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/smp_lock.h>
13 #include <linux/netdevice.h>
14 #include <linux/etherdevice.h>
15 #include <linux/string.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)
36 #define MAX_RETRIES 20000
38 static DEFINE_SPINLOCK(skb_list_lock);
40 static struct sk_buff *skbs;
42 static DEFINE_SPINLOCK(queue_lock);
43 static int queue_depth;
44 static struct sk_buff *queue_head, *queue_tail;
46 static atomic_t trapped;
48 #define NETPOLL_RX_ENABLED 1
49 #define NETPOLL_RX_DROP 2
51 #define MAX_SKB_SIZE \
52 (MAX_UDP_CHUNK + sizeof(struct udphdr) + \
53 sizeof(struct iphdr) + sizeof(struct ethhdr))
55 static void zap_completion_queue(void);
57 static void queue_process(void *p)
63 spin_lock_irqsave(&queue_lock, flags);
66 queue_head = skb->next;
67 if (skb == queue_tail)
72 spin_unlock_irqrestore(&queue_lock, flags);
78 static DECLARE_WORK(send_queue, queue_process, NULL);
80 void netpoll_queue(struct sk_buff *skb)
84 if (queue_depth == MAX_QUEUE_DEPTH) {
89 spin_lock_irqsave(&queue_lock, flags);
93 queue_tail->next = skb;
96 spin_unlock_irqrestore(&queue_lock, flags);
98 schedule_work(&send_queue);
101 static int checksum_udp(struct sk_buff *skb, struct udphdr *uh,
102 unsigned short ulen, u32 saddr, u32 daddr)
106 if (uh->check == 0 || skb->ip_summed == CHECKSUM_UNNECESSARY)
109 psum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0);
111 if (skb->ip_summed == CHECKSUM_HW &&
112 !(u16)csum_fold(csum_add(psum, skb->csum)))
117 return __skb_checksum_complete(skb);
121 * Check whether delayed processing was scheduled for our NIC. If so,
122 * we attempt to grab the poll lock and use ->poll() to pump the card.
123 * If this fails, either we've recursed in ->poll() or it's already
124 * running on another CPU.
126 * Note: we don't mask interrupts with this lock because we're using
127 * trylock here and interrupts are already disabled in the softirq
128 * case. Further, we test the poll_owner to avoid recursion on UP
129 * systems where the lock doesn't exist.
131 * In cases where there is bi-directional communications, reading only
132 * one message at a time can lead to packets being dropped by the
133 * network adapter, forcing superfluous retries and possibly timeouts.
134 * Thus, we set our budget to greater than 1.
136 static void poll_napi(struct netpoll *np)
138 struct netpoll_info *npinfo = np->dev->npinfo;
141 if (test_bit(__LINK_STATE_RX_SCHED, &np->dev->state) &&
142 npinfo->poll_owner != smp_processor_id() &&
143 spin_trylock(&npinfo->poll_lock)) {
144 npinfo->rx_flags |= NETPOLL_RX_DROP;
145 atomic_inc(&trapped);
147 np->dev->poll(np->dev, &budget);
149 atomic_dec(&trapped);
150 npinfo->rx_flags &= ~NETPOLL_RX_DROP;
151 spin_unlock(&npinfo->poll_lock);
155 void netpoll_poll(struct netpoll *np)
157 if(!np->dev || !netif_running(np->dev) || !np->dev->poll_controller)
160 /* Process pending work on NIC */
161 np->dev->poll_controller(np->dev);
165 zap_completion_queue();
168 static void refill_skbs(void)
173 spin_lock_irqsave(&skb_list_lock, flags);
174 while (nr_skbs < MAX_SKBS) {
175 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
183 spin_unlock_irqrestore(&skb_list_lock, flags);
186 static void zap_completion_queue(void)
189 struct softnet_data *sd = &get_cpu_var(softnet_data);
191 if (sd->completion_queue) {
192 struct sk_buff *clist;
194 local_irq_save(flags);
195 clist = sd->completion_queue;
196 sd->completion_queue = NULL;
197 local_irq_restore(flags);
199 while (clist != NULL) {
200 struct sk_buff *skb = clist;
203 dev_kfree_skb_any(skb); /* put this one back */
209 put_cpu_var(softnet_data);
212 static struct sk_buff * find_skb(struct netpoll *np, int len, int reserve)
214 int once = 1, count = 0;
216 struct sk_buff *skb = NULL;
218 zap_completion_queue();
220 if (nr_skbs < MAX_SKBS)
223 skb = alloc_skb(len, GFP_ATOMIC);
226 spin_lock_irqsave(&skb_list_lock, flags);
233 spin_unlock_irqrestore(&skb_list_lock, flags);
238 if (once && (count == 1000000)) {
239 printk("out of netpoll skbs!\n");
246 atomic_set(&skb->users, 1);
247 skb_reserve(skb, reserve);
251 static void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
254 struct netpoll_info *npinfo;
256 if (!np || !np->dev || !netif_running(np->dev)) {
261 npinfo = np->dev->npinfo;
263 /* avoid recursion */
264 if (npinfo->poll_owner == smp_processor_id() ||
265 np->dev->xmit_lock_owner == smp_processor_id()) {
275 spin_lock(&np->dev->xmit_lock);
276 np->dev->xmit_lock_owner = smp_processor_id();
279 * network drivers do not expect to be called if the queue is
282 if (netif_queue_stopped(np->dev)) {
283 np->dev->xmit_lock_owner = -1;
284 spin_unlock(&np->dev->xmit_lock);
290 status = np->dev->hard_start_xmit(skb, np->dev);
291 np->dev->xmit_lock_owner = -1;
292 spin_unlock(&np->dev->xmit_lock);
296 npinfo->tries = MAX_RETRIES; /* reset */
303 } while (npinfo->tries > 0);
306 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
308 int total_len, eth_len, ip_len, udp_len;
314 udp_len = len + sizeof(*udph);
315 ip_len = eth_len = udp_len + sizeof(*iph);
316 total_len = eth_len + ETH_HLEN + NET_IP_ALIGN;
318 skb = find_skb(np, total_len, total_len - len);
322 memcpy(skb->data, msg, len);
325 udph = (struct udphdr *) skb_push(skb, sizeof(*udph));
326 udph->source = htons(np->local_port);
327 udph->dest = htons(np->remote_port);
328 udph->len = htons(udp_len);
331 iph = (struct iphdr *)skb_push(skb, sizeof(*iph));
333 /* iph->version = 4; iph->ihl = 5; */
334 put_unaligned(0x45, (unsigned char *)iph);
336 put_unaligned(htons(ip_len), &(iph->tot_len));
340 iph->protocol = IPPROTO_UDP;
342 put_unaligned(htonl(np->local_ip), &(iph->saddr));
343 put_unaligned(htonl(np->remote_ip), &(iph->daddr));
344 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
346 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
348 eth->h_proto = htons(ETH_P_IP);
349 memcpy(eth->h_source, np->local_mac, 6);
350 memcpy(eth->h_dest, np->remote_mac, 6);
354 netpoll_send_skb(np, skb);
357 static void arp_reply(struct sk_buff *skb)
359 struct netpoll_info *npinfo = skb->dev->npinfo;
361 unsigned char *arp_ptr;
362 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->h.raw = skb->nh.raw = skb->data;
384 if ((arp->ar_hrd != htons(ARPHRD_ETHER) &&
385 arp->ar_hrd != htons(ARPHRD_IEEE802)) ||
386 arp->ar_pro != htons(ETH_P_IP) ||
387 arp->ar_op != htons(ARPOP_REQUEST))
390 arp_ptr = (unsigned char *)(arp+1) + skb->dev->addr_len;
391 memcpy(&sip, arp_ptr, 4);
392 arp_ptr += 4 + skb->dev->addr_len;
393 memcpy(&tip, arp_ptr, 4);
395 /* Should we ignore arp? */
396 if (tip != htonl(np->local_ip) || LOOPBACK(tip) || MULTICAST(tip))
399 size = sizeof(struct arphdr) + 2 * (skb->dev->addr_len + 4);
400 send_skb = find_skb(np, size + LL_RESERVED_SPACE(np->dev),
401 LL_RESERVED_SPACE(np->dev));
406 send_skb->nh.raw = send_skb->data;
407 arp = (struct arphdr *) skb_put(send_skb, size);
408 send_skb->dev = skb->dev;
409 send_skb->protocol = htons(ETH_P_ARP);
411 /* Fill the device header for the ARP frame */
413 if (np->dev->hard_header &&
414 np->dev->hard_header(send_skb, skb->dev, ptype,
415 np->remote_mac, np->local_mac,
416 send_skb->len) < 0) {
422 * Fill out the arp protocol part.
424 * we only support ethernet device type,
425 * which (according to RFC 1390) should always equal 1 (Ethernet).
428 arp->ar_hrd = htons(np->dev->type);
429 arp->ar_pro = htons(ETH_P_IP);
430 arp->ar_hln = np->dev->addr_len;
432 arp->ar_op = htons(type);
434 arp_ptr=(unsigned char *)(arp + 1);
435 memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
436 arp_ptr += np->dev->addr_len;
437 memcpy(arp_ptr, &tip, 4);
439 memcpy(arp_ptr, np->remote_mac, np->dev->addr_len);
440 arp_ptr += np->dev->addr_len;
441 memcpy(arp_ptr, &sip, 4);
443 netpoll_send_skb(np, send_skb);
446 int __netpoll_rx(struct sk_buff *skb)
448 int proto, len, ulen;
451 struct netpoll *np = skb->dev->npinfo->rx_np;
455 if (skb->dev->type != ARPHRD_ETHER)
458 /* check if netpoll clients need ARP */
459 if (skb->protocol == __constant_htons(ETH_P_ARP) &&
460 atomic_read(&trapped)) {
465 proto = ntohs(eth_hdr(skb)->h_proto);
466 if (proto != ETH_P_IP)
468 if (skb->pkt_type == PACKET_OTHERHOST)
473 iph = (struct iphdr *)skb->data;
474 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
476 if (iph->ihl < 5 || iph->version != 4)
478 if (!pskb_may_pull(skb, iph->ihl*4))
480 if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
483 len = ntohs(iph->tot_len);
484 if (skb->len < len || len < iph->ihl*4)
487 if (iph->protocol != IPPROTO_UDP)
491 uh = (struct udphdr *)(((char *)iph) + iph->ihl*4);
492 ulen = ntohs(uh->len);
496 if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr))
498 if (np->local_ip && np->local_ip != ntohl(iph->daddr))
500 if (np->remote_ip && np->remote_ip != ntohl(iph->saddr))
502 if (np->local_port && np->local_port != ntohs(uh->dest))
505 np->rx_hook(np, ntohs(uh->source),
507 ulen - sizeof(struct udphdr));
513 if (atomic_read(&trapped)) {
521 int netpoll_parse_options(struct netpoll *np, char *opt)
523 char *cur=opt, *delim;
526 if ((delim = strchr(cur, '@')) == NULL)
529 np->local_port=simple_strtol(cur, NULL, 10);
533 printk(KERN_INFO "%s: local port %d\n", np->name, np->local_port);
536 if ((delim = strchr(cur, '/')) == NULL)
539 np->local_ip=ntohl(in_aton(cur));
542 printk(KERN_INFO "%s: local IP %d.%d.%d.%d\n",
543 np->name, HIPQUAD(np->local_ip));
548 /* parse out dev name */
549 if ((delim = strchr(cur, ',')) == NULL)
552 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
557 printk(KERN_INFO "%s: interface %s\n", np->name, np->dev_name);
561 if ((delim = strchr(cur, '@')) == NULL)
564 np->remote_port=simple_strtol(cur, NULL, 10);
568 printk(KERN_INFO "%s: remote port %d\n", np->name, np->remote_port);
571 if ((delim = strchr(cur, '/')) == NULL)
574 np->remote_ip=ntohl(in_aton(cur));
577 printk(KERN_INFO "%s: remote IP %d.%d.%d.%d\n",
578 np->name, HIPQUAD(np->remote_ip));
583 if ((delim = strchr(cur, ':')) == NULL)
586 np->remote_mac[0]=simple_strtol(cur, NULL, 16);
588 if ((delim = strchr(cur, ':')) == NULL)
591 np->remote_mac[1]=simple_strtol(cur, NULL, 16);
593 if ((delim = strchr(cur, ':')) == NULL)
596 np->remote_mac[2]=simple_strtol(cur, NULL, 16);
598 if ((delim = strchr(cur, ':')) == NULL)
601 np->remote_mac[3]=simple_strtol(cur, NULL, 16);
603 if ((delim = strchr(cur, ':')) == NULL)
606 np->remote_mac[4]=simple_strtol(cur, NULL, 16);
608 np->remote_mac[5]=simple_strtol(cur, NULL, 16);
611 printk(KERN_INFO "%s: remote ethernet address "
612 "%02x:%02x:%02x:%02x:%02x:%02x\n",
624 printk(KERN_INFO "%s: couldn't parse config at %s!\n",
629 int netpoll_setup(struct netpoll *np)
631 struct net_device *ndev = NULL;
632 struct in_device *in_dev;
633 struct netpoll_info *npinfo;
637 ndev = dev_get_by_name(np->dev_name);
639 printk(KERN_ERR "%s: %s doesn't exist, aborting.\n",
640 np->name, np->dev_name);
646 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
650 npinfo->rx_flags = 0;
651 npinfo->rx_np = NULL;
652 spin_lock_init(&npinfo->poll_lock);
653 npinfo->poll_owner = -1;
654 npinfo->tries = MAX_RETRIES;
655 spin_lock_init(&npinfo->rx_lock);
657 npinfo = ndev->npinfo;
659 if (!ndev->poll_controller) {
660 printk(KERN_ERR "%s: %s doesn't support polling, aborting.\n",
661 np->name, np->dev_name);
665 if (!netif_running(ndev)) {
666 unsigned long atmost, atleast;
668 printk(KERN_INFO "%s: device %s not up yet, forcing it\n",
669 np->name, np->dev_name);
672 if (dev_change_flags(ndev, ndev->flags | IFF_UP) < 0) {
673 printk(KERN_ERR "%s: failed to open %s\n",
674 np->name, np->dev_name);
680 atleast = jiffies + HZ/10;
681 atmost = jiffies + 4*HZ;
682 while (!netif_carrier_ok(ndev)) {
683 if (time_after(jiffies, atmost)) {
685 "%s: timeout waiting for carrier\n",
692 /* If carrier appears to come up instantly, we don't
693 * trust it and pause so that we don't pump all our
694 * queued console messages into the bitbucket.
697 if (time_before(jiffies, atleast)) {
698 printk(KERN_NOTICE "%s: carrier detect appears"
699 " untrustworthy, waiting 4 seconds\n",
705 if (!memcmp(np->local_mac, "\0\0\0\0\0\0", 6) && ndev->dev_addr)
706 memcpy(np->local_mac, ndev->dev_addr, 6);
710 in_dev = __in_dev_get_rcu(ndev);
712 if (!in_dev || !in_dev->ifa_list) {
714 printk(KERN_ERR "%s: no IP address for %s, aborting\n",
715 np->name, np->dev_name);
719 np->local_ip = ntohl(in_dev->ifa_list->ifa_local);
721 printk(KERN_INFO "%s: local IP %d.%d.%d.%d\n",
722 np->name, HIPQUAD(np->local_ip));
726 spin_lock_irqsave(&npinfo->rx_lock, flags);
727 npinfo->rx_flags |= NETPOLL_RX_ENABLED;
729 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
732 /* fill up the skb queue */
735 /* last thing to do is link it to the net device structure */
736 ndev->npinfo = npinfo;
738 /* avoid racing with NAPI reading npinfo */
751 void netpoll_cleanup(struct netpoll *np)
753 struct netpoll_info *npinfo;
757 npinfo = np->dev->npinfo;
758 if (npinfo && npinfo->rx_np == np) {
759 spin_lock_irqsave(&npinfo->rx_lock, flags);
760 npinfo->rx_np = NULL;
761 npinfo->rx_flags &= ~NETPOLL_RX_ENABLED;
762 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
770 int netpoll_trap(void)
772 return atomic_read(&trapped);
775 void netpoll_set_trap(int trap)
778 atomic_inc(&trapped);
780 atomic_dec(&trapped);
783 EXPORT_SYMBOL(netpoll_set_trap);
784 EXPORT_SYMBOL(netpoll_trap);
785 EXPORT_SYMBOL(netpoll_parse_options);
786 EXPORT_SYMBOL(netpoll_setup);
787 EXPORT_SYMBOL(netpoll_cleanup);
788 EXPORT_SYMBOL(netpoll_send_udp);
789 EXPORT_SYMBOL(netpoll_poll);
790 EXPORT_SYMBOL(netpoll_queue);