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
56 static void zap_completion_queue(void);
57
58 static unsigned int carrier_timeout = 4;
59 module_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
netpoll_start_xmit(struct sk_buff * skb,struct net_device * dev,struct netdev_queue * txq)68 static 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
91 out:
92 return status;
93 }
94
queue_process(struct work_struct * work)95 static 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
netif_local_xmit_active(struct net_device * dev)135 static 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
poll_one_napi(struct napi_struct * napi)149 static 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
poll_napi(struct net_device * dev)170 static 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
netpoll_poll_dev(struct net_device * dev)183 void 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 }
213 EXPORT_SYMBOL(netpoll_poll_dev);
214
netpoll_poll_disable(struct net_device * dev)215 void 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
netpoll_poll_enable(struct net_device * dev)225 void 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
refill_skbs(struct netpoll * np)234 static 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
zap_completion_queue(void)253 static 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
find_skb(struct netpoll * np,int len,int reserve)281 static 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 repeat:
288
289 skb = alloc_skb(len, GFP_ATOMIC);
290 if (!skb) {
291 skb = skb_dequeue(&np->skb_pool);
292 schedule_work(&np->refill_wq);
293 }
294
295 if (!skb) {
296 if (++count < 10) {
297 netpoll_poll_dev(np->dev);
298 goto repeat;
299 }
300 return NULL;
301 }
302
303 refcount_set(&skb->users, 1);
304 skb_reserve(skb, reserve);
305 return skb;
306 }
307
netpoll_owner_active(struct net_device * dev)308 static int netpoll_owner_active(struct net_device *dev)
309 {
310 struct napi_struct *napi;
311
312 list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) {
313 if (READ_ONCE(napi->poll_owner) == smp_processor_id())
314 return 1;
315 }
316 return 0;
317 }
318
319 /* call with IRQ disabled */
__netpoll_send_skb(struct netpoll * np,struct sk_buff * skb)320 static netdev_tx_t __netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
321 {
322 netdev_tx_t status = NETDEV_TX_BUSY;
323 netdev_tx_t ret = NET_XMIT_DROP;
324 struct net_device *dev;
325 unsigned long tries;
326 /* It is up to the caller to keep npinfo alive. */
327 struct netpoll_info *npinfo;
328
329 lockdep_assert_irqs_disabled();
330
331 dev = np->dev;
332 rcu_read_lock();
333 npinfo = rcu_dereference_bh(dev->npinfo);
334
335 if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
336 dev_kfree_skb_irq(skb);
337 goto out;
338 }
339
340 /* don't get messages out of order, and no recursion */
341 if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
342 struct netdev_queue *txq;
343
344 txq = netdev_core_pick_tx(dev, skb, NULL);
345
346 /* try until next clock tick */
347 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
348 tries > 0; --tries) {
349 if (HARD_TX_TRYLOCK(dev, txq)) {
350 if (!netif_xmit_stopped(txq))
351 status = netpoll_start_xmit(skb, dev, txq);
352
353 HARD_TX_UNLOCK(dev, txq);
354
355 if (dev_xmit_complete(status))
356 break;
357
358 }
359
360 /* tickle device maybe there is some cleanup */
361 netpoll_poll_dev(np->dev);
362
363 udelay(USEC_PER_POLL);
364 }
365
366 WARN_ONCE(!irqs_disabled(),
367 "netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pS)\n",
368 dev->name, dev->netdev_ops->ndo_start_xmit);
369
370 }
371
372 if (!dev_xmit_complete(status)) {
373 skb_queue_tail(&npinfo->txq, skb);
374 schedule_delayed_work(&npinfo->tx_work,0);
375 }
376 ret = NETDEV_TX_OK;
377 out:
378 rcu_read_unlock();
379 return ret;
380 }
381
netpoll_send_skb(struct netpoll * np,struct sk_buff * skb)382 netdev_tx_t netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
383 {
384 unsigned long flags;
385 netdev_tx_t ret;
386
387 if (unlikely(!np)) {
388 dev_kfree_skb_irq(skb);
389 ret = NET_XMIT_DROP;
390 } else {
391 local_irq_save(flags);
392 ret = __netpoll_send_skb(np, skb);
393 local_irq_restore(flags);
394 }
395 return ret;
396 }
397 EXPORT_SYMBOL(netpoll_send_skb);
398
netpoll_send_udp(struct netpoll * np,const char * msg,int len)399 int netpoll_send_udp(struct netpoll *np, const char *msg, int len)
400 {
401 int total_len, ip_len, udp_len;
402 struct sk_buff *skb;
403 struct udphdr *udph;
404 struct iphdr *iph;
405 struct ethhdr *eth;
406 static atomic_t ip_ident;
407 struct ipv6hdr *ip6h;
408
409 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
410 WARN_ON_ONCE(!irqs_disabled());
411
412 udp_len = len + sizeof(*udph);
413 if (np->ipv6)
414 ip_len = udp_len + sizeof(*ip6h);
415 else
416 ip_len = udp_len + sizeof(*iph);
417
418 total_len = ip_len + LL_RESERVED_SPACE(np->dev);
419
420 skb = find_skb(np, total_len + np->dev->needed_tailroom,
421 total_len - len);
422 if (!skb)
423 return -ENOMEM;
424
425 skb_copy_to_linear_data(skb, msg, len);
426 skb_put(skb, len);
427
428 skb_push(skb, sizeof(*udph));
429 skb_reset_transport_header(skb);
430 udph = udp_hdr(skb);
431 udph->source = htons(np->local_port);
432 udph->dest = htons(np->remote_port);
433 udph->len = htons(udp_len);
434
435 udph->check = 0;
436 if (np->ipv6) {
437 udph->check = csum_ipv6_magic(&np->local_ip.in6,
438 &np->remote_ip.in6,
439 udp_len, IPPROTO_UDP,
440 csum_partial(udph, udp_len, 0));
441 if (udph->check == 0)
442 udph->check = CSUM_MANGLED_0;
443
444 skb_push(skb, sizeof(*ip6h));
445 skb_reset_network_header(skb);
446 ip6h = ipv6_hdr(skb);
447
448 /* ip6h->version = 6; ip6h->priority = 0; */
449 *(unsigned char *)ip6h = 0x60;
450 ip6h->flow_lbl[0] = 0;
451 ip6h->flow_lbl[1] = 0;
452 ip6h->flow_lbl[2] = 0;
453
454 ip6h->payload_len = htons(sizeof(struct udphdr) + len);
455 ip6h->nexthdr = IPPROTO_UDP;
456 ip6h->hop_limit = 32;
457 ip6h->saddr = np->local_ip.in6;
458 ip6h->daddr = np->remote_ip.in6;
459
460 eth = skb_push(skb, ETH_HLEN);
461 skb_reset_mac_header(skb);
462 skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
463 } else {
464 udph->check = csum_tcpudp_magic(np->local_ip.ip,
465 np->remote_ip.ip,
466 udp_len, IPPROTO_UDP,
467 csum_partial(udph, udp_len, 0));
468 if (udph->check == 0)
469 udph->check = CSUM_MANGLED_0;
470
471 skb_push(skb, sizeof(*iph));
472 skb_reset_network_header(skb);
473 iph = ip_hdr(skb);
474
475 /* iph->version = 4; iph->ihl = 5; */
476 *(unsigned char *)iph = 0x45;
477 iph->tos = 0;
478 put_unaligned(htons(ip_len), &(iph->tot_len));
479 iph->id = htons(atomic_inc_return(&ip_ident));
480 iph->frag_off = 0;
481 iph->ttl = 64;
482 iph->protocol = IPPROTO_UDP;
483 iph->check = 0;
484 put_unaligned(np->local_ip.ip, &(iph->saddr));
485 put_unaligned(np->remote_ip.ip, &(iph->daddr));
486 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
487
488 eth = skb_push(skb, ETH_HLEN);
489 skb_reset_mac_header(skb);
490 skb->protocol = eth->h_proto = htons(ETH_P_IP);
491 }
492
493 ether_addr_copy(eth->h_source, np->dev->dev_addr);
494 ether_addr_copy(eth->h_dest, np->remote_mac);
495
496 skb->dev = np->dev;
497
498 return (int)netpoll_send_skb(np, skb);
499 }
500 EXPORT_SYMBOL(netpoll_send_udp);
501
netpoll_print_options(struct netpoll * np)502 void netpoll_print_options(struct netpoll *np)
503 {
504 np_info(np, "local port %d\n", np->local_port);
505 if (np->ipv6)
506 np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
507 else
508 np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
509 np_info(np, "interface name '%s'\n", np->dev_name);
510 np_info(np, "local ethernet address '%pM'\n", np->dev_mac);
511 np_info(np, "remote port %d\n", np->remote_port);
512 if (np->ipv6)
513 np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
514 else
515 np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
516 np_info(np, "remote ethernet address %pM\n", np->remote_mac);
517 }
518 EXPORT_SYMBOL(netpoll_print_options);
519
netpoll_parse_ip_addr(const char * str,union inet_addr * addr)520 static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
521 {
522 const char *end;
523
524 if (!strchr(str, ':') &&
525 in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
526 if (!*end)
527 return 0;
528 }
529 if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
530 #if IS_ENABLED(CONFIG_IPV6)
531 if (!*end)
532 return 1;
533 #else
534 return -1;
535 #endif
536 }
537 return -1;
538 }
539
skb_pool_flush(struct netpoll * np)540 static void skb_pool_flush(struct netpoll *np)
541 {
542 struct sk_buff_head *skb_pool;
543
544 cancel_work_sync(&np->refill_wq);
545 skb_pool = &np->skb_pool;
546 skb_queue_purge_reason(skb_pool, SKB_CONSUMED);
547 }
548
netpoll_parse_options(struct netpoll * np,char * opt)549 int netpoll_parse_options(struct netpoll *np, char *opt)
550 {
551 char *cur=opt, *delim;
552 int ipv6;
553 bool ipversion_set = false;
554
555 if (*cur != '@') {
556 if ((delim = strchr(cur, '@')) == NULL)
557 goto parse_failed;
558 *delim = 0;
559 if (kstrtou16(cur, 10, &np->local_port))
560 goto parse_failed;
561 cur = delim;
562 }
563 cur++;
564
565 if (*cur != '/') {
566 ipversion_set = true;
567 if ((delim = strchr(cur, '/')) == NULL)
568 goto parse_failed;
569 *delim = 0;
570 ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
571 if (ipv6 < 0)
572 goto parse_failed;
573 else
574 np->ipv6 = (bool)ipv6;
575 cur = delim;
576 }
577 cur++;
578
579 if (*cur != ',') {
580 /* parse out dev_name or dev_mac */
581 if ((delim = strchr(cur, ',')) == NULL)
582 goto parse_failed;
583 *delim = 0;
584
585 np->dev_name[0] = '\0';
586 eth_broadcast_addr(np->dev_mac);
587 if (!strchr(cur, ':'))
588 strscpy(np->dev_name, cur, sizeof(np->dev_name));
589 else if (!mac_pton(cur, np->dev_mac))
590 goto parse_failed;
591
592 cur = delim;
593 }
594 cur++;
595
596 if (*cur != '@') {
597 /* dst port */
598 if ((delim = strchr(cur, '@')) == NULL)
599 goto parse_failed;
600 *delim = 0;
601 if (*cur == ' ' || *cur == '\t')
602 np_info(np, "warning: whitespace is not allowed\n");
603 if (kstrtou16(cur, 10, &np->remote_port))
604 goto parse_failed;
605 cur = delim;
606 }
607 cur++;
608
609 /* dst ip */
610 if ((delim = strchr(cur, '/')) == NULL)
611 goto parse_failed;
612 *delim = 0;
613 ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
614 if (ipv6 < 0)
615 goto parse_failed;
616 else if (ipversion_set && np->ipv6 != (bool)ipv6)
617 goto parse_failed;
618 else
619 np->ipv6 = (bool)ipv6;
620 cur = delim + 1;
621
622 if (*cur != 0) {
623 /* MAC address */
624 if (!mac_pton(cur, np->remote_mac))
625 goto parse_failed;
626 }
627
628 netpoll_print_options(np);
629
630 return 0;
631
632 parse_failed:
633 np_info(np, "couldn't parse config at '%s'!\n", cur);
634 return -1;
635 }
636 EXPORT_SYMBOL(netpoll_parse_options);
637
refill_skbs_work_handler(struct work_struct * work)638 static void refill_skbs_work_handler(struct work_struct *work)
639 {
640 struct netpoll *np =
641 container_of(work, struct netpoll, refill_wq);
642
643 refill_skbs(np);
644 }
645
__netpoll_setup(struct netpoll * np,struct net_device * ndev)646 int __netpoll_setup(struct netpoll *np, struct net_device *ndev)
647 {
648 struct netpoll_info *npinfo;
649 const struct net_device_ops *ops;
650 int err;
651
652 skb_queue_head_init(&np->skb_pool);
653
654 if (ndev->priv_flags & IFF_DISABLE_NETPOLL) {
655 np_err(np, "%s doesn't support polling, aborting\n",
656 ndev->name);
657 err = -ENOTSUPP;
658 goto out;
659 }
660
661 npinfo = rtnl_dereference(ndev->npinfo);
662 if (!npinfo) {
663 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
664 if (!npinfo) {
665 err = -ENOMEM;
666 goto out;
667 }
668
669 sema_init(&npinfo->dev_lock, 1);
670 skb_queue_head_init(&npinfo->txq);
671 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
672
673 refcount_set(&npinfo->refcnt, 1);
674
675 ops = ndev->netdev_ops;
676 if (ops->ndo_netpoll_setup) {
677 err = ops->ndo_netpoll_setup(ndev);
678 if (err)
679 goto free_npinfo;
680 }
681 } else {
682 refcount_inc(&npinfo->refcnt);
683 }
684
685 np->dev = ndev;
686 strscpy(np->dev_name, ndev->name, IFNAMSIZ);
687 npinfo->netpoll = np;
688
689 /* fill up the skb queue */
690 refill_skbs(np);
691 INIT_WORK(&np->refill_wq, refill_skbs_work_handler);
692
693 /* last thing to do is link it to the net device structure */
694 rcu_assign_pointer(ndev->npinfo, npinfo);
695
696 return 0;
697
698 free_npinfo:
699 kfree(npinfo);
700 out:
701 return err;
702 }
703 EXPORT_SYMBOL_GPL(__netpoll_setup);
704
705 /*
706 * Returns a pointer to a string representation of the identifier used
707 * to select the egress interface for the given netpoll instance. buf
708 * must be a buffer of length at least MAC_ADDR_STR_LEN + 1.
709 */
egress_dev(struct netpoll * np,char * buf)710 static char *egress_dev(struct netpoll *np, char *buf)
711 {
712 if (np->dev_name[0])
713 return np->dev_name;
714
715 snprintf(buf, MAC_ADDR_STR_LEN, "%pM", np->dev_mac);
716 return buf;
717 }
718
netpoll_setup(struct netpoll * np)719 int netpoll_setup(struct netpoll *np)
720 {
721 struct net *net = current->nsproxy->net_ns;
722 char buf[MAC_ADDR_STR_LEN + 1];
723 struct net_device *ndev = NULL;
724 bool ip_overwritten = false;
725 struct in_device *in_dev;
726 int err;
727
728 rtnl_lock();
729 if (np->dev_name[0])
730 ndev = __dev_get_by_name(net, np->dev_name);
731 else if (is_valid_ether_addr(np->dev_mac))
732 ndev = dev_getbyhwaddr(net, ARPHRD_ETHER, np->dev_mac);
733
734 if (!ndev) {
735 np_err(np, "%s doesn't exist, aborting\n", egress_dev(np, buf));
736 err = -ENODEV;
737 goto unlock;
738 }
739 netdev_hold(ndev, &np->dev_tracker, GFP_KERNEL);
740
741 if (netdev_master_upper_dev_get(ndev)) {
742 np_err(np, "%s is a slave device, aborting\n",
743 egress_dev(np, buf));
744 err = -EBUSY;
745 goto put;
746 }
747
748 if (!netif_running(ndev)) {
749 unsigned long atmost;
750
751 np_info(np, "device %s not up yet, forcing it\n",
752 egress_dev(np, buf));
753
754 err = dev_open(ndev, NULL);
755
756 if (err) {
757 np_err(np, "failed to open %s\n", ndev->name);
758 goto put;
759 }
760
761 rtnl_unlock();
762 atmost = jiffies + carrier_timeout * HZ;
763 while (!netif_carrier_ok(ndev)) {
764 if (time_after(jiffies, atmost)) {
765 np_notice(np, "timeout waiting for carrier\n");
766 break;
767 }
768 msleep(1);
769 }
770
771 rtnl_lock();
772 }
773
774 if (!np->local_ip.ip) {
775 if (!np->ipv6) {
776 const struct in_ifaddr *ifa;
777
778 in_dev = __in_dev_get_rtnl(ndev);
779 if (!in_dev)
780 goto put_noaddr;
781
782 ifa = rtnl_dereference(in_dev->ifa_list);
783 if (!ifa) {
784 put_noaddr:
785 np_err(np, "no IP address for %s, aborting\n",
786 egress_dev(np, buf));
787 err = -EDESTADDRREQ;
788 goto put;
789 }
790
791 np->local_ip.ip = ifa->ifa_local;
792 ip_overwritten = true;
793 np_info(np, "local IP %pI4\n", &np->local_ip.ip);
794 } else {
795 #if IS_ENABLED(CONFIG_IPV6)
796 struct inet6_dev *idev;
797
798 err = -EDESTADDRREQ;
799 idev = __in6_dev_get(ndev);
800 if (idev) {
801 struct inet6_ifaddr *ifp;
802
803 read_lock_bh(&idev->lock);
804 list_for_each_entry(ifp, &idev->addr_list, if_list) {
805 if (!!(ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL) !=
806 !!(ipv6_addr_type(&np->remote_ip.in6) & IPV6_ADDR_LINKLOCAL))
807 continue;
808 np->local_ip.in6 = ifp->addr;
809 ip_overwritten = true;
810 err = 0;
811 break;
812 }
813 read_unlock_bh(&idev->lock);
814 }
815 if (err) {
816 np_err(np, "no IPv6 address for %s, aborting\n",
817 egress_dev(np, buf));
818 goto put;
819 } else
820 np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
821 #else
822 np_err(np, "IPv6 is not supported %s, aborting\n",
823 egress_dev(np, buf));
824 err = -EINVAL;
825 goto put;
826 #endif
827 }
828 }
829
830 err = __netpoll_setup(np, ndev);
831 if (err)
832 goto flush;
833 rtnl_unlock();
834 return 0;
835
836 flush:
837 skb_pool_flush(np);
838 put:
839 DEBUG_NET_WARN_ON_ONCE(np->dev);
840 if (ip_overwritten)
841 memset(&np->local_ip, 0, sizeof(np->local_ip));
842 netdev_put(ndev, &np->dev_tracker);
843 unlock:
844 rtnl_unlock();
845 return err;
846 }
847 EXPORT_SYMBOL(netpoll_setup);
848
rcu_cleanup_netpoll_info(struct rcu_head * rcu_head)849 static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
850 {
851 struct netpoll_info *npinfo =
852 container_of(rcu_head, struct netpoll_info, rcu);
853
854 skb_queue_purge(&npinfo->txq);
855
856 /* we can't call cancel_delayed_work_sync here, as we are in softirq */
857 cancel_delayed_work(&npinfo->tx_work);
858
859 /* clean after last, unfinished work */
860 __skb_queue_purge(&npinfo->txq);
861 /* now cancel it again */
862 cancel_delayed_work(&npinfo->tx_work);
863 kfree(npinfo);
864 }
865
__netpoll_cleanup(struct netpoll * np)866 void __netpoll_cleanup(struct netpoll *np)
867 {
868 struct netpoll_info *npinfo;
869
870 npinfo = rtnl_dereference(np->dev->npinfo);
871 if (!npinfo)
872 return;
873
874 if (refcount_dec_and_test(&npinfo->refcnt)) {
875 const struct net_device_ops *ops;
876
877 ops = np->dev->netdev_ops;
878 if (ops->ndo_netpoll_cleanup)
879 ops->ndo_netpoll_cleanup(np->dev);
880
881 RCU_INIT_POINTER(np->dev->npinfo, NULL);
882 call_rcu(&npinfo->rcu, rcu_cleanup_netpoll_info);
883 } else
884 RCU_INIT_POINTER(np->dev->npinfo, NULL);
885
886 skb_pool_flush(np);
887 }
888 EXPORT_SYMBOL_GPL(__netpoll_cleanup);
889
__netpoll_free(struct netpoll * np)890 void __netpoll_free(struct netpoll *np)
891 {
892 ASSERT_RTNL();
893
894 /* Wait for transmitting packets to finish before freeing. */
895 synchronize_rcu();
896 __netpoll_cleanup(np);
897 kfree(np);
898 }
899 EXPORT_SYMBOL_GPL(__netpoll_free);
900
do_netpoll_cleanup(struct netpoll * np)901 void do_netpoll_cleanup(struct netpoll *np)
902 {
903 __netpoll_cleanup(np);
904 netdev_put(np->dev, &np->dev_tracker);
905 np->dev = NULL;
906 }
907 EXPORT_SYMBOL(do_netpoll_cleanup);
908
netpoll_cleanup(struct netpoll * np)909 void netpoll_cleanup(struct netpoll *np)
910 {
911 rtnl_lock();
912 if (!np->dev)
913 goto out;
914 do_netpoll_cleanup(np);
915 out:
916 rtnl_unlock();
917 }
918 EXPORT_SYMBOL(netpoll_cleanup);
919