net/core/netpoll.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Common framework for low-level network console, dump, and debugger code
   4  *
   5  * Sep 8 2003  Matt Mackall <mpm@selenic.com>
   6  *
   7  * based on the netconsole code from:
   8  *
   9  * Copyright (C) 2001  Ingo Molnar <mingo@redhat.com>
  10  * Copyright (C) 2002  Red Hat, Inc.
  11  */
  12
  13 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  14
  15 #include <linux/moduleparam.h>
  16 #include <linux/kernel.h>
  17 #include <linux/netdevice.h>
  18 #include <linux/etherdevice.h>
  19 #include <linux/string.h>
  20 #include <linux/if_arp.h>
  21 #include <linux/inetdevice.h>
  22 #include <linux/inet.h>
  23 #include <linux/interrupt.h>
  24 #include <linux/netpoll.h>
  25 #include <linux/sched.h>
  26 #include <linux/delay.h>
  27 #include <linux/rcupdate.h>
  28 #include <linux/workqueue.h>
  29 #include <linux/slab.h>
  30 #include <linux/export.h>
  31 #include <linux/if_vlan.h>
  32 #include <net/dsa.h>
  33 #include <net/tcp.h>
  34 #include <net/udp.h>
  35 #include <net/addrconf.h>
  36 #include <net/ndisc.h>
  37 #include <net/ip6_checksum.h>
  38 #include <asm/unaligned.h>
  39 #include <trace/events/napi.h>
  40
  41 /*
  42  * We maintain a small pool of fully-sized skbs, to make sure the
  43  * message gets out even in extreme OOM situations.
  44  */
  45
  46 #define MAX_UDP_CHUNK 1460
  47 #define MAX_SKBS 32
  48
  49 static struct sk_buff_head skb_pool;
  50
  51 DEFINE_STATIC_SRCU(netpoll_srcu);
  52
  53 #define USEC_PER_POLL   50
  54
  55 #define MAX_SKB_SIZE                                                    \
  56         (sizeof(struct ethhdr) +                                        \
  57          sizeof(struct iphdr) +                                         \
  58          sizeof(struct udphdr) +                                        \
  59          MAX_UDP_CHUNK)
  60
  61 static void zap_completion_queue(void);
  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 netdev_tx_t netpoll_start_xmit(struct sk_buff *skb,
  74                                       struct net_device *dev,
  75                                       struct netdev_queue *txq)
  76 {
  77         netdev_tx_t status = NETDEV_TX_OK;
  78         netdev_features_t features;
  79
  80         features = netif_skb_features(skb);
  81
  82         if (skb_vlan_tag_present(skb) &&
  83             !vlan_hw_offload_capable(features, skb->vlan_proto)) {
  84                 skb = __vlan_hwaccel_push_inside(skb);
  85                 if (unlikely(!skb)) {
  86                         /* This is actually a packet drop, but we
  87                          * don't want the code that calls this
  88                          * function to try and operate on a NULL skb.
  89                          */
  90                         goto out;
  91                 }
  92         }
  93
  94         status = netdev_start_xmit(skb, dev, txq, false);
  95
  96 out:
  97         return status;
  98 }
  99
 100 static void queue_process(struct work_struct *work)
 101 {
 102         struct netpoll_info *npinfo =
 103                 container_of(work, struct netpoll_info, tx_work.work);
 104         struct sk_buff *skb;
 105         unsigned long flags;
 106
 107         while ((skb = skb_dequeue(&npinfo->txq))) {
 108                 struct net_device *dev = skb->dev;
 109                 struct netdev_queue *txq;
 110                 unsigned int q_index;
 111
 112                 if (!netif_device_present(dev) || !netif_running(dev)) {
 113                         kfree_skb(skb);
 114                         continue;
 115                 }
 116
 117                 local_irq_save(flags);
 118                 /* check if skb->queue_mapping is still valid */
 119                 q_index = skb_get_queue_mapping(skb);
 120                 if (unlikely(q_index >= dev->real_num_tx_queues)) {
 121                         q_index = q_index % dev->real_num_tx_queues;
 122                         skb_set_queue_mapping(skb, q_index);
 123                 }
 124                 txq = netdev_get_tx_queue(dev, q_index);
 125                 HARD_TX_LOCK(dev, txq, smp_processor_id());
 126                 if (netif_xmit_frozen_or_stopped(txq) ||
 127                     !dev_xmit_complete(netpoll_start_xmit(skb, dev, txq))) {
 128                         skb_queue_head(&npinfo->txq, skb);
 129                         HARD_TX_UNLOCK(dev, txq);
 130                         local_irq_restore(flags);
 131
 132                         schedule_delayed_work(&npinfo->tx_work, HZ/10);
 133                         return;
 134                 }
 135                 HARD_TX_UNLOCK(dev, txq);
 136                 local_irq_restore(flags);
 137         }
 138 }
 139
 140 static void poll_one_napi(struct napi_struct *napi)
 141 {
 142         int work;
 143
 144         /* If we set this bit but see that it has already been set,
 145          * that indicates that napi has been disabled and we need
 146          * to abort this operation
 147          */
 148         if (test_and_set_bit(NAPI_STATE_NPSVC, &napi->state))
 149                 return;
 150
 151         /* We explicilty pass the polling call a budget of 0 to
 152          * indicate that we are clearing the Tx path only.
 153          */
 154         work = napi->poll(napi, 0);
 155         WARN_ONCE(work, "%pS exceeded budget in poll\n", napi->poll);
 156         trace_napi_poll(napi, work, 0);
 157
 158         clear_bit(NAPI_STATE_NPSVC, &napi->state);
 159 }
 160
 161 static void poll_napi(struct net_device *dev)
 162 {
 163         struct napi_struct *napi;
 164         int cpu = smp_processor_id();
 165
 166         list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) {
 167                 if (cmpxchg(&napi->poll_owner, -1, cpu) == -1) {
 168                         poll_one_napi(napi);
 169                         smp_store_release(&napi->poll_owner, -1);
 170                 }
 171         }
 172 }
 173
 174 void netpoll_poll_dev(struct net_device *dev)
 175 {
 176         struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
 177         const struct net_device_ops *ops;
 178
 179         /* Don't do any rx activity if the dev_lock mutex is held
 180          * the dev_open/close paths use this to block netpoll activity
 181          * while changing device state
 182          */
 183         if (!ni || down_trylock(&ni->dev_lock))
 184                 return;
 185
 186         if (!netif_running(dev)) {
 187                 up(&ni->dev_lock);
 188                 return;
 189         }
 190
 191         ops = dev->netdev_ops;
 192         if (ops->ndo_poll_controller)
 193                 ops->ndo_poll_controller(dev);
 194
 195         poll_napi(dev);
 196
 197         up(&ni->dev_lock);
 198
 199         zap_completion_queue();
 200 }
 201 EXPORT_SYMBOL(netpoll_poll_dev);
 202
 203 void netpoll_poll_disable(struct net_device *dev)
 204 {
 205         struct netpoll_info *ni;
 206         int idx;
 207         might_sleep();
 208         idx = srcu_read_lock(&netpoll_srcu);
 209         ni = srcu_dereference(dev->npinfo, &netpoll_srcu);
 210         if (ni)
 211                 down(&ni->dev_lock);
 212         srcu_read_unlock(&netpoll_srcu, idx);
 213 }
 214 EXPORT_SYMBOL(netpoll_poll_disable);
 215
 216 void netpoll_poll_enable(struct net_device *dev)
 217 {
 218         struct netpoll_info *ni;
 219         rcu_read_lock();
 220         ni = rcu_dereference(dev->npinfo);
 221         if (ni)
 222                 up(&ni->dev_lock);
 223         rcu_read_unlock();
 224 }
 225 EXPORT_SYMBOL(netpoll_poll_enable);
 226
 227 static void refill_skbs(void)
 228 {
 229         struct sk_buff *skb;
 230         unsigned long flags;
 231
 232         spin_lock_irqsave(&skb_pool.lock, flags);
 233         while (skb_pool.qlen < MAX_SKBS) {
 234                 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
 235                 if (!skb)
 236                         break;
 237
 238                 __skb_queue_tail(&skb_pool, skb);
 239         }
 240         spin_unlock_irqrestore(&skb_pool.lock, flags);
 241 }
 242
 243 static void zap_completion_queue(void)
 244 {
 245         unsigned long flags;
 246         struct softnet_data *sd = &get_cpu_var(softnet_data);
 247
 248         if (sd->completion_queue) {
 249                 struct sk_buff *clist;
 250
 251                 local_irq_save(flags);
 252                 clist = sd->completion_queue;
 253                 sd->completion_queue = NULL;
 254                 local_irq_restore(flags);
 255
 256                 while (clist != NULL) {
 257                         struct sk_buff *skb = clist;
 258                         clist = clist->next;
 259                         if (!skb_irq_freeable(skb)) {
 260                                 refcount_set(&skb->users, 1);
 261                                 dev_kfree_skb_any(skb); /* put this one back */
 262                         } else {
 263                                 __kfree_skb(skb);
 264                         }
 265                 }
 266         }
 267
 268         put_cpu_var(softnet_data);
 269 }
 270
 271 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
 272 {
 273         int count = 0;
 274         struct sk_buff *skb;
 275
 276         zap_completion_queue();
 277         refill_skbs();
 278 repeat:
 279
 280         skb = alloc_skb(len, GFP_ATOMIC);
 281         if (!skb)
 282                 skb = skb_dequeue(&skb_pool);
 283
 284         if (!skb) {
 285                 if (++count < 10) {
 286                         netpoll_poll_dev(np->dev);
 287                         goto repeat;
 288                 }
 289                 return NULL;
 290         }
 291
 292         refcount_set(&skb->users, 1);
 293         skb_reserve(skb, reserve);
 294         return skb;
 295 }
 296
 297 static int netpoll_owner_active(struct net_device *dev)
 298 {
 299         struct napi_struct *napi;
 300
 301         list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) {
 302                 if (napi->poll_owner == smp_processor_id())
 303                         return 1;
 304         }
 305         return 0;
 306 }
 307
 308 /* call with IRQ disabled */
 309 static netdev_tx_t __netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
 310 {
 311         netdev_tx_t status = NETDEV_TX_BUSY;
 312         struct net_device *dev;
 313         unsigned long tries;
 314         /* It is up to the caller to keep npinfo alive. */
 315         struct netpoll_info *npinfo;
 316
 317         lockdep_assert_irqs_disabled();
 318
 319         dev = np->dev;
 320         npinfo = rcu_dereference_bh(dev->npinfo);
 321
 322         if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
 323                 dev_kfree_skb_irq(skb);
 324                 return NET_XMIT_DROP;
 325         }
 326
 327         /* don't get messages out of order, and no recursion */
 328         if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
 329                 struct netdev_queue *txq;
 330
 331                 txq = netdev_core_pick_tx(dev, skb, NULL);
 332
 333                 /* try until next clock tick */
 334                 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
 335                      tries > 0; --tries) {
 336                         if (HARD_TX_TRYLOCK(dev, txq)) {
 337                                 if (!netif_xmit_stopped(txq))
 338                                         status = netpoll_start_xmit(skb, dev, txq);
 339
 340                                 HARD_TX_UNLOCK(dev, txq);
 341
 342                                 if (dev_xmit_complete(status))
 343                                         break;
 344
 345                         }
 346
 347                         /* tickle device maybe there is some cleanup */
 348                         netpoll_poll_dev(np->dev);
 349
 350                         udelay(USEC_PER_POLL);
 351                 }
 352
 353                 WARN_ONCE(!irqs_disabled(),
 354                         "netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pS)\n",
 355                         dev->name, dev->netdev_ops->ndo_start_xmit);
 356
 357         }
 358
 359         if (!dev_xmit_complete(status)) {
 360                 skb_queue_tail(&npinfo->txq, skb);
 361                 schedule_delayed_work(&npinfo->tx_work,0);
 362         }
 363         return NETDEV_TX_OK;
 364 }
 365
 366 netdev_tx_t netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
 367 {
 368         unsigned long flags;
 369         netdev_tx_t ret;
 370
 371         if (unlikely(!np)) {
 372                 dev_kfree_skb_irq(skb);
 373                 ret = NET_XMIT_DROP;
 374         } else {
 375                 local_irq_save(flags);
 376                 ret = __netpoll_send_skb(np, skb);
 377                 local_irq_restore(flags);
 378         }
 379         return ret;
 380 }
 381 EXPORT_SYMBOL(netpoll_send_skb);
 382
 383 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
 384 {
 385         int total_len, ip_len, udp_len;
 386         struct sk_buff *skb;
 387         struct udphdr *udph;
 388         struct iphdr *iph;
 389         struct ethhdr *eth;
 390         static atomic_t ip_ident;
 391         struct ipv6hdr *ip6h;
 392
 393         WARN_ON_ONCE(!irqs_disabled());
 394
 395         udp_len = len + sizeof(*udph);
 396         if (np->ipv6)
 397                 ip_len = udp_len + sizeof(*ip6h);
 398         else
 399                 ip_len = udp_len + sizeof(*iph);
 400
 401         total_len = ip_len + LL_RESERVED_SPACE(np->dev);
 402
 403         skb = find_skb(np, total_len + np->dev->needed_tailroom,
 404                        total_len - len);
 405         if (!skb)
 406                 return;
 407
 408         skb_copy_to_linear_data(skb, msg, len);
 409         skb_put(skb, len);
 410
 411         skb_push(skb, sizeof(*udph));
 412         skb_reset_transport_header(skb);
 413         udph = udp_hdr(skb);
 414         udph->source = htons(np->local_port);
 415         udph->dest = htons(np->remote_port);
 416         udph->len = htons(udp_len);
 417
 418         if (np->ipv6) {
 419                 udph->check = 0;
 420                 udph->check = csum_ipv6_magic(&np->local_ip.in6,
 421                                               &np->remote_ip.in6,
 422                                               udp_len, IPPROTO_UDP,
 423                                               csum_partial(udph, udp_len, 0));
 424                 if (udph->check == 0)
 425                         udph->check = CSUM_MANGLED_0;
 426
 427                 skb_push(skb, sizeof(*ip6h));
 428                 skb_reset_network_header(skb);
 429                 ip6h = ipv6_hdr(skb);
 430
 431                 /* ip6h->version = 6; ip6h->priority = 0; */
 432                 put_unaligned(0x60, (unsigned char *)ip6h);
 433                 ip6h->flow_lbl[0] = 0;
 434                 ip6h->flow_lbl[1] = 0;
 435                 ip6h->flow_lbl[2] = 0;
 436
 437                 ip6h->payload_len = htons(sizeof(struct udphdr) + len);
 438                 ip6h->nexthdr = IPPROTO_UDP;
 439                 ip6h->hop_limit = 32;
 440                 ip6h->saddr = np->local_ip.in6;
 441                 ip6h->daddr = np->remote_ip.in6;
 442
 443                 eth = skb_push(skb, ETH_HLEN);
 444                 skb_reset_mac_header(skb);
 445                 skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
 446         } else {
 447                 udph->check = 0;
 448                 udph->check = csum_tcpudp_magic(np->local_ip.ip,
 449                                                 np->remote_ip.ip,
 450                                                 udp_len, IPPROTO_UDP,
 451                                                 csum_partial(udph, udp_len, 0));
 452                 if (udph->check == 0)
 453                         udph->check = CSUM_MANGLED_0;
 454
 455                 skb_push(skb, sizeof(*iph));
 456                 skb_reset_network_header(skb);
 457                 iph = ip_hdr(skb);
 458
 459                 /* iph->version = 4; iph->ihl = 5; */
 460                 put_unaligned(0x45, (unsigned char *)iph);
 461                 iph->tos      = 0;
 462                 put_unaligned(htons(ip_len), &(iph->tot_len));
 463                 iph->id       = htons(atomic_inc_return(&ip_ident));
 464                 iph->frag_off = 0;
 465                 iph->ttl      = 64;
 466                 iph->protocol = IPPROTO_UDP;
 467                 iph->check    = 0;
 468                 put_unaligned(np->local_ip.ip, &(iph->saddr));
 469                 put_unaligned(np->remote_ip.ip, &(iph->daddr));
 470                 iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
 471
 472                 eth = skb_push(skb, ETH_HLEN);
 473                 skb_reset_mac_header(skb);
 474                 skb->protocol = eth->h_proto = htons(ETH_P_IP);
 475         }
 476
 477         ether_addr_copy(eth->h_source, np->dev->dev_addr);
 478         ether_addr_copy(eth->h_dest, np->remote_mac);
 479
 480         skb->dev = np->dev;
 481
 482         netpoll_send_skb(np, skb);
 483 }
 484 EXPORT_SYMBOL(netpoll_send_udp);
 485
 486 void netpoll_print_options(struct netpoll *np)
 487 {
 488         np_info(np, "local port %d\n", np->local_port);
 489         if (np->ipv6)
 490                 np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
 491         else
 492                 np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
 493         np_info(np, "interface '%s'\n", np->dev_name);
 494         np_info(np, "remote port %d\n", np->remote_port);
 495         if (np->ipv6)
 496                 np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
 497         else
 498                 np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
 499         np_info(np, "remote ethernet address %pM\n", np->remote_mac);
 500 }
 501 EXPORT_SYMBOL(netpoll_print_options);
 502
 503 static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
 504 {
 505         const char *end;
 506
 507         if (!strchr(str, ':') &&
 508             in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
 509                 if (!*end)
 510                         return 0;
 511         }
 512         if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
 513 #if IS_ENABLED(CONFIG_IPV6)
 514                 if (!*end)
 515                         return 1;
 516 #else
 517                 return -1;
 518 #endif
 519         }
 520         return -1;
 521 }
 522
 523 int netpoll_parse_options(struct netpoll *np, char *opt)
 524 {
 525         char *cur=opt, *delim;
 526         int ipv6;
 527         bool ipversion_set = false;
 528
 529         if (*cur != '@') {
 530                 if ((delim = strchr(cur, '@')) == NULL)
 531                         goto parse_failed;
 532                 *delim = 0;
 533                 if (kstrtou16(cur, 10, &np->local_port))
 534                         goto parse_failed;
 535                 cur = delim;
 536         }
 537         cur++;
 538
 539         if (*cur != '/') {
 540                 ipversion_set = true;
 541                 if ((delim = strchr(cur, '/')) == NULL)
 542                         goto parse_failed;
 543                 *delim = 0;
 544                 ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
 545                 if (ipv6 < 0)
 546                         goto parse_failed;
 547                 else
 548                         np->ipv6 = (bool)ipv6;
 549                 cur = delim;
 550         }
 551         cur++;
 552
 553         if (*cur != ',') {
 554                 /* parse out dev name */
 555                 if ((delim = strchr(cur, ',')) == NULL)
 556                         goto parse_failed;
 557                 *delim = 0;
 558                 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
 559                 cur = delim;
 560         }
 561         cur++;
 562
 563         if (*cur != '@') {
 564                 /* dst port */
 565                 if ((delim = strchr(cur, '@')) == NULL)
 566                         goto parse_failed;
 567                 *delim = 0;
 568                 if (*cur == ' ' || *cur == '\t')
 569                         np_info(np, "warning: whitespace is not allowed\n");
 570                 if (kstrtou16(cur, 10, &np->remote_port))
 571                         goto parse_failed;
 572                 cur = delim;
 573         }
 574         cur++;
 575
 576         /* dst ip */
 577         if ((delim = strchr(cur, '/')) == NULL)
 578                 goto parse_failed;
 579         *delim = 0;
 580         ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
 581         if (ipv6 < 0)
 582                 goto parse_failed;
 583         else if (ipversion_set && np->ipv6 != (bool)ipv6)
 584                 goto parse_failed;
 585         else
 586                 np->ipv6 = (bool)ipv6;
 587         cur = delim + 1;
 588
 589         if (*cur != 0) {
 590                 /* MAC address */
 591                 if (!mac_pton(cur, np->remote_mac))
 592                         goto parse_failed;
 593         }
 594
 595         netpoll_print_options(np);
 596
 597         return 0;
 598
 599  parse_failed:
 600         np_info(np, "couldn't parse config at '%s'!\n", cur);
 601         return -1;
 602 }
 603 EXPORT_SYMBOL(netpoll_parse_options);
 604
 605 int __netpoll_setup(struct netpoll *np, struct net_device *ndev)
 606 {
 607         struct netpoll_info *npinfo;
 608         const struct net_device_ops *ops;
 609         int err;
 610
 611         np->dev = ndev;
 612         strlcpy(np->dev_name, ndev->name, IFNAMSIZ);
 613
 614         if (ndev->priv_flags & IFF_DISABLE_NETPOLL) {
 615                 np_err(np, "%s doesn't support polling, aborting\n",
 616                        np->dev_name);
 617                 err = -ENOTSUPP;
 618                 goto out;
 619         }
 620
 621         if (!ndev->npinfo) {
 622                 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
 623                 if (!npinfo) {
 624                         err = -ENOMEM;
 625                         goto out;
 626                 }
 627
 628                 sema_init(&npinfo->dev_lock, 1);
 629                 skb_queue_head_init(&npinfo->txq);
 630                 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
 631
 632                 refcount_set(&npinfo->refcnt, 1);
 633
 634                 ops = np->dev->netdev_ops;
 635                 if (ops->ndo_netpoll_setup) {
 636                         err = ops->ndo_netpoll_setup(ndev, npinfo);
 637                         if (err)
 638                                 goto free_npinfo;
 639                 }
 640         } else {
 641                 npinfo = rtnl_dereference(ndev->npinfo);
 642                 refcount_inc(&npinfo->refcnt);
 643         }
 644
 645         npinfo->netpoll = np;
 646
 647         /* last thing to do is link it to the net device structure */
 648         rcu_assign_pointer(ndev->npinfo, npinfo);
 649
 650         return 0;
 651
 652 free_npinfo:
 653         kfree(npinfo);
 654 out:
 655         return err;
 656 }
 657 EXPORT_SYMBOL_GPL(__netpoll_setup);
 658
 659 int netpoll_setup(struct netpoll *np)
 660 {
 661         struct net_device *ndev = NULL, *dev = NULL;
 662         struct net *net = current->nsproxy->net_ns;
 663         struct in_device *in_dev;
 664         int err;
 665
 666         rtnl_lock();
 667         if (np->dev_name[0])
 668                 ndev = __dev_get_by_name(net, np->dev_name);
 669
 670         if (!ndev) {
 671                 np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
 672                 err = -ENODEV;
 673                 goto unlock;
 674         }
 675         dev_hold(ndev);
 676
 677         /* bring up DSA management network devices up first */
 678         for_each_netdev(net, dev) {
 679                 if (!netdev_uses_dsa(dev))
 680                         continue;
 681
 682                 err = dev_change_flags(dev, dev->flags | IFF_UP, NULL);
 683                 if (err < 0) {
 684                         np_err(np, "%s failed to open %s\n",
 685                                np->dev_name, dev->name);
 686                         goto put;
 687                 }
 688         }
 689
 690         if (netdev_master_upper_dev_get(ndev)) {
 691                 np_err(np, "%s is a slave device, aborting\n", np->dev_name);
 692                 err = -EBUSY;
 693                 goto put;
 694         }
 695
 696         if (!netif_running(ndev)) {
 697                 unsigned long atmost, atleast;
 698
 699                 np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
 700
 701                 err = dev_open(ndev, NULL);
 702
 703                 if (err) {
 704                         np_err(np, "failed to open %s\n", ndev->name);
 705                         goto put;
 706                 }
 707
 708                 rtnl_unlock();
 709                 atleast = jiffies + HZ/10;
 710                 atmost = jiffies + carrier_timeout * HZ;
 711                 while (!netif_carrier_ok(ndev)) {
 712                         if (time_after(jiffies, atmost)) {
 713                                 np_notice(np, "timeout waiting for carrier\n");
 714                                 break;
 715                         }
 716                         msleep(1);
 717                 }
 718
 719                 /* If carrier appears to come up instantly, we don't
 720                  * trust it and pause so that we don't pump all our
 721                  * queued console messages into the bitbucket.
 722                  */
 723
 724                 if (time_before(jiffies, atleast)) {
 725                         np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
 726                         msleep(4000);
 727                 }
 728                 rtnl_lock();
 729         }
 730
 731         if (!np->local_ip.ip) {
 732                 if (!np->ipv6) {
 733                         const struct in_ifaddr *ifa;
 734
 735                         in_dev = __in_dev_get_rtnl(ndev);
 736                         if (!in_dev)
 737                                 goto put_noaddr;
 738
 739                         ifa = rtnl_dereference(in_dev->ifa_list);
 740                         if (!ifa) {
 741 put_noaddr:
 742                                 np_err(np, "no IP address for %s, aborting\n",
 743                                        np->dev_name);
 744                                 err = -EDESTADDRREQ;
 745                                 goto put;
 746                         }
 747
 748                         np->local_ip.ip = ifa->ifa_local;
 749                         np_info(np, "local IP %pI4\n", &np->local_ip.ip);
 750                 } else {
 751 #if IS_ENABLED(CONFIG_IPV6)
 752                         struct inet6_dev *idev;
 753
 754                         err = -EDESTADDRREQ;
 755                         idev = __in6_dev_get(ndev);
 756                         if (idev) {
 757                                 struct inet6_ifaddr *ifp;
 758
 759                                 read_lock_bh(&idev->lock);
 760                                 list_for_each_entry(ifp, &idev->addr_list, if_list) {
 761                                         if (!!(ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL) !=
 762                                             !!(ipv6_addr_type(&np->remote_ip.in6) & IPV6_ADDR_LINKLOCAL))
 763                                                 continue;
 764                                         np->local_ip.in6 = ifp->addr;
 765                                         err = 0;
 766                                         break;
 767                                 }
 768                                 read_unlock_bh(&idev->lock);
 769                         }
 770                         if (err) {
 771                                 np_err(np, "no IPv6 address for %s, aborting\n",
 772                                        np->dev_name);
 773                                 goto put;
 774                         } else
 775                                 np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
 776 #else
 777                         np_err(np, "IPv6 is not supported %s, aborting\n",
 778                                np->dev_name);
 779                         err = -EINVAL;
 780                         goto put;
 781 #endif
 782                 }
 783         }
 784
 785         /* fill up the skb queue */
 786         refill_skbs();
 787
 788         err = __netpoll_setup(np, ndev);
 789         if (err)
 790                 goto put;
 791
 792         rtnl_unlock();
 793         return 0;
 794
 795 put:
 796         dev_put(ndev);
 797 unlock:
 798         rtnl_unlock();
 799         return err;
 800 }
 801 EXPORT_SYMBOL(netpoll_setup);
 802
 803 static int __init netpoll_init(void)
 804 {
 805         skb_queue_head_init(&skb_pool);
 806         return 0;
 807 }
 808 core_initcall(netpoll_init);
 809
 810 static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
 811 {
 812         struct netpoll_info *npinfo =
 813                         container_of(rcu_head, struct netpoll_info, rcu);
 814
 815         skb_queue_purge(&npinfo->txq);
 816
 817         /* we can't call cancel_delayed_work_sync here, as we are in softirq */
 818         cancel_delayed_work(&npinfo->tx_work);
 819
 820         /* clean after last, unfinished work */
 821         __skb_queue_purge(&npinfo->txq);
 822         /* now cancel it again */
 823         cancel_delayed_work(&npinfo->tx_work);
 824         kfree(npinfo);
 825 }
 826
 827 void __netpoll_cleanup(struct netpoll *np)
 828 {
 829         struct netpoll_info *npinfo;
 830
 831         npinfo = rtnl_dereference(np->dev->npinfo);
 832         if (!npinfo)
 833                 return;
 834
 835         synchronize_srcu(&netpoll_srcu);
 836
 837         if (refcount_dec_and_test(&npinfo->refcnt)) {
 838                 const struct net_device_ops *ops;
 839
 840                 ops = np->dev->netdev_ops;
 841                 if (ops->ndo_netpoll_cleanup)
 842                         ops->ndo_netpoll_cleanup(np->dev);
 843
 844                 RCU_INIT_POINTER(np->dev->npinfo, NULL);
 845                 call_rcu(&npinfo->rcu, rcu_cleanup_netpoll_info);
 846         } else
 847                 RCU_INIT_POINTER(np->dev->npinfo, NULL);
 848 }
 849 EXPORT_SYMBOL_GPL(__netpoll_cleanup);
 850
 851 void __netpoll_free(struct netpoll *np)
 852 {
 853         ASSERT_RTNL();
 854
 855         /* Wait for transmitting packets to finish before freeing. */
 856         synchronize_rcu();
 857         __netpoll_cleanup(np);
 858         kfree(np);
 859 }
 860 EXPORT_SYMBOL_GPL(__netpoll_free);
 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);