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)) ||
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 int poll_one_napi(struct netpoll_info *npinfo,
120 struct napi_struct *napi, int budget)
124 /* net_rx_action's ->poll() invocations and our's are
125 * synchronized by this test which is only made while
126 * holding the napi->poll_lock.
128 if (!test_bit(NAPI_STATE_SCHED, &napi->state))
131 npinfo->rx_flags |= NETPOLL_RX_DROP;
132 atomic_inc(&trapped);
134 work = napi->poll(napi, budget);
136 atomic_dec(&trapped);
137 npinfo->rx_flags &= ~NETPOLL_RX_DROP;
139 return budget - work;
142 static void poll_napi(struct net_device *dev)
144 struct napi_struct *napi;
147 list_for_each_entry(napi, &dev->napi_list, dev_list) {
148 if (napi->poll_owner != smp_processor_id() &&
149 spin_trylock(&napi->poll_lock)) {
150 budget = poll_one_napi(dev->npinfo, napi, budget);
151 spin_unlock(&napi->poll_lock);
159 static void service_arp_queue(struct netpoll_info *npi)
164 while ((skb = skb_dequeue(&npi->arp_tx)))
169 void netpoll_poll(struct netpoll *np)
171 struct net_device *dev = np->dev;
173 if (!dev || !netif_running(dev) || !dev->poll_controller)
176 /* Process pending work on NIC */
177 dev->poll_controller(dev);
181 service_arp_queue(dev->npinfo);
183 zap_completion_queue();
186 static void refill_skbs(void)
191 spin_lock_irqsave(&skb_pool.lock, flags);
192 while (skb_pool.qlen < MAX_SKBS) {
193 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
197 __skb_queue_tail(&skb_pool, skb);
199 spin_unlock_irqrestore(&skb_pool.lock, flags);
202 static void zap_completion_queue(void)
205 struct softnet_data *sd = &get_cpu_var(softnet_data);
207 if (sd->completion_queue) {
208 struct sk_buff *clist;
210 local_irq_save(flags);
211 clist = sd->completion_queue;
212 sd->completion_queue = NULL;
213 local_irq_restore(flags);
215 while (clist != NULL) {
216 struct sk_buff *skb = clist;
219 dev_kfree_skb_any(skb); /* put this one back */
225 put_cpu_var(softnet_data);
228 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
233 zap_completion_queue();
237 skb = alloc_skb(len, GFP_ATOMIC);
239 skb = skb_dequeue(&skb_pool);
249 atomic_set(&skb->users, 1);
250 skb_reserve(skb, reserve);
254 static int netpoll_owner_active(struct net_device *dev)
256 struct napi_struct *napi;
258 list_for_each_entry(napi, &dev->napi_list, dev_list) {
259 if (napi->poll_owner == smp_processor_id())
265 static void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
267 int status = NETDEV_TX_BUSY;
269 struct net_device *dev = np->dev;
270 struct netpoll_info *npinfo = np->dev->npinfo;
272 if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
277 /* don't get messages out of order, and no recursion */
278 if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
281 local_irq_save(flags);
282 /* try until next clock tick */
283 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
284 tries > 0; --tries) {
285 if (netif_tx_trylock(dev)) {
286 if (!netif_queue_stopped(dev) &&
287 !netif_subqueue_stopped(dev, skb))
288 status = dev->hard_start_xmit(skb, dev);
289 netif_tx_unlock(dev);
291 if (status == NETDEV_TX_OK)
296 /* tickle device maybe there is some cleanup */
299 udelay(USEC_PER_POLL);
301 local_irq_restore(flags);
304 if (status != NETDEV_TX_OK) {
305 skb_queue_tail(&npinfo->txq, skb);
306 schedule_delayed_work(&npinfo->tx_work,0);
310 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
312 int total_len, eth_len, ip_len, udp_len;
318 udp_len = len + sizeof(*udph);
319 ip_len = eth_len = udp_len + sizeof(*iph);
320 total_len = eth_len + ETH_HLEN + NET_IP_ALIGN;
322 skb = find_skb(np, total_len, total_len - len);
326 skb_copy_to_linear_data(skb, msg, len);
329 skb_push(skb, sizeof(*udph));
330 skb_reset_transport_header(skb);
332 udph->source = htons(np->local_port);
333 udph->dest = htons(np->remote_port);
334 udph->len = htons(udp_len);
336 udph->check = csum_tcpudp_magic(htonl(np->local_ip),
337 htonl(np->remote_ip),
338 udp_len, IPPROTO_UDP,
339 csum_partial((unsigned char *)udph, udp_len, 0));
340 if (udph->check == 0)
341 udph->check = CSUM_MANGLED_0;
343 skb_push(skb, sizeof(*iph));
344 skb_reset_network_header(skb);
347 /* iph->version = 4; iph->ihl = 5; */
348 put_unaligned(0x45, (unsigned char *)iph);
350 put_unaligned(htons(ip_len), &(iph->tot_len));
354 iph->protocol = IPPROTO_UDP;
356 put_unaligned(htonl(np->local_ip), &(iph->saddr));
357 put_unaligned(htonl(np->remote_ip), &(iph->daddr));
358 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
360 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
361 skb_reset_mac_header(skb);
362 skb->protocol = eth->h_proto = htons(ETH_P_IP);
363 memcpy(eth->h_source, np->local_mac, 6);
364 memcpy(eth->h_dest, np->remote_mac, 6);
368 netpoll_send_skb(np, skb);
371 static void arp_reply(struct sk_buff *skb)
373 struct netpoll_info *npinfo = skb->dev->npinfo;
375 unsigned char *arp_ptr;
376 int size, type = ARPOP_REPLY, ptype = ETH_P_ARP;
379 struct sk_buff *send_skb;
380 struct netpoll *np = NULL;
382 if (npinfo->rx_np && npinfo->rx_np->dev == skb->dev)
387 /* No arp on this interface */
388 if (skb->dev->flags & IFF_NOARP)
391 if (!pskb_may_pull(skb, (sizeof(struct arphdr) +
392 (2 * skb->dev->addr_len) +
396 skb_reset_network_header(skb);
397 skb_reset_transport_header(skb);
400 if ((arp->ar_hrd != htons(ARPHRD_ETHER) &&
401 arp->ar_hrd != htons(ARPHRD_IEEE802)) ||
402 arp->ar_pro != htons(ETH_P_IP) ||
403 arp->ar_op != htons(ARPOP_REQUEST))
406 arp_ptr = (unsigned char *)(arp+1);
407 /* save the location of the src hw addr */
409 arp_ptr += skb->dev->addr_len;
410 memcpy(&sip, arp_ptr, 4);
412 /* if we actually cared about dst hw addr, it would get copied here */
413 arp_ptr += skb->dev->addr_len;
414 memcpy(&tip, arp_ptr, 4);
416 /* Should we ignore arp? */
417 if (tip != htonl(np->local_ip) || LOOPBACK(tip) || MULTICAST(tip))
420 size = sizeof(struct arphdr) + 2 * (skb->dev->addr_len + 4);
421 send_skb = find_skb(np, size + LL_RESERVED_SPACE(np->dev),
422 LL_RESERVED_SPACE(np->dev));
427 skb_reset_network_header(send_skb);
428 arp = (struct arphdr *) skb_put(send_skb, size);
429 send_skb->dev = skb->dev;
430 send_skb->protocol = htons(ETH_P_ARP);
432 /* Fill the device header for the ARP frame */
433 if (dev_hard_header(send_skb, skb->dev, ptype,
435 send_skb->len) < 0) {
441 * Fill out the arp protocol part.
443 * we only support ethernet device type,
444 * which (according to RFC 1390) should always equal 1 (Ethernet).
447 arp->ar_hrd = htons(np->dev->type);
448 arp->ar_pro = htons(ETH_P_IP);
449 arp->ar_hln = np->dev->addr_len;
451 arp->ar_op = htons(type);
453 arp_ptr=(unsigned char *)(arp + 1);
454 memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
455 arp_ptr += np->dev->addr_len;
456 memcpy(arp_ptr, &tip, 4);
458 memcpy(arp_ptr, sha, np->dev->addr_len);
459 arp_ptr += np->dev->addr_len;
460 memcpy(arp_ptr, &sip, 4);
462 netpoll_send_skb(np, send_skb);
465 int __netpoll_rx(struct sk_buff *skb)
467 int proto, len, ulen;
470 struct netpoll_info *npi = skb->dev->npinfo;
471 struct netpoll *np = npi->rx_np;
475 if (skb->dev->type != ARPHRD_ETHER)
478 /* check if netpoll clients need ARP */
479 if (skb->protocol == htons(ETH_P_ARP) &&
480 atomic_read(&trapped)) {
481 skb_queue_tail(&npi->arp_tx, skb);
485 proto = ntohs(eth_hdr(skb)->h_proto);
486 if (proto != ETH_P_IP)
488 if (skb->pkt_type == PACKET_OTHERHOST)
493 iph = (struct iphdr *)skb->data;
494 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
496 if (iph->ihl < 5 || iph->version != 4)
498 if (!pskb_may_pull(skb, iph->ihl*4))
500 if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
503 len = ntohs(iph->tot_len);
504 if (skb->len < len || len < iph->ihl*4)
508 * Our transport medium may have padded the buffer out.
509 * Now We trim to the true length of the frame.
511 if (pskb_trim_rcsum(skb, len))
514 if (iph->protocol != IPPROTO_UDP)
518 uh = (struct udphdr *)(((char *)iph) + iph->ihl*4);
519 ulen = ntohs(uh->len);
523 if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr))
525 if (np->local_ip && np->local_ip != ntohl(iph->daddr))
527 if (np->remote_ip && np->remote_ip != ntohl(iph->saddr))
529 if (np->local_port && np->local_port != ntohs(uh->dest))
532 np->rx_hook(np, ntohs(uh->source),
534 ulen - sizeof(struct udphdr));
540 if (atomic_read(&trapped)) {
548 void netpoll_print_options(struct netpoll *np)
550 DECLARE_MAC_BUF(mac);
551 printk(KERN_INFO "%s: local port %d\n",
552 np->name, np->local_port);
553 printk(KERN_INFO "%s: local IP %d.%d.%d.%d\n",
554 np->name, HIPQUAD(np->local_ip));
555 printk(KERN_INFO "%s: interface %s\n",
556 np->name, np->dev_name);
557 printk(KERN_INFO "%s: remote port %d\n",
558 np->name, np->remote_port);
559 printk(KERN_INFO "%s: remote IP %d.%d.%d.%d\n",
560 np->name, HIPQUAD(np->remote_ip));
561 printk(KERN_INFO "%s: remote ethernet address %s\n",
562 np->name, print_mac(mac, np->remote_mac));
565 int netpoll_parse_options(struct netpoll *np, char *opt)
567 char *cur=opt, *delim;
570 if ((delim = strchr(cur, '@')) == NULL)
573 np->local_port = simple_strtol(cur, NULL, 10);
579 if ((delim = strchr(cur, '/')) == NULL)
582 np->local_ip = ntohl(in_aton(cur));
588 /* parse out dev name */
589 if ((delim = strchr(cur, ',')) == NULL)
592 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
599 if ((delim = strchr(cur, '@')) == NULL)
602 np->remote_port = simple_strtol(cur, NULL, 10);
608 if ((delim = strchr(cur, '/')) == NULL)
611 np->remote_ip = ntohl(in_aton(cur));
616 if ((delim = strchr(cur, ':')) == NULL)
619 np->remote_mac[0] = simple_strtol(cur, NULL, 16);
621 if ((delim = strchr(cur, ':')) == NULL)
624 np->remote_mac[1] = simple_strtol(cur, NULL, 16);
626 if ((delim = strchr(cur, ':')) == NULL)
629 np->remote_mac[2] = simple_strtol(cur, NULL, 16);
631 if ((delim = strchr(cur, ':')) == NULL)
634 np->remote_mac[3] = simple_strtol(cur, NULL, 16);
636 if ((delim = strchr(cur, ':')) == NULL)
639 np->remote_mac[4] = simple_strtol(cur, NULL, 16);
641 np->remote_mac[5] = simple_strtol(cur, NULL, 16);
644 netpoll_print_options(np);
649 printk(KERN_INFO "%s: couldn't parse config at %s!\n",
654 int netpoll_setup(struct netpoll *np)
656 struct net_device *ndev = NULL;
657 struct in_device *in_dev;
658 struct netpoll_info *npinfo;
663 ndev = dev_get_by_name(&init_net, np->dev_name);
665 printk(KERN_ERR "%s: %s doesn't exist, aborting.\n",
666 np->name, np->dev_name);
672 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
678 npinfo->rx_flags = 0;
679 npinfo->rx_np = NULL;
681 spin_lock_init(&npinfo->rx_lock);
682 skb_queue_head_init(&npinfo->arp_tx);
683 skb_queue_head_init(&npinfo->txq);
684 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
686 atomic_set(&npinfo->refcnt, 1);
688 npinfo = ndev->npinfo;
689 atomic_inc(&npinfo->refcnt);
692 if (!ndev->poll_controller) {
693 printk(KERN_ERR "%s: %s doesn't support polling, aborting.\n",
694 np->name, np->dev_name);
699 if (!netif_running(ndev)) {
700 unsigned long atmost, atleast;
702 printk(KERN_INFO "%s: device %s not up yet, forcing it\n",
703 np->name, np->dev_name);
706 err = dev_open(ndev);
710 printk(KERN_ERR "%s: failed to open %s\n",
711 np->name, ndev->name);
715 atleast = jiffies + HZ/10;
716 atmost = jiffies + 4*HZ;
717 while (!netif_carrier_ok(ndev)) {
718 if (time_after(jiffies, atmost)) {
720 "%s: timeout waiting for carrier\n",
727 /* If carrier appears to come up instantly, we don't
728 * trust it and pause so that we don't pump all our
729 * queued console messages into the bitbucket.
732 if (time_before(jiffies, atleast)) {
733 printk(KERN_NOTICE "%s: carrier detect appears"
734 " untrustworthy, waiting 4 seconds\n",
740 if (is_zero_ether_addr(np->local_mac) && ndev->dev_addr)
741 memcpy(np->local_mac, ndev->dev_addr, 6);
745 in_dev = __in_dev_get_rcu(ndev);
747 if (!in_dev || !in_dev->ifa_list) {
749 printk(KERN_ERR "%s: no IP address for %s, aborting\n",
750 np->name, np->dev_name);
755 np->local_ip = ntohl(in_dev->ifa_list->ifa_local);
757 printk(KERN_INFO "%s: local IP %d.%d.%d.%d\n",
758 np->name, HIPQUAD(np->local_ip));
762 spin_lock_irqsave(&npinfo->rx_lock, flags);
763 npinfo->rx_flags |= NETPOLL_RX_ENABLED;
765 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
768 /* fill up the skb queue */
771 /* last thing to do is link it to the net device structure */
772 ndev->npinfo = npinfo;
774 /* avoid racing with NAPI reading npinfo */
787 static int __init netpoll_init(void)
789 skb_queue_head_init(&skb_pool);
792 core_initcall(netpoll_init);
794 void netpoll_cleanup(struct netpoll *np)
796 struct netpoll_info *npinfo;
800 npinfo = np->dev->npinfo;
802 if (npinfo->rx_np == np) {
803 spin_lock_irqsave(&npinfo->rx_lock, flags);
804 npinfo->rx_np = NULL;
805 npinfo->rx_flags &= ~NETPOLL_RX_ENABLED;
806 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
809 if (atomic_dec_and_test(&npinfo->refcnt)) {
810 skb_queue_purge(&npinfo->arp_tx);
811 skb_queue_purge(&npinfo->txq);
812 cancel_rearming_delayed_work(&npinfo->tx_work);
814 /* clean after last, unfinished work */
815 __skb_queue_purge(&npinfo->txq);
817 np->dev->npinfo = NULL;
827 int netpoll_trap(void)
829 return atomic_read(&trapped);
832 void netpoll_set_trap(int trap)
835 atomic_inc(&trapped);
837 atomic_dec(&trapped);
840 EXPORT_SYMBOL(netpoll_set_trap);
841 EXPORT_SYMBOL(netpoll_trap);
842 EXPORT_SYMBOL(netpoll_print_options);
843 EXPORT_SYMBOL(netpoll_parse_options);
844 EXPORT_SYMBOL(netpoll_setup);
845 EXPORT_SYMBOL(netpoll_cleanup);
846 EXPORT_SYMBOL(netpoll_send_udp);
847 EXPORT_SYMBOL(netpoll_poll);