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
43 #define MAX_SKB_SIZE \
44 (MAX_UDP_CHUNK + sizeof(struct udphdr) + \
45 sizeof(struct iphdr) + sizeof(struct ethhdr))
47 static void zap_completion_queue(void);
48 static void arp_reply(struct sk_buff
*skb
);
50 static void queue_process(struct work_struct
*work
)
52 struct netpoll_info
*npinfo
=
53 container_of(work
, struct netpoll_info
, tx_work
.work
);
57 while ((skb
= skb_dequeue(&npinfo
->txq
))) {
58 struct net_device
*dev
= skb
->dev
;
60 if (!netif_device_present(dev
) || !netif_running(dev
)) {
65 local_irq_save(flags
);
67 if ((netif_queue_stopped(dev
) ||
68 netif_subqueue_stopped(dev
, skb
)) ||
69 dev
->hard_start_xmit(skb
, dev
) != NETDEV_TX_OK
) {
70 skb_queue_head(&npinfo
->txq
, skb
);
72 local_irq_restore(flags
);
74 schedule_delayed_work(&npinfo
->tx_work
, HZ
/10);
78 local_irq_restore(flags
);
82 static __sum16
checksum_udp(struct sk_buff
*skb
, struct udphdr
*uh
,
83 unsigned short ulen
, __be32 saddr
, __be32 daddr
)
87 if (uh
->check
== 0 || skb_csum_unnecessary(skb
))
90 psum
= csum_tcpudp_nofold(saddr
, daddr
, ulen
, IPPROTO_UDP
, 0);
92 if (skb
->ip_summed
== CHECKSUM_COMPLETE
&&
93 !csum_fold(csum_add(psum
, skb
->csum
)))
98 return __skb_checksum_complete(skb
);
102 * Check whether delayed processing was scheduled for our NIC. If so,
103 * we attempt to grab the poll lock and use ->poll() to pump the card.
104 * If this fails, either we've recursed in ->poll() or it's already
105 * running on another CPU.
107 * Note: we don't mask interrupts with this lock because we're using
108 * trylock here and interrupts are already disabled in the softirq
109 * case. Further, we test the poll_owner to avoid recursion on UP
110 * systems where the lock doesn't exist.
112 * In cases where there is bi-directional communications, reading only
113 * one message at a time can lead to packets being dropped by the
114 * network adapter, forcing superfluous retries and possibly timeouts.
115 * Thus, we set our budget to greater than 1.
117 static int poll_one_napi(struct netpoll_info
*npinfo
,
118 struct napi_struct
*napi
, int budget
)
122 /* net_rx_action's ->poll() invocations and our's are
123 * synchronized by this test which is only made while
124 * holding the napi->poll_lock.
126 if (!test_bit(NAPI_STATE_SCHED
, &napi
->state
))
129 atomic_inc(&trapped
);
131 work
= napi
->poll(napi
, budget
);
133 atomic_dec(&trapped
);
135 return budget
- work
;
138 static void poll_napi(struct net_device
*dev
)
140 struct napi_struct
*napi
;
143 list_for_each_entry(napi
, &dev
->napi_list
, dev_list
) {
144 if (napi
->poll_owner
!= smp_processor_id() &&
145 spin_trylock(&napi
->poll_lock
)) {
146 budget
= poll_one_napi(dev
->npinfo
, napi
, budget
);
147 spin_unlock(&napi
->poll_lock
);
155 static void service_arp_queue(struct netpoll_info
*npi
)
160 while ((skb
= skb_dequeue(&npi
->arp_tx
)))
165 void netpoll_poll(struct netpoll
*np
)
167 struct net_device
*dev
= np
->dev
;
169 if (!dev
|| !netif_running(dev
) || !dev
->poll_controller
)
172 /* Process pending work on NIC */
173 dev
->poll_controller(dev
);
177 service_arp_queue(dev
->npinfo
);
179 zap_completion_queue();
182 static void refill_skbs(void)
187 spin_lock_irqsave(&skb_pool
.lock
, flags
);
188 while (skb_pool
.qlen
< MAX_SKBS
) {
189 skb
= alloc_skb(MAX_SKB_SIZE
, GFP_ATOMIC
);
193 __skb_queue_tail(&skb_pool
, skb
);
195 spin_unlock_irqrestore(&skb_pool
.lock
, flags
);
198 static void zap_completion_queue(void)
201 struct softnet_data
*sd
= &get_cpu_var(softnet_data
);
203 if (sd
->completion_queue
) {
204 struct sk_buff
*clist
;
206 local_irq_save(flags
);
207 clist
= sd
->completion_queue
;
208 sd
->completion_queue
= NULL
;
209 local_irq_restore(flags
);
211 while (clist
!= NULL
) {
212 struct sk_buff
*skb
= clist
;
215 dev_kfree_skb_any(skb
); /* put this one back */
221 put_cpu_var(softnet_data
);
224 static struct sk_buff
*find_skb(struct netpoll
*np
, int len
, int reserve
)
229 zap_completion_queue();
233 skb
= alloc_skb(len
, GFP_ATOMIC
);
235 skb
= skb_dequeue(&skb_pool
);
245 atomic_set(&skb
->users
, 1);
246 skb_reserve(skb
, reserve
);
250 static int netpoll_owner_active(struct net_device
*dev
)
252 struct napi_struct
*napi
;
254 list_for_each_entry(napi
, &dev
->napi_list
, dev_list
) {
255 if (napi
->poll_owner
== smp_processor_id())
261 static void netpoll_send_skb(struct netpoll
*np
, struct sk_buff
*skb
)
263 int status
= NETDEV_TX_BUSY
;
265 struct net_device
*dev
= np
->dev
;
266 struct netpoll_info
*npinfo
= np
->dev
->npinfo
;
268 if (!npinfo
|| !netif_running(dev
) || !netif_device_present(dev
)) {
273 /* don't get messages out of order, and no recursion */
274 if (skb_queue_len(&npinfo
->txq
) == 0 && !netpoll_owner_active(dev
)) {
277 local_irq_save(flags
);
278 /* try until next clock tick */
279 for (tries
= jiffies_to_usecs(1)/USEC_PER_POLL
;
280 tries
> 0; --tries
) {
281 if (netif_tx_trylock(dev
)) {
282 if (!netif_queue_stopped(dev
) &&
283 !netif_subqueue_stopped(dev
, skb
))
284 status
= dev
->hard_start_xmit(skb
, dev
);
285 netif_tx_unlock(dev
);
287 if (status
== NETDEV_TX_OK
)
292 /* tickle device maybe there is some cleanup */
295 udelay(USEC_PER_POLL
);
297 local_irq_restore(flags
);
300 if (status
!= NETDEV_TX_OK
) {
301 skb_queue_tail(&npinfo
->txq
, skb
);
302 schedule_delayed_work(&npinfo
->tx_work
,0);
306 void netpoll_send_udp(struct netpoll
*np
, const char *msg
, int len
)
308 int total_len
, eth_len
, ip_len
, udp_len
;
314 udp_len
= len
+ sizeof(*udph
);
315 ip_len
= eth_len
= udp_len
+ sizeof(*iph
);
316 total_len
= eth_len
+ ETH_HLEN
+ NET_IP_ALIGN
;
318 skb
= find_skb(np
, total_len
, total_len
- len
);
322 skb_copy_to_linear_data(skb
, msg
, len
);
325 skb_push(skb
, sizeof(*udph
));
326 skb_reset_transport_header(skb
);
328 udph
->source
= htons(np
->local_port
);
329 udph
->dest
= htons(np
->remote_port
);
330 udph
->len
= htons(udp_len
);
332 udph
->check
= csum_tcpudp_magic(htonl(np
->local_ip
),
333 htonl(np
->remote_ip
),
334 udp_len
, IPPROTO_UDP
,
335 csum_partial((unsigned char *)udph
, udp_len
, 0));
336 if (udph
->check
== 0)
337 udph
->check
= CSUM_MANGLED_0
;
339 skb_push(skb
, sizeof(*iph
));
340 skb_reset_network_header(skb
);
343 /* iph->version = 4; iph->ihl = 5; */
344 put_unaligned(0x45, (unsigned char *)iph
);
346 put_unaligned(htons(ip_len
), &(iph
->tot_len
));
350 iph
->protocol
= IPPROTO_UDP
;
352 put_unaligned(htonl(np
->local_ip
), &(iph
->saddr
));
353 put_unaligned(htonl(np
->remote_ip
), &(iph
->daddr
));
354 iph
->check
= ip_fast_csum((unsigned char *)iph
, iph
->ihl
);
356 eth
= (struct ethhdr
*) skb_push(skb
, ETH_HLEN
);
357 skb_reset_mac_header(skb
);
358 skb
->protocol
= eth
->h_proto
= htons(ETH_P_IP
);
359 memcpy(eth
->h_source
, np
->dev
->dev_addr
, ETH_ALEN
);
360 memcpy(eth
->h_dest
, np
->remote_mac
, ETH_ALEN
);
364 netpoll_send_skb(np
, skb
);
367 static void arp_reply(struct sk_buff
*skb
)
369 struct netpoll_info
*npinfo
= skb
->dev
->npinfo
;
371 unsigned char *arp_ptr
;
372 int size
, type
= ARPOP_REPLY
, ptype
= ETH_P_ARP
;
375 struct sk_buff
*send_skb
;
376 struct netpoll
*np
= NULL
;
378 if (npinfo
->rx_np
&& npinfo
->rx_np
->dev
== skb
->dev
)
383 /* No arp on this interface */
384 if (skb
->dev
->flags
& IFF_NOARP
)
387 if (!pskb_may_pull(skb
, arp_hdr_len(skb
->dev
)))
390 skb_reset_network_header(skb
);
391 skb_reset_transport_header(skb
);
394 if ((arp
->ar_hrd
!= htons(ARPHRD_ETHER
) &&
395 arp
->ar_hrd
!= htons(ARPHRD_IEEE802
)) ||
396 arp
->ar_pro
!= htons(ETH_P_IP
) ||
397 arp
->ar_op
!= htons(ARPOP_REQUEST
))
400 arp_ptr
= (unsigned char *)(arp
+1);
401 /* save the location of the src hw addr */
403 arp_ptr
+= skb
->dev
->addr_len
;
404 memcpy(&sip
, arp_ptr
, 4);
406 /* if we actually cared about dst hw addr, it would get copied here */
407 arp_ptr
+= skb
->dev
->addr_len
;
408 memcpy(&tip
, arp_ptr
, 4);
410 /* Should we ignore arp? */
411 if (tip
!= htonl(np
->local_ip
) ||
412 ipv4_is_loopback(tip
) || ipv4_is_multicast(tip
))
415 size
= arp_hdr_len(skb
->dev
);
416 send_skb
= find_skb(np
, size
+ LL_RESERVED_SPACE(np
->dev
),
417 LL_RESERVED_SPACE(np
->dev
));
422 skb_reset_network_header(send_skb
);
423 arp
= (struct arphdr
*) skb_put(send_skb
, size
);
424 send_skb
->dev
= skb
->dev
;
425 send_skb
->protocol
= htons(ETH_P_ARP
);
427 /* Fill the device header for the ARP frame */
428 if (dev_hard_header(send_skb
, skb
->dev
, ptype
,
429 sha
, np
->dev
->dev_addr
,
430 send_skb
->len
) < 0) {
436 * Fill out the arp protocol part.
438 * we only support ethernet device type,
439 * which (according to RFC 1390) should always equal 1 (Ethernet).
442 arp
->ar_hrd
= htons(np
->dev
->type
);
443 arp
->ar_pro
= htons(ETH_P_IP
);
444 arp
->ar_hln
= np
->dev
->addr_len
;
446 arp
->ar_op
= htons(type
);
448 arp_ptr
=(unsigned char *)(arp
+ 1);
449 memcpy(arp_ptr
, np
->dev
->dev_addr
, np
->dev
->addr_len
);
450 arp_ptr
+= np
->dev
->addr_len
;
451 memcpy(arp_ptr
, &tip
, 4);
453 memcpy(arp_ptr
, sha
, np
->dev
->addr_len
);
454 arp_ptr
+= np
->dev
->addr_len
;
455 memcpy(arp_ptr
, &sip
, 4);
457 netpoll_send_skb(np
, send_skb
);
460 int __netpoll_rx(struct sk_buff
*skb
)
462 int proto
, len
, ulen
;
465 struct netpoll_info
*npi
= skb
->dev
->npinfo
;
466 struct netpoll
*np
= npi
->rx_np
;
470 if (skb
->dev
->type
!= ARPHRD_ETHER
)
473 /* if receive ARP during middle of NAPI poll, then queue */
474 if (skb
->protocol
== htons(ETH_P_ARP
) &&
475 atomic_read(&trapped
)) {
476 skb_queue_tail(&npi
->arp_tx
, skb
);
480 proto
= ntohs(eth_hdr(skb
)->h_proto
);
481 if (proto
!= ETH_P_IP
)
483 if (skb
->pkt_type
== PACKET_OTHERHOST
)
488 iph
= (struct iphdr
*)skb
->data
;
489 if (!pskb_may_pull(skb
, sizeof(struct iphdr
)))
491 if (iph
->ihl
< 5 || iph
->version
!= 4)
493 if (!pskb_may_pull(skb
, iph
->ihl
*4))
495 if (ip_fast_csum((u8
*)iph
, iph
->ihl
) != 0)
498 len
= ntohs(iph
->tot_len
);
499 if (skb
->len
< len
|| len
< iph
->ihl
*4)
503 * Our transport medium may have padded the buffer out.
504 * Now We trim to the true length of the frame.
506 if (pskb_trim_rcsum(skb
, len
))
509 if (iph
->protocol
!= IPPROTO_UDP
)
513 uh
= (struct udphdr
*)(((char *)iph
) + iph
->ihl
*4);
514 ulen
= ntohs(uh
->len
);
518 if (checksum_udp(skb
, uh
, ulen
, iph
->saddr
, iph
->daddr
))
520 if (np
->local_ip
&& np
->local_ip
!= ntohl(iph
->daddr
))
522 if (np
->remote_ip
&& np
->remote_ip
!= ntohl(iph
->saddr
))
524 if (np
->local_port
&& np
->local_port
!= ntohs(uh
->dest
))
527 np
->rx_hook(np
, ntohs(uh
->source
),
529 ulen
- sizeof(struct udphdr
));
535 /* If packet received while already in poll then just
538 if (atomic_read(&trapped
)) {
546 void netpoll_print_options(struct netpoll
*np
)
548 DECLARE_MAC_BUF(mac
);
549 printk(KERN_INFO
"%s: local port %d\n",
550 np
->name
, np
->local_port
);
551 printk(KERN_INFO
"%s: local IP %d.%d.%d.%d\n",
552 np
->name
, HIPQUAD(np
->local_ip
));
553 printk(KERN_INFO
"%s: interface %s\n",
554 np
->name
, np
->dev_name
);
555 printk(KERN_INFO
"%s: remote port %d\n",
556 np
->name
, np
->remote_port
);
557 printk(KERN_INFO
"%s: remote IP %d.%d.%d.%d\n",
558 np
->name
, HIPQUAD(np
->remote_ip
));
559 printk(KERN_INFO
"%s: remote ethernet address %s\n",
560 np
->name
, print_mac(mac
, np
->remote_mac
));
563 int netpoll_parse_options(struct netpoll
*np
, char *opt
)
565 char *cur
=opt
, *delim
;
568 if ((delim
= strchr(cur
, '@')) == NULL
)
571 np
->local_port
= simple_strtol(cur
, NULL
, 10);
577 if ((delim
= strchr(cur
, '/')) == NULL
)
580 np
->local_ip
= ntohl(in_aton(cur
));
586 /* parse out dev name */
587 if ((delim
= strchr(cur
, ',')) == NULL
)
590 strlcpy(np
->dev_name
, cur
, sizeof(np
->dev_name
));
597 if ((delim
= strchr(cur
, '@')) == NULL
)
600 np
->remote_port
= simple_strtol(cur
, NULL
, 10);
606 if ((delim
= strchr(cur
, '/')) == NULL
)
609 np
->remote_ip
= ntohl(in_aton(cur
));
614 if ((delim
= strchr(cur
, ':')) == NULL
)
617 np
->remote_mac
[0] = simple_strtol(cur
, NULL
, 16);
619 if ((delim
= strchr(cur
, ':')) == NULL
)
622 np
->remote_mac
[1] = simple_strtol(cur
, NULL
, 16);
624 if ((delim
= strchr(cur
, ':')) == NULL
)
627 np
->remote_mac
[2] = simple_strtol(cur
, NULL
, 16);
629 if ((delim
= strchr(cur
, ':')) == NULL
)
632 np
->remote_mac
[3] = simple_strtol(cur
, NULL
, 16);
634 if ((delim
= strchr(cur
, ':')) == NULL
)
637 np
->remote_mac
[4] = simple_strtol(cur
, NULL
, 16);
639 np
->remote_mac
[5] = simple_strtol(cur
, NULL
, 16);
642 netpoll_print_options(np
);
647 printk(KERN_INFO
"%s: couldn't parse config at %s!\n",
652 int netpoll_setup(struct netpoll
*np
)
654 struct net_device
*ndev
= NULL
;
655 struct in_device
*in_dev
;
656 struct netpoll_info
*npinfo
;
661 ndev
= dev_get_by_name(&init_net
, np
->dev_name
);
663 printk(KERN_ERR
"%s: %s doesn't exist, aborting.\n",
664 np
->name
, np
->dev_name
);
670 npinfo
= kmalloc(sizeof(*npinfo
), GFP_KERNEL
);
676 npinfo
->rx_np
= NULL
;
678 spin_lock_init(&npinfo
->rx_lock
);
679 skb_queue_head_init(&npinfo
->arp_tx
);
680 skb_queue_head_init(&npinfo
->txq
);
681 INIT_DELAYED_WORK(&npinfo
->tx_work
, queue_process
);
683 atomic_set(&npinfo
->refcnt
, 1);
685 npinfo
= ndev
->npinfo
;
686 atomic_inc(&npinfo
->refcnt
);
689 if (!ndev
->poll_controller
) {
690 printk(KERN_ERR
"%s: %s doesn't support polling, aborting.\n",
691 np
->name
, np
->dev_name
);
696 if (!netif_running(ndev
)) {
697 unsigned long atmost
, atleast
;
699 printk(KERN_INFO
"%s: device %s not up yet, forcing it\n",
700 np
->name
, np
->dev_name
);
703 err
= dev_open(ndev
);
707 printk(KERN_ERR
"%s: failed to open %s\n",
708 np
->name
, ndev
->name
);
712 atleast
= jiffies
+ HZ
/10;
713 atmost
= jiffies
+ 4*HZ
;
714 while (!netif_carrier_ok(ndev
)) {
715 if (time_after(jiffies
, atmost
)) {
717 "%s: timeout waiting for carrier\n",
724 /* If carrier appears to come up instantly, we don't
725 * trust it and pause so that we don't pump all our
726 * queued console messages into the bitbucket.
729 if (time_before(jiffies
, atleast
)) {
730 printk(KERN_NOTICE
"%s: carrier detect appears"
731 " untrustworthy, waiting 4 seconds\n",
739 in_dev
= __in_dev_get_rcu(ndev
);
741 if (!in_dev
|| !in_dev
->ifa_list
) {
743 printk(KERN_ERR
"%s: no IP address for %s, aborting\n",
744 np
->name
, np
->dev_name
);
749 np
->local_ip
= ntohl(in_dev
->ifa_list
->ifa_local
);
751 printk(KERN_INFO
"%s: local IP %d.%d.%d.%d\n",
752 np
->name
, HIPQUAD(np
->local_ip
));
756 spin_lock_irqsave(&npinfo
->rx_lock
, flags
);
758 spin_unlock_irqrestore(&npinfo
->rx_lock
, flags
);
761 /* fill up the skb queue */
764 /* last thing to do is link it to the net device structure */
765 ndev
->npinfo
= npinfo
;
767 /* avoid racing with NAPI reading npinfo */
780 static int __init
netpoll_init(void)
782 skb_queue_head_init(&skb_pool
);
785 core_initcall(netpoll_init
);
787 void netpoll_cleanup(struct netpoll
*np
)
789 struct netpoll_info
*npinfo
;
793 npinfo
= np
->dev
->npinfo
;
795 if (npinfo
->rx_np
== np
) {
796 spin_lock_irqsave(&npinfo
->rx_lock
, flags
);
797 npinfo
->rx_np
= NULL
;
798 spin_unlock_irqrestore(&npinfo
->rx_lock
, flags
);
801 if (atomic_dec_and_test(&npinfo
->refcnt
)) {
802 skb_queue_purge(&npinfo
->arp_tx
);
803 skb_queue_purge(&npinfo
->txq
);
804 cancel_rearming_delayed_work(&npinfo
->tx_work
);
806 /* clean after last, unfinished work */
807 __skb_queue_purge(&npinfo
->txq
);
809 np
->dev
->npinfo
= NULL
;
819 int netpoll_trap(void)
821 return atomic_read(&trapped
);
824 void netpoll_set_trap(int trap
)
827 atomic_inc(&trapped
);
829 atomic_dec(&trapped
);
832 EXPORT_SYMBOL(netpoll_set_trap
);
833 EXPORT_SYMBOL(netpoll_trap
);
834 EXPORT_SYMBOL(netpoll_print_options
);
835 EXPORT_SYMBOL(netpoll_parse_options
);
836 EXPORT_SYMBOL(netpoll_setup
);
837 EXPORT_SYMBOL(netpoll_cleanup
);
838 EXPORT_SYMBOL(netpoll_send_udp
);
839 EXPORT_SYMBOL(netpoll_poll
);
This page took 0.050187 seconds and 5 git commands to generate.