[CONNECTOR]: Update documentation to match reality.
[linux-2.6/verdex.git] / net / core / netpoll.c
blob802fe11efad0b2ad53942f28ce591435380e8008
1 /*
2 * Common framework for low-level network console, dump, and debugger code
4 * Sep 8 2003 Matt Mackall <mpm@selenic.com>
6 * based on the netconsole code from:
8 * Copyright (C) 2001 Ingo Molnar <mingo@redhat.com>
9 * Copyright (C) 2002 Red Hat, Inc.
12 #include <linux/smp_lock.h>
13 #include <linux/netdevice.h>
14 #include <linux/etherdevice.h>
15 #include <linux/string.h>
16 #include <linux/inetdevice.h>
17 #include <linux/inet.h>
18 #include <linux/interrupt.h>
19 #include <linux/netpoll.h>
20 #include <linux/sched.h>
21 #include <linux/delay.h>
22 #include <linux/rcupdate.h>
23 #include <linux/workqueue.h>
24 #include <net/tcp.h>
25 #include <net/udp.h>
26 #include <asm/unaligned.h>
29 * We maintain a small pool of fully-sized skbs, to make sure the
30 * message gets out even in extreme OOM situations.
33 #define MAX_UDP_CHUNK 1460
34 #define MAX_SKBS 32
35 #define MAX_QUEUE_DEPTH (MAX_SKBS / 2)
36 #define MAX_RETRIES 20000
38 static DEFINE_SPINLOCK(skb_list_lock);
39 static int nr_skbs;
40 static struct sk_buff *skbs;
42 static DEFINE_SPINLOCK(queue_lock);
43 static int queue_depth;
44 static struct sk_buff *queue_head, *queue_tail;
46 static atomic_t trapped;
48 #define NETPOLL_RX_ENABLED 1
49 #define NETPOLL_RX_DROP 2
51 #define MAX_SKB_SIZE \
52 (MAX_UDP_CHUNK + sizeof(struct udphdr) + \
53 sizeof(struct iphdr) + sizeof(struct ethhdr))
55 static void zap_completion_queue(void);
57 static void queue_process(void *p)
59 unsigned long flags;
60 struct sk_buff *skb;
62 while (queue_head) {
63 spin_lock_irqsave(&queue_lock, flags);
65 skb = queue_head;
66 queue_head = skb->next;
67 if (skb == queue_tail)
68 queue_head = NULL;
70 queue_depth--;
72 spin_unlock_irqrestore(&queue_lock, flags);
74 dev_queue_xmit(skb);
78 static DECLARE_WORK(send_queue, queue_process, NULL);
80 void netpoll_queue(struct sk_buff *skb)
82 unsigned long flags;
84 if (queue_depth == MAX_QUEUE_DEPTH) {
85 __kfree_skb(skb);
86 return;
89 spin_lock_irqsave(&queue_lock, flags);
90 if (!queue_head)
91 queue_head = skb;
92 else
93 queue_tail->next = skb;
94 queue_tail = skb;
95 queue_depth++;
96 spin_unlock_irqrestore(&queue_lock, flags);
98 schedule_work(&send_queue);
101 static int checksum_udp(struct sk_buff *skb, struct udphdr *uh,
102 unsigned short ulen, u32 saddr, u32 daddr)
104 if (uh->check == 0)
105 return 0;
107 if (skb->ip_summed == CHECKSUM_HW)
108 return csum_tcpudp_magic(
109 saddr, daddr, ulen, IPPROTO_UDP, skb->csum);
111 skb->csum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0);
113 return csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
117 * Check whether delayed processing was scheduled for our NIC. If so,
118 * we attempt to grab the poll lock and use ->poll() to pump the card.
119 * If this fails, either we've recursed in ->poll() or it's already
120 * running on another CPU.
122 * Note: we don't mask interrupts with this lock because we're using
123 * trylock here and interrupts are already disabled in the softirq
124 * case. Further, we test the poll_owner to avoid recursion on UP
125 * systems where the lock doesn't exist.
127 * In cases where there is bi-directional communications, reading only
128 * one message at a time can lead to packets being dropped by the
129 * network adapter, forcing superfluous retries and possibly timeouts.
130 * Thus, we set our budget to greater than 1.
132 static void poll_napi(struct netpoll *np)
134 struct netpoll_info *npinfo = np->dev->npinfo;
135 int budget = 16;
137 if (test_bit(__LINK_STATE_RX_SCHED, &np->dev->state) &&
138 npinfo->poll_owner != smp_processor_id() &&
139 spin_trylock(&npinfo->poll_lock)) {
140 npinfo->rx_flags |= NETPOLL_RX_DROP;
141 atomic_inc(&trapped);
143 np->dev->poll(np->dev, &budget);
145 atomic_dec(&trapped);
146 npinfo->rx_flags &= ~NETPOLL_RX_DROP;
147 spin_unlock(&npinfo->poll_lock);
151 void netpoll_poll(struct netpoll *np)
153 if(!np->dev || !netif_running(np->dev) || !np->dev->poll_controller)
154 return;
156 /* Process pending work on NIC */
157 np->dev->poll_controller(np->dev);
158 if (np->dev->poll)
159 poll_napi(np);
161 zap_completion_queue();
164 static void refill_skbs(void)
166 struct sk_buff *skb;
167 unsigned long flags;
169 spin_lock_irqsave(&skb_list_lock, flags);
170 while (nr_skbs < MAX_SKBS) {
171 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
172 if (!skb)
173 break;
175 skb->next = skbs;
176 skbs = skb;
177 nr_skbs++;
179 spin_unlock_irqrestore(&skb_list_lock, flags);
182 static void zap_completion_queue(void)
184 unsigned long flags;
185 struct softnet_data *sd = &get_cpu_var(softnet_data);
187 if (sd->completion_queue) {
188 struct sk_buff *clist;
190 local_irq_save(flags);
191 clist = sd->completion_queue;
192 sd->completion_queue = NULL;
193 local_irq_restore(flags);
195 while (clist != NULL) {
196 struct sk_buff *skb = clist;
197 clist = clist->next;
198 if(skb->destructor)
199 dev_kfree_skb_any(skb); /* put this one back */
200 else
201 __kfree_skb(skb);
205 put_cpu_var(softnet_data);
208 static struct sk_buff * find_skb(struct netpoll *np, int len, int reserve)
210 int once = 1, count = 0;
211 unsigned long flags;
212 struct sk_buff *skb = NULL;
214 zap_completion_queue();
215 repeat:
216 if (nr_skbs < MAX_SKBS)
217 refill_skbs();
219 skb = alloc_skb(len, GFP_ATOMIC);
221 if (!skb) {
222 spin_lock_irqsave(&skb_list_lock, flags);
223 skb = skbs;
224 if (skb) {
225 skbs = skb->next;
226 skb->next = NULL;
227 nr_skbs--;
229 spin_unlock_irqrestore(&skb_list_lock, flags);
232 if(!skb) {
233 count++;
234 if (once && (count == 1000000)) {
235 printk("out of netpoll skbs!\n");
236 once = 0;
238 netpoll_poll(np);
239 goto repeat;
242 atomic_set(&skb->users, 1);
243 skb_reserve(skb, reserve);
244 return skb;
247 static void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
249 int status;
250 struct netpoll_info *npinfo;
252 if (!np || !np->dev || !netif_running(np->dev)) {
253 __kfree_skb(skb);
254 return;
257 npinfo = np->dev->npinfo;
259 /* avoid recursion */
260 if (npinfo->poll_owner == smp_processor_id() ||
261 np->dev->xmit_lock_owner == smp_processor_id()) {
262 if (np->drop)
263 np->drop(skb);
264 else
265 __kfree_skb(skb);
266 return;
269 do {
270 npinfo->tries--;
271 spin_lock(&np->dev->xmit_lock);
272 np->dev->xmit_lock_owner = smp_processor_id();
275 * network drivers do not expect to be called if the queue is
276 * stopped.
278 if (netif_queue_stopped(np->dev)) {
279 np->dev->xmit_lock_owner = -1;
280 spin_unlock(&np->dev->xmit_lock);
281 netpoll_poll(np);
282 udelay(50);
283 continue;
286 status = np->dev->hard_start_xmit(skb, np->dev);
287 np->dev->xmit_lock_owner = -1;
288 spin_unlock(&np->dev->xmit_lock);
290 /* success */
291 if(!status) {
292 npinfo->tries = MAX_RETRIES; /* reset */
293 return;
296 /* transmit busy */
297 netpoll_poll(np);
298 udelay(50);
299 } while (npinfo->tries > 0);
302 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
304 int total_len, eth_len, ip_len, udp_len;
305 struct sk_buff *skb;
306 struct udphdr *udph;
307 struct iphdr *iph;
308 struct ethhdr *eth;
310 udp_len = len + sizeof(*udph);
311 ip_len = eth_len = udp_len + sizeof(*iph);
312 total_len = eth_len + ETH_HLEN + NET_IP_ALIGN;
314 skb = find_skb(np, total_len, total_len - len);
315 if (!skb)
316 return;
318 memcpy(skb->data, msg, len);
319 skb->len += len;
321 udph = (struct udphdr *) skb_push(skb, sizeof(*udph));
322 udph->source = htons(np->local_port);
323 udph->dest = htons(np->remote_port);
324 udph->len = htons(udp_len);
325 udph->check = 0;
327 iph = (struct iphdr *)skb_push(skb, sizeof(*iph));
329 /* iph->version = 4; iph->ihl = 5; */
330 put_unaligned(0x45, (unsigned char *)iph);
331 iph->tos = 0;
332 put_unaligned(htons(ip_len), &(iph->tot_len));
333 iph->id = 0;
334 iph->frag_off = 0;
335 iph->ttl = 64;
336 iph->protocol = IPPROTO_UDP;
337 iph->check = 0;
338 put_unaligned(htonl(np->local_ip), &(iph->saddr));
339 put_unaligned(htonl(np->remote_ip), &(iph->daddr));
340 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
342 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
344 eth->h_proto = htons(ETH_P_IP);
345 memcpy(eth->h_source, np->local_mac, 6);
346 memcpy(eth->h_dest, np->remote_mac, 6);
348 skb->dev = np->dev;
350 netpoll_send_skb(np, skb);
353 static void arp_reply(struct sk_buff *skb)
355 struct netpoll_info *npinfo = skb->dev->npinfo;
356 struct arphdr *arp;
357 unsigned char *arp_ptr;
358 int size, type = ARPOP_REPLY, ptype = ETH_P_ARP;
359 u32 sip, tip;
360 struct sk_buff *send_skb;
361 struct netpoll *np = NULL;
363 if (npinfo->rx_np && npinfo->rx_np->dev == skb->dev)
364 np = npinfo->rx_np;
365 if (!np)
366 return;
368 /* No arp on this interface */
369 if (skb->dev->flags & IFF_NOARP)
370 return;
372 if (!pskb_may_pull(skb, (sizeof(struct arphdr) +
373 (2 * skb->dev->addr_len) +
374 (2 * sizeof(u32)))))
375 return;
377 skb->h.raw = skb->nh.raw = skb->data;
378 arp = skb->nh.arph;
380 if ((arp->ar_hrd != htons(ARPHRD_ETHER) &&
381 arp->ar_hrd != htons(ARPHRD_IEEE802)) ||
382 arp->ar_pro != htons(ETH_P_IP) ||
383 arp->ar_op != htons(ARPOP_REQUEST))
384 return;
386 arp_ptr = (unsigned char *)(arp+1) + skb->dev->addr_len;
387 memcpy(&sip, arp_ptr, 4);
388 arp_ptr += 4 + skb->dev->addr_len;
389 memcpy(&tip, arp_ptr, 4);
391 /* Should we ignore arp? */
392 if (tip != htonl(np->local_ip) || LOOPBACK(tip) || MULTICAST(tip))
393 return;
395 size = sizeof(struct arphdr) + 2 * (skb->dev->addr_len + 4);
396 send_skb = find_skb(np, size + LL_RESERVED_SPACE(np->dev),
397 LL_RESERVED_SPACE(np->dev));
399 if (!send_skb)
400 return;
402 send_skb->nh.raw = send_skb->data;
403 arp = (struct arphdr *) skb_put(send_skb, size);
404 send_skb->dev = skb->dev;
405 send_skb->protocol = htons(ETH_P_ARP);
407 /* Fill the device header for the ARP frame */
409 if (np->dev->hard_header &&
410 np->dev->hard_header(send_skb, skb->dev, ptype,
411 np->remote_mac, np->local_mac,
412 send_skb->len) < 0) {
413 kfree_skb(send_skb);
414 return;
418 * Fill out the arp protocol part.
420 * we only support ethernet device type,
421 * which (according to RFC 1390) should always equal 1 (Ethernet).
424 arp->ar_hrd = htons(np->dev->type);
425 arp->ar_pro = htons(ETH_P_IP);
426 arp->ar_hln = np->dev->addr_len;
427 arp->ar_pln = 4;
428 arp->ar_op = htons(type);
430 arp_ptr=(unsigned char *)(arp + 1);
431 memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
432 arp_ptr += np->dev->addr_len;
433 memcpy(arp_ptr, &tip, 4);
434 arp_ptr += 4;
435 memcpy(arp_ptr, np->remote_mac, np->dev->addr_len);
436 arp_ptr += np->dev->addr_len;
437 memcpy(arp_ptr, &sip, 4);
439 netpoll_send_skb(np, send_skb);
442 int __netpoll_rx(struct sk_buff *skb)
444 int proto, len, ulen;
445 struct iphdr *iph;
446 struct udphdr *uh;
447 struct netpoll *np = skb->dev->npinfo->rx_np;
449 if (!np)
450 goto out;
451 if (skb->dev->type != ARPHRD_ETHER)
452 goto out;
454 /* check if netpoll clients need ARP */
455 if (skb->protocol == __constant_htons(ETH_P_ARP) &&
456 atomic_read(&trapped)) {
457 arp_reply(skb);
458 return 1;
461 proto = ntohs(eth_hdr(skb)->h_proto);
462 if (proto != ETH_P_IP)
463 goto out;
464 if (skb->pkt_type == PACKET_OTHERHOST)
465 goto out;
466 if (skb_shared(skb))
467 goto out;
469 iph = (struct iphdr *)skb->data;
470 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
471 goto out;
472 if (iph->ihl < 5 || iph->version != 4)
473 goto out;
474 if (!pskb_may_pull(skb, iph->ihl*4))
475 goto out;
476 if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
477 goto out;
479 len = ntohs(iph->tot_len);
480 if (skb->len < len || len < iph->ihl*4)
481 goto out;
483 if (iph->protocol != IPPROTO_UDP)
484 goto out;
486 len -= iph->ihl*4;
487 uh = (struct udphdr *)(((char *)iph) + iph->ihl*4);
488 ulen = ntohs(uh->len);
490 if (ulen != len)
491 goto out;
492 if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr) < 0)
493 goto out;
494 if (np->local_ip && np->local_ip != ntohl(iph->daddr))
495 goto out;
496 if (np->remote_ip && np->remote_ip != ntohl(iph->saddr))
497 goto out;
498 if (np->local_port && np->local_port != ntohs(uh->dest))
499 goto out;
501 np->rx_hook(np, ntohs(uh->source),
502 (char *)(uh+1),
503 ulen - sizeof(struct udphdr));
505 kfree_skb(skb);
506 return 1;
508 out:
509 if (atomic_read(&trapped)) {
510 kfree_skb(skb);
511 return 1;
514 return 0;
517 int netpoll_parse_options(struct netpoll *np, char *opt)
519 char *cur=opt, *delim;
521 if(*cur != '@') {
522 if ((delim = strchr(cur, '@')) == NULL)
523 goto parse_failed;
524 *delim=0;
525 np->local_port=simple_strtol(cur, NULL, 10);
526 cur=delim;
528 cur++;
529 printk(KERN_INFO "%s: local port %d\n", np->name, np->local_port);
531 if(*cur != '/') {
532 if ((delim = strchr(cur, '/')) == NULL)
533 goto parse_failed;
534 *delim=0;
535 np->local_ip=ntohl(in_aton(cur));
536 cur=delim;
538 printk(KERN_INFO "%s: local IP %d.%d.%d.%d\n",
539 np->name, HIPQUAD(np->local_ip));
541 cur++;
543 if ( *cur != ',') {
544 /* parse out dev name */
545 if ((delim = strchr(cur, ',')) == NULL)
546 goto parse_failed;
547 *delim=0;
548 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
549 cur=delim;
551 cur++;
553 printk(KERN_INFO "%s: interface %s\n", np->name, np->dev_name);
555 if ( *cur != '@' ) {
556 /* dst port */
557 if ((delim = strchr(cur, '@')) == NULL)
558 goto parse_failed;
559 *delim=0;
560 np->remote_port=simple_strtol(cur, NULL, 10);
561 cur=delim;
563 cur++;
564 printk(KERN_INFO "%s: remote port %d\n", np->name, np->remote_port);
566 /* dst ip */
567 if ((delim = strchr(cur, '/')) == NULL)
568 goto parse_failed;
569 *delim=0;
570 np->remote_ip=ntohl(in_aton(cur));
571 cur=delim+1;
573 printk(KERN_INFO "%s: remote IP %d.%d.%d.%d\n",
574 np->name, HIPQUAD(np->remote_ip));
576 if( *cur != 0 )
578 /* MAC address */
579 if ((delim = strchr(cur, ':')) == NULL)
580 goto parse_failed;
581 *delim=0;
582 np->remote_mac[0]=simple_strtol(cur, NULL, 16);
583 cur=delim+1;
584 if ((delim = strchr(cur, ':')) == NULL)
585 goto parse_failed;
586 *delim=0;
587 np->remote_mac[1]=simple_strtol(cur, NULL, 16);
588 cur=delim+1;
589 if ((delim = strchr(cur, ':')) == NULL)
590 goto parse_failed;
591 *delim=0;
592 np->remote_mac[2]=simple_strtol(cur, NULL, 16);
593 cur=delim+1;
594 if ((delim = strchr(cur, ':')) == NULL)
595 goto parse_failed;
596 *delim=0;
597 np->remote_mac[3]=simple_strtol(cur, NULL, 16);
598 cur=delim+1;
599 if ((delim = strchr(cur, ':')) == NULL)
600 goto parse_failed;
601 *delim=0;
602 np->remote_mac[4]=simple_strtol(cur, NULL, 16);
603 cur=delim+1;
604 np->remote_mac[5]=simple_strtol(cur, NULL, 16);
607 printk(KERN_INFO "%s: remote ethernet address "
608 "%02x:%02x:%02x:%02x:%02x:%02x\n",
609 np->name,
610 np->remote_mac[0],
611 np->remote_mac[1],
612 np->remote_mac[2],
613 np->remote_mac[3],
614 np->remote_mac[4],
615 np->remote_mac[5]);
617 return 0;
619 parse_failed:
620 printk(KERN_INFO "%s: couldn't parse config at %s!\n",
621 np->name, cur);
622 return -1;
625 int netpoll_setup(struct netpoll *np)
627 struct net_device *ndev = NULL;
628 struct in_device *in_dev;
629 struct netpoll_info *npinfo;
630 unsigned long flags;
632 if (np->dev_name)
633 ndev = dev_get_by_name(np->dev_name);
634 if (!ndev) {
635 printk(KERN_ERR "%s: %s doesn't exist, aborting.\n",
636 np->name, np->dev_name);
637 return -1;
640 np->dev = ndev;
641 if (!ndev->npinfo) {
642 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
643 if (!npinfo)
644 goto release;
646 npinfo->rx_flags = 0;
647 npinfo->rx_np = NULL;
648 spin_lock_init(&npinfo->poll_lock);
649 npinfo->poll_owner = -1;
650 npinfo->tries = MAX_RETRIES;
651 spin_lock_init(&npinfo->rx_lock);
652 } else
653 npinfo = ndev->npinfo;
655 if (!ndev->poll_controller) {
656 printk(KERN_ERR "%s: %s doesn't support polling, aborting.\n",
657 np->name, np->dev_name);
658 goto release;
661 if (!netif_running(ndev)) {
662 unsigned long atmost, atleast;
664 printk(KERN_INFO "%s: device %s not up yet, forcing it\n",
665 np->name, np->dev_name);
667 rtnl_shlock();
668 if (dev_change_flags(ndev, ndev->flags | IFF_UP) < 0) {
669 printk(KERN_ERR "%s: failed to open %s\n",
670 np->name, np->dev_name);
671 rtnl_shunlock();
672 goto release;
674 rtnl_shunlock();
676 atleast = jiffies + HZ/10;
677 atmost = jiffies + 4*HZ;
678 while (!netif_carrier_ok(ndev)) {
679 if (time_after(jiffies, atmost)) {
680 printk(KERN_NOTICE
681 "%s: timeout waiting for carrier\n",
682 np->name);
683 break;
685 cond_resched();
688 /* If carrier appears to come up instantly, we don't
689 * trust it and pause so that we don't pump all our
690 * queued console messages into the bitbucket.
693 if (time_before(jiffies, atleast)) {
694 printk(KERN_NOTICE "%s: carrier detect appears"
695 " untrustworthy, waiting 4 seconds\n",
696 np->name);
697 msleep(4000);
701 if (!memcmp(np->local_mac, "\0\0\0\0\0\0", 6) && ndev->dev_addr)
702 memcpy(np->local_mac, ndev->dev_addr, 6);
704 if (!np->local_ip) {
705 rcu_read_lock();
706 in_dev = __in_dev_get_rcu(ndev);
708 if (!in_dev || !in_dev->ifa_list) {
709 rcu_read_unlock();
710 printk(KERN_ERR "%s: no IP address for %s, aborting\n",
711 np->name, np->dev_name);
712 goto release;
715 np->local_ip = ntohl(in_dev->ifa_list->ifa_local);
716 rcu_read_unlock();
717 printk(KERN_INFO "%s: local IP %d.%d.%d.%d\n",
718 np->name, HIPQUAD(np->local_ip));
721 if (np->rx_hook) {
722 spin_lock_irqsave(&npinfo->rx_lock, flags);
723 npinfo->rx_flags |= NETPOLL_RX_ENABLED;
724 npinfo->rx_np = np;
725 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
728 /* fill up the skb queue */
729 refill_skbs();
731 /* last thing to do is link it to the net device structure */
732 ndev->npinfo = npinfo;
734 /* avoid racing with NAPI reading npinfo */
735 synchronize_rcu();
737 return 0;
739 release:
740 if (!ndev->npinfo)
741 kfree(npinfo);
742 np->dev = NULL;
743 dev_put(ndev);
744 return -1;
747 void netpoll_cleanup(struct netpoll *np)
749 struct netpoll_info *npinfo;
750 unsigned long flags;
752 if (np->dev) {
753 npinfo = np->dev->npinfo;
754 if (npinfo && npinfo->rx_np == np) {
755 spin_lock_irqsave(&npinfo->rx_lock, flags);
756 npinfo->rx_np = NULL;
757 npinfo->rx_flags &= ~NETPOLL_RX_ENABLED;
758 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
760 dev_put(np->dev);
763 np->dev = NULL;
766 int netpoll_trap(void)
768 return atomic_read(&trapped);
771 void netpoll_set_trap(int trap)
773 if (trap)
774 atomic_inc(&trapped);
775 else
776 atomic_dec(&trapped);
779 EXPORT_SYMBOL(netpoll_set_trap);
780 EXPORT_SYMBOL(netpoll_trap);
781 EXPORT_SYMBOL(netpoll_parse_options);
782 EXPORT_SYMBOL(netpoll_setup);
783 EXPORT_SYMBOL(netpoll_cleanup);
784 EXPORT_SYMBOL(netpoll_send_udp);
785 EXPORT_SYMBOL(netpoll_poll);
786 EXPORT_SYMBOL(netpoll_queue);