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/netdevice.h>
13 #include <linux/etherdevice.h>
14 #include <linux/string.h>
15 #include <linux/if_arp.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>
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
35 #define MAX_QUEUE_DEPTH (MAX_SKBS / 2)
37 static struct sk_buff_head skb_pool;
39 static atomic_t trapped;
41 #define USEC_PER_POLL 50
42 #define NETPOLL_RX_ENABLED 1
43 #define NETPOLL_RX_DROP 2
45 #define MAX_SKB_SIZE \
46 (MAX_UDP_CHUNK + sizeof(struct udphdr) + \
47 sizeof(struct iphdr) + sizeof(struct ethhdr))
49 static void zap_completion_queue(void);
50 static void arp_reply(struct sk_buff *skb);
52 static void queue_process(struct work_struct *work)
54 struct netpoll_info *npinfo =
55 container_of(work, struct netpoll_info, tx_work.work);
59 while ((skb = skb_dequeue(&npinfo->txq))) {
60 struct net_device *dev = skb->dev;
62 if (!netif_device_present(dev) || !netif_running(dev)) {
67 local_irq_save(flags);
69 if (netif_queue_stopped(dev) ||
70 dev->hard_start_xmit(skb, dev) != NETDEV_TX_OK) {
71 skb_queue_head(&npinfo->txq, skb);
73 local_irq_restore(flags);
75 if (atomic_read(&npinfo->refcnt))
76 schedule_delayed_work(&npinfo->tx_work, HZ/10);
80 local_irq_restore(flags);
84 static __sum16 checksum_udp(struct sk_buff *skb, struct udphdr *uh,
85 unsigned short ulen, __be32 saddr, __be32 daddr)
89 if (uh->check == 0 || skb_csum_unnecessary(skb))
92 psum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0);
94 if (skb->ip_summed == CHECKSUM_COMPLETE &&
95 !csum_fold(csum_add(psum, skb->csum)))
100 return __skb_checksum_complete(skb);
104 * Check whether delayed processing was scheduled for our NIC. If so,
105 * we attempt to grab the poll lock and use ->poll() to pump the card.
106 * If this fails, either we've recursed in ->poll() or it's already
107 * running on another CPU.
109 * Note: we don't mask interrupts with this lock because we're using
110 * trylock here and interrupts are already disabled in the softirq
111 * case. Further, we test the poll_owner to avoid recursion on UP
112 * systems where the lock doesn't exist.
114 * In cases where there is bi-directional communications, reading only
115 * one message at a time can lead to packets being dropped by the
116 * network adapter, forcing superfluous retries and possibly timeouts.
117 * Thus, we set our budget to greater than 1.
119 static void poll_napi(struct netpoll *np)
121 struct netpoll_info *npinfo = np->dev->npinfo;
124 if (test_bit(__LINK_STATE_RX_SCHED, &np->dev->state) &&
125 npinfo->poll_owner != smp_processor_id() &&
126 spin_trylock(&npinfo->poll_lock)) {
127 npinfo->rx_flags |= NETPOLL_RX_DROP;
128 atomic_inc(&trapped);
130 np->dev->poll(np->dev, &budget);
132 atomic_dec(&trapped);
133 npinfo->rx_flags &= ~NETPOLL_RX_DROP;
134 spin_unlock(&npinfo->poll_lock);
138 static void service_arp_queue(struct netpoll_info *npi)
145 skb = skb_dequeue(&npi->arp_tx);
147 while (skb != NULL) {
149 skb = skb_dequeue(&npi->arp_tx);
153 void netpoll_poll(struct netpoll *np)
155 if (!np->dev || !netif_running(np->dev) || !np->dev->poll_controller)
158 /* Process pending work on NIC */
159 np->dev->poll_controller(np->dev);
163 service_arp_queue(np->dev->npinfo);
165 zap_completion_queue();
168 static void refill_skbs(void)
173 spin_lock_irqsave(&skb_pool.lock, flags);
174 while (skb_pool.qlen < MAX_SKBS) {
175 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
179 __skb_queue_tail(&skb_pool, skb);
181 spin_unlock_irqrestore(&skb_pool.lock, flags);
184 static void zap_completion_queue(void)
187 struct softnet_data *sd = &get_cpu_var(softnet_data);
189 if (sd->completion_queue) {
190 struct sk_buff *clist;
192 local_irq_save(flags);
193 clist = sd->completion_queue;
194 sd->completion_queue = NULL;
195 local_irq_restore(flags);
197 while (clist != NULL) {
198 struct sk_buff *skb = clist;
201 dev_kfree_skb_any(skb); /* put this one back */
207 put_cpu_var(softnet_data);
210 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
215 zap_completion_queue();
219 skb = alloc_skb(len, GFP_ATOMIC);
221 skb = skb_dequeue(&skb_pool);
231 atomic_set(&skb->users, 1);
232 skb_reserve(skb, reserve);
236 static void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
238 int status = NETDEV_TX_BUSY;
240 struct net_device *dev = np->dev;
241 struct netpoll_info *npinfo = np->dev->npinfo;
243 if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
248 /* don't get messages out of order, and no recursion */
249 if (skb_queue_len(&npinfo->txq) == 0 &&
250 npinfo->poll_owner != smp_processor_id()) {
253 local_irq_save(flags);
254 /* try until next clock tick */
255 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
256 tries > 0; --tries) {
257 if (netif_tx_trylock(dev)) {
258 if (!netif_queue_stopped(dev))
259 status = dev->hard_start_xmit(skb, dev);
260 netif_tx_unlock(dev);
262 if (status == NETDEV_TX_OK)
267 /* tickle device maybe there is some cleanup */
270 udelay(USEC_PER_POLL);
272 local_irq_restore(flags);
275 if (status != NETDEV_TX_OK) {
276 skb_queue_tail(&npinfo->txq, skb);
277 schedule_delayed_work(&npinfo->tx_work,0);
281 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
283 int total_len, eth_len, ip_len, udp_len;
289 udp_len = len + sizeof(*udph);
290 ip_len = eth_len = udp_len + sizeof(*iph);
291 total_len = eth_len + ETH_HLEN + NET_IP_ALIGN;
293 skb = find_skb(np, total_len, total_len - len);
297 skb_copy_to_linear_data(skb, msg, len);
300 skb_push(skb, sizeof(*udph));
301 skb_reset_transport_header(skb);
303 udph->source = htons(np->local_port);
304 udph->dest = htons(np->remote_port);
305 udph->len = htons(udp_len);
307 udph->check = csum_tcpudp_magic(htonl(np->local_ip),
308 htonl(np->remote_ip),
309 udp_len, IPPROTO_UDP,
310 csum_partial((unsigned char *)udph, udp_len, 0));
311 if (udph->check == 0)
312 udph->check = CSUM_MANGLED_0;
314 skb_push(skb, sizeof(*iph));
315 skb_reset_network_header(skb);
318 /* iph->version = 4; iph->ihl = 5; */
319 put_unaligned(0x45, (unsigned char *)iph);
321 put_unaligned(htons(ip_len), &(iph->tot_len));
325 iph->protocol = IPPROTO_UDP;
327 put_unaligned(htonl(np->local_ip), &(iph->saddr));
328 put_unaligned(htonl(np->remote_ip), &(iph->daddr));
329 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
331 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
332 skb_reset_mac_header(skb);
333 skb->protocol = eth->h_proto = htons(ETH_P_IP);
334 memcpy(eth->h_source, np->local_mac, 6);
335 memcpy(eth->h_dest, np->remote_mac, 6);
339 netpoll_send_skb(np, skb);
342 static void arp_reply(struct sk_buff *skb)
344 struct netpoll_info *npinfo = skb->dev->npinfo;
346 unsigned char *arp_ptr;
347 int size, type = ARPOP_REPLY, ptype = ETH_P_ARP;
350 struct sk_buff *send_skb;
351 struct netpoll *np = NULL;
353 if (npinfo->rx_np && npinfo->rx_np->dev == skb->dev)
358 /* No arp on this interface */
359 if (skb->dev->flags & IFF_NOARP)
362 if (!pskb_may_pull(skb, (sizeof(struct arphdr) +
363 (2 * skb->dev->addr_len) +
367 skb_reset_network_header(skb);
368 skb_reset_transport_header(skb);
371 if ((arp->ar_hrd != htons(ARPHRD_ETHER) &&
372 arp->ar_hrd != htons(ARPHRD_IEEE802)) ||
373 arp->ar_pro != htons(ETH_P_IP) ||
374 arp->ar_op != htons(ARPOP_REQUEST))
377 arp_ptr = (unsigned char *)(arp+1);
378 /* save the location of the src hw addr */
380 arp_ptr += skb->dev->addr_len;
381 memcpy(&sip, arp_ptr, 4);
383 /* if we actually cared about dst hw addr, it would get copied here */
384 arp_ptr += skb->dev->addr_len;
385 memcpy(&tip, arp_ptr, 4);
387 /* Should we ignore arp? */
388 if (tip != htonl(np->local_ip) || LOOPBACK(tip) || MULTICAST(tip))
391 size = sizeof(struct arphdr) + 2 * (skb->dev->addr_len + 4);
392 send_skb = find_skb(np, size + LL_RESERVED_SPACE(np->dev),
393 LL_RESERVED_SPACE(np->dev));
398 skb_reset_network_header(send_skb);
399 arp = (struct arphdr *) skb_put(send_skb, size);
400 send_skb->dev = skb->dev;
401 send_skb->protocol = htons(ETH_P_ARP);
403 /* Fill the device header for the ARP frame */
405 if (np->dev->hard_header &&
406 np->dev->hard_header(send_skb, skb->dev, ptype,
408 send_skb->len) < 0) {
414 * Fill out the arp protocol part.
416 * we only support ethernet device type,
417 * which (according to RFC 1390) should always equal 1 (Ethernet).
420 arp->ar_hrd = htons(np->dev->type);
421 arp->ar_pro = htons(ETH_P_IP);
422 arp->ar_hln = np->dev->addr_len;
424 arp->ar_op = htons(type);
426 arp_ptr=(unsigned char *)(arp + 1);
427 memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
428 arp_ptr += np->dev->addr_len;
429 memcpy(arp_ptr, &tip, 4);
431 memcpy(arp_ptr, sha, np->dev->addr_len);
432 arp_ptr += np->dev->addr_len;
433 memcpy(arp_ptr, &sip, 4);
435 netpoll_send_skb(np, send_skb);
438 int __netpoll_rx(struct sk_buff *skb)
440 int proto, len, ulen;
443 struct netpoll_info *npi = skb->dev->npinfo;
444 struct netpoll *np = npi->rx_np;
448 if (skb->dev->type != ARPHRD_ETHER)
451 /* check if netpoll clients need ARP */
452 if (skb->protocol == htons(ETH_P_ARP) &&
453 atomic_read(&trapped)) {
454 skb_queue_tail(&npi->arp_tx, skb);
458 proto = ntohs(eth_hdr(skb)->h_proto);
459 if (proto != ETH_P_IP)
461 if (skb->pkt_type == PACKET_OTHERHOST)
466 iph = (struct iphdr *)skb->data;
467 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
469 if (iph->ihl < 5 || iph->version != 4)
471 if (!pskb_may_pull(skb, iph->ihl*4))
473 if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
476 len = ntohs(iph->tot_len);
477 if (skb->len < len || len < iph->ihl*4)
481 * Our transport medium may have padded the buffer out.
482 * Now We trim to the true length of the frame.
484 if (pskb_trim_rcsum(skb, len))
487 if (iph->protocol != IPPROTO_UDP)
491 uh = (struct udphdr *)(((char *)iph) + iph->ihl*4);
492 ulen = ntohs(uh->len);
496 if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr))
498 if (np->local_ip && np->local_ip != ntohl(iph->daddr))
500 if (np->remote_ip && np->remote_ip != ntohl(iph->saddr))
502 if (np->local_port && np->local_port != ntohs(uh->dest))
505 np->rx_hook(np, ntohs(uh->source),
507 ulen - sizeof(struct udphdr));
513 if (atomic_read(&trapped)) {
521 int netpoll_parse_options(struct netpoll *np, char *opt)
523 char *cur=opt, *delim;
526 if ((delim = strchr(cur, '@')) == NULL)
529 np->local_port = simple_strtol(cur, NULL, 10);
533 printk(KERN_INFO "%s: local port %d\n", np->name, np->local_port);
536 if ((delim = strchr(cur, '/')) == NULL)
539 np->local_ip = ntohl(in_aton(cur));
542 printk(KERN_INFO "%s: local IP %d.%d.%d.%d\n",
543 np->name, HIPQUAD(np->local_ip));
548 /* parse out dev name */
549 if ((delim = strchr(cur, ',')) == NULL)
552 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
557 printk(KERN_INFO "%s: interface %s\n", np->name, np->dev_name);
561 if ((delim = strchr(cur, '@')) == NULL)
564 np->remote_port = simple_strtol(cur, NULL, 10);
568 printk(KERN_INFO "%s: remote port %d\n", np->name, np->remote_port);
571 if ((delim = strchr(cur, '/')) == NULL)
574 np->remote_ip = ntohl(in_aton(cur));
577 printk(KERN_INFO "%s: remote IP %d.%d.%d.%d\n",
578 np->name, HIPQUAD(np->remote_ip));
582 if ((delim = strchr(cur, ':')) == NULL)
585 np->remote_mac[0] = simple_strtol(cur, NULL, 16);
587 if ((delim = strchr(cur, ':')) == NULL)
590 np->remote_mac[1] = simple_strtol(cur, NULL, 16);
592 if ((delim = strchr(cur, ':')) == NULL)
595 np->remote_mac[2] = simple_strtol(cur, NULL, 16);
597 if ((delim = strchr(cur, ':')) == NULL)
600 np->remote_mac[3] = simple_strtol(cur, NULL, 16);
602 if ((delim = strchr(cur, ':')) == NULL)
605 np->remote_mac[4] = simple_strtol(cur, NULL, 16);
607 np->remote_mac[5] = simple_strtol(cur, NULL, 16);
610 printk(KERN_INFO "%s: remote ethernet address "
611 "%02x:%02x:%02x:%02x:%02x:%02x\n",
623 printk(KERN_INFO "%s: couldn't parse config at %s!\n",
628 int netpoll_setup(struct netpoll *np)
630 struct net_device *ndev = NULL;
631 struct in_device *in_dev;
632 struct netpoll_info *npinfo;
637 ndev = dev_get_by_name(np->dev_name);
639 printk(KERN_ERR "%s: %s doesn't exist, aborting.\n",
640 np->name, np->dev_name);
646 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
652 npinfo->rx_flags = 0;
653 npinfo->rx_np = NULL;
654 spin_lock_init(&npinfo->poll_lock);
655 npinfo->poll_owner = -1;
657 spin_lock_init(&npinfo->rx_lock);
658 skb_queue_head_init(&npinfo->arp_tx);
659 skb_queue_head_init(&npinfo->txq);
660 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
662 atomic_set(&npinfo->refcnt, 1);
664 npinfo = ndev->npinfo;
665 atomic_inc(&npinfo->refcnt);
668 if (!ndev->poll_controller) {
669 printk(KERN_ERR "%s: %s doesn't support polling, aborting.\n",
670 np->name, np->dev_name);
675 if (!netif_running(ndev)) {
676 unsigned long atmost, atleast;
678 printk(KERN_INFO "%s: device %s not up yet, forcing it\n",
679 np->name, np->dev_name);
682 err = dev_open(ndev);
686 printk(KERN_ERR "%s: failed to open %s\n",
687 np->name, ndev->name);
691 atleast = jiffies + HZ/10;
692 atmost = jiffies + 4*HZ;
693 while (!netif_carrier_ok(ndev)) {
694 if (time_after(jiffies, atmost)) {
696 "%s: timeout waiting for carrier\n",
703 /* If carrier appears to come up instantly, we don't
704 * trust it and pause so that we don't pump all our
705 * queued console messages into the bitbucket.
708 if (time_before(jiffies, atleast)) {
709 printk(KERN_NOTICE "%s: carrier detect appears"
710 " untrustworthy, waiting 4 seconds\n",
716 if (is_zero_ether_addr(np->local_mac) && ndev->dev_addr)
717 memcpy(np->local_mac, ndev->dev_addr, 6);
721 in_dev = __in_dev_get_rcu(ndev);
723 if (!in_dev || !in_dev->ifa_list) {
725 printk(KERN_ERR "%s: no IP address for %s, aborting\n",
726 np->name, np->dev_name);
731 np->local_ip = ntohl(in_dev->ifa_list->ifa_local);
733 printk(KERN_INFO "%s: local IP %d.%d.%d.%d\n",
734 np->name, HIPQUAD(np->local_ip));
738 spin_lock_irqsave(&npinfo->rx_lock, flags);
739 npinfo->rx_flags |= NETPOLL_RX_ENABLED;
741 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
744 /* fill up the skb queue */
747 /* last thing to do is link it to the net device structure */
748 ndev->npinfo = npinfo;
750 /* avoid racing with NAPI reading npinfo */
763 static int __init netpoll_init(void)
765 skb_queue_head_init(&skb_pool);
768 core_initcall(netpoll_init);
770 void netpoll_cleanup(struct netpoll *np)
772 struct netpoll_info *npinfo;
776 npinfo = np->dev->npinfo;
778 if (npinfo->rx_np == np) {
779 spin_lock_irqsave(&npinfo->rx_lock, flags);
780 npinfo->rx_np = NULL;
781 npinfo->rx_flags &= ~NETPOLL_RX_ENABLED;
782 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
785 np->dev->npinfo = NULL;
786 if (atomic_dec_and_test(&npinfo->refcnt)) {
787 skb_queue_purge(&npinfo->arp_tx);
788 skb_queue_purge(&npinfo->txq);
789 cancel_delayed_work(&npinfo->tx_work);
790 flush_scheduled_work();
792 /* clean after last, unfinished work */
793 if (!skb_queue_empty(&npinfo->txq)) {
795 skb = __skb_dequeue(&npinfo->txq);
808 int netpoll_trap(void)
810 return atomic_read(&trapped);
813 void netpoll_set_trap(int trap)
816 atomic_inc(&trapped);
818 atomic_dec(&trapped);
821 EXPORT_SYMBOL(netpoll_set_trap);
822 EXPORT_SYMBOL(netpoll_trap);
823 EXPORT_SYMBOL(netpoll_parse_options);
824 EXPORT_SYMBOL(netpoll_setup);
825 EXPORT_SYMBOL(netpoll_cleanup);
826 EXPORT_SYMBOL(netpoll_send_udp);
827 EXPORT_SYMBOL(netpoll_poll);