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);
57 static void arp_reply(struct sk_buff
*skb
);
59 static void queue_process(void *p
)
65 spin_lock_irqsave(&queue_lock
, flags
);
68 queue_head
= skb
->next
;
69 if (skb
== queue_tail
)
74 spin_unlock_irqrestore(&queue_lock
, flags
);
80 static DECLARE_WORK(send_queue
, queue_process
, NULL
);
82 void netpoll_queue(struct sk_buff
*skb
)
86 if (queue_depth
== MAX_QUEUE_DEPTH
) {
91 spin_lock_irqsave(&queue_lock
, flags
);
95 queue_tail
->next
= skb
;
98 spin_unlock_irqrestore(&queue_lock
, flags
);
100 schedule_work(&send_queue
);
103 static int checksum_udp(struct sk_buff
*skb
, struct udphdr
*uh
,
104 unsigned short ulen
, u32 saddr
, u32 daddr
)
108 if (uh
->check
== 0 || skb
->ip_summed
== CHECKSUM_UNNECESSARY
)
111 psum
= csum_tcpudp_nofold(saddr
, daddr
, ulen
, IPPROTO_UDP
, 0);
113 if (skb
->ip_summed
== CHECKSUM_COMPLETE
&&
114 !(u16
)csum_fold(csum_add(psum
, skb
->csum
)))
119 return __skb_checksum_complete(skb
);
123 * Check whether delayed processing was scheduled for our NIC. If so,
124 * we attempt to grab the poll lock and use ->poll() to pump the card.
125 * If this fails, either we've recursed in ->poll() or it's already
126 * running on another CPU.
128 * Note: we don't mask interrupts with this lock because we're using
129 * trylock here and interrupts are already disabled in the softirq
130 * case. Further, we test the poll_owner to avoid recursion on UP
131 * systems where the lock doesn't exist.
133 * In cases where there is bi-directional communications, reading only
134 * one message at a time can lead to packets being dropped by the
135 * network adapter, forcing superfluous retries and possibly timeouts.
136 * Thus, we set our budget to greater than 1.
138 static void poll_napi(struct netpoll
*np
)
140 struct netpoll_info
*npinfo
= np
->dev
->npinfo
;
143 if (test_bit(__LINK_STATE_RX_SCHED
, &np
->dev
->state
) &&
144 npinfo
->poll_owner
!= smp_processor_id() &&
145 spin_trylock(&npinfo
->poll_lock
)) {
146 npinfo
->rx_flags
|= NETPOLL_RX_DROP
;
147 atomic_inc(&trapped
);
149 np
->dev
->poll(np
->dev
, &budget
);
151 atomic_dec(&trapped
);
152 npinfo
->rx_flags
&= ~NETPOLL_RX_DROP
;
153 spin_unlock(&npinfo
->poll_lock
);
157 static void service_arp_queue(struct netpoll_info
*npi
)
164 skb
= skb_dequeue(&npi
->arp_tx
);
166 while (skb
!= NULL
) {
168 skb
= skb_dequeue(&npi
->arp_tx
);
173 void netpoll_poll(struct netpoll
*np
)
175 if(!np
->dev
|| !netif_running(np
->dev
) || !np
->dev
->poll_controller
)
178 /* Process pending work on NIC */
179 np
->dev
->poll_controller(np
->dev
);
183 service_arp_queue(np
->dev
->npinfo
);
185 zap_completion_queue();
188 static void refill_skbs(void)
193 spin_lock_irqsave(&skb_list_lock
, flags
);
194 while (nr_skbs
< MAX_SKBS
) {
195 skb
= alloc_skb(MAX_SKB_SIZE
, GFP_ATOMIC
);
203 spin_unlock_irqrestore(&skb_list_lock
, flags
);
206 static void zap_completion_queue(void)
209 struct softnet_data
*sd
= &get_cpu_var(softnet_data
);
211 if (sd
->completion_queue
) {
212 struct sk_buff
*clist
;
214 local_irq_save(flags
);
215 clist
= sd
->completion_queue
;
216 sd
->completion_queue
= NULL
;
217 local_irq_restore(flags
);
219 while (clist
!= NULL
) {
220 struct sk_buff
*skb
= clist
;
223 dev_kfree_skb_any(skb
); /* put this one back */
229 put_cpu_var(softnet_data
);
232 static struct sk_buff
* find_skb(struct netpoll
*np
, int len
, int reserve
)
234 int once
= 1, count
= 0;
236 struct sk_buff
*skb
= NULL
;
238 zap_completion_queue();
240 if (nr_skbs
< MAX_SKBS
)
243 skb
= alloc_skb(len
, GFP_ATOMIC
);
246 spin_lock_irqsave(&skb_list_lock
, flags
);
253 spin_unlock_irqrestore(&skb_list_lock
, flags
);
258 if (once
&& (count
== 1000000)) {
259 printk("out of netpoll skbs!\n");
266 atomic_set(&skb
->users
, 1);
267 skb_reserve(skb
, reserve
);
271 static void netpoll_send_skb(struct netpoll
*np
, struct sk_buff
*skb
)
274 struct netpoll_info
*npinfo
;
276 if (!np
|| !np
->dev
|| !netif_running(np
->dev
)) {
281 npinfo
= np
->dev
->npinfo
;
283 /* avoid recursion */
284 if (npinfo
->poll_owner
== smp_processor_id() ||
285 np
->dev
->xmit_lock_owner
== smp_processor_id()) {
295 netif_tx_lock(np
->dev
);
298 * network drivers do not expect to be called if the queue is
301 status
= NETDEV_TX_BUSY
;
302 if (!netif_queue_stopped(np
->dev
))
303 status
= np
->dev
->hard_start_xmit(skb
, np
->dev
);
305 netif_tx_unlock(np
->dev
);
309 npinfo
->tries
= MAX_RETRIES
; /* reset */
316 } while (npinfo
->tries
> 0);
319 void netpoll_send_udp(struct netpoll
*np
, const char *msg
, int len
)
321 int total_len
, eth_len
, ip_len
, udp_len
;
327 udp_len
= len
+ sizeof(*udph
);
328 ip_len
= eth_len
= udp_len
+ sizeof(*iph
);
329 total_len
= eth_len
+ ETH_HLEN
+ NET_IP_ALIGN
;
331 skb
= find_skb(np
, total_len
, total_len
- len
);
335 memcpy(skb
->data
, msg
, len
);
338 skb
->h
.uh
= udph
= (struct udphdr
*) skb_push(skb
, sizeof(*udph
));
339 udph
->source
= htons(np
->local_port
);
340 udph
->dest
= htons(np
->remote_port
);
341 udph
->len
= htons(udp_len
);
343 udph
->check
= csum_tcpudp_magic(htonl(np
->local_ip
),
344 htonl(np
->remote_ip
),
345 udp_len
, IPPROTO_UDP
,
346 csum_partial((unsigned char *)udph
, udp_len
, 0));
347 if (udph
->check
== 0)
350 skb
->nh
.iph
= iph
= (struct iphdr
*)skb_push(skb
, sizeof(*iph
));
352 /* iph->version = 4; iph->ihl = 5; */
353 put_unaligned(0x45, (unsigned char *)iph
);
355 put_unaligned(htons(ip_len
), &(iph
->tot_len
));
359 iph
->protocol
= IPPROTO_UDP
;
361 put_unaligned(htonl(np
->local_ip
), &(iph
->saddr
));
362 put_unaligned(htonl(np
->remote_ip
), &(iph
->daddr
));
363 iph
->check
= ip_fast_csum((unsigned char *)iph
, iph
->ihl
);
365 eth
= (struct ethhdr
*) skb_push(skb
, ETH_HLEN
);
366 skb
->mac
.raw
= skb
->data
;
367 skb
->protocol
= eth
->h_proto
= htons(ETH_P_IP
);
368 memcpy(eth
->h_source
, np
->local_mac
, 6);
369 memcpy(eth
->h_dest
, np
->remote_mac
, 6);
373 netpoll_send_skb(np
, skb
);
376 static void arp_reply(struct sk_buff
*skb
)
378 struct netpoll_info
*npinfo
= skb
->dev
->npinfo
;
380 unsigned char *arp_ptr
;
381 int size
, type
= ARPOP_REPLY
, ptype
= ETH_P_ARP
;
383 struct sk_buff
*send_skb
;
384 struct netpoll
*np
= NULL
;
386 if (npinfo
->rx_np
&& npinfo
->rx_np
->dev
== skb
->dev
)
391 /* No arp on this interface */
392 if (skb
->dev
->flags
& IFF_NOARP
)
395 if (!pskb_may_pull(skb
, (sizeof(struct arphdr
) +
396 (2 * skb
->dev
->addr_len
) +
400 skb
->h
.raw
= skb
->nh
.raw
= skb
->data
;
403 if ((arp
->ar_hrd
!= htons(ARPHRD_ETHER
) &&
404 arp
->ar_hrd
!= htons(ARPHRD_IEEE802
)) ||
405 arp
->ar_pro
!= htons(ETH_P_IP
) ||
406 arp
->ar_op
!= htons(ARPOP_REQUEST
))
409 arp_ptr
= (unsigned char *)(arp
+1) + skb
->dev
->addr_len
;
410 memcpy(&sip
, arp_ptr
, 4);
411 arp_ptr
+= 4 + skb
->dev
->addr_len
;
412 memcpy(&tip
, arp_ptr
, 4);
414 /* Should we ignore arp? */
415 if (tip
!= htonl(np
->local_ip
) || LOOPBACK(tip
) || MULTICAST(tip
))
418 size
= sizeof(struct arphdr
) + 2 * (skb
->dev
->addr_len
+ 4);
419 send_skb
= find_skb(np
, size
+ LL_RESERVED_SPACE(np
->dev
),
420 LL_RESERVED_SPACE(np
->dev
));
425 send_skb
->nh
.raw
= send_skb
->data
;
426 arp
= (struct arphdr
*) skb_put(send_skb
, size
);
427 send_skb
->dev
= skb
->dev
;
428 send_skb
->protocol
= htons(ETH_P_ARP
);
430 /* Fill the device header for the ARP frame */
432 if (np
->dev
->hard_header
&&
433 np
->dev
->hard_header(send_skb
, skb
->dev
, ptype
,
434 np
->remote_mac
, np
->local_mac
,
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
, np
->remote_mac
, 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
;
476 if (skb
->dev
->type
!= ARPHRD_ETHER
)
479 /* check if netpoll clients need ARP */
480 if (skb
->protocol
== __constant_htons(ETH_P_ARP
) &&
481 atomic_read(&trapped
)) {
482 skb_queue_tail(&npi
->arp_tx
, skb
);
486 proto
= ntohs(eth_hdr(skb
)->h_proto
);
487 if (proto
!= ETH_P_IP
)
489 if (skb
->pkt_type
== PACKET_OTHERHOST
)
494 iph
= (struct iphdr
*)skb
->data
;
495 if (!pskb_may_pull(skb
, sizeof(struct iphdr
)))
497 if (iph
->ihl
< 5 || iph
->version
!= 4)
499 if (!pskb_may_pull(skb
, iph
->ihl
*4))
501 if (ip_fast_csum((u8
*)iph
, iph
->ihl
) != 0)
504 len
= ntohs(iph
->tot_len
);
505 if (skb
->len
< len
|| len
< iph
->ihl
*4)
508 if (iph
->protocol
!= IPPROTO_UDP
)
512 uh
= (struct udphdr
*)(((char *)iph
) + iph
->ihl
*4);
513 ulen
= ntohs(uh
->len
);
517 if (checksum_udp(skb
, uh
, ulen
, iph
->saddr
, iph
->daddr
))
519 if (np
->local_ip
&& np
->local_ip
!= ntohl(iph
->daddr
))
521 if (np
->remote_ip
&& np
->remote_ip
!= ntohl(iph
->saddr
))
523 if (np
->local_port
&& np
->local_port
!= ntohs(uh
->dest
))
526 np
->rx_hook(np
, ntohs(uh
->source
),
528 ulen
- sizeof(struct udphdr
));
534 if (atomic_read(&trapped
)) {
542 int netpoll_parse_options(struct netpoll
*np
, char *opt
)
544 char *cur
=opt
, *delim
;
547 if ((delim
= strchr(cur
, '@')) == NULL
)
550 np
->local_port
=simple_strtol(cur
, NULL
, 10);
554 printk(KERN_INFO
"%s: local port %d\n", np
->name
, np
->local_port
);
557 if ((delim
= strchr(cur
, '/')) == NULL
)
560 np
->local_ip
=ntohl(in_aton(cur
));
563 printk(KERN_INFO
"%s: local IP %d.%d.%d.%d\n",
564 np
->name
, HIPQUAD(np
->local_ip
));
569 /* parse out dev name */
570 if ((delim
= strchr(cur
, ',')) == NULL
)
573 strlcpy(np
->dev_name
, cur
, sizeof(np
->dev_name
));
578 printk(KERN_INFO
"%s: interface %s\n", np
->name
, np
->dev_name
);
582 if ((delim
= strchr(cur
, '@')) == NULL
)
585 np
->remote_port
=simple_strtol(cur
, NULL
, 10);
589 printk(KERN_INFO
"%s: remote port %d\n", np
->name
, np
->remote_port
);
592 if ((delim
= strchr(cur
, '/')) == NULL
)
595 np
->remote_ip
=ntohl(in_aton(cur
));
598 printk(KERN_INFO
"%s: remote IP %d.%d.%d.%d\n",
599 np
->name
, HIPQUAD(np
->remote_ip
));
604 if ((delim
= strchr(cur
, ':')) == NULL
)
607 np
->remote_mac
[0]=simple_strtol(cur
, NULL
, 16);
609 if ((delim
= strchr(cur
, ':')) == NULL
)
612 np
->remote_mac
[1]=simple_strtol(cur
, NULL
, 16);
614 if ((delim
= strchr(cur
, ':')) == NULL
)
617 np
->remote_mac
[2]=simple_strtol(cur
, NULL
, 16);
619 if ((delim
= strchr(cur
, ':')) == NULL
)
622 np
->remote_mac
[3]=simple_strtol(cur
, NULL
, 16);
624 if ((delim
= strchr(cur
, ':')) == NULL
)
627 np
->remote_mac
[4]=simple_strtol(cur
, NULL
, 16);
629 np
->remote_mac
[5]=simple_strtol(cur
, NULL
, 16);
632 printk(KERN_INFO
"%s: remote ethernet address "
633 "%02x:%02x:%02x:%02x:%02x:%02x\n",
645 printk(KERN_INFO
"%s: couldn't parse config at %s!\n",
650 int netpoll_setup(struct netpoll
*np
)
652 struct net_device
*ndev
= NULL
;
653 struct in_device
*in_dev
;
654 struct netpoll_info
*npinfo
;
658 ndev
= dev_get_by_name(np
->dev_name
);
660 printk(KERN_ERR
"%s: %s doesn't exist, aborting.\n",
661 np
->name
, np
->dev_name
);
667 npinfo
= kmalloc(sizeof(*npinfo
), GFP_KERNEL
);
671 npinfo
->rx_flags
= 0;
672 npinfo
->rx_np
= NULL
;
673 spin_lock_init(&npinfo
->poll_lock
);
674 npinfo
->poll_owner
= -1;
675 npinfo
->tries
= MAX_RETRIES
;
676 spin_lock_init(&npinfo
->rx_lock
);
677 skb_queue_head_init(&npinfo
->arp_tx
);
679 npinfo
= ndev
->npinfo
;
681 if (!ndev
->poll_controller
) {
682 printk(KERN_ERR
"%s: %s doesn't support polling, aborting.\n",
683 np
->name
, np
->dev_name
);
687 if (!netif_running(ndev
)) {
688 unsigned long atmost
, atleast
;
690 printk(KERN_INFO
"%s: device %s not up yet, forcing it\n",
691 np
->name
, np
->dev_name
);
694 if (dev_change_flags(ndev
, ndev
->flags
| IFF_UP
) < 0) {
695 printk(KERN_ERR
"%s: failed to open %s\n",
696 np
->name
, np
->dev_name
);
702 atleast
= jiffies
+ HZ
/10;
703 atmost
= jiffies
+ 4*HZ
;
704 while (!netif_carrier_ok(ndev
)) {
705 if (time_after(jiffies
, atmost
)) {
707 "%s: timeout waiting for carrier\n",
714 /* If carrier appears to come up instantly, we don't
715 * trust it and pause so that we don't pump all our
716 * queued console messages into the bitbucket.
719 if (time_before(jiffies
, atleast
)) {
720 printk(KERN_NOTICE
"%s: carrier detect appears"
721 " untrustworthy, waiting 4 seconds\n",
727 if (is_zero_ether_addr(np
->local_mac
) && ndev
->dev_addr
)
728 memcpy(np
->local_mac
, ndev
->dev_addr
, 6);
732 in_dev
= __in_dev_get_rcu(ndev
);
734 if (!in_dev
|| !in_dev
->ifa_list
) {
736 printk(KERN_ERR
"%s: no IP address for %s, aborting\n",
737 np
->name
, np
->dev_name
);
741 np
->local_ip
= ntohl(in_dev
->ifa_list
->ifa_local
);
743 printk(KERN_INFO
"%s: local IP %d.%d.%d.%d\n",
744 np
->name
, HIPQUAD(np
->local_ip
));
748 spin_lock_irqsave(&npinfo
->rx_lock
, flags
);
749 npinfo
->rx_flags
|= NETPOLL_RX_ENABLED
;
751 spin_unlock_irqrestore(&npinfo
->rx_lock
, flags
);
754 /* fill up the skb queue */
757 /* last thing to do is link it to the net device structure */
758 ndev
->npinfo
= npinfo
;
760 /* avoid racing with NAPI reading npinfo */
773 void netpoll_cleanup(struct netpoll
*np
)
775 struct netpoll_info
*npinfo
;
779 npinfo
= np
->dev
->npinfo
;
780 if (npinfo
&& npinfo
->rx_np
== np
) {
781 spin_lock_irqsave(&npinfo
->rx_lock
, flags
);
782 npinfo
->rx_np
= NULL
;
783 npinfo
->rx_flags
&= ~NETPOLL_RX_ENABLED
;
784 spin_unlock_irqrestore(&npinfo
->rx_lock
, flags
);
792 int netpoll_trap(void)
794 return atomic_read(&trapped
);
797 void netpoll_set_trap(int trap
)
800 atomic_inc(&trapped
);
802 atomic_dec(&trapped
);
805 EXPORT_SYMBOL(netpoll_set_trap
);
806 EXPORT_SYMBOL(netpoll_trap
);
807 EXPORT_SYMBOL(netpoll_parse_options
);
808 EXPORT_SYMBOL(netpoll_setup
);
809 EXPORT_SYMBOL(netpoll_cleanup
);
810 EXPORT_SYMBOL(netpoll_send_udp
);
811 EXPORT_SYMBOL(netpoll_poll
);
812 EXPORT_SYMBOL(netpoll_queue
);