Loading...
1/*
2 * Common framework for low-level network console, dump, and debugger code
3 *
4 * Sep 8 2003 Matt Mackall <mpm@selenic.com>
5 *
6 * based on the netconsole code from:
7 *
8 * Copyright (C) 2001 Ingo Molnar <mingo@redhat.com>
9 * Copyright (C) 2002 Red Hat, Inc.
10 */
11
12#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13
14#include <linux/moduleparam.h>
15#include <linux/kernel.h>
16#include <linux/netdevice.h>
17#include <linux/etherdevice.h>
18#include <linux/string.h>
19#include <linux/if_arp.h>
20#include <linux/inetdevice.h>
21#include <linux/inet.h>
22#include <linux/interrupt.h>
23#include <linux/netpoll.h>
24#include <linux/sched.h>
25#include <linux/delay.h>
26#include <linux/rcupdate.h>
27#include <linux/workqueue.h>
28#include <linux/slab.h>
29#include <linux/export.h>
30#include <linux/if_vlan.h>
31#include <net/tcp.h>
32#include <net/udp.h>
33#include <net/addrconf.h>
34#include <net/ndisc.h>
35#include <net/ip6_checksum.h>
36#include <asm/unaligned.h>
37#include <trace/events/napi.h>
38
39/*
40 * We maintain a small pool of fully-sized skbs, to make sure the
41 * message gets out even in extreme OOM situations.
42 */
43
44#define MAX_UDP_CHUNK 1460
45#define MAX_SKBS 32
46
47static struct sk_buff_head skb_pool;
48
49DEFINE_STATIC_SRCU(netpoll_srcu);
50
51#define USEC_PER_POLL 50
52
53#define MAX_SKB_SIZE \
54 (sizeof(struct ethhdr) + \
55 sizeof(struct iphdr) + \
56 sizeof(struct udphdr) + \
57 MAX_UDP_CHUNK)
58
59static void zap_completion_queue(void);
60static void netpoll_async_cleanup(struct work_struct *work);
61
62static unsigned int carrier_timeout = 4;
63module_param(carrier_timeout, uint, 0644);
64
65#define np_info(np, fmt, ...) \
66 pr_info("%s: " fmt, np->name, ##__VA_ARGS__)
67#define np_err(np, fmt, ...) \
68 pr_err("%s: " fmt, np->name, ##__VA_ARGS__)
69#define np_notice(np, fmt, ...) \
70 pr_notice("%s: " fmt, np->name, ##__VA_ARGS__)
71
72static int netpoll_start_xmit(struct sk_buff *skb, struct net_device *dev,
73 struct netdev_queue *txq)
74{
75 int status = NETDEV_TX_OK;
76 netdev_features_t features;
77
78 features = netif_skb_features(skb);
79
80 if (skb_vlan_tag_present(skb) &&
81 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
82 skb = __vlan_hwaccel_push_inside(skb);
83 if (unlikely(!skb)) {
84 /* This is actually a packet drop, but we
85 * don't want the code that calls this
86 * function to try and operate on a NULL skb.
87 */
88 goto out;
89 }
90 }
91
92 status = netdev_start_xmit(skb, dev, txq, false);
93
94out:
95 return status;
96}
97
98static void queue_process(struct work_struct *work)
99{
100 struct netpoll_info *npinfo =
101 container_of(work, struct netpoll_info, tx_work.work);
102 struct sk_buff *skb;
103 unsigned long flags;
104
105 while ((skb = skb_dequeue(&npinfo->txq))) {
106 struct net_device *dev = skb->dev;
107 struct netdev_queue *txq;
108
109 if (!netif_device_present(dev) || !netif_running(dev)) {
110 kfree_skb(skb);
111 continue;
112 }
113
114 txq = skb_get_tx_queue(dev, skb);
115
116 local_irq_save(flags);
117 HARD_TX_LOCK(dev, txq, smp_processor_id());
118 if (netif_xmit_frozen_or_stopped(txq) ||
119 netpoll_start_xmit(skb, dev, txq) != NETDEV_TX_OK) {
120 skb_queue_head(&npinfo->txq, skb);
121 HARD_TX_UNLOCK(dev, txq);
122 local_irq_restore(flags);
123
124 schedule_delayed_work(&npinfo->tx_work, HZ/10);
125 return;
126 }
127 HARD_TX_UNLOCK(dev, txq);
128 local_irq_restore(flags);
129 }
130}
131
132/*
133 * Check whether delayed processing was scheduled for our NIC. If so,
134 * we attempt to grab the poll lock and use ->poll() to pump the card.
135 * If this fails, either we've recursed in ->poll() or it's already
136 * running on another CPU.
137 *
138 * Note: we don't mask interrupts with this lock because we're using
139 * trylock here and interrupts are already disabled in the softirq
140 * case. Further, we test the poll_owner to avoid recursion on UP
141 * systems where the lock doesn't exist.
142 */
143static void poll_one_napi(struct napi_struct *napi)
144{
145 int work = 0;
146
147 /* net_rx_action's ->poll() invocations and our's are
148 * synchronized by this test which is only made while
149 * holding the napi->poll_lock.
150 */
151 if (!test_bit(NAPI_STATE_SCHED, &napi->state))
152 return;
153
154 /* If we set this bit but see that it has already been set,
155 * that indicates that napi has been disabled and we need
156 * to abort this operation
157 */
158 if (test_and_set_bit(NAPI_STATE_NPSVC, &napi->state))
159 return;
160
161 /* We explicilty pass the polling call a budget of 0 to
162 * indicate that we are clearing the Tx path only.
163 */
164 work = napi->poll(napi, 0);
165 WARN_ONCE(work, "%pF exceeded budget in poll\n", napi->poll);
166 trace_napi_poll(napi);
167
168 clear_bit(NAPI_STATE_NPSVC, &napi->state);
169}
170
171static void poll_napi(struct net_device *dev)
172{
173 struct napi_struct *napi;
174
175 list_for_each_entry(napi, &dev->napi_list, dev_list) {
176 if (napi->poll_owner != smp_processor_id() &&
177 spin_trylock(&napi->poll_lock)) {
178 poll_one_napi(napi);
179 spin_unlock(&napi->poll_lock);
180 }
181 }
182}
183
184static void netpoll_poll_dev(struct net_device *dev)
185{
186 const struct net_device_ops *ops;
187 struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
188
189 /* Don't do any rx activity if the dev_lock mutex is held
190 * the dev_open/close paths use this to block netpoll activity
191 * while changing device state
192 */
193 if (down_trylock(&ni->dev_lock))
194 return;
195
196 if (!netif_running(dev)) {
197 up(&ni->dev_lock);
198 return;
199 }
200
201 ops = dev->netdev_ops;
202 if (!ops->ndo_poll_controller) {
203 up(&ni->dev_lock);
204 return;
205 }
206
207 /* Process pending work on NIC */
208 ops->ndo_poll_controller(dev);
209
210 poll_napi(dev);
211
212 up(&ni->dev_lock);
213
214 zap_completion_queue();
215}
216
217void netpoll_poll_disable(struct net_device *dev)
218{
219 struct netpoll_info *ni;
220 int idx;
221 might_sleep();
222 idx = srcu_read_lock(&netpoll_srcu);
223 ni = srcu_dereference(dev->npinfo, &netpoll_srcu);
224 if (ni)
225 down(&ni->dev_lock);
226 srcu_read_unlock(&netpoll_srcu, idx);
227}
228EXPORT_SYMBOL(netpoll_poll_disable);
229
230void netpoll_poll_enable(struct net_device *dev)
231{
232 struct netpoll_info *ni;
233 rcu_read_lock();
234 ni = rcu_dereference(dev->npinfo);
235 if (ni)
236 up(&ni->dev_lock);
237 rcu_read_unlock();
238}
239EXPORT_SYMBOL(netpoll_poll_enable);
240
241static void refill_skbs(void)
242{
243 struct sk_buff *skb;
244 unsigned long flags;
245
246 spin_lock_irqsave(&skb_pool.lock, flags);
247 while (skb_pool.qlen < MAX_SKBS) {
248 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
249 if (!skb)
250 break;
251
252 __skb_queue_tail(&skb_pool, skb);
253 }
254 spin_unlock_irqrestore(&skb_pool.lock, flags);
255}
256
257static void zap_completion_queue(void)
258{
259 unsigned long flags;
260 struct softnet_data *sd = &get_cpu_var(softnet_data);
261
262 if (sd->completion_queue) {
263 struct sk_buff *clist;
264
265 local_irq_save(flags);
266 clist = sd->completion_queue;
267 sd->completion_queue = NULL;
268 local_irq_restore(flags);
269
270 while (clist != NULL) {
271 struct sk_buff *skb = clist;
272 clist = clist->next;
273 if (!skb_irq_freeable(skb)) {
274 atomic_inc(&skb->users);
275 dev_kfree_skb_any(skb); /* put this one back */
276 } else {
277 __kfree_skb(skb);
278 }
279 }
280 }
281
282 put_cpu_var(softnet_data);
283}
284
285static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
286{
287 int count = 0;
288 struct sk_buff *skb;
289
290 zap_completion_queue();
291 refill_skbs();
292repeat:
293
294 skb = alloc_skb(len, GFP_ATOMIC);
295 if (!skb)
296 skb = skb_dequeue(&skb_pool);
297
298 if (!skb) {
299 if (++count < 10) {
300 netpoll_poll_dev(np->dev);
301 goto repeat;
302 }
303 return NULL;
304 }
305
306 atomic_set(&skb->users, 1);
307 skb_reserve(skb, reserve);
308 return skb;
309}
310
311static int netpoll_owner_active(struct net_device *dev)
312{
313 struct napi_struct *napi;
314
315 list_for_each_entry(napi, &dev->napi_list, dev_list) {
316 if (napi->poll_owner == smp_processor_id())
317 return 1;
318 }
319 return 0;
320}
321
322/* call with IRQ disabled */
323void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb,
324 struct net_device *dev)
325{
326 int status = NETDEV_TX_BUSY;
327 unsigned long tries;
328 /* It is up to the caller to keep npinfo alive. */
329 struct netpoll_info *npinfo;
330
331 WARN_ON_ONCE(!irqs_disabled());
332
333 npinfo = rcu_dereference_bh(np->dev->npinfo);
334 if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
335 dev_kfree_skb_irq(skb);
336 return;
337 }
338
339 /* don't get messages out of order, and no recursion */
340 if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
341 struct netdev_queue *txq;
342
343 txq = netdev_pick_tx(dev, skb, NULL);
344
345 /* try until next clock tick */
346 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
347 tries > 0; --tries) {
348 if (HARD_TX_TRYLOCK(dev, txq)) {
349 if (!netif_xmit_stopped(txq))
350 status = netpoll_start_xmit(skb, dev, txq);
351
352 HARD_TX_UNLOCK(dev, txq);
353
354 if (status == NETDEV_TX_OK)
355 break;
356
357 }
358
359 /* tickle device maybe there is some cleanup */
360 netpoll_poll_dev(np->dev);
361
362 udelay(USEC_PER_POLL);
363 }
364
365 WARN_ONCE(!irqs_disabled(),
366 "netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pF)\n",
367 dev->name, dev->netdev_ops->ndo_start_xmit);
368
369 }
370
371 if (status != NETDEV_TX_OK) {
372 skb_queue_tail(&npinfo->txq, skb);
373 schedule_delayed_work(&npinfo->tx_work,0);
374 }
375}
376EXPORT_SYMBOL(netpoll_send_skb_on_dev);
377
378void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
379{
380 int total_len, ip_len, udp_len;
381 struct sk_buff *skb;
382 struct udphdr *udph;
383 struct iphdr *iph;
384 struct ethhdr *eth;
385 static atomic_t ip_ident;
386 struct ipv6hdr *ip6h;
387
388 WARN_ON_ONCE(!irqs_disabled());
389
390 udp_len = len + sizeof(*udph);
391 if (np->ipv6)
392 ip_len = udp_len + sizeof(*ip6h);
393 else
394 ip_len = udp_len + sizeof(*iph);
395
396 total_len = ip_len + LL_RESERVED_SPACE(np->dev);
397
398 skb = find_skb(np, total_len + np->dev->needed_tailroom,
399 total_len - len);
400 if (!skb)
401 return;
402
403 skb_copy_to_linear_data(skb, msg, len);
404 skb_put(skb, len);
405
406 skb_push(skb, sizeof(*udph));
407 skb_reset_transport_header(skb);
408 udph = udp_hdr(skb);
409 udph->source = htons(np->local_port);
410 udph->dest = htons(np->remote_port);
411 udph->len = htons(udp_len);
412
413 if (np->ipv6) {
414 udph->check = 0;
415 udph->check = csum_ipv6_magic(&np->local_ip.in6,
416 &np->remote_ip.in6,
417 udp_len, IPPROTO_UDP,
418 csum_partial(udph, udp_len, 0));
419 if (udph->check == 0)
420 udph->check = CSUM_MANGLED_0;
421
422 skb_push(skb, sizeof(*ip6h));
423 skb_reset_network_header(skb);
424 ip6h = ipv6_hdr(skb);
425
426 /* ip6h->version = 6; ip6h->priority = 0; */
427 put_unaligned(0x60, (unsigned char *)ip6h);
428 ip6h->flow_lbl[0] = 0;
429 ip6h->flow_lbl[1] = 0;
430 ip6h->flow_lbl[2] = 0;
431
432 ip6h->payload_len = htons(sizeof(struct udphdr) + len);
433 ip6h->nexthdr = IPPROTO_UDP;
434 ip6h->hop_limit = 32;
435 ip6h->saddr = np->local_ip.in6;
436 ip6h->daddr = np->remote_ip.in6;
437
438 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
439 skb_reset_mac_header(skb);
440 skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
441 } else {
442 udph->check = 0;
443 udph->check = csum_tcpudp_magic(np->local_ip.ip,
444 np->remote_ip.ip,
445 udp_len, IPPROTO_UDP,
446 csum_partial(udph, udp_len, 0));
447 if (udph->check == 0)
448 udph->check = CSUM_MANGLED_0;
449
450 skb_push(skb, sizeof(*iph));
451 skb_reset_network_header(skb);
452 iph = ip_hdr(skb);
453
454 /* iph->version = 4; iph->ihl = 5; */
455 put_unaligned(0x45, (unsigned char *)iph);
456 iph->tos = 0;
457 put_unaligned(htons(ip_len), &(iph->tot_len));
458 iph->id = htons(atomic_inc_return(&ip_ident));
459 iph->frag_off = 0;
460 iph->ttl = 64;
461 iph->protocol = IPPROTO_UDP;
462 iph->check = 0;
463 put_unaligned(np->local_ip.ip, &(iph->saddr));
464 put_unaligned(np->remote_ip.ip, &(iph->daddr));
465 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
466
467 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
468 skb_reset_mac_header(skb);
469 skb->protocol = eth->h_proto = htons(ETH_P_IP);
470 }
471
472 ether_addr_copy(eth->h_source, np->dev->dev_addr);
473 ether_addr_copy(eth->h_dest, np->remote_mac);
474
475 skb->dev = np->dev;
476
477 netpoll_send_skb(np, skb);
478}
479EXPORT_SYMBOL(netpoll_send_udp);
480
481void netpoll_print_options(struct netpoll *np)
482{
483 np_info(np, "local port %d\n", np->local_port);
484 if (np->ipv6)
485 np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
486 else
487 np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
488 np_info(np, "interface '%s'\n", np->dev_name);
489 np_info(np, "remote port %d\n", np->remote_port);
490 if (np->ipv6)
491 np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
492 else
493 np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
494 np_info(np, "remote ethernet address %pM\n", np->remote_mac);
495}
496EXPORT_SYMBOL(netpoll_print_options);
497
498static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
499{
500 const char *end;
501
502 if (!strchr(str, ':') &&
503 in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
504 if (!*end)
505 return 0;
506 }
507 if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
508#if IS_ENABLED(CONFIG_IPV6)
509 if (!*end)
510 return 1;
511#else
512 return -1;
513#endif
514 }
515 return -1;
516}
517
518int netpoll_parse_options(struct netpoll *np, char *opt)
519{
520 char *cur=opt, *delim;
521 int ipv6;
522 bool ipversion_set = false;
523
524 if (*cur != '@') {
525 if ((delim = strchr(cur, '@')) == NULL)
526 goto parse_failed;
527 *delim = 0;
528 if (kstrtou16(cur, 10, &np->local_port))
529 goto parse_failed;
530 cur = delim;
531 }
532 cur++;
533
534 if (*cur != '/') {
535 ipversion_set = true;
536 if ((delim = strchr(cur, '/')) == NULL)
537 goto parse_failed;
538 *delim = 0;
539 ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
540 if (ipv6 < 0)
541 goto parse_failed;
542 else
543 np->ipv6 = (bool)ipv6;
544 cur = delim;
545 }
546 cur++;
547
548 if (*cur != ',') {
549 /* parse out dev name */
550 if ((delim = strchr(cur, ',')) == NULL)
551 goto parse_failed;
552 *delim = 0;
553 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
554 cur = delim;
555 }
556 cur++;
557
558 if (*cur != '@') {
559 /* dst port */
560 if ((delim = strchr(cur, '@')) == NULL)
561 goto parse_failed;
562 *delim = 0;
563 if (*cur == ' ' || *cur == '\t')
564 np_info(np, "warning: whitespace is not allowed\n");
565 if (kstrtou16(cur, 10, &np->remote_port))
566 goto parse_failed;
567 cur = delim;
568 }
569 cur++;
570
571 /* dst ip */
572 if ((delim = strchr(cur, '/')) == NULL)
573 goto parse_failed;
574 *delim = 0;
575 ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
576 if (ipv6 < 0)
577 goto parse_failed;
578 else if (ipversion_set && np->ipv6 != (bool)ipv6)
579 goto parse_failed;
580 else
581 np->ipv6 = (bool)ipv6;
582 cur = delim + 1;
583
584 if (*cur != 0) {
585 /* MAC address */
586 if (!mac_pton(cur, np->remote_mac))
587 goto parse_failed;
588 }
589
590 netpoll_print_options(np);
591
592 return 0;
593
594 parse_failed:
595 np_info(np, "couldn't parse config at '%s'!\n", cur);
596 return -1;
597}
598EXPORT_SYMBOL(netpoll_parse_options);
599
600int __netpoll_setup(struct netpoll *np, struct net_device *ndev)
601{
602 struct netpoll_info *npinfo;
603 const struct net_device_ops *ops;
604 int err;
605
606 np->dev = ndev;
607 strlcpy(np->dev_name, ndev->name, IFNAMSIZ);
608 INIT_WORK(&np->cleanup_work, netpoll_async_cleanup);
609
610 if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) ||
611 !ndev->netdev_ops->ndo_poll_controller) {
612 np_err(np, "%s doesn't support polling, aborting\n",
613 np->dev_name);
614 err = -ENOTSUPP;
615 goto out;
616 }
617
618 if (!ndev->npinfo) {
619 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
620 if (!npinfo) {
621 err = -ENOMEM;
622 goto out;
623 }
624
625 sema_init(&npinfo->dev_lock, 1);
626 skb_queue_head_init(&npinfo->txq);
627 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
628
629 atomic_set(&npinfo->refcnt, 1);
630
631 ops = np->dev->netdev_ops;
632 if (ops->ndo_netpoll_setup) {
633 err = ops->ndo_netpoll_setup(ndev, npinfo);
634 if (err)
635 goto free_npinfo;
636 }
637 } else {
638 npinfo = rtnl_dereference(ndev->npinfo);
639 atomic_inc(&npinfo->refcnt);
640 }
641
642 npinfo->netpoll = np;
643
644 /* last thing to do is link it to the net device structure */
645 rcu_assign_pointer(ndev->npinfo, npinfo);
646
647 return 0;
648
649free_npinfo:
650 kfree(npinfo);
651out:
652 return err;
653}
654EXPORT_SYMBOL_GPL(__netpoll_setup);
655
656int netpoll_setup(struct netpoll *np)
657{
658 struct net_device *ndev = NULL;
659 struct in_device *in_dev;
660 int err;
661
662 rtnl_lock();
663 if (np->dev_name) {
664 struct net *net = current->nsproxy->net_ns;
665 ndev = __dev_get_by_name(net, np->dev_name);
666 }
667 if (!ndev) {
668 np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
669 err = -ENODEV;
670 goto unlock;
671 }
672 dev_hold(ndev);
673
674 if (netdev_master_upper_dev_get(ndev)) {
675 np_err(np, "%s is a slave device, aborting\n", np->dev_name);
676 err = -EBUSY;
677 goto put;
678 }
679
680 if (!netif_running(ndev)) {
681 unsigned long atmost, atleast;
682
683 np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
684
685 err = dev_open(ndev);
686
687 if (err) {
688 np_err(np, "failed to open %s\n", ndev->name);
689 goto put;
690 }
691
692 rtnl_unlock();
693 atleast = jiffies + HZ/10;
694 atmost = jiffies + carrier_timeout * HZ;
695 while (!netif_carrier_ok(ndev)) {
696 if (time_after(jiffies, atmost)) {
697 np_notice(np, "timeout waiting for carrier\n");
698 break;
699 }
700 msleep(1);
701 }
702
703 /* If carrier appears to come up instantly, we don't
704 * trust it and pause so that we don't pump all our
705 * queued console messages into the bitbucket.
706 */
707
708 if (time_before(jiffies, atleast)) {
709 np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
710 msleep(4000);
711 }
712 rtnl_lock();
713 }
714
715 if (!np->local_ip.ip) {
716 if (!np->ipv6) {
717 in_dev = __in_dev_get_rtnl(ndev);
718
719 if (!in_dev || !in_dev->ifa_list) {
720 np_err(np, "no IP address for %s, aborting\n",
721 np->dev_name);
722 err = -EDESTADDRREQ;
723 goto put;
724 }
725
726 np->local_ip.ip = in_dev->ifa_list->ifa_local;
727 np_info(np, "local IP %pI4\n", &np->local_ip.ip);
728 } else {
729#if IS_ENABLED(CONFIG_IPV6)
730 struct inet6_dev *idev;
731
732 err = -EDESTADDRREQ;
733 idev = __in6_dev_get(ndev);
734 if (idev) {
735 struct inet6_ifaddr *ifp;
736
737 read_lock_bh(&idev->lock);
738 list_for_each_entry(ifp, &idev->addr_list, if_list) {
739 if (ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL)
740 continue;
741 np->local_ip.in6 = ifp->addr;
742 err = 0;
743 break;
744 }
745 read_unlock_bh(&idev->lock);
746 }
747 if (err) {
748 np_err(np, "no IPv6 address for %s, aborting\n",
749 np->dev_name);
750 goto put;
751 } else
752 np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
753#else
754 np_err(np, "IPv6 is not supported %s, aborting\n",
755 np->dev_name);
756 err = -EINVAL;
757 goto put;
758#endif
759 }
760 }
761
762 /* fill up the skb queue */
763 refill_skbs();
764
765 err = __netpoll_setup(np, ndev);
766 if (err)
767 goto put;
768
769 rtnl_unlock();
770 return 0;
771
772put:
773 dev_put(ndev);
774unlock:
775 rtnl_unlock();
776 return err;
777}
778EXPORT_SYMBOL(netpoll_setup);
779
780static int __init netpoll_init(void)
781{
782 skb_queue_head_init(&skb_pool);
783 return 0;
784}
785core_initcall(netpoll_init);
786
787static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
788{
789 struct netpoll_info *npinfo =
790 container_of(rcu_head, struct netpoll_info, rcu);
791
792 skb_queue_purge(&npinfo->txq);
793
794 /* we can't call cancel_delayed_work_sync here, as we are in softirq */
795 cancel_delayed_work(&npinfo->tx_work);
796
797 /* clean after last, unfinished work */
798 __skb_queue_purge(&npinfo->txq);
799 /* now cancel it again */
800 cancel_delayed_work(&npinfo->tx_work);
801 kfree(npinfo);
802}
803
804void __netpoll_cleanup(struct netpoll *np)
805{
806 struct netpoll_info *npinfo;
807
808 /* rtnl_dereference would be preferable here but
809 * rcu_cleanup_netpoll path can put us in here safely without
810 * holding the rtnl, so plain rcu_dereference it is
811 */
812 npinfo = rtnl_dereference(np->dev->npinfo);
813 if (!npinfo)
814 return;
815
816 synchronize_srcu(&netpoll_srcu);
817
818 if (atomic_dec_and_test(&npinfo->refcnt)) {
819 const struct net_device_ops *ops;
820
821 ops = np->dev->netdev_ops;
822 if (ops->ndo_netpoll_cleanup)
823 ops->ndo_netpoll_cleanup(np->dev);
824
825 RCU_INIT_POINTER(np->dev->npinfo, NULL);
826 call_rcu_bh(&npinfo->rcu, rcu_cleanup_netpoll_info);
827 } else
828 RCU_INIT_POINTER(np->dev->npinfo, NULL);
829}
830EXPORT_SYMBOL_GPL(__netpoll_cleanup);
831
832static void netpoll_async_cleanup(struct work_struct *work)
833{
834 struct netpoll *np = container_of(work, struct netpoll, cleanup_work);
835
836 rtnl_lock();
837 __netpoll_cleanup(np);
838 rtnl_unlock();
839 kfree(np);
840}
841
842void __netpoll_free_async(struct netpoll *np)
843{
844 schedule_work(&np->cleanup_work);
845}
846EXPORT_SYMBOL_GPL(__netpoll_free_async);
847
848void netpoll_cleanup(struct netpoll *np)
849{
850 rtnl_lock();
851 if (!np->dev)
852 goto out;
853 __netpoll_cleanup(np);
854 dev_put(np->dev);
855 np->dev = NULL;
856out:
857 rtnl_unlock();
858}
859EXPORT_SYMBOL(netpoll_cleanup);
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);