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 <linux/unaligned.h>
38#include <trace/events/napi.h>
39#include <linux/kconfig.h>
40
41/*
42 * We maintain a small pool of fully-sized skbs, to make sure the
43 * message gets out even in extreme OOM situations.
44 */
45
46#define MAX_UDP_CHUNK 1460
47#define MAX_SKBS 32
48#define USEC_PER_POLL 50
49
50#define MAX_SKB_SIZE \
51 (sizeof(struct ethhdr) + \
52 sizeof(struct iphdr) + \
53 sizeof(struct udphdr) + \
54 MAX_UDP_CHUNK)
55
56static void zap_completion_queue(void);
57
58static unsigned int carrier_timeout = 4;
59module_param(carrier_timeout, uint, 0644);
60
61#define np_info(np, fmt, ...) \
62 pr_info("%s: " fmt, np->name, ##__VA_ARGS__)
63#define np_err(np, fmt, ...) \
64 pr_err("%s: " fmt, np->name, ##__VA_ARGS__)
65#define np_notice(np, fmt, ...) \
66 pr_notice("%s: " fmt, np->name, ##__VA_ARGS__)
67
68static netdev_tx_t netpoll_start_xmit(struct sk_buff *skb,
69 struct net_device *dev,
70 struct netdev_queue *txq)
71{
72 netdev_tx_t status = NETDEV_TX_OK;
73 netdev_features_t features;
74
75 features = netif_skb_features(skb);
76
77 if (skb_vlan_tag_present(skb) &&
78 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
79 skb = __vlan_hwaccel_push_inside(skb);
80 if (unlikely(!skb)) {
81 /* This is actually a packet drop, but we
82 * don't want the code that calls this
83 * function to try and operate on a NULL skb.
84 */
85 goto out;
86 }
87 }
88
89 status = netdev_start_xmit(skb, dev, txq, false);
90
91out:
92 return status;
93}
94
95static void queue_process(struct work_struct *work)
96{
97 struct netpoll_info *npinfo =
98 container_of(work, struct netpoll_info, tx_work.work);
99 struct sk_buff *skb;
100 unsigned long flags;
101
102 while ((skb = skb_dequeue(&npinfo->txq))) {
103 struct net_device *dev = skb->dev;
104 struct netdev_queue *txq;
105 unsigned int q_index;
106
107 if (!netif_device_present(dev) || !netif_running(dev)) {
108 kfree_skb(skb);
109 continue;
110 }
111
112 local_irq_save(flags);
113 /* check if skb->queue_mapping is still valid */
114 q_index = skb_get_queue_mapping(skb);
115 if (unlikely(q_index >= dev->real_num_tx_queues)) {
116 q_index = q_index % dev->real_num_tx_queues;
117 skb_set_queue_mapping(skb, q_index);
118 }
119 txq = netdev_get_tx_queue(dev, q_index);
120 HARD_TX_LOCK(dev, txq, smp_processor_id());
121 if (netif_xmit_frozen_or_stopped(txq) ||
122 !dev_xmit_complete(netpoll_start_xmit(skb, dev, txq))) {
123 skb_queue_head(&npinfo->txq, skb);
124 HARD_TX_UNLOCK(dev, txq);
125 local_irq_restore(flags);
126
127 schedule_delayed_work(&npinfo->tx_work, HZ/10);
128 return;
129 }
130 HARD_TX_UNLOCK(dev, txq);
131 local_irq_restore(flags);
132 }
133}
134
135static int netif_local_xmit_active(struct net_device *dev)
136{
137 int i;
138
139 for (i = 0; i < dev->num_tx_queues; i++) {
140 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
141
142 if (READ_ONCE(txq->xmit_lock_owner) == smp_processor_id())
143 return 1;
144 }
145
146 return 0;
147}
148
149static void poll_one_napi(struct napi_struct *napi)
150{
151 int work;
152
153 /* If we set this bit but see that it has already been set,
154 * that indicates that napi has been disabled and we need
155 * to abort this operation
156 */
157 if (test_and_set_bit(NAPI_STATE_NPSVC, &napi->state))
158 return;
159
160 /* We explicitly pass the polling call a budget of 0 to
161 * indicate that we are clearing the Tx path only.
162 */
163 work = napi->poll(napi, 0);
164 WARN_ONCE(work, "%pS exceeded budget in poll\n", napi->poll);
165 trace_napi_poll(napi, work, 0);
166
167 clear_bit(NAPI_STATE_NPSVC, &napi->state);
168}
169
170static void poll_napi(struct net_device *dev)
171{
172 struct napi_struct *napi;
173 int cpu = smp_processor_id();
174
175 list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) {
176 if (cmpxchg(&napi->poll_owner, -1, cpu) == -1) {
177 poll_one_napi(napi);
178 smp_store_release(&napi->poll_owner, -1);
179 }
180 }
181}
182
183void netpoll_poll_dev(struct net_device *dev)
184{
185 struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
186 const struct net_device_ops *ops;
187
188 /* Don't do any rx activity if the dev_lock mutex is held
189 * the dev_open/close paths use this to block netpoll activity
190 * while changing device state
191 */
192 if (!ni || down_trylock(&ni->dev_lock))
193 return;
194
195 /* Some drivers will take the same locks in poll and xmit,
196 * we can't poll if local CPU is already in xmit.
197 */
198 if (!netif_running(dev) || netif_local_xmit_active(dev)) {
199 up(&ni->dev_lock);
200 return;
201 }
202
203 ops = dev->netdev_ops;
204 if (ops->ndo_poll_controller)
205 ops->ndo_poll_controller(dev);
206
207 poll_napi(dev);
208
209 up(&ni->dev_lock);
210
211 zap_completion_queue();
212}
213EXPORT_SYMBOL(netpoll_poll_dev);
214
215void netpoll_poll_disable(struct net_device *dev)
216{
217 struct netpoll_info *ni;
218
219 might_sleep();
220 ni = rtnl_dereference(dev->npinfo);
221 if (ni)
222 down(&ni->dev_lock);
223}
224
225void netpoll_poll_enable(struct net_device *dev)
226{
227 struct netpoll_info *ni;
228
229 ni = rtnl_dereference(dev->npinfo);
230 if (ni)
231 up(&ni->dev_lock);
232}
233
234static void refill_skbs(struct netpoll *np)
235{
236 struct sk_buff_head *skb_pool;
237 struct sk_buff *skb;
238 unsigned long flags;
239
240 skb_pool = &np->skb_pool;
241
242 spin_lock_irqsave(&skb_pool->lock, flags);
243 while (skb_pool->qlen < MAX_SKBS) {
244 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
245 if (!skb)
246 break;
247
248 __skb_queue_tail(skb_pool, skb);
249 }
250 spin_unlock_irqrestore(&skb_pool->lock, flags);
251}
252
253static void zap_completion_queue(void)
254{
255 unsigned long flags;
256 struct softnet_data *sd = &get_cpu_var(softnet_data);
257
258 if (sd->completion_queue) {
259 struct sk_buff *clist;
260
261 local_irq_save(flags);
262 clist = sd->completion_queue;
263 sd->completion_queue = NULL;
264 local_irq_restore(flags);
265
266 while (clist != NULL) {
267 struct sk_buff *skb = clist;
268 clist = clist->next;
269 if (!skb_irq_freeable(skb)) {
270 refcount_set(&skb->users, 1);
271 dev_kfree_skb_any(skb); /* put this one back */
272 } else {
273 __kfree_skb(skb);
274 }
275 }
276 }
277
278 put_cpu_var(softnet_data);
279}
280
281static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
282{
283 int count = 0;
284 struct sk_buff *skb;
285
286 zap_completion_queue();
287 refill_skbs(np);
288repeat:
289
290 skb = alloc_skb(len, GFP_ATOMIC);
291 if (!skb)
292 skb = skb_dequeue(&np->skb_pool);
293
294 if (!skb) {
295 if (++count < 10) {
296 netpoll_poll_dev(np->dev);
297 goto repeat;
298 }
299 return NULL;
300 }
301
302 refcount_set(&skb->users, 1);
303 skb_reserve(skb, reserve);
304 return skb;
305}
306
307static int netpoll_owner_active(struct net_device *dev)
308{
309 struct napi_struct *napi;
310
311 list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) {
312 if (READ_ONCE(napi->poll_owner) == smp_processor_id())
313 return 1;
314 }
315 return 0;
316}
317
318/* call with IRQ disabled */
319static netdev_tx_t __netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
320{
321 netdev_tx_t status = NETDEV_TX_BUSY;
322 struct net_device *dev;
323 unsigned long tries;
324 /* It is up to the caller to keep npinfo alive. */
325 struct netpoll_info *npinfo;
326
327 lockdep_assert_irqs_disabled();
328
329 dev = np->dev;
330 npinfo = rcu_dereference_bh(dev->npinfo);
331
332 if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
333 dev_kfree_skb_irq(skb);
334 return NET_XMIT_DROP;
335 }
336
337 /* don't get messages out of order, and no recursion */
338 if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
339 struct netdev_queue *txq;
340
341 txq = netdev_core_pick_tx(dev, skb, NULL);
342
343 /* try until next clock tick */
344 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
345 tries > 0; --tries) {
346 if (HARD_TX_TRYLOCK(dev, txq)) {
347 if (!netif_xmit_stopped(txq))
348 status = netpoll_start_xmit(skb, dev, txq);
349
350 HARD_TX_UNLOCK(dev, txq);
351
352 if (dev_xmit_complete(status))
353 break;
354
355 }
356
357 /* tickle device maybe there is some cleanup */
358 netpoll_poll_dev(np->dev);
359
360 udelay(USEC_PER_POLL);
361 }
362
363 WARN_ONCE(!irqs_disabled(),
364 "netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pS)\n",
365 dev->name, dev->netdev_ops->ndo_start_xmit);
366
367 }
368
369 if (!dev_xmit_complete(status)) {
370 skb_queue_tail(&npinfo->txq, skb);
371 schedule_delayed_work(&npinfo->tx_work,0);
372 }
373 return NETDEV_TX_OK;
374}
375
376netdev_tx_t netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
377{
378 unsigned long flags;
379 netdev_tx_t ret;
380
381 if (unlikely(!np)) {
382 dev_kfree_skb_irq(skb);
383 ret = NET_XMIT_DROP;
384 } else {
385 local_irq_save(flags);
386 ret = __netpoll_send_skb(np, skb);
387 local_irq_restore(flags);
388 }
389 return ret;
390}
391EXPORT_SYMBOL(netpoll_send_skb);
392
393void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
394{
395 int total_len, ip_len, udp_len;
396 struct sk_buff *skb;
397 struct udphdr *udph;
398 struct iphdr *iph;
399 struct ethhdr *eth;
400 static atomic_t ip_ident;
401 struct ipv6hdr *ip6h;
402
403 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
404 WARN_ON_ONCE(!irqs_disabled());
405
406 udp_len = len + sizeof(*udph);
407 if (np->ipv6)
408 ip_len = udp_len + sizeof(*ip6h);
409 else
410 ip_len = udp_len + sizeof(*iph);
411
412 total_len = ip_len + LL_RESERVED_SPACE(np->dev);
413
414 skb = find_skb(np, total_len + np->dev->needed_tailroom,
415 total_len - len);
416 if (!skb)
417 return;
418
419 skb_copy_to_linear_data(skb, msg, len);
420 skb_put(skb, len);
421
422 skb_push(skb, sizeof(*udph));
423 skb_reset_transport_header(skb);
424 udph = udp_hdr(skb);
425 udph->source = htons(np->local_port);
426 udph->dest = htons(np->remote_port);
427 udph->len = htons(udp_len);
428
429 if (np->ipv6) {
430 udph->check = 0;
431 udph->check = csum_ipv6_magic(&np->local_ip.in6,
432 &np->remote_ip.in6,
433 udp_len, IPPROTO_UDP,
434 csum_partial(udph, udp_len, 0));
435 if (udph->check == 0)
436 udph->check = CSUM_MANGLED_0;
437
438 skb_push(skb, sizeof(*ip6h));
439 skb_reset_network_header(skb);
440 ip6h = ipv6_hdr(skb);
441
442 /* ip6h->version = 6; ip6h->priority = 0; */
443 *(unsigned char *)ip6h = 0x60;
444 ip6h->flow_lbl[0] = 0;
445 ip6h->flow_lbl[1] = 0;
446 ip6h->flow_lbl[2] = 0;
447
448 ip6h->payload_len = htons(sizeof(struct udphdr) + len);
449 ip6h->nexthdr = IPPROTO_UDP;
450 ip6h->hop_limit = 32;
451 ip6h->saddr = np->local_ip.in6;
452 ip6h->daddr = np->remote_ip.in6;
453
454 eth = skb_push(skb, ETH_HLEN);
455 skb_reset_mac_header(skb);
456 skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
457 } else {
458 udph->check = 0;
459 udph->check = csum_tcpudp_magic(np->local_ip.ip,
460 np->remote_ip.ip,
461 udp_len, IPPROTO_UDP,
462 csum_partial(udph, udp_len, 0));
463 if (udph->check == 0)
464 udph->check = CSUM_MANGLED_0;
465
466 skb_push(skb, sizeof(*iph));
467 skb_reset_network_header(skb);
468 iph = ip_hdr(skb);
469
470 /* iph->version = 4; iph->ihl = 5; */
471 *(unsigned char *)iph = 0x45;
472 iph->tos = 0;
473 put_unaligned(htons(ip_len), &(iph->tot_len));
474 iph->id = htons(atomic_inc_return(&ip_ident));
475 iph->frag_off = 0;
476 iph->ttl = 64;
477 iph->protocol = IPPROTO_UDP;
478 iph->check = 0;
479 put_unaligned(np->local_ip.ip, &(iph->saddr));
480 put_unaligned(np->remote_ip.ip, &(iph->daddr));
481 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
482
483 eth = skb_push(skb, ETH_HLEN);
484 skb_reset_mac_header(skb);
485 skb->protocol = eth->h_proto = htons(ETH_P_IP);
486 }
487
488 ether_addr_copy(eth->h_source, np->dev->dev_addr);
489 ether_addr_copy(eth->h_dest, np->remote_mac);
490
491 skb->dev = np->dev;
492
493 netpoll_send_skb(np, skb);
494}
495EXPORT_SYMBOL(netpoll_send_udp);
496
497void netpoll_print_options(struct netpoll *np)
498{
499 np_info(np, "local port %d\n", np->local_port);
500 if (np->ipv6)
501 np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
502 else
503 np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
504 np_info(np, "interface '%s'\n", np->dev_name);
505 np_info(np, "remote port %d\n", np->remote_port);
506 if (np->ipv6)
507 np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
508 else
509 np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
510 np_info(np, "remote ethernet address %pM\n", np->remote_mac);
511}
512EXPORT_SYMBOL(netpoll_print_options);
513
514static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
515{
516 const char *end;
517
518 if (!strchr(str, ':') &&
519 in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
520 if (!*end)
521 return 0;
522 }
523 if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
524#if IS_ENABLED(CONFIG_IPV6)
525 if (!*end)
526 return 1;
527#else
528 return -1;
529#endif
530 }
531 return -1;
532}
533
534static void skb_pool_flush(struct netpoll *np)
535{
536 struct sk_buff_head *skb_pool;
537
538 skb_pool = &np->skb_pool;
539 skb_queue_purge_reason(skb_pool, SKB_CONSUMED);
540}
541
542int netpoll_parse_options(struct netpoll *np, char *opt)
543{
544 char *cur=opt, *delim;
545 int ipv6;
546 bool ipversion_set = false;
547
548 if (*cur != '@') {
549 if ((delim = strchr(cur, '@')) == NULL)
550 goto parse_failed;
551 *delim = 0;
552 if (kstrtou16(cur, 10, &np->local_port))
553 goto parse_failed;
554 cur = delim;
555 }
556 cur++;
557
558 if (*cur != '/') {
559 ipversion_set = true;
560 if ((delim = strchr(cur, '/')) == NULL)
561 goto parse_failed;
562 *delim = 0;
563 ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
564 if (ipv6 < 0)
565 goto parse_failed;
566 else
567 np->ipv6 = (bool)ipv6;
568 cur = delim;
569 }
570 cur++;
571
572 if (*cur != ',') {
573 /* parse out dev name */
574 if ((delim = strchr(cur, ',')) == NULL)
575 goto parse_failed;
576 *delim = 0;
577 strscpy(np->dev_name, cur, sizeof(np->dev_name));
578 cur = delim;
579 }
580 cur++;
581
582 if (*cur != '@') {
583 /* dst port */
584 if ((delim = strchr(cur, '@')) == NULL)
585 goto parse_failed;
586 *delim = 0;
587 if (*cur == ' ' || *cur == '\t')
588 np_info(np, "warning: whitespace is not allowed\n");
589 if (kstrtou16(cur, 10, &np->remote_port))
590 goto parse_failed;
591 cur = delim;
592 }
593 cur++;
594
595 /* dst ip */
596 if ((delim = strchr(cur, '/')) == NULL)
597 goto parse_failed;
598 *delim = 0;
599 ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
600 if (ipv6 < 0)
601 goto parse_failed;
602 else if (ipversion_set && np->ipv6 != (bool)ipv6)
603 goto parse_failed;
604 else
605 np->ipv6 = (bool)ipv6;
606 cur = delim + 1;
607
608 if (*cur != 0) {
609 /* MAC address */
610 if (!mac_pton(cur, np->remote_mac))
611 goto parse_failed;
612 }
613
614 netpoll_print_options(np);
615
616 return 0;
617
618 parse_failed:
619 np_info(np, "couldn't parse config at '%s'!\n", cur);
620 return -1;
621}
622EXPORT_SYMBOL(netpoll_parse_options);
623
624int __netpoll_setup(struct netpoll *np, struct net_device *ndev)
625{
626 struct netpoll_info *npinfo;
627 const struct net_device_ops *ops;
628 int err;
629
630 skb_queue_head_init(&np->skb_pool);
631
632 if (ndev->priv_flags & IFF_DISABLE_NETPOLL) {
633 np_err(np, "%s doesn't support polling, aborting\n",
634 ndev->name);
635 err = -ENOTSUPP;
636 goto out;
637 }
638
639 if (!rcu_access_pointer(ndev->npinfo)) {
640 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
641 if (!npinfo) {
642 err = -ENOMEM;
643 goto out;
644 }
645
646 sema_init(&npinfo->dev_lock, 1);
647 skb_queue_head_init(&npinfo->txq);
648 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
649
650 refcount_set(&npinfo->refcnt, 1);
651
652 ops = ndev->netdev_ops;
653 if (ops->ndo_netpoll_setup) {
654 err = ops->ndo_netpoll_setup(ndev);
655 if (err)
656 goto free_npinfo;
657 }
658 } else {
659 npinfo = rtnl_dereference(ndev->npinfo);
660 refcount_inc(&npinfo->refcnt);
661 }
662
663 np->dev = ndev;
664 strscpy(np->dev_name, ndev->name, IFNAMSIZ);
665 npinfo->netpoll = np;
666
667 /* fill up the skb queue */
668 refill_skbs(np);
669
670 /* last thing to do is link it to the net device structure */
671 rcu_assign_pointer(ndev->npinfo, npinfo);
672
673 return 0;
674
675free_npinfo:
676 kfree(npinfo);
677out:
678 return err;
679}
680EXPORT_SYMBOL_GPL(__netpoll_setup);
681
682int netpoll_setup(struct netpoll *np)
683{
684 struct net_device *ndev = NULL;
685 bool ip_overwritten = false;
686 struct in_device *in_dev;
687 int err;
688
689 rtnl_lock();
690 if (np->dev_name[0]) {
691 struct net *net = current->nsproxy->net_ns;
692 ndev = __dev_get_by_name(net, np->dev_name);
693 }
694 if (!ndev) {
695 np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
696 err = -ENODEV;
697 goto unlock;
698 }
699 netdev_hold(ndev, &np->dev_tracker, GFP_KERNEL);
700
701 if (netdev_master_upper_dev_get(ndev)) {
702 np_err(np, "%s is a slave device, aborting\n", np->dev_name);
703 err = -EBUSY;
704 goto put;
705 }
706
707 if (!netif_running(ndev)) {
708 unsigned long atmost;
709
710 np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
711
712 err = dev_open(ndev, NULL);
713
714 if (err) {
715 np_err(np, "failed to open %s\n", ndev->name);
716 goto put;
717 }
718
719 rtnl_unlock();
720 atmost = jiffies + carrier_timeout * HZ;
721 while (!netif_carrier_ok(ndev)) {
722 if (time_after(jiffies, atmost)) {
723 np_notice(np, "timeout waiting for carrier\n");
724 break;
725 }
726 msleep(1);
727 }
728
729 rtnl_lock();
730 }
731
732 if (!np->local_ip.ip) {
733 if (!np->ipv6) {
734 const struct in_ifaddr *ifa;
735
736 in_dev = __in_dev_get_rtnl(ndev);
737 if (!in_dev)
738 goto put_noaddr;
739
740 ifa = rtnl_dereference(in_dev->ifa_list);
741 if (!ifa) {
742put_noaddr:
743 np_err(np, "no IP address for %s, aborting\n",
744 np->dev_name);
745 err = -EDESTADDRREQ;
746 goto put;
747 }
748
749 np->local_ip.ip = ifa->ifa_local;
750 ip_overwritten = true;
751 np_info(np, "local IP %pI4\n", &np->local_ip.ip);
752 } else {
753#if IS_ENABLED(CONFIG_IPV6)
754 struct inet6_dev *idev;
755
756 err = -EDESTADDRREQ;
757 idev = __in6_dev_get(ndev);
758 if (idev) {
759 struct inet6_ifaddr *ifp;
760
761 read_lock_bh(&idev->lock);
762 list_for_each_entry(ifp, &idev->addr_list, if_list) {
763 if (!!(ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL) !=
764 !!(ipv6_addr_type(&np->remote_ip.in6) & IPV6_ADDR_LINKLOCAL))
765 continue;
766 np->local_ip.in6 = ifp->addr;
767 ip_overwritten = true;
768 err = 0;
769 break;
770 }
771 read_unlock_bh(&idev->lock);
772 }
773 if (err) {
774 np_err(np, "no IPv6 address for %s, aborting\n",
775 np->dev_name);
776 goto put;
777 } else
778 np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
779#else
780 np_err(np, "IPv6 is not supported %s, aborting\n",
781 np->dev_name);
782 err = -EINVAL;
783 goto put;
784#endif
785 }
786 }
787
788 err = __netpoll_setup(np, ndev);
789 if (err)
790 goto flush;
791 rtnl_unlock();
792 return 0;
793
794flush:
795 skb_pool_flush(np);
796put:
797 DEBUG_NET_WARN_ON_ONCE(np->dev);
798 if (ip_overwritten)
799 memset(&np->local_ip, 0, sizeof(np->local_ip));
800 netdev_put(ndev, &np->dev_tracker);
801unlock:
802 rtnl_unlock();
803 return err;
804}
805EXPORT_SYMBOL(netpoll_setup);
806
807static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
808{
809 struct netpoll_info *npinfo =
810 container_of(rcu_head, struct netpoll_info, rcu);
811
812 skb_queue_purge(&npinfo->txq);
813
814 /* we can't call cancel_delayed_work_sync here, as we are in softirq */
815 cancel_delayed_work(&npinfo->tx_work);
816
817 /* clean after last, unfinished work */
818 __skb_queue_purge(&npinfo->txq);
819 /* now cancel it again */
820 cancel_delayed_work(&npinfo->tx_work);
821 kfree(npinfo);
822}
823
824void __netpoll_cleanup(struct netpoll *np)
825{
826 struct netpoll_info *npinfo;
827
828 npinfo = rtnl_dereference(np->dev->npinfo);
829 if (!npinfo)
830 return;
831
832 if (refcount_dec_and_test(&npinfo->refcnt)) {
833 const struct net_device_ops *ops;
834
835 ops = np->dev->netdev_ops;
836 if (ops->ndo_netpoll_cleanup)
837 ops->ndo_netpoll_cleanup(np->dev);
838
839 RCU_INIT_POINTER(np->dev->npinfo, NULL);
840 call_rcu(&npinfo->rcu, rcu_cleanup_netpoll_info);
841 } else
842 RCU_INIT_POINTER(np->dev->npinfo, NULL);
843
844 skb_pool_flush(np);
845}
846EXPORT_SYMBOL_GPL(__netpoll_cleanup);
847
848void __netpoll_free(struct netpoll *np)
849{
850 ASSERT_RTNL();
851
852 /* Wait for transmitting packets to finish before freeing. */
853 synchronize_rcu();
854 __netpoll_cleanup(np);
855 kfree(np);
856}
857EXPORT_SYMBOL_GPL(__netpoll_free);
858
859void do_netpoll_cleanup(struct netpoll *np)
860{
861 __netpoll_cleanup(np);
862 netdev_put(np->dev, &np->dev_tracker);
863 np->dev = NULL;
864}
865EXPORT_SYMBOL(do_netpoll_cleanup);
866
867void netpoll_cleanup(struct netpoll *np)
868{
869 rtnl_lock();
870 if (!np->dev)
871 goto out;
872 do_netpoll_cleanup(np);
873out:
874 rtnl_unlock();
875}
876EXPORT_SYMBOL(netpoll_cleanup);
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 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 !dev_xmit_complete(netpoll_start_xmit(skb, dev, txq))) {
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
138static void poll_one_napi(struct napi_struct *napi)
139{
140 int work;
141
142 /* If we set this bit but see that it has already been set,
143 * that indicates that napi has been disabled and we need
144 * to abort this operation
145 */
146 if (test_and_set_bit(NAPI_STATE_NPSVC, &napi->state))
147 return;
148
149 /* We explicilty pass the polling call a budget of 0 to
150 * indicate that we are clearing the Tx path only.
151 */
152 work = napi->poll(napi, 0);
153 WARN_ONCE(work, "%pS exceeded budget in poll\n", napi->poll);
154 trace_napi_poll(napi, work, 0);
155
156 clear_bit(NAPI_STATE_NPSVC, &napi->state);
157}
158
159static void poll_napi(struct net_device *dev)
160{
161 struct napi_struct *napi;
162 int cpu = smp_processor_id();
163
164 list_for_each_entry(napi, &dev->napi_list, dev_list) {
165 if (cmpxchg(&napi->poll_owner, -1, cpu) == -1) {
166 poll_one_napi(napi);
167 smp_store_release(&napi->poll_owner, -1);
168 }
169 }
170}
171
172void netpoll_poll_dev(struct net_device *dev)
173{
174 struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
175 const struct net_device_ops *ops;
176
177 /* Don't do any rx activity if the dev_lock mutex is held
178 * the dev_open/close paths use this to block netpoll activity
179 * while changing device state
180 */
181 if (!ni || down_trylock(&ni->dev_lock))
182 return;
183
184 if (!netif_running(dev)) {
185 up(&ni->dev_lock);
186 return;
187 }
188
189 ops = dev->netdev_ops;
190 if (ops->ndo_poll_controller)
191 ops->ndo_poll_controller(dev);
192
193 poll_napi(dev);
194
195 up(&ni->dev_lock);
196
197 zap_completion_queue();
198}
199EXPORT_SYMBOL(netpoll_poll_dev);
200
201void netpoll_poll_disable(struct net_device *dev)
202{
203 struct netpoll_info *ni;
204 int idx;
205 might_sleep();
206 idx = srcu_read_lock(&netpoll_srcu);
207 ni = srcu_dereference(dev->npinfo, &netpoll_srcu);
208 if (ni)
209 down(&ni->dev_lock);
210 srcu_read_unlock(&netpoll_srcu, idx);
211}
212EXPORT_SYMBOL(netpoll_poll_disable);
213
214void netpoll_poll_enable(struct net_device *dev)
215{
216 struct netpoll_info *ni;
217 rcu_read_lock();
218 ni = rcu_dereference(dev->npinfo);
219 if (ni)
220 up(&ni->dev_lock);
221 rcu_read_unlock();
222}
223EXPORT_SYMBOL(netpoll_poll_enable);
224
225static void refill_skbs(void)
226{
227 struct sk_buff *skb;
228 unsigned long flags;
229
230 spin_lock_irqsave(&skb_pool.lock, flags);
231 while (skb_pool.qlen < MAX_SKBS) {
232 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
233 if (!skb)
234 break;
235
236 __skb_queue_tail(&skb_pool, skb);
237 }
238 spin_unlock_irqrestore(&skb_pool.lock, flags);
239}
240
241static void zap_completion_queue(void)
242{
243 unsigned long flags;
244 struct softnet_data *sd = &get_cpu_var(softnet_data);
245
246 if (sd->completion_queue) {
247 struct sk_buff *clist;
248
249 local_irq_save(flags);
250 clist = sd->completion_queue;
251 sd->completion_queue = NULL;
252 local_irq_restore(flags);
253
254 while (clist != NULL) {
255 struct sk_buff *skb = clist;
256 clist = clist->next;
257 if (!skb_irq_freeable(skb)) {
258 refcount_set(&skb->users, 1);
259 dev_kfree_skb_any(skb); /* put this one back */
260 } else {
261 __kfree_skb(skb);
262 }
263 }
264 }
265
266 put_cpu_var(softnet_data);
267}
268
269static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
270{
271 int count = 0;
272 struct sk_buff *skb;
273
274 zap_completion_queue();
275 refill_skbs();
276repeat:
277
278 skb = alloc_skb(len, GFP_ATOMIC);
279 if (!skb)
280 skb = skb_dequeue(&skb_pool);
281
282 if (!skb) {
283 if (++count < 10) {
284 netpoll_poll_dev(np->dev);
285 goto repeat;
286 }
287 return NULL;
288 }
289
290 refcount_set(&skb->users, 1);
291 skb_reserve(skb, reserve);
292 return skb;
293}
294
295static int netpoll_owner_active(struct net_device *dev)
296{
297 struct napi_struct *napi;
298
299 list_for_each_entry(napi, &dev->napi_list, dev_list) {
300 if (napi->poll_owner == smp_processor_id())
301 return 1;
302 }
303 return 0;
304}
305
306/* call with IRQ disabled */
307void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb,
308 struct net_device *dev)
309{
310 int status = NETDEV_TX_BUSY;
311 unsigned long tries;
312 /* It is up to the caller to keep npinfo alive. */
313 struct netpoll_info *npinfo;
314
315 lockdep_assert_irqs_disabled();
316
317 npinfo = rcu_dereference_bh(np->dev->npinfo);
318 if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
319 dev_kfree_skb_irq(skb);
320 return;
321 }
322
323 /* don't get messages out of order, and no recursion */
324 if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
325 struct netdev_queue *txq;
326
327 txq = netdev_core_pick_tx(dev, skb, NULL);
328
329 /* try until next clock tick */
330 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
331 tries > 0; --tries) {
332 if (HARD_TX_TRYLOCK(dev, txq)) {
333 if (!netif_xmit_stopped(txq))
334 status = netpoll_start_xmit(skb, dev, txq);
335
336 HARD_TX_UNLOCK(dev, txq);
337
338 if (dev_xmit_complete(status))
339 break;
340
341 }
342
343 /* tickle device maybe there is some cleanup */
344 netpoll_poll_dev(np->dev);
345
346 udelay(USEC_PER_POLL);
347 }
348
349 WARN_ONCE(!irqs_disabled(),
350 "netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pS)\n",
351 dev->name, dev->netdev_ops->ndo_start_xmit);
352
353 }
354
355 if (!dev_xmit_complete(status)) {
356 skb_queue_tail(&npinfo->txq, skb);
357 schedule_delayed_work(&npinfo->tx_work,0);
358 }
359}
360EXPORT_SYMBOL(netpoll_send_skb_on_dev);
361
362void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
363{
364 int total_len, ip_len, udp_len;
365 struct sk_buff *skb;
366 struct udphdr *udph;
367 struct iphdr *iph;
368 struct ethhdr *eth;
369 static atomic_t ip_ident;
370 struct ipv6hdr *ip6h;
371
372 WARN_ON_ONCE(!irqs_disabled());
373
374 udp_len = len + sizeof(*udph);
375 if (np->ipv6)
376 ip_len = udp_len + sizeof(*ip6h);
377 else
378 ip_len = udp_len + sizeof(*iph);
379
380 total_len = ip_len + LL_RESERVED_SPACE(np->dev);
381
382 skb = find_skb(np, total_len + np->dev->needed_tailroom,
383 total_len - len);
384 if (!skb)
385 return;
386
387 skb_copy_to_linear_data(skb, msg, len);
388 skb_put(skb, len);
389
390 skb_push(skb, sizeof(*udph));
391 skb_reset_transport_header(skb);
392 udph = udp_hdr(skb);
393 udph->source = htons(np->local_port);
394 udph->dest = htons(np->remote_port);
395 udph->len = htons(udp_len);
396
397 if (np->ipv6) {
398 udph->check = 0;
399 udph->check = csum_ipv6_magic(&np->local_ip.in6,
400 &np->remote_ip.in6,
401 udp_len, IPPROTO_UDP,
402 csum_partial(udph, udp_len, 0));
403 if (udph->check == 0)
404 udph->check = CSUM_MANGLED_0;
405
406 skb_push(skb, sizeof(*ip6h));
407 skb_reset_network_header(skb);
408 ip6h = ipv6_hdr(skb);
409
410 /* ip6h->version = 6; ip6h->priority = 0; */
411 put_unaligned(0x60, (unsigned char *)ip6h);
412 ip6h->flow_lbl[0] = 0;
413 ip6h->flow_lbl[1] = 0;
414 ip6h->flow_lbl[2] = 0;
415
416 ip6h->payload_len = htons(sizeof(struct udphdr) + len);
417 ip6h->nexthdr = IPPROTO_UDP;
418 ip6h->hop_limit = 32;
419 ip6h->saddr = np->local_ip.in6;
420 ip6h->daddr = np->remote_ip.in6;
421
422 eth = skb_push(skb, ETH_HLEN);
423 skb_reset_mac_header(skb);
424 skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
425 } else {
426 udph->check = 0;
427 udph->check = csum_tcpudp_magic(np->local_ip.ip,
428 np->remote_ip.ip,
429 udp_len, IPPROTO_UDP,
430 csum_partial(udph, udp_len, 0));
431 if (udph->check == 0)
432 udph->check = CSUM_MANGLED_0;
433
434 skb_push(skb, sizeof(*iph));
435 skb_reset_network_header(skb);
436 iph = ip_hdr(skb);
437
438 /* iph->version = 4; iph->ihl = 5; */
439 put_unaligned(0x45, (unsigned char *)iph);
440 iph->tos = 0;
441 put_unaligned(htons(ip_len), &(iph->tot_len));
442 iph->id = htons(atomic_inc_return(&ip_ident));
443 iph->frag_off = 0;
444 iph->ttl = 64;
445 iph->protocol = IPPROTO_UDP;
446 iph->check = 0;
447 put_unaligned(np->local_ip.ip, &(iph->saddr));
448 put_unaligned(np->remote_ip.ip, &(iph->daddr));
449 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
450
451 eth = skb_push(skb, ETH_HLEN);
452 skb_reset_mac_header(skb);
453 skb->protocol = eth->h_proto = htons(ETH_P_IP);
454 }
455
456 ether_addr_copy(eth->h_source, np->dev->dev_addr);
457 ether_addr_copy(eth->h_dest, np->remote_mac);
458
459 skb->dev = np->dev;
460
461 netpoll_send_skb(np, skb);
462}
463EXPORT_SYMBOL(netpoll_send_udp);
464
465void netpoll_print_options(struct netpoll *np)
466{
467 np_info(np, "local port %d\n", np->local_port);
468 if (np->ipv6)
469 np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
470 else
471 np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
472 np_info(np, "interface '%s'\n", np->dev_name);
473 np_info(np, "remote port %d\n", np->remote_port);
474 if (np->ipv6)
475 np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
476 else
477 np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
478 np_info(np, "remote ethernet address %pM\n", np->remote_mac);
479}
480EXPORT_SYMBOL(netpoll_print_options);
481
482static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
483{
484 const char *end;
485
486 if (!strchr(str, ':') &&
487 in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
488 if (!*end)
489 return 0;
490 }
491 if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
492#if IS_ENABLED(CONFIG_IPV6)
493 if (!*end)
494 return 1;
495#else
496 return -1;
497#endif
498 }
499 return -1;
500}
501
502int netpoll_parse_options(struct netpoll *np, char *opt)
503{
504 char *cur=opt, *delim;
505 int ipv6;
506 bool ipversion_set = false;
507
508 if (*cur != '@') {
509 if ((delim = strchr(cur, '@')) == NULL)
510 goto parse_failed;
511 *delim = 0;
512 if (kstrtou16(cur, 10, &np->local_port))
513 goto parse_failed;
514 cur = delim;
515 }
516 cur++;
517
518 if (*cur != '/') {
519 ipversion_set = true;
520 if ((delim = strchr(cur, '/')) == NULL)
521 goto parse_failed;
522 *delim = 0;
523 ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
524 if (ipv6 < 0)
525 goto parse_failed;
526 else
527 np->ipv6 = (bool)ipv6;
528 cur = delim;
529 }
530 cur++;
531
532 if (*cur != ',') {
533 /* parse out dev name */
534 if ((delim = strchr(cur, ',')) == NULL)
535 goto parse_failed;
536 *delim = 0;
537 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
538 cur = delim;
539 }
540 cur++;
541
542 if (*cur != '@') {
543 /* dst port */
544 if ((delim = strchr(cur, '@')) == NULL)
545 goto parse_failed;
546 *delim = 0;
547 if (*cur == ' ' || *cur == '\t')
548 np_info(np, "warning: whitespace is not allowed\n");
549 if (kstrtou16(cur, 10, &np->remote_port))
550 goto parse_failed;
551 cur = delim;
552 }
553 cur++;
554
555 /* dst ip */
556 if ((delim = strchr(cur, '/')) == NULL)
557 goto parse_failed;
558 *delim = 0;
559 ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
560 if (ipv6 < 0)
561 goto parse_failed;
562 else if (ipversion_set && np->ipv6 != (bool)ipv6)
563 goto parse_failed;
564 else
565 np->ipv6 = (bool)ipv6;
566 cur = delim + 1;
567
568 if (*cur != 0) {
569 /* MAC address */
570 if (!mac_pton(cur, np->remote_mac))
571 goto parse_failed;
572 }
573
574 netpoll_print_options(np);
575
576 return 0;
577
578 parse_failed:
579 np_info(np, "couldn't parse config at '%s'!\n", cur);
580 return -1;
581}
582EXPORT_SYMBOL(netpoll_parse_options);
583
584int __netpoll_setup(struct netpoll *np, struct net_device *ndev)
585{
586 struct netpoll_info *npinfo;
587 const struct net_device_ops *ops;
588 int err;
589
590 np->dev = ndev;
591 strlcpy(np->dev_name, ndev->name, IFNAMSIZ);
592
593 if (ndev->priv_flags & IFF_DISABLE_NETPOLL) {
594 np_err(np, "%s doesn't support polling, aborting\n",
595 np->dev_name);
596 err = -ENOTSUPP;
597 goto out;
598 }
599
600 if (!ndev->npinfo) {
601 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
602 if (!npinfo) {
603 err = -ENOMEM;
604 goto out;
605 }
606
607 sema_init(&npinfo->dev_lock, 1);
608 skb_queue_head_init(&npinfo->txq);
609 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
610
611 refcount_set(&npinfo->refcnt, 1);
612
613 ops = np->dev->netdev_ops;
614 if (ops->ndo_netpoll_setup) {
615 err = ops->ndo_netpoll_setup(ndev, npinfo);
616 if (err)
617 goto free_npinfo;
618 }
619 } else {
620 npinfo = rtnl_dereference(ndev->npinfo);
621 refcount_inc(&npinfo->refcnt);
622 }
623
624 npinfo->netpoll = np;
625
626 /* last thing to do is link it to the net device structure */
627 rcu_assign_pointer(ndev->npinfo, npinfo);
628
629 return 0;
630
631free_npinfo:
632 kfree(npinfo);
633out:
634 return err;
635}
636EXPORT_SYMBOL_GPL(__netpoll_setup);
637
638int netpoll_setup(struct netpoll *np)
639{
640 struct net_device *ndev = NULL;
641 struct in_device *in_dev;
642 int err;
643
644 rtnl_lock();
645 if (np->dev_name[0]) {
646 struct net *net = current->nsproxy->net_ns;
647 ndev = __dev_get_by_name(net, np->dev_name);
648 }
649 if (!ndev) {
650 np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
651 err = -ENODEV;
652 goto unlock;
653 }
654 dev_hold(ndev);
655
656 if (netdev_master_upper_dev_get(ndev)) {
657 np_err(np, "%s is a slave device, aborting\n", np->dev_name);
658 err = -EBUSY;
659 goto put;
660 }
661
662 if (!netif_running(ndev)) {
663 unsigned long atmost, atleast;
664
665 np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
666
667 err = dev_open(ndev, NULL);
668
669 if (err) {
670 np_err(np, "failed to open %s\n", ndev->name);
671 goto put;
672 }
673
674 rtnl_unlock();
675 atleast = jiffies + HZ/10;
676 atmost = jiffies + carrier_timeout * HZ;
677 while (!netif_carrier_ok(ndev)) {
678 if (time_after(jiffies, atmost)) {
679 np_notice(np, "timeout waiting for carrier\n");
680 break;
681 }
682 msleep(1);
683 }
684
685 /* If carrier appears to come up instantly, we don't
686 * trust it and pause so that we don't pump all our
687 * queued console messages into the bitbucket.
688 */
689
690 if (time_before(jiffies, atleast)) {
691 np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
692 msleep(4000);
693 }
694 rtnl_lock();
695 }
696
697 if (!np->local_ip.ip) {
698 if (!np->ipv6) {
699 const struct in_ifaddr *ifa;
700
701 in_dev = __in_dev_get_rtnl(ndev);
702 if (!in_dev)
703 goto put_noaddr;
704
705 ifa = rtnl_dereference(in_dev->ifa_list);
706 if (!ifa) {
707put_noaddr:
708 np_err(np, "no IP address for %s, aborting\n",
709 np->dev_name);
710 err = -EDESTADDRREQ;
711 goto put;
712 }
713
714 np->local_ip.ip = ifa->ifa_local;
715 np_info(np, "local IP %pI4\n", &np->local_ip.ip);
716 } else {
717#if IS_ENABLED(CONFIG_IPV6)
718 struct inet6_dev *idev;
719
720 err = -EDESTADDRREQ;
721 idev = __in6_dev_get(ndev);
722 if (idev) {
723 struct inet6_ifaddr *ifp;
724
725 read_lock_bh(&idev->lock);
726 list_for_each_entry(ifp, &idev->addr_list, if_list) {
727 if (!!(ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL) !=
728 !!(ipv6_addr_type(&np->remote_ip.in6) & IPV6_ADDR_LINKLOCAL))
729 continue;
730 np->local_ip.in6 = ifp->addr;
731 err = 0;
732 break;
733 }
734 read_unlock_bh(&idev->lock);
735 }
736 if (err) {
737 np_err(np, "no IPv6 address for %s, aborting\n",
738 np->dev_name);
739 goto put;
740 } else
741 np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
742#else
743 np_err(np, "IPv6 is not supported %s, aborting\n",
744 np->dev_name);
745 err = -EINVAL;
746 goto put;
747#endif
748 }
749 }
750
751 /* fill up the skb queue */
752 refill_skbs();
753
754 err = __netpoll_setup(np, ndev);
755 if (err)
756 goto put;
757
758 rtnl_unlock();
759 return 0;
760
761put:
762 dev_put(ndev);
763unlock:
764 rtnl_unlock();
765 return err;
766}
767EXPORT_SYMBOL(netpoll_setup);
768
769static int __init netpoll_init(void)
770{
771 skb_queue_head_init(&skb_pool);
772 return 0;
773}
774core_initcall(netpoll_init);
775
776static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
777{
778 struct netpoll_info *npinfo =
779 container_of(rcu_head, struct netpoll_info, rcu);
780
781 skb_queue_purge(&npinfo->txq);
782
783 /* we can't call cancel_delayed_work_sync here, as we are in softirq */
784 cancel_delayed_work(&npinfo->tx_work);
785
786 /* clean after last, unfinished work */
787 __skb_queue_purge(&npinfo->txq);
788 /* now cancel it again */
789 cancel_delayed_work(&npinfo->tx_work);
790 kfree(npinfo);
791}
792
793void __netpoll_cleanup(struct netpoll *np)
794{
795 struct netpoll_info *npinfo;
796
797 npinfo = rtnl_dereference(np->dev->npinfo);
798 if (!npinfo)
799 return;
800
801 synchronize_srcu(&netpoll_srcu);
802
803 if (refcount_dec_and_test(&npinfo->refcnt)) {
804 const struct net_device_ops *ops;
805
806 ops = np->dev->netdev_ops;
807 if (ops->ndo_netpoll_cleanup)
808 ops->ndo_netpoll_cleanup(np->dev);
809
810 RCU_INIT_POINTER(np->dev->npinfo, NULL);
811 call_rcu(&npinfo->rcu, rcu_cleanup_netpoll_info);
812 } else
813 RCU_INIT_POINTER(np->dev->npinfo, NULL);
814}
815EXPORT_SYMBOL_GPL(__netpoll_cleanup);
816
817void __netpoll_free(struct netpoll *np)
818{
819 ASSERT_RTNL();
820
821 /* Wait for transmitting packets to finish before freeing. */
822 synchronize_rcu();
823 __netpoll_cleanup(np);
824 kfree(np);
825}
826EXPORT_SYMBOL_GPL(__netpoll_free);
827
828void netpoll_cleanup(struct netpoll *np)
829{
830 rtnl_lock();
831 if (!np->dev)
832 goto out;
833 __netpoll_cleanup(np);
834 dev_put(np->dev);
835 np->dev = NULL;
836out:
837 rtnl_unlock();
838}
839EXPORT_SYMBOL(netpoll_cleanup);