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
);
137 set_bit(NAPI_STATE_NPSVC
, &napi
->state
);
139 work
= napi
->poll(napi
, budget
);
141 clear_bit(NAPI_STATE_NPSVC
, &napi
->state
);
142 atomic_dec(&trapped
);
143 npinfo
->rx_flags
&= ~NETPOLL_RX_DROP
;
145 return budget
- work
;
148 static void poll_napi(struct net_device
*dev
)
150 struct napi_struct
*napi
;
153 list_for_each_entry(napi
, &dev
->napi_list
, dev_list
) {
154 if (napi
->poll_owner
!= smp_processor_id() &&
155 spin_trylock(&napi
->poll_lock
)) {
156 budget
= poll_one_napi(dev
->npinfo
, napi
, budget
);
157 spin_unlock(&napi
->poll_lock
);
165 static void service_arp_queue(struct netpoll_info
*npi
)
170 while ((skb
= skb_dequeue(&npi
->arp_tx
)))
175 void netpoll_poll(struct netpoll
*np
)
177 struct net_device
*dev
= np
->dev
;
178 const struct net_device_ops
*ops
= dev
->netdev_ops
;
180 if (!dev
|| !netif_running(dev
) || !ops
->ndo_poll_controller
)
183 /* Process pending work on NIC */
184 ops
->ndo_poll_controller(dev
);
188 service_arp_queue(dev
->npinfo
);
190 zap_completion_queue();
193 static void refill_skbs(void)
198 spin_lock_irqsave(&skb_pool
.lock
, flags
);
199 while (skb_pool
.qlen
< MAX_SKBS
) {
200 skb
= alloc_skb(MAX_SKB_SIZE
, GFP_ATOMIC
);
204 __skb_queue_tail(&skb_pool
, skb
);
206 spin_unlock_irqrestore(&skb_pool
.lock
, flags
);
209 static void zap_completion_queue(void)
212 struct softnet_data
*sd
= &get_cpu_var(softnet_data
);
214 if (sd
->completion_queue
) {
215 struct sk_buff
*clist
;
217 local_irq_save(flags
);
218 clist
= sd
->completion_queue
;
219 sd
->completion_queue
= NULL
;
220 local_irq_restore(flags
);
222 while (clist
!= NULL
) {
223 struct sk_buff
*skb
= clist
;
225 if (skb
->destructor
) {
226 atomic_inc(&skb
->users
);
227 dev_kfree_skb_any(skb
); /* put this one back */
234 put_cpu_var(softnet_data
);
237 static struct sk_buff
*find_skb(struct netpoll
*np
, int len
, int reserve
)
242 zap_completion_queue();
246 skb
= alloc_skb(len
, GFP_ATOMIC
);
248 skb
= skb_dequeue(&skb_pool
);
258 atomic_set(&skb
->users
, 1);
259 skb_reserve(skb
, reserve
);
263 static int netpoll_owner_active(struct net_device
*dev
)
265 struct napi_struct
*napi
;
267 list_for_each_entry(napi
, &dev
->napi_list
, dev_list
) {
268 if (napi
->poll_owner
== smp_processor_id())
274 static void netpoll_send_skb(struct netpoll
*np
, struct sk_buff
*skb
)
276 int status
= NETDEV_TX_BUSY
;
278 struct net_device
*dev
= np
->dev
;
279 const struct net_device_ops
*ops
= dev
->netdev_ops
;
280 struct netpoll_info
*npinfo
= np
->dev
->npinfo
;
282 if (!npinfo
|| !netif_running(dev
) || !netif_device_present(dev
)) {
287 /* don't get messages out of order, and no recursion */
288 if (skb_queue_len(&npinfo
->txq
) == 0 && !netpoll_owner_active(dev
)) {
289 struct netdev_queue
*txq
;
292 txq
= netdev_get_tx_queue(dev
, skb_get_queue_mapping(skb
));
294 local_irq_save(flags
);
295 /* try until next clock tick */
296 for (tries
= jiffies_to_usecs(1)/USEC_PER_POLL
;
297 tries
> 0; --tries
) {
298 if (__netif_tx_trylock(txq
)) {
299 if (!netif_tx_queue_stopped(txq
))
300 status
= ops
->ndo_start_xmit(skb
, dev
);
301 __netif_tx_unlock(txq
);
303 if (status
== NETDEV_TX_OK
)
308 /* tickle device maybe there is some cleanup */
311 udelay(USEC_PER_POLL
);
313 local_irq_restore(flags
);
316 if (status
!= NETDEV_TX_OK
) {
317 skb_queue_tail(&npinfo
->txq
, skb
);
318 schedule_delayed_work(&npinfo
->tx_work
,0);
322 void netpoll_send_udp(struct netpoll
*np
, const char *msg
, int len
)
324 int total_len
, eth_len
, ip_len
, udp_len
;
330 udp_len
= len
+ sizeof(*udph
);
331 ip_len
= eth_len
= udp_len
+ sizeof(*iph
);
332 total_len
= eth_len
+ ETH_HLEN
+ NET_IP_ALIGN
;
334 skb
= find_skb(np
, total_len
, total_len
- len
);
338 skb_copy_to_linear_data(skb
, msg
, len
);
341 skb_push(skb
, sizeof(*udph
));
342 skb_reset_transport_header(skb
);
344 udph
->source
= htons(np
->local_port
);
345 udph
->dest
= htons(np
->remote_port
);
346 udph
->len
= htons(udp_len
);
348 udph
->check
= csum_tcpudp_magic(htonl(np
->local_ip
),
349 htonl(np
->remote_ip
),
350 udp_len
, IPPROTO_UDP
,
351 csum_partial(udph
, udp_len
, 0));
352 if (udph
->check
== 0)
353 udph
->check
= CSUM_MANGLED_0
;
355 skb_push(skb
, sizeof(*iph
));
356 skb_reset_network_header(skb
);
359 /* iph->version = 4; iph->ihl = 5; */
360 put_unaligned(0x45, (unsigned char *)iph
);
362 put_unaligned(htons(ip_len
), &(iph
->tot_len
));
366 iph
->protocol
= IPPROTO_UDP
;
368 put_unaligned(htonl(np
->local_ip
), &(iph
->saddr
));
369 put_unaligned(htonl(np
->remote_ip
), &(iph
->daddr
));
370 iph
->check
= ip_fast_csum((unsigned char *)iph
, iph
->ihl
);
372 eth
= (struct ethhdr
*) skb_push(skb
, ETH_HLEN
);
373 skb_reset_mac_header(skb
);
374 skb
->protocol
= eth
->h_proto
= htons(ETH_P_IP
);
375 memcpy(eth
->h_source
, np
->dev
->dev_addr
, ETH_ALEN
);
376 memcpy(eth
->h_dest
, np
->remote_mac
, ETH_ALEN
);
380 netpoll_send_skb(np
, skb
);
383 static void arp_reply(struct sk_buff
*skb
)
385 struct netpoll_info
*npinfo
= skb
->dev
->npinfo
;
387 unsigned char *arp_ptr
;
388 int size
, type
= ARPOP_REPLY
, ptype
= ETH_P_ARP
;
391 struct sk_buff
*send_skb
;
392 struct netpoll
*np
= NULL
;
394 if (npinfo
->rx_np
&& npinfo
->rx_np
->dev
== skb
->dev
)
399 /* No arp on this interface */
400 if (skb
->dev
->flags
& IFF_NOARP
)
403 if (!pskb_may_pull(skb
, arp_hdr_len(skb
->dev
)))
406 skb_reset_network_header(skb
);
407 skb_reset_transport_header(skb
);
410 if ((arp
->ar_hrd
!= htons(ARPHRD_ETHER
) &&
411 arp
->ar_hrd
!= htons(ARPHRD_IEEE802
)) ||
412 arp
->ar_pro
!= htons(ETH_P_IP
) ||
413 arp
->ar_op
!= htons(ARPOP_REQUEST
))
416 arp_ptr
= (unsigned char *)(arp
+1);
417 /* save the location of the src hw addr */
419 arp_ptr
+= skb
->dev
->addr_len
;
420 memcpy(&sip
, arp_ptr
, 4);
422 /* if we actually cared about dst hw addr, it would get copied here */
423 arp_ptr
+= skb
->dev
->addr_len
;
424 memcpy(&tip
, arp_ptr
, 4);
426 /* Should we ignore arp? */
427 if (tip
!= htonl(np
->local_ip
) ||
428 ipv4_is_loopback(tip
) || ipv4_is_multicast(tip
))
431 size
= arp_hdr_len(skb
->dev
);
432 send_skb
= find_skb(np
, size
+ LL_ALLOCATED_SPACE(np
->dev
),
433 LL_RESERVED_SPACE(np
->dev
));
438 skb_reset_network_header(send_skb
);
439 arp
= (struct arphdr
*) skb_put(send_skb
, size
);
440 send_skb
->dev
= skb
->dev
;
441 send_skb
->protocol
= htons(ETH_P_ARP
);
443 /* Fill the device header for the ARP frame */
444 if (dev_hard_header(send_skb
, skb
->dev
, ptype
,
445 sha
, np
->dev
->dev_addr
,
446 send_skb
->len
) < 0) {
452 * Fill out the arp protocol part.
454 * we only support ethernet device type,
455 * which (according to RFC 1390) should always equal 1 (Ethernet).
458 arp
->ar_hrd
= htons(np
->dev
->type
);
459 arp
->ar_pro
= htons(ETH_P_IP
);
460 arp
->ar_hln
= np
->dev
->addr_len
;
462 arp
->ar_op
= htons(type
);
464 arp_ptr
=(unsigned char *)(arp
+ 1);
465 memcpy(arp_ptr
, np
->dev
->dev_addr
, np
->dev
->addr_len
);
466 arp_ptr
+= np
->dev
->addr_len
;
467 memcpy(arp_ptr
, &tip
, 4);
469 memcpy(arp_ptr
, sha
, np
->dev
->addr_len
);
470 arp_ptr
+= np
->dev
->addr_len
;
471 memcpy(arp_ptr
, &sip
, 4);
473 netpoll_send_skb(np
, send_skb
);
476 int __netpoll_rx(struct sk_buff
*skb
)
478 int proto
, len
, ulen
;
481 struct netpoll_info
*npi
= skb
->dev
->npinfo
;
482 struct netpoll
*np
= npi
->rx_np
;
486 if (skb
->dev
->type
!= ARPHRD_ETHER
)
489 /* check if netpoll clients need ARP */
490 if (skb
->protocol
== htons(ETH_P_ARP
) &&
491 atomic_read(&trapped
)) {
492 skb_queue_tail(&npi
->arp_tx
, skb
);
496 proto
= ntohs(eth_hdr(skb
)->h_proto
);
497 if (proto
!= ETH_P_IP
)
499 if (skb
->pkt_type
== PACKET_OTHERHOST
)
504 iph
= (struct iphdr
*)skb
->data
;
505 if (!pskb_may_pull(skb
, sizeof(struct iphdr
)))
507 if (iph
->ihl
< 5 || iph
->version
!= 4)
509 if (!pskb_may_pull(skb
, iph
->ihl
*4))
511 if (ip_fast_csum((u8
*)iph
, iph
->ihl
) != 0)
514 len
= ntohs(iph
->tot_len
);
515 if (skb
->len
< len
|| len
< iph
->ihl
*4)
519 * Our transport medium may have padded the buffer out.
520 * Now We trim to the true length of the frame.
522 if (pskb_trim_rcsum(skb
, len
))
525 if (iph
->protocol
!= IPPROTO_UDP
)
529 uh
= (struct udphdr
*)(((char *)iph
) + iph
->ihl
*4);
530 ulen
= ntohs(uh
->len
);
534 if (checksum_udp(skb
, uh
, ulen
, iph
->saddr
, iph
->daddr
))
536 if (np
->local_ip
&& np
->local_ip
!= ntohl(iph
->daddr
))
538 if (np
->remote_ip
&& np
->remote_ip
!= ntohl(iph
->saddr
))
540 if (np
->local_port
&& np
->local_port
!= ntohs(uh
->dest
))
543 np
->rx_hook(np
, ntohs(uh
->source
),
545 ulen
- sizeof(struct udphdr
));
551 if (atomic_read(&trapped
)) {
559 void netpoll_print_options(struct netpoll
*np
)
561 printk(KERN_INFO
"%s: local port %d\n",
562 np
->name
, np
->local_port
);
563 printk(KERN_INFO
"%s: local IP %d.%d.%d.%d\n",
564 np
->name
, HIPQUAD(np
->local_ip
));
565 printk(KERN_INFO
"%s: interface %s\n",
566 np
->name
, np
->dev_name
);
567 printk(KERN_INFO
"%s: remote port %d\n",
568 np
->name
, np
->remote_port
);
569 printk(KERN_INFO
"%s: remote IP %d.%d.%d.%d\n",
570 np
->name
, HIPQUAD(np
->remote_ip
));
571 printk(KERN_INFO
"%s: remote ethernet address %pM\n",
572 np
->name
, np
->remote_mac
);
575 int netpoll_parse_options(struct netpoll
*np
, char *opt
)
577 char *cur
=opt
, *delim
;
580 if ((delim
= strchr(cur
, '@')) == NULL
)
583 np
->local_port
= simple_strtol(cur
, NULL
, 10);
589 if ((delim
= strchr(cur
, '/')) == NULL
)
592 np
->local_ip
= ntohl(in_aton(cur
));
598 /* parse out dev name */
599 if ((delim
= strchr(cur
, ',')) == NULL
)
602 strlcpy(np
->dev_name
, cur
, sizeof(np
->dev_name
));
609 if ((delim
= strchr(cur
, '@')) == NULL
)
612 np
->remote_port
= simple_strtol(cur
, NULL
, 10);
618 if ((delim
= strchr(cur
, '/')) == NULL
)
621 np
->remote_ip
= ntohl(in_aton(cur
));
626 if ((delim
= strchr(cur
, ':')) == NULL
)
629 np
->remote_mac
[0] = simple_strtol(cur
, NULL
, 16);
631 if ((delim
= strchr(cur
, ':')) == NULL
)
634 np
->remote_mac
[1] = simple_strtol(cur
, NULL
, 16);
636 if ((delim
= strchr(cur
, ':')) == NULL
)
639 np
->remote_mac
[2] = simple_strtol(cur
, NULL
, 16);
641 if ((delim
= strchr(cur
, ':')) == NULL
)
644 np
->remote_mac
[3] = simple_strtol(cur
, NULL
, 16);
646 if ((delim
= strchr(cur
, ':')) == NULL
)
649 np
->remote_mac
[4] = simple_strtol(cur
, NULL
, 16);
651 np
->remote_mac
[5] = simple_strtol(cur
, NULL
, 16);
654 netpoll_print_options(np
);
659 printk(KERN_INFO
"%s: couldn't parse config at %s!\n",
664 int netpoll_setup(struct netpoll
*np
)
666 struct net_device
*ndev
= NULL
;
667 struct in_device
*in_dev
;
668 struct netpoll_info
*npinfo
;
673 ndev
= dev_get_by_name(&init_net
, np
->dev_name
);
675 printk(KERN_ERR
"%s: %s doesn't exist, aborting.\n",
676 np
->name
, np
->dev_name
);
682 npinfo
= kmalloc(sizeof(*npinfo
), GFP_KERNEL
);
688 npinfo
->rx_flags
= 0;
689 npinfo
->rx_np
= NULL
;
691 spin_lock_init(&npinfo
->rx_lock
);
692 skb_queue_head_init(&npinfo
->arp_tx
);
693 skb_queue_head_init(&npinfo
->txq
);
694 INIT_DELAYED_WORK(&npinfo
->tx_work
, queue_process
);
696 atomic_set(&npinfo
->refcnt
, 1);
698 npinfo
= ndev
->npinfo
;
699 atomic_inc(&npinfo
->refcnt
);
702 if (!ndev
->netdev_ops
->ndo_poll_controller
) {
703 printk(KERN_ERR
"%s: %s doesn't support polling, aborting.\n",
704 np
->name
, np
->dev_name
);
709 if (!netif_running(ndev
)) {
710 unsigned long atmost
, atleast
;
712 printk(KERN_INFO
"%s: device %s not up yet, forcing it\n",
713 np
->name
, np
->dev_name
);
716 err
= dev_open(ndev
);
720 printk(KERN_ERR
"%s: failed to open %s\n",
721 np
->name
, ndev
->name
);
725 atleast
= jiffies
+ HZ
/10;
726 atmost
= jiffies
+ 4*HZ
;
727 while (!netif_carrier_ok(ndev
)) {
728 if (time_after(jiffies
, atmost
)) {
730 "%s: timeout waiting for carrier\n",
737 /* If carrier appears to come up instantly, we don't
738 * trust it and pause so that we don't pump all our
739 * queued console messages into the bitbucket.
742 if (time_before(jiffies
, atleast
)) {
743 printk(KERN_NOTICE
"%s: carrier detect appears"
744 " untrustworthy, waiting 4 seconds\n",
752 in_dev
= __in_dev_get_rcu(ndev
);
754 if (!in_dev
|| !in_dev
->ifa_list
) {
756 printk(KERN_ERR
"%s: no IP address for %s, aborting\n",
757 np
->name
, np
->dev_name
);
762 np
->local_ip
= ntohl(in_dev
->ifa_list
->ifa_local
);
764 printk(KERN_INFO
"%s: local IP %d.%d.%d.%d\n",
765 np
->name
, HIPQUAD(np
->local_ip
));
769 spin_lock_irqsave(&npinfo
->rx_lock
, flags
);
770 npinfo
->rx_flags
|= NETPOLL_RX_ENABLED
;
772 spin_unlock_irqrestore(&npinfo
->rx_lock
, flags
);
775 /* fill up the skb queue */
778 /* last thing to do is link it to the net device structure */
779 ndev
->npinfo
= npinfo
;
781 /* avoid racing with NAPI reading npinfo */
794 static int __init
netpoll_init(void)
796 skb_queue_head_init(&skb_pool
);
799 core_initcall(netpoll_init
);
801 void netpoll_cleanup(struct netpoll
*np
)
803 struct netpoll_info
*npinfo
;
807 npinfo
= np
->dev
->npinfo
;
809 if (npinfo
->rx_np
== np
) {
810 spin_lock_irqsave(&npinfo
->rx_lock
, flags
);
811 npinfo
->rx_np
= NULL
;
812 npinfo
->rx_flags
&= ~NETPOLL_RX_ENABLED
;
813 spin_unlock_irqrestore(&npinfo
->rx_lock
, flags
);
816 if (atomic_dec_and_test(&npinfo
->refcnt
)) {
817 skb_queue_purge(&npinfo
->arp_tx
);
818 skb_queue_purge(&npinfo
->txq
);
819 cancel_rearming_delayed_work(&npinfo
->tx_work
);
821 /* clean after last, unfinished work */
822 __skb_queue_purge(&npinfo
->txq
);
824 np
->dev
->npinfo
= NULL
;
834 int netpoll_trap(void)
836 return atomic_read(&trapped
);
839 void netpoll_set_trap(int trap
)
842 atomic_inc(&trapped
);
844 atomic_dec(&trapped
);
847 EXPORT_SYMBOL(netpoll_set_trap
);
848 EXPORT_SYMBOL(netpoll_trap
);
849 EXPORT_SYMBOL(netpoll_print_options
);
850 EXPORT_SYMBOL(netpoll_parse_options
);
851 EXPORT_SYMBOL(netpoll_setup
);
852 EXPORT_SYMBOL(netpoll_cleanup
);
853 EXPORT_SYMBOL(netpoll_send_udp
);
854 EXPORT_SYMBOL(netpoll_poll
);