net/core/netpoll.c

   1 /*
   2  * Common framework for low-level network console, dump, and debugger code
   3  *
   4  * Sep 8 2003  Matt Mackall <mpm@selenic.com>
   5  *
   6  * based on the netconsole code from:
   7  *
   8  * Copyright (C) 2001  Ingo Molnar <mingo@redhat.com>
   9  * Copyright (C) 2002  Red Hat, Inc.
  10  */
  11
  12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  13
  14 #include <linux/moduleparam.h>
  15 #include <linux/netdevice.h>
  16 #include <linux/etherdevice.h>
  17 #include <linux/string.h>
  18 #include <linux/if_arp.h>
  19 #include <linux/inetdevice.h>
  20 #include <linux/inet.h>
  21 #include <linux/interrupt.h>
  22 #include <linux/netpoll.h>
  23 #include <linux/sched.h>
  24 #include <linux/delay.h>
  25 #include <linux/rcupdate.h>
  26 #include <linux/workqueue.h>
  27 #include <linux/slab.h>
  28 #include <linux/export.h>
  29 #include <linux/if_vlan.h>
  30 #include <net/tcp.h>
  31 #include <net/udp.h>
  32 #include <net/addrconf.h>
  33 #include <net/ndisc.h>
  34 #include <net/ip6_checksum.h>
  35 #include <asm/unaligned.h>
  36 #include <trace/events/napi.h>
  37
  38 /*
  39  * We maintain a small pool of fully-sized skbs, to make sure the
  40  * message gets out even in extreme OOM situations.
  41  */
  42
  43 #define MAX_UDP_CHUNK 1460
  44 #define MAX_SKBS 32
  45
  46 static struct sk_buff_head skb_pool;
  47
  48 static atomic_t trapped;
  49
  50 #define USEC_PER_POLL   50
  51 #define NETPOLL_RX_ENABLED  1
  52 #define NETPOLL_RX_DROP     2
  53
  54 #define MAX_SKB_SIZE                                                    \
  55         (sizeof(struct ethhdr) +                                        \
  56          sizeof(struct iphdr) +                                         \
  57          sizeof(struct udphdr) +                                        \
  58          MAX_UDP_CHUNK)
  59
  60 static void zap_completion_queue(void);
  61 static void netpoll_neigh_reply(struct sk_buff *skb, struct netpoll_info *npinfo);
  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 void queue_process(struct work_struct *work)
  74 {
  75         struct netpoll_info *npinfo =
  76                 container_of(work, struct netpoll_info, tx_work.work);
  77         struct sk_buff *skb;
  78         unsigned long flags;
  79
  80         while ((skb = skb_dequeue(&npinfo->txq))) {
  81                 struct net_device *dev = skb->dev;
  82                 const struct net_device_ops *ops = dev->netdev_ops;
  83                 struct netdev_queue *txq;
  84
  85                 if (!netif_device_present(dev) || !netif_running(dev)) {
  86                         __kfree_skb(skb);
  87                         continue;
  88                 }
  89
  90                 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
  91
  92                 local_irq_save(flags);
  93                 __netif_tx_lock(txq, smp_processor_id());
  94                 if (netif_xmit_frozen_or_stopped(txq) ||
  95                     ops->ndo_start_xmit(skb, dev) != NETDEV_TX_OK) {
  96                         skb_queue_head(&npinfo->txq, skb);
  97                         __netif_tx_unlock(txq);
  98                         local_irq_restore(flags);
  99
 100                         schedule_delayed_work(&npinfo->tx_work, HZ/10);
 101                         return;
 102                 }
 103                 __netif_tx_unlock(txq);
 104                 local_irq_restore(flags);
 105         }
 106 }
 107
 108 static __sum16 checksum_udp(struct sk_buff *skb, struct udphdr *uh,
 109                             unsigned short ulen, __be32 saddr, __be32 daddr)
 110 {
 111         __wsum psum;
 112
 113         if (uh->check == 0 || skb_csum_unnecessary(skb))
 114                 return 0;
 115
 116         psum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0);
 117
 118         if (skb->ip_summed == CHECKSUM_COMPLETE &&
 119             !csum_fold(csum_add(psum, skb->csum)))
 120                 return 0;
 121
 122         skb->csum = psum;
 123
 124         return __skb_checksum_complete(skb);
 125 }
 126
 127 /*
 128  * Check whether delayed processing was scheduled for our NIC. If so,
 129  * we attempt to grab the poll lock and use ->poll() to pump the card.
 130  * If this fails, either we've recursed in ->poll() or it's already
 131  * running on another CPU.
 132  *
 133  * Note: we don't mask interrupts with this lock because we're using
 134  * trylock here and interrupts are already disabled in the softirq
 135  * case. Further, we test the poll_owner to avoid recursion on UP
 136  * systems where the lock doesn't exist.
 137  *
 138  * In cases where there is bi-directional communications, reading only
 139  * one message at a time can lead to packets being dropped by the
 140  * network adapter, forcing superfluous retries and possibly timeouts.
 141  * Thus, we set our budget to greater than 1.
 142  */
 143 static int poll_one_napi(struct netpoll_info *npinfo,
 144                          struct napi_struct *napi, int budget)
 145 {
 146         int work;
 147
 148         /* net_rx_action's ->poll() invocations and our's are
 149          * synchronized by this test which is only made while
 150          * holding the napi->poll_lock.
 151          */
 152         if (!test_bit(NAPI_STATE_SCHED, &napi->state))
 153                 return budget;
 154
 155         npinfo->rx_flags |= NETPOLL_RX_DROP;
 156         atomic_inc(&trapped);
 157         set_bit(NAPI_STATE_NPSVC, &napi->state);
 158
 159         work = napi->poll(napi, budget);
 160         trace_napi_poll(napi);
 161
 162         clear_bit(NAPI_STATE_NPSVC, &napi->state);
 163         atomic_dec(&trapped);
 164         npinfo->rx_flags &= ~NETPOLL_RX_DROP;
 165
 166         return budget - work;
 167 }
 168
 169 static void poll_napi(struct net_device *dev)
 170 {
 171         struct napi_struct *napi;
 172         int budget = 16;
 173
 174         list_for_each_entry(napi, &dev->napi_list, dev_list) {
 175                 if (napi->poll_owner != smp_processor_id() &&
 176                     spin_trylock(&napi->poll_lock)) {
 177                         budget = poll_one_napi(rcu_dereference_bh(dev->npinfo),
 178                                                napi, budget);
 179                         spin_unlock(&napi->poll_lock);
 180
 181                         if (!budget)
 182                                 break;
 183                 }
 184         }
 185 }
 186
 187 static void service_neigh_queue(struct netpoll_info *npi)
 188 {
 189         if (npi) {
 190                 struct sk_buff *skb;
 191
 192                 while ((skb = skb_dequeue(&npi->neigh_tx)))
 193                         netpoll_neigh_reply(skb, npi);
 194         }
 195 }
 196
 197 static void netpoll_poll_dev(struct net_device *dev)
 198 {
 199         const struct net_device_ops *ops;
 200         struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
 201
 202         if (!dev || !netif_running(dev))
 203                 return;
 204
 205         ops = dev->netdev_ops;
 206         if (!ops->ndo_poll_controller)
 207                 return;
 208
 209         /* Process pending work on NIC */
 210         ops->ndo_poll_controller(dev);
 211
 212         poll_napi(dev);
 213
 214         if (dev->flags & IFF_SLAVE) {
 215                 if (ni) {
 216                         struct net_device *bond_dev;
 217                         struct sk_buff *skb;
 218                         struct netpoll_info *bond_ni;
 219
 220                         bond_dev = netdev_master_upper_dev_get_rcu(dev);
 221                         bond_ni = rcu_dereference_bh(bond_dev->npinfo);
 222                         while ((skb = skb_dequeue(&ni->neigh_tx))) {
 223                                 skb->dev = bond_dev;
 224                                 skb_queue_tail(&bond_ni->neigh_tx, skb);
 225                         }
 226                 }
 227         }
 228
 229         service_neigh_queue(ni);
 230
 231         zap_completion_queue();
 232 }
 233
 234 static void refill_skbs(void)
 235 {
 236         struct sk_buff *skb;
 237         unsigned long flags;
 238
 239         spin_lock_irqsave(&skb_pool.lock, flags);
 240         while (skb_pool.qlen < MAX_SKBS) {
 241                 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
 242                 if (!skb)
 243                         break;
 244
 245                 __skb_queue_tail(&skb_pool, skb);
 246         }
 247         spin_unlock_irqrestore(&skb_pool.lock, flags);
 248 }
 249
 250 static void zap_completion_queue(void)
 251 {
 252         unsigned long flags;
 253         struct softnet_data *sd = &get_cpu_var(softnet_data);
 254
 255         if (sd->completion_queue) {
 256                 struct sk_buff *clist;
 257
 258                 local_irq_save(flags);
 259                 clist = sd->completion_queue;
 260                 sd->completion_queue = NULL;
 261                 local_irq_restore(flags);
 262
 263                 while (clist != NULL) {
 264                         struct sk_buff *skb = clist;
 265                         clist = clist->next;
 266                         if (skb->destructor) {
 267                                 atomic_inc(&skb->users);
 268                                 dev_kfree_skb_any(skb); /* put this one back */
 269                         } else {
 270                                 __kfree_skb(skb);
 271                         }
 272                 }
 273         }
 274
 275         put_cpu_var(softnet_data);
 276 }
 277
 278 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
 279 {
 280         int count = 0;
 281         struct sk_buff *skb;
 282
 283         zap_completion_queue();
 284         refill_skbs();
 285 repeat:
 286
 287         skb = alloc_skb(len, GFP_ATOMIC);
 288         if (!skb)
 289                 skb = skb_dequeue(&skb_pool);
 290
 291         if (!skb) {
 292                 if (++count < 10) {
 293                         netpoll_poll_dev(np->dev);
 294                         goto repeat;
 295                 }
 296                 return NULL;
 297         }
 298
 299         atomic_set(&skb->users, 1);
 300         skb_reserve(skb, reserve);
 301         return skb;
 302 }
 303
 304 static int netpoll_owner_active(struct net_device *dev)
 305 {
 306         struct napi_struct *napi;
 307
 308         list_for_each_entry(napi, &dev->napi_list, dev_list) {
 309                 if (napi->poll_owner == smp_processor_id())
 310                         return 1;
 311         }
 312         return 0;
 313 }
 314
 315 /* call with IRQ disabled */
 316 void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb,
 317                              struct net_device *dev)
 318 {
 319         int status = NETDEV_TX_BUSY;
 320         unsigned long tries;
 321         const struct net_device_ops *ops = dev->netdev_ops;
 322         /* It is up to the caller to keep npinfo alive. */
 323         struct netpoll_info *npinfo;
 324
 325         WARN_ON_ONCE(!irqs_disabled());
 326
 327         npinfo = rcu_dereference_bh(np->dev->npinfo);
 328         if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
 329                 __kfree_skb(skb);
 330                 return;
 331         }
 332
 333         /* don't get messages out of order, and no recursion */
 334         if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
 335                 struct netdev_queue *txq;
 336
 337                 txq = netdev_pick_tx(dev, skb);
 338
 339                 /* try until next clock tick */
 340                 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
 341                      tries > 0; --tries) {
 342                         if (__netif_tx_trylock(txq)) {
 343                                 if (!netif_xmit_stopped(txq)) {
 344                                         if (vlan_tx_tag_present(skb) &&
 345                                             !(netif_skb_features(skb) & NETIF_F_HW_VLAN_TX)) {
 346                                                 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
 347                                                 if (unlikely(!skb))
 348                                                         break;
 349                                                 skb->vlan_tci = 0;
 350                                         }
 351
 352                                         status = ops->ndo_start_xmit(skb, dev);
 353                                         if (status == NETDEV_TX_OK)
 354                                                 txq_trans_update(txq);
 355                                 }
 356                                 __netif_tx_unlock(txq);
 357
 358                                 if (status == NETDEV_TX_OK)
 359                                         break;
 360
 361                         }
 362
 363                         /* tickle device maybe there is some cleanup */
 364                         netpoll_poll_dev(np->dev);
 365
 366                         udelay(USEC_PER_POLL);
 367                 }
 368
 369                 WARN_ONCE(!irqs_disabled(),
 370                         "netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pF)\n",
 371                         dev->name, ops->ndo_start_xmit);
 372
 373         }
 374
 375         if (status != NETDEV_TX_OK) {
 376                 skb_queue_tail(&npinfo->txq, skb);
 377                 schedule_delayed_work(&npinfo->tx_work,0);
 378         }
 379 }
 380 EXPORT_SYMBOL(netpoll_send_skb_on_dev);
 381
 382 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
 383 {
 384         int total_len, ip_len, udp_len;
 385         struct sk_buff *skb;
 386         struct udphdr *udph;
 387         struct iphdr *iph;
 388         struct ethhdr *eth;
 389         static atomic_t ip_ident;
 390         struct ipv6hdr *ip6h;
 391
 392         udp_len = len + sizeof(*udph);
 393         if (np->ipv6)
 394                 ip_len = udp_len + sizeof(*ip6h);
 395         else
 396                 ip_len = udp_len + sizeof(*iph);
 397
 398         total_len = ip_len + LL_RESERVED_SPACE(np->dev);
 399
 400         skb = find_skb(np, total_len + np->dev->needed_tailroom,
 401                        total_len - len);
 402         if (!skb)
 403                 return;
 404
 405         skb_copy_to_linear_data(skb, msg, len);
 406         skb_put(skb, len);
 407
 408         skb_push(skb, sizeof(*udph));
 409         skb_reset_transport_header(skb);
 410         udph = udp_hdr(skb);
 411         udph->source = htons(np->local_port);
 412         udph->dest = htons(np->remote_port);
 413         udph->len = htons(udp_len);
 414
 415         if (np->ipv6) {
 416                 udph->check = 0;
 417                 udph->check = csum_ipv6_magic(&np->local_ip.in6,
 418                                               &np->remote_ip.in6,
 419                                               udp_len, IPPROTO_UDP,
 420                                               csum_partial(udph, udp_len, 0));
 421                 if (udph->check == 0)
 422                         udph->check = CSUM_MANGLED_0;
 423
 424                 skb_push(skb, sizeof(*ip6h));
 425                 skb_reset_network_header(skb);
 426                 ip6h = ipv6_hdr(skb);
 427
 428                 /* ip6h->version = 6; ip6h->priority = 0; */
 429                 put_unaligned(0x60, (unsigned char *)ip6h);
 430                 ip6h->flow_lbl[0] = 0;
 431                 ip6h->flow_lbl[1] = 0;
 432                 ip6h->flow_lbl[2] = 0;
 433
 434                 ip6h->payload_len = htons(sizeof(struct udphdr) + len);
 435                 ip6h->nexthdr = IPPROTO_UDP;
 436                 ip6h->hop_limit = 32;
 437                 ip6h->saddr = np->local_ip.in6;
 438                 ip6h->daddr = np->remote_ip.in6;
 439
 440                 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
 441                 skb_reset_mac_header(skb);
 442                 skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
 443         } else {
 444                 udph->check = 0;
 445                 udph->check = csum_tcpudp_magic(np->local_ip.ip,
 446                                                 np->remote_ip.ip,
 447                                                 udp_len, IPPROTO_UDP,
 448                                                 csum_partial(udph, udp_len, 0));
 449                 if (udph->check == 0)
 450                         udph->check = CSUM_MANGLED_0;
 451
 452                 skb_push(skb, sizeof(*iph));
 453                 skb_reset_network_header(skb);
 454                 iph = ip_hdr(skb);
 455
 456                 /* iph->version = 4; iph->ihl = 5; */
 457                 put_unaligned(0x45, (unsigned char *)iph);
 458                 iph->tos      = 0;
 459                 put_unaligned(htons(ip_len), &(iph->tot_len));
 460                 iph->id       = htons(atomic_inc_return(&ip_ident));
 461                 iph->frag_off = 0;
 462                 iph->ttl      = 64;
 463                 iph->protocol = IPPROTO_UDP;
 464                 iph->check    = 0;
 465                 put_unaligned(np->local_ip.ip, &(iph->saddr));
 466                 put_unaligned(np->remote_ip.ip, &(iph->daddr));
 467                 iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
 468
 469                 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
 470                 skb_reset_mac_header(skb);
 471                 skb->protocol = eth->h_proto = htons(ETH_P_IP);
 472         }
 473
 474         memcpy(eth->h_source, np->dev->dev_addr, ETH_ALEN);
 475         memcpy(eth->h_dest, np->remote_mac, ETH_ALEN);
 476
 477         skb->dev = np->dev;
 478
 479         netpoll_send_skb(np, skb);
 480 }
 481 EXPORT_SYMBOL(netpoll_send_udp);
 482
 483 static void netpoll_neigh_reply(struct sk_buff *skb, struct netpoll_info *npinfo)
 484 {
 485         int size, type = ARPOP_REPLY;
 486         __be32 sip, tip;
 487         unsigned char *sha;
 488         struct sk_buff *send_skb;
 489         struct netpoll *np, *tmp;
 490         unsigned long flags;
 491         int hlen, tlen;
 492         int hits = 0, proto;
 493
 494         if (list_empty(&npinfo->rx_np))
 495                 return;
 496
 497         /* Before checking the packet, we do some early
 498            inspection whether this is interesting at all */
 499         spin_lock_irqsave(&npinfo->rx_lock, flags);
 500         list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
 501                 if (np->dev == skb->dev)
 502                         hits++;
 503         }
 504         spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 505
 506         /* No netpoll struct is using this dev */
 507         if (!hits)
 508                 return;
 509
 510         proto = ntohs(eth_hdr(skb)->h_proto);
 511         if (proto == ETH_P_IP) {
 512                 struct arphdr *arp;
 513                 unsigned char *arp_ptr;
 514                 /* No arp on this interface */
 515                 if (skb->dev->flags & IFF_NOARP)
 516                         return;
 517
 518                 if (!pskb_may_pull(skb, arp_hdr_len(skb->dev)))
 519                         return;
 520
 521                 skb_reset_network_header(skb);
 522                 skb_reset_transport_header(skb);
 523                 arp = arp_hdr(skb);
 524
 525                 if ((arp->ar_hrd != htons(ARPHRD_ETHER) &&
 526                      arp->ar_hrd != htons(ARPHRD_IEEE802)) ||
 527                     arp->ar_pro != htons(ETH_P_IP) ||
 528                     arp->ar_op != htons(ARPOP_REQUEST))
 529                         return;
 530
 531                 arp_ptr = (unsigned char *)(arp+1);
 532                 /* save the location of the src hw addr */
 533                 sha = arp_ptr;
 534                 arp_ptr += skb->dev->addr_len;
 535                 memcpy(&sip, arp_ptr, 4);
 536                 arp_ptr += 4;
 537                 /* If we actually cared about dst hw addr,
 538                    it would get copied here */
 539                 arp_ptr += skb->dev->addr_len;
 540                 memcpy(&tip, arp_ptr, 4);
 541
 542                 /* Should we ignore arp? */
 543                 if (ipv4_is_loopback(tip) || ipv4_is_multicast(tip))
 544                         return;
 545
 546                 size = arp_hdr_len(skb->dev);
 547
 548                 spin_lock_irqsave(&npinfo->rx_lock, flags);
 549                 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
 550                         if (tip != np->local_ip.ip)
 551                                 continue;
 552
 553                         hlen = LL_RESERVED_SPACE(np->dev);
 554                         tlen = np->dev->needed_tailroom;
 555                         send_skb = find_skb(np, size + hlen + tlen, hlen);
 556                         if (!send_skb)
 557                                 continue;
 558
 559                         skb_reset_network_header(send_skb);
 560                         arp = (struct arphdr *) skb_put(send_skb, size);
 561                         send_skb->dev = skb->dev;
 562                         send_skb->protocol = htons(ETH_P_ARP);
 563
 564                         /* Fill the device header for the ARP frame */
 565                         if (dev_hard_header(send_skb, skb->dev, ETH_P_ARP,
 566                                             sha, np->dev->dev_addr,
 567                                             send_skb->len) < 0) {
 568                                 kfree_skb(send_skb);
 569                                 continue;
 570                         }
 571
 572                         /*
 573                          * Fill out the arp protocol part.
 574                          *
 575                          * we only support ethernet device type,
 576                          * which (according to RFC 1390) should
 577                          * always equal 1 (Ethernet).
 578                          */
 579
 580                         arp->ar_hrd = htons(np->dev->type);
 581                         arp->ar_pro = htons(ETH_P_IP);
 582                         arp->ar_hln = np->dev->addr_len;
 583                         arp->ar_pln = 4;
 584                         arp->ar_op = htons(type);
 585
 586                         arp_ptr = (unsigned char *)(arp + 1);
 587                         memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
 588                         arp_ptr += np->dev->addr_len;
 589                         memcpy(arp_ptr, &tip, 4);
 590                         arp_ptr += 4;
 591                         memcpy(arp_ptr, sha, np->dev->addr_len);
 592                         arp_ptr += np->dev->addr_len;
 593                         memcpy(arp_ptr, &sip, 4);
 594
 595                         netpoll_send_skb(np, send_skb);
 596
 597                         /* If there are several rx_hooks for the same address,
 598                            we're fine by sending a single reply */
 599                         break;
 600                 }
 601                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 602         } else if( proto == ETH_P_IPV6) {
 603 #if IS_ENABLED(CONFIG_IPV6)
 604                 struct nd_msg *msg;
 605                 u8 *lladdr = NULL;
 606                 struct ipv6hdr *hdr;
 607                 struct icmp6hdr *icmp6h;
 608                 const struct in6_addr *saddr;
 609                 const struct in6_addr *daddr;
 610                 struct inet6_dev *in6_dev = NULL;
 611                 struct in6_addr *target;
 612
 613                 in6_dev = in6_dev_get(skb->dev);
 614                 if (!in6_dev || !in6_dev->cnf.accept_ra)
 615                         return;
 616
 617                 if (!pskb_may_pull(skb, skb->len))
 618                         return;
 619
 620                 msg = (struct nd_msg *)skb_transport_header(skb);
 621
 622                 __skb_push(skb, skb->data - skb_transport_header(skb));
 623
 624                 if (ipv6_hdr(skb)->hop_limit != 255)
 625                         return;
 626                 if (msg->icmph.icmp6_code != 0)
 627                         return;
 628                 if (msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
 629                         return;
 630
 631                 saddr = &ipv6_hdr(skb)->saddr;
 632                 daddr = &ipv6_hdr(skb)->daddr;
 633
 634                 size = sizeof(struct icmp6hdr) + sizeof(struct in6_addr);
 635
 636                 spin_lock_irqsave(&npinfo->rx_lock, flags);
 637                 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
 638                         if (!ipv6_addr_equal(daddr, &np->local_ip.in6))
 639                                 continue;
 640
 641                         hlen = LL_RESERVED_SPACE(np->dev);
 642                         tlen = np->dev->needed_tailroom;
 643                         send_skb = find_skb(np, size + hlen + tlen, hlen);
 644                         if (!send_skb)
 645                                 continue;
 646
 647                         send_skb->protocol = htons(ETH_P_IPV6);
 648                         send_skb->dev = skb->dev;
 649
 650                         skb_reset_network_header(send_skb);
 651                         skb_put(send_skb, sizeof(struct ipv6hdr));
 652                         hdr = ipv6_hdr(send_skb);
 653
 654                         *(__be32*)hdr = htonl(0x60000000);
 655
 656                         hdr->payload_len = htons(size);
 657                         hdr->nexthdr = IPPROTO_ICMPV6;
 658                         hdr->hop_limit = 255;
 659                         hdr->saddr = *saddr;
 660                         hdr->daddr = *daddr;
 661
 662                         send_skb->transport_header = send_skb->tail;
 663                         skb_put(send_skb, size);
 664
 665                         icmp6h = (struct icmp6hdr *)skb_transport_header(skb);
 666                         icmp6h->icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
 667                         icmp6h->icmp6_router = 0;
 668                         icmp6h->icmp6_solicited = 1;
 669                         target = (struct in6_addr *)skb_transport_header(send_skb) + sizeof(struct icmp6hdr);
 670                         *target = msg->target;
 671                         icmp6h->icmp6_cksum = csum_ipv6_magic(saddr, daddr, size,
 672                                                               IPPROTO_ICMPV6,
 673                                                               csum_partial(icmp6h,
 674                                                                            size, 0));
 675
 676                         if (dev_hard_header(send_skb, skb->dev, ETH_P_IPV6,
 677                                             lladdr, np->dev->dev_addr,
 678                                             send_skb->len) < 0) {
 679                                 kfree_skb(send_skb);
 680                                 continue;
 681                         }
 682
 683                         netpoll_send_skb(np, send_skb);
 684
 685                         /* If there are several rx_hooks for the same address,
 686                            we're fine by sending a single reply */
 687                         break;
 688                 }
 689                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 690 #endif
 691         }
 692 }
 693
 694 static bool pkt_is_ns(struct sk_buff *skb)
 695 {
 696         struct nd_msg *msg;
 697         struct ipv6hdr *hdr;
 698
 699         if (skb->protocol != htons(ETH_P_ARP))
 700                 return false;
 701         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr) + sizeof(struct nd_msg)))
 702                 return false;
 703
 704         msg = (struct nd_msg *)skb_transport_header(skb);
 705         __skb_push(skb, skb->data - skb_transport_header(skb));
 706         hdr = ipv6_hdr(skb);
 707
 708         if (hdr->nexthdr != IPPROTO_ICMPV6)
 709                 return false;
 710         if (hdr->hop_limit != 255)
 711                 return false;
 712         if (msg->icmph.icmp6_code != 0)
 713                 return false;
 714         if (msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
 715                 return false;
 716
 717         return true;
 718 }
 719
 720 int __netpoll_rx(struct sk_buff *skb, struct netpoll_info *npinfo)
 721 {
 722         int proto, len, ulen;
 723         int hits = 0;
 724         const struct iphdr *iph;
 725         struct udphdr *uh;
 726         struct netpoll *np, *tmp;
 727
 728         if (list_empty(&npinfo->rx_np))
 729                 goto out;
 730
 731         if (skb->dev->type != ARPHRD_ETHER)
 732                 goto out;
 733
 734         /* check if netpoll clients need ARP */
 735         if (skb->protocol == htons(ETH_P_ARP) && atomic_read(&trapped)) {
 736                 skb_queue_tail(&npinfo->neigh_tx, skb);
 737                 return 1;
 738         } else if (pkt_is_ns(skb) && atomic_read(&trapped)) {
 739                 skb_queue_tail(&npinfo->neigh_tx, skb);
 740                 return 1;
 741         }
 742
 743         if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
 744                 skb = vlan_untag(skb);
 745                 if (unlikely(!skb))
 746                         goto out;
 747         }
 748
 749         proto = ntohs(eth_hdr(skb)->h_proto);
 750         if (proto != ETH_P_IP && proto != ETH_P_IPV6)
 751                 goto out;
 752         if (skb->pkt_type == PACKET_OTHERHOST)
 753                 goto out;
 754         if (skb_shared(skb))
 755                 goto out;
 756
 757         if (proto == ETH_P_IP) {
 758                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 759                         goto out;
 760                 iph = (struct iphdr *)skb->data;
 761                 if (iph->ihl < 5 || iph->version != 4)
 762                         goto out;
 763                 if (!pskb_may_pull(skb, iph->ihl*4))
 764                         goto out;
 765                 iph = (struct iphdr *)skb->data;
 766                 if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
 767                         goto out;
 768
 769                 len = ntohs(iph->tot_len);
 770                 if (skb->len < len || len < iph->ihl*4)
 771                         goto out;
 772
 773                 /*
 774                  * Our transport medium may have padded the buffer out.
 775                  * Now We trim to the true length of the frame.
 776                  */
 777                 if (pskb_trim_rcsum(skb, len))
 778                         goto out;
 779
 780                 iph = (struct iphdr *)skb->data;
 781                 if (iph->protocol != IPPROTO_UDP)
 782                         goto out;
 783
 784                 len -= iph->ihl*4;
 785                 uh = (struct udphdr *)(((char *)iph) + iph->ihl*4);
 786                 ulen = ntohs(uh->len);
 787
 788                 if (ulen != len)
 789                         goto out;
 790                 if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr))
 791                         goto out;
 792                 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
 793                         if (np->local_ip.ip && np->local_ip.ip != iph->daddr)
 794                                 continue;
 795                         if (np->remote_ip.ip && np->remote_ip.ip != iph->saddr)
 796                                 continue;
 797                         if (np->local_port && np->local_port != ntohs(uh->dest))
 798                                 continue;
 799
 800                         np->rx_hook(np, ntohs(uh->source),
 801                                        (char *)(uh+1),
 802                                        ulen - sizeof(struct udphdr));
 803                         hits++;
 804                 }
 805         } else {
 806 #if IS_ENABLED(CONFIG_IPV6)
 807                 const struct ipv6hdr *ip6h;
 808
 809                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 810                         goto out;
 811                 ip6h = (struct ipv6hdr *)skb->data;
 812                 if (ip6h->version != 6)
 813                         goto out;
 814                 len = ntohs(ip6h->payload_len);
 815                 if (!len)
 816                         goto out;
 817                 if (len + sizeof(struct ipv6hdr) > skb->len)
 818                         goto out;
 819                 if (pskb_trim_rcsum(skb, len + sizeof(struct ipv6hdr)))
 820                         goto out;
 821                 ip6h = ipv6_hdr(skb);
 822                 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
 823                         goto out;
 824                 uh = udp_hdr(skb);
 825                 ulen = ntohs(uh->len);
 826                 if (ulen != skb->len)
 827                         goto out;
 828                 if (udp6_csum_init(skb, uh, IPPROTO_UDP))
 829                         goto out;
 830                 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
 831                         if (!ipv6_addr_equal(&np->local_ip.in6, &ip6h->daddr))
 832                                 continue;
 833                         if (!ipv6_addr_equal(&np->remote_ip.in6, &ip6h->saddr))
 834                                 continue;
 835                         if (np->local_port && np->local_port != ntohs(uh->dest))
 836                                 continue;
 837
 838                         np->rx_hook(np, ntohs(uh->source),
 839                                        (char *)(uh+1),
 840                                        ulen - sizeof(struct udphdr));
 841                         hits++;
 842                 }
 843 #endif
 844         }
 845
 846         if (!hits)
 847                 goto out;
 848
 849         kfree_skb(skb);
 850         return 1;
 851
 852 out:
 853         if (atomic_read(&trapped)) {
 854                 kfree_skb(skb);
 855                 return 1;
 856         }
 857
 858         return 0;
 859 }
 860
 861 void netpoll_print_options(struct netpoll *np)
 862 {
 863         np_info(np, "local port %d\n", np->local_port);
 864         if (np->ipv6)
 865                 np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
 866         else
 867                 np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
 868         np_info(np, "interface '%s'\n", np->dev_name);
 869         np_info(np, "remote port %d\n", np->remote_port);
 870         if (np->ipv6)
 871                 np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
 872         else
 873                 np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
 874         np_info(np, "remote ethernet address %pM\n", np->remote_mac);
 875 }
 876 EXPORT_SYMBOL(netpoll_print_options);
 877
 878 static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
 879 {
 880         const char *end;
 881
 882         if (!strchr(str, ':') &&
 883             in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
 884                 if (!*end)
 885                         return 0;
 886         }
 887         if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
 888 #if IS_ENABLED(CONFIG_IPV6)
 889                 if (!*end)
 890                         return 1;
 891 #else
 892                 return -1;
 893 #endif
 894         }
 895         return -1;
 896 }
 897
 898 int netpoll_parse_options(struct netpoll *np, char *opt)
 899 {
 900         char *cur=opt, *delim;
 901         int ipv6;
 902
 903         if (*cur != '@') {
 904                 if ((delim = strchr(cur, '@')) == NULL)
 905                         goto parse_failed;
 906                 *delim = 0;
 907                 if (kstrtou16(cur, 10, &np->local_port))
 908                         goto parse_failed;
 909                 cur = delim;
 910         }
 911         cur++;
 912
 913         if (*cur != '/') {
 914                 if ((delim = strchr(cur, '/')) == NULL)
 915                         goto parse_failed;
 916                 *delim = 0;
 917                 ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
 918                 if (ipv6 < 0)
 919                         goto parse_failed;
 920                 else
 921                         np->ipv6 = (bool)ipv6;
 922                 cur = delim;
 923         }
 924         cur++;
 925
 926         if (*cur != ',') {
 927                 /* parse out dev name */
 928                 if ((delim = strchr(cur, ',')) == NULL)
 929                         goto parse_failed;
 930                 *delim = 0;
 931                 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
 932                 cur = delim;
 933         }
 934         cur++;
 935
 936         if (*cur != '@') {
 937                 /* dst port */
 938                 if ((delim = strchr(cur, '@')) == NULL)
 939                         goto parse_failed;
 940                 *delim = 0;
 941                 if (*cur == ' ' || *cur == '\t')
 942                         np_info(np, "warning: whitespace is not allowed\n");
 943                 if (kstrtou16(cur, 10, &np->remote_port))
 944                         goto parse_failed;
 945                 cur = delim;
 946         }
 947         cur++;
 948
 949         /* dst ip */
 950         if ((delim = strchr(cur, '/')) == NULL)
 951                 goto parse_failed;
 952         *delim = 0;
 953         ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
 954         if (ipv6 < 0)
 955                 goto parse_failed;
 956         else if (np->ipv6 != (bool)ipv6)
 957                 goto parse_failed;
 958         else
 959                 np->ipv6 = (bool)ipv6;
 960         cur = delim + 1;
 961
 962         if (*cur != 0) {
 963                 /* MAC address */
 964                 if (!mac_pton(cur, np->remote_mac))
 965                         goto parse_failed;
 966         }
 967
 968         netpoll_print_options(np);
 969
 970         return 0;
 971
 972  parse_failed:
 973         np_info(np, "couldn't parse config at '%s'!\n", cur);
 974         return -1;
 975 }
 976 EXPORT_SYMBOL(netpoll_parse_options);
 977
 978 int __netpoll_setup(struct netpoll *np, struct net_device *ndev, gfp_t gfp)
 979 {
 980         struct netpoll_info *npinfo;
 981         const struct net_device_ops *ops;
 982         unsigned long flags;
 983         int err;
 984
 985         np->dev = ndev;
 986         strlcpy(np->dev_name, ndev->name, IFNAMSIZ);
 987
 988         if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) ||
 989             !ndev->netdev_ops->ndo_poll_controller) {
 990                 np_err(np, "%s doesn't support polling, aborting\n",
 991                        np->dev_name);
 992                 err = -ENOTSUPP;
 993                 goto out;
 994         }
 995
 996         if (!ndev->npinfo) {
 997                 npinfo = kmalloc(sizeof(*npinfo), gfp);
 998                 if (!npinfo) {
 999                         err = -ENOMEM;
1000                         goto out;
1001                 }
1002
1003                 npinfo->rx_flags = 0;
1004                 INIT_LIST_HEAD(&npinfo->rx_np);
1005
1006                 spin_lock_init(&npinfo->rx_lock);
1007                 skb_queue_head_init(&npinfo->neigh_tx);
1008                 skb_queue_head_init(&npinfo->txq);
1009                 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
1010
1011                 atomic_set(&npinfo->refcnt, 1);
1012
1013                 ops = np->dev->netdev_ops;
1014                 if (ops->ndo_netpoll_setup) {
1015                         err = ops->ndo_netpoll_setup(ndev, npinfo, gfp);
1016                         if (err)
1017                                 goto free_npinfo;
1018                 }
1019         } else {
1020                 npinfo = ndev->npinfo;
1021                 atomic_inc(&npinfo->refcnt);
1022         }
1023
1024         npinfo->netpoll = np;
1025
1026         if (np->rx_hook) {
1027                 spin_lock_irqsave(&npinfo->rx_lock, flags);
1028                 npinfo->rx_flags |= NETPOLL_RX_ENABLED;
1029                 list_add_tail(&np->rx, &npinfo->rx_np);
1030                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
1031         }
1032
1033         /* last thing to do is link it to the net device structure */
1034         rcu_assign_pointer(ndev->npinfo, npinfo);
1035
1036         return 0;
1037
1038 free_npinfo:
1039         kfree(npinfo);
1040 out:
1041         return err;
1042 }
1043 EXPORT_SYMBOL_GPL(__netpoll_setup);
1044
1045 int netpoll_setup(struct netpoll *np)
1046 {
1047         struct net_device *ndev = NULL;
1048         struct in_device *in_dev;
1049         int err;
1050
1051         rtnl_lock();
1052         if (np->dev_name) {
1053                 struct net *net = current->nsproxy->net_ns;
1054                 ndev = __dev_get_by_name(net, np->dev_name);
1055         }
1056         if (!ndev) {
1057                 np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
1058                 err = -ENODEV;
1059                 goto unlock;
1060         }
1061         dev_hold(ndev);
1062
1063         if (netdev_master_upper_dev_get(ndev)) {
1064                 np_err(np, "%s is a slave device, aborting\n", np->dev_name);
1065                 err = -EBUSY;
1066                 goto put;
1067         }
1068
1069         if (!netif_running(ndev)) {
1070                 unsigned long atmost, atleast;
1071
1072                 np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
1073
1074                 err = dev_open(ndev);
1075
1076                 if (err) {
1077                         np_err(np, "failed to open %s\n", ndev->name);
1078                         goto put;
1079                 }
1080
1081                 rtnl_unlock();
1082                 atleast = jiffies + HZ/10;
1083                 atmost = jiffies + carrier_timeout * HZ;
1084                 while (!netif_carrier_ok(ndev)) {
1085                         if (time_after(jiffies, atmost)) {
1086                                 np_notice(np, "timeout waiting for carrier\n");
1087                                 break;
1088                         }
1089                         msleep(1);
1090                 }
1091
1092                 /* If carrier appears to come up instantly, we don't
1093                  * trust it and pause so that we don't pump all our
1094                  * queued console messages into the bitbucket.
1095                  */
1096
1097                 if (time_before(jiffies, atleast)) {
1098                         np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
1099                         msleep(4000);
1100                 }
1101                 rtnl_lock();
1102         }
1103
1104         if (!np->local_ip.ip) {
1105                 if (!np->ipv6) {
1106                         in_dev = __in_dev_get_rtnl(ndev);
1107
1108                         if (!in_dev || !in_dev->ifa_list) {
1109                                 np_err(np, "no IP address for %s, aborting\n",
1110                                        np->dev_name);
1111                                 err = -EDESTADDRREQ;
1112                                 goto put;
1113                         }
1114
1115                         np->local_ip.ip = in_dev->ifa_list->ifa_local;
1116                         np_info(np, "local IP %pI4\n", &np->local_ip.ip);
1117                 } else {
1118 #if IS_ENABLED(CONFIG_IPV6)
1119                         struct inet6_dev *idev;
1120
1121                         err = -EDESTADDRREQ;
1122                         idev = __in6_dev_get(ndev);
1123                         if (idev) {
1124                                 struct inet6_ifaddr *ifp;
1125
1126                                 read_lock_bh(&idev->lock);
1127                                 list_for_each_entry(ifp, &idev->addr_list, if_list) {
1128                                         if (ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL)
1129                                                 continue;
1130                                         np->local_ip.in6 = ifp->addr;
1131                                         err = 0;
1132                                         break;
1133                                 }
1134                                 read_unlock_bh(&idev->lock);
1135                         }
1136                         if (err) {
1137                                 np_err(np, "no IPv6 address for %s, aborting\n",
1138                                        np->dev_name);
1139                                 goto put;
1140                         } else
1141                                 np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
1142 #else
1143                         np_err(np, "IPv6 is not supported %s, aborting\n",
1144                                np->dev_name);
1145                         err = -EINVAL;
1146                         goto put;
1147 #endif
1148                 }
1149         }
1150
1151         /* fill up the skb queue */
1152         refill_skbs();
1153
1154         err = __netpoll_setup(np, ndev, GFP_KERNEL);
1155         if (err)
1156                 goto put;
1157
1158         rtnl_unlock();
1159         return 0;
1160
1161 put:
1162         dev_put(ndev);
1163 unlock:
1164         rtnl_unlock();
1165         return err;
1166 }
1167 EXPORT_SYMBOL(netpoll_setup);
1168
1169 static int __init netpoll_init(void)
1170 {
1171         skb_queue_head_init(&skb_pool);
1172         return 0;
1173 }
1174 core_initcall(netpoll_init);
1175
1176 static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
1177 {
1178         struct netpoll_info *npinfo =
1179                         container_of(rcu_head, struct netpoll_info, rcu);
1180
1181         skb_queue_purge(&npinfo->neigh_tx);
1182         skb_queue_purge(&npinfo->txq);
1183
1184         /* we can't call cancel_delayed_work_sync here, as we are in softirq */
1185         cancel_delayed_work(&npinfo->tx_work);
1186
1187         /* clean after last, unfinished work */
1188         __skb_queue_purge(&npinfo->txq);
1189         /* now cancel it again */
1190         cancel_delayed_work(&npinfo->tx_work);
1191         kfree(npinfo);
1192 }
1193
1194 void __netpoll_cleanup(struct netpoll *np)
1195 {
1196         struct netpoll_info *npinfo;
1197         unsigned long flags;
1198
1199         npinfo = np->dev->npinfo;
1200         if (!npinfo)
1201                 return;
1202
1203         if (!list_empty(&npinfo->rx_np)) {
1204                 spin_lock_irqsave(&npinfo->rx_lock, flags);
1205                 list_del(&np->rx);
1206                 if (list_empty(&npinfo->rx_np))
1207                         npinfo->rx_flags &= ~NETPOLL_RX_ENABLED;
1208                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
1209         }
1210
1211         if (atomic_dec_and_test(&npinfo->refcnt)) {
1212                 const struct net_device_ops *ops;
1213
1214                 ops = np->dev->netdev_ops;
1215                 if (ops->ndo_netpoll_cleanup)
1216                         ops->ndo_netpoll_cleanup(np->dev);
1217
1218                 RCU_INIT_POINTER(np->dev->npinfo, NULL);
1219                 call_rcu_bh(&npinfo->rcu, rcu_cleanup_netpoll_info);
1220         }
1221 }
1222 EXPORT_SYMBOL_GPL(__netpoll_cleanup);
1223
1224 static void rcu_cleanup_netpoll(struct rcu_head *rcu_head)
1225 {
1226         struct netpoll *np = container_of(rcu_head, struct netpoll, rcu);
1227
1228         __netpoll_cleanup(np);
1229         kfree(np);
1230 }
1231
1232 void __netpoll_free_rcu(struct netpoll *np)
1233 {
1234         call_rcu_bh(&np->rcu, rcu_cleanup_netpoll);
1235 }
1236 EXPORT_SYMBOL_GPL(__netpoll_free_rcu);
1237
1238 void netpoll_cleanup(struct netpoll *np)
1239 {
1240         if (!np->dev)
1241                 return;
1242
1243         rtnl_lock();
1244         __netpoll_cleanup(np);
1245         rtnl_unlock();
1246
1247         dev_put(np->dev);
1248         np->dev = NULL;
1249 }
1250 EXPORT_SYMBOL(netpoll_cleanup);
1251
1252 int netpoll_trap(void)
1253 {
1254         return atomic_read(&trapped);
1255 }
1256 EXPORT_SYMBOL(netpoll_trap);
1257
1258 void netpoll_set_trap(int trap)
1259 {
1260         if (trap)
1261                 atomic_inc(&trapped);
1262         else
1263                 atomic_dec(&trapped);
1264 }
1265 EXPORT_SYMBOL(netpoll_set_trap);