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>
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);
61 static void netpoll_async_cleanup(struct work_struct *work);
63 static unsigned int carrier_timeout = 4;
64 module_param(carrier_timeout, uint, 0644);
66 #define np_info(np, fmt, ...) \
67 pr_info("%s: " fmt, np->name, ##__VA_ARGS__)
68 #define np_err(np, fmt, ...) \
69 pr_err("%s: " fmt, np->name, ##__VA_ARGS__)
70 #define np_notice(np, fmt, ...) \
71 pr_notice("%s: " fmt, np->name, ##__VA_ARGS__)
73 static int netpoll_start_xmit(struct sk_buff *skb, struct net_device *dev,
74 struct netdev_queue *txq)
76 int 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);
140 * Check whether delayed processing was scheduled for our NIC. If so,
141 * we attempt to grab the poll lock and use ->poll() to pump the card.
142 * If this fails, either we've recursed in ->poll() or it's already
143 * running on another CPU.
145 * Note: we don't mask interrupts with this lock because we're using
146 * trylock here and interrupts are already disabled in the softirq
147 * case. Further, we test the poll_owner to avoid recursion on UP
148 * systems where the lock doesn't exist.
150 static void poll_one_napi(struct napi_struct *napi)
154 /* net_rx_action's ->poll() invocations and our's are
155 * synchronized by this test which is only made while
156 * holding the napi->poll_lock.
158 if (!test_bit(NAPI_STATE_SCHED, &napi->state))
161 /* If we set this bit but see that it has already been set,
162 * that indicates that napi has been disabled and we need
163 * to abort this operation
165 if (test_and_set_bit(NAPI_STATE_NPSVC, &napi->state))
168 /* We explicilty pass the polling call a budget of 0 to
169 * indicate that we are clearing the Tx path only.
171 work = napi->poll(napi, 0);
172 WARN_ONCE(work, "%pF exceeded budget in poll\n", napi->poll);
173 trace_napi_poll(napi, work, 0);
175 clear_bit(NAPI_STATE_NPSVC, &napi->state);
178 static void poll_napi(struct net_device *dev)
180 struct napi_struct *napi;
182 list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) {
183 if (napi->poll_owner != smp_processor_id() &&
184 spin_trylock(&napi->poll_lock)) {
186 spin_unlock(&napi->poll_lock);
191 static void netpoll_poll_dev(struct net_device *dev)
193 const struct net_device_ops *ops;
194 struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
196 /* Don't do any rx activity if the dev_lock mutex is held
197 * the dev_open/close paths use this to block netpoll activity
198 * while changing device state
200 if (down_trylock(&ni->dev_lock))
203 if (!netif_running(dev)) {
208 ops = dev->netdev_ops;
209 if (!ops->ndo_poll_controller) {
214 /* Process pending work on NIC */
215 ops->ndo_poll_controller(dev);
221 zap_completion_queue();
224 void netpoll_poll_disable(struct net_device *dev)
226 struct netpoll_info *ni;
229 idx = srcu_read_lock(&netpoll_srcu);
230 ni = srcu_dereference(dev->npinfo, &netpoll_srcu);
233 srcu_read_unlock(&netpoll_srcu, idx);
235 EXPORT_SYMBOL(netpoll_poll_disable);
237 void netpoll_poll_enable(struct net_device *dev)
239 struct netpoll_info *ni;
241 ni = rcu_dereference(dev->npinfo);
246 EXPORT_SYMBOL(netpoll_poll_enable);
248 static void refill_skbs(void)
253 spin_lock_irqsave(&skb_pool.lock, flags);
254 while (skb_pool.qlen < MAX_SKBS) {
255 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
259 __skb_queue_tail(&skb_pool, skb);
261 spin_unlock_irqrestore(&skb_pool.lock, flags);
264 static void zap_completion_queue(void)
267 struct softnet_data *sd = &get_cpu_var(softnet_data);
269 if (sd->completion_queue) {
270 struct sk_buff *clist;
272 local_irq_save(flags);
273 clist = sd->completion_queue;
274 sd->completion_queue = NULL;
275 local_irq_restore(flags);
277 while (clist != NULL) {
278 struct sk_buff *skb = clist;
280 if (!skb_irq_freeable(skb)) {
281 atomic_inc(&skb->users);
282 dev_kfree_skb_any(skb); /* put this one back */
289 put_cpu_var(softnet_data);
292 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
297 zap_completion_queue();
301 skb = alloc_skb(len, GFP_ATOMIC);
303 skb = skb_dequeue(&skb_pool);
307 netpoll_poll_dev(np->dev);
313 atomic_set(&skb->users, 1);
314 skb_reserve(skb, reserve);
318 static int netpoll_owner_active(struct net_device *dev)
320 struct napi_struct *napi;
322 list_for_each_entry(napi, &dev->napi_list, dev_list) {
323 if (napi->poll_owner == smp_processor_id())
329 /* call with IRQ disabled */
330 void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb,
331 struct net_device *dev)
333 int status = NETDEV_TX_BUSY;
335 /* It is up to the caller to keep npinfo alive. */
336 struct netpoll_info *npinfo;
338 WARN_ON_ONCE(!irqs_disabled());
340 npinfo = rcu_dereference_bh(np->dev->npinfo);
341 if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
342 dev_kfree_skb_irq(skb);
346 /* don't get messages out of order, and no recursion */
347 if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
348 struct netdev_queue *txq;
350 txq = netdev_pick_tx(dev, skb, NULL);
352 /* try until next clock tick */
353 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
354 tries > 0; --tries) {
355 if (HARD_TX_TRYLOCK(dev, txq)) {
356 if (!netif_xmit_stopped(txq))
357 status = netpoll_start_xmit(skb, dev, txq);
359 HARD_TX_UNLOCK(dev, txq);
361 if (dev_xmit_complete(status))
366 /* tickle device maybe there is some cleanup */
367 netpoll_poll_dev(np->dev);
369 udelay(USEC_PER_POLL);
372 WARN_ONCE(!irqs_disabled(),
373 "netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pF)\n",
374 dev->name, dev->netdev_ops->ndo_start_xmit);
378 if (!dev_xmit_complete(status)) {
379 skb_queue_tail(&npinfo->txq, skb);
380 schedule_delayed_work(&npinfo->tx_work,0);
383 EXPORT_SYMBOL(netpoll_send_skb_on_dev);
385 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
387 int total_len, ip_len, udp_len;
392 static atomic_t ip_ident;
393 struct ipv6hdr *ip6h;
395 WARN_ON_ONCE(!irqs_disabled());
397 udp_len = len + sizeof(*udph);
399 ip_len = udp_len + sizeof(*ip6h);
401 ip_len = udp_len + sizeof(*iph);
403 total_len = ip_len + LL_RESERVED_SPACE(np->dev);
405 skb = find_skb(np, total_len + np->dev->needed_tailroom,
410 skb_copy_to_linear_data(skb, msg, len);
413 skb_push(skb, sizeof(*udph));
414 skb_reset_transport_header(skb);
416 udph->source = htons(np->local_port);
417 udph->dest = htons(np->remote_port);
418 udph->len = htons(udp_len);
422 udph->check = csum_ipv6_magic(&np->local_ip.in6,
424 udp_len, IPPROTO_UDP,
425 csum_partial(udph, udp_len, 0));
426 if (udph->check == 0)
427 udph->check = CSUM_MANGLED_0;
429 skb_push(skb, sizeof(*ip6h));
430 skb_reset_network_header(skb);
431 ip6h = ipv6_hdr(skb);
433 /* ip6h->version = 6; ip6h->priority = 0; */
434 put_unaligned(0x60, (unsigned char *)ip6h);
435 ip6h->flow_lbl[0] = 0;
436 ip6h->flow_lbl[1] = 0;
437 ip6h->flow_lbl[2] = 0;
439 ip6h->payload_len = htons(sizeof(struct udphdr) + len);
440 ip6h->nexthdr = IPPROTO_UDP;
441 ip6h->hop_limit = 32;
442 ip6h->saddr = np->local_ip.in6;
443 ip6h->daddr = np->remote_ip.in6;
445 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
446 skb_reset_mac_header(skb);
447 skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
450 udph->check = csum_tcpudp_magic(np->local_ip.ip,
452 udp_len, IPPROTO_UDP,
453 csum_partial(udph, udp_len, 0));
454 if (udph->check == 0)
455 udph->check = CSUM_MANGLED_0;
457 skb_push(skb, sizeof(*iph));
458 skb_reset_network_header(skb);
461 /* iph->version = 4; iph->ihl = 5; */
462 put_unaligned(0x45, (unsigned char *)iph);
464 put_unaligned(htons(ip_len), &(iph->tot_len));
465 iph->id = htons(atomic_inc_return(&ip_ident));
468 iph->protocol = IPPROTO_UDP;
470 put_unaligned(np->local_ip.ip, &(iph->saddr));
471 put_unaligned(np->remote_ip.ip, &(iph->daddr));
472 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
474 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
475 skb_reset_mac_header(skb);
476 skb->protocol = eth->h_proto = htons(ETH_P_IP);
479 ether_addr_copy(eth->h_source, np->dev->dev_addr);
480 ether_addr_copy(eth->h_dest, np->remote_mac);
484 netpoll_send_skb(np, skb);
486 EXPORT_SYMBOL(netpoll_send_udp);
488 void netpoll_print_options(struct netpoll *np)
490 np_info(np, "local port %d\n", np->local_port);
492 np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
494 np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
495 np_info(np, "interface '%s'\n", np->dev_name);
496 np_info(np, "remote port %d\n", np->remote_port);
498 np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
500 np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
501 np_info(np, "remote ethernet address %pM\n", np->remote_mac);
503 EXPORT_SYMBOL(netpoll_print_options);
505 static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
509 if (!strchr(str, ':') &&
510 in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
514 if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
515 #if IS_ENABLED(CONFIG_IPV6)
525 int netpoll_parse_options(struct netpoll *np, char *opt)
527 char *cur=opt, *delim;
529 bool ipversion_set = false;
532 if ((delim = strchr(cur, '@')) == NULL)
535 if (kstrtou16(cur, 10, &np->local_port))
542 ipversion_set = true;
543 if ((delim = strchr(cur, '/')) == NULL)
546 ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
550 np->ipv6 = (bool)ipv6;
556 /* parse out dev name */
557 if ((delim = strchr(cur, ',')) == NULL)
560 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
567 if ((delim = strchr(cur, '@')) == NULL)
570 if (*cur == ' ' || *cur == '\t')
571 np_info(np, "warning: whitespace is not allowed\n");
572 if (kstrtou16(cur, 10, &np->remote_port))
579 if ((delim = strchr(cur, '/')) == NULL)
582 ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
585 else if (ipversion_set && np->ipv6 != (bool)ipv6)
588 np->ipv6 = (bool)ipv6;
593 if (!mac_pton(cur, np->remote_mac))
597 netpoll_print_options(np);
602 np_info(np, "couldn't parse config at '%s'!\n", cur);
605 EXPORT_SYMBOL(netpoll_parse_options);
607 int __netpoll_setup(struct netpoll *np, struct net_device *ndev)
609 struct netpoll_info *npinfo;
610 const struct net_device_ops *ops;
614 strlcpy(np->dev_name, ndev->name, IFNAMSIZ);
615 INIT_WORK(&np->cleanup_work, netpoll_async_cleanup);
617 if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) ||
618 !ndev->netdev_ops->ndo_poll_controller) {
619 np_err(np, "%s doesn't support polling, aborting\n",
626 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
632 sema_init(&npinfo->dev_lock, 1);
633 skb_queue_head_init(&npinfo->txq);
634 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
636 atomic_set(&npinfo->refcnt, 1);
638 ops = np->dev->netdev_ops;
639 if (ops->ndo_netpoll_setup) {
640 err = ops->ndo_netpoll_setup(ndev, npinfo);
645 npinfo = rtnl_dereference(ndev->npinfo);
646 atomic_inc(&npinfo->refcnt);
649 npinfo->netpoll = np;
651 /* last thing to do is link it to the net device structure */
652 rcu_assign_pointer(ndev->npinfo, npinfo);
661 EXPORT_SYMBOL_GPL(__netpoll_setup);
663 int netpoll_setup(struct netpoll *np)
665 struct net_device *ndev = NULL, *dev = NULL;
666 struct net *net = current->nsproxy->net_ns;
667 struct in_device *in_dev;
672 ndev = __dev_get_by_name(net, np->dev_name);
675 np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
681 /* bring up DSA management network devices up first */
682 for_each_netdev(net, dev) {
683 if (!netdev_uses_dsa(dev))
686 err = dev_change_flags(dev, dev->flags | IFF_UP);
688 np_err(np, "%s failed to open %s\n",
689 np->dev_name, dev->name);
694 if (netdev_master_upper_dev_get(ndev)) {
695 np_err(np, "%s is a slave device, aborting\n", np->dev_name);
700 if (!netif_running(ndev)) {
701 unsigned long atmost, atleast;
703 np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
705 err = dev_open(ndev);
708 np_err(np, "failed to open %s\n", ndev->name);
713 atleast = jiffies + HZ/10;
714 atmost = jiffies + carrier_timeout * HZ;
715 while (!netif_carrier_ok(ndev)) {
716 if (time_after(jiffies, atmost)) {
717 np_notice(np, "timeout waiting for carrier\n");
723 /* If carrier appears to come up instantly, we don't
724 * trust it and pause so that we don't pump all our
725 * queued console messages into the bitbucket.
728 if (time_before(jiffies, atleast)) {
729 np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
735 if (!np->local_ip.ip) {
737 in_dev = __in_dev_get_rtnl(ndev);
739 if (!in_dev || !in_dev->ifa_list) {
740 np_err(np, "no IP address for %s, aborting\n",
746 np->local_ip.ip = in_dev->ifa_list->ifa_local;
747 np_info(np, "local IP %pI4\n", &np->local_ip.ip);
749 #if IS_ENABLED(CONFIG_IPV6)
750 struct inet6_dev *idev;
753 idev = __in6_dev_get(ndev);
755 struct inet6_ifaddr *ifp;
757 read_lock_bh(&idev->lock);
758 list_for_each_entry(ifp, &idev->addr_list, if_list) {
759 if (ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL)
761 np->local_ip.in6 = ifp->addr;
765 read_unlock_bh(&idev->lock);
768 np_err(np, "no IPv6 address for %s, aborting\n",
772 np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
774 np_err(np, "IPv6 is not supported %s, aborting\n",
782 /* fill up the skb queue */
785 err = __netpoll_setup(np, ndev);
798 EXPORT_SYMBOL(netpoll_setup);
800 static int __init netpoll_init(void)
802 skb_queue_head_init(&skb_pool);
805 core_initcall(netpoll_init);
807 static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
809 struct netpoll_info *npinfo =
810 container_of(rcu_head, struct netpoll_info, rcu);
812 skb_queue_purge(&npinfo->txq);
814 /* we can't call cancel_delayed_work_sync here, as we are in softirq */
815 cancel_delayed_work(&npinfo->tx_work);
817 /* clean after last, unfinished work */
818 __skb_queue_purge(&npinfo->txq);
819 /* now cancel it again */
820 cancel_delayed_work(&npinfo->tx_work);
824 void __netpoll_cleanup(struct netpoll *np)
826 struct netpoll_info *npinfo;
828 /* rtnl_dereference would be preferable here but
829 * rcu_cleanup_netpoll path can put us in here safely without
830 * holding the rtnl, so plain rcu_dereference it is
832 npinfo = rtnl_dereference(np->dev->npinfo);
836 synchronize_srcu(&netpoll_srcu);
838 if (atomic_dec_and_test(&npinfo->refcnt)) {
839 const struct net_device_ops *ops;
841 ops = np->dev->netdev_ops;
842 if (ops->ndo_netpoll_cleanup)
843 ops->ndo_netpoll_cleanup(np->dev);
845 RCU_INIT_POINTER(np->dev->npinfo, NULL);
846 call_rcu_bh(&npinfo->rcu, rcu_cleanup_netpoll_info);
848 RCU_INIT_POINTER(np->dev->npinfo, NULL);
850 EXPORT_SYMBOL_GPL(__netpoll_cleanup);
852 static void netpoll_async_cleanup(struct work_struct *work)
854 struct netpoll *np = container_of(work, struct netpoll, cleanup_work);
857 __netpoll_cleanup(np);
862 void __netpoll_free_async(struct netpoll *np)
864 schedule_work(&np->cleanup_work);
866 EXPORT_SYMBOL_GPL(__netpoll_free_async);
868 void netpoll_cleanup(struct netpoll *np)
873 __netpoll_cleanup(np);
879 EXPORT_SYMBOL(netpoll_cleanup);