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 void queue_process(struct work_struct *work)
74 struct netpoll_info *npinfo =
75 container_of(work, struct netpoll_info, tx_work.work);
79 while ((skb = skb_dequeue(&npinfo->txq))) {
80 struct net_device *dev = skb->dev;
81 const struct net_device_ops *ops = dev->netdev_ops;
82 struct netdev_queue *txq;
84 if (!netif_device_present(dev) || !netif_running(dev)) {
89 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
91 local_irq_save(flags);
92 __netif_tx_lock(txq, smp_processor_id());
93 if (netif_xmit_frozen_or_stopped(txq) ||
94 ops->ndo_start_xmit(skb, dev) != NETDEV_TX_OK) {
95 skb_queue_head(&npinfo->txq, skb);
96 __netif_tx_unlock(txq);
97 local_irq_restore(flags);
99 schedule_delayed_work(&npinfo->tx_work, HZ/10);
102 __netif_tx_unlock(txq);
103 local_irq_restore(flags);
108 * Check whether delayed processing was scheduled for our NIC. If so,
109 * we attempt to grab the poll lock and use ->poll() to pump the card.
110 * If this fails, either we've recursed in ->poll() or it's already
111 * running on another CPU.
113 * Note: we don't mask interrupts with this lock because we're using
114 * trylock here and interrupts are already disabled in the softirq
115 * case. Further, we test the poll_owner to avoid recursion on UP
116 * systems where the lock doesn't exist.
118 static int poll_one_napi(struct napi_struct *napi, int budget)
122 /* net_rx_action's ->poll() invocations and our's are
123 * synchronized by this test which is only made while
124 * holding the napi->poll_lock.
126 if (!test_bit(NAPI_STATE_SCHED, &napi->state))
129 set_bit(NAPI_STATE_NPSVC, &napi->state);
131 work = napi->poll(napi, budget);
132 WARN_ONCE(work > budget, "%pF exceeded budget in poll\n", napi->poll);
133 trace_napi_poll(napi);
135 clear_bit(NAPI_STATE_NPSVC, &napi->state);
137 return budget - work;
140 static void poll_napi(struct net_device *dev, int budget)
142 struct napi_struct *napi;
144 list_for_each_entry(napi, &dev->napi_list, dev_list) {
145 if (napi->poll_owner != smp_processor_id() &&
146 spin_trylock(&napi->poll_lock)) {
147 budget = poll_one_napi(napi, budget);
148 spin_unlock(&napi->poll_lock);
153 static void netpoll_poll_dev(struct net_device *dev)
155 const struct net_device_ops *ops;
156 struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
159 /* Don't do any rx activity if the dev_lock mutex is held
160 * the dev_open/close paths use this to block netpoll activity
161 * while changing device state
163 if (down_trylock(&ni->dev_lock))
166 if (!netif_running(dev)) {
171 ops = dev->netdev_ops;
172 if (!ops->ndo_poll_controller) {
177 /* Process pending work on NIC */
178 ops->ndo_poll_controller(dev);
180 poll_napi(dev, budget);
184 zap_completion_queue();
187 void netpoll_rx_disable(struct net_device *dev)
189 struct netpoll_info *ni;
192 idx = srcu_read_lock(&netpoll_srcu);
193 ni = srcu_dereference(dev->npinfo, &netpoll_srcu);
196 srcu_read_unlock(&netpoll_srcu, idx);
198 EXPORT_SYMBOL(netpoll_rx_disable);
200 void netpoll_rx_enable(struct net_device *dev)
202 struct netpoll_info *ni;
204 ni = rcu_dereference(dev->npinfo);
209 EXPORT_SYMBOL(netpoll_rx_enable);
211 static void refill_skbs(void)
216 spin_lock_irqsave(&skb_pool.lock, flags);
217 while (skb_pool.qlen < MAX_SKBS) {
218 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
222 __skb_queue_tail(&skb_pool, skb);
224 spin_unlock_irqrestore(&skb_pool.lock, flags);
227 static void zap_completion_queue(void)
230 struct softnet_data *sd = &get_cpu_var(softnet_data);
232 if (sd->completion_queue) {
233 struct sk_buff *clist;
235 local_irq_save(flags);
236 clist = sd->completion_queue;
237 sd->completion_queue = NULL;
238 local_irq_restore(flags);
240 while (clist != NULL) {
241 struct sk_buff *skb = clist;
243 if (skb->destructor) {
244 atomic_inc(&skb->users);
245 dev_kfree_skb_any(skb); /* put this one back */
252 put_cpu_var(softnet_data);
255 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
260 zap_completion_queue();
264 skb = alloc_skb(len, GFP_ATOMIC);
266 skb = skb_dequeue(&skb_pool);
270 netpoll_poll_dev(np->dev);
276 atomic_set(&skb->users, 1);
277 skb_reserve(skb, reserve);
281 static int netpoll_owner_active(struct net_device *dev)
283 struct napi_struct *napi;
285 list_for_each_entry(napi, &dev->napi_list, dev_list) {
286 if (napi->poll_owner == smp_processor_id())
292 /* call with IRQ disabled */
293 void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb,
294 struct net_device *dev)
296 int status = NETDEV_TX_BUSY;
298 const struct net_device_ops *ops = dev->netdev_ops;
299 /* It is up to the caller to keep npinfo alive. */
300 struct netpoll_info *npinfo;
302 WARN_ON_ONCE(!irqs_disabled());
304 npinfo = rcu_dereference_bh(np->dev->npinfo);
305 if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
310 /* don't get messages out of order, and no recursion */
311 if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
312 struct netdev_queue *txq;
314 txq = netdev_pick_tx(dev, skb, NULL);
316 /* try until next clock tick */
317 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
318 tries > 0; --tries) {
319 if (__netif_tx_trylock(txq)) {
320 if (!netif_xmit_stopped(txq)) {
321 if (vlan_tx_tag_present(skb) &&
322 !vlan_hw_offload_capable(netif_skb_features(skb),
324 skb = __vlan_put_tag(skb, skb->vlan_proto, vlan_tx_tag_get(skb));
325 if (unlikely(!skb)) {
326 /* This is actually a packet drop, but we
327 * don't want the code at the end of this
328 * function to try and re-queue a NULL skb.
330 status = NETDEV_TX_OK;
336 status = ops->ndo_start_xmit(skb, dev);
337 if (status == NETDEV_TX_OK)
338 txq_trans_update(txq);
341 __netif_tx_unlock(txq);
343 if (status == NETDEV_TX_OK)
348 /* tickle device maybe there is some cleanup */
349 netpoll_poll_dev(np->dev);
351 udelay(USEC_PER_POLL);
354 WARN_ONCE(!irqs_disabled(),
355 "netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pF)\n",
356 dev->name, ops->ndo_start_xmit);
360 if (status != NETDEV_TX_OK) {
361 skb_queue_tail(&npinfo->txq, skb);
362 schedule_delayed_work(&npinfo->tx_work,0);
365 EXPORT_SYMBOL(netpoll_send_skb_on_dev);
367 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
369 int total_len, ip_len, udp_len;
374 static atomic_t ip_ident;
375 struct ipv6hdr *ip6h;
377 udp_len = len + sizeof(*udph);
379 ip_len = udp_len + sizeof(*ip6h);
381 ip_len = udp_len + sizeof(*iph);
383 total_len = ip_len + LL_RESERVED_SPACE(np->dev);
385 skb = find_skb(np, total_len + np->dev->needed_tailroom,
390 skb_copy_to_linear_data(skb, msg, len);
393 skb_push(skb, sizeof(*udph));
394 skb_reset_transport_header(skb);
396 udph->source = htons(np->local_port);
397 udph->dest = htons(np->remote_port);
398 udph->len = htons(udp_len);
402 udph->check = csum_ipv6_magic(&np->local_ip.in6,
404 udp_len, IPPROTO_UDP,
405 csum_partial(udph, udp_len, 0));
406 if (udph->check == 0)
407 udph->check = CSUM_MANGLED_0;
409 skb_push(skb, sizeof(*ip6h));
410 skb_reset_network_header(skb);
411 ip6h = ipv6_hdr(skb);
413 /* ip6h->version = 6; ip6h->priority = 0; */
414 put_unaligned(0x60, (unsigned char *)ip6h);
415 ip6h->flow_lbl[0] = 0;
416 ip6h->flow_lbl[1] = 0;
417 ip6h->flow_lbl[2] = 0;
419 ip6h->payload_len = htons(sizeof(struct udphdr) + len);
420 ip6h->nexthdr = IPPROTO_UDP;
421 ip6h->hop_limit = 32;
422 ip6h->saddr = np->local_ip.in6;
423 ip6h->daddr = np->remote_ip.in6;
425 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
426 skb_reset_mac_header(skb);
427 skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
430 udph->check = csum_tcpudp_magic(np->local_ip.ip,
432 udp_len, IPPROTO_UDP,
433 csum_partial(udph, udp_len, 0));
434 if (udph->check == 0)
435 udph->check = CSUM_MANGLED_0;
437 skb_push(skb, sizeof(*iph));
438 skb_reset_network_header(skb);
441 /* iph->version = 4; iph->ihl = 5; */
442 put_unaligned(0x45, (unsigned char *)iph);
444 put_unaligned(htons(ip_len), &(iph->tot_len));
445 iph->id = htons(atomic_inc_return(&ip_ident));
448 iph->protocol = IPPROTO_UDP;
450 put_unaligned(np->local_ip.ip, &(iph->saddr));
451 put_unaligned(np->remote_ip.ip, &(iph->daddr));
452 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
454 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
455 skb_reset_mac_header(skb);
456 skb->protocol = eth->h_proto = htons(ETH_P_IP);
459 ether_addr_copy(eth->h_source, np->dev->dev_addr);
460 ether_addr_copy(eth->h_dest, np->remote_mac);
464 netpoll_send_skb(np, skb);
466 EXPORT_SYMBOL(netpoll_send_udp);
468 void netpoll_print_options(struct netpoll *np)
470 np_info(np, "local port %d\n", np->local_port);
472 np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
474 np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
475 np_info(np, "interface '%s'\n", np->dev_name);
476 np_info(np, "remote port %d\n", np->remote_port);
478 np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
480 np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
481 np_info(np, "remote ethernet address %pM\n", np->remote_mac);
483 EXPORT_SYMBOL(netpoll_print_options);
485 static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
489 if (!strchr(str, ':') &&
490 in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
494 if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
495 #if IS_ENABLED(CONFIG_IPV6)
505 int netpoll_parse_options(struct netpoll *np, char *opt)
507 char *cur=opt, *delim;
509 bool ipversion_set = false;
512 if ((delim = strchr(cur, '@')) == NULL)
515 if (kstrtou16(cur, 10, &np->local_port))
522 ipversion_set = true;
523 if ((delim = strchr(cur, '/')) == NULL)
526 ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
530 np->ipv6 = (bool)ipv6;
536 /* parse out dev name */
537 if ((delim = strchr(cur, ',')) == NULL)
540 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
547 if ((delim = strchr(cur, '@')) == NULL)
550 if (*cur == ' ' || *cur == '\t')
551 np_info(np, "warning: whitespace is not allowed\n");
552 if (kstrtou16(cur, 10, &np->remote_port))
559 if ((delim = strchr(cur, '/')) == NULL)
562 ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
565 else if (ipversion_set && np->ipv6 != (bool)ipv6)
568 np->ipv6 = (bool)ipv6;
573 if (!mac_pton(cur, np->remote_mac))
577 netpoll_print_options(np);
582 np_info(np, "couldn't parse config at '%s'!\n", cur);
585 EXPORT_SYMBOL(netpoll_parse_options);
587 int __netpoll_setup(struct netpoll *np, struct net_device *ndev, gfp_t gfp)
589 struct netpoll_info *npinfo;
590 const struct net_device_ops *ops;
594 strlcpy(np->dev_name, ndev->name, IFNAMSIZ);
595 INIT_WORK(&np->cleanup_work, netpoll_async_cleanup);
597 if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) ||
598 !ndev->netdev_ops->ndo_poll_controller) {
599 np_err(np, "%s doesn't support polling, aborting\n",
606 npinfo = kmalloc(sizeof(*npinfo), gfp);
612 sema_init(&npinfo->dev_lock, 1);
613 skb_queue_head_init(&npinfo->txq);
614 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
616 atomic_set(&npinfo->refcnt, 1);
618 ops = np->dev->netdev_ops;
619 if (ops->ndo_netpoll_setup) {
620 err = ops->ndo_netpoll_setup(ndev, npinfo, gfp);
625 npinfo = rtnl_dereference(ndev->npinfo);
626 atomic_inc(&npinfo->refcnt);
629 npinfo->netpoll = np;
631 /* last thing to do is link it to the net device structure */
632 rcu_assign_pointer(ndev->npinfo, npinfo);
641 EXPORT_SYMBOL_GPL(__netpoll_setup);
643 int netpoll_setup(struct netpoll *np)
645 struct net_device *ndev = NULL;
646 struct in_device *in_dev;
651 struct net *net = current->nsproxy->net_ns;
652 ndev = __dev_get_by_name(net, np->dev_name);
655 np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
661 if (netdev_master_upper_dev_get(ndev)) {
662 np_err(np, "%s is a slave device, aborting\n", np->dev_name);
667 if (!netif_running(ndev)) {
668 unsigned long atmost, atleast;
670 np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
672 err = dev_open(ndev);
675 np_err(np, "failed to open %s\n", ndev->name);
680 atleast = jiffies + HZ/10;
681 atmost = jiffies + carrier_timeout * HZ;
682 while (!netif_carrier_ok(ndev)) {
683 if (time_after(jiffies, atmost)) {
684 np_notice(np, "timeout waiting for carrier\n");
690 /* If carrier appears to come up instantly, we don't
691 * trust it and pause so that we don't pump all our
692 * queued console messages into the bitbucket.
695 if (time_before(jiffies, atleast)) {
696 np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
702 if (!np->local_ip.ip) {
704 in_dev = __in_dev_get_rtnl(ndev);
706 if (!in_dev || !in_dev->ifa_list) {
707 np_err(np, "no IP address for %s, aborting\n",
713 np->local_ip.ip = in_dev->ifa_list->ifa_local;
714 np_info(np, "local IP %pI4\n", &np->local_ip.ip);
716 #if IS_ENABLED(CONFIG_IPV6)
717 struct inet6_dev *idev;
720 idev = __in6_dev_get(ndev);
722 struct inet6_ifaddr *ifp;
724 read_lock_bh(&idev->lock);
725 list_for_each_entry(ifp, &idev->addr_list, if_list) {
726 if (ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL)
728 np->local_ip.in6 = ifp->addr;
732 read_unlock_bh(&idev->lock);
735 np_err(np, "no IPv6 address for %s, aborting\n",
739 np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
741 np_err(np, "IPv6 is not supported %s, aborting\n",
749 /* fill up the skb queue */
752 err = __netpoll_setup(np, ndev, GFP_KERNEL);
765 EXPORT_SYMBOL(netpoll_setup);
767 static int __init netpoll_init(void)
769 skb_queue_head_init(&skb_pool);
772 core_initcall(netpoll_init);
774 static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
776 struct netpoll_info *npinfo =
777 container_of(rcu_head, struct netpoll_info, rcu);
779 skb_queue_purge(&npinfo->txq);
781 /* we can't call cancel_delayed_work_sync here, as we are in softirq */
782 cancel_delayed_work(&npinfo->tx_work);
784 /* clean after last, unfinished work */
785 __skb_queue_purge(&npinfo->txq);
786 /* now cancel it again */
787 cancel_delayed_work(&npinfo->tx_work);
791 void __netpoll_cleanup(struct netpoll *np)
793 struct netpoll_info *npinfo;
795 /* rtnl_dereference would be preferable here but
796 * rcu_cleanup_netpoll path can put us in here safely without
797 * holding the rtnl, so plain rcu_dereference it is
799 npinfo = rtnl_dereference(np->dev->npinfo);
803 synchronize_srcu(&netpoll_srcu);
805 if (atomic_dec_and_test(&npinfo->refcnt)) {
806 const struct net_device_ops *ops;
808 ops = np->dev->netdev_ops;
809 if (ops->ndo_netpoll_cleanup)
810 ops->ndo_netpoll_cleanup(np->dev);
812 rcu_assign_pointer(np->dev->npinfo, NULL);
813 call_rcu_bh(&npinfo->rcu, rcu_cleanup_netpoll_info);
816 EXPORT_SYMBOL_GPL(__netpoll_cleanup);
818 static void netpoll_async_cleanup(struct work_struct *work)
820 struct netpoll *np = container_of(work, struct netpoll, cleanup_work);
823 __netpoll_cleanup(np);
828 void __netpoll_free_async(struct netpoll *np)
830 schedule_work(&np->cleanup_work);
832 EXPORT_SYMBOL_GPL(__netpoll_free_async);
834 void netpoll_cleanup(struct netpoll *np)
839 __netpoll_cleanup(np);
845 EXPORT_SYMBOL(netpoll_cleanup);