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>
27 #include <trace/napi.h>
30 * We maintain a small pool of fully-sized skbs, to make sure the
31 * message gets out even in extreme OOM situations.
34 #define MAX_UDP_CHUNK 1460
36 #define MAX_QUEUE_DEPTH (MAX_SKBS / 2)
38 static struct sk_buff_head skb_pool
;
40 static atomic_t trapped
;
42 #define USEC_PER_POLL 50
43 #define NETPOLL_RX_ENABLED 1
44 #define NETPOLL_RX_DROP 2
46 #define MAX_SKB_SIZE \
47 (MAX_UDP_CHUNK + sizeof(struct udphdr) + \
48 sizeof(struct iphdr) + sizeof(struct ethhdr))
50 static void zap_completion_queue(void);
51 static void arp_reply(struct sk_buff
*skb
);
53 static void queue_process(struct work_struct
*work
)
55 struct netpoll_info
*npinfo
=
56 container_of(work
, struct netpoll_info
, tx_work
.work
);
60 while ((skb
= skb_dequeue(&npinfo
->txq
))) {
61 struct net_device
*dev
= skb
->dev
;
62 const struct net_device_ops
*ops
= dev
->netdev_ops
;
63 struct netdev_queue
*txq
;
65 if (!netif_device_present(dev
) || !netif_running(dev
)) {
70 txq
= netdev_get_tx_queue(dev
, skb_get_queue_mapping(skb
));
72 local_irq_save(flags
);
73 __netif_tx_lock(txq
, smp_processor_id());
74 if (netif_tx_queue_stopped(txq
) ||
75 netif_tx_queue_frozen(txq
) ||
76 ops
->ndo_start_xmit(skb
, dev
) != NETDEV_TX_OK
) {
77 skb_queue_head(&npinfo
->txq
, skb
);
78 __netif_tx_unlock(txq
);
79 local_irq_restore(flags
);
81 schedule_delayed_work(&npinfo
->tx_work
, HZ
/10);
84 __netif_tx_unlock(txq
);
85 local_irq_restore(flags
);
89 static __sum16
checksum_udp(struct sk_buff
*skb
, struct udphdr
*uh
,
90 unsigned short ulen
, __be32 saddr
, __be32 daddr
)
94 if (uh
->check
== 0 || skb_csum_unnecessary(skb
))
97 psum
= csum_tcpudp_nofold(saddr
, daddr
, ulen
, IPPROTO_UDP
, 0);
99 if (skb
->ip_summed
== CHECKSUM_COMPLETE
&&
100 !csum_fold(csum_add(psum
, skb
->csum
)))
105 return __skb_checksum_complete(skb
);
109 * Check whether delayed processing was scheduled for our NIC. If so,
110 * we attempt to grab the poll lock and use ->poll() to pump the card.
111 * If this fails, either we've recursed in ->poll() or it's already
112 * running on another CPU.
114 * Note: we don't mask interrupts with this lock because we're using
115 * trylock here and interrupts are already disabled in the softirq
116 * case. Further, we test the poll_owner to avoid recursion on UP
117 * systems where the lock doesn't exist.
119 * In cases where there is bi-directional communications, reading only
120 * one message at a time can lead to packets being dropped by the
121 * network adapter, forcing superfluous retries and possibly timeouts.
122 * Thus, we set our budget to greater than 1.
124 static int poll_one_napi(struct netpoll_info
*npinfo
,
125 struct napi_struct
*napi
, int budget
)
129 /* net_rx_action's ->poll() invocations and our's are
130 * synchronized by this test which is only made while
131 * holding the napi->poll_lock.
133 if (!test_bit(NAPI_STATE_SCHED
, &napi
->state
))
136 npinfo
->rx_flags
|= NETPOLL_RX_DROP
;
137 atomic_inc(&trapped
);
138 set_bit(NAPI_STATE_NPSVC
, &napi
->state
);
140 work
= napi
->poll(napi
, budget
);
141 trace_napi_poll(napi
->dev
);
143 clear_bit(NAPI_STATE_NPSVC
, &napi
->state
);
144 atomic_dec(&trapped
);
145 npinfo
->rx_flags
&= ~NETPOLL_RX_DROP
;
147 return budget
- work
;
150 static void poll_napi(struct net_device
*dev
)
152 struct napi_struct
*napi
;
155 list_for_each_entry(napi
, &dev
->napi_list
, dev_list
) {
156 if (napi
->poll_owner
!= smp_processor_id() &&
157 spin_trylock(&napi
->poll_lock
)) {
158 budget
= poll_one_napi(dev
->npinfo
, napi
, budget
);
159 spin_unlock(&napi
->poll_lock
);
167 static void service_arp_queue(struct netpoll_info
*npi
)
172 while ((skb
= skb_dequeue(&npi
->arp_tx
)))
177 void netpoll_poll(struct netpoll
*np
)
179 struct net_device
*dev
= np
->dev
;
180 const struct net_device_ops
*ops
;
182 if (!dev
|| !netif_running(dev
))
185 ops
= dev
->netdev_ops
;
186 if (!ops
->ndo_poll_controller
)
189 /* Process pending work on NIC */
190 ops
->ndo_poll_controller(dev
);
194 service_arp_queue(dev
->npinfo
);
196 zap_completion_queue();
199 static void refill_skbs(void)
204 spin_lock_irqsave(&skb_pool
.lock
, flags
);
205 while (skb_pool
.qlen
< MAX_SKBS
) {
206 skb
= alloc_skb(MAX_SKB_SIZE
, GFP_ATOMIC
);
210 __skb_queue_tail(&skb_pool
, skb
);
212 spin_unlock_irqrestore(&skb_pool
.lock
, flags
);
215 static void zap_completion_queue(void)
218 struct softnet_data
*sd
= &get_cpu_var(softnet_data
);
220 if (sd
->completion_queue
) {
221 struct sk_buff
*clist
;
223 local_irq_save(flags
);
224 clist
= sd
->completion_queue
;
225 sd
->completion_queue
= NULL
;
226 local_irq_restore(flags
);
228 while (clist
!= NULL
) {
229 struct sk_buff
*skb
= clist
;
231 if (skb
->destructor
) {
232 atomic_inc(&skb
->users
);
233 dev_kfree_skb_any(skb
); /* put this one back */
240 put_cpu_var(softnet_data
);
243 static struct sk_buff
*find_skb(struct netpoll
*np
, int len
, int reserve
)
248 zap_completion_queue();
252 skb
= alloc_skb(len
, GFP_ATOMIC
);
254 skb
= skb_dequeue(&skb_pool
);
264 atomic_set(&skb
->users
, 1);
265 skb_reserve(skb
, reserve
);
269 static int netpoll_owner_active(struct net_device
*dev
)
271 struct napi_struct
*napi
;
273 list_for_each_entry(napi
, &dev
->napi_list
, dev_list
) {
274 if (napi
->poll_owner
== smp_processor_id())
280 static void netpoll_send_skb(struct netpoll
*np
, struct sk_buff
*skb
)
282 int status
= NETDEV_TX_BUSY
;
284 struct net_device
*dev
= np
->dev
;
285 const struct net_device_ops
*ops
= dev
->netdev_ops
;
286 struct netpoll_info
*npinfo
= np
->dev
->npinfo
;
288 if (!npinfo
|| !netif_running(dev
) || !netif_device_present(dev
)) {
293 /* don't get messages out of order, and no recursion */
294 if (skb_queue_len(&npinfo
->txq
) == 0 && !netpoll_owner_active(dev
)) {
295 struct netdev_queue
*txq
;
298 txq
= netdev_get_tx_queue(dev
, skb_get_queue_mapping(skb
));
300 local_irq_save(flags
);
301 /* try until next clock tick */
302 for (tries
= jiffies_to_usecs(1)/USEC_PER_POLL
;
303 tries
> 0; --tries
) {
304 if (__netif_tx_trylock(txq
)) {
305 if (!netif_tx_queue_stopped(txq
))
306 status
= ops
->ndo_start_xmit(skb
, dev
);
307 __netif_tx_unlock(txq
);
309 if (status
== NETDEV_TX_OK
)
314 /* tickle device maybe there is some cleanup */
317 udelay(USEC_PER_POLL
);
319 local_irq_restore(flags
);
322 if (status
!= NETDEV_TX_OK
) {
323 skb_queue_tail(&npinfo
->txq
, skb
);
324 schedule_delayed_work(&npinfo
->tx_work
,0);
328 void netpoll_send_udp(struct netpoll
*np
, const char *msg
, int len
)
330 int total_len
, eth_len
, ip_len
, udp_len
;
336 udp_len
= len
+ sizeof(*udph
);
337 ip_len
= eth_len
= udp_len
+ sizeof(*iph
);
338 total_len
= eth_len
+ ETH_HLEN
+ NET_IP_ALIGN
;
340 skb
= find_skb(np
, total_len
, total_len
- len
);
344 skb_copy_to_linear_data(skb
, msg
, len
);
347 skb_push(skb
, sizeof(*udph
));
348 skb_reset_transport_header(skb
);
350 udph
->source
= htons(np
->local_port
);
351 udph
->dest
= htons(np
->remote_port
);
352 udph
->len
= htons(udp_len
);
354 udph
->check
= csum_tcpudp_magic(np
->local_ip
,
356 udp_len
, IPPROTO_UDP
,
357 csum_partial(udph
, udp_len
, 0));
358 if (udph
->check
== 0)
359 udph
->check
= CSUM_MANGLED_0
;
361 skb_push(skb
, sizeof(*iph
));
362 skb_reset_network_header(skb
);
365 /* iph->version = 4; iph->ihl = 5; */
366 put_unaligned(0x45, (unsigned char *)iph
);
368 put_unaligned(htons(ip_len
), &(iph
->tot_len
));
372 iph
->protocol
= IPPROTO_UDP
;
374 put_unaligned(np
->local_ip
, &(iph
->saddr
));
375 put_unaligned(np
->remote_ip
, &(iph
->daddr
));
376 iph
->check
= ip_fast_csum((unsigned char *)iph
, iph
->ihl
);
378 eth
= (struct ethhdr
*) skb_push(skb
, ETH_HLEN
);
379 skb_reset_mac_header(skb
);
380 skb
->protocol
= eth
->h_proto
= htons(ETH_P_IP
);
381 memcpy(eth
->h_source
, np
->dev
->dev_addr
, ETH_ALEN
);
382 memcpy(eth
->h_dest
, np
->remote_mac
, ETH_ALEN
);
386 netpoll_send_skb(np
, skb
);
389 static void arp_reply(struct sk_buff
*skb
)
391 struct netpoll_info
*npinfo
= skb
->dev
->npinfo
;
393 unsigned char *arp_ptr
;
394 int size
, type
= ARPOP_REPLY
, ptype
= ETH_P_ARP
;
397 struct sk_buff
*send_skb
;
398 struct netpoll
*np
= NULL
;
400 if (npinfo
->rx_np
&& npinfo
->rx_np
->dev
== skb
->dev
)
405 /* No arp on this interface */
406 if (skb
->dev
->flags
& IFF_NOARP
)
409 if (!pskb_may_pull(skb
, arp_hdr_len(skb
->dev
)))
412 skb_reset_network_header(skb
);
413 skb_reset_transport_header(skb
);
416 if ((arp
->ar_hrd
!= htons(ARPHRD_ETHER
) &&
417 arp
->ar_hrd
!= htons(ARPHRD_IEEE802
)) ||
418 arp
->ar_pro
!= htons(ETH_P_IP
) ||
419 arp
->ar_op
!= htons(ARPOP_REQUEST
))
422 arp_ptr
= (unsigned char *)(arp
+1);
423 /* save the location of the src hw addr */
425 arp_ptr
+= skb
->dev
->addr_len
;
426 memcpy(&sip
, arp_ptr
, 4);
428 /* if we actually cared about dst hw addr, it would get copied here */
429 arp_ptr
+= skb
->dev
->addr_len
;
430 memcpy(&tip
, arp_ptr
, 4);
432 /* Should we ignore arp? */
433 if (tip
!= np
->local_ip
||
434 ipv4_is_loopback(tip
) || ipv4_is_multicast(tip
))
437 size
= arp_hdr_len(skb
->dev
);
438 send_skb
= find_skb(np
, size
+ LL_ALLOCATED_SPACE(np
->dev
),
439 LL_RESERVED_SPACE(np
->dev
));
444 skb_reset_network_header(send_skb
);
445 arp
= (struct arphdr
*) skb_put(send_skb
, size
);
446 send_skb
->dev
= skb
->dev
;
447 send_skb
->protocol
= htons(ETH_P_ARP
);
449 /* Fill the device header for the ARP frame */
450 if (dev_hard_header(send_skb
, skb
->dev
, ptype
,
451 sha
, np
->dev
->dev_addr
,
452 send_skb
->len
) < 0) {
458 * Fill out the arp protocol part.
460 * we only support ethernet device type,
461 * which (according to RFC 1390) should always equal 1 (Ethernet).
464 arp
->ar_hrd
= htons(np
->dev
->type
);
465 arp
->ar_pro
= htons(ETH_P_IP
);
466 arp
->ar_hln
= np
->dev
->addr_len
;
468 arp
->ar_op
= htons(type
);
470 arp_ptr
=(unsigned char *)(arp
+ 1);
471 memcpy(arp_ptr
, np
->dev
->dev_addr
, np
->dev
->addr_len
);
472 arp_ptr
+= np
->dev
->addr_len
;
473 memcpy(arp_ptr
, &tip
, 4);
475 memcpy(arp_ptr
, sha
, np
->dev
->addr_len
);
476 arp_ptr
+= np
->dev
->addr_len
;
477 memcpy(arp_ptr
, &sip
, 4);
479 netpoll_send_skb(np
, send_skb
);
482 int __netpoll_rx(struct sk_buff
*skb
)
484 int proto
, len
, ulen
;
487 struct netpoll_info
*npi
= skb
->dev
->npinfo
;
488 struct netpoll
*np
= npi
->rx_np
;
492 if (skb
->dev
->type
!= ARPHRD_ETHER
)
495 /* check if netpoll clients need ARP */
496 if (skb
->protocol
== htons(ETH_P_ARP
) &&
497 atomic_read(&trapped
)) {
498 skb_queue_tail(&npi
->arp_tx
, skb
);
502 proto
= ntohs(eth_hdr(skb
)->h_proto
);
503 if (proto
!= ETH_P_IP
)
505 if (skb
->pkt_type
== PACKET_OTHERHOST
)
510 iph
= (struct iphdr
*)skb
->data
;
511 if (!pskb_may_pull(skb
, sizeof(struct iphdr
)))
513 if (iph
->ihl
< 5 || iph
->version
!= 4)
515 if (!pskb_may_pull(skb
, iph
->ihl
*4))
517 if (ip_fast_csum((u8
*)iph
, iph
->ihl
) != 0)
520 len
= ntohs(iph
->tot_len
);
521 if (skb
->len
< len
|| len
< iph
->ihl
*4)
525 * Our transport medium may have padded the buffer out.
526 * Now We trim to the true length of the frame.
528 if (pskb_trim_rcsum(skb
, len
))
531 if (iph
->protocol
!= IPPROTO_UDP
)
535 uh
= (struct udphdr
*)(((char *)iph
) + iph
->ihl
*4);
536 ulen
= ntohs(uh
->len
);
540 if (checksum_udp(skb
, uh
, ulen
, iph
->saddr
, iph
->daddr
))
542 if (np
->local_ip
&& np
->local_ip
!= iph
->daddr
)
544 if (np
->remote_ip
&& np
->remote_ip
!= iph
->saddr
)
546 if (np
->local_port
&& np
->local_port
!= ntohs(uh
->dest
))
549 np
->rx_hook(np
, ntohs(uh
->source
),
551 ulen
- sizeof(struct udphdr
));
557 if (atomic_read(&trapped
)) {
565 void netpoll_print_options(struct netpoll
*np
)
567 printk(KERN_INFO
"%s: local port %d\n",
568 np
->name
, np
->local_port
);
569 printk(KERN_INFO
"%s: local IP %pI4\n",
570 np
->name
, &np
->local_ip
);
571 printk(KERN_INFO
"%s: interface %s\n",
572 np
->name
, np
->dev_name
);
573 printk(KERN_INFO
"%s: remote port %d\n",
574 np
->name
, np
->remote_port
);
575 printk(KERN_INFO
"%s: remote IP %pI4\n",
576 np
->name
, &np
->remote_ip
);
577 printk(KERN_INFO
"%s: remote ethernet address %pM\n",
578 np
->name
, np
->remote_mac
);
581 int netpoll_parse_options(struct netpoll
*np
, char *opt
)
583 char *cur
=opt
, *delim
;
586 if ((delim
= strchr(cur
, '@')) == NULL
)
589 np
->local_port
= simple_strtol(cur
, NULL
, 10);
595 if ((delim
= strchr(cur
, '/')) == NULL
)
598 np
->local_ip
= in_aton(cur
);
604 /* parse out dev name */
605 if ((delim
= strchr(cur
, ',')) == NULL
)
608 strlcpy(np
->dev_name
, cur
, sizeof(np
->dev_name
));
615 if ((delim
= strchr(cur
, '@')) == NULL
)
618 np
->remote_port
= simple_strtol(cur
, NULL
, 10);
624 if ((delim
= strchr(cur
, '/')) == NULL
)
627 np
->remote_ip
= in_aton(cur
);
632 if ((delim
= strchr(cur
, ':')) == NULL
)
635 np
->remote_mac
[0] = simple_strtol(cur
, NULL
, 16);
637 if ((delim
= strchr(cur
, ':')) == NULL
)
640 np
->remote_mac
[1] = simple_strtol(cur
, NULL
, 16);
642 if ((delim
= strchr(cur
, ':')) == NULL
)
645 np
->remote_mac
[2] = simple_strtol(cur
, NULL
, 16);
647 if ((delim
= strchr(cur
, ':')) == NULL
)
650 np
->remote_mac
[3] = simple_strtol(cur
, NULL
, 16);
652 if ((delim
= strchr(cur
, ':')) == NULL
)
655 np
->remote_mac
[4] = simple_strtol(cur
, NULL
, 16);
657 np
->remote_mac
[5] = simple_strtol(cur
, NULL
, 16);
660 netpoll_print_options(np
);
665 printk(KERN_INFO
"%s: couldn't parse config at %s!\n",
670 int netpoll_setup(struct netpoll
*np
)
672 struct net_device
*ndev
= NULL
;
673 struct in_device
*in_dev
;
674 struct netpoll_info
*npinfo
;
679 ndev
= dev_get_by_name(&init_net
, np
->dev_name
);
681 printk(KERN_ERR
"%s: %s doesn't exist, aborting.\n",
682 np
->name
, np
->dev_name
);
688 npinfo
= kmalloc(sizeof(*npinfo
), GFP_KERNEL
);
694 npinfo
->rx_flags
= 0;
695 npinfo
->rx_np
= NULL
;
697 spin_lock_init(&npinfo
->rx_lock
);
698 skb_queue_head_init(&npinfo
->arp_tx
);
699 skb_queue_head_init(&npinfo
->txq
);
700 INIT_DELAYED_WORK(&npinfo
->tx_work
, queue_process
);
702 atomic_set(&npinfo
->refcnt
, 1);
704 npinfo
= ndev
->npinfo
;
705 atomic_inc(&npinfo
->refcnt
);
708 if (!ndev
->netdev_ops
->ndo_poll_controller
) {
709 printk(KERN_ERR
"%s: %s doesn't support polling, aborting.\n",
710 np
->name
, np
->dev_name
);
715 if (!netif_running(ndev
)) {
716 unsigned long atmost
, atleast
;
718 printk(KERN_INFO
"%s: device %s not up yet, forcing it\n",
719 np
->name
, np
->dev_name
);
722 err
= dev_open(ndev
);
726 printk(KERN_ERR
"%s: failed to open %s\n",
727 np
->name
, ndev
->name
);
731 atleast
= jiffies
+ HZ
/10;
732 atmost
= jiffies
+ 4*HZ
;
733 while (!netif_carrier_ok(ndev
)) {
734 if (time_after(jiffies
, atmost
)) {
736 "%s: timeout waiting for carrier\n",
743 /* If carrier appears to come up instantly, we don't
744 * trust it and pause so that we don't pump all our
745 * queued console messages into the bitbucket.
748 if (time_before(jiffies
, atleast
)) {
749 printk(KERN_NOTICE
"%s: carrier detect appears"
750 " untrustworthy, waiting 4 seconds\n",
758 in_dev
= __in_dev_get_rcu(ndev
);
760 if (!in_dev
|| !in_dev
->ifa_list
) {
762 printk(KERN_ERR
"%s: no IP address for %s, aborting\n",
763 np
->name
, np
->dev_name
);
768 np
->local_ip
= in_dev
->ifa_list
->ifa_local
;
770 printk(KERN_INFO
"%s: local IP %pI4\n", np
->name
, &np
->local_ip
);
774 spin_lock_irqsave(&npinfo
->rx_lock
, flags
);
775 npinfo
->rx_flags
|= NETPOLL_RX_ENABLED
;
777 spin_unlock_irqrestore(&npinfo
->rx_lock
, flags
);
780 /* fill up the skb queue */
783 /* last thing to do is link it to the net device structure */
784 ndev
->npinfo
= npinfo
;
786 /* avoid racing with NAPI reading npinfo */
799 static int __init
netpoll_init(void)
801 skb_queue_head_init(&skb_pool
);
804 core_initcall(netpoll_init
);
806 void netpoll_cleanup(struct netpoll
*np
)
808 struct netpoll_info
*npinfo
;
812 npinfo
= np
->dev
->npinfo
;
814 if (npinfo
->rx_np
== np
) {
815 spin_lock_irqsave(&npinfo
->rx_lock
, flags
);
816 npinfo
->rx_np
= NULL
;
817 npinfo
->rx_flags
&= ~NETPOLL_RX_ENABLED
;
818 spin_unlock_irqrestore(&npinfo
->rx_lock
, flags
);
821 if (atomic_dec_and_test(&npinfo
->refcnt
)) {
822 skb_queue_purge(&npinfo
->arp_tx
);
823 skb_queue_purge(&npinfo
->txq
);
824 cancel_rearming_delayed_work(&npinfo
->tx_work
);
826 /* clean after last, unfinished work */
827 __skb_queue_purge(&npinfo
->txq
);
829 np
->dev
->npinfo
= NULL
;
839 int netpoll_trap(void)
841 return atomic_read(&trapped
);
844 void netpoll_set_trap(int trap
)
847 atomic_inc(&trapped
);
849 atomic_dec(&trapped
);
852 EXPORT_SYMBOL(netpoll_set_trap
);
853 EXPORT_SYMBOL(netpoll_trap
);
854 EXPORT_SYMBOL(netpoll_print_options
);
855 EXPORT_SYMBOL(netpoll_parse_options
);
856 EXPORT_SYMBOL(netpoll_setup
);
857 EXPORT_SYMBOL(netpoll_cleanup
);
858 EXPORT_SYMBOL(netpoll_send_udp
);
859 EXPORT_SYMBOL(netpoll_poll
);