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