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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14 #include <linux/moduleparam.h>
15 #include <linux/netdevice.h>
16 #include <linux/etherdevice.h>
17 #include <linux/string.h>
18 #include <linux/if_arp.h>
19 #include <linux/inetdevice.h>
20 #include <linux/inet.h>
21 #include <linux/interrupt.h>
22 #include <linux/netpoll.h>
23 #include <linux/sched.h>
24 #include <linux/delay.h>
25 #include <linux/rcupdate.h>
26 #include <linux/workqueue.h>
27 #include <linux/slab.h>
28 #include <linux/export.h>
29 #include <linux/if_vlan.h>
32 #include <net/addrconf.h>
33 #include <net/ndisc.h>
34 #include <net/ip6_checksum.h>
35 #include <asm/unaligned.h>
36 #include <trace/events/napi.h>
39 * We maintain a small pool of fully-sized skbs, to make sure the
40 * message gets out even in extreme OOM situations.
43 #define MAX_UDP_CHUNK 1460
46 static struct sk_buff_head skb_pool;
48 static atomic_t trapped;
50 DEFINE_STATIC_SRCU(netpoll_srcu);
52 #define USEC_PER_POLL 50
53 #define NETPOLL_RX_ENABLED 1
54 #define NETPOLL_RX_DROP 2
56 #define MAX_SKB_SIZE \
57 (sizeof(struct ethhdr) + \
58 sizeof(struct iphdr) + \
59 sizeof(struct udphdr) + \
62 static void zap_completion_queue(void);
63 static void netpoll_neigh_reply(struct sk_buff *skb, struct netpoll_info *npinfo);
64 static void netpoll_async_cleanup(struct work_struct *work);
66 static unsigned int carrier_timeout = 4;
67 module_param(carrier_timeout, uint, 0644);
69 #define np_info(np, fmt, ...) \
70 pr_info("%s: " fmt, np->name, ##__VA_ARGS__)
71 #define np_err(np, fmt, ...) \
72 pr_err("%s: " fmt, np->name, ##__VA_ARGS__)
73 #define np_notice(np, fmt, ...) \
74 pr_notice("%s: " fmt, np->name, ##__VA_ARGS__)
76 static void queue_process(struct work_struct *work)
78 struct netpoll_info *npinfo =
79 container_of(work, struct netpoll_info, tx_work.work);
83 while ((skb = skb_dequeue(&npinfo->txq))) {
84 struct net_device *dev = skb->dev;
85 const struct net_device_ops *ops = dev->netdev_ops;
86 struct netdev_queue *txq;
88 if (!netif_device_present(dev) || !netif_running(dev)) {
93 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
95 local_irq_save(flags);
96 __netif_tx_lock(txq, smp_processor_id());
97 if (netif_xmit_frozen_or_stopped(txq) ||
98 ops->ndo_start_xmit(skb, dev) != NETDEV_TX_OK) {
99 skb_queue_head(&npinfo->txq, skb);
100 __netif_tx_unlock(txq);
101 local_irq_restore(flags);
103 schedule_delayed_work(&npinfo->tx_work, HZ/10);
106 __netif_tx_unlock(txq);
107 local_irq_restore(flags);
111 static __sum16 checksum_udp(struct sk_buff *skb, struct udphdr *uh,
112 unsigned short ulen, __be32 saddr, __be32 daddr)
116 if (uh->check == 0 || skb_csum_unnecessary(skb))
119 psum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0);
121 if (skb->ip_summed == CHECKSUM_COMPLETE &&
122 !csum_fold(csum_add(psum, skb->csum)))
127 return __skb_checksum_complete(skb);
131 * Check whether delayed processing was scheduled for our NIC. If so,
132 * we attempt to grab the poll lock and use ->poll() to pump the card.
133 * If this fails, either we've recursed in ->poll() or it's already
134 * running on another CPU.
136 * Note: we don't mask interrupts with this lock because we're using
137 * trylock here and interrupts are already disabled in the softirq
138 * case. Further, we test the poll_owner to avoid recursion on UP
139 * systems where the lock doesn't exist.
141 * In cases where there is bi-directional communications, reading only
142 * one message at a time can lead to packets being dropped by the
143 * network adapter, forcing superfluous retries and possibly timeouts.
144 * Thus, we set our budget to greater than 1.
146 static int poll_one_napi(struct netpoll_info *npinfo,
147 struct napi_struct *napi, int budget)
151 /* net_rx_action's ->poll() invocations and our's are
152 * synchronized by this test which is only made while
153 * holding the napi->poll_lock.
155 if (!test_bit(NAPI_STATE_SCHED, &napi->state))
158 npinfo->rx_flags |= NETPOLL_RX_DROP;
159 atomic_inc(&trapped);
160 set_bit(NAPI_STATE_NPSVC, &napi->state);
162 work = napi->poll(napi, budget);
163 trace_napi_poll(napi);
165 clear_bit(NAPI_STATE_NPSVC, &napi->state);
166 atomic_dec(&trapped);
167 npinfo->rx_flags &= ~NETPOLL_RX_DROP;
169 return budget - work;
172 static void poll_napi(struct net_device *dev)
174 struct napi_struct *napi;
177 list_for_each_entry(napi, &dev->napi_list, dev_list) {
178 if (napi->poll_owner != smp_processor_id() &&
179 spin_trylock(&napi->poll_lock)) {
180 budget = poll_one_napi(rcu_dereference_bh(dev->npinfo),
182 spin_unlock(&napi->poll_lock);
190 static void service_neigh_queue(struct netpoll_info *npi)
195 while ((skb = skb_dequeue(&npi->neigh_tx)))
196 netpoll_neigh_reply(skb, npi);
200 static void netpoll_poll_dev(struct net_device *dev)
202 const struct net_device_ops *ops;
203 struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
205 /* Don't do any rx activity if the dev_lock mutex is held
206 * the dev_open/close paths use this to block netpoll activity
207 * while changing device state
209 if (down_trylock(&ni->dev_lock))
212 if (!netif_running(dev)) {
217 ops = dev->netdev_ops;
218 if (!ops->ndo_poll_controller) {
223 /* Process pending work on NIC */
224 ops->ndo_poll_controller(dev);
230 if (dev->flags & IFF_SLAVE) {
232 struct net_device *bond_dev;
234 struct netpoll_info *bond_ni;
236 bond_dev = netdev_master_upper_dev_get_rcu(dev);
237 bond_ni = rcu_dereference_bh(bond_dev->npinfo);
238 while ((skb = skb_dequeue(&ni->neigh_tx))) {
240 skb_queue_tail(&bond_ni->neigh_tx, skb);
245 service_neigh_queue(ni);
247 zap_completion_queue();
250 int netpoll_rx_disable(struct net_device *dev)
252 struct netpoll_info *ni;
255 idx = srcu_read_lock(&netpoll_srcu);
256 ni = srcu_dereference(dev->npinfo, &netpoll_srcu);
259 srcu_read_unlock(&netpoll_srcu, idx);
262 EXPORT_SYMBOL(netpoll_rx_disable);
264 void netpoll_rx_enable(struct net_device *dev)
266 struct netpoll_info *ni;
268 ni = rcu_dereference(dev->npinfo);
273 EXPORT_SYMBOL(netpoll_rx_enable);
275 static void refill_skbs(void)
280 spin_lock_irqsave(&skb_pool.lock, flags);
281 while (skb_pool.qlen < MAX_SKBS) {
282 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
286 __skb_queue_tail(&skb_pool, skb);
288 spin_unlock_irqrestore(&skb_pool.lock, flags);
291 static void zap_completion_queue(void)
294 struct softnet_data *sd = &get_cpu_var(softnet_data);
296 if (sd->completion_queue) {
297 struct sk_buff *clist;
299 local_irq_save(flags);
300 clist = sd->completion_queue;
301 sd->completion_queue = NULL;
302 local_irq_restore(flags);
304 while (clist != NULL) {
305 struct sk_buff *skb = clist;
307 if (skb->destructor) {
308 atomic_inc(&skb->users);
309 dev_kfree_skb_any(skb); /* put this one back */
316 put_cpu_var(softnet_data);
319 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
324 zap_completion_queue();
328 skb = alloc_skb(len, GFP_ATOMIC);
330 skb = skb_dequeue(&skb_pool);
334 netpoll_poll_dev(np->dev);
340 atomic_set(&skb->users, 1);
341 skb_reserve(skb, reserve);
345 static int netpoll_owner_active(struct net_device *dev)
347 struct napi_struct *napi;
349 list_for_each_entry(napi, &dev->napi_list, dev_list) {
350 if (napi->poll_owner == smp_processor_id())
356 /* call with IRQ disabled */
357 void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb,
358 struct net_device *dev)
360 int status = NETDEV_TX_BUSY;
362 const struct net_device_ops *ops = dev->netdev_ops;
363 /* It is up to the caller to keep npinfo alive. */
364 struct netpoll_info *npinfo;
366 WARN_ON_ONCE(!irqs_disabled());
368 npinfo = rcu_dereference_bh(np->dev->npinfo);
369 if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
374 /* don't get messages out of order, and no recursion */
375 if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
376 struct netdev_queue *txq;
378 txq = netdev_pick_tx(dev, skb);
380 /* try until next clock tick */
381 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
382 tries > 0; --tries) {
383 if (__netif_tx_trylock(txq)) {
384 if (!netif_xmit_stopped(txq)) {
385 if (vlan_tx_tag_present(skb) &&
386 !vlan_hw_offload_capable(netif_skb_features(skb),
388 skb = __vlan_put_tag(skb, skb->vlan_proto, vlan_tx_tag_get(skb));
389 if (unlikely(!skb)) {
390 /* This is actually a packet drop, but we
391 * don't want the code at the end of this
392 * function to try and re-queue a NULL skb.
394 status = NETDEV_TX_OK;
400 status = ops->ndo_start_xmit(skb, dev);
401 if (status == NETDEV_TX_OK)
402 txq_trans_update(txq);
405 __netif_tx_unlock(txq);
407 if (status == NETDEV_TX_OK)
412 /* tickle device maybe there is some cleanup */
413 netpoll_poll_dev(np->dev);
415 udelay(USEC_PER_POLL);
418 WARN_ONCE(!irqs_disabled(),
419 "netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pF)\n",
420 dev->name, ops->ndo_start_xmit);
424 if (status != NETDEV_TX_OK) {
425 skb_queue_tail(&npinfo->txq, skb);
426 schedule_delayed_work(&npinfo->tx_work,0);
429 EXPORT_SYMBOL(netpoll_send_skb_on_dev);
431 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
433 int total_len, ip_len, udp_len;
438 static atomic_t ip_ident;
439 struct ipv6hdr *ip6h;
441 udp_len = len + sizeof(*udph);
443 ip_len = udp_len + sizeof(*ip6h);
445 ip_len = udp_len + sizeof(*iph);
447 total_len = ip_len + LL_RESERVED_SPACE(np->dev);
449 skb = find_skb(np, total_len + np->dev->needed_tailroom,
454 skb_copy_to_linear_data(skb, msg, len);
457 skb_push(skb, sizeof(*udph));
458 skb_reset_transport_header(skb);
460 udph->source = htons(np->local_port);
461 udph->dest = htons(np->remote_port);
462 udph->len = htons(udp_len);
466 udph->check = csum_ipv6_magic(&np->local_ip.in6,
468 udp_len, IPPROTO_UDP,
469 csum_partial(udph, udp_len, 0));
470 if (udph->check == 0)
471 udph->check = CSUM_MANGLED_0;
473 skb_push(skb, sizeof(*ip6h));
474 skb_reset_network_header(skb);
475 ip6h = ipv6_hdr(skb);
477 /* ip6h->version = 6; ip6h->priority = 0; */
478 put_unaligned(0x60, (unsigned char *)ip6h);
479 ip6h->flow_lbl[0] = 0;
480 ip6h->flow_lbl[1] = 0;
481 ip6h->flow_lbl[2] = 0;
483 ip6h->payload_len = htons(sizeof(struct udphdr) + len);
484 ip6h->nexthdr = IPPROTO_UDP;
485 ip6h->hop_limit = 32;
486 ip6h->saddr = np->local_ip.in6;
487 ip6h->daddr = np->remote_ip.in6;
489 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
490 skb_reset_mac_header(skb);
491 skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
494 udph->check = csum_tcpudp_magic(np->local_ip.ip,
496 udp_len, IPPROTO_UDP,
497 csum_partial(udph, udp_len, 0));
498 if (udph->check == 0)
499 udph->check = CSUM_MANGLED_0;
501 skb_push(skb, sizeof(*iph));
502 skb_reset_network_header(skb);
505 /* iph->version = 4; iph->ihl = 5; */
506 put_unaligned(0x45, (unsigned char *)iph);
508 put_unaligned(htons(ip_len), &(iph->tot_len));
509 iph->id = htons(atomic_inc_return(&ip_ident));
512 iph->protocol = IPPROTO_UDP;
514 put_unaligned(np->local_ip.ip, &(iph->saddr));
515 put_unaligned(np->remote_ip.ip, &(iph->daddr));
516 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
518 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
519 skb_reset_mac_header(skb);
520 skb->protocol = eth->h_proto = htons(ETH_P_IP);
523 memcpy(eth->h_source, np->dev->dev_addr, ETH_ALEN);
524 memcpy(eth->h_dest, np->remote_mac, ETH_ALEN);
528 netpoll_send_skb(np, skb);
530 EXPORT_SYMBOL(netpoll_send_udp);
532 static void netpoll_neigh_reply(struct sk_buff *skb, struct netpoll_info *npinfo)
534 int size, type = ARPOP_REPLY;
537 struct sk_buff *send_skb;
538 struct netpoll *np, *tmp;
543 if (list_empty(&npinfo->rx_np))
546 /* Before checking the packet, we do some early
547 inspection whether this is interesting at all */
548 spin_lock_irqsave(&npinfo->rx_lock, flags);
549 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
550 if (np->dev == skb->dev)
553 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
555 /* No netpoll struct is using this dev */
559 proto = ntohs(eth_hdr(skb)->h_proto);
560 if (proto == ETH_P_ARP) {
562 unsigned char *arp_ptr;
563 /* No arp on this interface */
564 if (skb->dev->flags & IFF_NOARP)
567 if (!pskb_may_pull(skb, arp_hdr_len(skb->dev)))
570 skb_reset_network_header(skb);
571 skb_reset_transport_header(skb);
574 if ((arp->ar_hrd != htons(ARPHRD_ETHER) &&
575 arp->ar_hrd != htons(ARPHRD_IEEE802)) ||
576 arp->ar_pro != htons(ETH_P_IP) ||
577 arp->ar_op != htons(ARPOP_REQUEST))
580 arp_ptr = (unsigned char *)(arp+1);
581 /* save the location of the src hw addr */
583 arp_ptr += skb->dev->addr_len;
584 memcpy(&sip, arp_ptr, 4);
586 /* If we actually cared about dst hw addr,
587 it would get copied here */
588 arp_ptr += skb->dev->addr_len;
589 memcpy(&tip, arp_ptr, 4);
591 /* Should we ignore arp? */
592 if (ipv4_is_loopback(tip) || ipv4_is_multicast(tip))
595 size = arp_hdr_len(skb->dev);
597 spin_lock_irqsave(&npinfo->rx_lock, flags);
598 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
599 if (tip != np->local_ip.ip)
602 hlen = LL_RESERVED_SPACE(np->dev);
603 tlen = np->dev->needed_tailroom;
604 send_skb = find_skb(np, size + hlen + tlen, hlen);
608 skb_reset_network_header(send_skb);
609 arp = (struct arphdr *) skb_put(send_skb, size);
610 send_skb->dev = skb->dev;
611 send_skb->protocol = htons(ETH_P_ARP);
613 /* Fill the device header for the ARP frame */
614 if (dev_hard_header(send_skb, skb->dev, ETH_P_ARP,
615 sha, np->dev->dev_addr,
616 send_skb->len) < 0) {
622 * Fill out the arp protocol part.
624 * we only support ethernet device type,
625 * which (according to RFC 1390) should
626 * always equal 1 (Ethernet).
629 arp->ar_hrd = htons(np->dev->type);
630 arp->ar_pro = htons(ETH_P_IP);
631 arp->ar_hln = np->dev->addr_len;
633 arp->ar_op = htons(type);
635 arp_ptr = (unsigned char *)(arp + 1);
636 memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
637 arp_ptr += np->dev->addr_len;
638 memcpy(arp_ptr, &tip, 4);
640 memcpy(arp_ptr, sha, np->dev->addr_len);
641 arp_ptr += np->dev->addr_len;
642 memcpy(arp_ptr, &sip, 4);
644 netpoll_send_skb(np, send_skb);
646 /* If there are several rx_hooks for the same address,
647 we're fine by sending a single reply */
650 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
651 } else if( proto == ETH_P_IPV6) {
652 #if IS_ENABLED(CONFIG_IPV6)
656 struct icmp6hdr *icmp6h;
657 const struct in6_addr *saddr;
658 const struct in6_addr *daddr;
659 struct inet6_dev *in6_dev = NULL;
660 struct in6_addr *target;
662 in6_dev = in6_dev_get(skb->dev);
663 if (!in6_dev || !in6_dev->cnf.accept_ra)
666 if (!pskb_may_pull(skb, skb->len))
669 msg = (struct nd_msg *)skb_transport_header(skb);
671 __skb_push(skb, skb->data - skb_transport_header(skb));
673 if (ipv6_hdr(skb)->hop_limit != 255)
675 if (msg->icmph.icmp6_code != 0)
677 if (msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
680 saddr = &ipv6_hdr(skb)->saddr;
681 daddr = &ipv6_hdr(skb)->daddr;
683 size = sizeof(struct icmp6hdr) + sizeof(struct in6_addr);
685 spin_lock_irqsave(&npinfo->rx_lock, flags);
686 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
687 if (!ipv6_addr_equal(daddr, &np->local_ip.in6))
690 hlen = LL_RESERVED_SPACE(np->dev);
691 tlen = np->dev->needed_tailroom;
692 send_skb = find_skb(np, size + hlen + tlen, hlen);
696 send_skb->protocol = htons(ETH_P_IPV6);
697 send_skb->dev = skb->dev;
699 skb_reset_network_header(send_skb);
700 skb_put(send_skb, sizeof(struct ipv6hdr));
701 hdr = ipv6_hdr(send_skb);
703 *(__be32*)hdr = htonl(0x60000000);
705 hdr->payload_len = htons(size);
706 hdr->nexthdr = IPPROTO_ICMPV6;
707 hdr->hop_limit = 255;
711 send_skb->transport_header = send_skb->tail;
712 skb_put(send_skb, size);
714 icmp6h = (struct icmp6hdr *)skb_transport_header(skb);
715 icmp6h->icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
716 icmp6h->icmp6_router = 0;
717 icmp6h->icmp6_solicited = 1;
718 target = (struct in6_addr *)(skb_transport_header(send_skb) + sizeof(struct icmp6hdr));
719 *target = msg->target;
720 icmp6h->icmp6_cksum = csum_ipv6_magic(saddr, daddr, size,
725 if (dev_hard_header(send_skb, skb->dev, ETH_P_IPV6,
726 lladdr, np->dev->dev_addr,
727 send_skb->len) < 0) {
732 netpoll_send_skb(np, send_skb);
734 /* If there are several rx_hooks for the same address,
735 we're fine by sending a single reply */
738 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
743 static bool pkt_is_ns(struct sk_buff *skb)
748 if (skb->protocol != htons(ETH_P_ARP))
750 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr) + sizeof(struct nd_msg)))
753 msg = (struct nd_msg *)skb_transport_header(skb);
754 __skb_push(skb, skb->data - skb_transport_header(skb));
757 if (hdr->nexthdr != IPPROTO_ICMPV6)
759 if (hdr->hop_limit != 255)
761 if (msg->icmph.icmp6_code != 0)
763 if (msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
769 int __netpoll_rx(struct sk_buff *skb, struct netpoll_info *npinfo)
771 int proto, len, ulen;
773 const struct iphdr *iph;
775 struct netpoll *np, *tmp;
777 if (list_empty(&npinfo->rx_np))
780 if (skb->dev->type != ARPHRD_ETHER)
783 /* check if netpoll clients need ARP */
784 if (skb->protocol == htons(ETH_P_ARP) && atomic_read(&trapped)) {
785 skb_queue_tail(&npinfo->neigh_tx, skb);
787 } else if (pkt_is_ns(skb) && atomic_read(&trapped)) {
788 skb_queue_tail(&npinfo->neigh_tx, skb);
792 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
793 skb = vlan_untag(skb);
798 proto = ntohs(eth_hdr(skb)->h_proto);
799 if (proto != ETH_P_IP && proto != ETH_P_IPV6)
801 if (skb->pkt_type == PACKET_OTHERHOST)
806 if (proto == ETH_P_IP) {
807 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
809 iph = (struct iphdr *)skb->data;
810 if (iph->ihl < 5 || iph->version != 4)
812 if (!pskb_may_pull(skb, iph->ihl*4))
814 iph = (struct iphdr *)skb->data;
815 if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
818 len = ntohs(iph->tot_len);
819 if (skb->len < len || len < iph->ihl*4)
823 * Our transport medium may have padded the buffer out.
824 * Now We trim to the true length of the frame.
826 if (pskb_trim_rcsum(skb, len))
829 iph = (struct iphdr *)skb->data;
830 if (iph->protocol != IPPROTO_UDP)
834 uh = (struct udphdr *)(((char *)iph) + iph->ihl*4);
835 ulen = ntohs(uh->len);
839 if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr))
841 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
842 if (np->local_ip.ip && np->local_ip.ip != iph->daddr)
844 if (np->remote_ip.ip && np->remote_ip.ip != iph->saddr)
846 if (np->local_port && np->local_port != ntohs(uh->dest))
849 np->rx_hook(np, ntohs(uh->source),
851 ulen - sizeof(struct udphdr));
855 #if IS_ENABLED(CONFIG_IPV6)
856 const struct ipv6hdr *ip6h;
858 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
860 ip6h = (struct ipv6hdr *)skb->data;
861 if (ip6h->version != 6)
863 len = ntohs(ip6h->payload_len);
866 if (len + sizeof(struct ipv6hdr) > skb->len)
868 if (pskb_trim_rcsum(skb, len + sizeof(struct ipv6hdr)))
870 ip6h = ipv6_hdr(skb);
871 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
874 ulen = ntohs(uh->len);
875 if (ulen != skb->len)
877 if (udp6_csum_init(skb, uh, IPPROTO_UDP))
879 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
880 if (!ipv6_addr_equal(&np->local_ip.in6, &ip6h->daddr))
882 if (!ipv6_addr_equal(&np->remote_ip.in6, &ip6h->saddr))
884 if (np->local_port && np->local_port != ntohs(uh->dest))
887 np->rx_hook(np, ntohs(uh->source),
889 ulen - sizeof(struct udphdr));
902 if (atomic_read(&trapped)) {
910 void netpoll_print_options(struct netpoll *np)
912 np_info(np, "local port %d\n", np->local_port);
914 np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
916 np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
917 np_info(np, "interface '%s'\n", np->dev_name);
918 np_info(np, "remote port %d\n", np->remote_port);
920 np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
922 np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
923 np_info(np, "remote ethernet address %pM\n", np->remote_mac);
925 EXPORT_SYMBOL(netpoll_print_options);
927 static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
931 if (!strchr(str, ':') &&
932 in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
936 if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
937 #if IS_ENABLED(CONFIG_IPV6)
947 int netpoll_parse_options(struct netpoll *np, char *opt)
949 char *cur=opt, *delim;
953 if ((delim = strchr(cur, '@')) == NULL)
956 if (kstrtou16(cur, 10, &np->local_port))
963 if ((delim = strchr(cur, '/')) == NULL)
966 ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
970 np->ipv6 = (bool)ipv6;
976 /* parse out dev name */
977 if ((delim = strchr(cur, ',')) == NULL)
980 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
987 if ((delim = strchr(cur, '@')) == NULL)
990 if (*cur == ' ' || *cur == '\t')
991 np_info(np, "warning: whitespace is not allowed\n");
992 if (kstrtou16(cur, 10, &np->remote_port))
999 if ((delim = strchr(cur, '/')) == NULL)
1002 ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
1005 else if (np->ipv6 != (bool)ipv6)
1008 np->ipv6 = (bool)ipv6;
1013 if (!mac_pton(cur, np->remote_mac))
1017 netpoll_print_options(np);
1022 np_info(np, "couldn't parse config at '%s'!\n", cur);
1025 EXPORT_SYMBOL(netpoll_parse_options);
1027 int __netpoll_setup(struct netpoll *np, struct net_device *ndev, gfp_t gfp)
1029 struct netpoll_info *npinfo;
1030 const struct net_device_ops *ops;
1031 unsigned long flags;
1035 strlcpy(np->dev_name, ndev->name, IFNAMSIZ);
1036 INIT_WORK(&np->cleanup_work, netpoll_async_cleanup);
1038 if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) ||
1039 !ndev->netdev_ops->ndo_poll_controller) {
1040 np_err(np, "%s doesn't support polling, aborting\n",
1046 if (!ndev->npinfo) {
1047 npinfo = kmalloc(sizeof(*npinfo), gfp);
1053 npinfo->rx_flags = 0;
1054 INIT_LIST_HEAD(&npinfo->rx_np);
1056 spin_lock_init(&npinfo->rx_lock);
1057 sema_init(&npinfo->dev_lock, 1);
1058 skb_queue_head_init(&npinfo->neigh_tx);
1059 skb_queue_head_init(&npinfo->txq);
1060 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
1062 atomic_set(&npinfo->refcnt, 1);
1064 ops = np->dev->netdev_ops;
1065 if (ops->ndo_netpoll_setup) {
1066 err = ops->ndo_netpoll_setup(ndev, npinfo, gfp);
1071 npinfo = rtnl_dereference(ndev->npinfo);
1072 atomic_inc(&npinfo->refcnt);
1075 npinfo->netpoll = np;
1078 spin_lock_irqsave(&npinfo->rx_lock, flags);
1079 npinfo->rx_flags |= NETPOLL_RX_ENABLED;
1080 list_add_tail(&np->rx, &npinfo->rx_np);
1081 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
1084 /* last thing to do is link it to the net device structure */
1085 rcu_assign_pointer(ndev->npinfo, npinfo);
1094 EXPORT_SYMBOL_GPL(__netpoll_setup);
1096 int netpoll_setup(struct netpoll *np)
1098 struct net_device *ndev = NULL;
1099 struct in_device *in_dev;
1104 struct net *net = current->nsproxy->net_ns;
1105 ndev = __dev_get_by_name(net, np->dev_name);
1108 np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
1114 if (netdev_master_upper_dev_get(ndev)) {
1115 np_err(np, "%s is a slave device, aborting\n", np->dev_name);
1120 if (!netif_running(ndev)) {
1121 unsigned long atmost, atleast;
1123 np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
1125 err = dev_open(ndev);
1128 np_err(np, "failed to open %s\n", ndev->name);
1133 atleast = jiffies + HZ/10;
1134 atmost = jiffies + carrier_timeout * HZ;
1135 while (!netif_carrier_ok(ndev)) {
1136 if (time_after(jiffies, atmost)) {
1137 np_notice(np, "timeout waiting for carrier\n");
1143 /* If carrier appears to come up instantly, we don't
1144 * trust it and pause so that we don't pump all our
1145 * queued console messages into the bitbucket.
1148 if (time_before(jiffies, atleast)) {
1149 np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
1155 if (!np->local_ip.ip) {
1157 in_dev = __in_dev_get_rtnl(ndev);
1159 if (!in_dev || !in_dev->ifa_list) {
1160 np_err(np, "no IP address for %s, aborting\n",
1162 err = -EDESTADDRREQ;
1166 np->local_ip.ip = in_dev->ifa_list->ifa_local;
1167 np_info(np, "local IP %pI4\n", &np->local_ip.ip);
1169 #if IS_ENABLED(CONFIG_IPV6)
1170 struct inet6_dev *idev;
1172 err = -EDESTADDRREQ;
1173 idev = __in6_dev_get(ndev);
1175 struct inet6_ifaddr *ifp;
1177 read_lock_bh(&idev->lock);
1178 list_for_each_entry(ifp, &idev->addr_list, if_list) {
1179 if (ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL)
1181 np->local_ip.in6 = ifp->addr;
1185 read_unlock_bh(&idev->lock);
1188 np_err(np, "no IPv6 address for %s, aborting\n",
1192 np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
1194 np_err(np, "IPv6 is not supported %s, aborting\n",
1202 /* fill up the skb queue */
1205 err = __netpoll_setup(np, ndev, GFP_KERNEL);
1218 EXPORT_SYMBOL(netpoll_setup);
1220 static int __init netpoll_init(void)
1222 skb_queue_head_init(&skb_pool);
1225 core_initcall(netpoll_init);
1227 static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
1229 struct netpoll_info *npinfo =
1230 container_of(rcu_head, struct netpoll_info, rcu);
1232 skb_queue_purge(&npinfo->neigh_tx);
1233 skb_queue_purge(&npinfo->txq);
1235 /* we can't call cancel_delayed_work_sync here, as we are in softirq */
1236 cancel_delayed_work(&npinfo->tx_work);
1238 /* clean after last, unfinished work */
1239 __skb_queue_purge(&npinfo->txq);
1240 /* now cancel it again */
1241 cancel_delayed_work(&npinfo->tx_work);
1245 void __netpoll_cleanup(struct netpoll *np)
1247 struct netpoll_info *npinfo;
1248 unsigned long flags;
1250 /* rtnl_dereference would be preferable here but
1251 * rcu_cleanup_netpoll path can put us in here safely without
1252 * holding the rtnl, so plain rcu_dereference it is
1254 npinfo = rtnl_dereference(np->dev->npinfo);
1258 if (!list_empty(&npinfo->rx_np)) {
1259 spin_lock_irqsave(&npinfo->rx_lock, flags);
1261 if (list_empty(&npinfo->rx_np))
1262 npinfo->rx_flags &= ~NETPOLL_RX_ENABLED;
1263 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
1266 synchronize_srcu(&netpoll_srcu);
1268 if (atomic_dec_and_test(&npinfo->refcnt)) {
1269 const struct net_device_ops *ops;
1271 ops = np->dev->netdev_ops;
1272 if (ops->ndo_netpoll_cleanup)
1273 ops->ndo_netpoll_cleanup(np->dev);
1275 rcu_assign_pointer(np->dev->npinfo, NULL);
1276 call_rcu_bh(&npinfo->rcu, rcu_cleanup_netpoll_info);
1279 EXPORT_SYMBOL_GPL(__netpoll_cleanup);
1281 static void netpoll_async_cleanup(struct work_struct *work)
1283 struct netpoll *np = container_of(work, struct netpoll, cleanup_work);
1286 __netpoll_cleanup(np);
1291 void __netpoll_free_async(struct netpoll *np)
1293 schedule_work(&np->cleanup_work);
1295 EXPORT_SYMBOL_GPL(__netpoll_free_async);
1297 void netpoll_cleanup(struct netpoll *np)
1302 __netpoll_cleanup(np);
1308 EXPORT_SYMBOL(netpoll_cleanup);
1310 int netpoll_trap(void)
1312 return atomic_read(&trapped);
1314 EXPORT_SYMBOL(netpoll_trap);
1316 void netpoll_set_trap(int trap)
1319 atomic_inc(&trapped);
1321 atomic_dec(&trapped);
1323 EXPORT_SYMBOL(netpoll_set_trap);