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/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>
27 #include <asm/unaligned.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)
37 #define MAX_RETRIES 20000
39 static DEFINE_SPINLOCK(skb_list_lock
);
41 static struct sk_buff
*skbs
;
43 static DEFINE_SPINLOCK(queue_lock
);
44 static int queue_depth
;
45 static struct sk_buff
*queue_head
, *queue_tail
;
47 static atomic_t trapped
;
49 #define NETPOLL_RX_ENABLED 1
50 #define NETPOLL_RX_DROP 2
52 #define MAX_SKB_SIZE \
53 (MAX_UDP_CHUNK + sizeof(struct udphdr) + \
54 sizeof(struct iphdr) + sizeof(struct ethhdr))
56 static void zap_completion_queue(void);
58 static void queue_process(void *p
)
64 spin_lock_irqsave(&queue_lock
, flags
);
67 queue_head
= skb
->next
;
68 if (skb
== queue_tail
)
73 spin_unlock_irqrestore(&queue_lock
, flags
);
79 static DECLARE_WORK(send_queue
, queue_process
, NULL
);
81 void netpoll_queue(struct sk_buff
*skb
)
85 if (queue_depth
== MAX_QUEUE_DEPTH
) {
90 spin_lock_irqsave(&queue_lock
, flags
);
94 queue_tail
->next
= skb
;
97 spin_unlock_irqrestore(&queue_lock
, flags
);
99 schedule_work(&send_queue
);
102 static int checksum_udp(struct sk_buff
*skb
, struct udphdr
*uh
,
103 unsigned short ulen
, u32 saddr
, u32 daddr
)
107 if (uh
->check
== 0 || skb
->ip_summed
== CHECKSUM_UNNECESSARY
)
110 psum
= csum_tcpudp_nofold(saddr
, daddr
, ulen
, IPPROTO_UDP
, 0);
112 if (skb
->ip_summed
== CHECKSUM_HW
&&
113 !(u16
)csum_fold(csum_add(psum
, skb
->csum
)))
118 return __skb_checksum_complete(skb
);
122 * Check whether delayed processing was scheduled for our NIC. If so,
123 * we attempt to grab the poll lock and use ->poll() to pump the card.
124 * If this fails, either we've recursed in ->poll() or it's already
125 * running on another CPU.
127 * Note: we don't mask interrupts with this lock because we're using
128 * trylock here and interrupts are already disabled in the softirq
129 * case. Further, we test the poll_owner to avoid recursion on UP
130 * systems where the lock doesn't exist.
132 * In cases where there is bi-directional communications, reading only
133 * one message at a time can lead to packets being dropped by the
134 * network adapter, forcing superfluous retries and possibly timeouts.
135 * Thus, we set our budget to greater than 1.
137 static void poll_napi(struct netpoll
*np
)
139 struct netpoll_info
*npinfo
= np
->dev
->npinfo
;
142 if (test_bit(__LINK_STATE_RX_SCHED
, &np
->dev
->state
) &&
143 npinfo
->poll_owner
!= smp_processor_id() &&
144 spin_trylock(&npinfo
->poll_lock
)) {
145 npinfo
->rx_flags
|= NETPOLL_RX_DROP
;
146 atomic_inc(&trapped
);
148 np
->dev
->poll(np
->dev
, &budget
);
150 atomic_dec(&trapped
);
151 npinfo
->rx_flags
&= ~NETPOLL_RX_DROP
;
152 spin_unlock(&npinfo
->poll_lock
);
156 void netpoll_poll(struct netpoll
*np
)
158 if(!np
->dev
|| !netif_running(np
->dev
) || !np
->dev
->poll_controller
)
161 /* Process pending work on NIC */
162 np
->dev
->poll_controller(np
->dev
);
166 zap_completion_queue();
169 static void refill_skbs(void)
174 spin_lock_irqsave(&skb_list_lock
, flags
);
175 while (nr_skbs
< MAX_SKBS
) {
176 skb
= alloc_skb(MAX_SKB_SIZE
, GFP_ATOMIC
);
184 spin_unlock_irqrestore(&skb_list_lock
, flags
);
187 static void zap_completion_queue(void)
190 struct softnet_data
*sd
= &get_cpu_var(softnet_data
);
192 if (sd
->completion_queue
) {
193 struct sk_buff
*clist
;
195 local_irq_save(flags
);
196 clist
= sd
->completion_queue
;
197 sd
->completion_queue
= NULL
;
198 local_irq_restore(flags
);
200 while (clist
!= NULL
) {
201 struct sk_buff
*skb
= clist
;
204 dev_kfree_skb_any(skb
); /* put this one back */
210 put_cpu_var(softnet_data
);
213 static struct sk_buff
* find_skb(struct netpoll
*np
, int len
, int reserve
)
215 int once
= 1, count
= 0;
217 struct sk_buff
*skb
= NULL
;
219 zap_completion_queue();
221 if (nr_skbs
< MAX_SKBS
)
224 skb
= alloc_skb(len
, GFP_ATOMIC
);
227 spin_lock_irqsave(&skb_list_lock
, flags
);
234 spin_unlock_irqrestore(&skb_list_lock
, flags
);
239 if (once
&& (count
== 1000000)) {
240 printk("out of netpoll skbs!\n");
247 atomic_set(&skb
->users
, 1);
248 skb_reserve(skb
, reserve
);
252 static void netpoll_send_skb(struct netpoll
*np
, struct sk_buff
*skb
)
255 struct netpoll_info
*npinfo
;
257 if (!np
|| !np
->dev
|| !netif_running(np
->dev
)) {
262 npinfo
= np
->dev
->npinfo
;
264 /* avoid recursion */
265 if (npinfo
->poll_owner
== smp_processor_id() ||
266 np
->dev
->xmit_lock_owner
== smp_processor_id()) {
276 netif_tx_lock(np
->dev
);
279 * network drivers do not expect to be called if the queue is
282 if (netif_queue_stopped(np
->dev
)) {
283 netif_tx_unlock(np
->dev
);
289 status
= np
->dev
->hard_start_xmit(skb
, np
->dev
);
290 netif_tx_unlock(np
->dev
);
294 npinfo
->tries
= MAX_RETRIES
; /* reset */
301 } while (npinfo
->tries
> 0);
304 void netpoll_send_udp(struct netpoll
*np
, const char *msg
, int len
)
306 int total_len
, eth_len
, ip_len
, udp_len
;
312 udp_len
= len
+ sizeof(*udph
);
313 ip_len
= eth_len
= udp_len
+ sizeof(*iph
);
314 total_len
= eth_len
+ ETH_HLEN
+ NET_IP_ALIGN
;
316 skb
= find_skb(np
, total_len
, total_len
- len
);
320 memcpy(skb
->data
, msg
, len
);
323 udph
= (struct udphdr
*) skb_push(skb
, sizeof(*udph
));
324 udph
->source
= htons(np
->local_port
);
325 udph
->dest
= htons(np
->remote_port
);
326 udph
->len
= htons(udp_len
);
329 iph
= (struct iphdr
*)skb_push(skb
, sizeof(*iph
));
331 /* iph->version = 4; iph->ihl = 5; */
332 put_unaligned(0x45, (unsigned char *)iph
);
334 put_unaligned(htons(ip_len
), &(iph
->tot_len
));
338 iph
->protocol
= IPPROTO_UDP
;
340 put_unaligned(htonl(np
->local_ip
), &(iph
->saddr
));
341 put_unaligned(htonl(np
->remote_ip
), &(iph
->daddr
));
342 iph
->check
= ip_fast_csum((unsigned char *)iph
, iph
->ihl
);
344 eth
= (struct ethhdr
*) skb_push(skb
, ETH_HLEN
);
346 eth
->h_proto
= htons(ETH_P_IP
);
347 memcpy(eth
->h_source
, np
->local_mac
, 6);
348 memcpy(eth
->h_dest
, np
->remote_mac
, 6);
352 netpoll_send_skb(np
, skb
);
355 static void arp_reply(struct sk_buff
*skb
)
357 struct netpoll_info
*npinfo
= skb
->dev
->npinfo
;
359 unsigned char *arp_ptr
;
360 int size
, type
= ARPOP_REPLY
, ptype
= ETH_P_ARP
;
362 struct sk_buff
*send_skb
;
363 struct netpoll
*np
= NULL
;
365 if (npinfo
->rx_np
&& npinfo
->rx_np
->dev
== skb
->dev
)
370 /* No arp on this interface */
371 if (skb
->dev
->flags
& IFF_NOARP
)
374 if (!pskb_may_pull(skb
, (sizeof(struct arphdr
) +
375 (2 * skb
->dev
->addr_len
) +
379 skb
->h
.raw
= skb
->nh
.raw
= skb
->data
;
382 if ((arp
->ar_hrd
!= htons(ARPHRD_ETHER
) &&
383 arp
->ar_hrd
!= htons(ARPHRD_IEEE802
)) ||
384 arp
->ar_pro
!= htons(ETH_P_IP
) ||
385 arp
->ar_op
!= htons(ARPOP_REQUEST
))
388 arp_ptr
= (unsigned char *)(arp
+1) + skb
->dev
->addr_len
;
389 memcpy(&sip
, arp_ptr
, 4);
390 arp_ptr
+= 4 + skb
->dev
->addr_len
;
391 memcpy(&tip
, arp_ptr
, 4);
393 /* Should we ignore arp? */
394 if (tip
!= htonl(np
->local_ip
) || LOOPBACK(tip
) || MULTICAST(tip
))
397 size
= sizeof(struct arphdr
) + 2 * (skb
->dev
->addr_len
+ 4);
398 send_skb
= find_skb(np
, size
+ LL_RESERVED_SPACE(np
->dev
),
399 LL_RESERVED_SPACE(np
->dev
));
404 send_skb
->nh
.raw
= send_skb
->data
;
405 arp
= (struct arphdr
*) skb_put(send_skb
, size
);
406 send_skb
->dev
= skb
->dev
;
407 send_skb
->protocol
= htons(ETH_P_ARP
);
409 /* Fill the device header for the ARP frame */
411 if (np
->dev
->hard_header
&&
412 np
->dev
->hard_header(send_skb
, skb
->dev
, ptype
,
413 np
->remote_mac
, np
->local_mac
,
414 send_skb
->len
) < 0) {
420 * Fill out the arp protocol part.
422 * we only support ethernet device type,
423 * which (according to RFC 1390) should always equal 1 (Ethernet).
426 arp
->ar_hrd
= htons(np
->dev
->type
);
427 arp
->ar_pro
= htons(ETH_P_IP
);
428 arp
->ar_hln
= np
->dev
->addr_len
;
430 arp
->ar_op
= htons(type
);
432 arp_ptr
=(unsigned char *)(arp
+ 1);
433 memcpy(arp_ptr
, np
->dev
->dev_addr
, np
->dev
->addr_len
);
434 arp_ptr
+= np
->dev
->addr_len
;
435 memcpy(arp_ptr
, &tip
, 4);
437 memcpy(arp_ptr
, np
->remote_mac
, np
->dev
->addr_len
);
438 arp_ptr
+= np
->dev
->addr_len
;
439 memcpy(arp_ptr
, &sip
, 4);
441 netpoll_send_skb(np
, send_skb
);
444 int __netpoll_rx(struct sk_buff
*skb
)
446 int proto
, len
, ulen
;
449 struct netpoll
*np
= skb
->dev
->npinfo
->rx_np
;
453 if (skb
->dev
->type
!= ARPHRD_ETHER
)
456 /* check if netpoll clients need ARP */
457 if (skb
->protocol
== __constant_htons(ETH_P_ARP
) &&
458 atomic_read(&trapped
)) {
463 proto
= ntohs(eth_hdr(skb
)->h_proto
);
464 if (proto
!= ETH_P_IP
)
466 if (skb
->pkt_type
== PACKET_OTHERHOST
)
471 iph
= (struct iphdr
*)skb
->data
;
472 if (!pskb_may_pull(skb
, sizeof(struct iphdr
)))
474 if (iph
->ihl
< 5 || iph
->version
!= 4)
476 if (!pskb_may_pull(skb
, iph
->ihl
*4))
478 if (ip_fast_csum((u8
*)iph
, iph
->ihl
) != 0)
481 len
= ntohs(iph
->tot_len
);
482 if (skb
->len
< len
|| len
< iph
->ihl
*4)
485 if (iph
->protocol
!= IPPROTO_UDP
)
489 uh
= (struct udphdr
*)(((char *)iph
) + iph
->ihl
*4);
490 ulen
= ntohs(uh
->len
);
494 if (checksum_udp(skb
, uh
, ulen
, iph
->saddr
, iph
->daddr
))
496 if (np
->local_ip
&& np
->local_ip
!= ntohl(iph
->daddr
))
498 if (np
->remote_ip
&& np
->remote_ip
!= ntohl(iph
->saddr
))
500 if (np
->local_port
&& np
->local_port
!= ntohs(uh
->dest
))
503 np
->rx_hook(np
, ntohs(uh
->source
),
505 ulen
- sizeof(struct udphdr
));
511 if (atomic_read(&trapped
)) {
519 int netpoll_parse_options(struct netpoll
*np
, char *opt
)
521 char *cur
=opt
, *delim
;
524 if ((delim
= strchr(cur
, '@')) == NULL
)
527 np
->local_port
=simple_strtol(cur
, NULL
, 10);
531 printk(KERN_INFO
"%s: local port %d\n", np
->name
, np
->local_port
);
534 if ((delim
= strchr(cur
, '/')) == NULL
)
537 np
->local_ip
=ntohl(in_aton(cur
));
540 printk(KERN_INFO
"%s: local IP %d.%d.%d.%d\n",
541 np
->name
, HIPQUAD(np
->local_ip
));
546 /* parse out dev name */
547 if ((delim
= strchr(cur
, ',')) == NULL
)
550 strlcpy(np
->dev_name
, cur
, sizeof(np
->dev_name
));
555 printk(KERN_INFO
"%s: interface %s\n", np
->name
, np
->dev_name
);
559 if ((delim
= strchr(cur
, '@')) == NULL
)
562 np
->remote_port
=simple_strtol(cur
, NULL
, 10);
566 printk(KERN_INFO
"%s: remote port %d\n", np
->name
, np
->remote_port
);
569 if ((delim
= strchr(cur
, '/')) == NULL
)
572 np
->remote_ip
=ntohl(in_aton(cur
));
575 printk(KERN_INFO
"%s: remote IP %d.%d.%d.%d\n",
576 np
->name
, HIPQUAD(np
->remote_ip
));
581 if ((delim
= strchr(cur
, ':')) == NULL
)
584 np
->remote_mac
[0]=simple_strtol(cur
, NULL
, 16);
586 if ((delim
= strchr(cur
, ':')) == NULL
)
589 np
->remote_mac
[1]=simple_strtol(cur
, NULL
, 16);
591 if ((delim
= strchr(cur
, ':')) == NULL
)
594 np
->remote_mac
[2]=simple_strtol(cur
, NULL
, 16);
596 if ((delim
= strchr(cur
, ':')) == NULL
)
599 np
->remote_mac
[3]=simple_strtol(cur
, NULL
, 16);
601 if ((delim
= strchr(cur
, ':')) == NULL
)
604 np
->remote_mac
[4]=simple_strtol(cur
, NULL
, 16);
606 np
->remote_mac
[5]=simple_strtol(cur
, NULL
, 16);
609 printk(KERN_INFO
"%s: remote ethernet address "
610 "%02x:%02x:%02x:%02x:%02x:%02x\n",
622 printk(KERN_INFO
"%s: couldn't parse config at %s!\n",
627 int netpoll_setup(struct netpoll
*np
)
629 struct net_device
*ndev
= NULL
;
630 struct in_device
*in_dev
;
631 struct netpoll_info
*npinfo
;
635 ndev
= dev_get_by_name(np
->dev_name
);
637 printk(KERN_ERR
"%s: %s doesn't exist, aborting.\n",
638 np
->name
, np
->dev_name
);
644 npinfo
= kmalloc(sizeof(*npinfo
), GFP_KERNEL
);
648 npinfo
->rx_flags
= 0;
649 npinfo
->rx_np
= NULL
;
650 spin_lock_init(&npinfo
->poll_lock
);
651 npinfo
->poll_owner
= -1;
652 npinfo
->tries
= MAX_RETRIES
;
653 spin_lock_init(&npinfo
->rx_lock
);
655 npinfo
= ndev
->npinfo
;
657 if (!ndev
->poll_controller
) {
658 printk(KERN_ERR
"%s: %s doesn't support polling, aborting.\n",
659 np
->name
, np
->dev_name
);
663 if (!netif_running(ndev
)) {
664 unsigned long atmost
, atleast
;
666 printk(KERN_INFO
"%s: device %s not up yet, forcing it\n",
667 np
->name
, np
->dev_name
);
670 if (dev_change_flags(ndev
, ndev
->flags
| IFF_UP
) < 0) {
671 printk(KERN_ERR
"%s: failed to open %s\n",
672 np
->name
, np
->dev_name
);
678 atleast
= jiffies
+ HZ
/10;
679 atmost
= jiffies
+ 4*HZ
;
680 while (!netif_carrier_ok(ndev
)) {
681 if (time_after(jiffies
, atmost
)) {
683 "%s: timeout waiting for carrier\n",
690 /* If carrier appears to come up instantly, we don't
691 * trust it and pause so that we don't pump all our
692 * queued console messages into the bitbucket.
695 if (time_before(jiffies
, atleast
)) {
696 printk(KERN_NOTICE
"%s: carrier detect appears"
697 " untrustworthy, waiting 4 seconds\n",
703 if (is_zero_ether_addr(np
->local_mac
) && ndev
->dev_addr
)
704 memcpy(np
->local_mac
, ndev
->dev_addr
, 6);
708 in_dev
= __in_dev_get_rcu(ndev
);
710 if (!in_dev
|| !in_dev
->ifa_list
) {
712 printk(KERN_ERR
"%s: no IP address for %s, aborting\n",
713 np
->name
, np
->dev_name
);
717 np
->local_ip
= ntohl(in_dev
->ifa_list
->ifa_local
);
719 printk(KERN_INFO
"%s: local IP %d.%d.%d.%d\n",
720 np
->name
, HIPQUAD(np
->local_ip
));
724 spin_lock_irqsave(&npinfo
->rx_lock
, flags
);
725 npinfo
->rx_flags
|= NETPOLL_RX_ENABLED
;
727 spin_unlock_irqrestore(&npinfo
->rx_lock
, flags
);
730 /* fill up the skb queue */
733 /* last thing to do is link it to the net device structure */
734 ndev
->npinfo
= npinfo
;
736 /* avoid racing with NAPI reading npinfo */
749 void netpoll_cleanup(struct netpoll
*np
)
751 struct netpoll_info
*npinfo
;
755 npinfo
= np
->dev
->npinfo
;
756 if (npinfo
&& npinfo
->rx_np
== np
) {
757 spin_lock_irqsave(&npinfo
->rx_lock
, flags
);
758 npinfo
->rx_np
= NULL
;
759 npinfo
->rx_flags
&= ~NETPOLL_RX_ENABLED
;
760 spin_unlock_irqrestore(&npinfo
->rx_lock
, flags
);
768 int netpoll_trap(void)
770 return atomic_read(&trapped
);
773 void netpoll_set_trap(int trap
)
776 atomic_inc(&trapped
);
778 atomic_dec(&trapped
);
781 EXPORT_SYMBOL(netpoll_set_trap
);
782 EXPORT_SYMBOL(netpoll_trap
);
783 EXPORT_SYMBOL(netpoll_parse_options
);
784 EXPORT_SYMBOL(netpoll_setup
);
785 EXPORT_SYMBOL(netpoll_cleanup
);
786 EXPORT_SYMBOL(netpoll_send_udp
);
787 EXPORT_SYMBOL(netpoll_poll
);
788 EXPORT_SYMBOL(netpoll_queue
);