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/netdevice.h>
16#include <linux/etherdevice.h>
17#include <linux/string.h>
18#include <linux/if_arp.h>
19#include <linux/inetdevice.h>
20#include <linux/inet.h>
21#include <linux/interrupt.h>
22#include <linux/netpoll.h>
23#include <linux/sched.h>
24#include <linux/delay.h>
25#include <linux/rcupdate.h>
26#include <linux/workqueue.h>
27#include <linux/slab.h>
28#include <linux/export.h>
29#include <net/tcp.h>
30#include <net/udp.h>
31#include <asm/unaligned.h>
32#include <trace/events/napi.h>
33
34/*
35 * We maintain a small pool of fully-sized skbs, to make sure the
36 * message gets out even in extreme OOM situations.
37 */
38
39#define MAX_UDP_CHUNK 1460
40#define MAX_SKBS 32
41
42static struct sk_buff_head skb_pool;
43
44static atomic_t trapped;
45
46#define USEC_PER_POLL 50
47#define NETPOLL_RX_ENABLED 1
48#define NETPOLL_RX_DROP 2
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);
57static void arp_reply(struct sk_buff *skb);
58
59static unsigned int carrier_timeout = 4;
60module_param(carrier_timeout, uint, 0644);
61
62#define np_info(np, fmt, ...) \
63 pr_info("%s: " fmt, np->name, ##__VA_ARGS__)
64#define np_err(np, fmt, ...) \
65 pr_err("%s: " fmt, np->name, ##__VA_ARGS__)
66#define np_notice(np, fmt, ...) \
67 pr_notice("%s: " fmt, np->name, ##__VA_ARGS__)
68
69static void queue_process(struct work_struct *work)
70{
71 struct netpoll_info *npinfo =
72 container_of(work, struct netpoll_info, tx_work.work);
73 struct sk_buff *skb;
74 unsigned long flags;
75
76 while ((skb = skb_dequeue(&npinfo->txq))) {
77 struct net_device *dev = skb->dev;
78 const struct net_device_ops *ops = dev->netdev_ops;
79 struct netdev_queue *txq;
80
81 if (!netif_device_present(dev) || !netif_running(dev)) {
82 __kfree_skb(skb);
83 continue;
84 }
85
86 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
87
88 local_irq_save(flags);
89 __netif_tx_lock(txq, smp_processor_id());
90 if (netif_xmit_frozen_or_stopped(txq) ||
91 ops->ndo_start_xmit(skb, dev) != NETDEV_TX_OK) {
92 skb_queue_head(&npinfo->txq, skb);
93 __netif_tx_unlock(txq);
94 local_irq_restore(flags);
95
96 schedule_delayed_work(&npinfo->tx_work, HZ/10);
97 return;
98 }
99 __netif_tx_unlock(txq);
100 local_irq_restore(flags);
101 }
102}
103
104static __sum16 checksum_udp(struct sk_buff *skb, struct udphdr *uh,
105 unsigned short ulen, __be32 saddr, __be32 daddr)
106{
107 __wsum psum;
108
109 if (uh->check == 0 || skb_csum_unnecessary(skb))
110 return 0;
111
112 psum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0);
113
114 if (skb->ip_summed == CHECKSUM_COMPLETE &&
115 !csum_fold(csum_add(psum, skb->csum)))
116 return 0;
117
118 skb->csum = psum;
119
120 return __skb_checksum_complete(skb);
121}
122
123/*
124 * Check whether delayed processing was scheduled for our NIC. If so,
125 * we attempt to grab the poll lock and use ->poll() to pump the card.
126 * If this fails, either we've recursed in ->poll() or it's already
127 * running on another CPU.
128 *
129 * Note: we don't mask interrupts with this lock because we're using
130 * trylock here and interrupts are already disabled in the softirq
131 * case. Further, we test the poll_owner to avoid recursion on UP
132 * systems where the lock doesn't exist.
133 *
134 * In cases where there is bi-directional communications, reading only
135 * one message at a time can lead to packets being dropped by the
136 * network adapter, forcing superfluous retries and possibly timeouts.
137 * Thus, we set our budget to greater than 1.
138 */
139static int poll_one_napi(struct netpoll_info *npinfo,
140 struct napi_struct *napi, int budget)
141{
142 int work;
143
144 /* net_rx_action's ->poll() invocations and our's are
145 * synchronized by this test which is only made while
146 * holding the napi->poll_lock.
147 */
148 if (!test_bit(NAPI_STATE_SCHED, &napi->state))
149 return budget;
150
151 npinfo->rx_flags |= NETPOLL_RX_DROP;
152 atomic_inc(&trapped);
153 set_bit(NAPI_STATE_NPSVC, &napi->state);
154
155 work = napi->poll(napi, budget);
156 trace_napi_poll(napi);
157
158 clear_bit(NAPI_STATE_NPSVC, &napi->state);
159 atomic_dec(&trapped);
160 npinfo->rx_flags &= ~NETPOLL_RX_DROP;
161
162 return budget - work;
163}
164
165static void poll_napi(struct net_device *dev)
166{
167 struct napi_struct *napi;
168 int budget = 16;
169
170 list_for_each_entry(napi, &dev->napi_list, dev_list) {
171 if (napi->poll_owner != smp_processor_id() &&
172 spin_trylock(&napi->poll_lock)) {
173 budget = poll_one_napi(dev->npinfo, napi, budget);
174 spin_unlock(&napi->poll_lock);
175
176 if (!budget)
177 break;
178 }
179 }
180}
181
182static void service_arp_queue(struct netpoll_info *npi)
183{
184 if (npi) {
185 struct sk_buff *skb;
186
187 while ((skb = skb_dequeue(&npi->arp_tx)))
188 arp_reply(skb);
189 }
190}
191
192static void netpoll_poll_dev(struct net_device *dev)
193{
194 const struct net_device_ops *ops;
195
196 if (!dev || !netif_running(dev))
197 return;
198
199 ops = dev->netdev_ops;
200 if (!ops->ndo_poll_controller)
201 return;
202
203 /* Process pending work on NIC */
204 ops->ndo_poll_controller(dev);
205
206 poll_napi(dev);
207
208 if (dev->flags & IFF_SLAVE) {
209 if (dev->npinfo) {
210 struct net_device *bond_dev = dev->master;
211 struct sk_buff *skb;
212 while ((skb = skb_dequeue(&dev->npinfo->arp_tx))) {
213 skb->dev = bond_dev;
214 skb_queue_tail(&bond_dev->npinfo->arp_tx, skb);
215 }
216 }
217 }
218
219 service_arp_queue(dev->npinfo);
220
221 zap_completion_queue();
222}
223
224static void refill_skbs(void)
225{
226 struct sk_buff *skb;
227 unsigned long flags;
228
229 spin_lock_irqsave(&skb_pool.lock, flags);
230 while (skb_pool.qlen < MAX_SKBS) {
231 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
232 if (!skb)
233 break;
234
235 __skb_queue_tail(&skb_pool, skb);
236 }
237 spin_unlock_irqrestore(&skb_pool.lock, flags);
238}
239
240static void zap_completion_queue(void)
241{
242 unsigned long flags;
243 struct softnet_data *sd = &get_cpu_var(softnet_data);
244
245 if (sd->completion_queue) {
246 struct sk_buff *clist;
247
248 local_irq_save(flags);
249 clist = sd->completion_queue;
250 sd->completion_queue = NULL;
251 local_irq_restore(flags);
252
253 while (clist != NULL) {
254 struct sk_buff *skb = clist;
255 clist = clist->next;
256 if (skb->destructor) {
257 atomic_inc(&skb->users);
258 dev_kfree_skb_any(skb); /* put this one back */
259 } else {
260 __kfree_skb(skb);
261 }
262 }
263 }
264
265 put_cpu_var(softnet_data);
266}
267
268static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
269{
270 int count = 0;
271 struct sk_buff *skb;
272
273 zap_completion_queue();
274 refill_skbs();
275repeat:
276
277 skb = alloc_skb(len, GFP_ATOMIC);
278 if (!skb)
279 skb = skb_dequeue(&skb_pool);
280
281 if (!skb) {
282 if (++count < 10) {
283 netpoll_poll_dev(np->dev);
284 goto repeat;
285 }
286 return NULL;
287 }
288
289 atomic_set(&skb->users, 1);
290 skb_reserve(skb, reserve);
291 return skb;
292}
293
294static int netpoll_owner_active(struct net_device *dev)
295{
296 struct napi_struct *napi;
297
298 list_for_each_entry(napi, &dev->napi_list, dev_list) {
299 if (napi->poll_owner == smp_processor_id())
300 return 1;
301 }
302 return 0;
303}
304
305void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb,
306 struct net_device *dev)
307{
308 int status = NETDEV_TX_BUSY;
309 unsigned long tries;
310 const struct net_device_ops *ops = dev->netdev_ops;
311 /* It is up to the caller to keep npinfo alive. */
312 struct netpoll_info *npinfo = np->dev->npinfo;
313
314 if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
315 __kfree_skb(skb);
316 return;
317 }
318
319 /* don't get messages out of order, and no recursion */
320 if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
321 struct netdev_queue *txq;
322 unsigned long flags;
323
324 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
325
326 local_irq_save(flags);
327 /* try until next clock tick */
328 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
329 tries > 0; --tries) {
330 if (__netif_tx_trylock(txq)) {
331 if (!netif_xmit_stopped(txq)) {
332 status = ops->ndo_start_xmit(skb, dev);
333 if (status == NETDEV_TX_OK)
334 txq_trans_update(txq);
335 }
336 __netif_tx_unlock(txq);
337
338 if (status == NETDEV_TX_OK)
339 break;
340
341 }
342
343 /* tickle device maybe there is some cleanup */
344 netpoll_poll_dev(np->dev);
345
346 udelay(USEC_PER_POLL);
347 }
348
349 WARN_ONCE(!irqs_disabled(),
350 "netpoll_send_skb(): %s enabled interrupts in poll (%pF)\n",
351 dev->name, ops->ndo_start_xmit);
352
353 local_irq_restore(flags);
354 }
355
356 if (status != NETDEV_TX_OK) {
357 skb_queue_tail(&npinfo->txq, skb);
358 schedule_delayed_work(&npinfo->tx_work,0);
359 }
360}
361EXPORT_SYMBOL(netpoll_send_skb_on_dev);
362
363void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
364{
365 int total_len, ip_len, udp_len;
366 struct sk_buff *skb;
367 struct udphdr *udph;
368 struct iphdr *iph;
369 struct ethhdr *eth;
370
371 udp_len = len + sizeof(*udph);
372 ip_len = udp_len + sizeof(*iph);
373 total_len = ip_len + LL_RESERVED_SPACE(np->dev);
374
375 skb = find_skb(np, total_len + np->dev->needed_tailroom,
376 total_len - len);
377 if (!skb)
378 return;
379
380 skb_copy_to_linear_data(skb, msg, len);
381 skb_put(skb, len);
382
383 skb_push(skb, sizeof(*udph));
384 skb_reset_transport_header(skb);
385 udph = udp_hdr(skb);
386 udph->source = htons(np->local_port);
387 udph->dest = htons(np->remote_port);
388 udph->len = htons(udp_len);
389 udph->check = 0;
390 udph->check = csum_tcpudp_magic(np->local_ip,
391 np->remote_ip,
392 udp_len, IPPROTO_UDP,
393 csum_partial(udph, udp_len, 0));
394 if (udph->check == 0)
395 udph->check = CSUM_MANGLED_0;
396
397 skb_push(skb, sizeof(*iph));
398 skb_reset_network_header(skb);
399 iph = ip_hdr(skb);
400
401 /* iph->version = 4; iph->ihl = 5; */
402 put_unaligned(0x45, (unsigned char *)iph);
403 iph->tos = 0;
404 put_unaligned(htons(ip_len), &(iph->tot_len));
405 iph->id = 0;
406 iph->frag_off = 0;
407 iph->ttl = 64;
408 iph->protocol = IPPROTO_UDP;
409 iph->check = 0;
410 put_unaligned(np->local_ip, &(iph->saddr));
411 put_unaligned(np->remote_ip, &(iph->daddr));
412 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
413
414 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
415 skb_reset_mac_header(skb);
416 skb->protocol = eth->h_proto = htons(ETH_P_IP);
417 memcpy(eth->h_source, np->dev->dev_addr, ETH_ALEN);
418 memcpy(eth->h_dest, np->remote_mac, ETH_ALEN);
419
420 skb->dev = np->dev;
421
422 netpoll_send_skb(np, skb);
423}
424EXPORT_SYMBOL(netpoll_send_udp);
425
426static void arp_reply(struct sk_buff *skb)
427{
428 struct netpoll_info *npinfo = skb->dev->npinfo;
429 struct arphdr *arp;
430 unsigned char *arp_ptr;
431 int size, type = ARPOP_REPLY, ptype = ETH_P_ARP;
432 __be32 sip, tip;
433 unsigned char *sha;
434 struct sk_buff *send_skb;
435 struct netpoll *np, *tmp;
436 unsigned long flags;
437 int hlen, tlen;
438 int hits = 0;
439
440 if (list_empty(&npinfo->rx_np))
441 return;
442
443 /* Before checking the packet, we do some early
444 inspection whether this is interesting at all */
445 spin_lock_irqsave(&npinfo->rx_lock, flags);
446 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
447 if (np->dev == skb->dev)
448 hits++;
449 }
450 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
451
452 /* No netpoll struct is using this dev */
453 if (!hits)
454 return;
455
456 /* No arp on this interface */
457 if (skb->dev->flags & IFF_NOARP)
458 return;
459
460 if (!pskb_may_pull(skb, arp_hdr_len(skb->dev)))
461 return;
462
463 skb_reset_network_header(skb);
464 skb_reset_transport_header(skb);
465 arp = arp_hdr(skb);
466
467 if ((arp->ar_hrd != htons(ARPHRD_ETHER) &&
468 arp->ar_hrd != htons(ARPHRD_IEEE802)) ||
469 arp->ar_pro != htons(ETH_P_IP) ||
470 arp->ar_op != htons(ARPOP_REQUEST))
471 return;
472
473 arp_ptr = (unsigned char *)(arp+1);
474 /* save the location of the src hw addr */
475 sha = arp_ptr;
476 arp_ptr += skb->dev->addr_len;
477 memcpy(&sip, arp_ptr, 4);
478 arp_ptr += 4;
479 /* If we actually cared about dst hw addr,
480 it would get copied here */
481 arp_ptr += skb->dev->addr_len;
482 memcpy(&tip, arp_ptr, 4);
483
484 /* Should we ignore arp? */
485 if (ipv4_is_loopback(tip) || ipv4_is_multicast(tip))
486 return;
487
488 size = arp_hdr_len(skb->dev);
489
490 spin_lock_irqsave(&npinfo->rx_lock, flags);
491 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
492 if (tip != np->local_ip)
493 continue;
494
495 hlen = LL_RESERVED_SPACE(np->dev);
496 tlen = np->dev->needed_tailroom;
497 send_skb = find_skb(np, size + hlen + tlen, hlen);
498 if (!send_skb)
499 continue;
500
501 skb_reset_network_header(send_skb);
502 arp = (struct arphdr *) skb_put(send_skb, size);
503 send_skb->dev = skb->dev;
504 send_skb->protocol = htons(ETH_P_ARP);
505
506 /* Fill the device header for the ARP frame */
507 if (dev_hard_header(send_skb, skb->dev, ptype,
508 sha, np->dev->dev_addr,
509 send_skb->len) < 0) {
510 kfree_skb(send_skb);
511 continue;
512 }
513
514 /*
515 * Fill out the arp protocol part.
516 *
517 * we only support ethernet device type,
518 * which (according to RFC 1390) should
519 * always equal 1 (Ethernet).
520 */
521
522 arp->ar_hrd = htons(np->dev->type);
523 arp->ar_pro = htons(ETH_P_IP);
524 arp->ar_hln = np->dev->addr_len;
525 arp->ar_pln = 4;
526 arp->ar_op = htons(type);
527
528 arp_ptr = (unsigned char *)(arp + 1);
529 memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
530 arp_ptr += np->dev->addr_len;
531 memcpy(arp_ptr, &tip, 4);
532 arp_ptr += 4;
533 memcpy(arp_ptr, sha, np->dev->addr_len);
534 arp_ptr += np->dev->addr_len;
535 memcpy(arp_ptr, &sip, 4);
536
537 netpoll_send_skb(np, send_skb);
538
539 /* If there are several rx_hooks for the same address,
540 we're fine by sending a single reply */
541 break;
542 }
543 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
544}
545
546int __netpoll_rx(struct sk_buff *skb)
547{
548 int proto, len, ulen;
549 int hits = 0;
550 const struct iphdr *iph;
551 struct udphdr *uh;
552 struct netpoll_info *npinfo = skb->dev->npinfo;
553 struct netpoll *np, *tmp;
554
555 if (list_empty(&npinfo->rx_np))
556 goto out;
557
558 if (skb->dev->type != ARPHRD_ETHER)
559 goto out;
560
561 /* check if netpoll clients need ARP */
562 if (skb->protocol == htons(ETH_P_ARP) &&
563 atomic_read(&trapped)) {
564 skb_queue_tail(&npinfo->arp_tx, skb);
565 return 1;
566 }
567
568 proto = ntohs(eth_hdr(skb)->h_proto);
569 if (proto != ETH_P_IP)
570 goto out;
571 if (skb->pkt_type == PACKET_OTHERHOST)
572 goto out;
573 if (skb_shared(skb))
574 goto out;
575
576 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
577 goto out;
578 iph = (struct iphdr *)skb->data;
579 if (iph->ihl < 5 || iph->version != 4)
580 goto out;
581 if (!pskb_may_pull(skb, iph->ihl*4))
582 goto out;
583 iph = (struct iphdr *)skb->data;
584 if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
585 goto out;
586
587 len = ntohs(iph->tot_len);
588 if (skb->len < len || len < iph->ihl*4)
589 goto out;
590
591 /*
592 * Our transport medium may have padded the buffer out.
593 * Now We trim to the true length of the frame.
594 */
595 if (pskb_trim_rcsum(skb, len))
596 goto out;
597
598 iph = (struct iphdr *)skb->data;
599 if (iph->protocol != IPPROTO_UDP)
600 goto out;
601
602 len -= iph->ihl*4;
603 uh = (struct udphdr *)(((char *)iph) + iph->ihl*4);
604 ulen = ntohs(uh->len);
605
606 if (ulen != len)
607 goto out;
608 if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr))
609 goto out;
610
611 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
612 if (np->local_ip && np->local_ip != iph->daddr)
613 continue;
614 if (np->remote_ip && np->remote_ip != iph->saddr)
615 continue;
616 if (np->local_port && np->local_port != ntohs(uh->dest))
617 continue;
618
619 np->rx_hook(np, ntohs(uh->source),
620 (char *)(uh+1),
621 ulen - sizeof(struct udphdr));
622 hits++;
623 }
624
625 if (!hits)
626 goto out;
627
628 kfree_skb(skb);
629 return 1;
630
631out:
632 if (atomic_read(&trapped)) {
633 kfree_skb(skb);
634 return 1;
635 }
636
637 return 0;
638}
639
640void netpoll_print_options(struct netpoll *np)
641{
642 np_info(np, "local port %d\n", np->local_port);
643 np_info(np, "local IP %pI4\n", &np->local_ip);
644 np_info(np, "interface '%s'\n", np->dev_name);
645 np_info(np, "remote port %d\n", np->remote_port);
646 np_info(np, "remote IP %pI4\n", &np->remote_ip);
647 np_info(np, "remote ethernet address %pM\n", np->remote_mac);
648}
649EXPORT_SYMBOL(netpoll_print_options);
650
651int netpoll_parse_options(struct netpoll *np, char *opt)
652{
653 char *cur=opt, *delim;
654
655 if (*cur != '@') {
656 if ((delim = strchr(cur, '@')) == NULL)
657 goto parse_failed;
658 *delim = 0;
659 np->local_port = simple_strtol(cur, NULL, 10);
660 cur = delim;
661 }
662 cur++;
663
664 if (*cur != '/') {
665 if ((delim = strchr(cur, '/')) == NULL)
666 goto parse_failed;
667 *delim = 0;
668 np->local_ip = in_aton(cur);
669 cur = delim;
670 }
671 cur++;
672
673 if (*cur != ',') {
674 /* parse out dev name */
675 if ((delim = strchr(cur, ',')) == NULL)
676 goto parse_failed;
677 *delim = 0;
678 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
679 cur = delim;
680 }
681 cur++;
682
683 if (*cur != '@') {
684 /* dst port */
685 if ((delim = strchr(cur, '@')) == NULL)
686 goto parse_failed;
687 *delim = 0;
688 if (*cur == ' ' || *cur == '\t')
689 np_info(np, "warning: whitespace is not allowed\n");
690 np->remote_port = simple_strtol(cur, NULL, 10);
691 cur = delim;
692 }
693 cur++;
694
695 /* dst ip */
696 if ((delim = strchr(cur, '/')) == NULL)
697 goto parse_failed;
698 *delim = 0;
699 np->remote_ip = in_aton(cur);
700 cur = delim + 1;
701
702 if (*cur != 0) {
703 /* MAC address */
704 if (!mac_pton(cur, np->remote_mac))
705 goto parse_failed;
706 }
707
708 netpoll_print_options(np);
709
710 return 0;
711
712 parse_failed:
713 np_info(np, "couldn't parse config at '%s'!\n", cur);
714 return -1;
715}
716EXPORT_SYMBOL(netpoll_parse_options);
717
718int __netpoll_setup(struct netpoll *np)
719{
720 struct net_device *ndev = np->dev;
721 struct netpoll_info *npinfo;
722 const struct net_device_ops *ops;
723 unsigned long flags;
724 int err;
725
726 if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) ||
727 !ndev->netdev_ops->ndo_poll_controller) {
728 np_err(np, "%s doesn't support polling, aborting\n",
729 np->dev_name);
730 err = -ENOTSUPP;
731 goto out;
732 }
733
734 if (!ndev->npinfo) {
735 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
736 if (!npinfo) {
737 err = -ENOMEM;
738 goto out;
739 }
740
741 npinfo->rx_flags = 0;
742 INIT_LIST_HEAD(&npinfo->rx_np);
743
744 spin_lock_init(&npinfo->rx_lock);
745 skb_queue_head_init(&npinfo->arp_tx);
746 skb_queue_head_init(&npinfo->txq);
747 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
748
749 atomic_set(&npinfo->refcnt, 1);
750
751 ops = np->dev->netdev_ops;
752 if (ops->ndo_netpoll_setup) {
753 err = ops->ndo_netpoll_setup(ndev, npinfo);
754 if (err)
755 goto free_npinfo;
756 }
757 } else {
758 npinfo = ndev->npinfo;
759 atomic_inc(&npinfo->refcnt);
760 }
761
762 npinfo->netpoll = np;
763
764 if (np->rx_hook) {
765 spin_lock_irqsave(&npinfo->rx_lock, flags);
766 npinfo->rx_flags |= NETPOLL_RX_ENABLED;
767 list_add_tail(&np->rx, &npinfo->rx_np);
768 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
769 }
770
771 /* last thing to do is link it to the net device structure */
772 rcu_assign_pointer(ndev->npinfo, npinfo);
773
774 return 0;
775
776free_npinfo:
777 kfree(npinfo);
778out:
779 return err;
780}
781EXPORT_SYMBOL_GPL(__netpoll_setup);
782
783int netpoll_setup(struct netpoll *np)
784{
785 struct net_device *ndev = NULL;
786 struct in_device *in_dev;
787 int err;
788
789 if (np->dev_name)
790 ndev = dev_get_by_name(&init_net, np->dev_name);
791 if (!ndev) {
792 np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
793 return -ENODEV;
794 }
795
796 if (ndev->master) {
797 np_err(np, "%s is a slave device, aborting\n", np->dev_name);
798 err = -EBUSY;
799 goto put;
800 }
801
802 if (!netif_running(ndev)) {
803 unsigned long atmost, atleast;
804
805 np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
806
807 rtnl_lock();
808 err = dev_open(ndev);
809 rtnl_unlock();
810
811 if (err) {
812 np_err(np, "failed to open %s\n", ndev->name);
813 goto put;
814 }
815
816 atleast = jiffies + HZ/10;
817 atmost = jiffies + carrier_timeout * HZ;
818 while (!netif_carrier_ok(ndev)) {
819 if (time_after(jiffies, atmost)) {
820 np_notice(np, "timeout waiting for carrier\n");
821 break;
822 }
823 msleep(1);
824 }
825
826 /* If carrier appears to come up instantly, we don't
827 * trust it and pause so that we don't pump all our
828 * queued console messages into the bitbucket.
829 */
830
831 if (time_before(jiffies, atleast)) {
832 np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
833 msleep(4000);
834 }
835 }
836
837 if (!np->local_ip) {
838 rcu_read_lock();
839 in_dev = __in_dev_get_rcu(ndev);
840
841 if (!in_dev || !in_dev->ifa_list) {
842 rcu_read_unlock();
843 np_err(np, "no IP address for %s, aborting\n",
844 np->dev_name);
845 err = -EDESTADDRREQ;
846 goto put;
847 }
848
849 np->local_ip = in_dev->ifa_list->ifa_local;
850 rcu_read_unlock();
851 np_info(np, "local IP %pI4\n", &np->local_ip);
852 }
853
854 np->dev = ndev;
855
856 /* fill up the skb queue */
857 refill_skbs();
858
859 rtnl_lock();
860 err = __netpoll_setup(np);
861 rtnl_unlock();
862
863 if (err)
864 goto put;
865
866 return 0;
867
868put:
869 dev_put(ndev);
870 return err;
871}
872EXPORT_SYMBOL(netpoll_setup);
873
874static int __init netpoll_init(void)
875{
876 skb_queue_head_init(&skb_pool);
877 return 0;
878}
879core_initcall(netpoll_init);
880
881void __netpoll_cleanup(struct netpoll *np)
882{
883 struct netpoll_info *npinfo;
884 unsigned long flags;
885
886 npinfo = np->dev->npinfo;
887 if (!npinfo)
888 return;
889
890 if (!list_empty(&npinfo->rx_np)) {
891 spin_lock_irqsave(&npinfo->rx_lock, flags);
892 list_del(&np->rx);
893 if (list_empty(&npinfo->rx_np))
894 npinfo->rx_flags &= ~NETPOLL_RX_ENABLED;
895 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
896 }
897
898 if (atomic_dec_and_test(&npinfo->refcnt)) {
899 const struct net_device_ops *ops;
900
901 ops = np->dev->netdev_ops;
902 if (ops->ndo_netpoll_cleanup)
903 ops->ndo_netpoll_cleanup(np->dev);
904
905 RCU_INIT_POINTER(np->dev->npinfo, NULL);
906
907 /* avoid racing with NAPI reading npinfo */
908 synchronize_rcu_bh();
909
910 skb_queue_purge(&npinfo->arp_tx);
911 skb_queue_purge(&npinfo->txq);
912 cancel_delayed_work_sync(&npinfo->tx_work);
913
914 /* clean after last, unfinished work */
915 __skb_queue_purge(&npinfo->txq);
916 kfree(npinfo);
917 }
918}
919EXPORT_SYMBOL_GPL(__netpoll_cleanup);
920
921void netpoll_cleanup(struct netpoll *np)
922{
923 if (!np->dev)
924 return;
925
926 rtnl_lock();
927 __netpoll_cleanup(np);
928 rtnl_unlock();
929
930 dev_put(np->dev);
931 np->dev = NULL;
932}
933EXPORT_SYMBOL(netpoll_cleanup);
934
935int netpoll_trap(void)
936{
937 return atomic_read(&trapped);
938}
939EXPORT_SYMBOL(netpoll_trap);
940
941void netpoll_set_trap(int trap)
942{
943 if (trap)
944 atomic_inc(&trapped);
945 else
946 atomic_dec(&trapped);
947}
948EXPORT_SYMBOL(netpoll_set_trap);
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);