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 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         WARN_ON_ONCE(!irqs_disabled());
 411
 412         udp_len = len + sizeof(*udph);
 413         if (np->ipv6)
 414                 ip_len = udp_len + sizeof(*ip6h);
 415         else
 416                 ip_len = udp_len + sizeof(*iph);
 417
 418         total_len = ip_len + LL_RESERVED_SPACE(np->dev);
 419
 420         skb = find_skb(np, total_len + np->dev->needed_tailroom,
 421                        total_len - len);
 422         if (!skb)
 423                 return;
 424
 425         skb_copy_to_linear_data(skb, msg, len);
 426         skb_put(skb, len);
 427
 428         skb_push(skb, sizeof(*udph));
 429         skb_reset_transport_header(skb);
 430         udph = udp_hdr(skb);
 431         udph->source = htons(np->local_port);
 432         udph->dest = htons(np->remote_port);
 433         udph->len = htons(udp_len);
 434
 435         if (np->ipv6) {
 436                 udph->check = 0;
 437                 udph->check = csum_ipv6_magic(&np->local_ip.in6,
 438                                               &np->remote_ip.in6,
 439                                               udp_len, IPPROTO_UDP,
 440                                               csum_partial(udph, udp_len, 0));
 441                 if (udph->check == 0)
 442                         udph->check = CSUM_MANGLED_0;
 443
 444                 skb_push(skb, sizeof(*ip6h));
 445                 skb_reset_network_header(skb);
 446                 ip6h = ipv6_hdr(skb);
 447
 448                 /* ip6h->version = 6; ip6h->priority = 0; */
 449                 put_unaligned(0x60, (unsigned char *)ip6h);
 450                 ip6h->flow_lbl[0] = 0;
 451                 ip6h->flow_lbl[1] = 0;
 452                 ip6h->flow_lbl[2] = 0;
 453
 454                 ip6h->payload_len = htons(sizeof(struct udphdr) + len);
 455                 ip6h->nexthdr = IPPROTO_UDP;
 456                 ip6h->hop_limit = 32;
 457                 ip6h->saddr = np->local_ip.in6;
 458                 ip6h->daddr = np->remote_ip.in6;
 459
 460                 eth = skb_push(skb, ETH_HLEN);
 461                 skb_reset_mac_header(skb);
 462                 skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
 463         } else {
 464                 udph->check = 0;
 465                 udph->check = csum_tcpudp_magic(np->local_ip.ip,
 466                                                 np->remote_ip.ip,
 467                                                 udp_len, IPPROTO_UDP,
 468                                                 csum_partial(udph, udp_len, 0));
 469                 if (udph->check == 0)
 470                         udph->check = CSUM_MANGLED_0;
 471
 472                 skb_push(skb, sizeof(*iph));
 473                 skb_reset_network_header(skb);
 474                 iph = ip_hdr(skb);
 475
 476                 /* iph->version = 4; iph->ihl = 5; */
 477                 put_unaligned(0x45, (unsigned char *)iph);
 478                 iph->tos      = 0;
 479                 put_unaligned(htons(ip_len), &(iph->tot_len));
 480                 iph->id       = htons(atomic_inc_return(&ip_ident));
 481                 iph->frag_off = 0;
 482                 iph->ttl      = 64;
 483                 iph->protocol = IPPROTO_UDP;
 484                 iph->check    = 0;
 485                 put_unaligned(np->local_ip.ip, &(iph->saddr));
 486                 put_unaligned(np->remote_ip.ip, &(iph->daddr));
 487                 iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
 488
 489                 eth = skb_push(skb, ETH_HLEN);
 490                 skb_reset_mac_header(skb);
 491                 skb->protocol = eth->h_proto = htons(ETH_P_IP);
 492         }
 493
 494         ether_addr_copy(eth->h_source, np->dev->dev_addr);
 495         ether_addr_copy(eth->h_dest, np->remote_mac);
 496
 497         skb->dev = np->dev;
 498
 499         netpoll_send_skb(np, skb);
 500 }
 501 EXPORT_SYMBOL(netpoll_send_udp);
 502
 503 void netpoll_print_options(struct netpoll *np)
 504 {
 505         np_info(np, "local port %d\n", np->local_port);
 506         if (np->ipv6)
 507                 np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
 508         else
 509                 np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
 510         np_info(np, "interface '%s'\n", np->dev_name);
 511         np_info(np, "remote port %d\n", np->remote_port);
 512         if (np->ipv6)
 513                 np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
 514         else
 515                 np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
 516         np_info(np, "remote ethernet address %pM\n", np->remote_mac);
 517 }
 518 EXPORT_SYMBOL(netpoll_print_options);
 519
 520 static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
 521 {
 522         const char *end;
 523
 524         if (!strchr(str, ':') &&
 525             in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
 526                 if (!*end)
 527                         return 0;
 528         }
 529         if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
 530 #if IS_ENABLED(CONFIG_IPV6)
 531                 if (!*end)
 532                         return 1;
 533 #else
 534                 return -1;
 535 #endif
 536         }
 537         return -1;
 538 }
 539
 540 int netpoll_parse_options(struct netpoll *np, char *opt)
 541 {
 542         char *cur=opt, *delim;
 543         int ipv6;
 544         bool ipversion_set = false;
 545
 546         if (*cur != '@') {
 547                 if ((delim = strchr(cur, '@')) == NULL)
 548                         goto parse_failed;
 549                 *delim = 0;
 550                 if (kstrtou16(cur, 10, &np->local_port))
 551                         goto parse_failed;
 552                 cur = delim;
 553         }
 554         cur++;
 555
 556         if (*cur != '/') {
 557                 ipversion_set = true;
 558                 if ((delim = strchr(cur, '/')) == NULL)
 559                         goto parse_failed;
 560                 *delim = 0;
 561                 ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
 562                 if (ipv6 < 0)
 563                         goto parse_failed;
 564                 else
 565                         np->ipv6 = (bool)ipv6;
 566                 cur = delim;
 567         }
 568         cur++;
 569
 570         if (*cur != ',') {
 571                 /* parse out dev name */
 572                 if ((delim = strchr(cur, ',')) == NULL)
 573                         goto parse_failed;
 574                 *delim = 0;
 575                 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
 576                 cur = delim;
 577         }
 578         cur++;
 579
 580         if (*cur != '@') {
 581                 /* dst port */
 582                 if ((delim = strchr(cur, '@')) == NULL)
 583                         goto parse_failed;
 584                 *delim = 0;
 585                 if (*cur == ' ' || *cur == '\t')
 586                         np_info(np, "warning: whitespace is not allowed\n");
 587                 if (kstrtou16(cur, 10, &np->remote_port))
 588                         goto parse_failed;
 589                 cur = delim;
 590         }
 591         cur++;
 592
 593         /* dst ip */
 594         if ((delim = strchr(cur, '/')) == NULL)
 595                 goto parse_failed;
 596         *delim = 0;
 597         ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
 598         if (ipv6 < 0)
 599                 goto parse_failed;
 600         else if (ipversion_set && np->ipv6 != (bool)ipv6)
 601                 goto parse_failed;
 602         else
 603                 np->ipv6 = (bool)ipv6;
 604         cur = delim + 1;
 605
 606         if (*cur != 0) {
 607                 /* MAC address */
 608                 if (!mac_pton(cur, np->remote_mac))
 609                         goto parse_failed;
 610         }
 611
 612         netpoll_print_options(np);
 613
 614         return 0;
 615
 616  parse_failed:
 617         np_info(np, "couldn't parse config at '%s'!\n", cur);
 618         return -1;
 619 }
 620 EXPORT_SYMBOL(netpoll_parse_options);
 621
 622 int __netpoll_setup(struct netpoll *np, struct net_device *ndev)
 623 {
 624         struct netpoll_info *npinfo;
 625         const struct net_device_ops *ops;
 626         int err;
 627
 628         np->dev = ndev;
 629         strlcpy(np->dev_name, ndev->name, IFNAMSIZ);
 630
 631         if (ndev->priv_flags & IFF_DISABLE_NETPOLL) {
 632                 np_err(np, "%s doesn't support polling, aborting\n",
 633                        np->dev_name);
 634                 err = -ENOTSUPP;
 635                 goto out;
 636         }
 637
 638         if (!ndev->npinfo) {
 639                 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
 640                 if (!npinfo) {
 641                         err = -ENOMEM;
 642                         goto out;
 643                 }
 644
 645                 sema_init(&npinfo->dev_lock, 1);
 646                 skb_queue_head_init(&npinfo->txq);
 647                 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
 648
 649                 refcount_set(&npinfo->refcnt, 1);
 650
 651                 ops = np->dev->netdev_ops;
 652                 if (ops->ndo_netpoll_setup) {
 653                         err = ops->ndo_netpoll_setup(ndev, npinfo);
 654                         if (err)
 655                                 goto free_npinfo;
 656                 }
 657         } else {
 658                 npinfo = rtnl_dereference(ndev->npinfo);
 659                 refcount_inc(&npinfo->refcnt);
 660         }
 661
 662         npinfo->netpoll = np;
 663
 664         /* last thing to do is link it to the net device structure */
 665         rcu_assign_pointer(ndev->npinfo, npinfo);
 666
 667         return 0;
 668
 669 free_npinfo:
 670         kfree(npinfo);
 671 out:
 672         return err;
 673 }
 674 EXPORT_SYMBOL_GPL(__netpoll_setup);
 675
 676 int netpoll_setup(struct netpoll *np)
 677 {
 678         struct net_device *ndev = NULL, *dev = NULL;
 679         struct net *net = current->nsproxy->net_ns;
 680         struct in_device *in_dev;
 681         int err;
 682
 683         rtnl_lock();
 684         if (np->dev_name[0])
 685                 ndev = __dev_get_by_name(net, np->dev_name);
 686
 687         if (!ndev) {
 688                 np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
 689                 err = -ENODEV;
 690                 goto unlock;
 691         }
 692         dev_hold(ndev);
 693
 694         /* bring up DSA management network devices up first */
 695         for_each_netdev(net, dev) {
 696                 if (!netdev_uses_dsa(dev))
 697                         continue;
 698
 699                 err = dev_change_flags(dev, dev->flags | IFF_UP, NULL);
 700                 if (err < 0) {
 701                         np_err(np, "%s failed to open %s\n",
 702                                np->dev_name, dev->name);
 703                         goto put;
 704                 }
 705         }
 706
 707         if (netdev_master_upper_dev_get(ndev)) {
 708                 np_err(np, "%s is a slave device, aborting\n", np->dev_name);
 709                 err = -EBUSY;
 710                 goto put;
 711         }
 712
 713         if (!netif_running(ndev)) {
 714                 unsigned long atmost, atleast;
 715
 716                 np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
 717
 718                 err = dev_open(ndev, NULL);
 719
 720                 if (err) {
 721                         np_err(np, "failed to open %s\n", ndev->name);
 722                         goto put;
 723                 }
 724
 725                 rtnl_unlock();
 726                 atleast = jiffies + HZ/10;
 727                 atmost = jiffies + carrier_timeout * HZ;
 728                 while (!netif_carrier_ok(ndev)) {
 729                         if (time_after(jiffies, atmost)) {
 730                                 np_notice(np, "timeout waiting for carrier\n");
 731                                 break;
 732                         }
 733                         msleep(1);
 734                 }
 735
 736                 /* If carrier appears to come up instantly, we don't
 737                  * trust it and pause so that we don't pump all our
 738                  * queued console messages into the bitbucket.
 739                  */
 740
 741                 if (time_before(jiffies, atleast)) {
 742                         np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
 743                         msleep(4000);
 744                 }
 745                 rtnl_lock();
 746         }
 747
 748         if (!np->local_ip.ip) {
 749                 if (!np->ipv6) {
 750                         const struct in_ifaddr *ifa;
 751
 752                         in_dev = __in_dev_get_rtnl(ndev);
 753                         if (!in_dev)
 754                                 goto put_noaddr;
 755
 756                         ifa = rtnl_dereference(in_dev->ifa_list);
 757                         if (!ifa) {
 758 put_noaddr:
 759                                 np_err(np, "no IP address for %s, aborting\n",
 760                                        np->dev_name);
 761                                 err = -EDESTADDRREQ;
 762                                 goto put;
 763                         }
 764
 765                         np->local_ip.ip = ifa->ifa_local;
 766                         np_info(np, "local IP %pI4\n", &np->local_ip.ip);
 767                 } else {
 768 #if IS_ENABLED(CONFIG_IPV6)
 769                         struct inet6_dev *idev;
 770
 771                         err = -EDESTADDRREQ;
 772                         idev = __in6_dev_get(ndev);
 773                         if (idev) {
 774                                 struct inet6_ifaddr *ifp;
 775
 776                                 read_lock_bh(&idev->lock);
 777                                 list_for_each_entry(ifp, &idev->addr_list, if_list) {
 778                                         if (!!(ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL) !=
 779                                             !!(ipv6_addr_type(&np->remote_ip.in6) & IPV6_ADDR_LINKLOCAL))
 780                                                 continue;
 781                                         np->local_ip.in6 = ifp->addr;
 782                                         err = 0;
 783                                         break;
 784                                 }
 785                                 read_unlock_bh(&idev->lock);
 786                         }
 787                         if (err) {
 788                                 np_err(np, "no IPv6 address for %s, aborting\n",
 789                                        np->dev_name);
 790                                 goto put;
 791                         } else
 792                                 np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
 793 #else
 794                         np_err(np, "IPv6 is not supported %s, aborting\n",
 795                                np->dev_name);
 796                         err = -EINVAL;
 797                         goto put;
 798 #endif
 799                 }
 800         }
 801
 802         /* fill up the skb queue */
 803         refill_skbs();
 804
 805         err = __netpoll_setup(np, ndev);
 806         if (err)
 807                 goto put;
 808
 809         rtnl_unlock();
 810         return 0;
 811
 812 put:
 813         dev_put(ndev);
 814 unlock:
 815         rtnl_unlock();
 816         return err;
 817 }
 818 EXPORT_SYMBOL(netpoll_setup);
 819
 820 static int __init netpoll_init(void)
 821 {
 822         skb_queue_head_init(&skb_pool);
 823         return 0;
 824 }
 825 core_initcall(netpoll_init);
 826
 827 static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
 828 {
 829         struct netpoll_info *npinfo =
 830                         container_of(rcu_head, struct netpoll_info, rcu);
 831
 832         skb_queue_purge(&npinfo->txq);
 833
 834         /* we can't call cancel_delayed_work_sync here, as we are in softirq */
 835         cancel_delayed_work(&npinfo->tx_work);
 836
 837         /* clean after last, unfinished work */
 838         __skb_queue_purge(&npinfo->txq);
 839         /* now cancel it again */
 840         cancel_delayed_work(&npinfo->tx_work);
 841         kfree(npinfo);
 842 }
 843
 844 void __netpoll_cleanup(struct netpoll *np)
 845 {
 846         struct netpoll_info *npinfo;
 847
 848         npinfo = rtnl_dereference(np->dev->npinfo);
 849         if (!npinfo)
 850                 return;
 851
 852         synchronize_srcu(&netpoll_srcu);
 853
 854         if (refcount_dec_and_test(&npinfo->refcnt)) {
 855                 const struct net_device_ops *ops;
 856
 857                 ops = np->dev->netdev_ops;
 858                 if (ops->ndo_netpoll_cleanup)
 859                         ops->ndo_netpoll_cleanup(np->dev);
 860
 861                 RCU_INIT_POINTER(np->dev->npinfo, NULL);
 862                 call_rcu(&npinfo->rcu, rcu_cleanup_netpoll_info);
 863         } else
 864                 RCU_INIT_POINTER(np->dev->npinfo, NULL);
 865 }
 866 EXPORT_SYMBOL_GPL(__netpoll_cleanup);
 867
 868 void __netpoll_free(struct netpoll *np)
 869 {
 870         ASSERT_RTNL();
 871
 872         /* Wait for transmitting packets to finish before freeing. */
 873         synchronize_rcu();
 874         __netpoll_cleanup(np);
 875         kfree(np);
 876 }
 877 EXPORT_SYMBOL_GPL(__netpoll_free);
 878
 879 void netpoll_cleanup(struct netpoll *np)
 880 {
 881         rtnl_lock();
 882         if (!np->dev)
 883                 goto out;
 884         __netpoll_cleanup(np);
 885         dev_put(np->dev);
 886         np->dev = NULL;
 887 out:
 888         rtnl_unlock();
 889 }
 890 EXPORT_SYMBOL(netpoll_cleanup);