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
)) ||
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 int poll_one_napi(struct netpoll_info
*npinfo
,
120 struct napi_struct
*napi
, int budget
)
124 /* net_rx_action's ->poll() invocations and our's are
125 * synchronized by this test which is only made while
126 * holding the napi->poll_lock.
128 if (!test_bit(NAPI_STATE_SCHED
, &napi
->state
))
131 npinfo
->rx_flags
|= NETPOLL_RX_DROP
;
132 atomic_inc(&trapped
);
134 work
= napi
->poll(napi
, budget
);
136 atomic_dec(&trapped
);
137 npinfo
->rx_flags
&= ~NETPOLL_RX_DROP
;
139 return budget
- work
;
142 static void poll_napi(struct netpoll
*np
)
144 struct netpoll_info
*npinfo
= np
->dev
->npinfo
;
145 struct napi_struct
*napi
;
148 list_for_each_entry(napi
, &np
->dev
->napi_list
, dev_list
) {
149 if (napi
->poll_owner
!= smp_processor_id() &&
150 spin_trylock(&napi
->poll_lock
)) {
151 budget
= poll_one_napi(npinfo
, napi
, budget
);
152 spin_unlock(&napi
->poll_lock
);
160 static void service_arp_queue(struct netpoll_info
*npi
)
167 skb
= skb_dequeue(&npi
->arp_tx
);
169 while (skb
!= NULL
) {
171 skb
= skb_dequeue(&npi
->arp_tx
);
175 void netpoll_poll(struct netpoll
*np
)
177 if (!np
->dev
|| !netif_running(np
->dev
) || !np
->dev
->poll_controller
)
180 /* Process pending work on NIC */
181 np
->dev
->poll_controller(np
->dev
);
182 if (!list_empty(&np
->dev
->napi_list
))
185 service_arp_queue(np
->dev
->npinfo
);
187 zap_completion_queue();
190 static void refill_skbs(void)
195 spin_lock_irqsave(&skb_pool
.lock
, flags
);
196 while (skb_pool
.qlen
< MAX_SKBS
) {
197 skb
= alloc_skb(MAX_SKB_SIZE
, GFP_ATOMIC
);
201 __skb_queue_tail(&skb_pool
, skb
);
203 spin_unlock_irqrestore(&skb_pool
.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
)
237 zap_completion_queue();
241 skb
= alloc_skb(len
, GFP_ATOMIC
);
243 skb
= skb_dequeue(&skb_pool
);
253 atomic_set(&skb
->users
, 1);
254 skb_reserve(skb
, reserve
);
258 static int netpoll_owner_active(struct net_device
*dev
)
260 struct napi_struct
*napi
;
262 list_for_each_entry(napi
, &dev
->napi_list
, dev_list
) {
263 if (napi
->poll_owner
== smp_processor_id())
269 static void netpoll_send_skb(struct netpoll
*np
, struct sk_buff
*skb
)
271 int status
= NETDEV_TX_BUSY
;
273 struct net_device
*dev
= np
->dev
;
274 struct netpoll_info
*npinfo
= np
->dev
->npinfo
;
276 if (!npinfo
|| !netif_running(dev
) || !netif_device_present(dev
)) {
281 /* don't get messages out of order, and no recursion */
282 if (skb_queue_len(&npinfo
->txq
) == 0 && !netpoll_owner_active(dev
)) {
285 local_irq_save(flags
);
286 /* try until next clock tick */
287 for (tries
= jiffies_to_usecs(1)/USEC_PER_POLL
;
288 tries
> 0; --tries
) {
289 if (netif_tx_trylock(dev
)) {
290 if (!netif_queue_stopped(dev
) &&
291 !netif_subqueue_stopped(dev
, skb
))
292 status
= dev
->hard_start_xmit(skb
, dev
);
293 netif_tx_unlock(dev
);
295 if (status
== NETDEV_TX_OK
)
300 /* tickle device maybe there is some cleanup */
303 udelay(USEC_PER_POLL
);
305 local_irq_restore(flags
);
308 if (status
!= NETDEV_TX_OK
) {
309 skb_queue_tail(&npinfo
->txq
, skb
);
310 schedule_delayed_work(&npinfo
->tx_work
,0);
314 void netpoll_send_udp(struct netpoll
*np
, const char *msg
, int len
)
316 int total_len
, eth_len
, ip_len
, udp_len
;
322 udp_len
= len
+ sizeof(*udph
);
323 ip_len
= eth_len
= udp_len
+ sizeof(*iph
);
324 total_len
= eth_len
+ ETH_HLEN
+ NET_IP_ALIGN
;
326 skb
= find_skb(np
, total_len
, total_len
- len
);
330 skb_copy_to_linear_data(skb
, msg
, len
);
333 skb_push(skb
, sizeof(*udph
));
334 skb_reset_transport_header(skb
);
336 udph
->source
= htons(np
->local_port
);
337 udph
->dest
= htons(np
->remote_port
);
338 udph
->len
= htons(udp_len
);
340 udph
->check
= csum_tcpudp_magic(htonl(np
->local_ip
),
341 htonl(np
->remote_ip
),
342 udp_len
, IPPROTO_UDP
,
343 csum_partial((unsigned char *)udph
, udp_len
, 0));
344 if (udph
->check
== 0)
345 udph
->check
= CSUM_MANGLED_0
;
347 skb_push(skb
, sizeof(*iph
));
348 skb_reset_network_header(skb
);
351 /* iph->version = 4; iph->ihl = 5; */
352 put_unaligned(0x45, (unsigned char *)iph
);
354 put_unaligned(htons(ip_len
), &(iph
->tot_len
));
358 iph
->protocol
= IPPROTO_UDP
;
360 put_unaligned(htonl(np
->local_ip
), &(iph
->saddr
));
361 put_unaligned(htonl(np
->remote_ip
), &(iph
->daddr
));
362 iph
->check
= ip_fast_csum((unsigned char *)iph
, iph
->ihl
);
364 eth
= (struct ethhdr
*) skb_push(skb
, ETH_HLEN
);
365 skb_reset_mac_header(skb
);
366 skb
->protocol
= eth
->h_proto
= htons(ETH_P_IP
);
367 memcpy(eth
->h_source
, np
->local_mac
, 6);
368 memcpy(eth
->h_dest
, np
->remote_mac
, 6);
372 netpoll_send_skb(np
, skb
);
375 static void arp_reply(struct sk_buff
*skb
)
377 struct netpoll_info
*npinfo
= skb
->dev
->npinfo
;
379 unsigned char *arp_ptr
;
380 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_reset_network_header(skb
);
401 skb_reset_transport_header(skb
);
404 if ((arp
->ar_hrd
!= htons(ARPHRD_ETHER
) &&
405 arp
->ar_hrd
!= htons(ARPHRD_IEEE802
)) ||
406 arp
->ar_pro
!= htons(ETH_P_IP
) ||
407 arp
->ar_op
!= htons(ARPOP_REQUEST
))
410 arp_ptr
= (unsigned char *)(arp
+1);
411 /* save the location of the src hw addr */
413 arp_ptr
+= skb
->dev
->addr_len
;
414 memcpy(&sip
, arp_ptr
, 4);
416 /* if we actually cared about dst hw addr, it would get copied here */
417 arp_ptr
+= skb
->dev
->addr_len
;
418 memcpy(&tip
, arp_ptr
, 4);
420 /* Should we ignore arp? */
421 if (tip
!= htonl(np
->local_ip
) || LOOPBACK(tip
) || MULTICAST(tip
))
424 size
= sizeof(struct arphdr
) + 2 * (skb
->dev
->addr_len
+ 4);
425 send_skb
= find_skb(np
, size
+ LL_RESERVED_SPACE(np
->dev
),
426 LL_RESERVED_SPACE(np
->dev
));
431 skb_reset_network_header(send_skb
);
432 arp
= (struct arphdr
*) skb_put(send_skb
, size
);
433 send_skb
->dev
= skb
->dev
;
434 send_skb
->protocol
= htons(ETH_P_ARP
);
436 /* Fill the device header for the ARP frame */
437 if (dev_hard_header(send_skb
, skb
->dev
, ptype
,
439 send_skb
->len
) < 0) {
445 * Fill out the arp protocol part.
447 * we only support ethernet device type,
448 * which (according to RFC 1390) should always equal 1 (Ethernet).
451 arp
->ar_hrd
= htons(np
->dev
->type
);
452 arp
->ar_pro
= htons(ETH_P_IP
);
453 arp
->ar_hln
= np
->dev
->addr_len
;
455 arp
->ar_op
= htons(type
);
457 arp_ptr
=(unsigned char *)(arp
+ 1);
458 memcpy(arp_ptr
, np
->dev
->dev_addr
, np
->dev
->addr_len
);
459 arp_ptr
+= np
->dev
->addr_len
;
460 memcpy(arp_ptr
, &tip
, 4);
462 memcpy(arp_ptr
, sha
, np
->dev
->addr_len
);
463 arp_ptr
+= np
->dev
->addr_len
;
464 memcpy(arp_ptr
, &sip
, 4);
466 netpoll_send_skb(np
, send_skb
);
469 int __netpoll_rx(struct sk_buff
*skb
)
471 int proto
, len
, ulen
;
474 struct netpoll_info
*npi
= skb
->dev
->npinfo
;
475 struct netpoll
*np
= npi
->rx_np
;
479 if (skb
->dev
->type
!= ARPHRD_ETHER
)
482 /* check if netpoll clients need ARP */
483 if (skb
->protocol
== htons(ETH_P_ARP
) &&
484 atomic_read(&trapped
)) {
485 skb_queue_tail(&npi
->arp_tx
, skb
);
489 proto
= ntohs(eth_hdr(skb
)->h_proto
);
490 if (proto
!= ETH_P_IP
)
492 if (skb
->pkt_type
== PACKET_OTHERHOST
)
497 iph
= (struct iphdr
*)skb
->data
;
498 if (!pskb_may_pull(skb
, sizeof(struct iphdr
)))
500 if (iph
->ihl
< 5 || iph
->version
!= 4)
502 if (!pskb_may_pull(skb
, iph
->ihl
*4))
504 if (ip_fast_csum((u8
*)iph
, iph
->ihl
) != 0)
507 len
= ntohs(iph
->tot_len
);
508 if (skb
->len
< len
|| len
< iph
->ihl
*4)
512 * Our transport medium may have padded the buffer out.
513 * Now We trim to the true length of the frame.
515 if (pskb_trim_rcsum(skb
, len
))
518 if (iph
->protocol
!= IPPROTO_UDP
)
522 uh
= (struct udphdr
*)(((char *)iph
) + iph
->ihl
*4);
523 ulen
= ntohs(uh
->len
);
527 if (checksum_udp(skb
, uh
, ulen
, iph
->saddr
, iph
->daddr
))
529 if (np
->local_ip
&& np
->local_ip
!= ntohl(iph
->daddr
))
531 if (np
->remote_ip
&& np
->remote_ip
!= ntohl(iph
->saddr
))
533 if (np
->local_port
&& np
->local_port
!= ntohs(uh
->dest
))
536 np
->rx_hook(np
, ntohs(uh
->source
),
538 ulen
- sizeof(struct udphdr
));
544 if (atomic_read(&trapped
)) {
552 void netpoll_print_options(struct netpoll
*np
)
554 DECLARE_MAC_BUF(mac
);
555 printk(KERN_INFO
"%s: local port %d\n",
556 np
->name
, np
->local_port
);
557 printk(KERN_INFO
"%s: local IP %d.%d.%d.%d\n",
558 np
->name
, HIPQUAD(np
->local_ip
));
559 printk(KERN_INFO
"%s: interface %s\n",
560 np
->name
, np
->dev_name
);
561 printk(KERN_INFO
"%s: remote port %d\n",
562 np
->name
, np
->remote_port
);
563 printk(KERN_INFO
"%s: remote IP %d.%d.%d.%d\n",
564 np
->name
, HIPQUAD(np
->remote_ip
));
565 printk(KERN_INFO
"%s: remote ethernet address %s\n",
566 np
->name
, print_mac(mac
, np
->remote_mac
));
569 int netpoll_parse_options(struct netpoll
*np
, char *opt
)
571 char *cur
=opt
, *delim
;
574 if ((delim
= strchr(cur
, '@')) == NULL
)
577 np
->local_port
= simple_strtol(cur
, NULL
, 10);
583 if ((delim
= strchr(cur
, '/')) == NULL
)
586 np
->local_ip
= ntohl(in_aton(cur
));
592 /* parse out dev name */
593 if ((delim
= strchr(cur
, ',')) == NULL
)
596 strlcpy(np
->dev_name
, cur
, sizeof(np
->dev_name
));
603 if ((delim
= strchr(cur
, '@')) == NULL
)
606 np
->remote_port
= simple_strtol(cur
, NULL
, 10);
612 if ((delim
= strchr(cur
, '/')) == NULL
)
615 np
->remote_ip
= ntohl(in_aton(cur
));
620 if ((delim
= strchr(cur
, ':')) == NULL
)
623 np
->remote_mac
[0] = simple_strtol(cur
, NULL
, 16);
625 if ((delim
= strchr(cur
, ':')) == NULL
)
628 np
->remote_mac
[1] = simple_strtol(cur
, NULL
, 16);
630 if ((delim
= strchr(cur
, ':')) == NULL
)
633 np
->remote_mac
[2] = simple_strtol(cur
, NULL
, 16);
635 if ((delim
= strchr(cur
, ':')) == NULL
)
638 np
->remote_mac
[3] = simple_strtol(cur
, NULL
, 16);
640 if ((delim
= strchr(cur
, ':')) == NULL
)
643 np
->remote_mac
[4] = simple_strtol(cur
, NULL
, 16);
645 np
->remote_mac
[5] = simple_strtol(cur
, NULL
, 16);
648 netpoll_print_options(np
);
653 printk(KERN_INFO
"%s: couldn't parse config at %s!\n",
658 int netpoll_setup(struct netpoll
*np
)
660 struct net_device
*ndev
= NULL
;
661 struct in_device
*in_dev
;
662 struct netpoll_info
*npinfo
;
667 ndev
= dev_get_by_name(&init_net
, np
->dev_name
);
669 printk(KERN_ERR
"%s: %s doesn't exist, aborting.\n",
670 np
->name
, np
->dev_name
);
676 npinfo
= kmalloc(sizeof(*npinfo
), GFP_KERNEL
);
682 npinfo
->rx_flags
= 0;
683 npinfo
->rx_np
= NULL
;
685 spin_lock_init(&npinfo
->rx_lock
);
686 skb_queue_head_init(&npinfo
->arp_tx
);
687 skb_queue_head_init(&npinfo
->txq
);
688 INIT_DELAYED_WORK(&npinfo
->tx_work
, queue_process
);
690 atomic_set(&npinfo
->refcnt
, 1);
692 npinfo
= ndev
->npinfo
;
693 atomic_inc(&npinfo
->refcnt
);
696 if (!ndev
->poll_controller
) {
697 printk(KERN_ERR
"%s: %s doesn't support polling, aborting.\n",
698 np
->name
, np
->dev_name
);
703 if (!netif_running(ndev
)) {
704 unsigned long atmost
, atleast
;
706 printk(KERN_INFO
"%s: device %s not up yet, forcing it\n",
707 np
->name
, np
->dev_name
);
710 err
= dev_open(ndev
);
714 printk(KERN_ERR
"%s: failed to open %s\n",
715 np
->name
, ndev
->name
);
719 atleast
= jiffies
+ HZ
/10;
720 atmost
= jiffies
+ 4*HZ
;
721 while (!netif_carrier_ok(ndev
)) {
722 if (time_after(jiffies
, atmost
)) {
724 "%s: timeout waiting for carrier\n",
731 /* If carrier appears to come up instantly, we don't
732 * trust it and pause so that we don't pump all our
733 * queued console messages into the bitbucket.
736 if (time_before(jiffies
, atleast
)) {
737 printk(KERN_NOTICE
"%s: carrier detect appears"
738 " untrustworthy, waiting 4 seconds\n",
744 if (is_zero_ether_addr(np
->local_mac
) && ndev
->dev_addr
)
745 memcpy(np
->local_mac
, ndev
->dev_addr
, 6);
749 in_dev
= __in_dev_get_rcu(ndev
);
751 if (!in_dev
|| !in_dev
->ifa_list
) {
753 printk(KERN_ERR
"%s: no IP address for %s, aborting\n",
754 np
->name
, np
->dev_name
);
759 np
->local_ip
= ntohl(in_dev
->ifa_list
->ifa_local
);
761 printk(KERN_INFO
"%s: local IP %d.%d.%d.%d\n",
762 np
->name
, HIPQUAD(np
->local_ip
));
766 spin_lock_irqsave(&npinfo
->rx_lock
, flags
);
767 npinfo
->rx_flags
|= NETPOLL_RX_ENABLED
;
769 spin_unlock_irqrestore(&npinfo
->rx_lock
, flags
);
772 /* fill up the skb queue */
775 /* last thing to do is link it to the net device structure */
776 ndev
->npinfo
= npinfo
;
778 /* avoid racing with NAPI reading npinfo */
791 static int __init
netpoll_init(void)
793 skb_queue_head_init(&skb_pool
);
796 core_initcall(netpoll_init
);
798 void netpoll_cleanup(struct netpoll
*np
)
800 struct netpoll_info
*npinfo
;
804 npinfo
= np
->dev
->npinfo
;
806 if (npinfo
->rx_np
== np
) {
807 spin_lock_irqsave(&npinfo
->rx_lock
, flags
);
808 npinfo
->rx_np
= NULL
;
809 npinfo
->rx_flags
&= ~NETPOLL_RX_ENABLED
;
810 spin_unlock_irqrestore(&npinfo
->rx_lock
, flags
);
813 if (atomic_dec_and_test(&npinfo
->refcnt
)) {
814 skb_queue_purge(&npinfo
->arp_tx
);
815 skb_queue_purge(&npinfo
->txq
);
816 cancel_rearming_delayed_work(&npinfo
->tx_work
);
818 /* clean after last, unfinished work */
819 if (!skb_queue_empty(&npinfo
->txq
)) {
821 skb
= __skb_dequeue(&npinfo
->txq
);
825 np
->dev
->npinfo
= NULL
;
835 int netpoll_trap(void)
837 return atomic_read(&trapped
);
840 void netpoll_set_trap(int trap
)
843 atomic_inc(&trapped
);
845 atomic_dec(&trapped
);
848 EXPORT_SYMBOL(netpoll_set_trap
);
849 EXPORT_SYMBOL(netpoll_trap
);
850 EXPORT_SYMBOL(netpoll_print_options
);
851 EXPORT_SYMBOL(netpoll_parse_options
);
852 EXPORT_SYMBOL(netpoll_setup
);
853 EXPORT_SYMBOL(netpoll_cleanup
);
854 EXPORT_SYMBOL(netpoll_send_udp
);
855 EXPORT_SYMBOL(netpoll_poll
);