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/moduleparam.h>
13 #include <linux/netdevice.h>
14 #include <linux/etherdevice.h>
15 #include <linux/string.h>
16 #include <linux/if_arp.h>
17 #include <linux/inetdevice.h>
18 #include <linux/inet.h>
19 #include <linux/interrupt.h>
20 #include <linux/netpoll.h>
21 #include <linux/sched.h>
22 #include <linux/delay.h>
23 #include <linux/rcupdate.h>
24 #include <linux/workqueue.h>
27 #include <asm/unaligned.h>
28 #include <trace/events/napi.h>
31 * We maintain a small pool of fully-sized skbs, to make sure the
32 * message gets out even in extreme OOM situations.
35 #define MAX_UDP_CHUNK 1460
37 #define MAX_QUEUE_DEPTH (MAX_SKBS / 2)
39 static struct sk_buff_head skb_pool
;
41 static atomic_t trapped
;
43 #define USEC_PER_POLL 50
44 #define NETPOLL_RX_ENABLED 1
45 #define NETPOLL_RX_DROP 2
47 #define MAX_SKB_SIZE \
48 (MAX_UDP_CHUNK + sizeof(struct udphdr) + \
49 sizeof(struct iphdr) + sizeof(struct ethhdr))
51 static void zap_completion_queue(void);
52 static void arp_reply(struct sk_buff
*skb
);
54 static unsigned int carrier_timeout
= 4;
55 module_param(carrier_timeout
, uint
, 0644);
57 static void queue_process(struct work_struct
*work
)
59 struct netpoll_info
*npinfo
=
60 container_of(work
, struct netpoll_info
, tx_work
.work
);
64 while ((skb
= skb_dequeue(&npinfo
->txq
))) {
65 struct net_device
*dev
= skb
->dev
;
66 const struct net_device_ops
*ops
= dev
->netdev_ops
;
67 struct netdev_queue
*txq
;
69 if (!netif_device_present(dev
) || !netif_running(dev
)) {
74 txq
= netdev_get_tx_queue(dev
, skb_get_queue_mapping(skb
));
76 local_irq_save(flags
);
77 __netif_tx_lock(txq
, smp_processor_id());
78 if (netif_tx_queue_stopped(txq
) ||
79 netif_tx_queue_frozen(txq
) ||
80 ops
->ndo_start_xmit(skb
, dev
) != NETDEV_TX_OK
) {
81 skb_queue_head(&npinfo
->txq
, skb
);
82 __netif_tx_unlock(txq
);
83 local_irq_restore(flags
);
85 schedule_delayed_work(&npinfo
->tx_work
, HZ
/10);
88 __netif_tx_unlock(txq
);
89 local_irq_restore(flags
);
93 static __sum16
checksum_udp(struct sk_buff
*skb
, struct udphdr
*uh
,
94 unsigned short ulen
, __be32 saddr
, __be32 daddr
)
98 if (uh
->check
== 0 || skb_csum_unnecessary(skb
))
101 psum
= csum_tcpudp_nofold(saddr
, daddr
, ulen
, IPPROTO_UDP
, 0);
103 if (skb
->ip_summed
== CHECKSUM_COMPLETE
&&
104 !csum_fold(csum_add(psum
, skb
->csum
)))
109 return __skb_checksum_complete(skb
);
113 * Check whether delayed processing was scheduled for our NIC. If so,
114 * we attempt to grab the poll lock and use ->poll() to pump the card.
115 * If this fails, either we've recursed in ->poll() or it's already
116 * running on another CPU.
118 * Note: we don't mask interrupts with this lock because we're using
119 * trylock here and interrupts are already disabled in the softirq
120 * case. Further, we test the poll_owner to avoid recursion on UP
121 * systems where the lock doesn't exist.
123 * In cases where there is bi-directional communications, reading only
124 * one message at a time can lead to packets being dropped by the
125 * network adapter, forcing superfluous retries and possibly timeouts.
126 * Thus, we set our budget to greater than 1.
128 static int poll_one_napi(struct netpoll_info
*npinfo
,
129 struct napi_struct
*napi
, int budget
)
133 /* net_rx_action's ->poll() invocations and our's are
134 * synchronized by this test which is only made while
135 * holding the napi->poll_lock.
137 if (!test_bit(NAPI_STATE_SCHED
, &napi
->state
))
140 npinfo
->rx_flags
|= NETPOLL_RX_DROP
;
141 atomic_inc(&trapped
);
142 set_bit(NAPI_STATE_NPSVC
, &napi
->state
);
144 work
= napi
->poll(napi
, budget
);
145 trace_napi_poll(napi
);
147 clear_bit(NAPI_STATE_NPSVC
, &napi
->state
);
148 atomic_dec(&trapped
);
149 npinfo
->rx_flags
&= ~NETPOLL_RX_DROP
;
151 return budget
- work
;
154 static void poll_napi(struct net_device
*dev
)
156 struct napi_struct
*napi
;
159 list_for_each_entry(napi
, &dev
->napi_list
, dev_list
) {
160 if (napi
->poll_owner
!= smp_processor_id() &&
161 spin_trylock(&napi
->poll_lock
)) {
162 budget
= poll_one_napi(dev
->npinfo
, napi
, budget
);
163 spin_unlock(&napi
->poll_lock
);
171 static void service_arp_queue(struct netpoll_info
*npi
)
176 while ((skb
= skb_dequeue(&npi
->arp_tx
)))
181 void netpoll_poll(struct netpoll
*np
)
183 struct net_device
*dev
= np
->dev
;
184 const struct net_device_ops
*ops
;
186 if (!dev
|| !netif_running(dev
))
189 ops
= dev
->netdev_ops
;
190 if (!ops
->ndo_poll_controller
)
193 /* Process pending work on NIC */
194 ops
->ndo_poll_controller(dev
);
198 service_arp_queue(dev
->npinfo
);
200 zap_completion_queue();
203 static void refill_skbs(void)
208 spin_lock_irqsave(&skb_pool
.lock
, flags
);
209 while (skb_pool
.qlen
< MAX_SKBS
) {
210 skb
= alloc_skb(MAX_SKB_SIZE
, GFP_ATOMIC
);
214 __skb_queue_tail(&skb_pool
, skb
);
216 spin_unlock_irqrestore(&skb_pool
.lock
, flags
);
219 static void zap_completion_queue(void)
222 struct softnet_data
*sd
= &get_cpu_var(softnet_data
);
224 if (sd
->completion_queue
) {
225 struct sk_buff
*clist
;
227 local_irq_save(flags
);
228 clist
= sd
->completion_queue
;
229 sd
->completion_queue
= NULL
;
230 local_irq_restore(flags
);
232 while (clist
!= NULL
) {
233 struct sk_buff
*skb
= clist
;
235 if (skb
->destructor
) {
236 atomic_inc(&skb
->users
);
237 dev_kfree_skb_any(skb
); /* put this one back */
244 put_cpu_var(softnet_data
);
247 static struct sk_buff
*find_skb(struct netpoll
*np
, int len
, int reserve
)
252 zap_completion_queue();
256 skb
= alloc_skb(len
, GFP_ATOMIC
);
258 skb
= skb_dequeue(&skb_pool
);
268 atomic_set(&skb
->users
, 1);
269 skb_reserve(skb
, reserve
);
273 static int netpoll_owner_active(struct net_device
*dev
)
275 struct napi_struct
*napi
;
277 list_for_each_entry(napi
, &dev
->napi_list
, dev_list
) {
278 if (napi
->poll_owner
== smp_processor_id())
284 static void netpoll_send_skb(struct netpoll
*np
, struct sk_buff
*skb
)
286 int status
= NETDEV_TX_BUSY
;
288 struct net_device
*dev
= np
->dev
;
289 const struct net_device_ops
*ops
= dev
->netdev_ops
;
290 struct netpoll_info
*npinfo
= np
->dev
->npinfo
;
292 if (!npinfo
|| !netif_running(dev
) || !netif_device_present(dev
)) {
297 /* don't get messages out of order, and no recursion */
298 if (skb_queue_len(&npinfo
->txq
) == 0 && !netpoll_owner_active(dev
)) {
299 struct netdev_queue
*txq
;
302 txq
= netdev_get_tx_queue(dev
, skb_get_queue_mapping(skb
));
304 local_irq_save(flags
);
305 /* try until next clock tick */
306 for (tries
= jiffies_to_usecs(1)/USEC_PER_POLL
;
307 tries
> 0; --tries
) {
308 if (__netif_tx_trylock(txq
)) {
309 if (!netif_tx_queue_stopped(txq
)) {
310 status
= ops
->ndo_start_xmit(skb
, dev
);
311 if (status
== NETDEV_TX_OK
)
312 txq_trans_update(txq
);
314 __netif_tx_unlock(txq
);
316 if (status
== NETDEV_TX_OK
)
321 /* tickle device maybe there is some cleanup */
324 udelay(USEC_PER_POLL
);
327 WARN_ONCE(!irqs_disabled(),
328 "netpoll_send_skb(): %s enabled interrupts in poll (%pF)\n",
329 dev
->name
, ops
->ndo_start_xmit
);
331 local_irq_restore(flags
);
334 if (status
!= NETDEV_TX_OK
) {
335 skb_queue_tail(&npinfo
->txq
, skb
);
336 schedule_delayed_work(&npinfo
->tx_work
,0);
340 void netpoll_send_udp(struct netpoll
*np
, const char *msg
, int len
)
342 int total_len
, eth_len
, ip_len
, udp_len
;
348 udp_len
= len
+ sizeof(*udph
);
349 ip_len
= eth_len
= udp_len
+ sizeof(*iph
);
350 total_len
= eth_len
+ ETH_HLEN
+ NET_IP_ALIGN
;
352 skb
= find_skb(np
, total_len
, total_len
- len
);
356 skb_copy_to_linear_data(skb
, msg
, len
);
359 skb_push(skb
, sizeof(*udph
));
360 skb_reset_transport_header(skb
);
362 udph
->source
= htons(np
->local_port
);
363 udph
->dest
= htons(np
->remote_port
);
364 udph
->len
= htons(udp_len
);
366 udph
->check
= csum_tcpudp_magic(np
->local_ip
,
368 udp_len
, IPPROTO_UDP
,
369 csum_partial(udph
, udp_len
, 0));
370 if (udph
->check
== 0)
371 udph
->check
= CSUM_MANGLED_0
;
373 skb_push(skb
, sizeof(*iph
));
374 skb_reset_network_header(skb
);
377 /* iph->version = 4; iph->ihl = 5; */
378 put_unaligned(0x45, (unsigned char *)iph
);
380 put_unaligned(htons(ip_len
), &(iph
->tot_len
));
384 iph
->protocol
= IPPROTO_UDP
;
386 put_unaligned(np
->local_ip
, &(iph
->saddr
));
387 put_unaligned(np
->remote_ip
, &(iph
->daddr
));
388 iph
->check
= ip_fast_csum((unsigned char *)iph
, iph
->ihl
);
390 eth
= (struct ethhdr
*) skb_push(skb
, ETH_HLEN
);
391 skb_reset_mac_header(skb
);
392 skb
->protocol
= eth
->h_proto
= htons(ETH_P_IP
);
393 memcpy(eth
->h_source
, np
->dev
->dev_addr
, ETH_ALEN
);
394 memcpy(eth
->h_dest
, np
->remote_mac
, ETH_ALEN
);
398 netpoll_send_skb(np
, skb
);
401 static void arp_reply(struct sk_buff
*skb
)
403 struct netpoll_info
*npinfo
= skb
->dev
->npinfo
;
405 unsigned char *arp_ptr
;
406 int size
, type
= ARPOP_REPLY
, ptype
= ETH_P_ARP
;
409 struct sk_buff
*send_skb
;
410 struct netpoll
*np
= NULL
;
412 if (npinfo
->rx_np
&& npinfo
->rx_np
->dev
== skb
->dev
)
417 /* No arp on this interface */
418 if (skb
->dev
->flags
& IFF_NOARP
)
421 if (!pskb_may_pull(skb
, arp_hdr_len(skb
->dev
)))
424 skb_reset_network_header(skb
);
425 skb_reset_transport_header(skb
);
428 if ((arp
->ar_hrd
!= htons(ARPHRD_ETHER
) &&
429 arp
->ar_hrd
!= htons(ARPHRD_IEEE802
)) ||
430 arp
->ar_pro
!= htons(ETH_P_IP
) ||
431 arp
->ar_op
!= htons(ARPOP_REQUEST
))
434 arp_ptr
= (unsigned char *)(arp
+1);
435 /* save the location of the src hw addr */
437 arp_ptr
+= skb
->dev
->addr_len
;
438 memcpy(&sip
, arp_ptr
, 4);
440 /* if we actually cared about dst hw addr, it would get copied here */
441 arp_ptr
+= skb
->dev
->addr_len
;
442 memcpy(&tip
, arp_ptr
, 4);
444 /* Should we ignore arp? */
445 if (tip
!= np
->local_ip
||
446 ipv4_is_loopback(tip
) || ipv4_is_multicast(tip
))
449 size
= arp_hdr_len(skb
->dev
);
450 send_skb
= find_skb(np
, size
+ LL_ALLOCATED_SPACE(np
->dev
),
451 LL_RESERVED_SPACE(np
->dev
));
456 skb_reset_network_header(send_skb
);
457 arp
= (struct arphdr
*) skb_put(send_skb
, size
);
458 send_skb
->dev
= skb
->dev
;
459 send_skb
->protocol
= htons(ETH_P_ARP
);
461 /* Fill the device header for the ARP frame */
462 if (dev_hard_header(send_skb
, skb
->dev
, ptype
,
463 sha
, np
->dev
->dev_addr
,
464 send_skb
->len
) < 0) {
470 * Fill out the arp protocol part.
472 * we only support ethernet device type,
473 * which (according to RFC 1390) should always equal 1 (Ethernet).
476 arp
->ar_hrd
= htons(np
->dev
->type
);
477 arp
->ar_pro
= htons(ETH_P_IP
);
478 arp
->ar_hln
= np
->dev
->addr_len
;
480 arp
->ar_op
= htons(type
);
482 arp_ptr
=(unsigned char *)(arp
+ 1);
483 memcpy(arp_ptr
, np
->dev
->dev_addr
, np
->dev
->addr_len
);
484 arp_ptr
+= np
->dev
->addr_len
;
485 memcpy(arp_ptr
, &tip
, 4);
487 memcpy(arp_ptr
, sha
, np
->dev
->addr_len
);
488 arp_ptr
+= np
->dev
->addr_len
;
489 memcpy(arp_ptr
, &sip
, 4);
491 netpoll_send_skb(np
, send_skb
);
494 int __netpoll_rx(struct sk_buff
*skb
)
496 int proto
, len
, ulen
;
499 struct netpoll_info
*npi
= skb
->dev
->npinfo
;
500 struct netpoll
*np
= npi
->rx_np
;
504 if (skb
->dev
->type
!= ARPHRD_ETHER
)
507 /* check if netpoll clients need ARP */
508 if (skb
->protocol
== htons(ETH_P_ARP
) &&
509 atomic_read(&trapped
)) {
510 skb_queue_tail(&npi
->arp_tx
, skb
);
514 proto
= ntohs(eth_hdr(skb
)->h_proto
);
515 if (proto
!= ETH_P_IP
)
517 if (skb
->pkt_type
== PACKET_OTHERHOST
)
522 iph
= (struct iphdr
*)skb
->data
;
523 if (!pskb_may_pull(skb
, sizeof(struct iphdr
)))
525 if (iph
->ihl
< 5 || iph
->version
!= 4)
527 if (!pskb_may_pull(skb
, iph
->ihl
*4))
529 if (ip_fast_csum((u8
*)iph
, iph
->ihl
) != 0)
532 len
= ntohs(iph
->tot_len
);
533 if (skb
->len
< len
|| len
< iph
->ihl
*4)
537 * Our transport medium may have padded the buffer out.
538 * Now We trim to the true length of the frame.
540 if (pskb_trim_rcsum(skb
, len
))
543 if (iph
->protocol
!= IPPROTO_UDP
)
547 uh
= (struct udphdr
*)(((char *)iph
) + iph
->ihl
*4);
548 ulen
= ntohs(uh
->len
);
552 if (checksum_udp(skb
, uh
, ulen
, iph
->saddr
, iph
->daddr
))
554 if (np
->local_ip
&& np
->local_ip
!= iph
->daddr
)
556 if (np
->remote_ip
&& np
->remote_ip
!= iph
->saddr
)
558 if (np
->local_port
&& np
->local_port
!= ntohs(uh
->dest
))
561 np
->rx_hook(np
, ntohs(uh
->source
),
563 ulen
- sizeof(struct udphdr
));
569 if (atomic_read(&trapped
)) {
577 void netpoll_print_options(struct netpoll
*np
)
579 printk(KERN_INFO
"%s: local port %d\n",
580 np
->name
, np
->local_port
);
581 printk(KERN_INFO
"%s: local IP %pI4\n",
582 np
->name
, &np
->local_ip
);
583 printk(KERN_INFO
"%s: interface %s\n",
584 np
->name
, np
->dev_name
);
585 printk(KERN_INFO
"%s: remote port %d\n",
586 np
->name
, np
->remote_port
);
587 printk(KERN_INFO
"%s: remote IP %pI4\n",
588 np
->name
, &np
->remote_ip
);
589 printk(KERN_INFO
"%s: remote ethernet address %pM\n",
590 np
->name
, np
->remote_mac
);
593 int netpoll_parse_options(struct netpoll
*np
, char *opt
)
595 char *cur
=opt
, *delim
;
598 if ((delim
= strchr(cur
, '@')) == NULL
)
601 np
->local_port
= simple_strtol(cur
, NULL
, 10);
607 if ((delim
= strchr(cur
, '/')) == NULL
)
610 np
->local_ip
= in_aton(cur
);
616 /* parse out dev name */
617 if ((delim
= strchr(cur
, ',')) == NULL
)
620 strlcpy(np
->dev_name
, cur
, sizeof(np
->dev_name
));
627 if ((delim
= strchr(cur
, '@')) == NULL
)
630 np
->remote_port
= simple_strtol(cur
, NULL
, 10);
636 if ((delim
= strchr(cur
, '/')) == NULL
)
639 np
->remote_ip
= in_aton(cur
);
644 if ((delim
= strchr(cur
, ':')) == NULL
)
647 np
->remote_mac
[0] = simple_strtol(cur
, NULL
, 16);
649 if ((delim
= strchr(cur
, ':')) == NULL
)
652 np
->remote_mac
[1] = simple_strtol(cur
, NULL
, 16);
654 if ((delim
= strchr(cur
, ':')) == NULL
)
657 np
->remote_mac
[2] = simple_strtol(cur
, NULL
, 16);
659 if ((delim
= strchr(cur
, ':')) == NULL
)
662 np
->remote_mac
[3] = simple_strtol(cur
, NULL
, 16);
664 if ((delim
= strchr(cur
, ':')) == NULL
)
667 np
->remote_mac
[4] = simple_strtol(cur
, NULL
, 16);
669 np
->remote_mac
[5] = simple_strtol(cur
, NULL
, 16);
672 netpoll_print_options(np
);
677 printk(KERN_INFO
"%s: couldn't parse config at %s!\n",
682 int netpoll_setup(struct netpoll
*np
)
684 struct net_device
*ndev
= NULL
;
685 struct in_device
*in_dev
;
686 struct netpoll_info
*npinfo
;
691 ndev
= dev_get_by_name(&init_net
, np
->dev_name
);
693 printk(KERN_ERR
"%s: %s doesn't exist, aborting.\n",
694 np
->name
, np
->dev_name
);
700 npinfo
= kmalloc(sizeof(*npinfo
), GFP_KERNEL
);
706 npinfo
->rx_flags
= 0;
707 npinfo
->rx_np
= NULL
;
709 spin_lock_init(&npinfo
->rx_lock
);
710 skb_queue_head_init(&npinfo
->arp_tx
);
711 skb_queue_head_init(&npinfo
->txq
);
712 INIT_DELAYED_WORK(&npinfo
->tx_work
, queue_process
);
714 atomic_set(&npinfo
->refcnt
, 1);
716 npinfo
= ndev
->npinfo
;
717 atomic_inc(&npinfo
->refcnt
);
720 if (!ndev
->netdev_ops
->ndo_poll_controller
) {
721 printk(KERN_ERR
"%s: %s doesn't support polling, aborting.\n",
722 np
->name
, np
->dev_name
);
727 if (!netif_running(ndev
)) {
728 unsigned long atmost
, atleast
;
730 printk(KERN_INFO
"%s: device %s not up yet, forcing it\n",
731 np
->name
, np
->dev_name
);
734 err
= dev_open(ndev
);
738 printk(KERN_ERR
"%s: failed to open %s\n",
739 np
->name
, ndev
->name
);
743 atleast
= jiffies
+ HZ
/10;
744 atmost
= jiffies
+ carrier_timeout
* HZ
;
745 while (!netif_carrier_ok(ndev
)) {
746 if (time_after(jiffies
, atmost
)) {
748 "%s: timeout waiting for carrier\n",
755 /* If carrier appears to come up instantly, we don't
756 * trust it and pause so that we don't pump all our
757 * queued console messages into the bitbucket.
760 if (time_before(jiffies
, atleast
)) {
761 printk(KERN_NOTICE
"%s: carrier detect appears"
762 " untrustworthy, waiting 4 seconds\n",
770 in_dev
= __in_dev_get_rcu(ndev
);
772 if (!in_dev
|| !in_dev
->ifa_list
) {
774 printk(KERN_ERR
"%s: no IP address for %s, aborting\n",
775 np
->name
, np
->dev_name
);
780 np
->local_ip
= in_dev
->ifa_list
->ifa_local
;
782 printk(KERN_INFO
"%s: local IP %pI4\n", np
->name
, &np
->local_ip
);
786 spin_lock_irqsave(&npinfo
->rx_lock
, flags
);
787 npinfo
->rx_flags
|= NETPOLL_RX_ENABLED
;
789 spin_unlock_irqrestore(&npinfo
->rx_lock
, flags
);
792 /* fill up the skb queue */
795 /* last thing to do is link it to the net device structure */
796 ndev
->npinfo
= npinfo
;
798 /* avoid racing with NAPI reading npinfo */
811 static int __init
netpoll_init(void)
813 skb_queue_head_init(&skb_pool
);
816 core_initcall(netpoll_init
);
818 void netpoll_cleanup(struct netpoll
*np
)
820 struct netpoll_info
*npinfo
;
824 npinfo
= np
->dev
->npinfo
;
826 if (npinfo
->rx_np
== np
) {
827 spin_lock_irqsave(&npinfo
->rx_lock
, flags
);
828 npinfo
->rx_np
= NULL
;
829 npinfo
->rx_flags
&= ~NETPOLL_RX_ENABLED
;
830 spin_unlock_irqrestore(&npinfo
->rx_lock
, flags
);
833 if (atomic_dec_and_test(&npinfo
->refcnt
)) {
834 skb_queue_purge(&npinfo
->arp_tx
);
835 skb_queue_purge(&npinfo
->txq
);
836 cancel_rearming_delayed_work(&npinfo
->tx_work
);
838 /* clean after last, unfinished work */
839 __skb_queue_purge(&npinfo
->txq
);
841 np
->dev
->npinfo
= NULL
;
851 int netpoll_trap(void)
853 return atomic_read(&trapped
);
856 void netpoll_set_trap(int trap
)
859 atomic_inc(&trapped
);
861 atomic_dec(&trapped
);
864 EXPORT_SYMBOL(netpoll_set_trap
);
865 EXPORT_SYMBOL(netpoll_trap
);
866 EXPORT_SYMBOL(netpoll_print_options
);
867 EXPORT_SYMBOL(netpoll_parse_options
);
868 EXPORT_SYMBOL(netpoll_setup
);
869 EXPORT_SYMBOL(netpoll_cleanup
);
870 EXPORT_SYMBOL(netpoll_send_udp
);
871 EXPORT_SYMBOL(netpoll_poll
);