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 netif_subqueue_stopped(dev, skb->queue_mapping)) ||
71 dev->hard_start_xmit(skb, dev) != NETDEV_TX_OK) {
72 skb_queue_head(&npinfo->txq, skb);
74 local_irq_restore(flags);
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;
122 struct napi_struct *napi;
125 list_for_each_entry(napi, &np->dev->napi_list, dev_list) {
126 if (test_bit(NAPI_STATE_SCHED, &napi->state) &&
127 napi->poll_owner != smp_processor_id() &&
128 spin_trylock(&napi->poll_lock)) {
129 npinfo->rx_flags |= NETPOLL_RX_DROP;
130 atomic_inc(&trapped);
132 napi->poll(napi, budget);
134 atomic_dec(&trapped);
135 npinfo->rx_flags &= ~NETPOLL_RX_DROP;
136 spin_unlock(&napi->poll_lock);
141 static void service_arp_queue(struct netpoll_info *npi)
148 skb = skb_dequeue(&npi->arp_tx);
150 while (skb != NULL) {
152 skb = skb_dequeue(&npi->arp_tx);
156 void netpoll_poll(struct netpoll *np)
158 if (!np->dev || !netif_running(np->dev) || !np->dev->poll_controller)
161 /* Process pending work on NIC */
162 np->dev->poll_controller(np->dev);
163 if (!list_empty(&np->dev->napi_list))
166 service_arp_queue(np->dev->npinfo);
168 zap_completion_queue();
171 static void refill_skbs(void)
176 spin_lock_irqsave(&skb_pool.lock, flags);
177 while (skb_pool.qlen < MAX_SKBS) {
178 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
182 __skb_queue_tail(&skb_pool, skb);
184 spin_unlock_irqrestore(&skb_pool.lock, flags);
187 static void zap_completion_queue(void)
190 struct softnet_data *sd = &get_cpu_var(softnet_data);
192 if (sd->completion_queue) {
193 struct sk_buff *clist;
195 local_irq_save(flags);
196 clist = sd->completion_queue;
197 sd->completion_queue = NULL;
198 local_irq_restore(flags);
200 while (clist != NULL) {
201 struct sk_buff *skb = clist;
204 dev_kfree_skb_any(skb); /* put this one back */
210 put_cpu_var(softnet_data);
213 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
218 zap_completion_queue();
222 skb = alloc_skb(len, GFP_ATOMIC);
224 skb = skb_dequeue(&skb_pool);
234 atomic_set(&skb->users, 1);
235 skb_reserve(skb, reserve);
239 static int netpoll_owner_active(struct net_device *dev)
241 struct napi_struct *napi;
243 list_for_each_entry(napi, &dev->napi_list, dev_list) {
244 if (napi->poll_owner == smp_processor_id())
250 static void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
252 int status = NETDEV_TX_BUSY;
254 struct net_device *dev = np->dev;
255 struct netpoll_info *npinfo = np->dev->npinfo;
257 if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
262 /* don't get messages out of order, and no recursion */
263 if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
266 local_irq_save(flags);
267 /* try until next clock tick */
268 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
269 tries > 0; --tries) {
270 if (netif_tx_trylock(dev)) {
271 if (!netif_queue_stopped(dev) &&
272 !netif_subqueue_stopped(dev, skb->queue_mapping))
273 status = dev->hard_start_xmit(skb, dev);
274 netif_tx_unlock(dev);
276 if (status == NETDEV_TX_OK)
281 /* tickle device maybe there is some cleanup */
284 udelay(USEC_PER_POLL);
286 local_irq_restore(flags);
289 if (status != NETDEV_TX_OK) {
290 skb_queue_tail(&npinfo->txq, skb);
291 schedule_delayed_work(&npinfo->tx_work,0);
295 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
297 int total_len, eth_len, ip_len, udp_len;
303 udp_len = len + sizeof(*udph);
304 ip_len = eth_len = udp_len + sizeof(*iph);
305 total_len = eth_len + ETH_HLEN + NET_IP_ALIGN;
307 skb = find_skb(np, total_len, total_len - len);
311 skb_copy_to_linear_data(skb, msg, len);
314 skb_push(skb, sizeof(*udph));
315 skb_reset_transport_header(skb);
317 udph->source = htons(np->local_port);
318 udph->dest = htons(np->remote_port);
319 udph->len = htons(udp_len);
321 udph->check = csum_tcpudp_magic(htonl(np->local_ip),
322 htonl(np->remote_ip),
323 udp_len, IPPROTO_UDP,
324 csum_partial((unsigned char *)udph, udp_len, 0));
325 if (udph->check == 0)
326 udph->check = CSUM_MANGLED_0;
328 skb_push(skb, sizeof(*iph));
329 skb_reset_network_header(skb);
332 /* iph->version = 4; iph->ihl = 5; */
333 put_unaligned(0x45, (unsigned char *)iph);
335 put_unaligned(htons(ip_len), &(iph->tot_len));
339 iph->protocol = IPPROTO_UDP;
341 put_unaligned(htonl(np->local_ip), &(iph->saddr));
342 put_unaligned(htonl(np->remote_ip), &(iph->daddr));
343 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
345 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
346 skb_reset_mac_header(skb);
347 skb->protocol = eth->h_proto = htons(ETH_P_IP);
348 memcpy(eth->h_source, np->local_mac, 6);
349 memcpy(eth->h_dest, np->remote_mac, 6);
353 netpoll_send_skb(np, skb);
356 static void arp_reply(struct sk_buff *skb)
358 struct netpoll_info *npinfo = skb->dev->npinfo;
360 unsigned char *arp_ptr;
361 int size, type = ARPOP_REPLY, ptype = ETH_P_ARP;
364 struct sk_buff *send_skb;
365 struct netpoll *np = NULL;
367 if (npinfo->rx_np && npinfo->rx_np->dev == skb->dev)
372 /* No arp on this interface */
373 if (skb->dev->flags & IFF_NOARP)
376 if (!pskb_may_pull(skb, (sizeof(struct arphdr) +
377 (2 * skb->dev->addr_len) +
381 skb_reset_network_header(skb);
382 skb_reset_transport_header(skb);
385 if ((arp->ar_hrd != htons(ARPHRD_ETHER) &&
386 arp->ar_hrd != htons(ARPHRD_IEEE802)) ||
387 arp->ar_pro != htons(ETH_P_IP) ||
388 arp->ar_op != htons(ARPOP_REQUEST))
391 arp_ptr = (unsigned char *)(arp+1);
392 /* save the location of the src hw addr */
394 arp_ptr += skb->dev->addr_len;
395 memcpy(&sip, arp_ptr, 4);
397 /* if we actually cared about dst hw addr, it would get copied here */
398 arp_ptr += skb->dev->addr_len;
399 memcpy(&tip, arp_ptr, 4);
401 /* Should we ignore arp? */
402 if (tip != htonl(np->local_ip) || LOOPBACK(tip) || MULTICAST(tip))
405 size = sizeof(struct arphdr) + 2 * (skb->dev->addr_len + 4);
406 send_skb = find_skb(np, size + LL_RESERVED_SPACE(np->dev),
407 LL_RESERVED_SPACE(np->dev));
412 skb_reset_network_header(send_skb);
413 arp = (struct arphdr *) skb_put(send_skb, size);
414 send_skb->dev = skb->dev;
415 send_skb->protocol = htons(ETH_P_ARP);
417 /* Fill the device header for the ARP frame */
418 if (dev_hard_header(send_skb, skb->dev, ptype,
420 send_skb->len) < 0) {
426 * Fill out the arp protocol part.
428 * we only support ethernet device type,
429 * which (according to RFC 1390) should always equal 1 (Ethernet).
432 arp->ar_hrd = htons(np->dev->type);
433 arp->ar_pro = htons(ETH_P_IP);
434 arp->ar_hln = np->dev->addr_len;
436 arp->ar_op = htons(type);
438 arp_ptr=(unsigned char *)(arp + 1);
439 memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
440 arp_ptr += np->dev->addr_len;
441 memcpy(arp_ptr, &tip, 4);
443 memcpy(arp_ptr, sha, np->dev->addr_len);
444 arp_ptr += np->dev->addr_len;
445 memcpy(arp_ptr, &sip, 4);
447 netpoll_send_skb(np, send_skb);
450 int __netpoll_rx(struct sk_buff *skb)
452 int proto, len, ulen;
455 struct netpoll_info *npi = skb->dev->npinfo;
456 struct netpoll *np = npi->rx_np;
460 if (skb->dev->type != ARPHRD_ETHER)
463 /* check if netpoll clients need ARP */
464 if (skb->protocol == htons(ETH_P_ARP) &&
465 atomic_read(&trapped)) {
466 skb_queue_tail(&npi->arp_tx, skb);
470 proto = ntohs(eth_hdr(skb)->h_proto);
471 if (proto != ETH_P_IP)
473 if (skb->pkt_type == PACKET_OTHERHOST)
478 iph = (struct iphdr *)skb->data;
479 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
481 if (iph->ihl < 5 || iph->version != 4)
483 if (!pskb_may_pull(skb, iph->ihl*4))
485 if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
488 len = ntohs(iph->tot_len);
489 if (skb->len < len || len < iph->ihl*4)
493 * Our transport medium may have padded the buffer out.
494 * Now We trim to the true length of the frame.
496 if (pskb_trim_rcsum(skb, len))
499 if (iph->protocol != IPPROTO_UDP)
503 uh = (struct udphdr *)(((char *)iph) + iph->ihl*4);
504 ulen = ntohs(uh->len);
508 if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr))
510 if (np->local_ip && np->local_ip != ntohl(iph->daddr))
512 if (np->remote_ip && np->remote_ip != ntohl(iph->saddr))
514 if (np->local_port && np->local_port != ntohs(uh->dest))
517 np->rx_hook(np, ntohs(uh->source),
519 ulen - sizeof(struct udphdr));
525 if (atomic_read(&trapped)) {
533 void netpoll_print_options(struct netpoll *np)
535 DECLARE_MAC_BUF(mac);
536 printk(KERN_INFO "%s: local port %d\n",
537 np->name, np->local_port);
538 printk(KERN_INFO "%s: local IP %d.%d.%d.%d\n",
539 np->name, HIPQUAD(np->local_ip));
540 printk(KERN_INFO "%s: interface %s\n",
541 np->name, np->dev_name);
542 printk(KERN_INFO "%s: remote port %d\n",
543 np->name, np->remote_port);
544 printk(KERN_INFO "%s: remote IP %d.%d.%d.%d\n",
545 np->name, HIPQUAD(np->remote_ip));
546 printk(KERN_INFO "%s: remote ethernet address %s\n",
547 np->name, print_mac(mac, np->remote_mac));
550 int netpoll_parse_options(struct netpoll *np, char *opt)
552 char *cur=opt, *delim;
555 if ((delim = strchr(cur, '@')) == NULL)
558 np->local_port = simple_strtol(cur, NULL, 10);
564 if ((delim = strchr(cur, '/')) == NULL)
567 np->local_ip = ntohl(in_aton(cur));
573 /* parse out dev name */
574 if ((delim = strchr(cur, ',')) == NULL)
577 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
584 if ((delim = strchr(cur, '@')) == NULL)
587 np->remote_port = simple_strtol(cur, NULL, 10);
593 if ((delim = strchr(cur, '/')) == NULL)
596 np->remote_ip = ntohl(in_aton(cur));
601 if ((delim = strchr(cur, ':')) == NULL)
604 np->remote_mac[0] = simple_strtol(cur, NULL, 16);
606 if ((delim = strchr(cur, ':')) == NULL)
609 np->remote_mac[1] = simple_strtol(cur, NULL, 16);
611 if ((delim = strchr(cur, ':')) == NULL)
614 np->remote_mac[2] = simple_strtol(cur, NULL, 16);
616 if ((delim = strchr(cur, ':')) == NULL)
619 np->remote_mac[3] = simple_strtol(cur, NULL, 16);
621 if ((delim = strchr(cur, ':')) == NULL)
624 np->remote_mac[4] = simple_strtol(cur, NULL, 16);
626 np->remote_mac[5] = simple_strtol(cur, NULL, 16);
629 netpoll_print_options(np);
634 printk(KERN_INFO "%s: couldn't parse config at %s!\n",
639 int netpoll_setup(struct netpoll *np)
641 struct net_device *ndev = NULL;
642 struct in_device *in_dev;
643 struct netpoll_info *npinfo;
648 ndev = dev_get_by_name(&init_net, np->dev_name);
650 printk(KERN_ERR "%s: %s doesn't exist, aborting.\n",
651 np->name, np->dev_name);
657 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
663 npinfo->rx_flags = 0;
664 npinfo->rx_np = NULL;
666 spin_lock_init(&npinfo->rx_lock);
667 skb_queue_head_init(&npinfo->arp_tx);
668 skb_queue_head_init(&npinfo->txq);
669 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
671 atomic_set(&npinfo->refcnt, 1);
673 npinfo = ndev->npinfo;
674 atomic_inc(&npinfo->refcnt);
677 if (!ndev->poll_controller) {
678 printk(KERN_ERR "%s: %s doesn't support polling, aborting.\n",
679 np->name, np->dev_name);
684 if (!netif_running(ndev)) {
685 unsigned long atmost, atleast;
687 printk(KERN_INFO "%s: device %s not up yet, forcing it\n",
688 np->name, np->dev_name);
691 err = dev_open(ndev);
695 printk(KERN_ERR "%s: failed to open %s\n",
696 np->name, ndev->name);
700 atleast = jiffies + HZ/10;
701 atmost = jiffies + 4*HZ;
702 while (!netif_carrier_ok(ndev)) {
703 if (time_after(jiffies, atmost)) {
705 "%s: timeout waiting for carrier\n",
712 /* If carrier appears to come up instantly, we don't
713 * trust it and pause so that we don't pump all our
714 * queued console messages into the bitbucket.
717 if (time_before(jiffies, atleast)) {
718 printk(KERN_NOTICE "%s: carrier detect appears"
719 " untrustworthy, waiting 4 seconds\n",
725 if (is_zero_ether_addr(np->local_mac) && ndev->dev_addr)
726 memcpy(np->local_mac, ndev->dev_addr, 6);
730 in_dev = __in_dev_get_rcu(ndev);
732 if (!in_dev || !in_dev->ifa_list) {
734 printk(KERN_ERR "%s: no IP address for %s, aborting\n",
735 np->name, np->dev_name);
740 np->local_ip = ntohl(in_dev->ifa_list->ifa_local);
742 printk(KERN_INFO "%s: local IP %d.%d.%d.%d\n",
743 np->name, HIPQUAD(np->local_ip));
747 spin_lock_irqsave(&npinfo->rx_lock, flags);
748 npinfo->rx_flags |= NETPOLL_RX_ENABLED;
750 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
753 /* fill up the skb queue */
756 /* last thing to do is link it to the net device structure */
757 ndev->npinfo = npinfo;
759 /* avoid racing with NAPI reading npinfo */
772 static int __init netpoll_init(void)
774 skb_queue_head_init(&skb_pool);
777 core_initcall(netpoll_init);
779 void netpoll_cleanup(struct netpoll *np)
781 struct netpoll_info *npinfo;
785 npinfo = np->dev->npinfo;
787 if (npinfo->rx_np == np) {
788 spin_lock_irqsave(&npinfo->rx_lock, flags);
789 npinfo->rx_np = NULL;
790 npinfo->rx_flags &= ~NETPOLL_RX_ENABLED;
791 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
794 if (atomic_dec_and_test(&npinfo->refcnt)) {
795 skb_queue_purge(&npinfo->arp_tx);
796 skb_queue_purge(&npinfo->txq);
797 cancel_rearming_delayed_work(&npinfo->tx_work);
799 /* clean after last, unfinished work */
800 if (!skb_queue_empty(&npinfo->txq)) {
802 skb = __skb_dequeue(&npinfo->txq);
806 np->dev->npinfo = NULL;
816 int netpoll_trap(void)
818 return atomic_read(&trapped);
821 void netpoll_set_trap(int trap)
824 atomic_inc(&trapped);
826 atomic_dec(&trapped);
829 EXPORT_SYMBOL(netpoll_set_trap);
830 EXPORT_SYMBOL(netpoll_trap);
831 EXPORT_SYMBOL(netpoll_print_options);
832 EXPORT_SYMBOL(netpoll_parse_options);
833 EXPORT_SYMBOL(netpoll_setup);
834 EXPORT_SYMBOL(netpoll_cleanup);
835 EXPORT_SYMBOL(netpoll_send_udp);
836 EXPORT_SYMBOL(netpoll_poll);