~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/net/core/netpoll.c

Version: ~ [ linux-5.15-rc5 ] ~ [ linux-5.14.11 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.72 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.152 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.210 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.250 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.286 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.288 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

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

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp