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;
61 const struct net_device_ops *ops = dev->netdev_ops;
62 struct netdev_queue *txq;
64 if (!netif_device_present(dev) || !netif_running(dev)) {
69 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
71 local_irq_save(flags);
72 __netif_tx_lock(txq, smp_processor_id());
73 if (netif_tx_queue_stopped(txq) ||
74 netif_tx_queue_frozen(txq) ||
75 ops->ndo_start_xmit(skb, dev) != NETDEV_TX_OK) {
76 skb_queue_head(&npinfo->txq, skb);
77 __netif_tx_unlock(txq);
78 local_irq_restore(flags);
80 schedule_delayed_work(&npinfo->tx_work, HZ/10);
83 __netif_tx_unlock(txq);
84 local_irq_restore(flags);
88 static __sum16 checksum_udp(struct sk_buff *skb, struct udphdr *uh,
89 unsigned short ulen, __be32 saddr, __be32 daddr)
93 if (uh->check == 0 || skb_csum_unnecessary(skb))
96 psum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0);
98 if (skb->ip_summed == CHECKSUM_COMPLETE &&
99 !csum_fold(csum_add(psum, skb->csum)))
104 return __skb_checksum_complete(skb);
108 * Check whether delayed processing was scheduled for our NIC. If so,
109 * we attempt to grab the poll lock and use ->poll() to pump the card.
110 * If this fails, either we've recursed in ->poll() or it's already
111 * running on another CPU.
113 * Note: we don't mask interrupts with this lock because we're using
114 * trylock here and interrupts are already disabled in the softirq
115 * case. Further, we test the poll_owner to avoid recursion on UP
116 * systems where the lock doesn't exist.
118 * In cases where there is bi-directional communications, reading only
119 * one message at a time can lead to packets being dropped by the
120 * network adapter, forcing superfluous retries and possibly timeouts.
121 * Thus, we set our budget to greater than 1.
123 static int poll_one_napi(struct netpoll_info *npinfo,
124 struct napi_struct *napi, int budget)
128 /* net_rx_action's ->poll() invocations and our's are
129 * synchronized by this test which is only made while
130 * holding the napi->poll_lock.
132 if (!test_bit(NAPI_STATE_SCHED, &napi->state))
135 npinfo->rx_flags |= NETPOLL_RX_DROP;
136 atomic_inc(&trapped);
137 set_bit(NAPI_STATE_NPSVC, &napi->state);
139 work = napi->poll(napi, budget);
141 clear_bit(NAPI_STATE_NPSVC, &napi->state);
142 atomic_dec(&trapped);
143 npinfo->rx_flags &= ~NETPOLL_RX_DROP;
145 return budget - work;
148 static void poll_napi(struct net_device *dev)
150 struct napi_struct *napi;
153 list_for_each_entry(napi, &dev->napi_list, dev_list) {
154 if (napi->poll_owner != smp_processor_id() &&
155 spin_trylock(&napi->poll_lock)) {
156 budget = poll_one_napi(dev->npinfo, napi, budget);
157 spin_unlock(&napi->poll_lock);
165 static void service_arp_queue(struct netpoll_info *npi)
170 while ((skb = skb_dequeue(&npi->arp_tx)))
175 void netpoll_poll(struct netpoll *np)
177 struct net_device *dev = np->dev;
178 const struct net_device_ops *ops;
180 if (!dev || !netif_running(dev))
183 ops = dev->netdev_ops;
184 if (!ops->ndo_poll_controller)
187 /* Process pending work on NIC */
188 ops->ndo_poll_controller(dev);
192 service_arp_queue(dev->npinfo);
194 zap_completion_queue();
197 static void refill_skbs(void)
202 spin_lock_irqsave(&skb_pool.lock, flags);
203 while (skb_pool.qlen < MAX_SKBS) {
204 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
208 __skb_queue_tail(&skb_pool, skb);
210 spin_unlock_irqrestore(&skb_pool.lock, flags);
213 static void zap_completion_queue(void)
216 struct softnet_data *sd = &get_cpu_var(softnet_data);
218 if (sd->completion_queue) {
219 struct sk_buff *clist;
221 local_irq_save(flags);
222 clist = sd->completion_queue;
223 sd->completion_queue = NULL;
224 local_irq_restore(flags);
226 while (clist != NULL) {
227 struct sk_buff *skb = clist;
229 if (skb->destructor) {
230 atomic_inc(&skb->users);
231 dev_kfree_skb_any(skb); /* put this one back */
238 put_cpu_var(softnet_data);
241 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
246 zap_completion_queue();
250 skb = alloc_skb(len, GFP_ATOMIC);
252 skb = skb_dequeue(&skb_pool);
262 atomic_set(&skb->users, 1);
263 skb_reserve(skb, reserve);
267 static int netpoll_owner_active(struct net_device *dev)
269 struct napi_struct *napi;
271 list_for_each_entry(napi, &dev->napi_list, dev_list) {
272 if (napi->poll_owner == smp_processor_id())
278 static void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
280 int status = NETDEV_TX_BUSY;
282 struct net_device *dev = np->dev;
283 const struct net_device_ops *ops = dev->netdev_ops;
284 struct netpoll_info *npinfo = np->dev->npinfo;
286 if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
291 /* don't get messages out of order, and no recursion */
292 if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
293 struct netdev_queue *txq;
296 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
298 local_irq_save(flags);
299 /* try until next clock tick */
300 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
301 tries > 0; --tries) {
302 if (__netif_tx_trylock(txq)) {
303 if (!netif_tx_queue_stopped(txq))
304 status = ops->ndo_start_xmit(skb, dev);
305 __netif_tx_unlock(txq);
307 if (status == NETDEV_TX_OK)
312 /* tickle device maybe there is some cleanup */
315 udelay(USEC_PER_POLL);
317 local_irq_restore(flags);
320 if (status != NETDEV_TX_OK) {
321 skb_queue_tail(&npinfo->txq, skb);
322 schedule_delayed_work(&npinfo->tx_work,0);
326 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
328 int total_len, eth_len, ip_len, udp_len;
334 udp_len = len + sizeof(*udph);
335 ip_len = eth_len = udp_len + sizeof(*iph);
336 total_len = eth_len + ETH_HLEN + NET_IP_ALIGN;
338 skb = find_skb(np, total_len, total_len - len);
342 skb_copy_to_linear_data(skb, msg, len);
345 skb_push(skb, sizeof(*udph));
346 skb_reset_transport_header(skb);
348 udph->source = htons(np->local_port);
349 udph->dest = htons(np->remote_port);
350 udph->len = htons(udp_len);
352 udph->check = csum_tcpudp_magic(np->local_ip,
354 udp_len, IPPROTO_UDP,
355 csum_partial(udph, udp_len, 0));
356 if (udph->check == 0)
357 udph->check = CSUM_MANGLED_0;
359 skb_push(skb, sizeof(*iph));
360 skb_reset_network_header(skb);
363 /* iph->version = 4; iph->ihl = 5; */
364 put_unaligned(0x45, (unsigned char *)iph);
366 put_unaligned(htons(ip_len), &(iph->tot_len));
370 iph->protocol = IPPROTO_UDP;
372 put_unaligned(np->local_ip, &(iph->saddr));
373 put_unaligned(np->remote_ip, &(iph->daddr));
374 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
376 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
377 skb_reset_mac_header(skb);
378 skb->protocol = eth->h_proto = htons(ETH_P_IP);
379 memcpy(eth->h_source, np->dev->dev_addr, ETH_ALEN);
380 memcpy(eth->h_dest, np->remote_mac, ETH_ALEN);
384 netpoll_send_skb(np, skb);
387 static void arp_reply(struct sk_buff *skb)
389 struct netpoll_info *npinfo = skb->dev->npinfo;
391 unsigned char *arp_ptr;
392 int size, type = ARPOP_REPLY, ptype = ETH_P_ARP;
395 struct sk_buff *send_skb;
396 struct netpoll *np = NULL;
398 if (npinfo->rx_np && npinfo->rx_np->dev == skb->dev)
403 /* No arp on this interface */
404 if (skb->dev->flags & IFF_NOARP)
407 if (!pskb_may_pull(skb, arp_hdr_len(skb->dev)))
410 skb_reset_network_header(skb);
411 skb_reset_transport_header(skb);
414 if ((arp->ar_hrd != htons(ARPHRD_ETHER) &&
415 arp->ar_hrd != htons(ARPHRD_IEEE802)) ||
416 arp->ar_pro != htons(ETH_P_IP) ||
417 arp->ar_op != htons(ARPOP_REQUEST))
420 arp_ptr = (unsigned char *)(arp+1);
421 /* save the location of the src hw addr */
423 arp_ptr += skb->dev->addr_len;
424 memcpy(&sip, arp_ptr, 4);
426 /* if we actually cared about dst hw addr, it would get copied here */
427 arp_ptr += skb->dev->addr_len;
428 memcpy(&tip, arp_ptr, 4);
430 /* Should we ignore arp? */
431 if (tip != np->local_ip ||
432 ipv4_is_loopback(tip) || ipv4_is_multicast(tip))
435 size = arp_hdr_len(skb->dev);
436 send_skb = find_skb(np, size + LL_ALLOCATED_SPACE(np->dev),
437 LL_RESERVED_SPACE(np->dev));
442 skb_reset_network_header(send_skb);
443 arp = (struct arphdr *) skb_put(send_skb, size);
444 send_skb->dev = skb->dev;
445 send_skb->protocol = htons(ETH_P_ARP);
447 /* Fill the device header for the ARP frame */
448 if (dev_hard_header(send_skb, skb->dev, ptype,
449 sha, np->dev->dev_addr,
450 send_skb->len) < 0) {
456 * Fill out the arp protocol part.
458 * we only support ethernet device type,
459 * which (according to RFC 1390) should always equal 1 (Ethernet).
462 arp->ar_hrd = htons(np->dev->type);
463 arp->ar_pro = htons(ETH_P_IP);
464 arp->ar_hln = np->dev->addr_len;
466 arp->ar_op = htons(type);
468 arp_ptr=(unsigned char *)(arp + 1);
469 memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
470 arp_ptr += np->dev->addr_len;
471 memcpy(arp_ptr, &tip, 4);
473 memcpy(arp_ptr, sha, np->dev->addr_len);
474 arp_ptr += np->dev->addr_len;
475 memcpy(arp_ptr, &sip, 4);
477 netpoll_send_skb(np, send_skb);
480 int __netpoll_rx(struct sk_buff *skb)
482 int proto, len, ulen;
485 struct netpoll_info *npi = skb->dev->npinfo;
486 struct netpoll *np = npi->rx_np;
490 if (skb->dev->type != ARPHRD_ETHER)
493 /* check if netpoll clients need ARP */
494 if (skb->protocol == htons(ETH_P_ARP) &&
495 atomic_read(&trapped)) {
496 skb_queue_tail(&npi->arp_tx, skb);
500 proto = ntohs(eth_hdr(skb)->h_proto);
501 if (proto != ETH_P_IP)
503 if (skb->pkt_type == PACKET_OTHERHOST)
508 iph = (struct iphdr *)skb->data;
509 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
511 if (iph->ihl < 5 || iph->version != 4)
513 if (!pskb_may_pull(skb, iph->ihl*4))
515 if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
518 len = ntohs(iph->tot_len);
519 if (skb->len < len || len < iph->ihl*4)
523 * Our transport medium may have padded the buffer out.
524 * Now We trim to the true length of the frame.
526 if (pskb_trim_rcsum(skb, len))
529 if (iph->protocol != IPPROTO_UDP)
533 uh = (struct udphdr *)(((char *)iph) + iph->ihl*4);
534 ulen = ntohs(uh->len);
538 if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr))
540 if (np->local_ip && np->local_ip != iph->daddr)
542 if (np->remote_ip && np->remote_ip != iph->saddr)
544 if (np->local_port && np->local_port != ntohs(uh->dest))
547 np->rx_hook(np, ntohs(uh->source),
549 ulen - sizeof(struct udphdr));
555 if (atomic_read(&trapped)) {
563 void netpoll_print_options(struct netpoll *np)
565 printk(KERN_INFO "%s: local port %d\n",
566 np->name, np->local_port);
567 printk(KERN_INFO "%s: local IP %pI4\n",
568 np->name, &np->local_ip);
569 printk(KERN_INFO "%s: interface %s\n",
570 np->name, np->dev_name);
571 printk(KERN_INFO "%s: remote port %d\n",
572 np->name, np->remote_port);
573 printk(KERN_INFO "%s: remote IP %pI4\n",
574 np->name, &np->remote_ip);
575 printk(KERN_INFO "%s: remote ethernet address %pM\n",
576 np->name, np->remote_mac);
579 int netpoll_parse_options(struct netpoll *np, char *opt)
581 char *cur=opt, *delim;
584 if ((delim = strchr(cur, '@')) == NULL)
587 np->local_port = simple_strtol(cur, NULL, 10);
593 if ((delim = strchr(cur, '/')) == NULL)
596 np->local_ip = in_aton(cur);
602 /* parse out dev name */
603 if ((delim = strchr(cur, ',')) == NULL)
606 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
613 if ((delim = strchr(cur, '@')) == NULL)
616 np->remote_port = simple_strtol(cur, NULL, 10);
622 if ((delim = strchr(cur, '/')) == NULL)
625 np->remote_ip = in_aton(cur);
630 if ((delim = strchr(cur, ':')) == NULL)
633 np->remote_mac[0] = simple_strtol(cur, NULL, 16);
635 if ((delim = strchr(cur, ':')) == NULL)
638 np->remote_mac[1] = simple_strtol(cur, NULL, 16);
640 if ((delim = strchr(cur, ':')) == NULL)
643 np->remote_mac[2] = simple_strtol(cur, NULL, 16);
645 if ((delim = strchr(cur, ':')) == NULL)
648 np->remote_mac[3] = simple_strtol(cur, NULL, 16);
650 if ((delim = strchr(cur, ':')) == NULL)
653 np->remote_mac[4] = simple_strtol(cur, NULL, 16);
655 np->remote_mac[5] = simple_strtol(cur, NULL, 16);
658 netpoll_print_options(np);
663 printk(KERN_INFO "%s: couldn't parse config at %s!\n",
668 int netpoll_setup(struct netpoll *np)
670 struct net_device *ndev = NULL;
671 struct in_device *in_dev;
672 struct netpoll_info *npinfo;
677 ndev = dev_get_by_name(&init_net, np->dev_name);
679 printk(KERN_ERR "%s: %s doesn't exist, aborting.\n",
680 np->name, np->dev_name);
686 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
692 npinfo->rx_flags = 0;
693 npinfo->rx_np = NULL;
695 spin_lock_init(&npinfo->rx_lock);
696 skb_queue_head_init(&npinfo->arp_tx);
697 skb_queue_head_init(&npinfo->txq);
698 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
700 atomic_set(&npinfo->refcnt, 1);
702 npinfo = ndev->npinfo;
703 atomic_inc(&npinfo->refcnt);
706 if (!ndev->netdev_ops->ndo_poll_controller) {
707 printk(KERN_ERR "%s: %s doesn't support polling, aborting.\n",
708 np->name, np->dev_name);
713 if (!netif_running(ndev)) {
714 unsigned long atmost, atleast;
716 printk(KERN_INFO "%s: device %s not up yet, forcing it\n",
717 np->name, np->dev_name);
720 err = dev_open(ndev);
724 printk(KERN_ERR "%s: failed to open %s\n",
725 np->name, ndev->name);
729 atleast = jiffies + HZ/10;
730 atmost = jiffies + 4*HZ;
731 while (!netif_carrier_ok(ndev)) {
732 if (time_after(jiffies, atmost)) {
734 "%s: timeout waiting for carrier\n",
741 /* If carrier appears to come up instantly, we don't
742 * trust it and pause so that we don't pump all our
743 * queued console messages into the bitbucket.
746 if (time_before(jiffies, atleast)) {
747 printk(KERN_NOTICE "%s: carrier detect appears"
748 " untrustworthy, waiting 4 seconds\n",
756 in_dev = __in_dev_get_rcu(ndev);
758 if (!in_dev || !in_dev->ifa_list) {
760 printk(KERN_ERR "%s: no IP address for %s, aborting\n",
761 np->name, np->dev_name);
766 np->local_ip = in_dev->ifa_list->ifa_local;
768 printk(KERN_INFO "%s: local IP %pI4\n", np->name, &np->local_ip);
772 spin_lock_irqsave(&npinfo->rx_lock, flags);
773 npinfo->rx_flags |= NETPOLL_RX_ENABLED;
775 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
778 /* fill up the skb queue */
781 /* last thing to do is link it to the net device structure */
782 ndev->npinfo = npinfo;
784 /* avoid racing with NAPI reading npinfo */
797 static int __init netpoll_init(void)
799 skb_queue_head_init(&skb_pool);
802 core_initcall(netpoll_init);
804 void netpoll_cleanup(struct netpoll *np)
806 struct netpoll_info *npinfo;
810 npinfo = np->dev->npinfo;
812 if (npinfo->rx_np == np) {
813 spin_lock_irqsave(&npinfo->rx_lock, flags);
814 npinfo->rx_np = NULL;
815 npinfo->rx_flags &= ~NETPOLL_RX_ENABLED;
816 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
819 if (atomic_dec_and_test(&npinfo->refcnt)) {
820 skb_queue_purge(&npinfo->arp_tx);
821 skb_queue_purge(&npinfo->txq);
822 cancel_rearming_delayed_work(&npinfo->tx_work);
824 /* clean after last, unfinished work */
825 __skb_queue_purge(&npinfo->txq);
827 np->dev->npinfo = NULL;
837 int netpoll_trap(void)
839 return atomic_read(&trapped);
842 void netpoll_set_trap(int trap)
845 atomic_inc(&trapped);
847 atomic_dec(&trapped);
850 EXPORT_SYMBOL(netpoll_set_trap);
851 EXPORT_SYMBOL(netpoll_trap);
852 EXPORT_SYMBOL(netpoll_print_options);
853 EXPORT_SYMBOL(netpoll_parse_options);
854 EXPORT_SYMBOL(netpoll_setup);
855 EXPORT_SYMBOL(netpoll_cleanup);
856 EXPORT_SYMBOL(netpoll_send_udp);
857 EXPORT_SYMBOL(netpoll_poll);