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 npinfo
->rx_flags
|= NETPOLL_RX_DROP
;
128 atomic_inc(&trapped
);
130 np
->dev
->poll(np
->dev
, &budget
);
132 atomic_dec(&trapped
);
133 npinfo
->rx_flags
&= ~NETPOLL_RX_DROP
;
134 spin_unlock(&npinfo
->poll_lock
);
138 static void service_arp_queue(struct netpoll_info
*npi
)
145 skb
= skb_dequeue(&npi
->arp_tx
);
147 while (skb
!= NULL
) {
149 skb
= skb_dequeue(&npi
->arp_tx
);
153 void netpoll_poll(struct netpoll
*np
)
155 if (!np
->dev
|| !netif_running(np
->dev
) || !np
->dev
->poll_controller
)
158 /* Process pending work on NIC */
159 np
->dev
->poll_controller(np
->dev
);
163 service_arp_queue(np
->dev
->npinfo
);
165 zap_completion_queue();
168 static void refill_skbs(void)
173 spin_lock_irqsave(&skb_pool
.lock
, flags
);
174 while (skb_pool
.qlen
< MAX_SKBS
) {
175 skb
= alloc_skb(MAX_SKB_SIZE
, GFP_ATOMIC
);
179 __skb_queue_tail(&skb_pool
, skb
);
181 spin_unlock_irqrestore(&skb_pool
.lock
, flags
);
184 static void zap_completion_queue(void)
187 struct softnet_data
*sd
= &get_cpu_var(softnet_data
);
189 if (sd
->completion_queue
) {
190 struct sk_buff
*clist
;
192 local_irq_save(flags
);
193 clist
= sd
->completion_queue
;
194 sd
->completion_queue
= NULL
;
195 local_irq_restore(flags
);
197 while (clist
!= NULL
) {
198 struct sk_buff
*skb
= clist
;
201 dev_kfree_skb_any(skb
); /* put this one back */
207 put_cpu_var(softnet_data
);
210 static struct sk_buff
*find_skb(struct netpoll
*np
, int len
, int reserve
)
215 zap_completion_queue();
219 skb
= alloc_skb(len
, GFP_ATOMIC
);
221 skb
= skb_dequeue(&skb_pool
);
231 atomic_set(&skb
->users
, 1);
232 skb_reserve(skb
, reserve
);
236 static void netpoll_send_skb(struct netpoll
*np
, struct sk_buff
*skb
)
238 int status
= NETDEV_TX_BUSY
;
240 struct net_device
*dev
= np
->dev
;
241 struct netpoll_info
*npinfo
= np
->dev
->npinfo
;
243 if (!npinfo
|| !netif_running(dev
) || !netif_device_present(dev
)) {
248 /* don't get messages out of order, and no recursion */
249 if (skb_queue_len(&npinfo
->txq
) == 0 &&
250 npinfo
->poll_owner
!= smp_processor_id()) {
253 local_irq_save(flags
);
254 /* try until next clock tick */
255 for (tries
= jiffies_to_usecs(1)/USEC_PER_POLL
;
256 tries
> 0; --tries
) {
257 if (netif_tx_trylock(dev
)) {
258 if (!netif_queue_stopped(dev
) &&
259 !netif_subqueue_stopped(dev
, skb
->queue_mapping
))
260 status
= dev
->hard_start_xmit(skb
, dev
);
261 netif_tx_unlock(dev
);
263 if (status
== NETDEV_TX_OK
)
268 /* tickle device maybe there is some cleanup */
271 udelay(USEC_PER_POLL
);
273 local_irq_restore(flags
);
276 if (status
!= NETDEV_TX_OK
) {
277 skb_queue_tail(&npinfo
->txq
, skb
);
278 schedule_delayed_work(&npinfo
->tx_work
,0);
282 void netpoll_send_udp(struct netpoll
*np
, const char *msg
, int len
)
284 int total_len
, eth_len
, ip_len
, udp_len
;
290 udp_len
= len
+ sizeof(*udph
);
291 ip_len
= eth_len
= udp_len
+ sizeof(*iph
);
292 total_len
= eth_len
+ ETH_HLEN
+ NET_IP_ALIGN
;
294 skb
= find_skb(np
, total_len
, total_len
- len
);
298 skb_copy_to_linear_data(skb
, msg
, len
);
301 skb_push(skb
, sizeof(*udph
));
302 skb_reset_transport_header(skb
);
304 udph
->source
= htons(np
->local_port
);
305 udph
->dest
= htons(np
->remote_port
);
306 udph
->len
= htons(udp_len
);
308 udph
->check
= csum_tcpudp_magic(htonl(np
->local_ip
),
309 htonl(np
->remote_ip
),
310 udp_len
, IPPROTO_UDP
,
311 csum_partial((unsigned char *)udph
, udp_len
, 0));
312 if (udph
->check
== 0)
313 udph
->check
= CSUM_MANGLED_0
;
315 skb_push(skb
, sizeof(*iph
));
316 skb_reset_network_header(skb
);
319 /* iph->version = 4; iph->ihl = 5; */
320 put_unaligned(0x45, (unsigned char *)iph
);
322 put_unaligned(htons(ip_len
), &(iph
->tot_len
));
326 iph
->protocol
= IPPROTO_UDP
;
328 put_unaligned(htonl(np
->local_ip
), &(iph
->saddr
));
329 put_unaligned(htonl(np
->remote_ip
), &(iph
->daddr
));
330 iph
->check
= ip_fast_csum((unsigned char *)iph
, iph
->ihl
);
332 eth
= (struct ethhdr
*) skb_push(skb
, ETH_HLEN
);
333 skb_reset_mac_header(skb
);
334 skb
->protocol
= eth
->h_proto
= htons(ETH_P_IP
);
335 memcpy(eth
->h_source
, np
->local_mac
, 6);
336 memcpy(eth
->h_dest
, np
->remote_mac
, 6);
340 netpoll_send_skb(np
, skb
);
343 static void arp_reply(struct sk_buff
*skb
)
345 struct netpoll_info
*npinfo
= skb
->dev
->npinfo
;
347 unsigned char *arp_ptr
;
348 int size
, type
= ARPOP_REPLY
, ptype
= ETH_P_ARP
;
351 struct sk_buff
*send_skb
;
352 struct netpoll
*np
= NULL
;
354 if (npinfo
->rx_np
&& npinfo
->rx_np
->dev
== skb
->dev
)
359 /* No arp on this interface */
360 if (skb
->dev
->flags
& IFF_NOARP
)
363 if (!pskb_may_pull(skb
, (sizeof(struct arphdr
) +
364 (2 * skb
->dev
->addr_len
) +
368 skb_reset_network_header(skb
);
369 skb_reset_transport_header(skb
);
372 if ((arp
->ar_hrd
!= htons(ARPHRD_ETHER
) &&
373 arp
->ar_hrd
!= htons(ARPHRD_IEEE802
)) ||
374 arp
->ar_pro
!= htons(ETH_P_IP
) ||
375 arp
->ar_op
!= htons(ARPOP_REQUEST
))
378 arp_ptr
= (unsigned char *)(arp
+1);
379 /* save the location of the src hw addr */
381 arp_ptr
+= skb
->dev
->addr_len
;
382 memcpy(&sip
, arp_ptr
, 4);
384 /* if we actually cared about dst hw addr, it would get copied here */
385 arp_ptr
+= skb
->dev
->addr_len
;
386 memcpy(&tip
, arp_ptr
, 4);
388 /* Should we ignore arp? */
389 if (tip
!= htonl(np
->local_ip
) || LOOPBACK(tip
) || MULTICAST(tip
))
392 size
= sizeof(struct arphdr
) + 2 * (skb
->dev
->addr_len
+ 4);
393 send_skb
= find_skb(np
, size
+ LL_RESERVED_SPACE(np
->dev
),
394 LL_RESERVED_SPACE(np
->dev
));
399 skb_reset_network_header(send_skb
);
400 arp
= (struct arphdr
*) skb_put(send_skb
, size
);
401 send_skb
->dev
= skb
->dev
;
402 send_skb
->protocol
= htons(ETH_P_ARP
);
404 /* Fill the device header for the ARP frame */
406 if (np
->dev
->hard_header
&&
407 np
->dev
->hard_header(send_skb
, skb
->dev
, ptype
,
409 send_skb
->len
) < 0) {
415 * Fill out the arp protocol part.
417 * we only support ethernet device type,
418 * which (according to RFC 1390) should always equal 1 (Ethernet).
421 arp
->ar_hrd
= htons(np
->dev
->type
);
422 arp
->ar_pro
= htons(ETH_P_IP
);
423 arp
->ar_hln
= np
->dev
->addr_len
;
425 arp
->ar_op
= htons(type
);
427 arp_ptr
=(unsigned char *)(arp
+ 1);
428 memcpy(arp_ptr
, np
->dev
->dev_addr
, np
->dev
->addr_len
);
429 arp_ptr
+= np
->dev
->addr_len
;
430 memcpy(arp_ptr
, &tip
, 4);
432 memcpy(arp_ptr
, sha
, np
->dev
->addr_len
);
433 arp_ptr
+= np
->dev
->addr_len
;
434 memcpy(arp_ptr
, &sip
, 4);
436 netpoll_send_skb(np
, send_skb
);
439 int __netpoll_rx(struct sk_buff
*skb
)
441 int proto
, len
, ulen
;
444 struct netpoll_info
*npi
= skb
->dev
->npinfo
;
445 struct netpoll
*np
= npi
->rx_np
;
449 if (skb
->dev
->type
!= ARPHRD_ETHER
)
452 /* check if netpoll clients need ARP */
453 if (skb
->protocol
== htons(ETH_P_ARP
) &&
454 atomic_read(&trapped
)) {
455 skb_queue_tail(&npi
->arp_tx
, skb
);
459 proto
= ntohs(eth_hdr(skb
)->h_proto
);
460 if (proto
!= ETH_P_IP
)
462 if (skb
->pkt_type
== PACKET_OTHERHOST
)
467 iph
= (struct iphdr
*)skb
->data
;
468 if (!pskb_may_pull(skb
, sizeof(struct iphdr
)))
470 if (iph
->ihl
< 5 || iph
->version
!= 4)
472 if (!pskb_may_pull(skb
, iph
->ihl
*4))
474 if (ip_fast_csum((u8
*)iph
, iph
->ihl
) != 0)
477 len
= ntohs(iph
->tot_len
);
478 if (skb
->len
< len
|| len
< iph
->ihl
*4)
482 * Our transport medium may have padded the buffer out.
483 * Now We trim to the true length of the frame.
485 if (pskb_trim_rcsum(skb
, len
))
488 if (iph
->protocol
!= IPPROTO_UDP
)
492 uh
= (struct udphdr
*)(((char *)iph
) + iph
->ihl
*4);
493 ulen
= ntohs(uh
->len
);
497 if (checksum_udp(skb
, uh
, ulen
, iph
->saddr
, iph
->daddr
))
499 if (np
->local_ip
&& np
->local_ip
!= ntohl(iph
->daddr
))
501 if (np
->remote_ip
&& np
->remote_ip
!= ntohl(iph
->saddr
))
503 if (np
->local_port
&& np
->local_port
!= ntohs(uh
->dest
))
506 np
->rx_hook(np
, ntohs(uh
->source
),
508 ulen
- sizeof(struct udphdr
));
514 if (atomic_read(&trapped
)) {
522 int netpoll_parse_options(struct netpoll
*np
, char *opt
)
524 char *cur
=opt
, *delim
;
527 if ((delim
= strchr(cur
, '@')) == NULL
)
530 np
->local_port
= simple_strtol(cur
, NULL
, 10);
534 printk(KERN_INFO
"%s: local port %d\n", np
->name
, np
->local_port
);
537 if ((delim
= strchr(cur
, '/')) == NULL
)
540 np
->local_ip
= ntohl(in_aton(cur
));
543 printk(KERN_INFO
"%s: local IP %d.%d.%d.%d\n",
544 np
->name
, HIPQUAD(np
->local_ip
));
549 /* parse out dev name */
550 if ((delim
= strchr(cur
, ',')) == NULL
)
553 strlcpy(np
->dev_name
, cur
, sizeof(np
->dev_name
));
558 printk(KERN_INFO
"%s: interface %s\n", np
->name
, np
->dev_name
);
562 if ((delim
= strchr(cur
, '@')) == NULL
)
565 np
->remote_port
= simple_strtol(cur
, NULL
, 10);
569 printk(KERN_INFO
"%s: remote port %d\n", np
->name
, np
->remote_port
);
572 if ((delim
= strchr(cur
, '/')) == NULL
)
575 np
->remote_ip
= ntohl(in_aton(cur
));
578 printk(KERN_INFO
"%s: remote IP %d.%d.%d.%d\n",
579 np
->name
, HIPQUAD(np
->remote_ip
));
583 if ((delim
= strchr(cur
, ':')) == NULL
)
586 np
->remote_mac
[0] = simple_strtol(cur
, NULL
, 16);
588 if ((delim
= strchr(cur
, ':')) == NULL
)
591 np
->remote_mac
[1] = simple_strtol(cur
, NULL
, 16);
593 if ((delim
= strchr(cur
, ':')) == NULL
)
596 np
->remote_mac
[2] = simple_strtol(cur
, NULL
, 16);
598 if ((delim
= strchr(cur
, ':')) == NULL
)
601 np
->remote_mac
[3] = simple_strtol(cur
, NULL
, 16);
603 if ((delim
= strchr(cur
, ':')) == NULL
)
606 np
->remote_mac
[4] = simple_strtol(cur
, NULL
, 16);
608 np
->remote_mac
[5] = simple_strtol(cur
, NULL
, 16);
611 printk(KERN_INFO
"%s: remote ethernet address "
612 "%02x:%02x:%02x:%02x:%02x:%02x\n",
624 printk(KERN_INFO
"%s: couldn't parse config at %s!\n",
629 int netpoll_setup(struct netpoll
*np
)
631 struct net_device
*ndev
= NULL
;
632 struct in_device
*in_dev
;
633 struct netpoll_info
*npinfo
;
638 ndev
= dev_get_by_name(np
->dev_name
);
640 printk(KERN_ERR
"%s: %s doesn't exist, aborting.\n",
641 np
->name
, np
->dev_name
);
647 npinfo
= kmalloc(sizeof(*npinfo
), GFP_KERNEL
);
653 npinfo
->rx_flags
= 0;
654 npinfo
->rx_np
= NULL
;
655 spin_lock_init(&npinfo
->poll_lock
);
656 npinfo
->poll_owner
= -1;
658 spin_lock_init(&npinfo
->rx_lock
);
659 skb_queue_head_init(&npinfo
->arp_tx
);
660 skb_queue_head_init(&npinfo
->txq
);
661 INIT_DELAYED_WORK(&npinfo
->tx_work
, queue_process
);
663 atomic_set(&npinfo
->refcnt
, 1);
665 npinfo
= ndev
->npinfo
;
666 atomic_inc(&npinfo
->refcnt
);
669 if (!ndev
->poll_controller
) {
670 printk(KERN_ERR
"%s: %s doesn't support polling, aborting.\n",
671 np
->name
, np
->dev_name
);
676 if (!netif_running(ndev
)) {
677 unsigned long atmost
, atleast
;
679 printk(KERN_INFO
"%s: device %s not up yet, forcing it\n",
680 np
->name
, np
->dev_name
);
683 err
= dev_open(ndev
);
687 printk(KERN_ERR
"%s: failed to open %s\n",
688 np
->name
, ndev
->name
);
692 atleast
= jiffies
+ HZ
/10;
693 atmost
= jiffies
+ 4*HZ
;
694 while (!netif_carrier_ok(ndev
)) {
695 if (time_after(jiffies
, atmost
)) {
697 "%s: timeout waiting for carrier\n",
704 /* If carrier appears to come up instantly, we don't
705 * trust it and pause so that we don't pump all our
706 * queued console messages into the bitbucket.
709 if (time_before(jiffies
, atleast
)) {
710 printk(KERN_NOTICE
"%s: carrier detect appears"
711 " untrustworthy, waiting 4 seconds\n",
717 if (is_zero_ether_addr(np
->local_mac
) && ndev
->dev_addr
)
718 memcpy(np
->local_mac
, ndev
->dev_addr
, 6);
722 in_dev
= __in_dev_get_rcu(ndev
);
724 if (!in_dev
|| !in_dev
->ifa_list
) {
726 printk(KERN_ERR
"%s: no IP address for %s, aborting\n",
727 np
->name
, np
->dev_name
);
732 np
->local_ip
= ntohl(in_dev
->ifa_list
->ifa_local
);
734 printk(KERN_INFO
"%s: local IP %d.%d.%d.%d\n",
735 np
->name
, HIPQUAD(np
->local_ip
));
739 spin_lock_irqsave(&npinfo
->rx_lock
, flags
);
740 npinfo
->rx_flags
|= NETPOLL_RX_ENABLED
;
742 spin_unlock_irqrestore(&npinfo
->rx_lock
, flags
);
745 /* fill up the skb queue */
748 /* last thing to do is link it to the net device structure */
749 ndev
->npinfo
= npinfo
;
751 /* avoid racing with NAPI reading npinfo */
764 static int __init
netpoll_init(void)
766 skb_queue_head_init(&skb_pool
);
769 core_initcall(netpoll_init
);
771 void netpoll_cleanup(struct netpoll
*np
)
773 struct netpoll_info
*npinfo
;
777 npinfo
= np
->dev
->npinfo
;
779 if (npinfo
->rx_np
== np
) {
780 spin_lock_irqsave(&npinfo
->rx_lock
, flags
);
781 npinfo
->rx_np
= NULL
;
782 npinfo
->rx_flags
&= ~NETPOLL_RX_ENABLED
;
783 spin_unlock_irqrestore(&npinfo
->rx_lock
, flags
);
786 if (atomic_dec_and_test(&npinfo
->refcnt
)) {
787 skb_queue_purge(&npinfo
->arp_tx
);
788 skb_queue_purge(&npinfo
->txq
);
789 cancel_rearming_delayed_work(&npinfo
->tx_work
);
791 /* clean after last, unfinished work */
792 if (!skb_queue_empty(&npinfo
->txq
)) {
794 skb
= __skb_dequeue(&npinfo
->txq
);
798 np
->dev
->npinfo
= NULL
;
808 int netpoll_trap(void)
810 return atomic_read(&trapped
);
813 void netpoll_set_trap(int trap
)
816 atomic_inc(&trapped
);
818 atomic_dec(&trapped
);
821 EXPORT_SYMBOL(netpoll_set_trap
);
822 EXPORT_SYMBOL(netpoll_trap
);
823 EXPORT_SYMBOL(netpoll_parse_options
);
824 EXPORT_SYMBOL(netpoll_setup
);
825 EXPORT_SYMBOL(netpoll_cleanup
);
826 EXPORT_SYMBOL(netpoll_send_udp
);
827 EXPORT_SYMBOL(netpoll_poll
);