Loading...
1/*
2 * Common framework for low-level network console, dump, and debugger code
3 *
4 * Sep 8 2003 Matt Mackall <mpm@selenic.com>
5 *
6 * based on the netconsole code from:
7 *
8 * Copyright (C) 2001 Ingo Molnar <mingo@redhat.com>
9 * Copyright (C) 2002 Red Hat, Inc.
10 */
11
12#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13
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>
31#include <net/tcp.h>
32#include <net/udp.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>
38
39/*
40 * We maintain a small pool of fully-sized skbs, to make sure the
41 * message gets out even in extreme OOM situations.
42 */
43
44#define MAX_UDP_CHUNK 1460
45#define MAX_SKBS 32
46
47static struct sk_buff_head skb_pool;
48
49DEFINE_STATIC_SRCU(netpoll_srcu);
50
51#define USEC_PER_POLL 50
52
53#define MAX_SKB_SIZE \
54 (sizeof(struct ethhdr) + \
55 sizeof(struct iphdr) + \
56 sizeof(struct udphdr) + \
57 MAX_UDP_CHUNK)
58
59static void zap_completion_queue(void);
60static void netpoll_async_cleanup(struct work_struct *work);
61
62static unsigned int carrier_timeout = 4;
63module_param(carrier_timeout, uint, 0644);
64
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__)
71
72static int netpoll_start_xmit(struct sk_buff *skb, struct net_device *dev,
73 struct netdev_queue *txq)
74{
75 int status = NETDEV_TX_OK;
76 netdev_features_t features;
77
78 features = netif_skb_features(skb);
79
80 if (skb_vlan_tag_present(skb) &&
81 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
82 skb = __vlan_hwaccel_push_inside(skb);
83 if (unlikely(!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.
87 */
88 goto out;
89 }
90 }
91
92 status = netdev_start_xmit(skb, dev, txq, false);
93
94out:
95 return status;
96}
97
98static void queue_process(struct work_struct *work)
99{
100 struct netpoll_info *npinfo =
101 container_of(work, struct netpoll_info, tx_work.work);
102 struct sk_buff *skb;
103 unsigned long flags;
104
105 while ((skb = skb_dequeue(&npinfo->txq))) {
106 struct net_device *dev = skb->dev;
107 struct netdev_queue *txq;
108 unsigned int q_index;
109
110 if (!netif_device_present(dev) || !netif_running(dev)) {
111 kfree_skb(skb);
112 continue;
113 }
114
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);
121 }
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);
129
130 schedule_delayed_work(&npinfo->tx_work, HZ/10);
131 return;
132 }
133 HARD_TX_UNLOCK(dev, txq);
134 local_irq_restore(flags);
135 }
136}
137
138/*
139 * Check whether delayed processing was scheduled for our NIC. If so,
140 * we attempt to grab the poll lock and use ->poll() to pump the card.
141 * If this fails, either we've recursed in ->poll() or it's already
142 * running on another CPU.
143 *
144 * Note: we don't mask interrupts with this lock because we're using
145 * trylock here and interrupts are already disabled in the softirq
146 * case. Further, we test the poll_owner to avoid recursion on UP
147 * systems where the lock doesn't exist.
148 */
149static void poll_one_napi(struct napi_struct *napi)
150{
151 int work = 0;
152
153 /* net_rx_action's ->poll() invocations and our's are
154 * synchronized by this test which is only made while
155 * holding the napi->poll_lock.
156 */
157 if (!test_bit(NAPI_STATE_SCHED, &napi->state))
158 return;
159
160 /* If we set this bit but see that it has already been set,
161 * that indicates that napi has been disabled and we need
162 * to abort this operation
163 */
164 if (test_and_set_bit(NAPI_STATE_NPSVC, &napi->state))
165 return;
166
167 /* We explicilty pass the polling call a budget of 0 to
168 * indicate that we are clearing the Tx path only.
169 */
170 work = napi->poll(napi, 0);
171 WARN_ONCE(work, "%pF exceeded budget in poll\n", napi->poll);
172 trace_napi_poll(napi, work, 0);
173
174 clear_bit(NAPI_STATE_NPSVC, &napi->state);
175}
176
177static void poll_napi(struct net_device *dev)
178{
179 struct napi_struct *napi;
180 int cpu = smp_processor_id();
181
182 list_for_each_entry(napi, &dev->napi_list, dev_list) {
183 if (cmpxchg(&napi->poll_owner, -1, cpu) == -1) {
184 poll_one_napi(napi);
185 smp_store_release(&napi->poll_owner, -1);
186 }
187 }
188}
189
190static void netpoll_poll_dev(struct net_device *dev)
191{
192 const struct net_device_ops *ops;
193 struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
194
195 /* Don't do any rx activity if the dev_lock mutex is held
196 * the dev_open/close paths use this to block netpoll activity
197 * while changing device state
198 */
199 if (down_trylock(&ni->dev_lock))
200 return;
201
202 if (!netif_running(dev)) {
203 up(&ni->dev_lock);
204 return;
205 }
206
207 ops = dev->netdev_ops;
208 if (!ops->ndo_poll_controller) {
209 up(&ni->dev_lock);
210 return;
211 }
212
213 /* Process pending work on NIC */
214 ops->ndo_poll_controller(dev);
215
216 poll_napi(dev);
217
218 up(&ni->dev_lock);
219
220 zap_completion_queue();
221}
222
223void netpoll_poll_disable(struct net_device *dev)
224{
225 struct netpoll_info *ni;
226 int idx;
227 might_sleep();
228 idx = srcu_read_lock(&netpoll_srcu);
229 ni = srcu_dereference(dev->npinfo, &netpoll_srcu);
230 if (ni)
231 down(&ni->dev_lock);
232 srcu_read_unlock(&netpoll_srcu, idx);
233}
234EXPORT_SYMBOL(netpoll_poll_disable);
235
236void netpoll_poll_enable(struct net_device *dev)
237{
238 struct netpoll_info *ni;
239 rcu_read_lock();
240 ni = rcu_dereference(dev->npinfo);
241 if (ni)
242 up(&ni->dev_lock);
243 rcu_read_unlock();
244}
245EXPORT_SYMBOL(netpoll_poll_enable);
246
247static void refill_skbs(void)
248{
249 struct sk_buff *skb;
250 unsigned long flags;
251
252 spin_lock_irqsave(&skb_pool.lock, flags);
253 while (skb_pool.qlen < MAX_SKBS) {
254 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
255 if (!skb)
256 break;
257
258 __skb_queue_tail(&skb_pool, skb);
259 }
260 spin_unlock_irqrestore(&skb_pool.lock, flags);
261}
262
263static void zap_completion_queue(void)
264{
265 unsigned long flags;
266 struct softnet_data *sd = &get_cpu_var(softnet_data);
267
268 if (sd->completion_queue) {
269 struct sk_buff *clist;
270
271 local_irq_save(flags);
272 clist = sd->completion_queue;
273 sd->completion_queue = NULL;
274 local_irq_restore(flags);
275
276 while (clist != NULL) {
277 struct sk_buff *skb = clist;
278 clist = clist->next;
279 if (!skb_irq_freeable(skb)) {
280 refcount_set(&skb->users, 1);
281 dev_kfree_skb_any(skb); /* put this one back */
282 } else {
283 __kfree_skb(skb);
284 }
285 }
286 }
287
288 put_cpu_var(softnet_data);
289}
290
291static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
292{
293 int count = 0;
294 struct sk_buff *skb;
295
296 zap_completion_queue();
297 refill_skbs();
298repeat:
299
300 skb = alloc_skb(len, GFP_ATOMIC);
301 if (!skb)
302 skb = skb_dequeue(&skb_pool);
303
304 if (!skb) {
305 if (++count < 10) {
306 netpoll_poll_dev(np->dev);
307 goto repeat;
308 }
309 return NULL;
310 }
311
312 refcount_set(&skb->users, 1);
313 skb_reserve(skb, reserve);
314 return skb;
315}
316
317static int netpoll_owner_active(struct net_device *dev)
318{
319 struct napi_struct *napi;
320
321 list_for_each_entry(napi, &dev->napi_list, dev_list) {
322 if (napi->poll_owner == smp_processor_id())
323 return 1;
324 }
325 return 0;
326}
327
328/* call with IRQ disabled */
329void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb,
330 struct net_device *dev)
331{
332 int status = NETDEV_TX_BUSY;
333 unsigned long tries;
334 /* It is up to the caller to keep npinfo alive. */
335 struct netpoll_info *npinfo;
336
337 lockdep_assert_irqs_disabled();
338
339 npinfo = rcu_dereference_bh(np->dev->npinfo);
340 if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
341 dev_kfree_skb_irq(skb);
342 return;
343 }
344
345 /* don't get messages out of order, and no recursion */
346 if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
347 struct netdev_queue *txq;
348
349 txq = netdev_pick_tx(dev, skb, NULL);
350
351 /* try until next clock tick */
352 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
353 tries > 0; --tries) {
354 if (HARD_TX_TRYLOCK(dev, txq)) {
355 if (!netif_xmit_stopped(txq))
356 status = netpoll_start_xmit(skb, dev, txq);
357
358 HARD_TX_UNLOCK(dev, txq);
359
360 if (status == NETDEV_TX_OK)
361 break;
362
363 }
364
365 /* tickle device maybe there is some cleanup */
366 netpoll_poll_dev(np->dev);
367
368 udelay(USEC_PER_POLL);
369 }
370
371 WARN_ONCE(!irqs_disabled(),
372 "netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pF)\n",
373 dev->name, dev->netdev_ops->ndo_start_xmit);
374
375 }
376
377 if (status != NETDEV_TX_OK) {
378 skb_queue_tail(&npinfo->txq, skb);
379 schedule_delayed_work(&npinfo->tx_work,0);
380 }
381}
382EXPORT_SYMBOL(netpoll_send_skb_on_dev);
383
384void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
385{
386 int total_len, ip_len, udp_len;
387 struct sk_buff *skb;
388 struct udphdr *udph;
389 struct iphdr *iph;
390 struct ethhdr *eth;
391 static atomic_t ip_ident;
392 struct ipv6hdr *ip6h;
393
394 WARN_ON_ONCE(!irqs_disabled());
395
396 udp_len = len + sizeof(*udph);
397 if (np->ipv6)
398 ip_len = udp_len + sizeof(*ip6h);
399 else
400 ip_len = udp_len + sizeof(*iph);
401
402 total_len = ip_len + LL_RESERVED_SPACE(np->dev);
403
404 skb = find_skb(np, total_len + np->dev->needed_tailroom,
405 total_len - len);
406 if (!skb)
407 return;
408
409 skb_copy_to_linear_data(skb, msg, len);
410 skb_put(skb, len);
411
412 skb_push(skb, sizeof(*udph));
413 skb_reset_transport_header(skb);
414 udph = udp_hdr(skb);
415 udph->source = htons(np->local_port);
416 udph->dest = htons(np->remote_port);
417 udph->len = htons(udp_len);
418
419 if (np->ipv6) {
420 udph->check = 0;
421 udph->check = csum_ipv6_magic(&np->local_ip.in6,
422 &np->remote_ip.in6,
423 udp_len, IPPROTO_UDP,
424 csum_partial(udph, udp_len, 0));
425 if (udph->check == 0)
426 udph->check = CSUM_MANGLED_0;
427
428 skb_push(skb, sizeof(*ip6h));
429 skb_reset_network_header(skb);
430 ip6h = ipv6_hdr(skb);
431
432 /* ip6h->version = 6; ip6h->priority = 0; */
433 put_unaligned(0x60, (unsigned char *)ip6h);
434 ip6h->flow_lbl[0] = 0;
435 ip6h->flow_lbl[1] = 0;
436 ip6h->flow_lbl[2] = 0;
437
438 ip6h->payload_len = htons(sizeof(struct udphdr) + len);
439 ip6h->nexthdr = IPPROTO_UDP;
440 ip6h->hop_limit = 32;
441 ip6h->saddr = np->local_ip.in6;
442 ip6h->daddr = np->remote_ip.in6;
443
444 eth = skb_push(skb, ETH_HLEN);
445 skb_reset_mac_header(skb);
446 skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
447 } else {
448 udph->check = 0;
449 udph->check = csum_tcpudp_magic(np->local_ip.ip,
450 np->remote_ip.ip,
451 udp_len, IPPROTO_UDP,
452 csum_partial(udph, udp_len, 0));
453 if (udph->check == 0)
454 udph->check = CSUM_MANGLED_0;
455
456 skb_push(skb, sizeof(*iph));
457 skb_reset_network_header(skb);
458 iph = ip_hdr(skb);
459
460 /* iph->version = 4; iph->ihl = 5; */
461 put_unaligned(0x45, (unsigned char *)iph);
462 iph->tos = 0;
463 put_unaligned(htons(ip_len), &(iph->tot_len));
464 iph->id = htons(atomic_inc_return(&ip_ident));
465 iph->frag_off = 0;
466 iph->ttl = 64;
467 iph->protocol = IPPROTO_UDP;
468 iph->check = 0;
469 put_unaligned(np->local_ip.ip, &(iph->saddr));
470 put_unaligned(np->remote_ip.ip, &(iph->daddr));
471 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
472
473 eth = skb_push(skb, ETH_HLEN);
474 skb_reset_mac_header(skb);
475 skb->protocol = eth->h_proto = htons(ETH_P_IP);
476 }
477
478 ether_addr_copy(eth->h_source, np->dev->dev_addr);
479 ether_addr_copy(eth->h_dest, np->remote_mac);
480
481 skb->dev = np->dev;
482
483 netpoll_send_skb(np, skb);
484}
485EXPORT_SYMBOL(netpoll_send_udp);
486
487void netpoll_print_options(struct netpoll *np)
488{
489 np_info(np, "local port %d\n", np->local_port);
490 if (np->ipv6)
491 np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
492 else
493 np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
494 np_info(np, "interface '%s'\n", np->dev_name);
495 np_info(np, "remote port %d\n", np->remote_port);
496 if (np->ipv6)
497 np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
498 else
499 np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
500 np_info(np, "remote ethernet address %pM\n", np->remote_mac);
501}
502EXPORT_SYMBOL(netpoll_print_options);
503
504static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
505{
506 const char *end;
507
508 if (!strchr(str, ':') &&
509 in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
510 if (!*end)
511 return 0;
512 }
513 if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
514#if IS_ENABLED(CONFIG_IPV6)
515 if (!*end)
516 return 1;
517#else
518 return -1;
519#endif
520 }
521 return -1;
522}
523
524int netpoll_parse_options(struct netpoll *np, char *opt)
525{
526 char *cur=opt, *delim;
527 int ipv6;
528 bool ipversion_set = false;
529
530 if (*cur != '@') {
531 if ((delim = strchr(cur, '@')) == NULL)
532 goto parse_failed;
533 *delim = 0;
534 if (kstrtou16(cur, 10, &np->local_port))
535 goto parse_failed;
536 cur = delim;
537 }
538 cur++;
539
540 if (*cur != '/') {
541 ipversion_set = true;
542 if ((delim = strchr(cur, '/')) == NULL)
543 goto parse_failed;
544 *delim = 0;
545 ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
546 if (ipv6 < 0)
547 goto parse_failed;
548 else
549 np->ipv6 = (bool)ipv6;
550 cur = delim;
551 }
552 cur++;
553
554 if (*cur != ',') {
555 /* parse out dev name */
556 if ((delim = strchr(cur, ',')) == NULL)
557 goto parse_failed;
558 *delim = 0;
559 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
560 cur = delim;
561 }
562 cur++;
563
564 if (*cur != '@') {
565 /* dst port */
566 if ((delim = strchr(cur, '@')) == NULL)
567 goto parse_failed;
568 *delim = 0;
569 if (*cur == ' ' || *cur == '\t')
570 np_info(np, "warning: whitespace is not allowed\n");
571 if (kstrtou16(cur, 10, &np->remote_port))
572 goto parse_failed;
573 cur = delim;
574 }
575 cur++;
576
577 /* dst ip */
578 if ((delim = strchr(cur, '/')) == NULL)
579 goto parse_failed;
580 *delim = 0;
581 ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
582 if (ipv6 < 0)
583 goto parse_failed;
584 else if (ipversion_set && np->ipv6 != (bool)ipv6)
585 goto parse_failed;
586 else
587 np->ipv6 = (bool)ipv6;
588 cur = delim + 1;
589
590 if (*cur != 0) {
591 /* MAC address */
592 if (!mac_pton(cur, np->remote_mac))
593 goto parse_failed;
594 }
595
596 netpoll_print_options(np);
597
598 return 0;
599
600 parse_failed:
601 np_info(np, "couldn't parse config at '%s'!\n", cur);
602 return -1;
603}
604EXPORT_SYMBOL(netpoll_parse_options);
605
606int __netpoll_setup(struct netpoll *np, struct net_device *ndev)
607{
608 struct netpoll_info *npinfo;
609 const struct net_device_ops *ops;
610 int err;
611
612 np->dev = ndev;
613 strlcpy(np->dev_name, ndev->name, IFNAMSIZ);
614 INIT_WORK(&np->cleanup_work, netpoll_async_cleanup);
615
616 if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) ||
617 !ndev->netdev_ops->ndo_poll_controller) {
618 np_err(np, "%s doesn't support polling, aborting\n",
619 np->dev_name);
620 err = -ENOTSUPP;
621 goto out;
622 }
623
624 if (!ndev->npinfo) {
625 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
626 if (!npinfo) {
627 err = -ENOMEM;
628 goto out;
629 }
630
631 sema_init(&npinfo->dev_lock, 1);
632 skb_queue_head_init(&npinfo->txq);
633 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
634
635 refcount_set(&npinfo->refcnt, 1);
636
637 ops = np->dev->netdev_ops;
638 if (ops->ndo_netpoll_setup) {
639 err = ops->ndo_netpoll_setup(ndev, npinfo);
640 if (err)
641 goto free_npinfo;
642 }
643 } else {
644 npinfo = rtnl_dereference(ndev->npinfo);
645 refcount_inc(&npinfo->refcnt);
646 }
647
648 npinfo->netpoll = np;
649
650 /* last thing to do is link it to the net device structure */
651 rcu_assign_pointer(ndev->npinfo, npinfo);
652
653 return 0;
654
655free_npinfo:
656 kfree(npinfo);
657out:
658 return err;
659}
660EXPORT_SYMBOL_GPL(__netpoll_setup);
661
662int netpoll_setup(struct netpoll *np)
663{
664 struct net_device *ndev = NULL;
665 struct in_device *in_dev;
666 int err;
667
668 rtnl_lock();
669 if (np->dev_name[0]) {
670 struct net *net = current->nsproxy->net_ns;
671 ndev = __dev_get_by_name(net, np->dev_name);
672 }
673 if (!ndev) {
674 np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
675 err = -ENODEV;
676 goto unlock;
677 }
678 dev_hold(ndev);
679
680 if (netdev_master_upper_dev_get(ndev)) {
681 np_err(np, "%s is a slave device, aborting\n", np->dev_name);
682 err = -EBUSY;
683 goto put;
684 }
685
686 if (!netif_running(ndev)) {
687 unsigned long atmost, atleast;
688
689 np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
690
691 err = dev_open(ndev);
692
693 if (err) {
694 np_err(np, "failed to open %s\n", ndev->name);
695 goto put;
696 }
697
698 rtnl_unlock();
699 atleast = jiffies + HZ/10;
700 atmost = jiffies + carrier_timeout * HZ;
701 while (!netif_carrier_ok(ndev)) {
702 if (time_after(jiffies, atmost)) {
703 np_notice(np, "timeout waiting for carrier\n");
704 break;
705 }
706 msleep(1);
707 }
708
709 /* If carrier appears to come up instantly, we don't
710 * trust it and pause so that we don't pump all our
711 * queued console messages into the bitbucket.
712 */
713
714 if (time_before(jiffies, atleast)) {
715 np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
716 msleep(4000);
717 }
718 rtnl_lock();
719 }
720
721 if (!np->local_ip.ip) {
722 if (!np->ipv6) {
723 in_dev = __in_dev_get_rtnl(ndev);
724
725 if (!in_dev || !in_dev->ifa_list) {
726 np_err(np, "no IP address for %s, aborting\n",
727 np->dev_name);
728 err = -EDESTADDRREQ;
729 goto put;
730 }
731
732 np->local_ip.ip = in_dev->ifa_list->ifa_local;
733 np_info(np, "local IP %pI4\n", &np->local_ip.ip);
734 } else {
735#if IS_ENABLED(CONFIG_IPV6)
736 struct inet6_dev *idev;
737
738 err = -EDESTADDRREQ;
739 idev = __in6_dev_get(ndev);
740 if (idev) {
741 struct inet6_ifaddr *ifp;
742
743 read_lock_bh(&idev->lock);
744 list_for_each_entry(ifp, &idev->addr_list, if_list) {
745 if (ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL)
746 continue;
747 np->local_ip.in6 = ifp->addr;
748 err = 0;
749 break;
750 }
751 read_unlock_bh(&idev->lock);
752 }
753 if (err) {
754 np_err(np, "no IPv6 address for %s, aborting\n",
755 np->dev_name);
756 goto put;
757 } else
758 np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
759#else
760 np_err(np, "IPv6 is not supported %s, aborting\n",
761 np->dev_name);
762 err = -EINVAL;
763 goto put;
764#endif
765 }
766 }
767
768 /* fill up the skb queue */
769 refill_skbs();
770
771 err = __netpoll_setup(np, ndev);
772 if (err)
773 goto put;
774
775 rtnl_unlock();
776 return 0;
777
778put:
779 dev_put(ndev);
780unlock:
781 rtnl_unlock();
782 return err;
783}
784EXPORT_SYMBOL(netpoll_setup);
785
786static int __init netpoll_init(void)
787{
788 skb_queue_head_init(&skb_pool);
789 return 0;
790}
791core_initcall(netpoll_init);
792
793static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
794{
795 struct netpoll_info *npinfo =
796 container_of(rcu_head, struct netpoll_info, rcu);
797
798 skb_queue_purge(&npinfo->txq);
799
800 /* we can't call cancel_delayed_work_sync here, as we are in softirq */
801 cancel_delayed_work(&npinfo->tx_work);
802
803 /* clean after last, unfinished work */
804 __skb_queue_purge(&npinfo->txq);
805 /* now cancel it again */
806 cancel_delayed_work(&npinfo->tx_work);
807 kfree(npinfo);
808}
809
810void __netpoll_cleanup(struct netpoll *np)
811{
812 struct netpoll_info *npinfo;
813
814 /* rtnl_dereference would be preferable here but
815 * rcu_cleanup_netpoll path can put us in here safely without
816 * holding the rtnl, so plain rcu_dereference it is
817 */
818 npinfo = rtnl_dereference(np->dev->npinfo);
819 if (!npinfo)
820 return;
821
822 synchronize_srcu(&netpoll_srcu);
823
824 if (refcount_dec_and_test(&npinfo->refcnt)) {
825 const struct net_device_ops *ops;
826
827 ops = np->dev->netdev_ops;
828 if (ops->ndo_netpoll_cleanup)
829 ops->ndo_netpoll_cleanup(np->dev);
830
831 RCU_INIT_POINTER(np->dev->npinfo, NULL);
832 call_rcu_bh(&npinfo->rcu, rcu_cleanup_netpoll_info);
833 } else
834 RCU_INIT_POINTER(np->dev->npinfo, NULL);
835}
836EXPORT_SYMBOL_GPL(__netpoll_cleanup);
837
838static void netpoll_async_cleanup(struct work_struct *work)
839{
840 struct netpoll *np = container_of(work, struct netpoll, cleanup_work);
841
842 rtnl_lock();
843 __netpoll_cleanup(np);
844 rtnl_unlock();
845 kfree(np);
846}
847
848void __netpoll_free_async(struct netpoll *np)
849{
850 schedule_work(&np->cleanup_work);
851}
852EXPORT_SYMBOL_GPL(__netpoll_free_async);
853
854void netpoll_cleanup(struct netpoll *np)
855{
856 rtnl_lock();
857 if (!np->dev)
858 goto out;
859 __netpoll_cleanup(np);
860 dev_put(np->dev);
861 np->dev = NULL;
862out:
863 rtnl_unlock();
864}
865EXPORT_SYMBOL(netpoll_cleanup);
1/*
2 * Common framework for low-level network console, dump, and debugger code
3 *
4 * Sep 8 2003 Matt Mackall <mpm@selenic.com>
5 *
6 * based on the netconsole code from:
7 *
8 * Copyright (C) 2001 Ingo Molnar <mingo@redhat.com>
9 * Copyright (C) 2002 Red Hat, Inc.
10 */
11
12#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13
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>
31#include <net/tcp.h>
32#include <net/udp.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>
38
39/*
40 * We maintain a small pool of fully-sized skbs, to make sure the
41 * message gets out even in extreme OOM situations.
42 */
43
44#define MAX_UDP_CHUNK 1460
45#define MAX_SKBS 32
46
47static struct sk_buff_head skb_pool;
48
49DEFINE_STATIC_SRCU(netpoll_srcu);
50
51#define USEC_PER_POLL 50
52
53#define MAX_SKB_SIZE \
54 (sizeof(struct ethhdr) + \
55 sizeof(struct iphdr) + \
56 sizeof(struct udphdr) + \
57 MAX_UDP_CHUNK)
58
59static void zap_completion_queue(void);
60static void netpoll_async_cleanup(struct work_struct *work);
61
62static unsigned int carrier_timeout = 4;
63module_param(carrier_timeout, uint, 0644);
64
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__)
71
72static int netpoll_start_xmit(struct sk_buff *skb, struct net_device *dev,
73 struct netdev_queue *txq)
74{
75 int status = NETDEV_TX_OK;
76 netdev_features_t features;
77
78 features = netif_skb_features(skb);
79
80 if (skb_vlan_tag_present(skb) &&
81 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
82 skb = __vlan_hwaccel_push_inside(skb);
83 if (unlikely(!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.
87 */
88 goto out;
89 }
90 }
91
92 status = netdev_start_xmit(skb, dev, txq, false);
93
94out:
95 return status;
96}
97
98static void queue_process(struct work_struct *work)
99{
100 struct netpoll_info *npinfo =
101 container_of(work, struct netpoll_info, tx_work.work);
102 struct sk_buff *skb;
103 unsigned long flags;
104
105 while ((skb = skb_dequeue(&npinfo->txq))) {
106 struct net_device *dev = skb->dev;
107 struct netdev_queue *txq;
108
109 if (!netif_device_present(dev) || !netif_running(dev)) {
110 kfree_skb(skb);
111 continue;
112 }
113
114 txq = skb_get_tx_queue(dev, skb);
115
116 local_irq_save(flags);
117 HARD_TX_LOCK(dev, txq, smp_processor_id());
118 if (netif_xmit_frozen_or_stopped(txq) ||
119 netpoll_start_xmit(skb, dev, txq) != NETDEV_TX_OK) {
120 skb_queue_head(&npinfo->txq, skb);
121 HARD_TX_UNLOCK(dev, txq);
122 local_irq_restore(flags);
123
124 schedule_delayed_work(&npinfo->tx_work, HZ/10);
125 return;
126 }
127 HARD_TX_UNLOCK(dev, txq);
128 local_irq_restore(flags);
129 }
130}
131
132/*
133 * Check whether delayed processing was scheduled for our NIC. If so,
134 * we attempt to grab the poll lock and use ->poll() to pump the card.
135 * If this fails, either we've recursed in ->poll() or it's already
136 * running on another CPU.
137 *
138 * Note: we don't mask interrupts with this lock because we're using
139 * trylock here and interrupts are already disabled in the softirq
140 * case. Further, we test the poll_owner to avoid recursion on UP
141 * systems where the lock doesn't exist.
142 */
143static void poll_one_napi(struct napi_struct *napi)
144{
145 int work = 0;
146
147 /* net_rx_action's ->poll() invocations and our's are
148 * synchronized by this test which is only made while
149 * holding the napi->poll_lock.
150 */
151 if (!test_bit(NAPI_STATE_SCHED, &napi->state))
152 return;
153
154 /* If we set this bit but see that it has already been set,
155 * that indicates that napi has been disabled and we need
156 * to abort this operation
157 */
158 if (test_and_set_bit(NAPI_STATE_NPSVC, &napi->state))
159 return;
160
161 /* We explicilty pass the polling call a budget of 0 to
162 * indicate that we are clearing the Tx path only.
163 */
164 work = napi->poll(napi, 0);
165 WARN_ONCE(work, "%pF exceeded budget in poll\n", napi->poll);
166 trace_napi_poll(napi);
167
168 clear_bit(NAPI_STATE_NPSVC, &napi->state);
169}
170
171static void poll_napi(struct net_device *dev)
172{
173 struct napi_struct *napi;
174
175 list_for_each_entry(napi, &dev->napi_list, dev_list) {
176 if (napi->poll_owner != smp_processor_id() &&
177 spin_trylock(&napi->poll_lock)) {
178 poll_one_napi(napi);
179 spin_unlock(&napi->poll_lock);
180 }
181 }
182}
183
184static void netpoll_poll_dev(struct net_device *dev)
185{
186 const struct net_device_ops *ops;
187 struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
188
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
192 */
193 if (down_trylock(&ni->dev_lock))
194 return;
195
196 if (!netif_running(dev)) {
197 up(&ni->dev_lock);
198 return;
199 }
200
201 ops = dev->netdev_ops;
202 if (!ops->ndo_poll_controller) {
203 up(&ni->dev_lock);
204 return;
205 }
206
207 /* Process pending work on NIC */
208 ops->ndo_poll_controller(dev);
209
210 poll_napi(dev);
211
212 up(&ni->dev_lock);
213
214 zap_completion_queue();
215}
216
217void netpoll_poll_disable(struct net_device *dev)
218{
219 struct netpoll_info *ni;
220 int idx;
221 might_sleep();
222 idx = srcu_read_lock(&netpoll_srcu);
223 ni = srcu_dereference(dev->npinfo, &netpoll_srcu);
224 if (ni)
225 down(&ni->dev_lock);
226 srcu_read_unlock(&netpoll_srcu, idx);
227}
228EXPORT_SYMBOL(netpoll_poll_disable);
229
230void netpoll_poll_enable(struct net_device *dev)
231{
232 struct netpoll_info *ni;
233 rcu_read_lock();
234 ni = rcu_dereference(dev->npinfo);
235 if (ni)
236 up(&ni->dev_lock);
237 rcu_read_unlock();
238}
239EXPORT_SYMBOL(netpoll_poll_enable);
240
241static void refill_skbs(void)
242{
243 struct sk_buff *skb;
244 unsigned long flags;
245
246 spin_lock_irqsave(&skb_pool.lock, flags);
247 while (skb_pool.qlen < MAX_SKBS) {
248 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
249 if (!skb)
250 break;
251
252 __skb_queue_tail(&skb_pool, skb);
253 }
254 spin_unlock_irqrestore(&skb_pool.lock, flags);
255}
256
257static void zap_completion_queue(void)
258{
259 unsigned long flags;
260 struct softnet_data *sd = &get_cpu_var(softnet_data);
261
262 if (sd->completion_queue) {
263 struct sk_buff *clist;
264
265 local_irq_save(flags);
266 clist = sd->completion_queue;
267 sd->completion_queue = NULL;
268 local_irq_restore(flags);
269
270 while (clist != NULL) {
271 struct sk_buff *skb = clist;
272 clist = clist->next;
273 if (!skb_irq_freeable(skb)) {
274 atomic_inc(&skb->users);
275 dev_kfree_skb_any(skb); /* put this one back */
276 } else {
277 __kfree_skb(skb);
278 }
279 }
280 }
281
282 put_cpu_var(softnet_data);
283}
284
285static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
286{
287 int count = 0;
288 struct sk_buff *skb;
289
290 zap_completion_queue();
291 refill_skbs();
292repeat:
293
294 skb = alloc_skb(len, GFP_ATOMIC);
295 if (!skb)
296 skb = skb_dequeue(&skb_pool);
297
298 if (!skb) {
299 if (++count < 10) {
300 netpoll_poll_dev(np->dev);
301 goto repeat;
302 }
303 return NULL;
304 }
305
306 atomic_set(&skb->users, 1);
307 skb_reserve(skb, reserve);
308 return skb;
309}
310
311static int netpoll_owner_active(struct net_device *dev)
312{
313 struct napi_struct *napi;
314
315 list_for_each_entry(napi, &dev->napi_list, dev_list) {
316 if (napi->poll_owner == smp_processor_id())
317 return 1;
318 }
319 return 0;
320}
321
322/* call with IRQ disabled */
323void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb,
324 struct net_device *dev)
325{
326 int status = NETDEV_TX_BUSY;
327 unsigned long tries;
328 /* It is up to the caller to keep npinfo alive. */
329 struct netpoll_info *npinfo;
330
331 WARN_ON_ONCE(!irqs_disabled());
332
333 npinfo = rcu_dereference_bh(np->dev->npinfo);
334 if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
335 dev_kfree_skb_irq(skb);
336 return;
337 }
338
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;
342
343 txq = netdev_pick_tx(dev, skb, NULL);
344
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);
351
352 HARD_TX_UNLOCK(dev, txq);
353
354 if (status == NETDEV_TX_OK)
355 break;
356
357 }
358
359 /* tickle device maybe there is some cleanup */
360 netpoll_poll_dev(np->dev);
361
362 udelay(USEC_PER_POLL);
363 }
364
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);
368
369 }
370
371 if (status != NETDEV_TX_OK) {
372 skb_queue_tail(&npinfo->txq, skb);
373 schedule_delayed_work(&npinfo->tx_work,0);
374 }
375}
376EXPORT_SYMBOL(netpoll_send_skb_on_dev);
377
378void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
379{
380 int total_len, ip_len, udp_len;
381 struct sk_buff *skb;
382 struct udphdr *udph;
383 struct iphdr *iph;
384 struct ethhdr *eth;
385 static atomic_t ip_ident;
386 struct ipv6hdr *ip6h;
387
388 WARN_ON_ONCE(!irqs_disabled());
389
390 udp_len = len + sizeof(*udph);
391 if (np->ipv6)
392 ip_len = udp_len + sizeof(*ip6h);
393 else
394 ip_len = udp_len + sizeof(*iph);
395
396 total_len = ip_len + LL_RESERVED_SPACE(np->dev);
397
398 skb = find_skb(np, total_len + np->dev->needed_tailroom,
399 total_len - len);
400 if (!skb)
401 return;
402
403 skb_copy_to_linear_data(skb, msg, len);
404 skb_put(skb, len);
405
406 skb_push(skb, sizeof(*udph));
407 skb_reset_transport_header(skb);
408 udph = udp_hdr(skb);
409 udph->source = htons(np->local_port);
410 udph->dest = htons(np->remote_port);
411 udph->len = htons(udp_len);
412
413 if (np->ipv6) {
414 udph->check = 0;
415 udph->check = csum_ipv6_magic(&np->local_ip.in6,
416 &np->remote_ip.in6,
417 udp_len, IPPROTO_UDP,
418 csum_partial(udph, udp_len, 0));
419 if (udph->check == 0)
420 udph->check = CSUM_MANGLED_0;
421
422 skb_push(skb, sizeof(*ip6h));
423 skb_reset_network_header(skb);
424 ip6h = ipv6_hdr(skb);
425
426 /* ip6h->version = 6; ip6h->priority = 0; */
427 put_unaligned(0x60, (unsigned char *)ip6h);
428 ip6h->flow_lbl[0] = 0;
429 ip6h->flow_lbl[1] = 0;
430 ip6h->flow_lbl[2] = 0;
431
432 ip6h->payload_len = htons(sizeof(struct udphdr) + len);
433 ip6h->nexthdr = IPPROTO_UDP;
434 ip6h->hop_limit = 32;
435 ip6h->saddr = np->local_ip.in6;
436 ip6h->daddr = np->remote_ip.in6;
437
438 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
439 skb_reset_mac_header(skb);
440 skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
441 } else {
442 udph->check = 0;
443 udph->check = csum_tcpudp_magic(np->local_ip.ip,
444 np->remote_ip.ip,
445 udp_len, IPPROTO_UDP,
446 csum_partial(udph, udp_len, 0));
447 if (udph->check == 0)
448 udph->check = CSUM_MANGLED_0;
449
450 skb_push(skb, sizeof(*iph));
451 skb_reset_network_header(skb);
452 iph = ip_hdr(skb);
453
454 /* iph->version = 4; iph->ihl = 5; */
455 put_unaligned(0x45, (unsigned char *)iph);
456 iph->tos = 0;
457 put_unaligned(htons(ip_len), &(iph->tot_len));
458 iph->id = htons(atomic_inc_return(&ip_ident));
459 iph->frag_off = 0;
460 iph->ttl = 64;
461 iph->protocol = IPPROTO_UDP;
462 iph->check = 0;
463 put_unaligned(np->local_ip.ip, &(iph->saddr));
464 put_unaligned(np->remote_ip.ip, &(iph->daddr));
465 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
466
467 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
468 skb_reset_mac_header(skb);
469 skb->protocol = eth->h_proto = htons(ETH_P_IP);
470 }
471
472 ether_addr_copy(eth->h_source, np->dev->dev_addr);
473 ether_addr_copy(eth->h_dest, np->remote_mac);
474
475 skb->dev = np->dev;
476
477 netpoll_send_skb(np, skb);
478}
479EXPORT_SYMBOL(netpoll_send_udp);
480
481void netpoll_print_options(struct netpoll *np)
482{
483 np_info(np, "local port %d\n", np->local_port);
484 if (np->ipv6)
485 np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
486 else
487 np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
488 np_info(np, "interface '%s'\n", np->dev_name);
489 np_info(np, "remote port %d\n", np->remote_port);
490 if (np->ipv6)
491 np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
492 else
493 np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
494 np_info(np, "remote ethernet address %pM\n", np->remote_mac);
495}
496EXPORT_SYMBOL(netpoll_print_options);
497
498static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
499{
500 const char *end;
501
502 if (!strchr(str, ':') &&
503 in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
504 if (!*end)
505 return 0;
506 }
507 if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
508#if IS_ENABLED(CONFIG_IPV6)
509 if (!*end)
510 return 1;
511#else
512 return -1;
513#endif
514 }
515 return -1;
516}
517
518int netpoll_parse_options(struct netpoll *np, char *opt)
519{
520 char *cur=opt, *delim;
521 int ipv6;
522 bool ipversion_set = false;
523
524 if (*cur != '@') {
525 if ((delim = strchr(cur, '@')) == NULL)
526 goto parse_failed;
527 *delim = 0;
528 if (kstrtou16(cur, 10, &np->local_port))
529 goto parse_failed;
530 cur = delim;
531 }
532 cur++;
533
534 if (*cur != '/') {
535 ipversion_set = true;
536 if ((delim = strchr(cur, '/')) == NULL)
537 goto parse_failed;
538 *delim = 0;
539 ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
540 if (ipv6 < 0)
541 goto parse_failed;
542 else
543 np->ipv6 = (bool)ipv6;
544 cur = delim;
545 }
546 cur++;
547
548 if (*cur != ',') {
549 /* parse out dev name */
550 if ((delim = strchr(cur, ',')) == NULL)
551 goto parse_failed;
552 *delim = 0;
553 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
554 cur = delim;
555 }
556 cur++;
557
558 if (*cur != '@') {
559 /* dst port */
560 if ((delim = strchr(cur, '@')) == NULL)
561 goto parse_failed;
562 *delim = 0;
563 if (*cur == ' ' || *cur == '\t')
564 np_info(np, "warning: whitespace is not allowed\n");
565 if (kstrtou16(cur, 10, &np->remote_port))
566 goto parse_failed;
567 cur = delim;
568 }
569 cur++;
570
571 /* dst ip */
572 if ((delim = strchr(cur, '/')) == NULL)
573 goto parse_failed;
574 *delim = 0;
575 ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
576 if (ipv6 < 0)
577 goto parse_failed;
578 else if (ipversion_set && np->ipv6 != (bool)ipv6)
579 goto parse_failed;
580 else
581 np->ipv6 = (bool)ipv6;
582 cur = delim + 1;
583
584 if (*cur != 0) {
585 /* MAC address */
586 if (!mac_pton(cur, np->remote_mac))
587 goto parse_failed;
588 }
589
590 netpoll_print_options(np);
591
592 return 0;
593
594 parse_failed:
595 np_info(np, "couldn't parse config at '%s'!\n", cur);
596 return -1;
597}
598EXPORT_SYMBOL(netpoll_parse_options);
599
600int __netpoll_setup(struct netpoll *np, struct net_device *ndev)
601{
602 struct netpoll_info *npinfo;
603 const struct net_device_ops *ops;
604 int err;
605
606 np->dev = ndev;
607 strlcpy(np->dev_name, ndev->name, IFNAMSIZ);
608 INIT_WORK(&np->cleanup_work, netpoll_async_cleanup);
609
610 if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) ||
611 !ndev->netdev_ops->ndo_poll_controller) {
612 np_err(np, "%s doesn't support polling, aborting\n",
613 np->dev_name);
614 err = -ENOTSUPP;
615 goto out;
616 }
617
618 if (!ndev->npinfo) {
619 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
620 if (!npinfo) {
621 err = -ENOMEM;
622 goto out;
623 }
624
625 sema_init(&npinfo->dev_lock, 1);
626 skb_queue_head_init(&npinfo->txq);
627 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
628
629 atomic_set(&npinfo->refcnt, 1);
630
631 ops = np->dev->netdev_ops;
632 if (ops->ndo_netpoll_setup) {
633 err = ops->ndo_netpoll_setup(ndev, npinfo);
634 if (err)
635 goto free_npinfo;
636 }
637 } else {
638 npinfo = rtnl_dereference(ndev->npinfo);
639 atomic_inc(&npinfo->refcnt);
640 }
641
642 npinfo->netpoll = np;
643
644 /* last thing to do is link it to the net device structure */
645 rcu_assign_pointer(ndev->npinfo, npinfo);
646
647 return 0;
648
649free_npinfo:
650 kfree(npinfo);
651out:
652 return err;
653}
654EXPORT_SYMBOL_GPL(__netpoll_setup);
655
656int netpoll_setup(struct netpoll *np)
657{
658 struct net_device *ndev = NULL;
659 struct in_device *in_dev;
660 int err;
661
662 rtnl_lock();
663 if (np->dev_name) {
664 struct net *net = current->nsproxy->net_ns;
665 ndev = __dev_get_by_name(net, np->dev_name);
666 }
667 if (!ndev) {
668 np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
669 err = -ENODEV;
670 goto unlock;
671 }
672 dev_hold(ndev);
673
674 if (netdev_master_upper_dev_get(ndev)) {
675 np_err(np, "%s is a slave device, aborting\n", np->dev_name);
676 err = -EBUSY;
677 goto put;
678 }
679
680 if (!netif_running(ndev)) {
681 unsigned long atmost, atleast;
682
683 np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
684
685 err = dev_open(ndev);
686
687 if (err) {
688 np_err(np, "failed to open %s\n", ndev->name);
689 goto put;
690 }
691
692 rtnl_unlock();
693 atleast = jiffies + HZ/10;
694 atmost = jiffies + carrier_timeout * HZ;
695 while (!netif_carrier_ok(ndev)) {
696 if (time_after(jiffies, atmost)) {
697 np_notice(np, "timeout waiting for carrier\n");
698 break;
699 }
700 msleep(1);
701 }
702
703 /* If carrier appears to come up instantly, we don't
704 * trust it and pause so that we don't pump all our
705 * queued console messages into the bitbucket.
706 */
707
708 if (time_before(jiffies, atleast)) {
709 np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
710 msleep(4000);
711 }
712 rtnl_lock();
713 }
714
715 if (!np->local_ip.ip) {
716 if (!np->ipv6) {
717 in_dev = __in_dev_get_rtnl(ndev);
718
719 if (!in_dev || !in_dev->ifa_list) {
720 np_err(np, "no IP address for %s, aborting\n",
721 np->dev_name);
722 err = -EDESTADDRREQ;
723 goto put;
724 }
725
726 np->local_ip.ip = in_dev->ifa_list->ifa_local;
727 np_info(np, "local IP %pI4\n", &np->local_ip.ip);
728 } else {
729#if IS_ENABLED(CONFIG_IPV6)
730 struct inet6_dev *idev;
731
732 err = -EDESTADDRREQ;
733 idev = __in6_dev_get(ndev);
734 if (idev) {
735 struct inet6_ifaddr *ifp;
736
737 read_lock_bh(&idev->lock);
738 list_for_each_entry(ifp, &idev->addr_list, if_list) {
739 if (ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL)
740 continue;
741 np->local_ip.in6 = ifp->addr;
742 err = 0;
743 break;
744 }
745 read_unlock_bh(&idev->lock);
746 }
747 if (err) {
748 np_err(np, "no IPv6 address for %s, aborting\n",
749 np->dev_name);
750 goto put;
751 } else
752 np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
753#else
754 np_err(np, "IPv6 is not supported %s, aborting\n",
755 np->dev_name);
756 err = -EINVAL;
757 goto put;
758#endif
759 }
760 }
761
762 /* fill up the skb queue */
763 refill_skbs();
764
765 err = __netpoll_setup(np, ndev);
766 if (err)
767 goto put;
768
769 rtnl_unlock();
770 return 0;
771
772put:
773 dev_put(ndev);
774unlock:
775 rtnl_unlock();
776 return err;
777}
778EXPORT_SYMBOL(netpoll_setup);
779
780static int __init netpoll_init(void)
781{
782 skb_queue_head_init(&skb_pool);
783 return 0;
784}
785core_initcall(netpoll_init);
786
787static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
788{
789 struct netpoll_info *npinfo =
790 container_of(rcu_head, struct netpoll_info, rcu);
791
792 skb_queue_purge(&npinfo->txq);
793
794 /* we can't call cancel_delayed_work_sync here, as we are in softirq */
795 cancel_delayed_work(&npinfo->tx_work);
796
797 /* clean after last, unfinished work */
798 __skb_queue_purge(&npinfo->txq);
799 /* now cancel it again */
800 cancel_delayed_work(&npinfo->tx_work);
801 kfree(npinfo);
802}
803
804void __netpoll_cleanup(struct netpoll *np)
805{
806 struct netpoll_info *npinfo;
807
808 /* rtnl_dereference would be preferable here but
809 * rcu_cleanup_netpoll path can put us in here safely without
810 * holding the rtnl, so plain rcu_dereference it is
811 */
812 npinfo = rtnl_dereference(np->dev->npinfo);
813 if (!npinfo)
814 return;
815
816 synchronize_srcu(&netpoll_srcu);
817
818 if (atomic_dec_and_test(&npinfo->refcnt)) {
819 const struct net_device_ops *ops;
820
821 ops = np->dev->netdev_ops;
822 if (ops->ndo_netpoll_cleanup)
823 ops->ndo_netpoll_cleanup(np->dev);
824
825 RCU_INIT_POINTER(np->dev->npinfo, NULL);
826 call_rcu_bh(&npinfo->rcu, rcu_cleanup_netpoll_info);
827 } else
828 RCU_INIT_POINTER(np->dev->npinfo, NULL);
829}
830EXPORT_SYMBOL_GPL(__netpoll_cleanup);
831
832static void netpoll_async_cleanup(struct work_struct *work)
833{
834 struct netpoll *np = container_of(work, struct netpoll, cleanup_work);
835
836 rtnl_lock();
837 __netpoll_cleanup(np);
838 rtnl_unlock();
839 kfree(np);
840}
841
842void __netpoll_free_async(struct netpoll *np)
843{
844 schedule_work(&np->cleanup_work);
845}
846EXPORT_SYMBOL_GPL(__netpoll_free_async);
847
848void netpoll_cleanup(struct netpoll *np)
849{
850 rtnl_lock();
851 if (!np->dev)
852 goto out;
853 __netpoll_cleanup(np);
854 dev_put(np->dev);
855 np->dev = NULL;
856out:
857 rtnl_unlock();
858}
859EXPORT_SYMBOL(netpoll_cleanup);