1 // SPDX-License-Identifier: GPL-2.0-only
3 * Common framework for low-level network console, dump, and debugger code
7 * based on the netconsole code from:
10 * Copyright (C) 2002 Red Hat, Inc.
13 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15 #include <linux/moduleparam.h>
16 #include <linux/kernel.h>
17 #include <linux/netdevice.h>
18 #include <linux/etherdevice.h>
19 #include <linux/string.h>
20 #include <linux/if_arp.h>
21 #include <linux/inetdevice.h>
22 #include <linux/inet.h>
23 #include <linux/interrupt.h>
24 #include <linux/netpoll.h>
25 #include <linux/sched.h>
26 #include <linux/delay.h>
27 #include <linux/rcupdate.h>
28 #include <linux/workqueue.h>
29 #include <linux/slab.h>
30 #include <linux/export.h>
31 #include <linux/if_vlan.h>
34 #include <net/addrconf.h>
35 #include <net/ndisc.h>
36 #include <net/ip6_checksum.h>
37 #include <asm/unaligned.h>
38 #include <trace/events/napi.h>
41 * We maintain a small pool of fully-sized skbs, to make sure the
42 * message gets out even in extreme OOM situations.
45 #define MAX_UDP_CHUNK 1460
48 static struct sk_buff_head skb_pool;
50 DEFINE_STATIC_SRCU(netpoll_srcu);
52 #define USEC_PER_POLL 50
54 #define MAX_SKB_SIZE \
55 (sizeof(struct ethhdr) + \
56 sizeof(struct iphdr) + \
57 sizeof(struct udphdr) + \
60 static void zap_completion_queue(void);
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 netdev_tx_t netpoll_start_xmit(struct sk_buff *skb,
73 struct net_device *dev,
74 struct netdev_queue *txq)
76 netdev_tx_t status = NETDEV_TX_OK;
77 netdev_features_t features;
79 features = netif_skb_features(skb);
81 if (skb_vlan_tag_present(skb) &&
82 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
83 skb = __vlan_hwaccel_push_inside(skb);
85 /* This is actually a packet drop, but we
86 * don't want the code that calls this
87 * function to try and operate on a NULL skb.
93 status = netdev_start_xmit(skb, dev, txq, false);
99 static void queue_process(struct work_struct *work)
101 struct netpoll_info *npinfo =
102 container_of(work, struct netpoll_info, tx_work.work);
106 while ((skb = skb_dequeue(&npinfo->txq))) {
107 struct net_device *dev = skb->dev;
108 struct netdev_queue *txq;
109 unsigned int q_index;
111 if (!netif_device_present(dev) || !netif_running(dev)) {
116 local_irq_save(flags);
117 /* check if skb->queue_mapping is still valid */
118 q_index = skb_get_queue_mapping(skb);
119 if (unlikely(q_index >= dev->real_num_tx_queues)) {
120 q_index = q_index % dev->real_num_tx_queues;
121 skb_set_queue_mapping(skb, q_index);
123 txq = netdev_get_tx_queue(dev, q_index);
124 HARD_TX_LOCK(dev, txq, smp_processor_id());
125 if (netif_xmit_frozen_or_stopped(txq) ||
126 !dev_xmit_complete(netpoll_start_xmit(skb, dev, txq))) {
127 skb_queue_head(&npinfo->txq, skb);
128 HARD_TX_UNLOCK(dev, txq);
129 local_irq_restore(flags);
131 schedule_delayed_work(&npinfo->tx_work, HZ/10);
134 HARD_TX_UNLOCK(dev, txq);
135 local_irq_restore(flags);
139 static void poll_one_napi(struct napi_struct *napi)
143 /* If we set this bit but see that it has already been set,
144 * that indicates that napi has been disabled and we need
145 * to abort this operation
147 if (test_and_set_bit(NAPI_STATE_NPSVC, &napi->state))
150 /* We explicilty pass the polling call a budget of 0 to
151 * indicate that we are clearing the Tx path only.
153 work = napi->poll(napi, 0);
154 WARN_ONCE(work, "%pS exceeded budget in poll\n", napi->poll);
155 trace_napi_poll(napi, work, 0);
157 clear_bit(NAPI_STATE_NPSVC, &napi->state);
160 static void poll_napi(struct net_device *dev)
162 struct napi_struct *napi;
163 int cpu = smp_processor_id();
165 list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) {
166 if (cmpxchg(&napi->poll_owner, -1, cpu) == -1) {
168 smp_store_release(&napi->poll_owner, -1);
173 void netpoll_poll_dev(struct net_device *dev)
175 struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
176 const struct net_device_ops *ops;
178 /* Don't do any rx activity if the dev_lock mutex is held
179 * the dev_open/close paths use this to block netpoll activity
180 * while changing device state
182 if (!ni || down_trylock(&ni->dev_lock))
185 if (!netif_running(dev)) {
190 ops = dev->netdev_ops;
191 if (ops->ndo_poll_controller)
192 ops->ndo_poll_controller(dev);
198 zap_completion_queue();
200 EXPORT_SYMBOL(netpoll_poll_dev);
202 void netpoll_poll_disable(struct net_device *dev)
204 struct netpoll_info *ni;
207 idx = srcu_read_lock(&netpoll_srcu);
208 ni = srcu_dereference(dev->npinfo, &netpoll_srcu);
211 srcu_read_unlock(&netpoll_srcu, idx);
213 EXPORT_SYMBOL(netpoll_poll_disable);
215 void netpoll_poll_enable(struct net_device *dev)
217 struct netpoll_info *ni;
219 ni = rcu_dereference(dev->npinfo);
224 EXPORT_SYMBOL(netpoll_poll_enable);
226 static void refill_skbs(void)
231 spin_lock_irqsave(&skb_pool.lock, flags);
232 while (skb_pool.qlen < MAX_SKBS) {
233 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
237 __skb_queue_tail(&skb_pool, skb);
239 spin_unlock_irqrestore(&skb_pool.lock, flags);
242 static void zap_completion_queue(void)
245 struct softnet_data *sd = &get_cpu_var(softnet_data);
247 if (sd->completion_queue) {
248 struct sk_buff *clist;
250 local_irq_save(flags);
251 clist = sd->completion_queue;
252 sd->completion_queue = NULL;
253 local_irq_restore(flags);
255 while (clist != NULL) {
256 struct sk_buff *skb = clist;
258 if (!skb_irq_freeable(skb)) {
259 refcount_set(&skb->users, 1);
260 dev_kfree_skb_any(skb); /* put this one back */
267 put_cpu_var(softnet_data);
270 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
275 zap_completion_queue();
279 skb = alloc_skb(len, GFP_ATOMIC);
281 skb = skb_dequeue(&skb_pool);
285 netpoll_poll_dev(np->dev);
291 refcount_set(&skb->users, 1);
292 skb_reserve(skb, reserve);
296 static int netpoll_owner_active(struct net_device *dev)
298 struct napi_struct *napi;
300 list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) {
301 if (napi->poll_owner == smp_processor_id())
307 /* call with IRQ disabled */
308 static netdev_tx_t __netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
310 netdev_tx_t status = NETDEV_TX_BUSY;
311 struct net_device *dev;
313 /* It is up to the caller to keep npinfo alive. */
314 struct netpoll_info *npinfo;
316 lockdep_assert_irqs_disabled();
319 npinfo = rcu_dereference_bh(dev->npinfo);
321 if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
322 dev_kfree_skb_irq(skb);
323 return NET_XMIT_DROP;
326 /* don't get messages out of order, and no recursion */
327 if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
328 struct netdev_queue *txq;
330 txq = netdev_core_pick_tx(dev, skb, NULL);
332 /* try until next clock tick */
333 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
334 tries > 0; --tries) {
335 if (HARD_TX_TRYLOCK(dev, txq)) {
336 if (!netif_xmit_stopped(txq))
337 status = netpoll_start_xmit(skb, dev, txq);
339 HARD_TX_UNLOCK(dev, txq);
341 if (dev_xmit_complete(status))
346 /* tickle device maybe there is some cleanup */
347 netpoll_poll_dev(np->dev);
349 udelay(USEC_PER_POLL);
352 WARN_ONCE(!irqs_disabled(),
353 "netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pS)\n",
354 dev->name, dev->netdev_ops->ndo_start_xmit);
358 if (!dev_xmit_complete(status)) {
359 skb_queue_tail(&npinfo->txq, skb);
360 schedule_delayed_work(&npinfo->tx_work,0);
365 netdev_tx_t netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
371 dev_kfree_skb_irq(skb);
374 local_irq_save(flags);
375 ret = __netpoll_send_skb(np, skb);
376 local_irq_restore(flags);
380 EXPORT_SYMBOL(netpoll_send_skb);
382 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
384 int total_len, ip_len, udp_len;
389 static atomic_t ip_ident;
390 struct ipv6hdr *ip6h;
392 WARN_ON_ONCE(!irqs_disabled());
394 udp_len = len + sizeof(*udph);
396 ip_len = udp_len + sizeof(*ip6h);
398 ip_len = udp_len + sizeof(*iph);
400 total_len = ip_len + LL_RESERVED_SPACE(np->dev);
402 skb = find_skb(np, total_len + np->dev->needed_tailroom,
407 skb_copy_to_linear_data(skb, msg, len);
410 skb_push(skb, sizeof(*udph));
411 skb_reset_transport_header(skb);
413 udph->source = htons(np->local_port);
414 udph->dest = htons(np->remote_port);
415 udph->len = htons(udp_len);
419 udph->check = csum_ipv6_magic(&np->local_ip.in6,
421 udp_len, IPPROTO_UDP,
422 csum_partial(udph, udp_len, 0));
423 if (udph->check == 0)
424 udph->check = CSUM_MANGLED_0;
426 skb_push(skb, sizeof(*ip6h));
427 skb_reset_network_header(skb);
428 ip6h = ipv6_hdr(skb);
430 /* ip6h->version = 6; ip6h->priority = 0; */
431 put_unaligned(0x60, (unsigned char *)ip6h);
432 ip6h->flow_lbl[0] = 0;
433 ip6h->flow_lbl[1] = 0;
434 ip6h->flow_lbl[2] = 0;
436 ip6h->payload_len = htons(sizeof(struct udphdr) + len);
437 ip6h->nexthdr = IPPROTO_UDP;
438 ip6h->hop_limit = 32;
439 ip6h->saddr = np->local_ip.in6;
440 ip6h->daddr = np->remote_ip.in6;
442 eth = skb_push(skb, ETH_HLEN);
443 skb_reset_mac_header(skb);
444 skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
447 udph->check = csum_tcpudp_magic(np->local_ip.ip,
449 udp_len, IPPROTO_UDP,
450 csum_partial(udph, udp_len, 0));
451 if (udph->check == 0)
452 udph->check = CSUM_MANGLED_0;
454 skb_push(skb, sizeof(*iph));
455 skb_reset_network_header(skb);
458 /* iph->version = 4; iph->ihl = 5; */
459 put_unaligned(0x45, (unsigned char *)iph);
461 put_unaligned(htons(ip_len), &(iph->tot_len));
462 iph->id = htons(atomic_inc_return(&ip_ident));
465 iph->protocol = IPPROTO_UDP;
467 put_unaligned(np->local_ip.ip, &(iph->saddr));
468 put_unaligned(np->remote_ip.ip, &(iph->daddr));
469 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
471 eth = skb_push(skb, ETH_HLEN);
472 skb_reset_mac_header(skb);
473 skb->protocol = eth->h_proto = htons(ETH_P_IP);
476 ether_addr_copy(eth->h_source, np->dev->dev_addr);
477 ether_addr_copy(eth->h_dest, np->remote_mac);
481 netpoll_send_skb(np, skb);
483 EXPORT_SYMBOL(netpoll_send_udp);
485 void netpoll_print_options(struct netpoll *np)
487 np_info(np, "local port %d\n", np->local_port);
489 np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
491 np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
492 np_info(np, "interface '%s'\n", np->dev_name);
493 np_info(np, "remote port %d\n", np->remote_port);
495 np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
497 np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
498 np_info(np, "remote ethernet address %pM\n", np->remote_mac);
500 EXPORT_SYMBOL(netpoll_print_options);
502 static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
506 if (!strchr(str, ':') &&
507 in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
511 if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
512 #if IS_ENABLED(CONFIG_IPV6)
522 int netpoll_parse_options(struct netpoll *np, char *opt)
524 char *cur=opt, *delim;
526 bool ipversion_set = false;
529 if ((delim = strchr(cur, '@')) == NULL)
532 if (kstrtou16(cur, 10, &np->local_port))
539 ipversion_set = true;
540 if ((delim = strchr(cur, '/')) == NULL)
543 ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
547 np->ipv6 = (bool)ipv6;
553 /* parse out dev name */
554 if ((delim = strchr(cur, ',')) == NULL)
557 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
564 if ((delim = strchr(cur, '@')) == NULL)
567 if (*cur == ' ' || *cur == '\t')
568 np_info(np, "warning: whitespace is not allowed\n");
569 if (kstrtou16(cur, 10, &np->remote_port))
576 if ((delim = strchr(cur, '/')) == NULL)
579 ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
582 else if (ipversion_set && np->ipv6 != (bool)ipv6)
585 np->ipv6 = (bool)ipv6;
590 if (!mac_pton(cur, np->remote_mac))
594 netpoll_print_options(np);
599 np_info(np, "couldn't parse config at '%s'!\n", cur);
602 EXPORT_SYMBOL(netpoll_parse_options);
604 int __netpoll_setup(struct netpoll *np, struct net_device *ndev)
606 struct netpoll_info *npinfo;
607 const struct net_device_ops *ops;
611 strlcpy(np->dev_name, ndev->name, IFNAMSIZ);
613 if (ndev->priv_flags & IFF_DISABLE_NETPOLL) {
614 np_err(np, "%s doesn't support polling, aborting\n",
621 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
627 sema_init(&npinfo->dev_lock, 1);
628 skb_queue_head_init(&npinfo->txq);
629 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
631 refcount_set(&npinfo->refcnt, 1);
633 ops = np->dev->netdev_ops;
634 if (ops->ndo_netpoll_setup) {
635 err = ops->ndo_netpoll_setup(ndev, npinfo);
640 npinfo = rtnl_dereference(ndev->npinfo);
641 refcount_inc(&npinfo->refcnt);
644 npinfo->netpoll = np;
646 /* last thing to do is link it to the net device structure */
647 rcu_assign_pointer(ndev->npinfo, npinfo);
656 EXPORT_SYMBOL_GPL(__netpoll_setup);
658 int netpoll_setup(struct netpoll *np)
660 struct net_device *ndev = NULL;
661 struct in_device *in_dev;
665 if (np->dev_name[0]) {
666 struct net *net = current->nsproxy->net_ns;
667 ndev = __dev_get_by_name(net, np->dev_name);
670 np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
676 if (netdev_master_upper_dev_get(ndev)) {
677 np_err(np, "%s is a slave device, aborting\n", np->dev_name);
682 if (!netif_running(ndev)) {
683 unsigned long atmost, atleast;
685 np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
687 err = dev_open(ndev, NULL);
690 np_err(np, "failed to open %s\n", ndev->name);
695 atleast = jiffies + HZ/10;
696 atmost = jiffies + carrier_timeout * HZ;
697 while (!netif_carrier_ok(ndev)) {
698 if (time_after(jiffies, atmost)) {
699 np_notice(np, "timeout waiting for carrier\n");
705 /* If carrier appears to come up instantly, we don't
706 * trust it and pause so that we don't pump all our
707 * queued console messages into the bitbucket.
710 if (time_before(jiffies, atleast)) {
711 np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
717 if (!np->local_ip.ip) {
719 const struct in_ifaddr *ifa;
721 in_dev = __in_dev_get_rtnl(ndev);
725 ifa = rtnl_dereference(in_dev->ifa_list);
728 np_err(np, "no IP address for %s, aborting\n",
734 np->local_ip.ip = ifa->ifa_local;
735 np_info(np, "local IP %pI4\n", &np->local_ip.ip);
737 #if IS_ENABLED(CONFIG_IPV6)
738 struct inet6_dev *idev;
741 idev = __in6_dev_get(ndev);
743 struct inet6_ifaddr *ifp;
745 read_lock_bh(&idev->lock);
746 list_for_each_entry(ifp, &idev->addr_list, if_list) {
747 if (!!(ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL) !=
748 !!(ipv6_addr_type(&np->remote_ip.in6) & IPV6_ADDR_LINKLOCAL))
750 np->local_ip.in6 = ifp->addr;
754 read_unlock_bh(&idev->lock);
757 np_err(np, "no IPv6 address for %s, aborting\n",
761 np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
763 np_err(np, "IPv6 is not supported %s, aborting\n",
771 /* fill up the skb queue */
774 err = __netpoll_setup(np, ndev);
787 EXPORT_SYMBOL(netpoll_setup);
789 static int __init netpoll_init(void)
791 skb_queue_head_init(&skb_pool);
794 core_initcall(netpoll_init);
796 static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
798 struct netpoll_info *npinfo =
799 container_of(rcu_head, struct netpoll_info, rcu);
801 skb_queue_purge(&npinfo->txq);
803 /* we can't call cancel_delayed_work_sync here, as we are in softirq */
804 cancel_delayed_work(&npinfo->tx_work);
806 /* clean after last, unfinished work */
807 __skb_queue_purge(&npinfo->txq);
808 /* now cancel it again */
809 cancel_delayed_work(&npinfo->tx_work);
813 void __netpoll_cleanup(struct netpoll *np)
815 struct netpoll_info *npinfo;
817 npinfo = rtnl_dereference(np->dev->npinfo);
821 synchronize_srcu(&netpoll_srcu);
823 if (refcount_dec_and_test(&npinfo->refcnt)) {
824 const struct net_device_ops *ops;
826 ops = np->dev->netdev_ops;
827 if (ops->ndo_netpoll_cleanup)
828 ops->ndo_netpoll_cleanup(np->dev);
830 RCU_INIT_POINTER(np->dev->npinfo, NULL);
831 call_rcu(&npinfo->rcu, rcu_cleanup_netpoll_info);
833 RCU_INIT_POINTER(np->dev->npinfo, NULL);
835 EXPORT_SYMBOL_GPL(__netpoll_cleanup);
837 void __netpoll_free(struct netpoll *np)
841 /* Wait for transmitting packets to finish before freeing. */
843 __netpoll_cleanup(np);
846 EXPORT_SYMBOL_GPL(__netpoll_free);
848 void netpoll_cleanup(struct netpoll *np)
853 __netpoll_cleanup(np);
859 EXPORT_SYMBOL(netpoll_cleanup);