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/*
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);