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/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 <linux/slab.h>
28 #include <asm/unaligned.h>
29 #include <trace/events/napi.h>
32 * We maintain a small pool of fully-sized skbs, to make sure the
33 * message gets out even in extreme OOM situations.
36 #define MAX_UDP_CHUNK 1460
38 #define MAX_QUEUE_DEPTH (MAX_SKBS / 2)
40 static struct sk_buff_head skb_pool;
42 static atomic_t trapped;
44 #define USEC_PER_POLL 50
45 #define NETPOLL_RX_ENABLED 1
46 #define NETPOLL_RX_DROP 2
48 #define MAX_SKB_SIZE \
49 (MAX_UDP_CHUNK + sizeof(struct udphdr) + \
50 sizeof(struct iphdr) + sizeof(struct ethhdr))
52 static void zap_completion_queue(void);
53 static void arp_reply(struct sk_buff *skb);
55 static unsigned int carrier_timeout = 4;
56 module_param(carrier_timeout, uint, 0644);
58 static void queue_process(struct work_struct *work)
60 struct netpoll_info *npinfo =
61 container_of(work, struct netpoll_info, tx_work.work);
65 while ((skb = skb_dequeue(&npinfo->txq))) {
66 struct net_device *dev = skb->dev;
67 const struct net_device_ops *ops = dev->netdev_ops;
68 struct netdev_queue *txq;
70 if (!netif_device_present(dev) || !netif_running(dev)) {
75 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
77 local_irq_save(flags);
78 __netif_tx_lock(txq, smp_processor_id());
79 if (netif_tx_queue_stopped(txq) ||
80 netif_tx_queue_frozen(txq) ||
81 ops->ndo_start_xmit(skb, dev) != NETDEV_TX_OK) {
82 skb_queue_head(&npinfo->txq, skb);
83 __netif_tx_unlock(txq);
84 local_irq_restore(flags);
86 schedule_delayed_work(&npinfo->tx_work, HZ/10);
89 __netif_tx_unlock(txq);
90 local_irq_restore(flags);
94 static __sum16 checksum_udp(struct sk_buff *skb, struct udphdr *uh,
95 unsigned short ulen, __be32 saddr, __be32 daddr)
99 if (uh->check == 0 || skb_csum_unnecessary(skb))
102 psum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0);
104 if (skb->ip_summed == CHECKSUM_COMPLETE &&
105 !csum_fold(csum_add(psum, skb->csum)))
110 return __skb_checksum_complete(skb);
114 * Check whether delayed processing was scheduled for our NIC. If so,
115 * we attempt to grab the poll lock and use ->poll() to pump the card.
116 * If this fails, either we've recursed in ->poll() or it's already
117 * running on another CPU.
119 * Note: we don't mask interrupts with this lock because we're using
120 * trylock here and interrupts are already disabled in the softirq
121 * case. Further, we test the poll_owner to avoid recursion on UP
122 * systems where the lock doesn't exist.
124 * In cases where there is bi-directional communications, reading only
125 * one message at a time can lead to packets being dropped by the
126 * network adapter, forcing superfluous retries and possibly timeouts.
127 * Thus, we set our budget to greater than 1.
129 static int poll_one_napi(struct netpoll_info *npinfo,
130 struct napi_struct *napi, int budget)
134 /* net_rx_action's ->poll() invocations and our's are
135 * synchronized by this test which is only made while
136 * holding the napi->poll_lock.
138 if (!test_bit(NAPI_STATE_SCHED, &napi->state))
141 npinfo->rx_flags |= NETPOLL_RX_DROP;
142 atomic_inc(&trapped);
143 set_bit(NAPI_STATE_NPSVC, &napi->state);
145 work = napi->poll(napi, budget);
146 trace_napi_poll(napi);
148 clear_bit(NAPI_STATE_NPSVC, &napi->state);
149 atomic_dec(&trapped);
150 npinfo->rx_flags &= ~NETPOLL_RX_DROP;
152 return budget - work;
155 static void poll_napi(struct net_device *dev)
157 struct napi_struct *napi;
160 list_for_each_entry(napi, &dev->napi_list, dev_list) {
161 if (napi->poll_owner != smp_processor_id() &&
162 spin_trylock(&napi->poll_lock)) {
163 budget = poll_one_napi(dev->npinfo, napi, budget);
164 spin_unlock(&napi->poll_lock);
172 static void service_arp_queue(struct netpoll_info *npi)
177 while ((skb = skb_dequeue(&npi->arp_tx)))
182 void netpoll_poll_dev(struct net_device *dev)
184 const struct net_device_ops *ops;
186 if (!dev || !netif_running(dev))
189 ops = dev->netdev_ops;
190 if (!ops->ndo_poll_controller)
193 /* Process pending work on NIC */
194 ops->ndo_poll_controller(dev);
198 service_arp_queue(dev->npinfo);
200 zap_completion_queue();
203 void netpoll_poll(struct netpoll *np)
205 netpoll_poll_dev(np->dev);
208 static void refill_skbs(void)
213 spin_lock_irqsave(&skb_pool.lock, flags);
214 while (skb_pool.qlen < MAX_SKBS) {
215 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
219 __skb_queue_tail(&skb_pool, skb);
221 spin_unlock_irqrestore(&skb_pool.lock, flags);
224 static void zap_completion_queue(void)
227 struct softnet_data *sd = &get_cpu_var(softnet_data);
229 if (sd->completion_queue) {
230 struct sk_buff *clist;
232 local_irq_save(flags);
233 clist = sd->completion_queue;
234 sd->completion_queue = NULL;
235 local_irq_restore(flags);
237 while (clist != NULL) {
238 struct sk_buff *skb = clist;
240 if (skb->destructor) {
241 atomic_inc(&skb->users);
242 dev_kfree_skb_any(skb); /* put this one back */
249 put_cpu_var(softnet_data);
252 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
257 zap_completion_queue();
261 skb = alloc_skb(len, GFP_ATOMIC);
263 skb = skb_dequeue(&skb_pool);
273 atomic_set(&skb->users, 1);
274 skb_reserve(skb, reserve);
278 static int netpoll_owner_active(struct net_device *dev)
280 struct napi_struct *napi;
282 list_for_each_entry(napi, &dev->napi_list, dev_list) {
283 if (napi->poll_owner == smp_processor_id())
289 void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
291 int status = NETDEV_TX_BUSY;
293 struct net_device *dev = np->dev;
294 const struct net_device_ops *ops = dev->netdev_ops;
295 struct netpoll_info *npinfo = np->dev->npinfo;
297 if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
302 /* don't get messages out of order, and no recursion */
303 if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
304 struct netdev_queue *txq;
307 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
309 local_irq_save(flags);
310 /* try until next clock tick */
311 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
312 tries > 0; --tries) {
313 if (__netif_tx_trylock(txq)) {
314 if (!netif_tx_queue_stopped(txq)) {
315 dev->priv_flags |= IFF_IN_NETPOLL;
316 status = ops->ndo_start_xmit(skb, dev);
317 dev->priv_flags &= ~IFF_IN_NETPOLL;
318 if (status == NETDEV_TX_OK)
319 txq_trans_update(txq);
321 __netif_tx_unlock(txq);
323 if (status == NETDEV_TX_OK)
328 /* tickle device maybe there is some cleanup */
331 udelay(USEC_PER_POLL);
334 WARN_ONCE(!irqs_disabled(),
335 "netpoll_send_skb(): %s enabled interrupts in poll (%pF)\n",
336 dev->name, ops->ndo_start_xmit);
338 local_irq_restore(flags);
341 if (status != NETDEV_TX_OK) {
342 skb_queue_tail(&npinfo->txq, skb);
343 schedule_delayed_work(&npinfo->tx_work,0);
347 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
349 int total_len, eth_len, ip_len, udp_len;
355 udp_len = len + sizeof(*udph);
356 ip_len = eth_len = udp_len + sizeof(*iph);
357 total_len = eth_len + ETH_HLEN + NET_IP_ALIGN;
359 skb = find_skb(np, total_len, total_len - len);
363 skb_copy_to_linear_data(skb, msg, len);
366 skb_push(skb, sizeof(*udph));
367 skb_reset_transport_header(skb);
369 udph->source = htons(np->local_port);
370 udph->dest = htons(np->remote_port);
371 udph->len = htons(udp_len);
373 udph->check = csum_tcpudp_magic(np->local_ip,
375 udp_len, IPPROTO_UDP,
376 csum_partial(udph, udp_len, 0));
377 if (udph->check == 0)
378 udph->check = CSUM_MANGLED_0;
380 skb_push(skb, sizeof(*iph));
381 skb_reset_network_header(skb);
384 /* iph->version = 4; iph->ihl = 5; */
385 put_unaligned(0x45, (unsigned char *)iph);
387 put_unaligned(htons(ip_len), &(iph->tot_len));
391 iph->protocol = IPPROTO_UDP;
393 put_unaligned(np->local_ip, &(iph->saddr));
394 put_unaligned(np->remote_ip, &(iph->daddr));
395 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
397 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
398 skb_reset_mac_header(skb);
399 skb->protocol = eth->h_proto = htons(ETH_P_IP);
400 memcpy(eth->h_source, np->dev->dev_addr, ETH_ALEN);
401 memcpy(eth->h_dest, np->remote_mac, ETH_ALEN);
405 netpoll_send_skb(np, skb);
408 static void arp_reply(struct sk_buff *skb)
410 struct netpoll_info *npinfo = skb->dev->npinfo;
412 unsigned char *arp_ptr;
413 int size, type = ARPOP_REPLY, ptype = ETH_P_ARP;
416 struct sk_buff *send_skb;
417 struct netpoll *np, *tmp;
421 if (list_empty(&npinfo->rx_np))
424 /* Before checking the packet, we do some early
425 inspection whether this is interesting at all */
426 spin_lock_irqsave(&npinfo->rx_lock, flags);
427 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
428 if (np->dev == skb->dev)
431 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
433 /* No netpoll struct is using this dev */
437 /* No arp on this interface */
438 if (skb->dev->flags & IFF_NOARP)
441 if (!pskb_may_pull(skb, arp_hdr_len(skb->dev)))
444 skb_reset_network_header(skb);
445 skb_reset_transport_header(skb);
448 if ((arp->ar_hrd != htons(ARPHRD_ETHER) &&
449 arp->ar_hrd != htons(ARPHRD_IEEE802)) ||
450 arp->ar_pro != htons(ETH_P_IP) ||
451 arp->ar_op != htons(ARPOP_REQUEST))
454 arp_ptr = (unsigned char *)(arp+1);
455 /* save the location of the src hw addr */
457 arp_ptr += skb->dev->addr_len;
458 memcpy(&sip, arp_ptr, 4);
460 /* If we actually cared about dst hw addr,
461 it would get copied here */
462 arp_ptr += skb->dev->addr_len;
463 memcpy(&tip, arp_ptr, 4);
465 /* Should we ignore arp? */
466 if (ipv4_is_loopback(tip) || ipv4_is_multicast(tip))
469 size = arp_hdr_len(skb->dev);
471 spin_lock_irqsave(&npinfo->rx_lock, flags);
472 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
473 if (tip != np->local_ip)
476 send_skb = find_skb(np, size + LL_ALLOCATED_SPACE(np->dev),
477 LL_RESERVED_SPACE(np->dev));
481 skb_reset_network_header(send_skb);
482 arp = (struct arphdr *) skb_put(send_skb, size);
483 send_skb->dev = skb->dev;
484 send_skb->protocol = htons(ETH_P_ARP);
486 /* Fill the device header for the ARP frame */
487 if (dev_hard_header(send_skb, skb->dev, ptype,
488 sha, np->dev->dev_addr,
489 send_skb->len) < 0) {
495 * Fill out the arp protocol part.
497 * we only support ethernet device type,
498 * which (according to RFC 1390) should
499 * always equal 1 (Ethernet).
502 arp->ar_hrd = htons(np->dev->type);
503 arp->ar_pro = htons(ETH_P_IP);
504 arp->ar_hln = np->dev->addr_len;
506 arp->ar_op = htons(type);
508 arp_ptr = (unsigned char *)(arp + 1);
509 memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
510 arp_ptr += np->dev->addr_len;
511 memcpy(arp_ptr, &tip, 4);
513 memcpy(arp_ptr, sha, np->dev->addr_len);
514 arp_ptr += np->dev->addr_len;
515 memcpy(arp_ptr, &sip, 4);
517 netpoll_send_skb(np, send_skb);
519 /* If there are several rx_hooks for the same address,
520 we're fine by sending a single reply */
523 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
526 int __netpoll_rx(struct sk_buff *skb)
528 int proto, len, ulen;
532 struct netpoll_info *npinfo = skb->dev->npinfo;
533 struct netpoll *np, *tmp;
535 if (list_empty(&npinfo->rx_np))
538 if (skb->dev->type != ARPHRD_ETHER)
541 /* check if netpoll clients need ARP */
542 if (skb->protocol == htons(ETH_P_ARP) &&
543 atomic_read(&trapped)) {
544 skb_queue_tail(&npinfo->arp_tx, skb);
548 proto = ntohs(eth_hdr(skb)->h_proto);
549 if (proto != ETH_P_IP)
551 if (skb->pkt_type == PACKET_OTHERHOST)
556 iph = (struct iphdr *)skb->data;
557 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
559 if (iph->ihl < 5 || iph->version != 4)
561 if (!pskb_may_pull(skb, iph->ihl*4))
563 if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
566 len = ntohs(iph->tot_len);
567 if (skb->len < len || len < iph->ihl*4)
571 * Our transport medium may have padded the buffer out.
572 * Now We trim to the true length of the frame.
574 if (pskb_trim_rcsum(skb, len))
577 if (iph->protocol != IPPROTO_UDP)
581 uh = (struct udphdr *)(((char *)iph) + iph->ihl*4);
582 ulen = ntohs(uh->len);
586 if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr))
589 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
590 if (np->local_ip && np->local_ip != iph->daddr)
592 if (np->remote_ip && np->remote_ip != iph->saddr)
594 if (np->local_port && np->local_port != ntohs(uh->dest))
597 np->rx_hook(np, ntohs(uh->source),
599 ulen - sizeof(struct udphdr));
610 if (atomic_read(&trapped)) {
618 void netpoll_print_options(struct netpoll *np)
620 printk(KERN_INFO "%s: local port %d\n",
621 np->name, np->local_port);
622 printk(KERN_INFO "%s: local IP %pI4\n",
623 np->name, &np->local_ip);
624 printk(KERN_INFO "%s: interface '%s'\n",
625 np->name, np->dev_name);
626 printk(KERN_INFO "%s: remote port %d\n",
627 np->name, np->remote_port);
628 printk(KERN_INFO "%s: remote IP %pI4\n",
629 np->name, &np->remote_ip);
630 printk(KERN_INFO "%s: remote ethernet address %pM\n",
631 np->name, np->remote_mac);
634 int netpoll_parse_options(struct netpoll *np, char *opt)
636 char *cur=opt, *delim;
639 if ((delim = strchr(cur, '@')) == NULL)
642 np->local_port = simple_strtol(cur, NULL, 10);
648 if ((delim = strchr(cur, '/')) == NULL)
651 np->local_ip = in_aton(cur);
657 /* parse out dev name */
658 if ((delim = strchr(cur, ',')) == NULL)
661 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
668 if ((delim = strchr(cur, '@')) == NULL)
671 if (*cur == ' ' || *cur == '\t')
672 printk(KERN_INFO "%s: warning: whitespace"
673 "is not allowed\n", np->name);
674 np->remote_port = simple_strtol(cur, NULL, 10);
680 if ((delim = strchr(cur, '/')) == NULL)
683 np->remote_ip = in_aton(cur);
688 if ((delim = strchr(cur, ':')) == NULL)
691 np->remote_mac[0] = simple_strtol(cur, NULL, 16);
693 if ((delim = strchr(cur, ':')) == NULL)
696 np->remote_mac[1] = simple_strtol(cur, NULL, 16);
698 if ((delim = strchr(cur, ':')) == NULL)
701 np->remote_mac[2] = simple_strtol(cur, NULL, 16);
703 if ((delim = strchr(cur, ':')) == NULL)
706 np->remote_mac[3] = simple_strtol(cur, NULL, 16);
708 if ((delim = strchr(cur, ':')) == NULL)
711 np->remote_mac[4] = simple_strtol(cur, NULL, 16);
713 np->remote_mac[5] = simple_strtol(cur, NULL, 16);
716 netpoll_print_options(np);
721 printk(KERN_INFO "%s: couldn't parse config at '%s'!\n",
726 int netpoll_setup(struct netpoll *np)
728 struct net_device *ndev = NULL;
729 struct in_device *in_dev;
730 struct netpoll_info *npinfo;
731 struct netpoll *npe, *tmp;
736 ndev = dev_get_by_name(&init_net, np->dev_name);
738 printk(KERN_ERR "%s: %s doesn't exist, aborting.\n",
739 np->name, np->dev_name);
745 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
751 npinfo->rx_flags = 0;
752 INIT_LIST_HEAD(&npinfo->rx_np);
754 spin_lock_init(&npinfo->rx_lock);
755 skb_queue_head_init(&npinfo->arp_tx);
756 skb_queue_head_init(&npinfo->txq);
757 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
759 atomic_set(&npinfo->refcnt, 1);
761 npinfo = ndev->npinfo;
762 atomic_inc(&npinfo->refcnt);
765 npinfo->netpoll = np;
767 if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) ||
768 !ndev->netdev_ops->ndo_poll_controller) {
769 printk(KERN_ERR "%s: %s doesn't support polling, aborting.\n",
770 np->name, np->dev_name);
775 if (!netif_running(ndev)) {
776 unsigned long atmost, atleast;
778 printk(KERN_INFO "%s: device %s not up yet, forcing it\n",
779 np->name, np->dev_name);
782 err = dev_open(ndev);
786 printk(KERN_ERR "%s: failed to open %s\n",
787 np->name, ndev->name);
791 atleast = jiffies + HZ/10;
792 atmost = jiffies + carrier_timeout * HZ;
793 while (!netif_carrier_ok(ndev)) {
794 if (time_after(jiffies, atmost)) {
796 "%s: timeout waiting for carrier\n",
803 /* If carrier appears to come up instantly, we don't
804 * trust it and pause so that we don't pump all our
805 * queued console messages into the bitbucket.
808 if (time_before(jiffies, atleast)) {
809 printk(KERN_NOTICE "%s: carrier detect appears"
810 " untrustworthy, waiting 4 seconds\n",
818 in_dev = __in_dev_get_rcu(ndev);
820 if (!in_dev || !in_dev->ifa_list) {
822 printk(KERN_ERR "%s: no IP address for %s, aborting\n",
823 np->name, np->dev_name);
828 np->local_ip = in_dev->ifa_list->ifa_local;
830 printk(KERN_INFO "%s: local IP %pI4\n", np->name, &np->local_ip);
834 spin_lock_irqsave(&npinfo->rx_lock, flags);
835 npinfo->rx_flags |= NETPOLL_RX_ENABLED;
836 list_add_tail(&np->rx, &npinfo->rx_np);
837 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
840 /* fill up the skb queue */
843 /* last thing to do is link it to the net device structure */
844 ndev->npinfo = npinfo;
846 /* avoid racing with NAPI reading npinfo */
853 spin_lock_irqsave(&npinfo->rx_lock, flags);
854 list_for_each_entry_safe(npe, tmp, &npinfo->rx_np, rx) {
857 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
866 static int __init netpoll_init(void)
868 skb_queue_head_init(&skb_pool);
871 core_initcall(netpoll_init);
873 void netpoll_cleanup(struct netpoll *np)
875 struct netpoll_info *npinfo;
879 npinfo = np->dev->npinfo;
881 if (!list_empty(&npinfo->rx_np)) {
882 spin_lock_irqsave(&npinfo->rx_lock, flags);
884 if (list_empty(&npinfo->rx_np))
885 npinfo->rx_flags &= ~NETPOLL_RX_ENABLED;
886 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
889 if (atomic_dec_and_test(&npinfo->refcnt)) {
890 const struct net_device_ops *ops;
891 skb_queue_purge(&npinfo->arp_tx);
892 skb_queue_purge(&npinfo->txq);
893 cancel_rearming_delayed_work(&npinfo->tx_work);
895 /* clean after last, unfinished work */
896 __skb_queue_purge(&npinfo->txq);
898 ops = np->dev->netdev_ops;
899 if (ops->ndo_netpoll_cleanup)
900 ops->ndo_netpoll_cleanup(np->dev);
902 np->dev->npinfo = NULL;
912 int netpoll_trap(void)
914 return atomic_read(&trapped);
917 void netpoll_set_trap(int trap)
920 atomic_inc(&trapped);
922 atomic_dec(&trapped);
925 EXPORT_SYMBOL(netpoll_send_skb);
926 EXPORT_SYMBOL(netpoll_set_trap);
927 EXPORT_SYMBOL(netpoll_trap);
928 EXPORT_SYMBOL(netpoll_print_options);
929 EXPORT_SYMBOL(netpoll_parse_options);
930 EXPORT_SYMBOL(netpoll_setup);
931 EXPORT_SYMBOL(netpoll_cleanup);
932 EXPORT_SYMBOL(netpoll_send_udp);
933 EXPORT_SYMBOL(netpoll_poll_dev);
934 EXPORT_SYMBOL(netpoll_poll);