net/core/netpoll.c

   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/dsa.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 <asm/unaligned.h>
  38 #include <trace/events/napi.h>
  39
  40 /*
  41  * We maintain a small pool of fully-sized skbs, to make sure the
  42  * message gets out even in extreme OOM situations.
  43  */
  44
  45 #define MAX_UDP_CHUNK 1460
  46 #define MAX_SKBS 32
  47
  48 static struct sk_buff_head skb_pool;
  49
  50 DEFINE_STATIC_SRCU(netpoll_srcu);
  51
  52 #define USEC_PER_POLL   50
  53
  54 #define MAX_SKB_SIZE                                                    \
  55         (sizeof(struct ethhdr) +                                        \
  56          sizeof(struct iphdr) +                                         \
  57          sizeof(struct udphdr) +                                        \
  58          MAX_UDP_CHUNK)
  59
  60 static void zap_completion_queue(void);
  61 static void netpoll_async_cleanup(struct work_struct *work);
  62
  63 static unsigned int carrier_timeout = 4;
  64 module_param(carrier_timeout, uint, 0644);
  65
  66 #define np_info(np, fmt, ...)                           \
  67         pr_info("%s: " fmt, np->name, ##__VA_ARGS__)
  68 #define np_err(np, fmt, ...)                            \
  69         pr_err("%s: " fmt, np->name, ##__VA_ARGS__)
  70 #define np_notice(np, fmt, ...)                         \
  71         pr_notice("%s: " fmt, np->name, ##__VA_ARGS__)
  72
  73 static int netpoll_start_xmit(struct sk_buff *skb, struct net_device *dev,
  74                               struct netdev_queue *txq)
  75 {
  76         int status = NETDEV_TX_OK;
  77         netdev_features_t features;
  78
  79         features = netif_skb_features(skb);
  80
  81         if (skb_vlan_tag_present(skb) &&
  82             !vlan_hw_offload_capable(features, skb->vlan_proto)) {
  83                 skb = __vlan_hwaccel_push_inside(skb);
  84                 if (unlikely(!skb)) {
  85                         /* This is actually a packet drop, but we
  86                          * don't want the code that calls this
  87                          * function to try and operate on a NULL skb.
  88                          */
  89                         goto out;
  90                 }
  91         }
  92
  93         status = netdev_start_xmit(skb, dev, txq, false);
  94
  95 out:
  96         return status;
  97 }
  98
  99 static void queue_process(struct work_struct *work)
 100 {
 101         struct netpoll_info *npinfo =
 102                 container_of(work, struct netpoll_info, tx_work.work);
 103         struct sk_buff *skb;
 104         unsigned long flags;
 105
 106         while ((skb = skb_dequeue(&npinfo->txq))) {
 107                 struct net_device *dev = skb->dev;
 108                 struct netdev_queue *txq;
 109                 unsigned int q_index;
 110
 111                 if (!netif_device_present(dev) || !netif_running(dev)) {
 112                         kfree_skb(skb);
 113                         continue;
 114                 }
 115
 116                 local_irq_save(flags);
 117                 /* check if skb->queue_mapping is still valid */
 118                 q_index = skb_get_queue_mapping(skb);
 119                 if (unlikely(q_index >= dev->real_num_tx_queues)) {
 120                         q_index = q_index % dev->real_num_tx_queues;
 121                         skb_set_queue_mapping(skb, q_index);
 122                 }
 123                 txq = netdev_get_tx_queue(dev, q_index);
 124                 HARD_TX_LOCK(dev, txq, smp_processor_id());
 125                 if (netif_xmit_frozen_or_stopped(txq) ||
 126                     !dev_xmit_complete(netpoll_start_xmit(skb, dev, txq))) {
 127                         skb_queue_head(&npinfo->txq, skb);
 128                         HARD_TX_UNLOCK(dev, txq);
 129                         local_irq_restore(flags);
 130
 131                         schedule_delayed_work(&npinfo->tx_work, HZ/10);
 132                         return;
 133                 }
 134                 HARD_TX_UNLOCK(dev, txq);
 135                 local_irq_restore(flags);
 136         }
 137 }
 138
 139 static int netif_local_xmit_active(struct net_device *dev)
 140 {
 141         int i;
 142
 143         for (i = 0; i < dev->num_tx_queues; i++) {
 144                 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
 145
 146                 if (READ_ONCE(txq->xmit_lock_owner) == smp_processor_id())
 147                         return 1;
 148         }
 149
 150         return 0;
 151 }
 152
 153 static void poll_one_napi(struct napi_struct *napi)
 154 {
 155         int work;
 156
 157         /* If we set this bit but see that it has already been set,
 158          * that indicates that napi has been disabled and we need
 159          * to abort this operation
 160          */
 161         if (test_and_set_bit(NAPI_STATE_NPSVC, &napi->state))
 162                 return;
 163
 164         /* We explicilty pass the polling call a budget of 0 to
 165          * indicate that we are clearing the Tx path only.
 166          */
 167         work = napi->poll(napi, 0);
 168         WARN_ONCE(work, "%pF exceeded budget in poll\n", napi->poll);
 169         trace_napi_poll(napi, work, 0);
 170
 171         clear_bit(NAPI_STATE_NPSVC, &napi->state);
 172 }
 173
 174 static void poll_napi(struct net_device *dev)
 175 {
 176         struct napi_struct *napi;
 177         int cpu = smp_processor_id();
 178
 179         list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) {
 180                 if (cmpxchg(&napi->poll_owner, -1, cpu) == -1) {
 181                         poll_one_napi(napi);
 182                         smp_store_release(&napi->poll_owner, -1);
 183                 }
 184         }
 185 }
 186
 187 void netpoll_poll_dev(struct net_device *dev)
 188 {
 189         struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
 190         const struct net_device_ops *ops;
 191
 192         /* Don't do any rx activity if the dev_lock mutex is held
 193          * the dev_open/close paths use this to block netpoll activity
 194          * while changing device state
 195          */
 196         if (!ni || down_trylock(&ni->dev_lock))
 197                 return;
 198
 199         /* Some drivers will take the same locks in poll and xmit,
 200          * we can't poll if local CPU is already in xmit.
 201          */
 202         if (!netif_running(dev) || netif_local_xmit_active(dev)) {
 203                 up(&ni->dev_lock);
 204                 return;
 205         }
 206
 207         ops = dev->netdev_ops;
 208         if (ops->ndo_poll_controller)
 209                 ops->ndo_poll_controller(dev);
 210
 211         poll_napi(dev);
 212
 213         up(&ni->dev_lock);
 214
 215         zap_completion_queue();
 216 }
 217 EXPORT_SYMBOL(netpoll_poll_dev);
 218
 219 void netpoll_poll_disable(struct net_device *dev)
 220 {
 221         struct netpoll_info *ni;
 222         int idx;
 223         might_sleep();
 224         idx = srcu_read_lock(&netpoll_srcu);
 225         ni = srcu_dereference(dev->npinfo, &netpoll_srcu);
 226         if (ni)
 227                 down(&ni->dev_lock);
 228         srcu_read_unlock(&netpoll_srcu, idx);
 229 }
 230 EXPORT_SYMBOL(netpoll_poll_disable);
 231
 232 void netpoll_poll_enable(struct net_device *dev)
 233 {
 234         struct netpoll_info *ni;
 235         rcu_read_lock();
 236         ni = rcu_dereference(dev->npinfo);
 237         if (ni)
 238                 up(&ni->dev_lock);
 239         rcu_read_unlock();
 240 }
 241 EXPORT_SYMBOL(netpoll_poll_enable);
 242
 243 static void refill_skbs(void)
 244 {
 245         struct sk_buff *skb;
 246         unsigned long flags;
 247
 248         spin_lock_irqsave(&skb_pool.lock, flags);
 249         while (skb_pool.qlen < MAX_SKBS) {
 250                 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
 251                 if (!skb)
 252                         break;
 253
 254                 __skb_queue_tail(&skb_pool, skb);
 255         }
 256         spin_unlock_irqrestore(&skb_pool.lock, flags);
 257 }
 258
 259 static void zap_completion_queue(void)
 260 {
 261         unsigned long flags;
 262         struct softnet_data *sd = &get_cpu_var(softnet_data);
 263
 264         if (sd->completion_queue) {
 265                 struct sk_buff *clist;
 266
 267                 local_irq_save(flags);
 268                 clist = sd->completion_queue;
 269                 sd->completion_queue = NULL;
 270                 local_irq_restore(flags);
 271
 272                 while (clist != NULL) {
 273                         struct sk_buff *skb = clist;
 274                         clist = clist->next;
 275                         if (!skb_irq_freeable(skb)) {
 276                                 refcount_set(&skb->users, 1);
 277                                 dev_kfree_skb_any(skb); /* put this one back */
 278                         } else {
 279                                 __kfree_skb(skb);
 280                         }
 281                 }
 282         }
 283
 284         put_cpu_var(softnet_data);
 285 }
 286
 287 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
 288 {
 289         int count = 0;
 290         struct sk_buff *skb;
 291
 292         zap_completion_queue();
 293         refill_skbs();
 294 repeat:
 295
 296         skb = alloc_skb(len, GFP_ATOMIC);
 297         if (!skb)
 298                 skb = skb_dequeue(&skb_pool);
 299
 300         if (!skb) {
 301                 if (++count < 10) {
 302                         netpoll_poll_dev(np->dev);
 303                         goto repeat;
 304                 }
 305                 return NULL;
 306         }
 307
 308         refcount_set(&skb->users, 1);
 309         skb_reserve(skb, reserve);
 310         return skb;
 311 }
 312
 313 static int netpoll_owner_active(struct net_device *dev)
 314 {
 315         struct napi_struct *napi;
 316
 317         list_for_each_entry(napi, &dev->napi_list, dev_list) {
 318                 if (napi->poll_owner == smp_processor_id())
 319                         return 1;
 320         }
 321         return 0;
 322 }
 323
 324 /* call with IRQ disabled */
 325 void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb,
 326                              struct net_device *dev)
 327 {
 328         int status = NETDEV_TX_BUSY;
 329         unsigned long tries;
 330         /* It is up to the caller to keep npinfo alive. */
 331         struct netpoll_info *npinfo;
 332
 333         lockdep_assert_irqs_disabled();
 334
 335         npinfo = rcu_dereference_bh(np->dev->npinfo);
 336         if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
 337                 dev_kfree_skb_irq(skb);
 338                 return;
 339         }
 340
 341         /* don't get messages out of order, and no recursion */
 342         if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
 343                 struct netdev_queue *txq;
 344
 345                 txq = netdev_pick_tx(dev, skb, NULL);
 346
 347                 /* try until next clock tick */
 348                 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
 349                      tries > 0; --tries) {
 350                         if (HARD_TX_TRYLOCK(dev, txq)) {
 351                                 if (!netif_xmit_stopped(txq))
 352                                         status = netpoll_start_xmit(skb, dev, txq);
 353
 354                                 HARD_TX_UNLOCK(dev, txq);
 355
 356                                 if (dev_xmit_complete(status))
 357                                         break;
 358
 359                         }
 360
 361                         /* tickle device maybe there is some cleanup */
 362                         netpoll_poll_dev(np->dev);
 363
 364                         udelay(USEC_PER_POLL);
 365                 }
 366
 367                 WARN_ONCE(!irqs_disabled(),
 368                         "netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pF)\n",
 369                         dev->name, dev->netdev_ops->ndo_start_xmit);
 370
 371         }
 372
 373         if (!dev_xmit_complete(status)) {
 374                 skb_queue_tail(&npinfo->txq, skb);
 375                 schedule_delayed_work(&npinfo->tx_work,0);
 376         }
 377 }
 378 EXPORT_SYMBOL(netpoll_send_skb_on_dev);
 379
 380 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
 381 {
 382         int total_len, ip_len, udp_len;
 383         struct sk_buff *skb;
 384         struct udphdr *udph;
 385         struct iphdr *iph;
 386         struct ethhdr *eth;
 387         static atomic_t ip_ident;
 388         struct ipv6hdr *ip6h;
 389
 390         WARN_ON_ONCE(!irqs_disabled());
 391
 392         udp_len = len + sizeof(*udph);
 393         if (np->ipv6)
 394                 ip_len = udp_len + sizeof(*ip6h);
 395         else
 396                 ip_len = udp_len + sizeof(*iph);
 397
 398         total_len = ip_len + LL_RESERVED_SPACE(np->dev);
 399
 400         skb = find_skb(np, total_len + np->dev->needed_tailroom,
 401                        total_len - len);
 402         if (!skb)
 403                 return;
 404
 405         skb_copy_to_linear_data(skb, msg, len);
 406         skb_put(skb, len);
 407
 408         skb_push(skb, sizeof(*udph));
 409         skb_reset_transport_header(skb);
 410         udph = udp_hdr(skb);
 411         udph->source = htons(np->local_port);
 412         udph->dest = htons(np->remote_port);
 413         udph->len = htons(udp_len);
 414
 415         if (np->ipv6) {
 416                 udph->check = 0;
 417                 udph->check = csum_ipv6_magic(&np->local_ip.in6,
 418                                               &np->remote_ip.in6,
 419                                               udp_len, IPPROTO_UDP,
 420                                               csum_partial(udph, udp_len, 0));
 421                 if (udph->check == 0)
 422                         udph->check = CSUM_MANGLED_0;
 423
 424                 skb_push(skb, sizeof(*ip6h));
 425                 skb_reset_network_header(skb);
 426                 ip6h = ipv6_hdr(skb);
 427
 428                 /* ip6h->version = 6; ip6h->priority = 0; */
 429                 put_unaligned(0x60, (unsigned char *)ip6h);
 430                 ip6h->flow_lbl[0] = 0;
 431                 ip6h->flow_lbl[1] = 0;
 432                 ip6h->flow_lbl[2] = 0;
 433
 434                 ip6h->payload_len = htons(sizeof(struct udphdr) + len);
 435                 ip6h->nexthdr = IPPROTO_UDP;
 436                 ip6h->hop_limit = 32;
 437                 ip6h->saddr = np->local_ip.in6;
 438                 ip6h->daddr = np->remote_ip.in6;
 439
 440                 eth = skb_push(skb, ETH_HLEN);
 441                 skb_reset_mac_header(skb);
 442                 skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
 443         } else {
 444                 udph->check = 0;
 445                 udph->check = csum_tcpudp_magic(np->local_ip.ip,
 446                                                 np->remote_ip.ip,
 447                                                 udp_len, IPPROTO_UDP,
 448                                                 csum_partial(udph, udp_len, 0));
 449                 if (udph->check == 0)
 450                         udph->check = CSUM_MANGLED_0;
 451
 452                 skb_push(skb, sizeof(*iph));
 453                 skb_reset_network_header(skb);
 454                 iph = ip_hdr(skb);
 455
 456                 /* iph->version = 4; iph->ihl = 5; */
 457                 put_unaligned(0x45, (unsigned char *)iph);
 458                 iph->tos      = 0;
 459                 put_unaligned(htons(ip_len), &(iph->tot_len));
 460                 iph->id       = htons(atomic_inc_return(&ip_ident));
 461                 iph->frag_off = 0;
 462                 iph->ttl      = 64;
 463                 iph->protocol = IPPROTO_UDP;
 464                 iph->check    = 0;
 465                 put_unaligned(np->local_ip.ip, &(iph->saddr));
 466                 put_unaligned(np->remote_ip.ip, &(iph->daddr));
 467                 iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
 468
 469                 eth = skb_push(skb, ETH_HLEN);
 470                 skb_reset_mac_header(skb);
 471                 skb->protocol = eth->h_proto = htons(ETH_P_IP);
 472         }
 473
 474         ether_addr_copy(eth->h_source, np->dev->dev_addr);
 475         ether_addr_copy(eth->h_dest, np->remote_mac);
 476
 477         skb->dev = np->dev;
 478
 479         netpoll_send_skb(np, skb);
 480 }
 481 EXPORT_SYMBOL(netpoll_send_udp);
 482
 483 void netpoll_print_options(struct netpoll *np)
 484 {
 485         np_info(np, "local port %d\n", np->local_port);
 486         if (np->ipv6)
 487                 np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
 488         else
 489                 np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
 490         np_info(np, "interface '%s'\n", np->dev_name);
 491         np_info(np, "remote port %d\n", np->remote_port);
 492         if (np->ipv6)
 493                 np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
 494         else
 495                 np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
 496         np_info(np, "remote ethernet address %pM\n", np->remote_mac);
 497 }
 498 EXPORT_SYMBOL(netpoll_print_options);
 499
 500 static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
 501 {
 502         const char *end;
 503
 504         if (!strchr(str, ':') &&
 505             in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
 506                 if (!*end)
 507                         return 0;
 508         }
 509         if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
 510 #if IS_ENABLED(CONFIG_IPV6)
 511                 if (!*end)
 512                         return 1;
 513 #else
 514                 return -1;
 515 #endif
 516         }
 517         return -1;
 518 }
 519
 520 int netpoll_parse_options(struct netpoll *np, char *opt)
 521 {
 522         char *cur=opt, *delim;
 523         int ipv6;
 524         bool ipversion_set = false;
 525
 526         if (*cur != '@') {
 527                 if ((delim = strchr(cur, '@')) == NULL)
 528                         goto parse_failed;
 529                 *delim = 0;
 530                 if (kstrtou16(cur, 10, &np->local_port))
 531                         goto parse_failed;
 532                 cur = delim;
 533         }
 534         cur++;
 535
 536         if (*cur != '/') {
 537                 ipversion_set = true;
 538                 if ((delim = strchr(cur, '/')) == NULL)
 539                         goto parse_failed;
 540                 *delim = 0;
 541                 ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
 542                 if (ipv6 < 0)
 543                         goto parse_failed;
 544                 else
 545                         np->ipv6 = (bool)ipv6;
 546                 cur = delim;
 547         }
 548         cur++;
 549
 550         if (*cur != ',') {
 551                 /* parse out dev name */
 552                 if ((delim = strchr(cur, ',')) == NULL)
 553                         goto parse_failed;
 554                 *delim = 0;
 555                 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
 556                 cur = delim;
 557         }
 558         cur++;
 559
 560         if (*cur != '@') {
 561                 /* dst port */
 562                 if ((delim = strchr(cur, '@')) == NULL)
 563                         goto parse_failed;
 564                 *delim = 0;
 565                 if (*cur == ' ' || *cur == '\t')
 566                         np_info(np, "warning: whitespace is not allowed\n");
 567                 if (kstrtou16(cur, 10, &np->remote_port))
 568                         goto parse_failed;
 569                 cur = delim;
 570         }
 571         cur++;
 572
 573         /* dst ip */
 574         if ((delim = strchr(cur, '/')) == NULL)
 575                 goto parse_failed;
 576         *delim = 0;
 577         ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
 578         if (ipv6 < 0)
 579                 goto parse_failed;
 580         else if (ipversion_set && np->ipv6 != (bool)ipv6)
 581                 goto parse_failed;
 582         else
 583                 np->ipv6 = (bool)ipv6;
 584         cur = delim + 1;
 585
 586         if (*cur != 0) {
 587                 /* MAC address */
 588                 if (!mac_pton(cur, np->remote_mac))
 589                         goto parse_failed;
 590         }
 591
 592         netpoll_print_options(np);
 593
 594         return 0;
 595
 596  parse_failed:
 597         np_info(np, "couldn't parse config at '%s'!\n", cur);
 598         return -1;
 599 }
 600 EXPORT_SYMBOL(netpoll_parse_options);
 601
 602 int __netpoll_setup(struct netpoll *np, struct net_device *ndev)
 603 {
 604         struct netpoll_info *npinfo;
 605         const struct net_device_ops *ops;
 606         int err;
 607
 608         np->dev = ndev;
 609         strlcpy(np->dev_name, ndev->name, IFNAMSIZ);
 610         INIT_WORK(&np->cleanup_work, netpoll_async_cleanup);
 611
 612         if (ndev->priv_flags & IFF_DISABLE_NETPOLL) {
 613                 np_err(np, "%s doesn't support polling, aborting\n",
 614                        np->dev_name);
 615                 err = -ENOTSUPP;
 616                 goto out;
 617         }
 618
 619         if (!ndev->npinfo) {
 620                 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
 621                 if (!npinfo) {
 622                         err = -ENOMEM;
 623                         goto out;
 624                 }
 625
 626                 sema_init(&npinfo->dev_lock, 1);
 627                 skb_queue_head_init(&npinfo->txq);
 628                 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
 629
 630                 refcount_set(&npinfo->refcnt, 1);
 631
 632                 ops = np->dev->netdev_ops;
 633                 if (ops->ndo_netpoll_setup) {
 634                         err = ops->ndo_netpoll_setup(ndev, npinfo);
 635                         if (err)
 636                                 goto free_npinfo;
 637                 }
 638         } else {
 639                 npinfo = rtnl_dereference(ndev->npinfo);
 640                 refcount_inc(&npinfo->refcnt);
 641         }
 642
 643         npinfo->netpoll = np;
 644
 645         /* last thing to do is link it to the net device structure */
 646         rcu_assign_pointer(ndev->npinfo, npinfo);
 647
 648         return 0;
 649
 650 free_npinfo:
 651         kfree(npinfo);
 652 out:
 653         return err;
 654 }
 655 EXPORT_SYMBOL_GPL(__netpoll_setup);
 656
 657 int netpoll_setup(struct netpoll *np)
 658 {
 659         struct net_device *ndev = NULL, *dev = NULL;
 660         struct net *net = current->nsproxy->net_ns;
 661         struct in_device *in_dev;
 662         int err;
 663
 664         rtnl_lock();
 665         if (np->dev_name[0])
 666                 ndev = __dev_get_by_name(net, np->dev_name);
 667
 668         if (!ndev) {
 669                 np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
 670                 err = -ENODEV;
 671                 goto unlock;
 672         }
 673         dev_hold(ndev);
 674
 675         /* bring up DSA management network devices up first */
 676         for_each_netdev(net, dev) {
 677                 if (!netdev_uses_dsa(dev))
 678                         continue;
 679
 680                 err = dev_change_flags(dev, dev->flags | IFF_UP);
 681                 if (err < 0) {
 682                         np_err(np, "%s failed to open %s\n",
 683                                np->dev_name, dev->name);
 684                         goto put;
 685                 }
 686         }
 687
 688         if (netdev_master_upper_dev_get(ndev)) {
 689                 np_err(np, "%s is a slave device, aborting\n", np->dev_name);
 690                 err = -EBUSY;
 691                 goto put;
 692         }
 693
 694         if (!netif_running(ndev)) {
 695                 unsigned long atmost, atleast;
 696
 697                 np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
 698
 699                 err = dev_open(ndev);
 700
 701                 if (err) {
 702                         np_err(np, "failed to open %s\n", ndev->name);
 703                         goto put;
 704                 }
 705
 706                 rtnl_unlock();
 707                 atleast = jiffies + HZ/10;
 708                 atmost = jiffies + carrier_timeout * HZ;
 709                 while (!netif_carrier_ok(ndev)) {
 710                         if (time_after(jiffies, atmost)) {
 711                                 np_notice(np, "timeout waiting for carrier\n");
 712                                 break;
 713                         }
 714                         msleep(1);
 715                 }
 716
 717                 /* If carrier appears to come up instantly, we don't
 718                  * trust it and pause so that we don't pump all our
 719                  * queued console messages into the bitbucket.
 720                  */
 721
 722                 if (time_before(jiffies, atleast)) {
 723                         np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
 724                         msleep(4000);
 725                 }
 726                 rtnl_lock();
 727         }
 728
 729         if (!np->local_ip.ip) {
 730                 if (!np->ipv6) {
 731                         in_dev = __in_dev_get_rtnl(ndev);
 732
 733                         if (!in_dev || !in_dev->ifa_list) {
 734                                 np_err(np, "no IP address for %s, aborting\n",
 735                                        np->dev_name);
 736                                 err = -EDESTADDRREQ;
 737                                 goto put;
 738                         }
 739
 740                         np->local_ip.ip = in_dev->ifa_list->ifa_local;
 741                         np_info(np, "local IP %pI4\n", &np->local_ip.ip);
 742                 } else {
 743 #if IS_ENABLED(CONFIG_IPV6)
 744                         struct inet6_dev *idev;
 745
 746                         err = -EDESTADDRREQ;
 747                         idev = __in6_dev_get(ndev);
 748                         if (idev) {
 749                                 struct inet6_ifaddr *ifp;
 750
 751                                 read_lock_bh(&idev->lock);
 752                                 list_for_each_entry(ifp, &idev->addr_list, if_list) {
 753                                         if (ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL)
 754                                                 continue;
 755                                         np->local_ip.in6 = ifp->addr;
 756                                         err = 0;
 757                                         break;
 758                                 }
 759                                 read_unlock_bh(&idev->lock);
 760                         }
 761                         if (err) {
 762                                 np_err(np, "no IPv6 address for %s, aborting\n",
 763                                        np->dev_name);
 764                                 goto put;
 765                         } else
 766                                 np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
 767 #else
 768                         np_err(np, "IPv6 is not supported %s, aborting\n",
 769                                np->dev_name);
 770                         err = -EINVAL;
 771                         goto put;
 772 #endif
 773                 }
 774         }
 775
 776         /* fill up the skb queue */
 777         refill_skbs();
 778
 779         err = __netpoll_setup(np, ndev);
 780         if (err)
 781                 goto put;
 782
 783         rtnl_unlock();
 784         return 0;
 785
 786 put:
 787         dev_put(ndev);
 788 unlock:
 789         rtnl_unlock();
 790         return err;
 791 }
 792 EXPORT_SYMBOL(netpoll_setup);
 793
 794 static int __init netpoll_init(void)
 795 {
 796         skb_queue_head_init(&skb_pool);
 797         return 0;
 798 }
 799 core_initcall(netpoll_init);
 800
 801 static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
 802 {
 803         struct netpoll_info *npinfo =
 804                         container_of(rcu_head, struct netpoll_info, rcu);
 805
 806         skb_queue_purge(&npinfo->txq);
 807
 808         /* we can't call cancel_delayed_work_sync here, as we are in softirq */
 809         cancel_delayed_work(&npinfo->tx_work);
 810
 811         /* clean after last, unfinished work */
 812         __skb_queue_purge(&npinfo->txq);
 813         /* now cancel it again */
 814         cancel_delayed_work(&npinfo->tx_work);
 815         kfree(npinfo);
 816 }
 817
 818 void __netpoll_cleanup(struct netpoll *np)
 819 {
 820         struct netpoll_info *npinfo;
 821
 822         /* rtnl_dereference would be preferable here but
 823          * rcu_cleanup_netpoll path can put us in here safely without
 824          * holding the rtnl, so plain rcu_dereference it is
 825          */
 826         npinfo = rtnl_dereference(np->dev->npinfo);
 827         if (!npinfo)
 828                 return;
 829
 830         synchronize_srcu(&netpoll_srcu);
 831
 832         if (refcount_dec_and_test(&npinfo->refcnt)) {
 833                 const struct net_device_ops *ops;
 834
 835                 ops = np->dev->netdev_ops;
 836                 if (ops->ndo_netpoll_cleanup)
 837                         ops->ndo_netpoll_cleanup(np->dev);
 838
 839                 RCU_INIT_POINTER(np->dev->npinfo, NULL);
 840                 call_rcu_bh(&npinfo->rcu, rcu_cleanup_netpoll_info);
 841         } else
 842                 RCU_INIT_POINTER(np->dev->npinfo, NULL);
 843 }
 844 EXPORT_SYMBOL_GPL(__netpoll_cleanup);
 845
 846 static void netpoll_async_cleanup(struct work_struct *work)
 847 {
 848         struct netpoll *np = container_of(work, struct netpoll, cleanup_work);
 849
 850         rtnl_lock();
 851         __netpoll_cleanup(np);
 852         rtnl_unlock();
 853         kfree(np);
 854 }
 855
 856 void __netpoll_free_async(struct netpoll *np)
 857 {
 858         schedule_work(&np->cleanup_work);
 859 }
 860 EXPORT_SYMBOL_GPL(__netpoll_free_async);
 861
 862 void netpoll_cleanup(struct netpoll *np)
 863 {
 864         rtnl_lock();
 865         if (!np->dev)
 866                 goto out;
 867         __netpoll_cleanup(np);
 868         dev_put(np->dev);
 869         np->dev = NULL;
 870 out:
 871         rtnl_unlock();
 872 }
 873 EXPORT_SYMBOL(netpoll_cleanup);