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)
37 static DEFINE_SPINLOCK(skb_list_lock
);
39 static struct sk_buff
*skbs
;
41 static DEFINE_SPINLOCK(queue_lock
);
42 static int queue_depth
;
43 static struct sk_buff
*queue_head
, *queue_tail
;
45 static atomic_t trapped
;
47 #define NETPOLL_RX_ENABLED 1
48 #define NETPOLL_RX_DROP 2
50 #define MAX_SKB_SIZE \
51 (MAX_UDP_CHUNK + sizeof(struct udphdr) + \
52 sizeof(struct iphdr) + sizeof(struct ethhdr))
54 static void zap_completion_queue(void);
56 static void queue_process(void *p
)
62 spin_lock_irqsave(&queue_lock
, flags
);
65 queue_head
= skb
->next
;
66 if (skb
== queue_tail
)
71 spin_unlock_irqrestore(&queue_lock
, flags
);
77 static DECLARE_WORK(send_queue
, queue_process
, NULL
);
79 void netpoll_queue(struct sk_buff
*skb
)
83 if (queue_depth
== MAX_QUEUE_DEPTH
) {
88 spin_lock_irqsave(&queue_lock
, flags
);
92 queue_tail
->next
= skb
;
95 spin_unlock_irqrestore(&queue_lock
, flags
);
97 schedule_work(&send_queue
);
100 static int checksum_udp(struct sk_buff
*skb
, struct udphdr
*uh
,
101 unsigned short ulen
, u32 saddr
, u32 daddr
)
106 if (skb
->ip_summed
== CHECKSUM_HW
)
107 return csum_tcpudp_magic(
108 saddr
, daddr
, ulen
, IPPROTO_UDP
, skb
->csum
);
110 skb
->csum
= csum_tcpudp_nofold(saddr
, daddr
, ulen
, IPPROTO_UDP
, 0);
112 return csum_fold(skb_checksum(skb
, 0, skb
->len
, skb
->csum
));
116 * Check whether delayed processing was scheduled for our NIC. If so,
117 * we attempt to grab the poll lock and use ->poll() to pump the card.
118 * If this fails, either we've recursed in ->poll() or it's already
119 * running on another CPU.
121 * Note: we don't mask interrupts with this lock because we're using
122 * trylock here and interrupts are already disabled in the softirq
123 * case. Further, we test the poll_owner to avoid recursion on UP
124 * systems where the lock doesn't exist.
126 * In cases where there is bi-directional communications, reading only
127 * one message at a time can lead to packets being dropped by the
128 * network adapter, forcing superfluous retries and possibly timeouts.
129 * Thus, we set our budget to greater than 1.
131 static void poll_napi(struct netpoll
*np
)
135 if (test_bit(__LINK_STATE_RX_SCHED
, &np
->dev
->state
) &&
136 np
->poll_owner
!= smp_processor_id() &&
137 spin_trylock(&np
->poll_lock
)) {
138 np
->rx_flags
|= NETPOLL_RX_DROP
;
139 atomic_inc(&trapped
);
141 np
->dev
->poll(np
->dev
, &budget
);
143 atomic_dec(&trapped
);
144 np
->rx_flags
&= ~NETPOLL_RX_DROP
;
145 spin_unlock(&np
->poll_lock
);
149 void netpoll_poll(struct netpoll
*np
)
151 if(!np
->dev
|| !netif_running(np
->dev
) || !np
->dev
->poll_controller
)
154 /* Process pending work on NIC */
155 np
->dev
->poll_controller(np
->dev
);
159 zap_completion_queue();
162 static void refill_skbs(void)
167 spin_lock_irqsave(&skb_list_lock
, flags
);
168 while (nr_skbs
< MAX_SKBS
) {
169 skb
= alloc_skb(MAX_SKB_SIZE
, GFP_ATOMIC
);
177 spin_unlock_irqrestore(&skb_list_lock
, flags
);
180 static void zap_completion_queue(void)
183 struct softnet_data
*sd
= &get_cpu_var(softnet_data
);
185 if (sd
->completion_queue
) {
186 struct sk_buff
*clist
;
188 local_irq_save(flags
);
189 clist
= sd
->completion_queue
;
190 sd
->completion_queue
= NULL
;
191 local_irq_restore(flags
);
193 while (clist
!= NULL
) {
194 struct sk_buff
*skb
= clist
;
197 dev_kfree_skb_any(skb
); /* put this one back */
203 put_cpu_var(softnet_data
);
206 static struct sk_buff
* find_skb(struct netpoll
*np
, int len
, int reserve
)
208 int once
= 1, count
= 0;
210 struct sk_buff
*skb
= NULL
;
212 zap_completion_queue();
214 if (nr_skbs
< MAX_SKBS
)
217 skb
= alloc_skb(len
, GFP_ATOMIC
);
220 spin_lock_irqsave(&skb_list_lock
, flags
);
227 spin_unlock_irqrestore(&skb_list_lock
, flags
);
232 if (once
&& (count
== 1000000)) {
233 printk("out of netpoll skbs!\n");
240 atomic_set(&skb
->users
, 1);
241 skb_reserve(skb
, reserve
);
245 static void netpoll_send_skb(struct netpoll
*np
, struct sk_buff
*skb
)
250 if(!np
|| !np
->dev
|| !netif_running(np
->dev
)) {
255 /* avoid recursion */
256 if(np
->poll_owner
== smp_processor_id() ||
257 np
->dev
->xmit_lock_owner
== smp_processor_id()) {
265 spin_lock(&np
->dev
->xmit_lock
);
266 np
->dev
->xmit_lock_owner
= smp_processor_id();
269 * network drivers do not expect to be called if the queue is
272 if (netif_queue_stopped(np
->dev
)) {
273 np
->dev
->xmit_lock_owner
= -1;
274 spin_unlock(&np
->dev
->xmit_lock
);
280 status
= np
->dev
->hard_start_xmit(skb
, np
->dev
);
281 np
->dev
->xmit_lock_owner
= -1;
282 spin_unlock(&np
->dev
->xmit_lock
);
291 void netpoll_send_udp(struct netpoll
*np
, const char *msg
, int len
)
293 int total_len
, eth_len
, ip_len
, udp_len
;
299 udp_len
= len
+ sizeof(*udph
);
300 ip_len
= eth_len
= udp_len
+ sizeof(*iph
);
301 total_len
= eth_len
+ ETH_HLEN
+ NET_IP_ALIGN
;
303 skb
= find_skb(np
, total_len
, total_len
- len
);
307 memcpy(skb
->data
, msg
, len
);
310 udph
= (struct udphdr
*) skb_push(skb
, sizeof(*udph
));
311 udph
->source
= htons(np
->local_port
);
312 udph
->dest
= htons(np
->remote_port
);
313 udph
->len
= htons(udp_len
);
316 iph
= (struct iphdr
*)skb_push(skb
, sizeof(*iph
));
318 /* iph->version = 4; iph->ihl = 5; */
319 put_unaligned(0x45, (unsigned char *)iph
);
321 put_unaligned(htons(ip_len
), &(iph
->tot_len
));
325 iph
->protocol
= IPPROTO_UDP
;
327 put_unaligned(htonl(np
->local_ip
), &(iph
->saddr
));
328 put_unaligned(htonl(np
->remote_ip
), &(iph
->daddr
));
329 iph
->check
= ip_fast_csum((unsigned char *)iph
, iph
->ihl
);
331 eth
= (struct ethhdr
*) skb_push(skb
, ETH_HLEN
);
333 eth
->h_proto
= htons(ETH_P_IP
);
334 memcpy(eth
->h_source
, np
->local_mac
, 6);
335 memcpy(eth
->h_dest
, np
->remote_mac
, 6);
339 netpoll_send_skb(np
, skb
);
342 static void arp_reply(struct sk_buff
*skb
)
345 unsigned char *arp_ptr
;
346 int size
, type
= ARPOP_REPLY
, ptype
= ETH_P_ARP
;
348 struct sk_buff
*send_skb
;
349 struct netpoll
*np
= skb
->dev
->np
;
353 /* No arp on this interface */
354 if (skb
->dev
->flags
& IFF_NOARP
)
357 if (!pskb_may_pull(skb
, (sizeof(struct arphdr
) +
358 (2 * skb
->dev
->addr_len
) +
362 skb
->h
.raw
= skb
->nh
.raw
= skb
->data
;
365 if ((arp
->ar_hrd
!= htons(ARPHRD_ETHER
) &&
366 arp
->ar_hrd
!= htons(ARPHRD_IEEE802
)) ||
367 arp
->ar_pro
!= htons(ETH_P_IP
) ||
368 arp
->ar_op
!= htons(ARPOP_REQUEST
))
371 arp_ptr
= (unsigned char *)(arp
+1) + skb
->dev
->addr_len
;
372 memcpy(&sip
, arp_ptr
, 4);
373 arp_ptr
+= 4 + skb
->dev
->addr_len
;
374 memcpy(&tip
, arp_ptr
, 4);
376 /* Should we ignore arp? */
377 if (tip
!= htonl(np
->local_ip
) || LOOPBACK(tip
) || MULTICAST(tip
))
380 size
= sizeof(struct arphdr
) + 2 * (skb
->dev
->addr_len
+ 4);
381 send_skb
= find_skb(np
, size
+ LL_RESERVED_SPACE(np
->dev
),
382 LL_RESERVED_SPACE(np
->dev
));
387 send_skb
->nh
.raw
= send_skb
->data
;
388 arp
= (struct arphdr
*) skb_put(send_skb
, size
);
389 send_skb
->dev
= skb
->dev
;
390 send_skb
->protocol
= htons(ETH_P_ARP
);
392 /* Fill the device header for the ARP frame */
394 if (np
->dev
->hard_header
&&
395 np
->dev
->hard_header(send_skb
, skb
->dev
, ptype
,
396 np
->remote_mac
, np
->local_mac
,
397 send_skb
->len
) < 0) {
403 * Fill out the arp protocol part.
405 * we only support ethernet device type,
406 * which (according to RFC 1390) should always equal 1 (Ethernet).
409 arp
->ar_hrd
= htons(np
->dev
->type
);
410 arp
->ar_pro
= htons(ETH_P_IP
);
411 arp
->ar_hln
= np
->dev
->addr_len
;
413 arp
->ar_op
= htons(type
);
415 arp_ptr
=(unsigned char *)(arp
+ 1);
416 memcpy(arp_ptr
, np
->dev
->dev_addr
, np
->dev
->addr_len
);
417 arp_ptr
+= np
->dev
->addr_len
;
418 memcpy(arp_ptr
, &tip
, 4);
420 memcpy(arp_ptr
, np
->remote_mac
, np
->dev
->addr_len
);
421 arp_ptr
+= np
->dev
->addr_len
;
422 memcpy(arp_ptr
, &sip
, 4);
424 netpoll_send_skb(np
, send_skb
);
427 int __netpoll_rx(struct sk_buff
*skb
)
429 int proto
, len
, ulen
;
432 struct netpoll
*np
= skb
->dev
->np
;
436 if (skb
->dev
->type
!= ARPHRD_ETHER
)
439 /* check if netpoll clients need ARP */
440 if (skb
->protocol
== __constant_htons(ETH_P_ARP
) &&
441 atomic_read(&trapped
)) {
446 proto
= ntohs(eth_hdr(skb
)->h_proto
);
447 if (proto
!= ETH_P_IP
)
449 if (skb
->pkt_type
== PACKET_OTHERHOST
)
454 iph
= (struct iphdr
*)skb
->data
;
455 if (!pskb_may_pull(skb
, sizeof(struct iphdr
)))
457 if (iph
->ihl
< 5 || iph
->version
!= 4)
459 if (!pskb_may_pull(skb
, iph
->ihl
*4))
461 if (ip_fast_csum((u8
*)iph
, iph
->ihl
) != 0)
464 len
= ntohs(iph
->tot_len
);
465 if (skb
->len
< len
|| len
< iph
->ihl
*4)
468 if (iph
->protocol
!= IPPROTO_UDP
)
472 uh
= (struct udphdr
*)(((char *)iph
) + iph
->ihl
*4);
473 ulen
= ntohs(uh
->len
);
477 if (checksum_udp(skb
, uh
, ulen
, iph
->saddr
, iph
->daddr
) < 0)
479 if (np
->local_ip
&& np
->local_ip
!= ntohl(iph
->daddr
))
481 if (np
->remote_ip
&& np
->remote_ip
!= ntohl(iph
->saddr
))
483 if (np
->local_port
&& np
->local_port
!= ntohs(uh
->dest
))
486 np
->rx_hook(np
, ntohs(uh
->source
),
488 ulen
- sizeof(struct udphdr
));
494 if (atomic_read(&trapped
)) {
502 int netpoll_parse_options(struct netpoll
*np
, char *opt
)
504 char *cur
=opt
, *delim
;
507 if ((delim
= strchr(cur
, '@')) == NULL
)
510 np
->local_port
=simple_strtol(cur
, NULL
, 10);
514 printk(KERN_INFO
"%s: local port %d\n", np
->name
, np
->local_port
);
517 if ((delim
= strchr(cur
, '/')) == NULL
)
520 np
->local_ip
=ntohl(in_aton(cur
));
523 printk(KERN_INFO
"%s: local IP %d.%d.%d.%d\n",
524 np
->name
, HIPQUAD(np
->local_ip
));
529 /* parse out dev name */
530 if ((delim
= strchr(cur
, ',')) == NULL
)
533 strlcpy(np
->dev_name
, cur
, sizeof(np
->dev_name
));
538 printk(KERN_INFO
"%s: interface %s\n", np
->name
, np
->dev_name
);
542 if ((delim
= strchr(cur
, '@')) == NULL
)
545 np
->remote_port
=simple_strtol(cur
, NULL
, 10);
549 printk(KERN_INFO
"%s: remote port %d\n", np
->name
, np
->remote_port
);
552 if ((delim
= strchr(cur
, '/')) == NULL
)
555 np
->remote_ip
=ntohl(in_aton(cur
));
558 printk(KERN_INFO
"%s: remote IP %d.%d.%d.%d\n",
559 np
->name
, HIPQUAD(np
->remote_ip
));
564 if ((delim
= strchr(cur
, ':')) == NULL
)
567 np
->remote_mac
[0]=simple_strtol(cur
, NULL
, 16);
569 if ((delim
= strchr(cur
, ':')) == NULL
)
572 np
->remote_mac
[1]=simple_strtol(cur
, NULL
, 16);
574 if ((delim
= strchr(cur
, ':')) == NULL
)
577 np
->remote_mac
[2]=simple_strtol(cur
, NULL
, 16);
579 if ((delim
= strchr(cur
, ':')) == NULL
)
582 np
->remote_mac
[3]=simple_strtol(cur
, NULL
, 16);
584 if ((delim
= strchr(cur
, ':')) == NULL
)
587 np
->remote_mac
[4]=simple_strtol(cur
, NULL
, 16);
589 np
->remote_mac
[5]=simple_strtol(cur
, NULL
, 16);
592 printk(KERN_INFO
"%s: remote ethernet address "
593 "%02x:%02x:%02x:%02x:%02x:%02x\n",
605 printk(KERN_INFO
"%s: couldn't parse config at %s!\n",
610 int netpoll_setup(struct netpoll
*np
)
612 struct net_device
*ndev
= NULL
;
613 struct in_device
*in_dev
;
615 np
->poll_lock
= SPIN_LOCK_UNLOCKED
;
619 ndev
= dev_get_by_name(np
->dev_name
);
621 printk(KERN_ERR
"%s: %s doesn't exist, aborting.\n",
622 np
->name
, np
->dev_name
);
629 if (!ndev
->poll_controller
) {
630 printk(KERN_ERR
"%s: %s doesn't support polling, aborting.\n",
631 np
->name
, np
->dev_name
);
635 if (!netif_running(ndev
)) {
636 unsigned long atmost
, atleast
;
638 printk(KERN_INFO
"%s: device %s not up yet, forcing it\n",
639 np
->name
, np
->dev_name
);
642 if (dev_change_flags(ndev
, ndev
->flags
| IFF_UP
) < 0) {
643 printk(KERN_ERR
"%s: failed to open %s\n",
644 np
->name
, np
->dev_name
);
650 atleast
= jiffies
+ HZ
/10;
651 atmost
= jiffies
+ 4*HZ
;
652 while (!netif_carrier_ok(ndev
)) {
653 if (time_after(jiffies
, atmost
)) {
655 "%s: timeout waiting for carrier\n",
662 /* If carrier appears to come up instantly, we don't
663 * trust it and pause so that we don't pump all our
664 * queued console messages into the bitbucket.
667 if (time_before(jiffies
, atleast
)) {
668 printk(KERN_NOTICE
"%s: carrier detect appears"
669 " untrustworthy, waiting 4 seconds\n",
675 if (!memcmp(np
->local_mac
, "\0\0\0\0\0\0", 6) && ndev
->dev_addr
)
676 memcpy(np
->local_mac
, ndev
->dev_addr
, 6);
680 in_dev
= __in_dev_get(ndev
);
682 if (!in_dev
|| !in_dev
->ifa_list
) {
684 printk(KERN_ERR
"%s: no IP address for %s, aborting\n",
685 np
->name
, np
->dev_name
);
689 np
->local_ip
= ntohl(in_dev
->ifa_list
->ifa_local
);
691 printk(KERN_INFO
"%s: local IP %d.%d.%d.%d\n",
692 np
->name
, HIPQUAD(np
->local_ip
));
696 np
->rx_flags
= NETPOLL_RX_ENABLED
;
707 void netpoll_cleanup(struct netpoll
*np
)
715 int netpoll_trap(void)
717 return atomic_read(&trapped
);
720 void netpoll_set_trap(int trap
)
723 atomic_inc(&trapped
);
725 atomic_dec(&trapped
);
728 EXPORT_SYMBOL(netpoll_set_trap
);
729 EXPORT_SYMBOL(netpoll_trap
);
730 EXPORT_SYMBOL(netpoll_parse_options
);
731 EXPORT_SYMBOL(netpoll_setup
);
732 EXPORT_SYMBOL(netpoll_cleanup
);
733 EXPORT_SYMBOL(netpoll_send_udp
);
734 EXPORT_SYMBOL(netpoll_poll
);
735 EXPORT_SYMBOL(netpoll_queue
);