2 * Common framework for low-level network console, dump, and debugger code
6 * based on the netconsole code from:
9 * Copyright (C) 2002 Red Hat, Inc.
12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14 #include <linux/moduleparam.h>
15 #include <linux/kernel.h>
16 #include <linux/netdevice.h>
17 #include <linux/etherdevice.h>
18 #include <linux/string.h>
19 #include <linux/if_arp.h>
20 #include <linux/inetdevice.h>
21 #include <linux/inet.h>
22 #include <linux/interrupt.h>
23 #include <linux/netpoll.h>
24 #include <linux/sched.h>
25 #include <linux/delay.h>
26 #include <linux/rcupdate.h>
27 #include <linux/workqueue.h>
28 #include <linux/slab.h>
29 #include <linux/export.h>
30 #include <linux/if_vlan.h>
33 #include <net/addrconf.h>
34 #include <net/ndisc.h>
35 #include <net/ip6_checksum.h>
36 #include <asm/unaligned.h>
37 #include <trace/events/napi.h>
40 * We maintain a small pool of fully-sized skbs, to make sure the
41 * message gets out even in extreme OOM situations.
44 #define MAX_UDP_CHUNK 1460
47 static struct sk_buff_head skb_pool;
49 DEFINE_STATIC_SRCU(netpoll_srcu);
51 #define USEC_PER_POLL 50
53 #define MAX_SKB_SIZE \
54 (sizeof(struct ethhdr) + \
55 sizeof(struct iphdr) + \
56 sizeof(struct udphdr) + \
59 static void zap_completion_queue(void);
60 static void netpoll_async_cleanup(struct work_struct *work);
62 static unsigned int carrier_timeout = 4;
63 module_param(carrier_timeout, uint, 0644);
65 #define np_info(np, fmt, ...) \
66 pr_info("%s: " fmt, np->name, ##__VA_ARGS__)
67 #define np_err(np, fmt, ...) \
68 pr_err("%s: " fmt, np->name, ##__VA_ARGS__)
69 #define np_notice(np, fmt, ...) \
70 pr_notice("%s: " fmt, np->name, ##__VA_ARGS__)
72 static int netpoll_start_xmit(struct sk_buff *skb, struct net_device *dev,
73 struct netdev_queue *txq)
75 int status = NETDEV_TX_OK;
76 netdev_features_t features;
78 features = netif_skb_features(skb);
80 if (skb_vlan_tag_present(skb) &&
81 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
82 skb = __vlan_hwaccel_push_inside(skb);
84 /* This is actually a packet drop, but we
85 * don't want the code that calls this
86 * function to try and operate on a NULL skb.
92 status = netdev_start_xmit(skb, dev, txq, false);
98 static void queue_process(struct work_struct *work)
100 struct netpoll_info *npinfo =
101 container_of(work, struct netpoll_info, tx_work.work);
105 while ((skb = skb_dequeue(&npinfo->txq))) {
106 struct net_device *dev = skb->dev;
107 struct netdev_queue *txq;
109 if (!netif_device_present(dev) || !netif_running(dev)) {
114 txq = skb_get_tx_queue(dev, skb);
116 local_irq_save(flags);
117 HARD_TX_LOCK(dev, txq, smp_processor_id());
118 if (netif_xmit_frozen_or_stopped(txq) ||
119 netpoll_start_xmit(skb, dev, txq) != NETDEV_TX_OK) {
120 skb_queue_head(&npinfo->txq, skb);
121 HARD_TX_UNLOCK(dev, txq);
122 local_irq_restore(flags);
124 schedule_delayed_work(&npinfo->tx_work, HZ/10);
127 HARD_TX_UNLOCK(dev, txq);
128 local_irq_restore(flags);
133 * Check whether delayed processing was scheduled for our NIC. If so,
134 * we attempt to grab the poll lock and use ->poll() to pump the card.
135 * If this fails, either we've recursed in ->poll() or it's already
136 * running on another CPU.
138 * Note: we don't mask interrupts with this lock because we're using
139 * trylock here and interrupts are already disabled in the softirq
140 * case. Further, we test the poll_owner to avoid recursion on UP
141 * systems where the lock doesn't exist.
143 static int poll_one_napi(struct napi_struct *napi, int budget)
147 /* net_rx_action's ->poll() invocations and our's are
148 * synchronized by this test which is only made while
149 * holding the napi->poll_lock.
151 if (!test_bit(NAPI_STATE_SCHED, &napi->state))
154 set_bit(NAPI_STATE_NPSVC, &napi->state);
156 work = napi->poll(napi, budget);
157 WARN_ONCE(work > budget, "%pF exceeded budget in poll\n", napi->poll);
158 trace_napi_poll(napi);
160 clear_bit(NAPI_STATE_NPSVC, &napi->state);
162 return budget - work;
165 static void poll_napi(struct net_device *dev, int budget)
167 struct napi_struct *napi;
169 list_for_each_entry(napi, &dev->napi_list, dev_list) {
170 if (napi->poll_owner != smp_processor_id() &&
171 spin_trylock(&napi->poll_lock)) {
172 budget = poll_one_napi(napi, budget);
173 spin_unlock(&napi->poll_lock);
178 static void netpoll_poll_dev(struct net_device *dev)
180 const struct net_device_ops *ops;
181 struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
184 /* Don't do any rx activity if the dev_lock mutex is held
185 * the dev_open/close paths use this to block netpoll activity
186 * while changing device state
188 if (down_trylock(&ni->dev_lock))
191 if (!netif_running(dev)) {
196 ops = dev->netdev_ops;
197 if (!ops->ndo_poll_controller) {
202 /* Process pending work on NIC */
203 ops->ndo_poll_controller(dev);
205 poll_napi(dev, budget);
209 zap_completion_queue();
212 void netpoll_poll_disable(struct net_device *dev)
214 struct netpoll_info *ni;
217 idx = srcu_read_lock(&netpoll_srcu);
218 ni = srcu_dereference(dev->npinfo, &netpoll_srcu);
221 srcu_read_unlock(&netpoll_srcu, idx);
223 EXPORT_SYMBOL(netpoll_poll_disable);
225 void netpoll_poll_enable(struct net_device *dev)
227 struct netpoll_info *ni;
229 ni = rcu_dereference(dev->npinfo);
234 EXPORT_SYMBOL(netpoll_poll_enable);
236 static void refill_skbs(void)
241 spin_lock_irqsave(&skb_pool.lock, flags);
242 while (skb_pool.qlen < MAX_SKBS) {
243 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
247 __skb_queue_tail(&skb_pool, skb);
249 spin_unlock_irqrestore(&skb_pool.lock, flags);
252 static void zap_completion_queue(void)
255 struct softnet_data *sd = &get_cpu_var(softnet_data);
257 if (sd->completion_queue) {
258 struct sk_buff *clist;
260 local_irq_save(flags);
261 clist = sd->completion_queue;
262 sd->completion_queue = NULL;
263 local_irq_restore(flags);
265 while (clist != NULL) {
266 struct sk_buff *skb = clist;
268 if (!skb_irq_freeable(skb)) {
269 atomic_inc(&skb->users);
270 dev_kfree_skb_any(skb); /* put this one back */
277 put_cpu_var(softnet_data);
280 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
285 zap_completion_queue();
289 skb = alloc_skb(len, GFP_ATOMIC);
291 skb = skb_dequeue(&skb_pool);
295 netpoll_poll_dev(np->dev);
301 atomic_set(&skb->users, 1);
302 skb_reserve(skb, reserve);
306 static int netpoll_owner_active(struct net_device *dev)
308 struct napi_struct *napi;
310 list_for_each_entry(napi, &dev->napi_list, dev_list) {
311 if (napi->poll_owner == smp_processor_id())
317 /* call with IRQ disabled */
318 void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb,
319 struct net_device *dev)
321 int status = NETDEV_TX_BUSY;
323 /* It is up to the caller to keep npinfo alive. */
324 struct netpoll_info *npinfo;
326 WARN_ON_ONCE(!irqs_disabled());
328 npinfo = rcu_dereference_bh(np->dev->npinfo);
329 if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
330 dev_kfree_skb_irq(skb);
334 /* don't get messages out of order, and no recursion */
335 if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
336 struct netdev_queue *txq;
338 txq = netdev_pick_tx(dev, skb, NULL);
340 /* try until next clock tick */
341 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
342 tries > 0; --tries) {
343 if (HARD_TX_TRYLOCK(dev, txq)) {
344 if (!netif_xmit_stopped(txq))
345 status = netpoll_start_xmit(skb, dev, txq);
347 HARD_TX_UNLOCK(dev, txq);
349 if (status == NETDEV_TX_OK)
354 /* tickle device maybe there is some cleanup */
355 netpoll_poll_dev(np->dev);
357 udelay(USEC_PER_POLL);
360 WARN_ONCE(!irqs_disabled(),
361 "netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pF)\n",
362 dev->name, dev->netdev_ops->ndo_start_xmit);
366 if (status != NETDEV_TX_OK) {
367 skb_queue_tail(&npinfo->txq, skb);
368 schedule_delayed_work(&npinfo->tx_work,0);
371 EXPORT_SYMBOL(netpoll_send_skb_on_dev);
373 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
375 int total_len, ip_len, udp_len;
380 static atomic_t ip_ident;
381 struct ipv6hdr *ip6h;
383 udp_len = len + sizeof(*udph);
385 ip_len = udp_len + sizeof(*ip6h);
387 ip_len = udp_len + sizeof(*iph);
389 total_len = ip_len + LL_RESERVED_SPACE(np->dev);
391 skb = find_skb(np, total_len + np->dev->needed_tailroom,
396 skb_copy_to_linear_data(skb, msg, len);
399 skb_push(skb, sizeof(*udph));
400 skb_reset_transport_header(skb);
402 udph->source = htons(np->local_port);
403 udph->dest = htons(np->remote_port);
404 udph->len = htons(udp_len);
408 udph->check = csum_ipv6_magic(&np->local_ip.in6,
410 udp_len, IPPROTO_UDP,
411 csum_partial(udph, udp_len, 0));
412 if (udph->check == 0)
413 udph->check = CSUM_MANGLED_0;
415 skb_push(skb, sizeof(*ip6h));
416 skb_reset_network_header(skb);
417 ip6h = ipv6_hdr(skb);
419 /* ip6h->version = 6; ip6h->priority = 0; */
420 put_unaligned(0x60, (unsigned char *)ip6h);
421 ip6h->flow_lbl[0] = 0;
422 ip6h->flow_lbl[1] = 0;
423 ip6h->flow_lbl[2] = 0;
425 ip6h->payload_len = htons(sizeof(struct udphdr) + len);
426 ip6h->nexthdr = IPPROTO_UDP;
427 ip6h->hop_limit = 32;
428 ip6h->saddr = np->local_ip.in6;
429 ip6h->daddr = np->remote_ip.in6;
431 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
432 skb_reset_mac_header(skb);
433 skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
436 udph->check = csum_tcpudp_magic(np->local_ip.ip,
438 udp_len, IPPROTO_UDP,
439 csum_partial(udph, udp_len, 0));
440 if (udph->check == 0)
441 udph->check = CSUM_MANGLED_0;
443 skb_push(skb, sizeof(*iph));
444 skb_reset_network_header(skb);
447 /* iph->version = 4; iph->ihl = 5; */
448 put_unaligned(0x45, (unsigned char *)iph);
450 put_unaligned(htons(ip_len), &(iph->tot_len));
451 iph->id = htons(atomic_inc_return(&ip_ident));
454 iph->protocol = IPPROTO_UDP;
456 put_unaligned(np->local_ip.ip, &(iph->saddr));
457 put_unaligned(np->remote_ip.ip, &(iph->daddr));
458 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
460 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
461 skb_reset_mac_header(skb);
462 skb->protocol = eth->h_proto = htons(ETH_P_IP);
465 ether_addr_copy(eth->h_source, np->dev->dev_addr);
466 ether_addr_copy(eth->h_dest, np->remote_mac);
470 netpoll_send_skb(np, skb);
472 EXPORT_SYMBOL(netpoll_send_udp);
474 void netpoll_print_options(struct netpoll *np)
476 np_info(np, "local port %d\n", np->local_port);
478 np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
480 np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
481 np_info(np, "interface '%s'\n", np->dev_name);
482 np_info(np, "remote port %d\n", np->remote_port);
484 np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
486 np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
487 np_info(np, "remote ethernet address %pM\n", np->remote_mac);
489 EXPORT_SYMBOL(netpoll_print_options);
491 static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
495 if (!strchr(str, ':') &&
496 in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
500 if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
501 #if IS_ENABLED(CONFIG_IPV6)
511 int netpoll_parse_options(struct netpoll *np, char *opt)
513 char *cur=opt, *delim;
515 bool ipversion_set = false;
518 if ((delim = strchr(cur, '@')) == NULL)
521 if (kstrtou16(cur, 10, &np->local_port))
528 ipversion_set = true;
529 if ((delim = strchr(cur, '/')) == NULL)
532 ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
536 np->ipv6 = (bool)ipv6;
542 /* parse out dev name */
543 if ((delim = strchr(cur, ',')) == NULL)
546 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
553 if ((delim = strchr(cur, '@')) == NULL)
556 if (*cur == ' ' || *cur == '\t')
557 np_info(np, "warning: whitespace is not allowed\n");
558 if (kstrtou16(cur, 10, &np->remote_port))
565 if ((delim = strchr(cur, '/')) == NULL)
568 ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
571 else if (ipversion_set && np->ipv6 != (bool)ipv6)
574 np->ipv6 = (bool)ipv6;
579 if (!mac_pton(cur, np->remote_mac))
583 netpoll_print_options(np);
588 np_info(np, "couldn't parse config at '%s'!\n", cur);
591 EXPORT_SYMBOL(netpoll_parse_options);
593 int __netpoll_setup(struct netpoll *np, struct net_device *ndev)
595 struct netpoll_info *npinfo;
596 const struct net_device_ops *ops;
600 strlcpy(np->dev_name, ndev->name, IFNAMSIZ);
601 INIT_WORK(&np->cleanup_work, netpoll_async_cleanup);
603 if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) ||
604 !ndev->netdev_ops->ndo_poll_controller) {
605 np_err(np, "%s doesn't support polling, aborting\n",
612 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
618 sema_init(&npinfo->dev_lock, 1);
619 skb_queue_head_init(&npinfo->txq);
620 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
622 atomic_set(&npinfo->refcnt, 1);
624 ops = np->dev->netdev_ops;
625 if (ops->ndo_netpoll_setup) {
626 err = ops->ndo_netpoll_setup(ndev, npinfo);
631 npinfo = rtnl_dereference(ndev->npinfo);
632 atomic_inc(&npinfo->refcnt);
635 npinfo->netpoll = np;
637 /* last thing to do is link it to the net device structure */
638 rcu_assign_pointer(ndev->npinfo, npinfo);
647 EXPORT_SYMBOL_GPL(__netpoll_setup);
649 int netpoll_setup(struct netpoll *np)
651 struct net_device *ndev = NULL;
652 struct in_device *in_dev;
657 struct net *net = current->nsproxy->net_ns;
658 ndev = __dev_get_by_name(net, np->dev_name);
661 np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
667 if (netdev_master_upper_dev_get(ndev)) {
668 np_err(np, "%s is a slave device, aborting\n", np->dev_name);
673 if (!netif_running(ndev)) {
674 unsigned long atmost, atleast;
676 np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
678 err = dev_open(ndev);
681 np_err(np, "failed to open %s\n", ndev->name);
686 atleast = jiffies + HZ/10;
687 atmost = jiffies + carrier_timeout * HZ;
688 while (!netif_carrier_ok(ndev)) {
689 if (time_after(jiffies, atmost)) {
690 np_notice(np, "timeout waiting for carrier\n");
696 /* If carrier appears to come up instantly, we don't
697 * trust it and pause so that we don't pump all our
698 * queued console messages into the bitbucket.
701 if (time_before(jiffies, atleast)) {
702 np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
708 if (!np->local_ip.ip) {
710 in_dev = __in_dev_get_rtnl(ndev);
712 if (!in_dev || !in_dev->ifa_list) {
713 np_err(np, "no IP address for %s, aborting\n",
719 np->local_ip.ip = in_dev->ifa_list->ifa_local;
720 np_info(np, "local IP %pI4\n", &np->local_ip.ip);
722 #if IS_ENABLED(CONFIG_IPV6)
723 struct inet6_dev *idev;
726 idev = __in6_dev_get(ndev);
728 struct inet6_ifaddr *ifp;
730 read_lock_bh(&idev->lock);
731 list_for_each_entry(ifp, &idev->addr_list, if_list) {
732 if (ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL)
734 np->local_ip.in6 = ifp->addr;
738 read_unlock_bh(&idev->lock);
741 np_err(np, "no IPv6 address for %s, aborting\n",
745 np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
747 np_err(np, "IPv6 is not supported %s, aborting\n",
755 /* fill up the skb queue */
758 err = __netpoll_setup(np, ndev);
771 EXPORT_SYMBOL(netpoll_setup);
773 static int __init netpoll_init(void)
775 skb_queue_head_init(&skb_pool);
778 core_initcall(netpoll_init);
780 static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
782 struct netpoll_info *npinfo =
783 container_of(rcu_head, struct netpoll_info, rcu);
785 skb_queue_purge(&npinfo->txq);
787 /* we can't call cancel_delayed_work_sync here, as we are in softirq */
788 cancel_delayed_work(&npinfo->tx_work);
790 /* clean after last, unfinished work */
791 __skb_queue_purge(&npinfo->txq);
792 /* now cancel it again */
793 cancel_delayed_work(&npinfo->tx_work);
797 void __netpoll_cleanup(struct netpoll *np)
799 struct netpoll_info *npinfo;
801 /* rtnl_dereference would be preferable here but
802 * rcu_cleanup_netpoll path can put us in here safely without
803 * holding the rtnl, so plain rcu_dereference it is
805 npinfo = rtnl_dereference(np->dev->npinfo);
809 synchronize_srcu(&netpoll_srcu);
811 if (atomic_dec_and_test(&npinfo->refcnt)) {
812 const struct net_device_ops *ops;
814 ops = np->dev->netdev_ops;
815 if (ops->ndo_netpoll_cleanup)
816 ops->ndo_netpoll_cleanup(np->dev);
818 RCU_INIT_POINTER(np->dev->npinfo, NULL);
819 call_rcu_bh(&npinfo->rcu, rcu_cleanup_netpoll_info);
821 RCU_INIT_POINTER(np->dev->npinfo, NULL);
823 EXPORT_SYMBOL_GPL(__netpoll_cleanup);
825 static void netpoll_async_cleanup(struct work_struct *work)
827 struct netpoll *np = container_of(work, struct netpoll, cleanup_work);
830 __netpoll_cleanup(np);
835 void __netpoll_free_async(struct netpoll *np)
837 schedule_work(&np->cleanup_work);
839 EXPORT_SYMBOL_GPL(__netpoll_free_async);
841 void netpoll_cleanup(struct netpoll *np)
846 __netpoll_cleanup(np);
852 EXPORT_SYMBOL(netpoll_cleanup);