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/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 const struct net_device_ops *ops = dev->netdev_ops;
76 int status = NETDEV_TX_OK;
77 netdev_features_t features;
79 features = netif_skb_features(skb);
81 if (vlan_tx_tag_present(skb) &&
82 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
83 skb = __vlan_put_tag(skb, skb->vlan_proto,
84 vlan_tx_tag_get(skb));
86 /* This is actually a packet drop, but we
87 * don't want the code that calls this
88 * function to try and operate on a NULL skb.
95 status = ops->ndo_start_xmit(skb, dev);
96 if (status == NETDEV_TX_OK)
97 txq_trans_update(txq);
103 static void queue_process(struct work_struct *work)
105 struct netpoll_info *npinfo =
106 container_of(work, struct netpoll_info, tx_work.work);
110 while ((skb = skb_dequeue(&npinfo->txq))) {
111 struct net_device *dev = skb->dev;
112 struct netdev_queue *txq;
114 if (!netif_device_present(dev) || !netif_running(dev)) {
119 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
121 local_irq_save(flags);
122 HARD_TX_LOCK(dev, txq, smp_processor_id());
123 if (netif_xmit_frozen_or_stopped(txq) ||
124 netpoll_start_xmit(skb, dev, txq) != NETDEV_TX_OK) {
125 skb_queue_head(&npinfo->txq, skb);
126 HARD_TX_UNLOCK(dev, txq);
127 local_irq_restore(flags);
129 schedule_delayed_work(&npinfo->tx_work, HZ/10);
132 HARD_TX_UNLOCK(dev, txq);
133 local_irq_restore(flags);
138 * Check whether delayed processing was scheduled for our NIC. If so,
139 * we attempt to grab the poll lock and use ->poll() to pump the card.
140 * If this fails, either we've recursed in ->poll() or it's already
141 * running on another CPU.
143 * Note: we don't mask interrupts with this lock because we're using
144 * trylock here and interrupts are already disabled in the softirq
145 * case. Further, we test the poll_owner to avoid recursion on UP
146 * systems where the lock doesn't exist.
148 static int poll_one_napi(struct napi_struct *napi, int budget)
152 /* net_rx_action's ->poll() invocations and our's are
153 * synchronized by this test which is only made while
154 * holding the napi->poll_lock.
156 if (!test_bit(NAPI_STATE_SCHED, &napi->state))
159 set_bit(NAPI_STATE_NPSVC, &napi->state);
161 work = napi->poll(napi, budget);
162 WARN_ONCE(work > budget, "%pF exceeded budget in poll\n", napi->poll);
163 trace_napi_poll(napi);
165 clear_bit(NAPI_STATE_NPSVC, &napi->state);
167 return budget - work;
170 static void poll_napi(struct net_device *dev, int budget)
172 struct napi_struct *napi;
174 list_for_each_entry(napi, &dev->napi_list, dev_list) {
175 if (napi->poll_owner != smp_processor_id() &&
176 spin_trylock(&napi->poll_lock)) {
177 budget = poll_one_napi(napi, budget);
178 spin_unlock(&napi->poll_lock);
183 static void netpoll_poll_dev(struct net_device *dev)
185 const struct net_device_ops *ops;
186 struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
189 /* Don't do any rx activity if the dev_lock mutex is held
190 * the dev_open/close paths use this to block netpoll activity
191 * while changing device state
193 if (down_trylock(&ni->dev_lock))
196 if (!netif_running(dev)) {
201 ops = dev->netdev_ops;
202 if (!ops->ndo_poll_controller) {
207 /* Process pending work on NIC */
208 ops->ndo_poll_controller(dev);
210 poll_napi(dev, budget);
214 zap_completion_queue();
217 void netpoll_poll_disable(struct net_device *dev)
219 struct netpoll_info *ni;
222 idx = srcu_read_lock(&netpoll_srcu);
223 ni = srcu_dereference(dev->npinfo, &netpoll_srcu);
226 srcu_read_unlock(&netpoll_srcu, idx);
228 EXPORT_SYMBOL(netpoll_poll_disable);
230 void netpoll_poll_enable(struct net_device *dev)
232 struct netpoll_info *ni;
234 ni = rcu_dereference(dev->npinfo);
239 EXPORT_SYMBOL(netpoll_poll_enable);
241 static void refill_skbs(void)
246 spin_lock_irqsave(&skb_pool.lock, flags);
247 while (skb_pool.qlen < MAX_SKBS) {
248 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
252 __skb_queue_tail(&skb_pool, skb);
254 spin_unlock_irqrestore(&skb_pool.lock, flags);
257 static void zap_completion_queue(void)
260 struct softnet_data *sd = &get_cpu_var(softnet_data);
262 if (sd->completion_queue) {
263 struct sk_buff *clist;
265 local_irq_save(flags);
266 clist = sd->completion_queue;
267 sd->completion_queue = NULL;
268 local_irq_restore(flags);
270 while (clist != NULL) {
271 struct sk_buff *skb = clist;
273 if (!skb_irq_freeable(skb)) {
274 atomic_inc(&skb->users);
275 dev_kfree_skb_any(skb); /* put this one back */
282 put_cpu_var(softnet_data);
285 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
290 zap_completion_queue();
294 skb = alloc_skb(len, GFP_ATOMIC);
296 skb = skb_dequeue(&skb_pool);
300 netpoll_poll_dev(np->dev);
306 atomic_set(&skb->users, 1);
307 skb_reserve(skb, reserve);
311 static int netpoll_owner_active(struct net_device *dev)
313 struct napi_struct *napi;
315 list_for_each_entry(napi, &dev->napi_list, dev_list) {
316 if (napi->poll_owner == smp_processor_id())
322 /* call with IRQ disabled */
323 void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb,
324 struct net_device *dev)
326 int status = NETDEV_TX_BUSY;
328 /* It is up to the caller to keep npinfo alive. */
329 struct netpoll_info *npinfo;
331 WARN_ON_ONCE(!irqs_disabled());
333 npinfo = rcu_dereference_bh(np->dev->npinfo);
334 if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
335 dev_kfree_skb_irq(skb);
339 /* don't get messages out of order, and no recursion */
340 if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
341 struct netdev_queue *txq;
343 txq = netdev_pick_tx(dev, skb, NULL);
345 /* try until next clock tick */
346 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
347 tries > 0; --tries) {
348 if (HARD_TX_TRYLOCK(dev, txq)) {
349 if (!netif_xmit_stopped(txq))
350 status = netpoll_start_xmit(skb, dev, txq);
352 HARD_TX_UNLOCK(dev, txq);
354 if (status == NETDEV_TX_OK)
359 /* tickle device maybe there is some cleanup */
360 netpoll_poll_dev(np->dev);
362 udelay(USEC_PER_POLL);
365 WARN_ONCE(!irqs_disabled(),
366 "netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pF)\n",
367 dev->name, dev->netdev_ops->ndo_start_xmit);
371 if (status != NETDEV_TX_OK) {
372 skb_queue_tail(&npinfo->txq, skb);
373 schedule_delayed_work(&npinfo->tx_work,0);
376 EXPORT_SYMBOL(netpoll_send_skb_on_dev);
378 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
380 int total_len, ip_len, udp_len;
385 static atomic_t ip_ident;
386 struct ipv6hdr *ip6h;
388 udp_len = len + sizeof(*udph);
390 ip_len = udp_len + sizeof(*ip6h);
392 ip_len = udp_len + sizeof(*iph);
394 total_len = ip_len + LL_RESERVED_SPACE(np->dev);
396 skb = find_skb(np, total_len + np->dev->needed_tailroom,
401 skb_copy_to_linear_data(skb, msg, len);
404 skb_push(skb, sizeof(*udph));
405 skb_reset_transport_header(skb);
407 udph->source = htons(np->local_port);
408 udph->dest = htons(np->remote_port);
409 udph->len = htons(udp_len);
413 udph->check = csum_ipv6_magic(&np->local_ip.in6,
415 udp_len, IPPROTO_UDP,
416 csum_partial(udph, udp_len, 0));
417 if (udph->check == 0)
418 udph->check = CSUM_MANGLED_0;
420 skb_push(skb, sizeof(*ip6h));
421 skb_reset_network_header(skb);
422 ip6h = ipv6_hdr(skb);
424 /* ip6h->version = 6; ip6h->priority = 0; */
425 put_unaligned(0x60, (unsigned char *)ip6h);
426 ip6h->flow_lbl[0] = 0;
427 ip6h->flow_lbl[1] = 0;
428 ip6h->flow_lbl[2] = 0;
430 ip6h->payload_len = htons(sizeof(struct udphdr) + len);
431 ip6h->nexthdr = IPPROTO_UDP;
432 ip6h->hop_limit = 32;
433 ip6h->saddr = np->local_ip.in6;
434 ip6h->daddr = np->remote_ip.in6;
436 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
437 skb_reset_mac_header(skb);
438 skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
441 udph->check = csum_tcpudp_magic(np->local_ip.ip,
443 udp_len, IPPROTO_UDP,
444 csum_partial(udph, udp_len, 0));
445 if (udph->check == 0)
446 udph->check = CSUM_MANGLED_0;
448 skb_push(skb, sizeof(*iph));
449 skb_reset_network_header(skb);
452 /* iph->version = 4; iph->ihl = 5; */
453 put_unaligned(0x45, (unsigned char *)iph);
455 put_unaligned(htons(ip_len), &(iph->tot_len));
456 iph->id = htons(atomic_inc_return(&ip_ident));
459 iph->protocol = IPPROTO_UDP;
461 put_unaligned(np->local_ip.ip, &(iph->saddr));
462 put_unaligned(np->remote_ip.ip, &(iph->daddr));
463 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
465 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
466 skb_reset_mac_header(skb);
467 skb->protocol = eth->h_proto = htons(ETH_P_IP);
470 ether_addr_copy(eth->h_source, np->dev->dev_addr);
471 ether_addr_copy(eth->h_dest, np->remote_mac);
475 netpoll_send_skb(np, skb);
477 EXPORT_SYMBOL(netpoll_send_udp);
479 void netpoll_print_options(struct netpoll *np)
481 np_info(np, "local port %d\n", np->local_port);
483 np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
485 np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
486 np_info(np, "interface '%s'\n", np->dev_name);
487 np_info(np, "remote port %d\n", np->remote_port);
489 np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
491 np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
492 np_info(np, "remote ethernet address %pM\n", np->remote_mac);
494 EXPORT_SYMBOL(netpoll_print_options);
496 static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
500 if (!strchr(str, ':') &&
501 in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
505 if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
506 #if IS_ENABLED(CONFIG_IPV6)
516 int netpoll_parse_options(struct netpoll *np, char *opt)
518 char *cur=opt, *delim;
520 bool ipversion_set = false;
523 if ((delim = strchr(cur, '@')) == NULL)
526 if (kstrtou16(cur, 10, &np->local_port))
533 ipversion_set = true;
534 if ((delim = strchr(cur, '/')) == NULL)
537 ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
541 np->ipv6 = (bool)ipv6;
547 /* parse out dev name */
548 if ((delim = strchr(cur, ',')) == NULL)
551 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
558 if ((delim = strchr(cur, '@')) == NULL)
561 if (*cur == ' ' || *cur == '\t')
562 np_info(np, "warning: whitespace is not allowed\n");
563 if (kstrtou16(cur, 10, &np->remote_port))
570 if ((delim = strchr(cur, '/')) == NULL)
573 ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
576 else if (ipversion_set && np->ipv6 != (bool)ipv6)
579 np->ipv6 = (bool)ipv6;
584 if (!mac_pton(cur, np->remote_mac))
588 netpoll_print_options(np);
593 np_info(np, "couldn't parse config at '%s'!\n", cur);
596 EXPORT_SYMBOL(netpoll_parse_options);
598 int __netpoll_setup(struct netpoll *np, struct net_device *ndev)
600 struct netpoll_info *npinfo;
601 const struct net_device_ops *ops;
605 strlcpy(np->dev_name, ndev->name, IFNAMSIZ);
606 INIT_WORK(&np->cleanup_work, netpoll_async_cleanup);
608 if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) ||
609 !ndev->netdev_ops->ndo_poll_controller) {
610 np_err(np, "%s doesn't support polling, aborting\n",
617 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
623 sema_init(&npinfo->dev_lock, 1);
624 skb_queue_head_init(&npinfo->txq);
625 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
627 atomic_set(&npinfo->refcnt, 1);
629 ops = np->dev->netdev_ops;
630 if (ops->ndo_netpoll_setup) {
631 err = ops->ndo_netpoll_setup(ndev, npinfo);
636 npinfo = rtnl_dereference(ndev->npinfo);
637 atomic_inc(&npinfo->refcnt);
640 npinfo->netpoll = np;
642 /* last thing to do is link it to the net device structure */
643 rcu_assign_pointer(ndev->npinfo, npinfo);
652 EXPORT_SYMBOL_GPL(__netpoll_setup);
654 int netpoll_setup(struct netpoll *np)
656 struct net_device *ndev = NULL;
657 struct in_device *in_dev;
662 struct net *net = current->nsproxy->net_ns;
663 ndev = __dev_get_by_name(net, np->dev_name);
666 np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
672 if (netdev_master_upper_dev_get(ndev)) {
673 np_err(np, "%s is a slave device, aborting\n", np->dev_name);
678 if (!netif_running(ndev)) {
679 unsigned long atmost, atleast;
681 np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
683 err = dev_open(ndev);
686 np_err(np, "failed to open %s\n", ndev->name);
691 atleast = jiffies + HZ/10;
692 atmost = jiffies + carrier_timeout * HZ;
693 while (!netif_carrier_ok(ndev)) {
694 if (time_after(jiffies, atmost)) {
695 np_notice(np, "timeout waiting for carrier\n");
701 /* If carrier appears to come up instantly, we don't
702 * trust it and pause so that we don't pump all our
703 * queued console messages into the bitbucket.
706 if (time_before(jiffies, atleast)) {
707 np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
713 if (!np->local_ip.ip) {
715 in_dev = __in_dev_get_rtnl(ndev);
717 if (!in_dev || !in_dev->ifa_list) {
718 np_err(np, "no IP address for %s, aborting\n",
724 np->local_ip.ip = in_dev->ifa_list->ifa_local;
725 np_info(np, "local IP %pI4\n", &np->local_ip.ip);
727 #if IS_ENABLED(CONFIG_IPV6)
728 struct inet6_dev *idev;
731 idev = __in6_dev_get(ndev);
733 struct inet6_ifaddr *ifp;
735 read_lock_bh(&idev->lock);
736 list_for_each_entry(ifp, &idev->addr_list, if_list) {
737 if (ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL)
739 np->local_ip.in6 = ifp->addr;
743 read_unlock_bh(&idev->lock);
746 np_err(np, "no IPv6 address for %s, aborting\n",
750 np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
752 np_err(np, "IPv6 is not supported %s, aborting\n",
760 /* fill up the skb queue */
763 err = __netpoll_setup(np, ndev);
776 EXPORT_SYMBOL(netpoll_setup);
778 static int __init netpoll_init(void)
780 skb_queue_head_init(&skb_pool);
783 core_initcall(netpoll_init);
785 static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
787 struct netpoll_info *npinfo =
788 container_of(rcu_head, struct netpoll_info, rcu);
790 skb_queue_purge(&npinfo->txq);
792 /* we can't call cancel_delayed_work_sync here, as we are in softirq */
793 cancel_delayed_work(&npinfo->tx_work);
795 /* clean after last, unfinished work */
796 __skb_queue_purge(&npinfo->txq);
797 /* now cancel it again */
798 cancel_delayed_work(&npinfo->tx_work);
802 void __netpoll_cleanup(struct netpoll *np)
804 struct netpoll_info *npinfo;
806 /* rtnl_dereference would be preferable here but
807 * rcu_cleanup_netpoll path can put us in here safely without
808 * holding the rtnl, so plain rcu_dereference it is
810 npinfo = rtnl_dereference(np->dev->npinfo);
814 synchronize_srcu(&netpoll_srcu);
816 if (atomic_dec_and_test(&npinfo->refcnt)) {
817 const struct net_device_ops *ops;
819 ops = np->dev->netdev_ops;
820 if (ops->ndo_netpoll_cleanup)
821 ops->ndo_netpoll_cleanup(np->dev);
823 RCU_INIT_POINTER(np->dev->npinfo, NULL);
824 call_rcu_bh(&npinfo->rcu, rcu_cleanup_netpoll_info);
827 EXPORT_SYMBOL_GPL(__netpoll_cleanup);
829 static void netpoll_async_cleanup(struct work_struct *work)
831 struct netpoll *np = container_of(work, struct netpoll, cleanup_work);
834 __netpoll_cleanup(np);
839 void __netpoll_free_async(struct netpoll *np)
841 schedule_work(&np->cleanup_work);
843 EXPORT_SYMBOL_GPL(__netpoll_free_async);
845 void netpoll_cleanup(struct netpoll *np)
850 __netpoll_cleanup(np);
856 EXPORT_SYMBOL(netpoll_cleanup);