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
;
61 const struct net_device_ops
*ops
= dev
->netdev_ops
;
62 struct netdev_queue
*txq
;
64 if (!netif_device_present(dev
) || !netif_running(dev
)) {
69 txq
= netdev_get_tx_queue(dev
, skb_get_queue_mapping(skb
));
71 local_irq_save(flags
);
72 __netif_tx_lock(txq
, smp_processor_id());
73 if (netif_tx_queue_stopped(txq
) ||
74 netif_tx_queue_frozen(txq
) ||
75 ops
->ndo_start_xmit(skb
, dev
) != NETDEV_TX_OK
) {
76 skb_queue_head(&npinfo
->txq
, skb
);
77 __netif_tx_unlock(txq
);
78 local_irq_restore(flags
);
80 schedule_delayed_work(&npinfo
->tx_work
, HZ
/10);
83 __netif_tx_unlock(txq
);
84 local_irq_restore(flags
);
88 static __sum16
checksum_udp(struct sk_buff
*skb
, struct udphdr
*uh
,
89 unsigned short ulen
, __be32 saddr
, __be32 daddr
)
93 if (uh
->check
== 0 || skb_csum_unnecessary(skb
))
96 psum
= csum_tcpudp_nofold(saddr
, daddr
, ulen
, IPPROTO_UDP
, 0);
98 if (skb
->ip_summed
== CHECKSUM_COMPLETE
&&
99 !csum_fold(csum_add(psum
, skb
->csum
)))
104 return __skb_checksum_complete(skb
);
108 * Check whether delayed processing was scheduled for our NIC. If so,
109 * we attempt to grab the poll lock and use ->poll() to pump the card.
110 * If this fails, either we've recursed in ->poll() or it's already
111 * running on another CPU.
113 * Note: we don't mask interrupts with this lock because we're using
114 * trylock here and interrupts are already disabled in the softirq
115 * case. Further, we test the poll_owner to avoid recursion on UP
116 * systems where the lock doesn't exist.
118 * In cases where there is bi-directional communications, reading only
119 * one message at a time can lead to packets being dropped by the
120 * network adapter, forcing superfluous retries and possibly timeouts.
121 * Thus, we set our budget to greater than 1.
123 static int poll_one_napi(struct netpoll_info
*npinfo
,
124 struct napi_struct
*napi
, int budget
)
128 /* net_rx_action's ->poll() invocations and our's are
129 * synchronized by this test which is only made while
130 * holding the napi->poll_lock.
132 if (!test_bit(NAPI_STATE_SCHED
, &napi
->state
))
135 npinfo
->rx_flags
|= NETPOLL_RX_DROP
;
136 atomic_inc(&trapped
);
138 work
= napi
->poll(napi
, budget
);
140 atomic_dec(&trapped
);
141 npinfo
->rx_flags
&= ~NETPOLL_RX_DROP
;
143 return budget
- work
;
146 static void poll_napi(struct net_device
*dev
)
148 struct napi_struct
*napi
;
151 list_for_each_entry(napi
, &dev
->napi_list
, dev_list
) {
152 if (napi
->poll_owner
!= smp_processor_id() &&
153 spin_trylock(&napi
->poll_lock
)) {
154 budget
= poll_one_napi(dev
->npinfo
, napi
, budget
);
155 spin_unlock(&napi
->poll_lock
);
163 static void service_arp_queue(struct netpoll_info
*npi
)
168 while ((skb
= skb_dequeue(&npi
->arp_tx
)))
173 void netpoll_poll(struct netpoll
*np
)
175 struct net_device
*dev
= np
->dev
;
176 const struct net_device_ops
*ops
= dev
->netdev_ops
;
178 if (!dev
|| !netif_running(dev
) || !ops
->ndo_poll_controller
)
181 /* Process pending work on NIC */
182 ops
->ndo_poll_controller(dev
);
186 service_arp_queue(dev
->npinfo
);
188 zap_completion_queue();
191 static void refill_skbs(void)
196 spin_lock_irqsave(&skb_pool
.lock
, flags
);
197 while (skb_pool
.qlen
< MAX_SKBS
) {
198 skb
= alloc_skb(MAX_SKB_SIZE
, GFP_ATOMIC
);
202 __skb_queue_tail(&skb_pool
, skb
);
204 spin_unlock_irqrestore(&skb_pool
.lock
, flags
);
207 static void zap_completion_queue(void)
210 struct softnet_data
*sd
= &get_cpu_var(softnet_data
);
212 if (sd
->completion_queue
) {
213 struct sk_buff
*clist
;
215 local_irq_save(flags
);
216 clist
= sd
->completion_queue
;
217 sd
->completion_queue
= NULL
;
218 local_irq_restore(flags
);
220 while (clist
!= NULL
) {
221 struct sk_buff
*skb
= clist
;
223 if (skb
->destructor
) {
224 atomic_inc(&skb
->users
);
225 dev_kfree_skb_any(skb
); /* put this one back */
232 put_cpu_var(softnet_data
);
235 static struct sk_buff
*find_skb(struct netpoll
*np
, int len
, int reserve
)
240 zap_completion_queue();
244 skb
= alloc_skb(len
, GFP_ATOMIC
);
246 skb
= skb_dequeue(&skb_pool
);
256 atomic_set(&skb
->users
, 1);
257 skb_reserve(skb
, reserve
);
261 static int netpoll_owner_active(struct net_device
*dev
)
263 struct napi_struct
*napi
;
265 list_for_each_entry(napi
, &dev
->napi_list
, dev_list
) {
266 if (napi
->poll_owner
== smp_processor_id())
272 static void netpoll_send_skb(struct netpoll
*np
, struct sk_buff
*skb
)
274 int status
= NETDEV_TX_BUSY
;
276 struct net_device
*dev
= np
->dev
;
277 const struct net_device_ops
*ops
= dev
->netdev_ops
;
278 struct netpoll_info
*npinfo
= np
->dev
->npinfo
;
280 if (!npinfo
|| !netif_running(dev
) || !netif_device_present(dev
)) {
285 /* don't get messages out of order, and no recursion */
286 if (skb_queue_len(&npinfo
->txq
) == 0 && !netpoll_owner_active(dev
)) {
287 struct netdev_queue
*txq
;
290 txq
= netdev_get_tx_queue(dev
, skb_get_queue_mapping(skb
));
292 local_irq_save(flags
);
293 /* try until next clock tick */
294 for (tries
= jiffies_to_usecs(1)/USEC_PER_POLL
;
295 tries
> 0; --tries
) {
296 if (__netif_tx_trylock(txq
)) {
297 if (!netif_tx_queue_stopped(txq
))
298 status
= ops
->ndo_start_xmit(skb
, dev
);
299 __netif_tx_unlock(txq
);
301 if (status
== NETDEV_TX_OK
)
306 /* tickle device maybe there is some cleanup */
309 udelay(USEC_PER_POLL
);
311 local_irq_restore(flags
);
314 if (status
!= NETDEV_TX_OK
) {
315 skb_queue_tail(&npinfo
->txq
, skb
);
316 schedule_delayed_work(&npinfo
->tx_work
,0);
320 void netpoll_send_udp(struct netpoll
*np
, const char *msg
, int len
)
322 int total_len
, eth_len
, ip_len
, udp_len
;
328 udp_len
= len
+ sizeof(*udph
);
329 ip_len
= eth_len
= udp_len
+ sizeof(*iph
);
330 total_len
= eth_len
+ ETH_HLEN
+ NET_IP_ALIGN
;
332 skb
= find_skb(np
, total_len
, total_len
- len
);
336 skb_copy_to_linear_data(skb
, msg
, len
);
339 skb_push(skb
, sizeof(*udph
));
340 skb_reset_transport_header(skb
);
342 udph
->source
= htons(np
->local_port
);
343 udph
->dest
= htons(np
->remote_port
);
344 udph
->len
= htons(udp_len
);
346 udph
->check
= csum_tcpudp_magic(htonl(np
->local_ip
),
347 htonl(np
->remote_ip
),
348 udp_len
, IPPROTO_UDP
,
349 csum_partial(udph
, udp_len
, 0));
350 if (udph
->check
== 0)
351 udph
->check
= CSUM_MANGLED_0
;
353 skb_push(skb
, sizeof(*iph
));
354 skb_reset_network_header(skb
);
357 /* iph->version = 4; iph->ihl = 5; */
358 put_unaligned(0x45, (unsigned char *)iph
);
360 put_unaligned(htons(ip_len
), &(iph
->tot_len
));
364 iph
->protocol
= IPPROTO_UDP
;
366 put_unaligned(htonl(np
->local_ip
), &(iph
->saddr
));
367 put_unaligned(htonl(np
->remote_ip
), &(iph
->daddr
));
368 iph
->check
= ip_fast_csum((unsigned char *)iph
, iph
->ihl
);
370 eth
= (struct ethhdr
*) skb_push(skb
, ETH_HLEN
);
371 skb_reset_mac_header(skb
);
372 skb
->protocol
= eth
->h_proto
= htons(ETH_P_IP
);
373 memcpy(eth
->h_source
, np
->dev
->dev_addr
, ETH_ALEN
);
374 memcpy(eth
->h_dest
, np
->remote_mac
, ETH_ALEN
);
378 netpoll_send_skb(np
, skb
);
381 static void arp_reply(struct sk_buff
*skb
)
383 struct netpoll_info
*npinfo
= skb
->dev
->npinfo
;
385 unsigned char *arp_ptr
;
386 int size
, type
= ARPOP_REPLY
, ptype
= ETH_P_ARP
;
389 struct sk_buff
*send_skb
;
390 struct netpoll
*np
= NULL
;
392 if (npinfo
->rx_np
&& npinfo
->rx_np
->dev
== skb
->dev
)
397 /* No arp on this interface */
398 if (skb
->dev
->flags
& IFF_NOARP
)
401 if (!pskb_may_pull(skb
, arp_hdr_len(skb
->dev
)))
404 skb_reset_network_header(skb
);
405 skb_reset_transport_header(skb
);
408 if ((arp
->ar_hrd
!= htons(ARPHRD_ETHER
) &&
409 arp
->ar_hrd
!= htons(ARPHRD_IEEE802
)) ||
410 arp
->ar_pro
!= htons(ETH_P_IP
) ||
411 arp
->ar_op
!= htons(ARPOP_REQUEST
))
414 arp_ptr
= (unsigned char *)(arp
+1);
415 /* save the location of the src hw addr */
417 arp_ptr
+= skb
->dev
->addr_len
;
418 memcpy(&sip
, arp_ptr
, 4);
420 /* if we actually cared about dst hw addr, it would get copied here */
421 arp_ptr
+= skb
->dev
->addr_len
;
422 memcpy(&tip
, arp_ptr
, 4);
424 /* Should we ignore arp? */
425 if (tip
!= htonl(np
->local_ip
) ||
426 ipv4_is_loopback(tip
) || ipv4_is_multicast(tip
))
429 size
= arp_hdr_len(skb
->dev
);
430 send_skb
= find_skb(np
, size
+ LL_ALLOCATED_SPACE(np
->dev
),
431 LL_RESERVED_SPACE(np
->dev
));
436 skb_reset_network_header(send_skb
);
437 arp
= (struct arphdr
*) skb_put(send_skb
, size
);
438 send_skb
->dev
= skb
->dev
;
439 send_skb
->protocol
= htons(ETH_P_ARP
);
441 /* Fill the device header for the ARP frame */
442 if (dev_hard_header(send_skb
, skb
->dev
, ptype
,
443 sha
, np
->dev
->dev_addr
,
444 send_skb
->len
) < 0) {
450 * Fill out the arp protocol part.
452 * we only support ethernet device type,
453 * which (according to RFC 1390) should always equal 1 (Ethernet).
456 arp
->ar_hrd
= htons(np
->dev
->type
);
457 arp
->ar_pro
= htons(ETH_P_IP
);
458 arp
->ar_hln
= np
->dev
->addr_len
;
460 arp
->ar_op
= htons(type
);
462 arp_ptr
=(unsigned char *)(arp
+ 1);
463 memcpy(arp_ptr
, np
->dev
->dev_addr
, np
->dev
->addr_len
);
464 arp_ptr
+= np
->dev
->addr_len
;
465 memcpy(arp_ptr
, &tip
, 4);
467 memcpy(arp_ptr
, sha
, np
->dev
->addr_len
);
468 arp_ptr
+= np
->dev
->addr_len
;
469 memcpy(arp_ptr
, &sip
, 4);
471 netpoll_send_skb(np
, send_skb
);
474 int __netpoll_rx(struct sk_buff
*skb
)
476 int proto
, len
, ulen
;
479 struct netpoll_info
*npi
= skb
->dev
->npinfo
;
480 struct netpoll
*np
= npi
->rx_np
;
484 if (skb
->dev
->type
!= ARPHRD_ETHER
)
487 /* check if netpoll clients need ARP */
488 if (skb
->protocol
== htons(ETH_P_ARP
) &&
489 atomic_read(&trapped
)) {
490 skb_queue_tail(&npi
->arp_tx
, skb
);
494 proto
= ntohs(eth_hdr(skb
)->h_proto
);
495 if (proto
!= ETH_P_IP
)
497 if (skb
->pkt_type
== PACKET_OTHERHOST
)
502 iph
= (struct iphdr
*)skb
->data
;
503 if (!pskb_may_pull(skb
, sizeof(struct iphdr
)))
505 if (iph
->ihl
< 5 || iph
->version
!= 4)
507 if (!pskb_may_pull(skb
, iph
->ihl
*4))
509 if (ip_fast_csum((u8
*)iph
, iph
->ihl
) != 0)
512 len
= ntohs(iph
->tot_len
);
513 if (skb
->len
< len
|| len
< iph
->ihl
*4)
517 * Our transport medium may have padded the buffer out.
518 * Now We trim to the true length of the frame.
520 if (pskb_trim_rcsum(skb
, len
))
523 if (iph
->protocol
!= IPPROTO_UDP
)
527 uh
= (struct udphdr
*)(((char *)iph
) + iph
->ihl
*4);
528 ulen
= ntohs(uh
->len
);
532 if (checksum_udp(skb
, uh
, ulen
, iph
->saddr
, iph
->daddr
))
534 if (np
->local_ip
&& np
->local_ip
!= ntohl(iph
->daddr
))
536 if (np
->remote_ip
&& np
->remote_ip
!= ntohl(iph
->saddr
))
538 if (np
->local_port
&& np
->local_port
!= ntohs(uh
->dest
))
541 np
->rx_hook(np
, ntohs(uh
->source
),
543 ulen
- sizeof(struct udphdr
));
549 if (atomic_read(&trapped
)) {
557 void netpoll_print_options(struct netpoll
*np
)
559 printk(KERN_INFO
"%s: local port %d\n",
560 np
->name
, np
->local_port
);
561 printk(KERN_INFO
"%s: local IP %d.%d.%d.%d\n",
562 np
->name
, HIPQUAD(np
->local_ip
));
563 printk(KERN_INFO
"%s: interface %s\n",
564 np
->name
, np
->dev_name
);
565 printk(KERN_INFO
"%s: remote port %d\n",
566 np
->name
, np
->remote_port
);
567 printk(KERN_INFO
"%s: remote IP %d.%d.%d.%d\n",
568 np
->name
, HIPQUAD(np
->remote_ip
));
569 printk(KERN_INFO
"%s: remote ethernet address %pM\n",
570 np
->name
, np
->remote_mac
);
573 int netpoll_parse_options(struct netpoll
*np
, char *opt
)
575 char *cur
=opt
, *delim
;
578 if ((delim
= strchr(cur
, '@')) == NULL
)
581 np
->local_port
= simple_strtol(cur
, NULL
, 10);
587 if ((delim
= strchr(cur
, '/')) == NULL
)
590 np
->local_ip
= ntohl(in_aton(cur
));
596 /* parse out dev name */
597 if ((delim
= strchr(cur
, ',')) == NULL
)
600 strlcpy(np
->dev_name
, cur
, sizeof(np
->dev_name
));
607 if ((delim
= strchr(cur
, '@')) == NULL
)
610 np
->remote_port
= simple_strtol(cur
, NULL
, 10);
616 if ((delim
= strchr(cur
, '/')) == NULL
)
619 np
->remote_ip
= ntohl(in_aton(cur
));
624 if ((delim
= strchr(cur
, ':')) == NULL
)
627 np
->remote_mac
[0] = simple_strtol(cur
, NULL
, 16);
629 if ((delim
= strchr(cur
, ':')) == NULL
)
632 np
->remote_mac
[1] = simple_strtol(cur
, NULL
, 16);
634 if ((delim
= strchr(cur
, ':')) == NULL
)
637 np
->remote_mac
[2] = simple_strtol(cur
, NULL
, 16);
639 if ((delim
= strchr(cur
, ':')) == NULL
)
642 np
->remote_mac
[3] = simple_strtol(cur
, NULL
, 16);
644 if ((delim
= strchr(cur
, ':')) == NULL
)
647 np
->remote_mac
[4] = simple_strtol(cur
, NULL
, 16);
649 np
->remote_mac
[5] = simple_strtol(cur
, NULL
, 16);
652 netpoll_print_options(np
);
657 printk(KERN_INFO
"%s: couldn't parse config at %s!\n",
662 int netpoll_setup(struct netpoll
*np
)
664 struct net_device
*ndev
= NULL
;
665 struct in_device
*in_dev
;
666 struct netpoll_info
*npinfo
;
671 ndev
= dev_get_by_name(&init_net
, np
->dev_name
);
673 printk(KERN_ERR
"%s: %s doesn't exist, aborting.\n",
674 np
->name
, np
->dev_name
);
680 npinfo
= kmalloc(sizeof(*npinfo
), GFP_KERNEL
);
686 npinfo
->rx_flags
= 0;
687 npinfo
->rx_np
= NULL
;
689 spin_lock_init(&npinfo
->rx_lock
);
690 skb_queue_head_init(&npinfo
->arp_tx
);
691 skb_queue_head_init(&npinfo
->txq
);
692 INIT_DELAYED_WORK(&npinfo
->tx_work
, queue_process
);
694 atomic_set(&npinfo
->refcnt
, 1);
696 npinfo
= ndev
->npinfo
;
697 atomic_inc(&npinfo
->refcnt
);
700 if (!ndev
->netdev_ops
->ndo_poll_controller
) {
701 printk(KERN_ERR
"%s: %s doesn't support polling, aborting.\n",
702 np
->name
, np
->dev_name
);
707 if (!netif_running(ndev
)) {
708 unsigned long atmost
, atleast
;
710 printk(KERN_INFO
"%s: device %s not up yet, forcing it\n",
711 np
->name
, np
->dev_name
);
714 err
= dev_open(ndev
);
718 printk(KERN_ERR
"%s: failed to open %s\n",
719 np
->name
, ndev
->name
);
723 atleast
= jiffies
+ HZ
/10;
724 atmost
= jiffies
+ 4*HZ
;
725 while (!netif_carrier_ok(ndev
)) {
726 if (time_after(jiffies
, atmost
)) {
728 "%s: timeout waiting for carrier\n",
735 /* If carrier appears to come up instantly, we don't
736 * trust it and pause so that we don't pump all our
737 * queued console messages into the bitbucket.
740 if (time_before(jiffies
, atleast
)) {
741 printk(KERN_NOTICE
"%s: carrier detect appears"
742 " untrustworthy, waiting 4 seconds\n",
750 in_dev
= __in_dev_get_rcu(ndev
);
752 if (!in_dev
|| !in_dev
->ifa_list
) {
754 printk(KERN_ERR
"%s: no IP address for %s, aborting\n",
755 np
->name
, np
->dev_name
);
760 np
->local_ip
= ntohl(in_dev
->ifa_list
->ifa_local
);
762 printk(KERN_INFO
"%s: local IP %d.%d.%d.%d\n",
763 np
->name
, HIPQUAD(np
->local_ip
));
767 spin_lock_irqsave(&npinfo
->rx_lock
, flags
);
768 npinfo
->rx_flags
|= NETPOLL_RX_ENABLED
;
770 spin_unlock_irqrestore(&npinfo
->rx_lock
, flags
);
773 /* fill up the skb queue */
776 /* last thing to do is link it to the net device structure */
777 ndev
->npinfo
= npinfo
;
779 /* avoid racing with NAPI reading npinfo */
792 static int __init
netpoll_init(void)
794 skb_queue_head_init(&skb_pool
);
797 core_initcall(netpoll_init
);
799 void netpoll_cleanup(struct netpoll
*np
)
801 struct netpoll_info
*npinfo
;
805 npinfo
= np
->dev
->npinfo
;
807 if (npinfo
->rx_np
== np
) {
808 spin_lock_irqsave(&npinfo
->rx_lock
, flags
);
809 npinfo
->rx_np
= NULL
;
810 npinfo
->rx_flags
&= ~NETPOLL_RX_ENABLED
;
811 spin_unlock_irqrestore(&npinfo
->rx_lock
, flags
);
814 if (atomic_dec_and_test(&npinfo
->refcnt
)) {
815 skb_queue_purge(&npinfo
->arp_tx
);
816 skb_queue_purge(&npinfo
->txq
);
817 cancel_rearming_delayed_work(&npinfo
->tx_work
);
819 /* clean after last, unfinished work */
820 __skb_queue_purge(&npinfo
->txq
);
822 np
->dev
->npinfo
= NULL
;
832 int netpoll_trap(void)
834 return atomic_read(&trapped
);
837 void netpoll_set_trap(int trap
)
840 atomic_inc(&trapped
);
842 atomic_dec(&trapped
);
845 EXPORT_SYMBOL(netpoll_set_trap
);
846 EXPORT_SYMBOL(netpoll_trap
);
847 EXPORT_SYMBOL(netpoll_print_options
);
848 EXPORT_SYMBOL(netpoll_parse_options
);
849 EXPORT_SYMBOL(netpoll_setup
);
850 EXPORT_SYMBOL(netpoll_cleanup
);
851 EXPORT_SYMBOL(netpoll_send_udp
);
852 EXPORT_SYMBOL(netpoll_poll
);