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/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 #include <linux/kconfig.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 int netif_local_xmit_active(struct net_device *dev)
 141 {
 142         int i;
 143
 144         for (i = 0; i < dev->num_tx_queues; i++) {
 145                 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
 146
 147                 if (READ_ONCE(txq->xmit_lock_owner) == smp_processor_id())
 148                         return 1;
 149         }
 150
 151         return 0;
 152 }
 153
 154 static void poll_one_napi(struct napi_struct *napi)
 155 {
 156         int work;
 157
 158         /* If we set this bit but see that it has already been set,
 159          * that indicates that napi has been disabled and we need
 160          * to abort this operation
 161          */
 162         if (test_and_set_bit(NAPI_STATE_NPSVC, &napi->state))
 163                 return;
 164
 165         /* We explicilty pass the polling call a budget of 0 to
 166          * indicate that we are clearing the Tx path only.
 167          */
 168         work = napi->poll(napi, 0);
 169         WARN_ONCE(work, "%pS exceeded budget in poll\n", napi->poll);
 170         trace_napi_poll(napi, work, 0);
 171
 172         clear_bit(NAPI_STATE_NPSVC, &napi->state);
 173 }
 174
 175 static void poll_napi(struct net_device *dev)
 176 {
 177         struct napi_struct *napi;
 178         int cpu = smp_processor_id();
 179
 180         list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) {
 181                 if (cmpxchg(&napi->poll_owner, -1, cpu) == -1) {
 182                         poll_one_napi(napi);
 183                         smp_store_release(&napi->poll_owner, -1);
 184                 }
 185         }
 186 }
 187
 188 void netpoll_poll_dev(struct net_device *dev)
 189 {
 190         struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
 191         const struct net_device_ops *ops;
 192
 193         /* Don't do any rx activity if the dev_lock mutex is held
 194          * the dev_open/close paths use this to block netpoll activity
 195          * while changing device state
 196          */
 197         if (!ni || down_trylock(&ni->dev_lock))
 198                 return;
 199
 200         /* Some drivers will take the same locks in poll and xmit,
 201          * we can't poll if local CPU is already in xmit.
 202          */
 203         if (!netif_running(dev) || netif_local_xmit_active(dev)) {
 204                 up(&ni->dev_lock);
 205                 return;
 206         }
 207
 208         ops = dev->netdev_ops;
 209         if (ops->ndo_poll_controller)
 210                 ops->ndo_poll_controller(dev);
 211
 212         poll_napi(dev);
 213
 214         up(&ni->dev_lock);
 215
 216         zap_completion_queue();
 217 }
 218 EXPORT_SYMBOL(netpoll_poll_dev);
 219
 220 void netpoll_poll_disable(struct net_device *dev)
 221 {
 222         struct netpoll_info *ni;
 223         int idx;
 224         might_sleep();
 225         idx = srcu_read_lock(&netpoll_srcu);
 226         ni = srcu_dereference(dev->npinfo, &netpoll_srcu);
 227         if (ni)
 228                 down(&ni->dev_lock);
 229         srcu_read_unlock(&netpoll_srcu, idx);
 230 }
 231 EXPORT_SYMBOL(netpoll_poll_disable);
 232
 233 void netpoll_poll_enable(struct net_device *dev)
 234 {
 235         struct netpoll_info *ni;
 236         rcu_read_lock();
 237         ni = rcu_dereference(dev->npinfo);
 238         if (ni)
 239                 up(&ni->dev_lock);
 240         rcu_read_unlock();
 241 }
 242 EXPORT_SYMBOL(netpoll_poll_enable);
 243
 244 static void refill_skbs(void)
 245 {
 246         struct sk_buff *skb;
 247         unsigned long flags;
 248
 249         spin_lock_irqsave(&skb_pool.lock, flags);
 250         while (skb_pool.qlen < MAX_SKBS) {
 251                 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
 252                 if (!skb)
 253                         break;
 254
 255                 __skb_queue_tail(&skb_pool, skb);
 256         }
 257         spin_unlock_irqrestore(&skb_pool.lock, flags);
 258 }
 259
 260 static void zap_completion_queue(void)
 261 {
 262         unsigned long flags;
 263         struct softnet_data *sd = &get_cpu_var(softnet_data);
 264
 265         if (sd->completion_queue) {
 266                 struct sk_buff *clist;
 267
 268                 local_irq_save(flags);
 269                 clist = sd->completion_queue;
 270                 sd->completion_queue = NULL;
 271                 local_irq_restore(flags);
 272
 273                 while (clist != NULL) {
 274                         struct sk_buff *skb = clist;
 275                         clist = clist->next;
 276                         if (!skb_irq_freeable(skb)) {
 277                                 refcount_set(&skb->users, 1);
 278                                 dev_kfree_skb_any(skb); /* put this one back */
 279                         } else {
 280                                 __kfree_skb(skb);
 281                         }
 282                 }
 283         }
 284
 285         put_cpu_var(softnet_data);
 286 }
 287
 288 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
 289 {
 290         int count = 0;
 291         struct sk_buff *skb;
 292
 293         zap_completion_queue();
 294         refill_skbs();
 295 repeat:
 296
 297         skb = alloc_skb(len, GFP_ATOMIC);
 298         if (!skb)
 299                 skb = skb_dequeue(&skb_pool);
 300
 301         if (!skb) {
 302                 if (++count < 10) {
 303                         netpoll_poll_dev(np->dev);
 304                         goto repeat;
 305                 }
 306                 return NULL;
 307         }
 308
 309         refcount_set(&skb->users, 1);
 310         skb_reserve(skb, reserve);
 311         return skb;
 312 }
 313
 314 static int netpoll_owner_active(struct net_device *dev)
 315 {
 316         struct napi_struct *napi;
 317
 318         list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) {
 319                 if (napi->poll_owner == smp_processor_id())
 320                         return 1;
 321         }
 322         return 0;
 323 }
 324
 325 /* call with IRQ disabled */
 326 static netdev_tx_t __netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
 327 {
 328         netdev_tx_t status = NETDEV_TX_BUSY;
 329         struct net_device *dev;
 330         unsigned long tries;
 331         /* It is up to the caller to keep npinfo alive. */
 332         struct netpoll_info *npinfo;
 333
 334         lockdep_assert_irqs_disabled();
 335
 336         dev = np->dev;
 337         npinfo = rcu_dereference_bh(dev->npinfo);
 338
 339         if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
 340                 dev_kfree_skb_irq(skb);
 341                 return NET_XMIT_DROP;
 342         }
 343
 344         /* don't get messages out of order, and no recursion */
 345         if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
 346                 struct netdev_queue *txq;
 347
 348                 txq = netdev_core_pick_tx(dev, skb, NULL);
 349
 350                 /* try until next clock tick */
 351                 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
 352                      tries > 0; --tries) {
 353                         if (HARD_TX_TRYLOCK(dev, txq)) {
 354                                 if (!netif_xmit_stopped(txq))
 355                                         status = netpoll_start_xmit(skb, dev, txq);
 356
 357                                 HARD_TX_UNLOCK(dev, txq);
 358
 359                                 if (dev_xmit_complete(status))
 360                                         break;
 361
 362                         }
 363
 364                         /* tickle device maybe there is some cleanup */
 365                         netpoll_poll_dev(np->dev);
 366
 367                         udelay(USEC_PER_POLL);
 368                 }
 369
 370                 WARN_ONCE(!irqs_disabled(),
 371                         "netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pS)\n",
 372                         dev->name, dev->netdev_ops->ndo_start_xmit);
 373
 374         }
 375
 376         if (!dev_xmit_complete(status)) {
 377                 skb_queue_tail(&npinfo->txq, skb);
 378                 schedule_delayed_work(&npinfo->tx_work,0);
 379         }
 380         return NETDEV_TX_OK;
 381 }
 382
 383 netdev_tx_t netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
 384 {
 385         unsigned long flags;
 386         netdev_tx_t ret;
 387
 388         if (unlikely(!np)) {
 389                 dev_kfree_skb_irq(skb);
 390                 ret = NET_XMIT_DROP;
 391         } else {
 392                 local_irq_save(flags);
 393                 ret = __netpoll_send_skb(np, skb);
 394                 local_irq_restore(flags);
 395         }
 396         return ret;
 397 }
 398 EXPORT_SYMBOL(netpoll_send_skb);
 399
 400 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
 401 {
 402         int total_len, ip_len, udp_len;
 403         struct sk_buff *skb;
 404         struct udphdr *udph;
 405         struct iphdr *iph;
 406         struct ethhdr *eth;
 407         static atomic_t ip_ident;
 408         struct ipv6hdr *ip6h;
 409
 410         if (!IS_ENABLED(CONFIG_PREEMPT_RT))
 411                 WARN_ON_ONCE(!irqs_disabled());
 412
 413         udp_len = len + sizeof(*udph);
 414         if (np->ipv6)
 415                 ip_len = udp_len + sizeof(*ip6h);
 416         else
 417                 ip_len = udp_len + sizeof(*iph);
 418
 419         total_len = ip_len + LL_RESERVED_SPACE(np->dev);
 420
 421         skb = find_skb(np, total_len + np->dev->needed_tailroom,
 422                        total_len - len);
 423         if (!skb)
 424                 return;
 425
 426         skb_copy_to_linear_data(skb, msg, len);
 427         skb_put(skb, len);
 428
 429         skb_push(skb, sizeof(*udph));
 430         skb_reset_transport_header(skb);
 431         udph = udp_hdr(skb);
 432         udph->source = htons(np->local_port);
 433         udph->dest = htons(np->remote_port);
 434         udph->len = htons(udp_len);
 435
 436         if (np->ipv6) {
 437                 udph->check = 0;
 438                 udph->check = csum_ipv6_magic(&np->local_ip.in6,
 439                                               &np->remote_ip.in6,
 440                                               udp_len, IPPROTO_UDP,
 441                                               csum_partial(udph, udp_len, 0));
 442                 if (udph->check == 0)
 443                         udph->check = CSUM_MANGLED_0;
 444
 445                 skb_push(skb, sizeof(*ip6h));
 446                 skb_reset_network_header(skb);
 447                 ip6h = ipv6_hdr(skb);
 448
 449                 /* ip6h->version = 6; ip6h->priority = 0; */
 450                 *(unsigned char *)ip6h = 0x60;
 451                 ip6h->flow_lbl[0] = 0;
 452                 ip6h->flow_lbl[1] = 0;
 453                 ip6h->flow_lbl[2] = 0;
 454
 455                 ip6h->payload_len = htons(sizeof(struct udphdr) + len);
 456                 ip6h->nexthdr = IPPROTO_UDP;
 457                 ip6h->hop_limit = 32;
 458                 ip6h->saddr = np->local_ip.in6;
 459                 ip6h->daddr = np->remote_ip.in6;
 460
 461                 eth = skb_push(skb, ETH_HLEN);
 462                 skb_reset_mac_header(skb);
 463                 skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
 464         } else {
 465                 udph->check = 0;
 466                 udph->check = csum_tcpudp_magic(np->local_ip.ip,
 467                                                 np->remote_ip.ip,
 468                                                 udp_len, IPPROTO_UDP,
 469                                                 csum_partial(udph, udp_len, 0));
 470                 if (udph->check == 0)
 471                         udph->check = CSUM_MANGLED_0;
 472
 473                 skb_push(skb, sizeof(*iph));
 474                 skb_reset_network_header(skb);
 475                 iph = ip_hdr(skb);
 476
 477                 /* iph->version = 4; iph->ihl = 5; */
 478                 *(unsigned char *)iph = 0x45;
 479                 iph->tos      = 0;
 480                 put_unaligned(htons(ip_len), &(iph->tot_len));
 481                 iph->id       = htons(atomic_inc_return(&ip_ident));
 482                 iph->frag_off = 0;
 483                 iph->ttl      = 64;
 484                 iph->protocol = IPPROTO_UDP;
 485                 iph->check    = 0;
 486                 put_unaligned(np->local_ip.ip, &(iph->saddr));
 487                 put_unaligned(np->remote_ip.ip, &(iph->daddr));
 488                 iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
 489
 490                 eth = skb_push(skb, ETH_HLEN);
 491                 skb_reset_mac_header(skb);
 492                 skb->protocol = eth->h_proto = htons(ETH_P_IP);
 493         }
 494
 495         ether_addr_copy(eth->h_source, np->dev->dev_addr);
 496         ether_addr_copy(eth->h_dest, np->remote_mac);
 497
 498         skb->dev = np->dev;
 499
 500         netpoll_send_skb(np, skb);
 501 }
 502 EXPORT_SYMBOL(netpoll_send_udp);
 503
 504 void netpoll_print_options(struct netpoll *np)
 505 {
 506         np_info(np, "local port %d\n", np->local_port);
 507         if (np->ipv6)
 508                 np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
 509         else
 510                 np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
 511         np_info(np, "interface '%s'\n", np->dev_name);
 512         np_info(np, "remote port %d\n", np->remote_port);
 513         if (np->ipv6)
 514                 np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
 515         else
 516                 np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
 517         np_info(np, "remote ethernet address %pM\n", np->remote_mac);
 518 }
 519 EXPORT_SYMBOL(netpoll_print_options);
 520
 521 static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
 522 {
 523         const char *end;
 524
 525         if (!strchr(str, ':') &&
 526             in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
 527                 if (!*end)
 528                         return 0;
 529         }
 530         if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
 531 #if IS_ENABLED(CONFIG_IPV6)
 532                 if (!*end)
 533                         return 1;
 534 #else
 535                 return -1;
 536 #endif
 537         }
 538         return -1;
 539 }
 540
 541 int netpoll_parse_options(struct netpoll *np, char *opt)
 542 {
 543         char *cur=opt, *delim;
 544         int ipv6;
 545         bool ipversion_set = false;
 546
 547         if (*cur != '@') {
 548                 if ((delim = strchr(cur, '@')) == NULL)
 549                         goto parse_failed;
 550                 *delim = 0;
 551                 if (kstrtou16(cur, 10, &np->local_port))
 552                         goto parse_failed;
 553                 cur = delim;
 554         }
 555         cur++;
 556
 557         if (*cur != '/') {
 558                 ipversion_set = true;
 559                 if ((delim = strchr(cur, '/')) == NULL)
 560                         goto parse_failed;
 561                 *delim = 0;
 562                 ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
 563                 if (ipv6 < 0)
 564                         goto parse_failed;
 565                 else
 566                         np->ipv6 = (bool)ipv6;
 567                 cur = delim;
 568         }
 569         cur++;
 570
 571         if (*cur != ',') {
 572                 /* parse out dev name */
 573                 if ((delim = strchr(cur, ',')) == NULL)
 574                         goto parse_failed;
 575                 *delim = 0;
 576                 strscpy(np->dev_name, cur, sizeof(np->dev_name));
 577                 cur = delim;
 578         }
 579         cur++;
 580
 581         if (*cur != '@') {
 582                 /* dst port */
 583                 if ((delim = strchr(cur, '@')) == NULL)
 584                         goto parse_failed;
 585                 *delim = 0;
 586                 if (*cur == ' ' || *cur == '\t')
 587                         np_info(np, "warning: whitespace is not allowed\n");
 588                 if (kstrtou16(cur, 10, &np->remote_port))
 589                         goto parse_failed;
 590                 cur = delim;
 591         }
 592         cur++;
 593
 594         /* dst ip */
 595         if ((delim = strchr(cur, '/')) == NULL)
 596                 goto parse_failed;
 597         *delim = 0;
 598         ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
 599         if (ipv6 < 0)
 600                 goto parse_failed;
 601         else if (ipversion_set && np->ipv6 != (bool)ipv6)
 602                 goto parse_failed;
 603         else
 604                 np->ipv6 = (bool)ipv6;
 605         cur = delim + 1;
 606
 607         if (*cur != 0) {
 608                 /* MAC address */
 609                 if (!mac_pton(cur, np->remote_mac))
 610                         goto parse_failed;
 611         }
 612
 613         netpoll_print_options(np);
 614
 615         return 0;
 616
 617  parse_failed:
 618         np_info(np, "couldn't parse config at '%s'!\n", cur);
 619         return -1;
 620 }
 621 EXPORT_SYMBOL(netpoll_parse_options);
 622
 623 int __netpoll_setup(struct netpoll *np, struct net_device *ndev)
 624 {
 625         struct netpoll_info *npinfo;
 626         const struct net_device_ops *ops;
 627         int err;
 628
 629         np->dev = ndev;
 630         strscpy(np->dev_name, ndev->name, IFNAMSIZ);
 631
 632         if (ndev->priv_flags & IFF_DISABLE_NETPOLL) {
 633                 np_err(np, "%s doesn't support polling, aborting\n",
 634                        np->dev_name);
 635                 err = -ENOTSUPP;
 636                 goto out;
 637         }
 638
 639         if (!ndev->npinfo) {
 640                 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
 641                 if (!npinfo) {
 642                         err = -ENOMEM;
 643                         goto out;
 644                 }
 645
 646                 sema_init(&npinfo->dev_lock, 1);
 647                 skb_queue_head_init(&npinfo->txq);
 648                 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
 649
 650                 refcount_set(&npinfo->refcnt, 1);
 651
 652                 ops = np->dev->netdev_ops;
 653                 if (ops->ndo_netpoll_setup) {
 654                         err = ops->ndo_netpoll_setup(ndev, npinfo);
 655                         if (err)
 656                                 goto free_npinfo;
 657                 }
 658         } else {
 659                 npinfo = rtnl_dereference(ndev->npinfo);
 660                 refcount_inc(&npinfo->refcnt);
 661         }
 662
 663         npinfo->netpoll = np;
 664
 665         /* last thing to do is link it to the net device structure */
 666         rcu_assign_pointer(ndev->npinfo, npinfo);
 667
 668         return 0;
 669
 670 free_npinfo:
 671         kfree(npinfo);
 672 out:
 673         return err;
 674 }
 675 EXPORT_SYMBOL_GPL(__netpoll_setup);
 676
 677 int netpoll_setup(struct netpoll *np)
 678 {
 679         struct net_device *ndev = NULL;
 680         struct in_device *in_dev;
 681         int err;
 682
 683         rtnl_lock();
 684         if (np->dev_name[0]) {
 685                 struct net *net = current->nsproxy->net_ns;
 686                 ndev = __dev_get_by_name(net, np->dev_name);
 687         }
 688         if (!ndev) {
 689                 np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
 690                 err = -ENODEV;
 691                 goto unlock;
 692         }
 693         dev_hold(ndev);
 694
 695         if (netdev_master_upper_dev_get(ndev)) {
 696                 np_err(np, "%s is a slave device, aborting\n", np->dev_name);
 697                 err = -EBUSY;
 698                 goto put;
 699         }
 700
 701         if (!netif_running(ndev)) {
 702                 unsigned long atmost, atleast;
 703
 704                 np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
 705
 706                 err = dev_open(ndev, NULL);
 707
 708                 if (err) {
 709                         np_err(np, "failed to open %s\n", ndev->name);
 710                         goto put;
 711                 }
 712
 713                 rtnl_unlock();
 714                 atleast = jiffies + HZ/10;
 715                 atmost = jiffies + carrier_timeout * HZ;
 716                 while (!netif_carrier_ok(ndev)) {
 717                         if (time_after(jiffies, atmost)) {
 718                                 np_notice(np, "timeout waiting for carrier\n");
 719                                 break;
 720                         }
 721                         msleep(1);
 722                 }
 723
 724                 /* If carrier appears to come up instantly, we don't
 725                  * trust it and pause so that we don't pump all our
 726                  * queued console messages into the bitbucket.
 727                  */
 728
 729                 if (time_before(jiffies, atleast)) {
 730                         np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
 731                         msleep(4000);
 732                 }
 733                 rtnl_lock();
 734         }
 735
 736         if (!np->local_ip.ip) {
 737                 if (!np->ipv6) {
 738                         const struct in_ifaddr *ifa;
 739
 740                         in_dev = __in_dev_get_rtnl(ndev);
 741                         if (!in_dev)
 742                                 goto put_noaddr;
 743
 744                         ifa = rtnl_dereference(in_dev->ifa_list);
 745                         if (!ifa) {
 746 put_noaddr:
 747                                 np_err(np, "no IP address for %s, aborting\n",
 748                                        np->dev_name);
 749                                 err = -EDESTADDRREQ;
 750                                 goto put;
 751                         }
 752
 753                         np->local_ip.ip = ifa->ifa_local;
 754                         np_info(np, "local IP %pI4\n", &np->local_ip.ip);
 755                 } else {
 756 #if IS_ENABLED(CONFIG_IPV6)
 757                         struct inet6_dev *idev;
 758
 759                         err = -EDESTADDRREQ;
 760                         idev = __in6_dev_get(ndev);
 761                         if (idev) {
 762                                 struct inet6_ifaddr *ifp;
 763
 764                                 read_lock_bh(&idev->lock);
 765                                 list_for_each_entry(ifp, &idev->addr_list, if_list) {
 766                                         if (!!(ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL) !=
 767                                             !!(ipv6_addr_type(&np->remote_ip.in6) & IPV6_ADDR_LINKLOCAL))
 768                                                 continue;
 769                                         np->local_ip.in6 = ifp->addr;
 770                                         err = 0;
 771                                         break;
 772                                 }
 773                                 read_unlock_bh(&idev->lock);
 774                         }
 775                         if (err) {
 776                                 np_err(np, "no IPv6 address for %s, aborting\n",
 777                                        np->dev_name);
 778                                 goto put;
 779                         } else
 780                                 np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
 781 #else
 782                         np_err(np, "IPv6 is not supported %s, aborting\n",
 783                                np->dev_name);
 784                         err = -EINVAL;
 785                         goto put;
 786 #endif
 787                 }
 788         }
 789
 790         /* fill up the skb queue */
 791         refill_skbs();
 792
 793         err = __netpoll_setup(np, ndev);
 794         if (err)
 795                 goto put;
 796
 797         rtnl_unlock();
 798         return 0;
 799
 800 put:
 801         dev_put(ndev);
 802 unlock:
 803         rtnl_unlock();
 804         return err;
 805 }
 806 EXPORT_SYMBOL(netpoll_setup);
 807
 808 static int __init netpoll_init(void)
 809 {
 810         skb_queue_head_init(&skb_pool);
 811         return 0;
 812 }
 813 core_initcall(netpoll_init);
 814
 815 static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
 816 {
 817         struct netpoll_info *npinfo =
 818                         container_of(rcu_head, struct netpoll_info, rcu);
 819
 820         skb_queue_purge(&npinfo->txq);
 821
 822         /* we can't call cancel_delayed_work_sync here, as we are in softirq */
 823         cancel_delayed_work(&npinfo->tx_work);
 824
 825         /* clean after last, unfinished work */
 826         __skb_queue_purge(&npinfo->txq);
 827         /* now cancel it again */
 828         cancel_delayed_work(&npinfo->tx_work);
 829         kfree(npinfo);
 830 }
 831
 832 void __netpoll_cleanup(struct netpoll *np)
 833 {
 834         struct netpoll_info *npinfo;
 835
 836         npinfo = rtnl_dereference(np->dev->npinfo);
 837         if (!npinfo)
 838                 return;
 839
 840         synchronize_srcu(&netpoll_srcu);
 841
 842         if (refcount_dec_and_test(&npinfo->refcnt)) {
 843                 const struct net_device_ops *ops;
 844
 845                 ops = np->dev->netdev_ops;
 846                 if (ops->ndo_netpoll_cleanup)
 847                         ops->ndo_netpoll_cleanup(np->dev);
 848
 849                 RCU_INIT_POINTER(np->dev->npinfo, NULL);
 850                 call_rcu(&npinfo->rcu, rcu_cleanup_netpoll_info);
 851         } else
 852                 RCU_INIT_POINTER(np->dev->npinfo, NULL);
 853 }
 854 EXPORT_SYMBOL_GPL(__netpoll_cleanup);
 855
 856 void __netpoll_free(struct netpoll *np)
 857 {
 858         ASSERT_RTNL();
 859
 860         /* Wait for transmitting packets to finish before freeing. */
 861         synchronize_rcu();
 862         __netpoll_cleanup(np);
 863         kfree(np);
 864 }
 865 EXPORT_SYMBOL_GPL(__netpoll_free);
 866
 867 void netpoll_cleanup(struct netpoll *np)
 868 {
 869         rtnl_lock();
 870         if (!np->dev)
 871                 goto out;
 872         __netpoll_cleanup(np);
 873         dev_put(np->dev);
 874         np->dev = NULL;
 875 out:
 876         rtnl_unlock();
 877 }
 878 EXPORT_SYMBOL(netpoll_cleanup);