Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Common framework for low-level network console, dump, and debugger code
4 *
5 * Sep 8 2003 Matt Mackall <mpm@selenic.com>
6 *
7 * based on the netconsole code from:
8 *
9 * Copyright (C) 2001 Ingo Molnar <mingo@redhat.com>
10 * Copyright (C) 2002 Red Hat, Inc.
11 */
12
13#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14
15#include <linux/moduleparam.h>
16#include <linux/kernel.h>
17#include <linux/netdevice.h>
18#include <linux/etherdevice.h>
19#include <linux/string.h>
20#include <linux/if_arp.h>
21#include <linux/inetdevice.h>
22#include <linux/inet.h>
23#include <linux/interrupt.h>
24#include <linux/netpoll.h>
25#include <linux/sched.h>
26#include <linux/delay.h>
27#include <linux/rcupdate.h>
28#include <linux/workqueue.h>
29#include <linux/slab.h>
30#include <linux/export.h>
31#include <linux/if_vlan.h>
32#include <net/tcp.h>
33#include <net/udp.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>
39
40/*
41 * We maintain a small pool of fully-sized skbs, to make sure the
42 * message gets out even in extreme OOM situations.
43 */
44
45#define MAX_UDP_CHUNK 1460
46#define MAX_SKBS 32
47
48static struct sk_buff_head skb_pool;
49
50DEFINE_STATIC_SRCU(netpoll_srcu);
51
52#define USEC_PER_POLL 50
53
54#define MAX_SKB_SIZE \
55 (sizeof(struct ethhdr) + \
56 sizeof(struct iphdr) + \
57 sizeof(struct udphdr) + \
58 MAX_UDP_CHUNK)
59
60static void zap_completion_queue(void);
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 netdev_tx_t netpoll_start_xmit(struct sk_buff *skb,
73 struct net_device *dev,
74 struct netdev_queue *txq)
75{
76 netdev_tx_t status = NETDEV_TX_OK;
77 netdev_features_t features;
78
79 features = netif_skb_features(skb);
80
81 if (skb_vlan_tag_present(skb) &&
82 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
83 skb = __vlan_hwaccel_push_inside(skb);
84 if (unlikely(!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.
88 */
89 goto out;
90 }
91 }
92
93 status = netdev_start_xmit(skb, dev, txq, false);
94
95out:
96 return status;
97}
98
99static void queue_process(struct work_struct *work)
100{
101 struct netpoll_info *npinfo =
102 container_of(work, struct netpoll_info, tx_work.work);
103 struct sk_buff *skb;
104 unsigned long flags;
105
106 while ((skb = skb_dequeue(&npinfo->txq))) {
107 struct net_device *dev = skb->dev;
108 struct netdev_queue *txq;
109 unsigned int q_index;
110
111 if (!netif_device_present(dev) || !netif_running(dev)) {
112 kfree_skb(skb);
113 continue;
114 }
115
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);
122 }
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);
130
131 schedule_delayed_work(&npinfo->tx_work, HZ/10);
132 return;
133 }
134 HARD_TX_UNLOCK(dev, txq);
135 local_irq_restore(flags);
136 }
137}
138
139static void poll_one_napi(struct napi_struct *napi)
140{
141 int work;
142
143 /* If we set this bit but see that it has already been set,
144 * that indicates that napi has been disabled and we need
145 * to abort this operation
146 */
147 if (test_and_set_bit(NAPI_STATE_NPSVC, &napi->state))
148 return;
149
150 /* We explicilty pass the polling call a budget of 0 to
151 * indicate that we are clearing the Tx path only.
152 */
153 work = napi->poll(napi, 0);
154 WARN_ONCE(work, "%pS exceeded budget in poll\n", napi->poll);
155 trace_napi_poll(napi, work, 0);
156
157 clear_bit(NAPI_STATE_NPSVC, &napi->state);
158}
159
160static void poll_napi(struct net_device *dev)
161{
162 struct napi_struct *napi;
163 int cpu = smp_processor_id();
164
165 list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) {
166 if (cmpxchg(&napi->poll_owner, -1, cpu) == -1) {
167 poll_one_napi(napi);
168 smp_store_release(&napi->poll_owner, -1);
169 }
170 }
171}
172
173void netpoll_poll_dev(struct net_device *dev)
174{
175 struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
176 const struct net_device_ops *ops;
177
178 /* Don't do any rx activity if the dev_lock mutex is held
179 * the dev_open/close paths use this to block netpoll activity
180 * while changing device state
181 */
182 if (!ni || down_trylock(&ni->dev_lock))
183 return;
184
185 if (!netif_running(dev)) {
186 up(&ni->dev_lock);
187 return;
188 }
189
190 ops = dev->netdev_ops;
191 if (ops->ndo_poll_controller)
192 ops->ndo_poll_controller(dev);
193
194 poll_napi(dev);
195
196 up(&ni->dev_lock);
197
198 zap_completion_queue();
199}
200EXPORT_SYMBOL(netpoll_poll_dev);
201
202void netpoll_poll_disable(struct net_device *dev)
203{
204 struct netpoll_info *ni;
205 int idx;
206 might_sleep();
207 idx = srcu_read_lock(&netpoll_srcu);
208 ni = srcu_dereference(dev->npinfo, &netpoll_srcu);
209 if (ni)
210 down(&ni->dev_lock);
211 srcu_read_unlock(&netpoll_srcu, idx);
212}
213EXPORT_SYMBOL(netpoll_poll_disable);
214
215void netpoll_poll_enable(struct net_device *dev)
216{
217 struct netpoll_info *ni;
218 rcu_read_lock();
219 ni = rcu_dereference(dev->npinfo);
220 if (ni)
221 up(&ni->dev_lock);
222 rcu_read_unlock();
223}
224EXPORT_SYMBOL(netpoll_poll_enable);
225
226static void refill_skbs(void)
227{
228 struct sk_buff *skb;
229 unsigned long flags;
230
231 spin_lock_irqsave(&skb_pool.lock, flags);
232 while (skb_pool.qlen < MAX_SKBS) {
233 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
234 if (!skb)
235 break;
236
237 __skb_queue_tail(&skb_pool, skb);
238 }
239 spin_unlock_irqrestore(&skb_pool.lock, flags);
240}
241
242static void zap_completion_queue(void)
243{
244 unsigned long flags;
245 struct softnet_data *sd = &get_cpu_var(softnet_data);
246
247 if (sd->completion_queue) {
248 struct sk_buff *clist;
249
250 local_irq_save(flags);
251 clist = sd->completion_queue;
252 sd->completion_queue = NULL;
253 local_irq_restore(flags);
254
255 while (clist != NULL) {
256 struct sk_buff *skb = clist;
257 clist = clist->next;
258 if (!skb_irq_freeable(skb)) {
259 refcount_set(&skb->users, 1);
260 dev_kfree_skb_any(skb); /* put this one back */
261 } else {
262 __kfree_skb(skb);
263 }
264 }
265 }
266
267 put_cpu_var(softnet_data);
268}
269
270static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
271{
272 int count = 0;
273 struct sk_buff *skb;
274
275 zap_completion_queue();
276 refill_skbs();
277repeat:
278
279 skb = alloc_skb(len, GFP_ATOMIC);
280 if (!skb)
281 skb = skb_dequeue(&skb_pool);
282
283 if (!skb) {
284 if (++count < 10) {
285 netpoll_poll_dev(np->dev);
286 goto repeat;
287 }
288 return NULL;
289 }
290
291 refcount_set(&skb->users, 1);
292 skb_reserve(skb, reserve);
293 return skb;
294}
295
296static int netpoll_owner_active(struct net_device *dev)
297{
298 struct napi_struct *napi;
299
300 list_for_each_entry(napi, &dev->napi_list, dev_list) {
301 if (napi->poll_owner == smp_processor_id())
302 return 1;
303 }
304 return 0;
305}
306
307/* call with IRQ disabled */
308static netdev_tx_t __netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
309{
310 netdev_tx_t status = NETDEV_TX_BUSY;
311 struct net_device *dev;
312 unsigned long tries;
313 /* It is up to the caller to keep npinfo alive. */
314 struct netpoll_info *npinfo;
315
316 lockdep_assert_irqs_disabled();
317
318 dev = np->dev;
319 npinfo = rcu_dereference_bh(dev->npinfo);
320
321 if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
322 dev_kfree_skb_irq(skb);
323 return NET_XMIT_DROP;
324 }
325
326 /* don't get messages out of order, and no recursion */
327 if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
328 struct netdev_queue *txq;
329
330 txq = netdev_core_pick_tx(dev, skb, NULL);
331
332 /* try until next clock tick */
333 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
334 tries > 0; --tries) {
335 if (HARD_TX_TRYLOCK(dev, txq)) {
336 if (!netif_xmit_stopped(txq))
337 status = netpoll_start_xmit(skb, dev, txq);
338
339 HARD_TX_UNLOCK(dev, txq);
340
341 if (dev_xmit_complete(status))
342 break;
343
344 }
345
346 /* tickle device maybe there is some cleanup */
347 netpoll_poll_dev(np->dev);
348
349 udelay(USEC_PER_POLL);
350 }
351
352 WARN_ONCE(!irqs_disabled(),
353 "netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pS)\n",
354 dev->name, dev->netdev_ops->ndo_start_xmit);
355
356 }
357
358 if (!dev_xmit_complete(status)) {
359 skb_queue_tail(&npinfo->txq, skb);
360 schedule_delayed_work(&npinfo->tx_work,0);
361 }
362 return NETDEV_TX_OK;
363}
364
365netdev_tx_t netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
366{
367 unsigned long flags;
368 netdev_tx_t ret;
369
370 if (unlikely(!np)) {
371 dev_kfree_skb_irq(skb);
372 ret = NET_XMIT_DROP;
373 } else {
374 local_irq_save(flags);
375 ret = __netpoll_send_skb(np, skb);
376 local_irq_restore(flags);
377 }
378 return ret;
379}
380EXPORT_SYMBOL(netpoll_send_skb);
381
382void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
383{
384 int total_len, ip_len, udp_len;
385 struct sk_buff *skb;
386 struct udphdr *udph;
387 struct iphdr *iph;
388 struct ethhdr *eth;
389 static atomic_t ip_ident;
390 struct ipv6hdr *ip6h;
391
392 WARN_ON_ONCE(!irqs_disabled());
393
394 udp_len = len + sizeof(*udph);
395 if (np->ipv6)
396 ip_len = udp_len + sizeof(*ip6h);
397 else
398 ip_len = udp_len + sizeof(*iph);
399
400 total_len = ip_len + LL_RESERVED_SPACE(np->dev);
401
402 skb = find_skb(np, total_len + np->dev->needed_tailroom,
403 total_len - len);
404 if (!skb)
405 return;
406
407 skb_copy_to_linear_data(skb, msg, len);
408 skb_put(skb, len);
409
410 skb_push(skb, sizeof(*udph));
411 skb_reset_transport_header(skb);
412 udph = udp_hdr(skb);
413 udph->source = htons(np->local_port);
414 udph->dest = htons(np->remote_port);
415 udph->len = htons(udp_len);
416
417 if (np->ipv6) {
418 udph->check = 0;
419 udph->check = csum_ipv6_magic(&np->local_ip.in6,
420 &np->remote_ip.in6,
421 udp_len, IPPROTO_UDP,
422 csum_partial(udph, udp_len, 0));
423 if (udph->check == 0)
424 udph->check = CSUM_MANGLED_0;
425
426 skb_push(skb, sizeof(*ip6h));
427 skb_reset_network_header(skb);
428 ip6h = ipv6_hdr(skb);
429
430 /* ip6h->version = 6; ip6h->priority = 0; */
431 put_unaligned(0x60, (unsigned char *)ip6h);
432 ip6h->flow_lbl[0] = 0;
433 ip6h->flow_lbl[1] = 0;
434 ip6h->flow_lbl[2] = 0;
435
436 ip6h->payload_len = htons(sizeof(struct udphdr) + len);
437 ip6h->nexthdr = IPPROTO_UDP;
438 ip6h->hop_limit = 32;
439 ip6h->saddr = np->local_ip.in6;
440 ip6h->daddr = np->remote_ip.in6;
441
442 eth = skb_push(skb, ETH_HLEN);
443 skb_reset_mac_header(skb);
444 skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
445 } else {
446 udph->check = 0;
447 udph->check = csum_tcpudp_magic(np->local_ip.ip,
448 np->remote_ip.ip,
449 udp_len, IPPROTO_UDP,
450 csum_partial(udph, udp_len, 0));
451 if (udph->check == 0)
452 udph->check = CSUM_MANGLED_0;
453
454 skb_push(skb, sizeof(*iph));
455 skb_reset_network_header(skb);
456 iph = ip_hdr(skb);
457
458 /* iph->version = 4; iph->ihl = 5; */
459 put_unaligned(0x45, (unsigned char *)iph);
460 iph->tos = 0;
461 put_unaligned(htons(ip_len), &(iph->tot_len));
462 iph->id = htons(atomic_inc_return(&ip_ident));
463 iph->frag_off = 0;
464 iph->ttl = 64;
465 iph->protocol = IPPROTO_UDP;
466 iph->check = 0;
467 put_unaligned(np->local_ip.ip, &(iph->saddr));
468 put_unaligned(np->remote_ip.ip, &(iph->daddr));
469 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
470
471 eth = skb_push(skb, ETH_HLEN);
472 skb_reset_mac_header(skb);
473 skb->protocol = eth->h_proto = htons(ETH_P_IP);
474 }
475
476 ether_addr_copy(eth->h_source, np->dev->dev_addr);
477 ether_addr_copy(eth->h_dest, np->remote_mac);
478
479 skb->dev = np->dev;
480
481 netpoll_send_skb(np, skb);
482}
483EXPORT_SYMBOL(netpoll_send_udp);
484
485void netpoll_print_options(struct netpoll *np)
486{
487 np_info(np, "local port %d\n", np->local_port);
488 if (np->ipv6)
489 np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
490 else
491 np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
492 np_info(np, "interface '%s'\n", np->dev_name);
493 np_info(np, "remote port %d\n", np->remote_port);
494 if (np->ipv6)
495 np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
496 else
497 np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
498 np_info(np, "remote ethernet address %pM\n", np->remote_mac);
499}
500EXPORT_SYMBOL(netpoll_print_options);
501
502static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
503{
504 const char *end;
505
506 if (!strchr(str, ':') &&
507 in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
508 if (!*end)
509 return 0;
510 }
511 if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
512#if IS_ENABLED(CONFIG_IPV6)
513 if (!*end)
514 return 1;
515#else
516 return -1;
517#endif
518 }
519 return -1;
520}
521
522int netpoll_parse_options(struct netpoll *np, char *opt)
523{
524 char *cur=opt, *delim;
525 int ipv6;
526 bool ipversion_set = false;
527
528 if (*cur != '@') {
529 if ((delim = strchr(cur, '@')) == NULL)
530 goto parse_failed;
531 *delim = 0;
532 if (kstrtou16(cur, 10, &np->local_port))
533 goto parse_failed;
534 cur = delim;
535 }
536 cur++;
537
538 if (*cur != '/') {
539 ipversion_set = true;
540 if ((delim = strchr(cur, '/')) == NULL)
541 goto parse_failed;
542 *delim = 0;
543 ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
544 if (ipv6 < 0)
545 goto parse_failed;
546 else
547 np->ipv6 = (bool)ipv6;
548 cur = delim;
549 }
550 cur++;
551
552 if (*cur != ',') {
553 /* parse out dev name */
554 if ((delim = strchr(cur, ',')) == NULL)
555 goto parse_failed;
556 *delim = 0;
557 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
558 cur = delim;
559 }
560 cur++;
561
562 if (*cur != '@') {
563 /* dst port */
564 if ((delim = strchr(cur, '@')) == NULL)
565 goto parse_failed;
566 *delim = 0;
567 if (*cur == ' ' || *cur == '\t')
568 np_info(np, "warning: whitespace is not allowed\n");
569 if (kstrtou16(cur, 10, &np->remote_port))
570 goto parse_failed;
571 cur = delim;
572 }
573 cur++;
574
575 /* dst ip */
576 if ((delim = strchr(cur, '/')) == NULL)
577 goto parse_failed;
578 *delim = 0;
579 ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
580 if (ipv6 < 0)
581 goto parse_failed;
582 else if (ipversion_set && np->ipv6 != (bool)ipv6)
583 goto parse_failed;
584 else
585 np->ipv6 = (bool)ipv6;
586 cur = delim + 1;
587
588 if (*cur != 0) {
589 /* MAC address */
590 if (!mac_pton(cur, np->remote_mac))
591 goto parse_failed;
592 }
593
594 netpoll_print_options(np);
595
596 return 0;
597
598 parse_failed:
599 np_info(np, "couldn't parse config at '%s'!\n", cur);
600 return -1;
601}
602EXPORT_SYMBOL(netpoll_parse_options);
603
604int __netpoll_setup(struct netpoll *np, struct net_device *ndev)
605{
606 struct netpoll_info *npinfo;
607 const struct net_device_ops *ops;
608 int err;
609
610 np->dev = ndev;
611 strlcpy(np->dev_name, ndev->name, IFNAMSIZ);
612
613 if (ndev->priv_flags & IFF_DISABLE_NETPOLL) {
614 np_err(np, "%s doesn't support polling, aborting\n",
615 np->dev_name);
616 err = -ENOTSUPP;
617 goto out;
618 }
619
620 if (!ndev->npinfo) {
621 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
622 if (!npinfo) {
623 err = -ENOMEM;
624 goto out;
625 }
626
627 sema_init(&npinfo->dev_lock, 1);
628 skb_queue_head_init(&npinfo->txq);
629 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
630
631 refcount_set(&npinfo->refcnt, 1);
632
633 ops = np->dev->netdev_ops;
634 if (ops->ndo_netpoll_setup) {
635 err = ops->ndo_netpoll_setup(ndev, npinfo);
636 if (err)
637 goto free_npinfo;
638 }
639 } else {
640 npinfo = rtnl_dereference(ndev->npinfo);
641 refcount_inc(&npinfo->refcnt);
642 }
643
644 npinfo->netpoll = np;
645
646 /* last thing to do is link it to the net device structure */
647 rcu_assign_pointer(ndev->npinfo, npinfo);
648
649 return 0;
650
651free_npinfo:
652 kfree(npinfo);
653out:
654 return err;
655}
656EXPORT_SYMBOL_GPL(__netpoll_setup);
657
658int netpoll_setup(struct netpoll *np)
659{
660 struct net_device *ndev = NULL;
661 struct in_device *in_dev;
662 int err;
663
664 rtnl_lock();
665 if (np->dev_name[0]) {
666 struct net *net = current->nsproxy->net_ns;
667 ndev = __dev_get_by_name(net, np->dev_name);
668 }
669 if (!ndev) {
670 np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
671 err = -ENODEV;
672 goto unlock;
673 }
674 dev_hold(ndev);
675
676 if (netdev_master_upper_dev_get(ndev)) {
677 np_err(np, "%s is a slave device, aborting\n", np->dev_name);
678 err = -EBUSY;
679 goto put;
680 }
681
682 if (!netif_running(ndev)) {
683 unsigned long atmost, atleast;
684
685 np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
686
687 err = dev_open(ndev, NULL);
688
689 if (err) {
690 np_err(np, "failed to open %s\n", ndev->name);
691 goto put;
692 }
693
694 rtnl_unlock();
695 atleast = jiffies + HZ/10;
696 atmost = jiffies + carrier_timeout * HZ;
697 while (!netif_carrier_ok(ndev)) {
698 if (time_after(jiffies, atmost)) {
699 np_notice(np, "timeout waiting for carrier\n");
700 break;
701 }
702 msleep(1);
703 }
704
705 /* If carrier appears to come up instantly, we don't
706 * trust it and pause so that we don't pump all our
707 * queued console messages into the bitbucket.
708 */
709
710 if (time_before(jiffies, atleast)) {
711 np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
712 msleep(4000);
713 }
714 rtnl_lock();
715 }
716
717 if (!np->local_ip.ip) {
718 if (!np->ipv6) {
719 const struct in_ifaddr *ifa;
720
721 in_dev = __in_dev_get_rtnl(ndev);
722 if (!in_dev)
723 goto put_noaddr;
724
725 ifa = rtnl_dereference(in_dev->ifa_list);
726 if (!ifa) {
727put_noaddr:
728 np_err(np, "no IP address for %s, aborting\n",
729 np->dev_name);
730 err = -EDESTADDRREQ;
731 goto put;
732 }
733
734 np->local_ip.ip = ifa->ifa_local;
735 np_info(np, "local IP %pI4\n", &np->local_ip.ip);
736 } else {
737#if IS_ENABLED(CONFIG_IPV6)
738 struct inet6_dev *idev;
739
740 err = -EDESTADDRREQ;
741 idev = __in6_dev_get(ndev);
742 if (idev) {
743 struct inet6_ifaddr *ifp;
744
745 read_lock_bh(&idev->lock);
746 list_for_each_entry(ifp, &idev->addr_list, if_list) {
747 if (!!(ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL) !=
748 !!(ipv6_addr_type(&np->remote_ip.in6) & IPV6_ADDR_LINKLOCAL))
749 continue;
750 np->local_ip.in6 = ifp->addr;
751 err = 0;
752 break;
753 }
754 read_unlock_bh(&idev->lock);
755 }
756 if (err) {
757 np_err(np, "no IPv6 address for %s, aborting\n",
758 np->dev_name);
759 goto put;
760 } else
761 np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
762#else
763 np_err(np, "IPv6 is not supported %s, aborting\n",
764 np->dev_name);
765 err = -EINVAL;
766 goto put;
767#endif
768 }
769 }
770
771 /* fill up the skb queue */
772 refill_skbs();
773
774 err = __netpoll_setup(np, ndev);
775 if (err)
776 goto put;
777
778 rtnl_unlock();
779 return 0;
780
781put:
782 dev_put(ndev);
783unlock:
784 rtnl_unlock();
785 return err;
786}
787EXPORT_SYMBOL(netpoll_setup);
788
789static int __init netpoll_init(void)
790{
791 skb_queue_head_init(&skb_pool);
792 return 0;
793}
794core_initcall(netpoll_init);
795
796static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
797{
798 struct netpoll_info *npinfo =
799 container_of(rcu_head, struct netpoll_info, rcu);
800
801 skb_queue_purge(&npinfo->txq);
802
803 /* we can't call cancel_delayed_work_sync here, as we are in softirq */
804 cancel_delayed_work(&npinfo->tx_work);
805
806 /* clean after last, unfinished work */
807 __skb_queue_purge(&npinfo->txq);
808 /* now cancel it again */
809 cancel_delayed_work(&npinfo->tx_work);
810 kfree(npinfo);
811}
812
813void __netpoll_cleanup(struct netpoll *np)
814{
815 struct netpoll_info *npinfo;
816
817 npinfo = rtnl_dereference(np->dev->npinfo);
818 if (!npinfo)
819 return;
820
821 synchronize_srcu(&netpoll_srcu);
822
823 if (refcount_dec_and_test(&npinfo->refcnt)) {
824 const struct net_device_ops *ops;
825
826 ops = np->dev->netdev_ops;
827 if (ops->ndo_netpoll_cleanup)
828 ops->ndo_netpoll_cleanup(np->dev);
829
830 RCU_INIT_POINTER(np->dev->npinfo, NULL);
831 call_rcu(&npinfo->rcu, rcu_cleanup_netpoll_info);
832 } else
833 RCU_INIT_POINTER(np->dev->npinfo, NULL);
834}
835EXPORT_SYMBOL_GPL(__netpoll_cleanup);
836
837void __netpoll_free(struct netpoll *np)
838{
839 ASSERT_RTNL();
840
841 /* Wait for transmitting packets to finish before freeing. */
842 synchronize_rcu();
843 __netpoll_cleanup(np);
844 kfree(np);
845}
846EXPORT_SYMBOL_GPL(__netpoll_free);
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);
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 const struct net_device_ops *ops = dev->netdev_ops;
76 int status = NETDEV_TX_OK;
77 netdev_features_t features;
78
79 features = netif_skb_features(skb);
80
81 if (vlan_tx_tag_present(skb) &&
82 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
83 skb = __vlan_put_tag(skb, skb->vlan_proto,
84 vlan_tx_tag_get(skb));
85 if (unlikely(!skb)) {
86 /* This is actually a packet drop, but we
87 * don't want the code that calls this
88 * function to try and operate on a NULL skb.
89 */
90 goto out;
91 }
92 skb->vlan_tci = 0;
93 }
94
95 status = ops->ndo_start_xmit(skb, dev);
96 if (status == NETDEV_TX_OK)
97 txq_trans_update(txq);
98
99out:
100 return status;
101}
102
103static void queue_process(struct work_struct *work)
104{
105 struct netpoll_info *npinfo =
106 container_of(work, struct netpoll_info, tx_work.work);
107 struct sk_buff *skb;
108 unsigned long flags;
109
110 while ((skb = skb_dequeue(&npinfo->txq))) {
111 struct net_device *dev = skb->dev;
112 struct netdev_queue *txq;
113
114 if (!netif_device_present(dev) || !netif_running(dev)) {
115 kfree_skb(skb);
116 continue;
117 }
118
119 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
120
121 local_irq_save(flags);
122 HARD_TX_LOCK(dev, txq, smp_processor_id());
123 if (netif_xmit_frozen_or_stopped(txq) ||
124 netpoll_start_xmit(skb, dev, txq) != NETDEV_TX_OK) {
125 skb_queue_head(&npinfo->txq, skb);
126 HARD_TX_UNLOCK(dev, txq);
127 local_irq_restore(flags);
128
129 schedule_delayed_work(&npinfo->tx_work, HZ/10);
130 return;
131 }
132 HARD_TX_UNLOCK(dev, txq);
133 local_irq_restore(flags);
134 }
135}
136
137/*
138 * Check whether delayed processing was scheduled for our NIC. If so,
139 * we attempt to grab the poll lock and use ->poll() to pump the card.
140 * If this fails, either we've recursed in ->poll() or it's already
141 * running on another CPU.
142 *
143 * Note: we don't mask interrupts with this lock because we're using
144 * trylock here and interrupts are already disabled in the softirq
145 * case. Further, we test the poll_owner to avoid recursion on UP
146 * systems where the lock doesn't exist.
147 */
148static int poll_one_napi(struct napi_struct *napi, int budget)
149{
150 int work;
151
152 /* net_rx_action's ->poll() invocations and our's are
153 * synchronized by this test which is only made while
154 * holding the napi->poll_lock.
155 */
156 if (!test_bit(NAPI_STATE_SCHED, &napi->state))
157 return budget;
158
159 set_bit(NAPI_STATE_NPSVC, &napi->state);
160
161 work = napi->poll(napi, budget);
162 WARN_ONCE(work > budget, "%pF exceeded budget in poll\n", napi->poll);
163 trace_napi_poll(napi);
164
165 clear_bit(NAPI_STATE_NPSVC, &napi->state);
166
167 return budget - work;
168}
169
170static void poll_napi(struct net_device *dev, int budget)
171{
172 struct napi_struct *napi;
173
174 list_for_each_entry(napi, &dev->napi_list, dev_list) {
175 if (napi->poll_owner != smp_processor_id() &&
176 spin_trylock(&napi->poll_lock)) {
177 budget = poll_one_napi(napi, budget);
178 spin_unlock(&napi->poll_lock);
179 }
180 }
181}
182
183static void netpoll_poll_dev(struct net_device *dev)
184{
185 const struct net_device_ops *ops;
186 struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
187 int budget = 0;
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, budget);
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 udp_len = len + sizeof(*udph);
389 if (np->ipv6)
390 ip_len = udp_len + sizeof(*ip6h);
391 else
392 ip_len = udp_len + sizeof(*iph);
393
394 total_len = ip_len + LL_RESERVED_SPACE(np->dev);
395
396 skb = find_skb(np, total_len + np->dev->needed_tailroom,
397 total_len - len);
398 if (!skb)
399 return;
400
401 skb_copy_to_linear_data(skb, msg, len);
402 skb_put(skb, len);
403
404 skb_push(skb, sizeof(*udph));
405 skb_reset_transport_header(skb);
406 udph = udp_hdr(skb);
407 udph->source = htons(np->local_port);
408 udph->dest = htons(np->remote_port);
409 udph->len = htons(udp_len);
410
411 if (np->ipv6) {
412 udph->check = 0;
413 udph->check = csum_ipv6_magic(&np->local_ip.in6,
414 &np->remote_ip.in6,
415 udp_len, IPPROTO_UDP,
416 csum_partial(udph, udp_len, 0));
417 if (udph->check == 0)
418 udph->check = CSUM_MANGLED_0;
419
420 skb_push(skb, sizeof(*ip6h));
421 skb_reset_network_header(skb);
422 ip6h = ipv6_hdr(skb);
423
424 /* ip6h->version = 6; ip6h->priority = 0; */
425 put_unaligned(0x60, (unsigned char *)ip6h);
426 ip6h->flow_lbl[0] = 0;
427 ip6h->flow_lbl[1] = 0;
428 ip6h->flow_lbl[2] = 0;
429
430 ip6h->payload_len = htons(sizeof(struct udphdr) + len);
431 ip6h->nexthdr = IPPROTO_UDP;
432 ip6h->hop_limit = 32;
433 ip6h->saddr = np->local_ip.in6;
434 ip6h->daddr = np->remote_ip.in6;
435
436 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
437 skb_reset_mac_header(skb);
438 skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
439 } else {
440 udph->check = 0;
441 udph->check = csum_tcpudp_magic(np->local_ip.ip,
442 np->remote_ip.ip,
443 udp_len, IPPROTO_UDP,
444 csum_partial(udph, udp_len, 0));
445 if (udph->check == 0)
446 udph->check = CSUM_MANGLED_0;
447
448 skb_push(skb, sizeof(*iph));
449 skb_reset_network_header(skb);
450 iph = ip_hdr(skb);
451
452 /* iph->version = 4; iph->ihl = 5; */
453 put_unaligned(0x45, (unsigned char *)iph);
454 iph->tos = 0;
455 put_unaligned(htons(ip_len), &(iph->tot_len));
456 iph->id = htons(atomic_inc_return(&ip_ident));
457 iph->frag_off = 0;
458 iph->ttl = 64;
459 iph->protocol = IPPROTO_UDP;
460 iph->check = 0;
461 put_unaligned(np->local_ip.ip, &(iph->saddr));
462 put_unaligned(np->remote_ip.ip, &(iph->daddr));
463 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
464
465 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
466 skb_reset_mac_header(skb);
467 skb->protocol = eth->h_proto = htons(ETH_P_IP);
468 }
469
470 ether_addr_copy(eth->h_source, np->dev->dev_addr);
471 ether_addr_copy(eth->h_dest, np->remote_mac);
472
473 skb->dev = np->dev;
474
475 netpoll_send_skb(np, skb);
476}
477EXPORT_SYMBOL(netpoll_send_udp);
478
479void netpoll_print_options(struct netpoll *np)
480{
481 np_info(np, "local port %d\n", np->local_port);
482 if (np->ipv6)
483 np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
484 else
485 np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
486 np_info(np, "interface '%s'\n", np->dev_name);
487 np_info(np, "remote port %d\n", np->remote_port);
488 if (np->ipv6)
489 np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
490 else
491 np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
492 np_info(np, "remote ethernet address %pM\n", np->remote_mac);
493}
494EXPORT_SYMBOL(netpoll_print_options);
495
496static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
497{
498 const char *end;
499
500 if (!strchr(str, ':') &&
501 in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
502 if (!*end)
503 return 0;
504 }
505 if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
506#if IS_ENABLED(CONFIG_IPV6)
507 if (!*end)
508 return 1;
509#else
510 return -1;
511#endif
512 }
513 return -1;
514}
515
516int netpoll_parse_options(struct netpoll *np, char *opt)
517{
518 char *cur=opt, *delim;
519 int ipv6;
520 bool ipversion_set = false;
521
522 if (*cur != '@') {
523 if ((delim = strchr(cur, '@')) == NULL)
524 goto parse_failed;
525 *delim = 0;
526 if (kstrtou16(cur, 10, &np->local_port))
527 goto parse_failed;
528 cur = delim;
529 }
530 cur++;
531
532 if (*cur != '/') {
533 ipversion_set = true;
534 if ((delim = strchr(cur, '/')) == NULL)
535 goto parse_failed;
536 *delim = 0;
537 ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
538 if (ipv6 < 0)
539 goto parse_failed;
540 else
541 np->ipv6 = (bool)ipv6;
542 cur = delim;
543 }
544 cur++;
545
546 if (*cur != ',') {
547 /* parse out dev name */
548 if ((delim = strchr(cur, ',')) == NULL)
549 goto parse_failed;
550 *delim = 0;
551 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
552 cur = delim;
553 }
554 cur++;
555
556 if (*cur != '@') {
557 /* dst port */
558 if ((delim = strchr(cur, '@')) == NULL)
559 goto parse_failed;
560 *delim = 0;
561 if (*cur == ' ' || *cur == '\t')
562 np_info(np, "warning: whitespace is not allowed\n");
563 if (kstrtou16(cur, 10, &np->remote_port))
564 goto parse_failed;
565 cur = delim;
566 }
567 cur++;
568
569 /* dst ip */
570 if ((delim = strchr(cur, '/')) == NULL)
571 goto parse_failed;
572 *delim = 0;
573 ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
574 if (ipv6 < 0)
575 goto parse_failed;
576 else if (ipversion_set && np->ipv6 != (bool)ipv6)
577 goto parse_failed;
578 else
579 np->ipv6 = (bool)ipv6;
580 cur = delim + 1;
581
582 if (*cur != 0) {
583 /* MAC address */
584 if (!mac_pton(cur, np->remote_mac))
585 goto parse_failed;
586 }
587
588 netpoll_print_options(np);
589
590 return 0;
591
592 parse_failed:
593 np_info(np, "couldn't parse config at '%s'!\n", cur);
594 return -1;
595}
596EXPORT_SYMBOL(netpoll_parse_options);
597
598int __netpoll_setup(struct netpoll *np, struct net_device *ndev)
599{
600 struct netpoll_info *npinfo;
601 const struct net_device_ops *ops;
602 int err;
603
604 np->dev = ndev;
605 strlcpy(np->dev_name, ndev->name, IFNAMSIZ);
606 INIT_WORK(&np->cleanup_work, netpoll_async_cleanup);
607
608 if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) ||
609 !ndev->netdev_ops->ndo_poll_controller) {
610 np_err(np, "%s doesn't support polling, aborting\n",
611 np->dev_name);
612 err = -ENOTSUPP;
613 goto out;
614 }
615
616 if (!ndev->npinfo) {
617 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
618 if (!npinfo) {
619 err = -ENOMEM;
620 goto out;
621 }
622
623 sema_init(&npinfo->dev_lock, 1);
624 skb_queue_head_init(&npinfo->txq);
625 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
626
627 atomic_set(&npinfo->refcnt, 1);
628
629 ops = np->dev->netdev_ops;
630 if (ops->ndo_netpoll_setup) {
631 err = ops->ndo_netpoll_setup(ndev, npinfo);
632 if (err)
633 goto free_npinfo;
634 }
635 } else {
636 npinfo = rtnl_dereference(ndev->npinfo);
637 atomic_inc(&npinfo->refcnt);
638 }
639
640 npinfo->netpoll = np;
641
642 /* last thing to do is link it to the net device structure */
643 rcu_assign_pointer(ndev->npinfo, npinfo);
644
645 return 0;
646
647free_npinfo:
648 kfree(npinfo);
649out:
650 return err;
651}
652EXPORT_SYMBOL_GPL(__netpoll_setup);
653
654int netpoll_setup(struct netpoll *np)
655{
656 struct net_device *ndev = NULL;
657 struct in_device *in_dev;
658 int err;
659
660 rtnl_lock();
661 if (np->dev_name) {
662 struct net *net = current->nsproxy->net_ns;
663 ndev = __dev_get_by_name(net, np->dev_name);
664 }
665 if (!ndev) {
666 np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
667 err = -ENODEV;
668 goto unlock;
669 }
670 dev_hold(ndev);
671
672 if (netdev_master_upper_dev_get(ndev)) {
673 np_err(np, "%s is a slave device, aborting\n", np->dev_name);
674 err = -EBUSY;
675 goto put;
676 }
677
678 if (!netif_running(ndev)) {
679 unsigned long atmost, atleast;
680
681 np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
682
683 err = dev_open(ndev);
684
685 if (err) {
686 np_err(np, "failed to open %s\n", ndev->name);
687 goto put;
688 }
689
690 rtnl_unlock();
691 atleast = jiffies + HZ/10;
692 atmost = jiffies + carrier_timeout * HZ;
693 while (!netif_carrier_ok(ndev)) {
694 if (time_after(jiffies, atmost)) {
695 np_notice(np, "timeout waiting for carrier\n");
696 break;
697 }
698 msleep(1);
699 }
700
701 /* If carrier appears to come up instantly, we don't
702 * trust it and pause so that we don't pump all our
703 * queued console messages into the bitbucket.
704 */
705
706 if (time_before(jiffies, atleast)) {
707 np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
708 msleep(4000);
709 }
710 rtnl_lock();
711 }
712
713 if (!np->local_ip.ip) {
714 if (!np->ipv6) {
715 in_dev = __in_dev_get_rtnl(ndev);
716
717 if (!in_dev || !in_dev->ifa_list) {
718 np_err(np, "no IP address for %s, aborting\n",
719 np->dev_name);
720 err = -EDESTADDRREQ;
721 goto put;
722 }
723
724 np->local_ip.ip = in_dev->ifa_list->ifa_local;
725 np_info(np, "local IP %pI4\n", &np->local_ip.ip);
726 } else {
727#if IS_ENABLED(CONFIG_IPV6)
728 struct inet6_dev *idev;
729
730 err = -EDESTADDRREQ;
731 idev = __in6_dev_get(ndev);
732 if (idev) {
733 struct inet6_ifaddr *ifp;
734
735 read_lock_bh(&idev->lock);
736 list_for_each_entry(ifp, &idev->addr_list, if_list) {
737 if (ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL)
738 continue;
739 np->local_ip.in6 = ifp->addr;
740 err = 0;
741 break;
742 }
743 read_unlock_bh(&idev->lock);
744 }
745 if (err) {
746 np_err(np, "no IPv6 address for %s, aborting\n",
747 np->dev_name);
748 goto put;
749 } else
750 np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
751#else
752 np_err(np, "IPv6 is not supported %s, aborting\n",
753 np->dev_name);
754 err = -EINVAL;
755 goto put;
756#endif
757 }
758 }
759
760 /* fill up the skb queue */
761 refill_skbs();
762
763 err = __netpoll_setup(np, ndev);
764 if (err)
765 goto put;
766
767 rtnl_unlock();
768 return 0;
769
770put:
771 dev_put(ndev);
772unlock:
773 rtnl_unlock();
774 return err;
775}
776EXPORT_SYMBOL(netpoll_setup);
777
778static int __init netpoll_init(void)
779{
780 skb_queue_head_init(&skb_pool);
781 return 0;
782}
783core_initcall(netpoll_init);
784
785static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
786{
787 struct netpoll_info *npinfo =
788 container_of(rcu_head, struct netpoll_info, rcu);
789
790 skb_queue_purge(&npinfo->txq);
791
792 /* we can't call cancel_delayed_work_sync here, as we are in softirq */
793 cancel_delayed_work(&npinfo->tx_work);
794
795 /* clean after last, unfinished work */
796 __skb_queue_purge(&npinfo->txq);
797 /* now cancel it again */
798 cancel_delayed_work(&npinfo->tx_work);
799 kfree(npinfo);
800}
801
802void __netpoll_cleanup(struct netpoll *np)
803{
804 struct netpoll_info *npinfo;
805
806 /* rtnl_dereference would be preferable here but
807 * rcu_cleanup_netpoll path can put us in here safely without
808 * holding the rtnl, so plain rcu_dereference it is
809 */
810 npinfo = rtnl_dereference(np->dev->npinfo);
811 if (!npinfo)
812 return;
813
814 synchronize_srcu(&netpoll_srcu);
815
816 if (atomic_dec_and_test(&npinfo->refcnt)) {
817 const struct net_device_ops *ops;
818
819 ops = np->dev->netdev_ops;
820 if (ops->ndo_netpoll_cleanup)
821 ops->ndo_netpoll_cleanup(np->dev);
822
823 RCU_INIT_POINTER(np->dev->npinfo, NULL);
824 call_rcu_bh(&npinfo->rcu, rcu_cleanup_netpoll_info);
825 }
826}
827EXPORT_SYMBOL_GPL(__netpoll_cleanup);
828
829static void netpoll_async_cleanup(struct work_struct *work)
830{
831 struct netpoll *np = container_of(work, struct netpoll, cleanup_work);
832
833 rtnl_lock();
834 __netpoll_cleanup(np);
835 rtnl_unlock();
836 kfree(np);
837}
838
839void __netpoll_free_async(struct netpoll *np)
840{
841 schedule_work(&np->cleanup_work);
842}
843EXPORT_SYMBOL_GPL(__netpoll_free_async);
844
845void netpoll_cleanup(struct netpoll *np)
846{
847 rtnl_lock();
848 if (!np->dev)
849 goto out;
850 __netpoll_cleanup(np);
851 dev_put(np->dev);
852 np->dev = NULL;
853out:
854 rtnl_unlock();
855}
856EXPORT_SYMBOL(netpoll_cleanup);