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>
27 #include <trace/events/napi.h>
30 * We maintain a small pool of fully-sized skbs, to make sure the
31 * message gets out even in extreme OOM situations.
34 #define MAX_UDP_CHUNK 1460
36 #define MAX_QUEUE_DEPTH (MAX_SKBS / 2)
38 static struct sk_buff_head skb_pool;
40 static atomic_t trapped;
42 #define USEC_PER_POLL 50
43 #define NETPOLL_RX_ENABLED 1
44 #define NETPOLL_RX_DROP 2
46 #define MAX_SKB_SIZE \
47 (MAX_UDP_CHUNK + sizeof(struct udphdr) + \
48 sizeof(struct iphdr) + sizeof(struct ethhdr))
50 static void zap_completion_queue(void);
51 static void arp_reply(struct sk_buff *skb);
53 static void queue_process(struct work_struct *work)
55 struct netpoll_info *npinfo =
56 container_of(work, struct netpoll_info, tx_work.work);
60 while ((skb = skb_dequeue(&npinfo->txq))) {
61 struct net_device *dev = skb->dev;
62 const struct net_device_ops *ops = dev->netdev_ops;
63 struct netdev_queue *txq;
65 if (!netif_device_present(dev) || !netif_running(dev)) {
70 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
72 local_irq_save(flags);
73 __netif_tx_lock(txq, smp_processor_id());
74 if (netif_tx_queue_stopped(txq) ||
75 netif_tx_queue_frozen(txq) ||
76 ops->ndo_start_xmit(skb, dev) != NETDEV_TX_OK) {
77 skb_queue_head(&npinfo->txq, skb);
78 __netif_tx_unlock(txq);
79 local_irq_restore(flags);
81 schedule_delayed_work(&npinfo->tx_work, HZ/10);
84 __netif_tx_unlock(txq);
85 local_irq_restore(flags);
89 static __sum16 checksum_udp(struct sk_buff *skb, struct udphdr *uh,
90 unsigned short ulen, __be32 saddr, __be32 daddr)
94 if (uh->check == 0 || skb_csum_unnecessary(skb))
97 psum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0);
99 if (skb->ip_summed == CHECKSUM_COMPLETE &&
100 !csum_fold(csum_add(psum, skb->csum)))
105 return __skb_checksum_complete(skb);
109 * Check whether delayed processing was scheduled for our NIC. If so,
110 * we attempt to grab the poll lock and use ->poll() to pump the card.
111 * If this fails, either we've recursed in ->poll() or it's already
112 * running on another CPU.
114 * Note: we don't mask interrupts with this lock because we're using
115 * trylock here and interrupts are already disabled in the softirq
116 * case. Further, we test the poll_owner to avoid recursion on UP
117 * systems where the lock doesn't exist.
119 * In cases where there is bi-directional communications, reading only
120 * one message at a time can lead to packets being dropped by the
121 * network adapter, forcing superfluous retries and possibly timeouts.
122 * Thus, we set our budget to greater than 1.
124 static int poll_one_napi(struct netpoll_info *npinfo,
125 struct napi_struct *napi, int budget)
129 /* net_rx_action's ->poll() invocations and our's are
130 * synchronized by this test which is only made while
131 * holding the napi->poll_lock.
133 if (!test_bit(NAPI_STATE_SCHED, &napi->state))
136 npinfo->rx_flags |= NETPOLL_RX_DROP;
137 atomic_inc(&trapped);
138 set_bit(NAPI_STATE_NPSVC, &napi->state);
140 work = napi->poll(napi, budget);
141 trace_napi_poll(napi);
143 clear_bit(NAPI_STATE_NPSVC, &napi->state);
144 atomic_dec(&trapped);
145 npinfo->rx_flags &= ~NETPOLL_RX_DROP;
147 return budget - work;
150 static void poll_napi(struct net_device *dev)
152 struct napi_struct *napi;
155 list_for_each_entry(napi, &dev->napi_list, dev_list) {
156 if (napi->poll_owner != smp_processor_id() &&
157 spin_trylock(&napi->poll_lock)) {
158 budget = poll_one_napi(dev->npinfo, napi, budget);
159 spin_unlock(&napi->poll_lock);
167 static void service_arp_queue(struct netpoll_info *npi)
172 while ((skb = skb_dequeue(&npi->arp_tx)))
177 void netpoll_poll(struct netpoll *np)
179 struct net_device *dev = np->dev;
180 const struct net_device_ops *ops;
182 if (!dev || !netif_running(dev))
185 ops = dev->netdev_ops;
186 if (!ops->ndo_poll_controller)
189 /* Process pending work on NIC */
190 ops->ndo_poll_controller(dev);
194 service_arp_queue(dev->npinfo);
196 zap_completion_queue();
199 static void refill_skbs(void)
204 spin_lock_irqsave(&skb_pool.lock, flags);
205 while (skb_pool.qlen < MAX_SKBS) {
206 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
210 __skb_queue_tail(&skb_pool, skb);
212 spin_unlock_irqrestore(&skb_pool.lock, flags);
215 static void zap_completion_queue(void)
218 struct softnet_data *sd = &get_cpu_var(softnet_data);
220 if (sd->completion_queue) {
221 struct sk_buff *clist;
223 local_irq_save(flags);
224 clist = sd->completion_queue;
225 sd->completion_queue = NULL;
226 local_irq_restore(flags);
228 while (clist != NULL) {
229 struct sk_buff *skb = clist;
231 if (skb->destructor) {
232 atomic_inc(&skb->users);
233 dev_kfree_skb_any(skb); /* put this one back */
240 put_cpu_var(softnet_data);
243 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
248 zap_completion_queue();
252 skb = alloc_skb(len, GFP_ATOMIC);
254 skb = skb_dequeue(&skb_pool);
264 atomic_set(&skb->users, 1);
265 skb_reserve(skb, reserve);
269 static int netpoll_owner_active(struct net_device *dev)
271 struct napi_struct *napi;
273 list_for_each_entry(napi, &dev->napi_list, dev_list) {
274 if (napi->poll_owner == smp_processor_id())
280 static void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
282 int status = NETDEV_TX_BUSY;
284 struct net_device *dev = np->dev;
285 const struct net_device_ops *ops = dev->netdev_ops;
286 struct netpoll_info *npinfo = np->dev->npinfo;
288 if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
293 /* don't get messages out of order, and no recursion */
294 if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
295 struct netdev_queue *txq;
298 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
300 local_irq_save(flags);
301 /* try until next clock tick */
302 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
303 tries > 0; --tries) {
304 if (__netif_tx_trylock(txq)) {
305 if (!netif_tx_queue_stopped(txq)) {
306 status = ops->ndo_start_xmit(skb, dev);
307 if (status == NETDEV_TX_OK)
308 txq_trans_update(txq);
310 __netif_tx_unlock(txq);
312 if (status == NETDEV_TX_OK)
317 /* tickle device maybe there is some cleanup */
320 udelay(USEC_PER_POLL);
323 WARN_ONCE(!irqs_disabled(),
324 "netpoll_send_skb(): %s enabled interrupts in poll (%pF)\n",
325 dev->name, ops->ndo_start_xmit);
327 local_irq_restore(flags);
330 if (status != NETDEV_TX_OK) {
331 skb_queue_tail(&npinfo->txq, skb);
332 schedule_delayed_work(&npinfo->tx_work,0);
336 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
338 int total_len, eth_len, ip_len, udp_len;
344 udp_len = len + sizeof(*udph);
345 ip_len = eth_len = udp_len + sizeof(*iph);
346 total_len = eth_len + ETH_HLEN + NET_IP_ALIGN;
348 skb = find_skb(np, total_len, total_len - len);
352 skb_copy_to_linear_data(skb, msg, len);
355 skb_push(skb, sizeof(*udph));
356 skb_reset_transport_header(skb);
358 udph->source = htons(np->local_port);
359 udph->dest = htons(np->remote_port);
360 udph->len = htons(udp_len);
362 udph->check = csum_tcpudp_magic(np->local_ip,
364 udp_len, IPPROTO_UDP,
365 csum_partial(udph, udp_len, 0));
366 if (udph->check == 0)
367 udph->check = CSUM_MANGLED_0;
369 skb_push(skb, sizeof(*iph));
370 skb_reset_network_header(skb);
373 /* iph->version = 4; iph->ihl = 5; */
374 put_unaligned(0x45, (unsigned char *)iph);
376 put_unaligned(htons(ip_len), &(iph->tot_len));
380 iph->protocol = IPPROTO_UDP;
382 put_unaligned(np->local_ip, &(iph->saddr));
383 put_unaligned(np->remote_ip, &(iph->daddr));
384 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
386 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
387 skb_reset_mac_header(skb);
388 skb->protocol = eth->h_proto = htons(ETH_P_IP);
389 memcpy(eth->h_source, np->dev->dev_addr, ETH_ALEN);
390 memcpy(eth->h_dest, np->remote_mac, ETH_ALEN);
394 netpoll_send_skb(np, skb);
397 static void arp_reply(struct sk_buff *skb)
399 struct netpoll_info *npinfo = skb->dev->npinfo;
401 unsigned char *arp_ptr;
402 int size, type = ARPOP_REPLY, ptype = ETH_P_ARP;
405 struct sk_buff *send_skb;
406 struct netpoll *np = NULL;
408 if (npinfo->rx_np && npinfo->rx_np->dev == skb->dev)
413 /* No arp on this interface */
414 if (skb->dev->flags & IFF_NOARP)
417 if (!pskb_may_pull(skb, arp_hdr_len(skb->dev)))
420 skb_reset_network_header(skb);
421 skb_reset_transport_header(skb);
424 if ((arp->ar_hrd != htons(ARPHRD_ETHER) &&
425 arp->ar_hrd != htons(ARPHRD_IEEE802)) ||
426 arp->ar_pro != htons(ETH_P_IP) ||
427 arp->ar_op != htons(ARPOP_REQUEST))
430 arp_ptr = (unsigned char *)(arp+1);
431 /* save the location of the src hw addr */
433 arp_ptr += skb->dev->addr_len;
434 memcpy(&sip, arp_ptr, 4);
436 /* if we actually cared about dst hw addr, it would get copied here */
437 arp_ptr += skb->dev->addr_len;
438 memcpy(&tip, arp_ptr, 4);
440 /* Should we ignore arp? */
441 if (tip != np->local_ip ||
442 ipv4_is_loopback(tip) || ipv4_is_multicast(tip))
445 size = arp_hdr_len(skb->dev);
446 send_skb = find_skb(np, size + LL_ALLOCATED_SPACE(np->dev),
447 LL_RESERVED_SPACE(np->dev));
452 skb_reset_network_header(send_skb);
453 arp = (struct arphdr *) skb_put(send_skb, size);
454 send_skb->dev = skb->dev;
455 send_skb->protocol = htons(ETH_P_ARP);
457 /* Fill the device header for the ARP frame */
458 if (dev_hard_header(send_skb, skb->dev, ptype,
459 sha, np->dev->dev_addr,
460 send_skb->len) < 0) {
466 * Fill out the arp protocol part.
468 * we only support ethernet device type,
469 * which (according to RFC 1390) should always equal 1 (Ethernet).
472 arp->ar_hrd = htons(np->dev->type);
473 arp->ar_pro = htons(ETH_P_IP);
474 arp->ar_hln = np->dev->addr_len;
476 arp->ar_op = htons(type);
478 arp_ptr=(unsigned char *)(arp + 1);
479 memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
480 arp_ptr += np->dev->addr_len;
481 memcpy(arp_ptr, &tip, 4);
483 memcpy(arp_ptr, sha, np->dev->addr_len);
484 arp_ptr += np->dev->addr_len;
485 memcpy(arp_ptr, &sip, 4);
487 netpoll_send_skb(np, send_skb);
490 int __netpoll_rx(struct sk_buff *skb)
492 int proto, len, ulen;
495 struct netpoll_info *npi = skb->dev->npinfo;
496 struct netpoll *np = npi->rx_np;
500 if (skb->dev->type != ARPHRD_ETHER)
503 /* check if netpoll clients need ARP */
504 if (skb->protocol == htons(ETH_P_ARP) &&
505 atomic_read(&trapped)) {
506 skb_queue_tail(&npi->arp_tx, skb);
510 proto = ntohs(eth_hdr(skb)->h_proto);
511 if (proto != ETH_P_IP)
513 if (skb->pkt_type == PACKET_OTHERHOST)
518 iph = (struct iphdr *)skb->data;
519 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
521 if (iph->ihl < 5 || iph->version != 4)
523 if (!pskb_may_pull(skb, iph->ihl*4))
525 if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
528 len = ntohs(iph->tot_len);
529 if (skb->len < len || len < iph->ihl*4)
533 * Our transport medium may have padded the buffer out.
534 * Now We trim to the true length of the frame.
536 if (pskb_trim_rcsum(skb, len))
539 if (iph->protocol != IPPROTO_UDP)
543 uh = (struct udphdr *)(((char *)iph) + iph->ihl*4);
544 ulen = ntohs(uh->len);
548 if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr))
550 if (np->local_ip && np->local_ip != iph->daddr)
552 if (np->remote_ip && np->remote_ip != iph->saddr)
554 if (np->local_port && np->local_port != ntohs(uh->dest))
557 np->rx_hook(np, ntohs(uh->source),
559 ulen - sizeof(struct udphdr));
565 if (atomic_read(&trapped)) {
573 void netpoll_print_options(struct netpoll *np)
575 printk(KERN_INFO "%s: local port %d\n",
576 np->name, np->local_port);
577 printk(KERN_INFO "%s: local IP %pI4\n",
578 np->name, &np->local_ip);
579 printk(KERN_INFO "%s: interface %s\n",
580 np->name, np->dev_name);
581 printk(KERN_INFO "%s: remote port %d\n",
582 np->name, np->remote_port);
583 printk(KERN_INFO "%s: remote IP %pI4\n",
584 np->name, &np->remote_ip);
585 printk(KERN_INFO "%s: remote ethernet address %pM\n",
586 np->name, np->remote_mac);
589 int netpoll_parse_options(struct netpoll *np, char *opt)
591 char *cur=opt, *delim;
594 if ((delim = strchr(cur, '@')) == NULL)
597 np->local_port = simple_strtol(cur, NULL, 10);
603 if ((delim = strchr(cur, '/')) == NULL)
606 np->local_ip = in_aton(cur);
612 /* parse out dev name */
613 if ((delim = strchr(cur, ',')) == NULL)
616 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
623 if ((delim = strchr(cur, '@')) == NULL)
626 np->remote_port = simple_strtol(cur, NULL, 10);
632 if ((delim = strchr(cur, '/')) == NULL)
635 np->remote_ip = in_aton(cur);
640 if ((delim = strchr(cur, ':')) == NULL)
643 np->remote_mac[0] = simple_strtol(cur, NULL, 16);
645 if ((delim = strchr(cur, ':')) == NULL)
648 np->remote_mac[1] = simple_strtol(cur, NULL, 16);
650 if ((delim = strchr(cur, ':')) == NULL)
653 np->remote_mac[2] = simple_strtol(cur, NULL, 16);
655 if ((delim = strchr(cur, ':')) == NULL)
658 np->remote_mac[3] = simple_strtol(cur, NULL, 16);
660 if ((delim = strchr(cur, ':')) == NULL)
663 np->remote_mac[4] = simple_strtol(cur, NULL, 16);
665 np->remote_mac[5] = simple_strtol(cur, NULL, 16);
668 netpoll_print_options(np);
673 printk(KERN_INFO "%s: couldn't parse config at %s!\n",
678 int netpoll_setup(struct netpoll *np)
680 struct net_device *ndev = NULL;
681 struct in_device *in_dev;
682 struct netpoll_info *npinfo;
687 ndev = dev_get_by_name(&init_net, np->dev_name);
689 printk(KERN_ERR "%s: %s doesn't exist, aborting.\n",
690 np->name, np->dev_name);
696 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
702 npinfo->rx_flags = 0;
703 npinfo->rx_np = NULL;
705 spin_lock_init(&npinfo->rx_lock);
706 skb_queue_head_init(&npinfo->arp_tx);
707 skb_queue_head_init(&npinfo->txq);
708 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
710 atomic_set(&npinfo->refcnt, 1);
712 npinfo = ndev->npinfo;
713 atomic_inc(&npinfo->refcnt);
716 if (!ndev->netdev_ops->ndo_poll_controller) {
717 printk(KERN_ERR "%s: %s doesn't support polling, aborting.\n",
718 np->name, np->dev_name);
723 if (!netif_running(ndev)) {
724 unsigned long atmost, atleast;
726 printk(KERN_INFO "%s: device %s not up yet, forcing it\n",
727 np->name, np->dev_name);
730 err = dev_open(ndev);
734 printk(KERN_ERR "%s: failed to open %s\n",
735 np->name, ndev->name);
739 atleast = jiffies + HZ/10;
740 atmost = jiffies + 4*HZ;
741 while (!netif_carrier_ok(ndev)) {
742 if (time_after(jiffies, atmost)) {
744 "%s: timeout waiting for carrier\n",
751 /* If carrier appears to come up instantly, we don't
752 * trust it and pause so that we don't pump all our
753 * queued console messages into the bitbucket.
756 if (time_before(jiffies, atleast)) {
757 printk(KERN_NOTICE "%s: carrier detect appears"
758 " untrustworthy, waiting 4 seconds\n",
766 in_dev = __in_dev_get_rcu(ndev);
768 if (!in_dev || !in_dev->ifa_list) {
770 printk(KERN_ERR "%s: no IP address for %s, aborting\n",
771 np->name, np->dev_name);
776 np->local_ip = in_dev->ifa_list->ifa_local;
778 printk(KERN_INFO "%s: local IP %pI4\n", np->name, &np->local_ip);
782 spin_lock_irqsave(&npinfo->rx_lock, flags);
783 npinfo->rx_flags |= NETPOLL_RX_ENABLED;
785 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
788 /* fill up the skb queue */
791 /* last thing to do is link it to the net device structure */
792 ndev->npinfo = npinfo;
794 /* avoid racing with NAPI reading npinfo */
807 static int __init netpoll_init(void)
809 skb_queue_head_init(&skb_pool);
812 core_initcall(netpoll_init);
814 void netpoll_cleanup(struct netpoll *np)
816 struct netpoll_info *npinfo;
820 npinfo = np->dev->npinfo;
822 if (npinfo->rx_np == np) {
823 spin_lock_irqsave(&npinfo->rx_lock, flags);
824 npinfo->rx_np = NULL;
825 npinfo->rx_flags &= ~NETPOLL_RX_ENABLED;
826 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
829 if (atomic_dec_and_test(&npinfo->refcnt)) {
830 skb_queue_purge(&npinfo->arp_tx);
831 skb_queue_purge(&npinfo->txq);
832 cancel_rearming_delayed_work(&npinfo->tx_work);
834 /* clean after last, unfinished work */
835 __skb_queue_purge(&npinfo->txq);
837 np->dev->npinfo = NULL;
847 int netpoll_trap(void)
849 return atomic_read(&trapped);
852 void netpoll_set_trap(int trap)
855 atomic_inc(&trapped);
857 atomic_dec(&trapped);
860 EXPORT_SYMBOL(netpoll_set_trap);
861 EXPORT_SYMBOL(netpoll_trap);
862 EXPORT_SYMBOL(netpoll_print_options);
863 EXPORT_SYMBOL(netpoll_parse_options);
864 EXPORT_SYMBOL(netpoll_setup);
865 EXPORT_SYMBOL(netpoll_cleanup);
866 EXPORT_SYMBOL(netpoll_send_udp);
867 EXPORT_SYMBOL(netpoll_poll);