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 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;
108 unsigned int q_index;
110 if (!netif_device_present(dev) || !netif_running(dev)) {
115 local_irq_save(flags);
116 /* check if skb->queue_mapping is still valid */
117 q_index = skb_get_queue_mapping(skb);
118 if (unlikely(q_index >= dev->real_num_tx_queues)) {
119 q_index = q_index % dev->real_num_tx_queues;
120 skb_set_queue_mapping(skb, q_index);
122 txq = netdev_get_tx_queue(dev, q_index);
123 HARD_TX_LOCK(dev, txq, smp_processor_id());
124 if (netif_xmit_frozen_or_stopped(txq) ||
125 netpoll_start_xmit(skb, dev, txq) != NETDEV_TX_OK) {
126 skb_queue_head(&npinfo->txq, skb);
127 HARD_TX_UNLOCK(dev, txq);
128 local_irq_restore(flags);
130 schedule_delayed_work(&npinfo->tx_work, HZ/10);
133 HARD_TX_UNLOCK(dev, txq);
134 local_irq_restore(flags);
138 static void poll_one_napi(struct napi_struct *napi)
142 /* If we set this bit but see that it has already been set,
143 * that indicates that napi has been disabled and we need
144 * to abort this operation
146 if (test_and_set_bit(NAPI_STATE_NPSVC, &napi->state))
149 /* We explicilty pass the polling call a budget of 0 to
150 * indicate that we are clearing the Tx path only.
152 work = napi->poll(napi, 0);
153 WARN_ONCE(work, "%pF exceeded budget in poll\n", napi->poll);
154 trace_napi_poll(napi, work, 0);
156 clear_bit(NAPI_STATE_NPSVC, &napi->state);
159 static void poll_napi(struct net_device *dev)
161 struct napi_struct *napi;
162 int cpu = smp_processor_id();
164 list_for_each_entry(napi, &dev->napi_list, dev_list) {
165 if (cmpxchg(&napi->poll_owner, -1, cpu) == -1) {
167 smp_store_release(&napi->poll_owner, -1);
172 void netpoll_poll_dev(struct net_device *dev)
174 struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
175 const struct net_device_ops *ops;
177 /* Don't do any rx activity if the dev_lock mutex is held
178 * the dev_open/close paths use this to block netpoll activity
179 * while changing device state
181 if (!ni || down_trylock(&ni->dev_lock))
184 if (!netif_running(dev)) {
189 ops = dev->netdev_ops;
190 if (ops->ndo_poll_controller)
191 ops->ndo_poll_controller(dev);
197 zap_completion_queue();
199 EXPORT_SYMBOL(netpoll_poll_dev);
201 void netpoll_poll_disable(struct net_device *dev)
203 struct netpoll_info *ni;
206 idx = srcu_read_lock(&netpoll_srcu);
207 ni = srcu_dereference(dev->npinfo, &netpoll_srcu);
210 srcu_read_unlock(&netpoll_srcu, idx);
212 EXPORT_SYMBOL(netpoll_poll_disable);
214 void netpoll_poll_enable(struct net_device *dev)
216 struct netpoll_info *ni;
218 ni = rcu_dereference(dev->npinfo);
223 EXPORT_SYMBOL(netpoll_poll_enable);
225 static void refill_skbs(void)
230 spin_lock_irqsave(&skb_pool.lock, flags);
231 while (skb_pool.qlen < MAX_SKBS) {
232 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
236 __skb_queue_tail(&skb_pool, skb);
238 spin_unlock_irqrestore(&skb_pool.lock, flags);
241 static void zap_completion_queue(void)
244 struct softnet_data *sd = &get_cpu_var(softnet_data);
246 if (sd->completion_queue) {
247 struct sk_buff *clist;
249 local_irq_save(flags);
250 clist = sd->completion_queue;
251 sd->completion_queue = NULL;
252 local_irq_restore(flags);
254 while (clist != NULL) {
255 struct sk_buff *skb = clist;
257 if (!skb_irq_freeable(skb)) {
258 refcount_set(&skb->users, 1);
259 dev_kfree_skb_any(skb); /* put this one back */
266 put_cpu_var(softnet_data);
269 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
274 zap_completion_queue();
278 skb = alloc_skb(len, GFP_ATOMIC);
280 skb = skb_dequeue(&skb_pool);
284 netpoll_poll_dev(np->dev);
290 refcount_set(&skb->users, 1);
291 skb_reserve(skb, reserve);
295 static int netpoll_owner_active(struct net_device *dev)
297 struct napi_struct *napi;
299 list_for_each_entry(napi, &dev->napi_list, dev_list) {
300 if (napi->poll_owner == smp_processor_id())
306 /* call with IRQ disabled */
307 void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb,
308 struct net_device *dev)
310 int status = NETDEV_TX_BUSY;
312 /* It is up to the caller to keep npinfo alive. */
313 struct netpoll_info *npinfo;
316 lockdep_assert_irqs_disabled();
318 npinfo = rcu_dereference_bh(np->dev->npinfo);
319 if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
320 dev_kfree_skb_irq(skb);
324 /* don't get messages out of order, and no recursion */
325 if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
326 struct netdev_queue *txq;
328 txq = netdev_pick_tx(dev, skb, NULL);
330 /* try until next clock tick */
331 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
332 tries > 0; --tries) {
333 if (HARD_TX_TRYLOCK(dev, txq)) {
334 if (!netif_xmit_stopped(txq))
335 status = netpoll_start_xmit(skb, dev, txq);
337 HARD_TX_UNLOCK(dev, txq);
339 if (status == NETDEV_TX_OK)
344 /* tickle device maybe there is some cleanup */
345 netpoll_poll_dev(np->dev);
347 udelay(USEC_PER_POLL);
350 WARN_ONCE(!irqs_disabled(),
351 "netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pF)\n",
352 dev->name, dev->netdev_ops->ndo_start_xmit);
356 if (status != NETDEV_TX_OK) {
357 skb_queue_tail(&npinfo->txq, skb);
358 schedule_delayed_work(&npinfo->tx_work,0);
360 rcu_read_unlock_bh();
362 EXPORT_SYMBOL(netpoll_send_skb_on_dev);
364 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
366 int total_len, ip_len, udp_len;
371 static atomic_t ip_ident;
372 struct ipv6hdr *ip6h;
374 WARN_ON_ONCE(!irqs_disabled());
376 udp_len = len + sizeof(*udph);
378 ip_len = udp_len + sizeof(*ip6h);
380 ip_len = udp_len + sizeof(*iph);
382 total_len = ip_len + LL_RESERVED_SPACE(np->dev);
384 skb = find_skb(np, total_len + np->dev->needed_tailroom,
389 skb_copy_to_linear_data(skb, msg, len);
392 skb_push(skb, sizeof(*udph));
393 skb_reset_transport_header(skb);
395 udph->source = htons(np->local_port);
396 udph->dest = htons(np->remote_port);
397 udph->len = htons(udp_len);
401 udph->check = csum_ipv6_magic(&np->local_ip.in6,
403 udp_len, IPPROTO_UDP,
404 csum_partial(udph, udp_len, 0));
405 if (udph->check == 0)
406 udph->check = CSUM_MANGLED_0;
408 skb_push(skb, sizeof(*ip6h));
409 skb_reset_network_header(skb);
410 ip6h = ipv6_hdr(skb);
412 /* ip6h->version = 6; ip6h->priority = 0; */
413 put_unaligned(0x60, (unsigned char *)ip6h);
414 ip6h->flow_lbl[0] = 0;
415 ip6h->flow_lbl[1] = 0;
416 ip6h->flow_lbl[2] = 0;
418 ip6h->payload_len = htons(sizeof(struct udphdr) + len);
419 ip6h->nexthdr = IPPROTO_UDP;
420 ip6h->hop_limit = 32;
421 ip6h->saddr = np->local_ip.in6;
422 ip6h->daddr = np->remote_ip.in6;
424 eth = skb_push(skb, ETH_HLEN);
425 skb_reset_mac_header(skb);
426 skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
429 udph->check = csum_tcpudp_magic(np->local_ip.ip,
431 udp_len, IPPROTO_UDP,
432 csum_partial(udph, udp_len, 0));
433 if (udph->check == 0)
434 udph->check = CSUM_MANGLED_0;
436 skb_push(skb, sizeof(*iph));
437 skb_reset_network_header(skb);
440 /* iph->version = 4; iph->ihl = 5; */
441 put_unaligned(0x45, (unsigned char *)iph);
443 put_unaligned(htons(ip_len), &(iph->tot_len));
444 iph->id = htons(atomic_inc_return(&ip_ident));
447 iph->protocol = IPPROTO_UDP;
449 put_unaligned(np->local_ip.ip, &(iph->saddr));
450 put_unaligned(np->remote_ip.ip, &(iph->daddr));
451 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
453 eth = skb_push(skb, ETH_HLEN);
454 skb_reset_mac_header(skb);
455 skb->protocol = eth->h_proto = htons(ETH_P_IP);
458 ether_addr_copy(eth->h_source, np->dev->dev_addr);
459 ether_addr_copy(eth->h_dest, np->remote_mac);
463 netpoll_send_skb(np, skb);
465 EXPORT_SYMBOL(netpoll_send_udp);
467 void netpoll_print_options(struct netpoll *np)
469 np_info(np, "local port %d\n", np->local_port);
471 np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
473 np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
474 np_info(np, "interface '%s'\n", np->dev_name);
475 np_info(np, "remote port %d\n", np->remote_port);
477 np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
479 np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
480 np_info(np, "remote ethernet address %pM\n", np->remote_mac);
482 EXPORT_SYMBOL(netpoll_print_options);
484 static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
488 if (!strchr(str, ':') &&
489 in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
493 if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
494 #if IS_ENABLED(CONFIG_IPV6)
504 int netpoll_parse_options(struct netpoll *np, char *opt)
506 char *cur=opt, *delim;
508 bool ipversion_set = false;
511 if ((delim = strchr(cur, '@')) == NULL)
514 if (kstrtou16(cur, 10, &np->local_port))
521 ipversion_set = true;
522 if ((delim = strchr(cur, '/')) == NULL)
525 ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
529 np->ipv6 = (bool)ipv6;
535 /* parse out dev name */
536 if ((delim = strchr(cur, ',')) == NULL)
539 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
546 if ((delim = strchr(cur, '@')) == NULL)
549 if (*cur == ' ' || *cur == '\t')
550 np_info(np, "warning: whitespace is not allowed\n");
551 if (kstrtou16(cur, 10, &np->remote_port))
558 if ((delim = strchr(cur, '/')) == NULL)
561 ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
564 else if (ipversion_set && np->ipv6 != (bool)ipv6)
567 np->ipv6 = (bool)ipv6;
572 if (!mac_pton(cur, np->remote_mac))
576 netpoll_print_options(np);
581 np_info(np, "couldn't parse config at '%s'!\n", cur);
584 EXPORT_SYMBOL(netpoll_parse_options);
586 int __netpoll_setup(struct netpoll *np, struct net_device *ndev)
588 struct netpoll_info *npinfo;
589 const struct net_device_ops *ops;
593 strlcpy(np->dev_name, ndev->name, IFNAMSIZ);
594 INIT_WORK(&np->cleanup_work, netpoll_async_cleanup);
596 if (ndev->priv_flags & IFF_DISABLE_NETPOLL) {
597 np_err(np, "%s doesn't support polling, aborting\n",
604 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
610 sema_init(&npinfo->dev_lock, 1);
611 skb_queue_head_init(&npinfo->txq);
612 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
614 refcount_set(&npinfo->refcnt, 1);
616 ops = np->dev->netdev_ops;
617 if (ops->ndo_netpoll_setup) {
618 err = ops->ndo_netpoll_setup(ndev, npinfo);
623 npinfo = rtnl_dereference(ndev->npinfo);
624 refcount_inc(&npinfo->refcnt);
627 npinfo->netpoll = np;
629 /* last thing to do is link it to the net device structure */
630 rcu_assign_pointer(ndev->npinfo, npinfo);
639 EXPORT_SYMBOL_GPL(__netpoll_setup);
641 int netpoll_setup(struct netpoll *np)
643 struct net_device *ndev = NULL;
644 struct in_device *in_dev;
648 if (np->dev_name[0]) {
649 struct net *net = current->nsproxy->net_ns;
650 ndev = __dev_get_by_name(net, np->dev_name);
653 np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
659 if (netdev_master_upper_dev_get(ndev)) {
660 np_err(np, "%s is a slave device, aborting\n", np->dev_name);
665 if (!netif_running(ndev)) {
666 unsigned long atmost, atleast;
668 np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
670 err = dev_open(ndev);
673 np_err(np, "failed to open %s\n", ndev->name);
678 atleast = jiffies + HZ/10;
679 atmost = jiffies + carrier_timeout * HZ;
680 while (!netif_carrier_ok(ndev)) {
681 if (time_after(jiffies, atmost)) {
682 np_notice(np, "timeout waiting for carrier\n");
688 /* If carrier appears to come up instantly, we don't
689 * trust it and pause so that we don't pump all our
690 * queued console messages into the bitbucket.
693 if (time_before(jiffies, atleast)) {
694 np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
700 if (!np->local_ip.ip) {
702 in_dev = __in_dev_get_rtnl(ndev);
704 if (!in_dev || !in_dev->ifa_list) {
705 np_err(np, "no IP address for %s, aborting\n",
711 np->local_ip.ip = in_dev->ifa_list->ifa_local;
712 np_info(np, "local IP %pI4\n", &np->local_ip.ip);
714 #if IS_ENABLED(CONFIG_IPV6)
715 struct inet6_dev *idev;
718 idev = __in6_dev_get(ndev);
720 struct inet6_ifaddr *ifp;
722 read_lock_bh(&idev->lock);
723 list_for_each_entry(ifp, &idev->addr_list, if_list) {
724 if (ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL)
726 np->local_ip.in6 = ifp->addr;
730 read_unlock_bh(&idev->lock);
733 np_err(np, "no IPv6 address for %s, aborting\n",
737 np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
739 np_err(np, "IPv6 is not supported %s, aborting\n",
747 /* fill up the skb queue */
750 err = __netpoll_setup(np, ndev);
763 EXPORT_SYMBOL(netpoll_setup);
765 static int __init netpoll_init(void)
767 skb_queue_head_init(&skb_pool);
770 core_initcall(netpoll_init);
772 static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
774 struct netpoll_info *npinfo =
775 container_of(rcu_head, struct netpoll_info, rcu);
777 skb_queue_purge(&npinfo->txq);
779 /* we can't call cancel_delayed_work_sync here, as we are in softirq */
780 cancel_delayed_work(&npinfo->tx_work);
782 /* clean after last, unfinished work */
783 __skb_queue_purge(&npinfo->txq);
784 /* now cancel it again */
785 cancel_delayed_work(&npinfo->tx_work);
789 void __netpoll_cleanup(struct netpoll *np)
791 struct netpoll_info *npinfo;
793 /* rtnl_dereference would be preferable here but
794 * rcu_cleanup_netpoll path can put us in here safely without
795 * holding the rtnl, so plain rcu_dereference it is
797 npinfo = rtnl_dereference(np->dev->npinfo);
801 synchronize_srcu(&netpoll_srcu);
803 if (refcount_dec_and_test(&npinfo->refcnt)) {
804 const struct net_device_ops *ops;
806 ops = np->dev->netdev_ops;
807 if (ops->ndo_netpoll_cleanup)
808 ops->ndo_netpoll_cleanup(np->dev);
810 RCU_INIT_POINTER(np->dev->npinfo, NULL);
811 call_rcu_bh(&npinfo->rcu, rcu_cleanup_netpoll_info);
813 RCU_INIT_POINTER(np->dev->npinfo, NULL);
815 EXPORT_SYMBOL_GPL(__netpoll_cleanup);
817 static void netpoll_async_cleanup(struct work_struct *work)
819 struct netpoll *np = container_of(work, struct netpoll, cleanup_work);
822 __netpoll_cleanup(np);
827 void __netpoll_free_async(struct netpoll *np)
829 schedule_work(&np->cleanup_work);
831 EXPORT_SYMBOL_GPL(__netpoll_free_async);
833 void netpoll_cleanup(struct netpoll *np)
838 __netpoll_cleanup(np);
844 EXPORT_SYMBOL(netpoll_cleanup);