simple_strtol(cur, NULL, 16);
707         }
708
709         netpoll_print_options(np);
710
711         return 0;
712
713  parse_failed:
714         printk(KERN_INFO "%s: couldn't parse config at '%s'!\n",
715                np->name, cur);
716         return -1;
717 }
718
719 int netpoll_setup(struct netpoll *np)
720 {
721         struct net_device *ndev = NULL;
722         struct in_device *in_dev;
723         struct netpoll_info *npinfo;
724         struct netpoll *npe, *tmp;
725         unsigned long flags;
726         int err;
727
728         if (np->dev_name)
729                 ndev = dev_get_by_name(&init_net, np->dev_name);
730         if (!ndev) {
731                 printk(KERN_ERR "%s: %s doesn't exist, aborting.\n",
732                        np->name, np->dev_name);
733                 return -ENODEV;
734         }
735
736         np->dev = ndev;
737         if (!ndev->npinfo) {
738                 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
739                 if (!npinfo) {
740                         err = -ENOMEM;
741                         goto put;
742                 }
743
744                 npinfo->rx_flags = 0;
745                 INIT_LIST_HEAD(&npinfo->rx_np);
746
747                 spin_lock_init(&npinfo->rx_lock);
748                 skb_queue_head_init(&npinfo->arp_tx);
749                 skb_queue_head_init(&npinfo->txq);
750                 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
751
752                 atomic_set(&npinfo->refcnt, 1);
753         } else {
754                 npinfo = ndev->npinfo;
755                 atomic_inc(&npinfo->refcnt);
756         }
757
758         if (!ndev->netdev_ops->ndo_poll_controller) {
759                 printk(KERN_ERR "%s: %s doesn't support polling, aborting.\n",
760                        np->name, np->dev_name);
761                 err = -ENOTSUPP;
762                 goto release;
763         }
764
765         if (!netif_running(ndev)) {
766                 unsigned long atmost, atleast;
767
768                 printk(KERN_INFO "%s: device %s not up yet, forcing it\n",
769                        np->name, np->dev_name);
770
771                 rtnl_lock();
772                 err = dev_open(ndev);
773                 rtnl_unlock();
774
775                 if (err) {
776                         printk(KERN_ERR "%s: failed to open %s\n",
777                                np->name, ndev->name);
778                         goto release;
779                 }
780
781                 atleast = jiffies + HZ/10;
782                 atmost = jiffies + carrier_timeout * HZ;
783                 while (!netif_carrier_ok(ndev)) {
784                         if (time_after(jiffies, atmost)) {
785                                 printk(KERN_NOTICE
786                                        "%s: timeout waiting for carrier\n",
787                                        np->name);
788                                 break;
789                         }
790                         msleep(1);
791                 }
792
793                 /* If carrier appears to come up instantly, we don't
794                  * trust it and pause so that we don't pump all our
795                  * queued console messages into the bitbucket.
796                  */
797
798                 if (time_before(jiffies, atleast)) {
799                         printk(KERN_NOTICE "%s: carrier detect appears"
800                                " untrustworthy, waiting 4 seconds\n",
801                                np->name);
802                         msleep(4000);
803                 }
804         }
805
806         if (!np->local_ip) {
807                 rcu_read_lock();
808                 in_dev = __in_dev_get_rcu(ndev);
809
810                 if (!in_dev || !in_dev->ifa_list) {
811                         rcu_read_unlock();
812                         printk(KERN_ERR "%s: no IP address for %s, aborting\n",
813                                np->name, np->dev_name);
814                         err = -EDESTADDRREQ;
815                         goto release;
816                 }
817
818                 np->local_ip = in_dev->ifa_list->ifa_local;
819                 rcu_read_unlock();
820                 printk(KERN_INFO "%s: local IP %pI4\n", np->name, &np->local_ip);
821         }
822
823         if (np->rx_hook) {
824                 spin_lock_irqsave(&npinfo->rx_lock, flags);
825                 npinfo->rx_flags |= NETPOLL_RX_ENABLED;
826                 list_add_tail(&np->rx, &npinfo->rx_np);
827                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
828         }
829
830         /* fill up the skb queue */
831         refill_skbs();
832
833         /* last thing to do is link it to the net device structure */
834         ndev->npinfo = npinfo;
835
836         /* avoid racing with NAPI reading npinfo */
837         synchronize_rcu();
838
839         return 0;
840
841  release:
842         if (!ndev->npinfo) {
843                 spin_lock_irqsave(&npinfo->rx_lock, flags);
844                 list_for_each_entry_safe(npe, tmp, &npinfo->rx_np, rx) {
845                         npe->dev = NULL;
846                 }
847                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
848
849                 kfree(npinfo);
850         }
851 put:
852         dev_put(ndev);
853         return err;
854 }
855
856 static int __init netpoll_init(void)
857 {
858         skb_queue_head_init(&skb_pool);
859         return 0;
860 }
861 core_initcall(netpoll_init);
862
863 void netpoll_cleanup(struct netpoll *np)
864 {
865         struct netpoll_info *npinfo;
866         unsigned long flags;
867
868         if (np->dev) {
869                 npinfo = np->dev->npinfo;
870                 if (npinfo) {
871                         if (!list_empty(&npinfo->rx_np)) {
872                                 spin_lock_irqsave(&npinfo->rx_lock, flags);
873                                 list_del(&np->rx);
874                                 if (list_empty(&npinfo->rx_np))
875                                         npinfo->rx_flags &= ~NETPOLL_RX_ENABLED;
876                                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
877                         }
878
879                         if (atomic_dec_and_test(&npinfo->refcnt)) {
880                                 skb_queue_purge(&npinfo->arp_tx);
881                                 skb_queue_purge(&npinfo->txq);
882                                 cancel_rearming_delayed_work(&npinfo->tx_work);
883
884                                 /* clean after last, unfinished work */
885                                 __skb_queue_purge(&npinfo->txq);
886                                 kfree(npinfo);
887                                 np->dev->npinfo = NULL;
888                         }
889                 }
890
891                 dev_put(np->dev);
892         }
893
894         np->dev = NULL;
895 }
896
897 int netpoll_trap(void)
898 {
899         return atomic_read(&trapped);
900 }
901
902 void netpoll_set_trap(int trap)
903 {
904         if (trap)
905                 atomic_inc(&trapped);
906         else
907                 atomic_dec(&trapped);
908 }
909
910 EXPORT_SYMBOL(netpoll_set_trap);
911 EXPORT_SYMBOL(netpoll_trap);
912 EXPORT_SYMBOL(netpoll_print_options);
913 EXPORT_SYMBOL(netpoll_parse_options);
914 EXPORT_SYMBOL(netpoll_setup);
915 EXPORT_SYMBOL(netpoll_cleanup);
916 EXPORT_SYMBOL(netpoll_send_udp);
917 EXPORT_SYMBOL(netpoll_poll);