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