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>
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 struct sk_buff_head skb_pool
;
39 static atomic_t trapped
;
41 #define USEC_PER_POLL 50
42 #define NETPOLL_RX_ENABLED 1
43 #define NETPOLL_RX_DROP 2
45 #define MAX_SKB_SIZE \
46 (MAX_UDP_CHUNK + sizeof(struct udphdr) + \
47 sizeof(struct iphdr) + sizeof(struct ethhdr))
49 static void zap_completion_queue(void);
50 static void arp_reply(struct sk_buff
*skb
);
52 static void queue_process(struct work_struct
*work
)
54 struct netpoll_info
*npinfo
=
55 container_of(work
, struct netpoll_info
, tx_work
.work
);
59 while ((skb
= skb_dequeue(&npinfo
->txq
))) {
60 struct net_device
*dev
= skb
->dev
;
62 if (!netif_device_present(dev
) || !netif_running(dev
)) {
67 local_irq_save(flags
);
69 if ((netif_queue_stopped(dev
) ||
70 netif_subqueue_stopped(dev
, skb
->queue_mapping
)) ||
71 dev
->hard_start_xmit(skb
, dev
) != NETDEV_TX_OK
) {
72 skb_queue_head(&npinfo
->txq
, skb
);
74 local_irq_restore(flags
);
76 schedule_delayed_work(&npinfo
->tx_work
, HZ
/10);
80 local_irq_restore(flags
);
84 static __sum16
checksum_udp(struct sk_buff
*skb
, struct udphdr
*uh
,
85 unsigned short ulen
, __be32 saddr
, __be32 daddr
)
89 if (uh
->check
== 0 || skb_csum_unnecessary(skb
))
92 psum
= csum_tcpudp_nofold(saddr
, daddr
, ulen
, IPPROTO_UDP
, 0);
94 if (skb
->ip_summed
== CHECKSUM_COMPLETE
&&
95 !csum_fold(csum_add(psum
, skb
->csum
)))
100 return __skb_checksum_complete(skb
);
104 * Check whether delayed processing was scheduled for our NIC. If so,
105 * we attempt to grab the poll lock and use ->poll() to pump the card.
106 * If this fails, either we've recursed in ->poll() or it's already
107 * running on another CPU.
109 * Note: we don't mask interrupts with this lock because we're using
110 * trylock here and interrupts are already disabled in the softirq
111 * case. Further, we test the poll_owner to avoid recursion on UP
112 * systems where the lock doesn't exist.
114 * In cases where there is bi-directional communications, reading only
115 * one message at a time can lead to packets being dropped by the
116 * network adapter, forcing superfluous retries and possibly timeouts.
117 * Thus, we set our budget to greater than 1.
119 static void poll_napi(struct netpoll
*np
)
121 struct netpoll_info
*npinfo
= np
->dev
->npinfo
;
124 if (test_bit(__LINK_STATE_RX_SCHED
, &np
->dev
->state
) &&
125 npinfo
->poll_owner
!= smp_processor_id() &&
126 spin_trylock(&npinfo
->poll_lock
)) {
127 /* When calling dev->poll from poll_napi, we may end up in
128 * netif_rx_complete. However, only the CPU to which the
129 * device was queued is allowed to remove it from poll_list.
130 * Setting POLL_LIST_FROZEN tells netif_rx_complete
131 * to leave the NAPI state alone.
133 set_bit(__LINK_STATE_POLL_LIST_FROZEN
, &np
->dev
->state
);
134 npinfo
->rx_flags
|= NETPOLL_RX_DROP
;
135 atomic_inc(&trapped
);
137 np
->dev
->poll(np
->dev
, &budget
);
139 atomic_dec(&trapped
);
140 npinfo
->rx_flags
&= ~NETPOLL_RX_DROP
;
141 clear_bit(__LINK_STATE_POLL_LIST_FROZEN
, &np
->dev
->state
);
142 spin_unlock(&npinfo
->poll_lock
);
146 static void service_arp_queue(struct netpoll_info
*npi
)
153 skb
= skb_dequeue(&npi
->arp_tx
);
155 while (skb
!= NULL
) {
157 skb
= skb_dequeue(&npi
->arp_tx
);
161 void netpoll_poll(struct netpoll
*np
)
163 if (!np
->dev
|| !netif_running(np
->dev
) || !np
->dev
->poll_controller
)
166 /* Process pending work on NIC */
167 np
->dev
->poll_controller(np
->dev
);
171 service_arp_queue(np
->dev
->npinfo
);
173 zap_completion_queue();
176 static void refill_skbs(void)
181 spin_lock_irqsave(&skb_pool
.lock
, flags
);
182 while (skb_pool
.qlen
< MAX_SKBS
) {
183 skb
= alloc_skb(MAX_SKB_SIZE
, GFP_ATOMIC
);
187 __skb_queue_tail(&skb_pool
, skb
);
189 spin_unlock_irqrestore(&skb_pool
.lock
, flags
);
192 static void zap_completion_queue(void)
195 struct softnet_data
*sd
= &get_cpu_var(softnet_data
);
197 if (sd
->completion_queue
) {
198 struct sk_buff
*clist
;
200 local_irq_save(flags
);
201 clist
= sd
->completion_queue
;
202 sd
->completion_queue
= NULL
;
203 local_irq_restore(flags
);
205 while (clist
!= NULL
) {
206 struct sk_buff
*skb
= clist
;
209 dev_kfree_skb_any(skb
); /* put this one back */
215 put_cpu_var(softnet_data
);
218 static struct sk_buff
*find_skb(struct netpoll
*np
, int len
, int reserve
)
223 zap_completion_queue();
227 skb
= alloc_skb(len
, GFP_ATOMIC
);
229 skb
= skb_dequeue(&skb_pool
);
239 atomic_set(&skb
->users
, 1);
240 skb_reserve(skb
, reserve
);
244 static void netpoll_send_skb(struct netpoll
*np
, struct sk_buff
*skb
)
246 int status
= NETDEV_TX_BUSY
;
248 struct net_device
*dev
= np
->dev
;
249 struct netpoll_info
*npinfo
= np
->dev
->npinfo
;
251 if (!npinfo
|| !netif_running(dev
) || !netif_device_present(dev
)) {
256 /* don't get messages out of order, and no recursion */
257 if (skb_queue_len(&npinfo
->txq
) == 0 &&
258 npinfo
->poll_owner
!= smp_processor_id()) {
261 local_irq_save(flags
);
262 /* try until next clock tick */
263 for (tries
= jiffies_to_usecs(1)/USEC_PER_POLL
;
264 tries
> 0; --tries
) {
265 if (netif_tx_trylock(dev
)) {
266 if (!netif_queue_stopped(dev
) &&
267 !netif_subqueue_stopped(dev
, skb
->queue_mapping
))
268 status
= dev
->hard_start_xmit(skb
, dev
);
269 netif_tx_unlock(dev
);
271 if (status
== NETDEV_TX_OK
)
276 /* tickle device maybe there is some cleanup */
279 udelay(USEC_PER_POLL
);
281 local_irq_restore(flags
);
284 if (status
!= NETDEV_TX_OK
) {
285 skb_queue_tail(&npinfo
->txq
, skb
);
286 schedule_delayed_work(&npinfo
->tx_work
,0);
290 void netpoll_send_udp(struct netpoll
*np
, const char *msg
, int len
)
292 int total_len
, eth_len
, ip_len
, udp_len
;
298 udp_len
= len
+ sizeof(*udph
);
299 ip_len
= eth_len
= udp_len
+ sizeof(*iph
);
300 total_len
= eth_len
+ ETH_HLEN
+ NET_IP_ALIGN
;
302 skb
= find_skb(np
, total_len
, total_len
- len
);
306 skb_copy_to_linear_data(skb
, msg
, len
);
309 skb_push(skb
, sizeof(*udph
));
310 skb_reset_transport_header(skb
);
312 udph
->source
= htons(np
->local_port
);
313 udph
->dest
= htons(np
->remote_port
);
314 udph
->len
= htons(udp_len
);
316 udph
->check
= csum_tcpudp_magic(htonl(np
->local_ip
),
317 htonl(np
->remote_ip
),
318 udp_len
, IPPROTO_UDP
,
319 csum_partial((unsigned char *)udph
, udp_len
, 0));
320 if (udph
->check
== 0)
321 udph
->check
= CSUM_MANGLED_0
;
323 skb_push(skb
, sizeof(*iph
));
324 skb_reset_network_header(skb
);
327 /* iph->version = 4; iph->ihl = 5; */
328 put_unaligned(0x45, (unsigned char *)iph
);
330 put_unaligned(htons(ip_len
), &(iph
->tot_len
));
334 iph
->protocol
= IPPROTO_UDP
;
336 put_unaligned(htonl(np
->local_ip
), &(iph
->saddr
));
337 put_unaligned(htonl(np
->remote_ip
), &(iph
->daddr
));
338 iph
->check
= ip_fast_csum((unsigned char *)iph
, iph
->ihl
);
340 eth
= (struct ethhdr
*) skb_push(skb
, ETH_HLEN
);
341 skb_reset_mac_header(skb
);
342 skb
->protocol
= eth
->h_proto
= htons(ETH_P_IP
);
343 memcpy(eth
->h_source
, np
->local_mac
, 6);
344 memcpy(eth
->h_dest
, np
->remote_mac
, 6);
348 netpoll_send_skb(np
, skb
);
351 static void arp_reply(struct sk_buff
*skb
)
353 struct netpoll_info
*npinfo
= skb
->dev
->npinfo
;
355 unsigned char *arp_ptr
;
356 int size
, type
= ARPOP_REPLY
, ptype
= ETH_P_ARP
;
359 struct sk_buff
*send_skb
;
360 struct netpoll
*np
= NULL
;
362 if (npinfo
->rx_np
&& npinfo
->rx_np
->dev
== skb
->dev
)
367 /* No arp on this interface */
368 if (skb
->dev
->flags
& IFF_NOARP
)
371 if (!pskb_may_pull(skb
, (sizeof(struct arphdr
) +
372 (2 * skb
->dev
->addr_len
) +
376 skb_reset_network_header(skb
);
377 skb_reset_transport_header(skb
);
380 if ((arp
->ar_hrd
!= htons(ARPHRD_ETHER
) &&
381 arp
->ar_hrd
!= htons(ARPHRD_IEEE802
)) ||
382 arp
->ar_pro
!= htons(ETH_P_IP
) ||
383 arp
->ar_op
!= htons(ARPOP_REQUEST
))
386 arp_ptr
= (unsigned char *)(arp
+1);
387 /* save the location of the src hw addr */
389 arp_ptr
+= skb
->dev
->addr_len
;
390 memcpy(&sip
, arp_ptr
, 4);
392 /* if we actually cared about dst hw addr, it would get copied here */
393 arp_ptr
+= skb
->dev
->addr_len
;
394 memcpy(&tip
, arp_ptr
, 4);
396 /* Should we ignore arp? */
397 if (tip
!= htonl(np
->local_ip
) || LOOPBACK(tip
) || MULTICAST(tip
))
400 size
= sizeof(struct arphdr
) + 2 * (skb
->dev
->addr_len
+ 4);
401 send_skb
= find_skb(np
, size
+ LL_RESERVED_SPACE(np
->dev
),
402 LL_RESERVED_SPACE(np
->dev
));
407 skb_reset_network_header(send_skb
);
408 arp
= (struct arphdr
*) skb_put(send_skb
, size
);
409 send_skb
->dev
= skb
->dev
;
410 send_skb
->protocol
= htons(ETH_P_ARP
);
412 /* Fill the device header for the ARP frame */
414 if (np
->dev
->hard_header
&&
415 np
->dev
->hard_header(send_skb
, skb
->dev
, ptype
,
417 send_skb
->len
) < 0) {
423 * Fill out the arp protocol part.
425 * we only support ethernet device type,
426 * which (according to RFC 1390) should always equal 1 (Ethernet).
429 arp
->ar_hrd
= htons(np
->dev
->type
);
430 arp
->ar_pro
= htons(ETH_P_IP
);
431 arp
->ar_hln
= np
->dev
->addr_len
;
433 arp
->ar_op
= htons(type
);
435 arp_ptr
=(unsigned char *)(arp
+ 1);
436 memcpy(arp_ptr
, np
->dev
->dev_addr
, np
->dev
->addr_len
);
437 arp_ptr
+= np
->dev
->addr_len
;
438 memcpy(arp_ptr
, &tip
, 4);
440 memcpy(arp_ptr
, sha
, np
->dev
->addr_len
);
441 arp_ptr
+= np
->dev
->addr_len
;
442 memcpy(arp_ptr
, &sip
, 4);
444 netpoll_send_skb(np
, send_skb
);
447 int __netpoll_rx(struct sk_buff
*skb
)
449 int proto
, len
, ulen
;
452 struct netpoll_info
*npi
= skb
->dev
->npinfo
;
453 struct netpoll
*np
= npi
->rx_np
;
457 if (skb
->dev
->type
!= ARPHRD_ETHER
)
460 /* check if netpoll clients need ARP */
461 if (skb
->protocol
== htons(ETH_P_ARP
) &&
462 atomic_read(&trapped
)) {
463 skb_queue_tail(&npi
->arp_tx
, skb
);
467 proto
= ntohs(eth_hdr(skb
)->h_proto
);
468 if (proto
!= ETH_P_IP
)
470 if (skb
->pkt_type
== PACKET_OTHERHOST
)
475 iph
= (struct iphdr
*)skb
->data
;
476 if (!pskb_may_pull(skb
, sizeof(struct iphdr
)))
478 if (iph
->ihl
< 5 || iph
->version
!= 4)
480 if (!pskb_may_pull(skb
, iph
->ihl
*4))
482 if (ip_fast_csum((u8
*)iph
, iph
->ihl
) != 0)
485 len
= ntohs(iph
->tot_len
);
486 if (skb
->len
< len
|| len
< iph
->ihl
*4)
490 * Our transport medium may have padded the buffer out.
491 * Now We trim to the true length of the frame.
493 if (pskb_trim_rcsum(skb
, len
))
496 if (iph
->protocol
!= IPPROTO_UDP
)
500 uh
= (struct udphdr
*)(((char *)iph
) + iph
->ihl
*4);
501 ulen
= ntohs(uh
->len
);
505 if (checksum_udp(skb
, uh
, ulen
, iph
->saddr
, iph
->daddr
))
507 if (np
->local_ip
&& np
->local_ip
!= ntohl(iph
->daddr
))
509 if (np
->remote_ip
&& np
->remote_ip
!= ntohl(iph
->saddr
))
511 if (np
->local_port
&& np
->local_port
!= ntohs(uh
->dest
))
514 np
->rx_hook(np
, ntohs(uh
->source
),
516 ulen
- sizeof(struct udphdr
));
522 if (atomic_read(&trapped
)) {
530 int netpoll_parse_options(struct netpoll
*np
, char *opt
)
532 char *cur
=opt
, *delim
;
535 if ((delim
= strchr(cur
, '@')) == NULL
)
538 np
->local_port
= simple_strtol(cur
, NULL
, 10);
542 printk(KERN_INFO
"%s: local port %d\n", np
->name
, np
->local_port
);
545 if ((delim
= strchr(cur
, '/')) == NULL
)
548 np
->local_ip
= ntohl(in_aton(cur
));
551 printk(KERN_INFO
"%s: local IP %d.%d.%d.%d\n",
552 np
->name
, HIPQUAD(np
->local_ip
));
557 /* parse out dev name */
558 if ((delim
= strchr(cur
, ',')) == NULL
)
561 strlcpy(np
->dev_name
, cur
, sizeof(np
->dev_name
));
566 printk(KERN_INFO
"%s: interface %s\n", np
->name
, np
->dev_name
);
570 if ((delim
= strchr(cur
, '@')) == NULL
)
573 np
->remote_port
= simple_strtol(cur
, NULL
, 10);
577 printk(KERN_INFO
"%s: remote port %d\n", np
->name
, np
->remote_port
);
580 if ((delim
= strchr(cur
, '/')) == NULL
)
583 np
->remote_ip
= ntohl(in_aton(cur
));
586 printk(KERN_INFO
"%s: remote IP %d.%d.%d.%d\n",
587 np
->name
, HIPQUAD(np
->remote_ip
));
591 if ((delim
= strchr(cur
, ':')) == NULL
)
594 np
->remote_mac
[0] = simple_strtol(cur
, NULL
, 16);
596 if ((delim
= strchr(cur
, ':')) == NULL
)
599 np
->remote_mac
[1] = simple_strtol(cur
, NULL
, 16);
601 if ((delim
= strchr(cur
, ':')) == NULL
)
604 np
->remote_mac
[2] = simple_strtol(cur
, NULL
, 16);
606 if ((delim
= strchr(cur
, ':')) == NULL
)
609 np
->remote_mac
[3] = simple_strtol(cur
, NULL
, 16);
611 if ((delim
= strchr(cur
, ':')) == NULL
)
614 np
->remote_mac
[4] = simple_strtol(cur
, NULL
, 16);
616 np
->remote_mac
[5] = simple_strtol(cur
, NULL
, 16);
619 printk(KERN_INFO
"%s: remote ethernet address "
620 "%02x:%02x:%02x:%02x:%02x:%02x\n",
632 printk(KERN_INFO
"%s: couldn't parse config at %s!\n",
637 int netpoll_setup(struct netpoll
*np
)
639 struct net_device
*ndev
= NULL
;
640 struct in_device
*in_dev
;
641 struct netpoll_info
*npinfo
;
646 ndev
= dev_get_by_name(np
->dev_name
);
648 printk(KERN_ERR
"%s: %s doesn't exist, aborting.\n",
649 np
->name
, np
->dev_name
);
655 npinfo
= kmalloc(sizeof(*npinfo
), GFP_KERNEL
);
661 npinfo
->rx_flags
= 0;
662 npinfo
->rx_np
= NULL
;
663 spin_lock_init(&npinfo
->poll_lock
);
664 npinfo
->poll_owner
= -1;
666 spin_lock_init(&npinfo
->rx_lock
);
667 skb_queue_head_init(&npinfo
->arp_tx
);
668 skb_queue_head_init(&npinfo
->txq
);
669 INIT_DELAYED_WORK(&npinfo
->tx_work
, queue_process
);
671 atomic_set(&npinfo
->refcnt
, 1);
673 npinfo
= ndev
->npinfo
;
674 atomic_inc(&npinfo
->refcnt
);
677 if (!ndev
->poll_controller
) {
678 printk(KERN_ERR
"%s: %s doesn't support polling, aborting.\n",
679 np
->name
, np
->dev_name
);
684 if (!netif_running(ndev
)) {
685 unsigned long atmost
, atleast
;
687 printk(KERN_INFO
"%s: device %s not up yet, forcing it\n",
688 np
->name
, np
->dev_name
);
691 err
= dev_open(ndev
);
695 printk(KERN_ERR
"%s: failed to open %s\n",
696 np
->name
, ndev
->name
);
700 atleast
= jiffies
+ HZ
/10;
701 atmost
= jiffies
+ 4*HZ
;
702 while (!netif_carrier_ok(ndev
)) {
703 if (time_after(jiffies
, atmost
)) {
705 "%s: timeout waiting for carrier\n",
712 /* If carrier appears to come up instantly, we don't
713 * trust it and pause so that we don't pump all our
714 * queued console messages into the bitbucket.
717 if (time_before(jiffies
, atleast
)) {
718 printk(KERN_NOTICE
"%s: carrier detect appears"
719 " untrustworthy, waiting 4 seconds\n",
725 if (is_zero_ether_addr(np
->local_mac
) && ndev
->dev_addr
)
726 memcpy(np
->local_mac
, ndev
->dev_addr
, 6);
730 in_dev
= __in_dev_get_rcu(ndev
);
732 if (!in_dev
|| !in_dev
->ifa_list
) {
734 printk(KERN_ERR
"%s: no IP address for %s, aborting\n",
735 np
->name
, np
->dev_name
);
740 np
->local_ip
= ntohl(in_dev
->ifa_list
->ifa_local
);
742 printk(KERN_INFO
"%s: local IP %d.%d.%d.%d\n",
743 np
->name
, HIPQUAD(np
->local_ip
));
747 spin_lock_irqsave(&npinfo
->rx_lock
, flags
);
748 npinfo
->rx_flags
|= NETPOLL_RX_ENABLED
;
750 spin_unlock_irqrestore(&npinfo
->rx_lock
, flags
);
753 /* fill up the skb queue */
756 /* last thing to do is link it to the net device structure */
757 ndev
->npinfo
= npinfo
;
759 /* avoid racing with NAPI reading npinfo */
772 static int __init
netpoll_init(void)
774 skb_queue_head_init(&skb_pool
);
777 core_initcall(netpoll_init
);
779 void netpoll_cleanup(struct netpoll
*np
)
781 struct netpoll_info
*npinfo
;
785 npinfo
= np
->dev
->npinfo
;
787 if (npinfo
->rx_np
== np
) {
788 spin_lock_irqsave(&npinfo
->rx_lock
, flags
);
789 npinfo
->rx_np
= NULL
;
790 npinfo
->rx_flags
&= ~NETPOLL_RX_ENABLED
;
791 spin_unlock_irqrestore(&npinfo
->rx_lock
, flags
);
794 if (atomic_dec_and_test(&npinfo
->refcnt
)) {
795 skb_queue_purge(&npinfo
->arp_tx
);
796 skb_queue_purge(&npinfo
->txq
);
797 cancel_rearming_delayed_work(&npinfo
->tx_work
);
799 /* clean after last, unfinished work */
800 if (!skb_queue_empty(&npinfo
->txq
)) {
802 skb
= __skb_dequeue(&npinfo
->txq
);
806 np
->dev
->npinfo
= NULL
;
816 int netpoll_trap(void)
818 return atomic_read(&trapped
);
821 void netpoll_set_trap(int trap
)
824 atomic_inc(&trapped
);
826 atomic_dec(&trapped
);
829 EXPORT_SYMBOL(netpoll_set_trap
);
830 EXPORT_SYMBOL(netpoll_trap
);
831 EXPORT_SYMBOL(netpoll_parse_options
);
832 EXPORT_SYMBOL(netpoll_setup
);
833 EXPORT_SYMBOL(netpoll_cleanup
);
834 EXPORT_SYMBOL(netpoll_send_udp
);
835 EXPORT_SYMBOL(netpoll_poll
);