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 #include <linux/moduleparam.h>
  13 #include <linux/netdevice.h>
  14 #include <linux/etherdevice.h>
  15 #include <linux/string.h>
  16 #include <linux/if_arp.h>
  17 #include <linux/inetdevice.h>
  18 #include <linux/inet.h>
  19 #include <linux/interrupt.h>
  20 #include <linux/netpoll.h>
  21 #include <linux/sched.h>
  22 #include <linux/delay.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/workqueue.h>
  25 #include <linux/slab.h>
  26 #include <net/tcp.h>
  27 #include <net/udp.h>
  28 #include <asm/unaligned.h>
  29 #include <trace/events/napi.h>
  30
  31 /*
  32  * We maintain a small pool of fully-sized skbs, to make sure the
  33  * message gets out even in extreme OOM situations.
  34  */
  35
  36 #define MAX_UDP_CHUNK 1460
  37 #define MAX_SKBS 32
  38 #define MAX_QUEUE_DEPTH (MAX_SKBS / 2)
  39
  40 static struct sk_buff_head skb_pool;
  41
  42 static atomic_t trapped;
  43
  44 #define USEC_PER_POLL   50
  45 #define NETPOLL_RX_ENABLED  1
  46 #define NETPOLL_RX_DROP     2
  47
  48 #define MAX_SKB_SIZE \
  49                 (MAX_UDP_CHUNK + sizeof(struct udphdr) + \
  50                                 sizeof(struct iphdr) + sizeof(struct ethhdr))
  51
  52 static void arp_reply(struct sk_buff *skb);
  53
  54 static unsigned int carrier_timeout = 4;
  55 module_param(carrier_timeout, uint, 0644);
  56
  57 static void queue_process(struct work_struct *work)
  58 {
  59         struct netpoll_info *npinfo =
  60                 container_of(work, struct netpoll_info, tx_work.work);
  61         struct sk_buff *skb;
  62         unsigned long flags;
  63
  64         while ((skb = skb_dequeue(&npinfo->txq))) {
  65                 struct net_device *dev = skb->dev;
  66                 const struct net_device_ops *ops = dev->netdev_ops;
  67                 struct netdev_queue *txq;
  68
  69                 if (!netif_device_present(dev) || !netif_running(dev)) {
  70                         __kfree_skb(skb);
  71                         continue;
  72                 }
  73
  74                 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
  75
  76                 local_irq_save(flags);
  77                 __netif_tx_lock(txq, smp_processor_id());
  78                 if (netif_tx_queue_stopped(txq) ||
  79                     netif_tx_queue_frozen(txq) ||
  80                     ops->ndo_start_xmit(skb, dev) != NETDEV_TX_OK) {
  81                         skb_queue_head(&npinfo->txq, skb);
  82                         __netif_tx_unlock(txq);
  83                         local_irq_restore(flags);
  84
  85                         schedule_delayed_work(&npinfo->tx_work, HZ/10);
  86                         return;
  87                 }
  88                 __netif_tx_unlock(txq);
  89                 local_irq_restore(flags);
  90         }
  91 }
  92
  93 static __sum16 checksum_udp(struct sk_buff *skb, struct udphdr *uh,
  94                             unsigned short ulen, __be32 saddr, __be32 daddr)
  95 {
  96         __wsum psum;
  97
  98         if (uh->check == 0 || skb_csum_unnecessary(skb))
  99                 return 0;
 100
 101         psum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0);
 102
 103         if (skb->ip_summed == CHECKSUM_COMPLETE &&
 104             !csum_fold(csum_add(psum, skb->csum)))
 105                 return 0;
 106
 107         skb->csum = psum;
 108
 109         return __skb_checksum_complete(skb);
 110 }
 111
 112 /*
 113  * Check whether delayed processing was scheduled for our NIC. If so,
 114  * we attempt to grab the poll lock and use ->poll() to pump the card.
 115  * If this fails, either we've recursed in ->poll() or it's already
 116  * running on another CPU.
 117  *
 118  * Note: we don't mask interrupts with this lock because we're using
 119  * trylock here and interrupts are already disabled in the softirq
 120  * case. Further, we test the poll_owner to avoid recursion on UP
 121  * systems where the lock doesn't exist.
 122  *
 123  * In cases where there is bi-directional communications, reading only
 124  * one message at a time can lead to packets being dropped by the
 125  * network adapter, forcing superfluous retries and possibly timeouts.
 126  * Thus, we set our budget to greater than 1.
 127  */
 128 static int poll_one_napi(struct netpoll_info *npinfo,
 129                          struct napi_struct *napi, int budget)
 130 {
 131         int work;
 132
 133         /* net_rx_action's ->poll() invocations and our's are
 134          * synchronized by this test which is only made while
 135          * holding the napi->poll_lock.
 136          */
 137         if (!test_bit(NAPI_STATE_SCHED, &napi->state))
 138                 return budget;
 139
 140         npinfo->rx_flags |= NETPOLL_RX_DROP;
 141         atomic_inc(&trapped);
 142         set_bit(NAPI_STATE_NPSVC, &napi->state);
 143
 144         work = napi->poll(napi, budget);
 145         trace_napi_poll(napi);
 146
 147         clear_bit(NAPI_STATE_NPSVC, &napi->state);
 148         atomic_dec(&trapped);
 149         npinfo->rx_flags &= ~NETPOLL_RX_DROP;
 150
 151         return budget - work;
 152 }
 153
 154 static void poll_napi(struct net_device *dev)
 155 {
 156         struct napi_struct *napi;
 157         int budget = 16;
 158
 159         list_for_each_entry(napi, &dev->napi_list, dev_list) {
 160                 if (napi->poll_owner != smp_processor_id() &&
 161                     spin_trylock(&napi->poll_lock)) {
 162                         budget = poll_one_napi(dev->npinfo, napi, budget);
 163                         spin_unlock(&napi->poll_lock);
 164
 165                         if (!budget)
 166                                 break;
 167                 }
 168         }
 169 }
 170
 171 static void service_arp_queue(struct netpoll_info *npi)
 172 {
 173         if (npi) {
 174                 struct sk_buff *skb;
 175
 176                 while ((skb = skb_dequeue(&npi->arp_tx)))
 177                         arp_reply(skb);
 178         }
 179 }
 180
 181 void netpoll_poll_dev(struct net_device *dev)
 182 {
 183         const struct net_device_ops *ops;
 184
 185         if (!dev || !netif_running(dev))
 186                 return;
 187
 188         ops = dev->netdev_ops;
 189         if (!ops->ndo_poll_controller)
 190                 return;
 191
 192         /* Process pending work on NIC */
 193         ops->ndo_poll_controller(dev);
 194
 195         poll_napi(dev);
 196
 197         service_arp_queue(dev->npinfo);
 198
 199 }
 200 EXPORT_SYMBOL(netpoll_poll_dev);
 201
 202 void netpoll_poll(struct netpoll *np)
 203 {
 204         netpoll_poll_dev(np->dev);
 205 }
 206 EXPORT_SYMBOL(netpoll_poll);
 207
 208 static void refill_skbs(void)
 209 {
 210         struct sk_buff *skb;
 211         unsigned long flags;
 212
 213         spin_lock_irqsave(&skb_pool.lock, flags);
 214         while (skb_pool.qlen < MAX_SKBS) {
 215                 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
 216                 if (!skb)
 217                         break;
 218
 219                 __skb_queue_tail(&skb_pool, skb);
 220         }
 221         spin_unlock_irqrestore(&skb_pool.lock, flags);
 222 }
 223
 224 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
 225 {
 226         int count = 0;
 227         struct sk_buff *skb;
 228
 229         refill_skbs();
 230 repeat:
 231
 232         skb = alloc_skb(len, GFP_ATOMIC);
 233         if (!skb)
 234                 skb = skb_dequeue(&skb_pool);
 235
 236         if (!skb) {
 237                 if (++count < 10) {
 238                         netpoll_poll(np);
 239                         goto repeat;
 240                 }
 241                 return NULL;
 242         }
 243
 244         atomic_set(&skb->users, 1);
 245         skb_reserve(skb, reserve);
 246         return skb;
 247 }
 248
 249 static int netpoll_owner_active(struct net_device *dev)
 250 {
 251         struct napi_struct *napi;
 252
 253         list_for_each_entry(napi, &dev->napi_list, dev_list) {
 254                 if (napi->poll_owner == smp_processor_id())
 255                         return 1;
 256         }
 257         return 0;
 258 }
 259
 260 void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
 261 {
 262         int status = NETDEV_TX_BUSY;
 263         unsigned long tries;
 264         struct net_device *dev = np->dev;
 265         const struct net_device_ops *ops = dev->netdev_ops;
 266         /* It is up to the caller to keep npinfo alive. */
 267         struct netpoll_info *npinfo = np->dev->npinfo;
 268
 269         if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
 270                 __kfree_skb(skb);
 271                 return;
 272         }
 273
 274         /* don't get messages out of order, and no recursion */
 275         if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
 276                 struct netdev_queue *txq;
 277                 unsigned long flags;
 278
 279                 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
 280
 281                 local_irq_save(flags);
 282                 /* try until next clock tick */
 283                 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
 284                      tries > 0; --tries) {
 285                         if (__netif_tx_trylock(txq)) {
 286                                 if (!netif_tx_queue_stopped(txq)) {
 287                                         dev->priv_flags |= IFF_IN_NETPOLL;
 288                                         status = ops->ndo_start_xmit(skb, dev);
 289                                         dev->priv_flags &= ~IFF_IN_NETPOLL;
 290                                         if (status == NETDEV_TX_OK)
 291                                                 txq_trans_update(txq);
 292                                 }
 293                                 __netif_tx_unlock(txq);
 294
 295                                 if (status == NETDEV_TX_OK)
 296                                         break;
 297
 298                         }
 299
 300                         /* tickle device maybe there is some cleanup */
 301                         netpoll_poll(np);
 302
 303                         udelay(USEC_PER_POLL);
 304                 }
 305
 306                 WARN_ONCE(!irqs_disabled(),
 307                         "netpoll_send_skb(): %s enabled interrupts in poll (%pF)\n",
 308                         dev->name, ops->ndo_start_xmit);
 309
 310                 local_irq_restore(flags);
 311         }
 312
 313         if (status != NETDEV_TX_OK) {
 314                 skb_queue_tail(&npinfo->txq, skb);
 315                 schedule_delayed_work(&npinfo->tx_work,0);
 316         }
 317 }
 318 EXPORT_SYMBOL(netpoll_send_skb);
 319
 320 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
 321 {
 322         int total_len, eth_len, ip_len, udp_len;
 323         struct sk_buff *skb;
 324         struct udphdr *udph;
 325         struct iphdr *iph;
 326         struct ethhdr *eth;
 327
 328         udp_len = len + sizeof(*udph);
 329         ip_len = eth_len = udp_len + sizeof(*iph);
 330         total_len = eth_len + ETH_HLEN + NET_IP_ALIGN;
 331
 332         skb = find_skb(np, total_len, total_len - len);
 333         if (!skb)
 334                 return;
 335
 336         skb_copy_to_linear_data(skb, msg, len);
 337         skb->len += len;
 338
 339         skb_push(skb, sizeof(*udph));
 340         skb_reset_transport_header(skb);
 341         udph = udp_hdr(skb);
 342         udph->source = htons(np->local_port);
 343         udph->dest = htons(np->remote_port);
 344         udph->len = htons(udp_len);
 345         udph->check = 0;
 346         udph->check = csum_tcpudp_magic(np->local_ip,
 347                                         np->remote_ip,
 348                                         udp_len, IPPROTO_UDP,
 349                                         csum_partial(udph, udp_len, 0));
 350         if (udph->check == 0)
 351                 udph->check = CSUM_MANGLED_0;
 352
 353         skb_push(skb, sizeof(*iph));
 354         skb_reset_network_header(skb);
 355         iph = ip_hdr(skb);
 356
 357         /* iph->version = 4; iph->ihl = 5; */
 358         put_unaligned(0x45, (unsigned char *)iph);
 359         iph->tos      = 0;
 360         put_unaligned(htons(ip_len), &(iph->tot_len));
 361         iph->id       = 0;
 362         iph->frag_off = 0;
 363         iph->ttl      = 64;
 364         iph->protocol = IPPROTO_UDP;
 365         iph->check    = 0;
 366         put_unaligned(np->local_ip, &(iph->saddr));
 367         put_unaligned(np->remote_ip, &(iph->daddr));
 368         iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
 369
 370         eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
 371         skb_reset_mac_header(skb);
 372         skb->protocol = eth->h_proto = htons(ETH_P_IP);
 373         memcpy(eth->h_source, np->dev->dev_addr, ETH_ALEN);
 374         memcpy(eth->h_dest, np->remote_mac, ETH_ALEN);
 375
 376         skb->dev = np->dev;
 377
 378         netpoll_send_skb(np, skb);
 379 }
 380 EXPORT_SYMBOL(netpoll_send_udp);
 381
 382 static void arp_reply(struct sk_buff *skb)
 383 {
 384         struct netpoll_info *npinfo = skb->dev->npinfo;
 385         struct arphdr *arp;
 386         unsigned char *arp_ptr;
 387         int size, type = ARPOP_REPLY, ptype = ETH_P_ARP;
 388         __be32 sip, tip;
 389         unsigned char *sha;
 390         struct sk_buff *send_skb;
 391         struct netpoll *np, *tmp;
 392         unsigned long flags;
 393         int hits = 0;
 394
 395         if (list_empty(&npinfo->rx_np))
 396                 return;
 397
 398         /* Before checking the packet, we do some early
 399            inspection whether this is interesting at all */
 400         spin_lock_irqsave(&npinfo->rx_lock, flags);
 401         list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
 402                 if (np->dev == skb->dev)
 403                         hits++;
 404         }
 405         spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 406
 407         /* No netpoll struct is using this dev */
 408         if (!hits)
 409                 return;
 410
 411         /* No arp on this interface */
 412         if (skb->dev->flags & IFF_NOARP)
 413                 return;
 414
 415         if (!pskb_may_pull(skb, arp_hdr_len(skb->dev)))
 416                 return;
 417
 418         skb_reset_network_header(skb);
 419         skb_reset_transport_header(skb);
 420         arp = arp_hdr(skb);
 421
 422         if ((arp->ar_hrd != htons(ARPHRD_ETHER) &&
 423              arp->ar_hrd != htons(ARPHRD_IEEE802)) ||
 424             arp->ar_pro != htons(ETH_P_IP) ||
 425             arp->ar_op != htons(ARPOP_REQUEST))
 426                 return;
 427
 428         arp_ptr = (unsigned char *)(arp+1);
 429         /* save the location of the src hw addr */
 430         sha = arp_ptr;
 431         arp_ptr += skb->dev->addr_len;
 432         memcpy(&sip, arp_ptr, 4);
 433         arp_ptr += 4;
 434         /* If we actually cared about dst hw addr,
 435            it would get copied here */
 436         arp_ptr += skb->dev->addr_len;
 437         memcpy(&tip, arp_ptr, 4);
 438
 439         /* Should we ignore arp? */
 440         if (ipv4_is_loopback(tip) || ipv4_is_multicast(tip))
 441                 return;
 442
 443         size = arp_hdr_len(skb->dev);
 444
 445         spin_lock_irqsave(&npinfo->rx_lock, flags);
 446         list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
 447                 if (tip != np->local_ip)
 448                         continue;
 449
 450                 send_skb = find_skb(np, size + LL_ALLOCATED_SPACE(np->dev),
 451                                     LL_RESERVED_SPACE(np->dev));
 452                 if (!send_skb)
 453                         continue;
 454
 455                 skb_reset_network_header(send_skb);
 456                 arp = (struct arphdr *) skb_put(send_skb, size);
 457                 send_skb->dev = skb->dev;
 458                 send_skb->protocol = htons(ETH_P_ARP);
 459
 460                 /* Fill the device header for the ARP frame */
 461                 if (dev_hard_header(send_skb, skb->dev, ptype,
 462                                     sha, np->dev->dev_addr,
 463                                     send_skb->len) < 0) {
 464                         kfree_skb(send_skb);
 465                         continue;
 466                 }
 467
 468                 /*
 469                  * Fill out the arp protocol part.
 470                  *
 471                  * we only support ethernet device type,
 472                  * which (according to RFC 1390) should
 473                  * always equal 1 (Ethernet).
 474                  */
 475
 476                 arp->ar_hrd = htons(np->dev->type);
 477                 arp->ar_pro = htons(ETH_P_IP);
 478                 arp->ar_hln = np->dev->addr_len;
 479                 arp->ar_pln = 4;
 480                 arp->ar_op = htons(type);
 481
 482                 arp_ptr = (unsigned char *)(arp + 1);
 483                 memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
 484                 arp_ptr += np->dev->addr_len;
 485                 memcpy(arp_ptr, &tip, 4);
 486                 arp_ptr += 4;
 487                 memcpy(arp_ptr, sha, np->dev->addr_len);
 488                 arp_ptr += np->dev->addr_len;
 489                 memcpy(arp_ptr, &sip, 4);
 490
 491                 netpoll_send_skb(np, send_skb);
 492
 493                 /* If there are several rx_hooks for the same address,
 494                    we're fine by sending a single reply */
 495                 break;
 496         }
 497         spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 498 }
 499
 500 int __netpoll_rx(struct sk_buff *skb)
 501 {
 502         int proto, len, ulen;
 503         int hits = 0;
 504         struct iphdr *iph;
 505         struct udphdr *uh;
 506         struct netpoll_info *npinfo = skb->dev->npinfo;
 507         struct netpoll *np, *tmp;
 508
 509         if (list_empty(&npinfo->rx_np))
 510                 goto out;
 511
 512         if (skb->dev->type != ARPHRD_ETHER)
 513                 goto out;
 514
 515         /* check if netpoll clients need ARP */
 516         if (skb->protocol == htons(ETH_P_ARP) &&
 517             atomic_read(&trapped)) {
 518                 skb_queue_tail(&npinfo->arp_tx, skb);
 519                 return 1;
 520         }
 521
 522         proto = ntohs(eth_hdr(skb)->h_proto);
 523         if (proto != ETH_P_IP)
 524                 goto out;
 525         if (skb->pkt_type == PACKET_OTHERHOST)
 526                 goto out;
 527         if (skb_shared(skb))
 528                 goto out;
 529
 530         iph = (struct iphdr *)skb->data;
 531         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 532                 goto out;
 533         if (iph->ihl < 5 || iph->version != 4)
 534                 goto out;
 535         if (!pskb_may_pull(skb, iph->ihl*4))
 536                 goto out;
 537         if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
 538                 goto out;
 539
 540         len = ntohs(iph->tot_len);
 541         if (skb->len < len || len < iph->ihl*4)
 542                 goto out;
 543
 544         /*
 545          * Our transport medium may have padded the buffer out.
 546          * Now We trim to the true length of the frame.
 547          */
 548         if (pskb_trim_rcsum(skb, len))
 549                 goto out;
 550
 551         if (iph->protocol != IPPROTO_UDP)
 552                 goto out;
 553
 554         len -= iph->ihl*4;
 555         uh = (struct udphdr *)(((char *)iph) + iph->ihl*4);
 556         ulen = ntohs(uh->len);
 557
 558         if (ulen != len)
 559                 goto out;
 560         if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr))
 561                 goto out;
 562
 563         list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
 564                 if (np->local_ip && np->local_ip != iph->daddr)
 565                         continue;
 566                 if (np->remote_ip && np->remote_ip != iph->saddr)
 567                         continue;
 568                 if (np->local_port && np->local_port != ntohs(uh->dest))
 569                         continue;
 570
 571                 np->rx_hook(np, ntohs(uh->source),
 572                                (char *)(uh+1),
 573                                ulen - sizeof(struct udphdr));
 574                 hits++;
 575         }
 576
 577         if (!hits)
 578                 goto out;
 579
 580         kfree_skb(skb);
 581         return 1;
 582
 583 out:
 584         if (atomic_read(&trapped)) {
 585                 kfree_skb(skb);
 586                 return 1;
 587         }
 588
 589         return 0;
 590 }
 591
 592 void netpoll_print_options(struct netpoll *np)
 593 {
 594         printk(KERN_INFO "%s: local port %d\n",
 595                          np->name, np->local_port);
 596         printk(KERN_INFO "%s: local IP %pI4\n",
 597                          np->name, &np->local_ip);
 598         printk(KERN_INFO "%s: interface '%s'\n",
 599                          np->name, np->dev_name);
 600         printk(KERN_INFO "%s: remote port %d\n",
 601                          np->name, np->remote_port);
 602         printk(KERN_INFO "%s: remote IP %pI4\n",
 603                          np->name, &np->remote_ip);
 604         printk(KERN_INFO "%s: remote ethernet address %pM\n",
 605                          np->name, np->remote_mac);
 606 }
 607 EXPORT_SYMBOL(netpoll_print_options);
 608
 609 int netpoll_parse_options(struct netpoll *np, char *opt)
 610 {
 611         char *cur=opt, *delim;
 612
 613         if (*cur != '@') {
 614                 if ((delim = strchr(cur, '@')) == NULL)
 615                         goto parse_failed;
 616                 *delim = 0;
 617                 np->local_port = simple_strtol(cur, NULL, 10);
 618                 cur = delim;
 619         }
 620         cur++;
 621
 622         if (*cur != '/') {
 623                 if ((delim = strchr(cur, '/')) == NULL)
 624                         goto parse_failed;
 625                 *delim = 0;
 626                 np->local_ip = in_aton(cur);
 627                 cur = delim;
 628         }
 629         cur++;
 630
 631         if (*cur != ',') {
 632                 /* parse out dev name */
 633                 if ((delim = strchr(cur, ',')) == NULL)
 634                         goto parse_failed;
 635                 *delim = 0;
 636                 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
 637                 cur = delim;
 638         }
 639         cur++;
 640
 641         if (*cur != '@') {
 642                 /* dst port */
 643                 if ((delim = strchr(cur, '@')) == NULL)
 644                         goto parse_failed;
 645                 *delim = 0;
 646                 if (*cur == ' ' || *cur == '\t')
 647                         printk(KERN_INFO "%s: warning: whitespace"
 648                                         "is not allowed\n", np->name);
 649                 np->remote_port = simple_strtol(cur, NULL, 10);
 650                 cur = delim;
 651         }
 652         cur++;
 653
 654         /* dst ip */
 655         if ((delim = strchr(cur, '/')) == NULL)
 656                 goto parse_failed;
 657         *delim = 0;
 658         np->remote_ip = in_aton(cur);
 659         cur = delim + 1;
 660
 661         if (*cur != 0) {
 662                 /* MAC address */
 663                 if ((delim = strchr(cur, ':')) == NULL)
 664                         goto parse_failed;
 665                 *delim = 0;
 666                 np->remote_mac[0] = simple_strtol(cur, NULL, 16);
 667                 cur = delim + 1;
 668                 if ((delim = strchr(cur, ':')) == NULL)
 669                         goto parse_failed;
 670                 *delim = 0;
 671                 np->remote_mac[1] = simple_strtol(cur, NULL, 16);
 672                 cur = delim + 1;
 673                 if ((delim = strchr(cur, ':')) == NULL)
 674                         goto parse_failed;
 675                 *delim = 0;
 676                 np->remote_mac[2] = simple_strtol(cur, NULL, 16);
 677                 cur = delim + 1;
 678                 if ((delim = strchr(cur, ':')) == NULL)
 679                         goto parse_failed;
 680                 *delim = 0;
 681                 np->remote_mac[3] = simple_strtol(cur, NULL, 16);
 682                 cur = delim + 1;
 683                 if ((delim = strchr(cur, ':')) == NULL)
 684                         goto parse_failed;
 685                 *delim = 0;
 686                 np->remote_mac[4] = simple_strtol(cur, NULL, 16);
 687                 cur = delim + 1;
 688                 np->remote_mac[5] = simple_strtol(cur, NULL, 16);
 689         }
 690
 691         netpoll_print_options(np);
 692
 693         return 0;
 694
 695  parse_failed:
 696         printk(KERN_INFO "%s: couldn't parse config at '%s'!\n",
 697                np->name, cur);
 698         return -1;
 699 }
 700 EXPORT_SYMBOL(netpoll_parse_options);
 701
 702 int __netpoll_setup(struct netpoll *np)
 703 {
 704         struct net_device *ndev = np->dev;
 705         struct netpoll_info *npinfo;
 706         const struct net_device_ops *ops;
 707         unsigned long flags;
 708         int err;
 709
 710         if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) ||
 711             !ndev->netdev_ops->ndo_poll_controller) {
 712                 printk(KERN_ERR "%s: %s doesn't support polling, aborting.\n",
 713                        np->name, np->dev_name);
 714                 err = -ENOTSUPP;
 715                 goto out;
 716         }
 717
 718         if (!ndev->npinfo) {
 719                 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
 720                 if (!npinfo) {
 721                         err = -ENOMEM;
 722                         goto out;
 723                 }
 724
 725                 npinfo->rx_flags = 0;
 726                 INIT_LIST_HEAD(&npinfo->rx_np);
 727
 728                 spin_lock_init(&npinfo->rx_lock);
 729                 skb_queue_head_init(&npinfo->arp_tx);
 730                 skb_queue_head_init(&npinfo->txq);
 731                 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
 732
 733                 atomic_set(&npinfo->refcnt, 1);
 734
 735                 ops = np->dev->netdev_ops;
 736                 if (ops->ndo_netpoll_setup) {
 737                         err = ops->ndo_netpoll_setup(ndev, npinfo);
 738                         if (err)
 739                                 goto free_npinfo;
 740                 }
 741         } else {
 742                 npinfo = ndev->npinfo;
 743                 atomic_inc(&npinfo->refcnt);
 744         }
 745
 746         npinfo->netpoll = np;
 747
 748         if (np->rx_hook) {
 749                 spin_lock_irqsave(&npinfo->rx_lock, flags);
 750                 npinfo->rx_flags |= NETPOLL_RX_ENABLED;
 751                 list_add_tail(&np->rx, &npinfo->rx_np);
 752                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 753         }
 754
 755         /* last thing to do is link it to the net device structure */
 756         rcu_assign_pointer(ndev->npinfo, npinfo);
 757
 758         return 0;
 759
 760 free_npinfo:
 761         kfree(npinfo);
 762 out:
 763         return err;
 764 }
 765 EXPORT_SYMBOL_GPL(__netpoll_setup);
 766
 767 int netpoll_setup(struct netpoll *np)
 768 {
 769         struct net_device *ndev = NULL;
 770         struct in_device *in_dev;
 771         int err;
 772
 773         if (np->dev_name)
 774                 ndev = dev_get_by_name(&init_net, np->dev_name);
 775         if (!ndev) {
 776                 printk(KERN_ERR "%s: %s doesn't exist, aborting.\n",
 777                        np->name, np->dev_name);
 778                 return -ENODEV;
 779         }
 780
 781         if (!netif_running(ndev)) {
 782                 unsigned long atmost, atleast;
 783
 784                 printk(KERN_INFO "%s: device %s not up yet, forcing it\n",
 785                        np->name, np->dev_name);
 786
 787                 rtnl_lock();
 788                 err = dev_open(ndev);
 789                 rtnl_unlock();
 790
 791                 if (err) {
 792                         printk(KERN_ERR "%s: failed to open %s\n",
 793                                np->name, ndev->name);
 794                         goto put;
 795                 }
 796
 797                 atleast = jiffies + HZ/10;
 798                 atmost = jiffies + carrier_timeout * HZ;
 799                 while (!netif_carrier_ok(ndev)) {
 800                         if (time_after(jiffies, atmost)) {
 801                                 printk(KERN_NOTICE
 802                                        "%s: timeout waiting for carrier\n",
 803                                        np->name);
 804                                 break;
 805                         }
 806                         msleep(1);
 807                 }
 808
 809                 /* If carrier appears to come up instantly, we don't
 810                  * trust it and pause so that we don't pump all our
 811                  * queued console messages into the bitbucket.
 812                  */
 813
 814                 if (time_before(jiffies, atleast)) {
 815                         printk(KERN_NOTICE "%s: carrier detect appears"
 816                                " untrustworthy, waiting 4 seconds\n",
 817                                np->name);
 818                         msleep(4000);
 819                 }
 820         }
 821
 822         if (!np->local_ip) {
 823                 rcu_read_lock();
 824                 in_dev = __in_dev_get_rcu(ndev);
 825
 826                 if (!in_dev || !in_dev->ifa_list) {
 827                         rcu_read_unlock();
 828                         printk(KERN_ERR "%s: no IP address for %s, aborting\n",
 829                                np->name, np->dev_name);
 830                         err = -EDESTADDRREQ;
 831                         goto put;
 832                 }
 833
 834                 np->local_ip = in_dev->ifa_list->ifa_local;
 835                 rcu_read_unlock();
 836                 printk(KERN_INFO "%s: local IP %pI4\n", np->name, &np->local_ip);
 837         }
 838
 839         np->dev = ndev;
 840
 841         /* fill up the skb queue */
 842         refill_skbs();
 843
 844         rtnl_lock();
 845         err = __netpoll_setup(np);
 846         rtnl_unlock();
 847
 848         if (err)
 849                 goto put;
 850
 851         return 0;
 852
 853 put:
 854         dev_put(ndev);
 855         return err;
 856 }
 857 EXPORT_SYMBOL(netpoll_setup);
 858
 859 static int __init netpoll_init(void)
 860 {
 861         skb_queue_head_init(&skb_pool);
 862         return 0;
 863 }
 864 core_initcall(netpoll_init);
 865
 866 void __netpoll_cleanup(struct netpoll *np)
 867 {
 868         struct netpoll_info *npinfo;
 869         unsigned long flags;
 870
 871         npinfo = np->dev->npinfo;
 872         if (!npinfo)
 873                 return;
 874
 875         if (!list_empty(&npinfo->rx_np)) {
 876                 spin_lock_irqsave(&npinfo->rx_lock, flags);
 877                 list_del(&np->rx);
 878                 if (list_empty(&npinfo->rx_np))
 879                         npinfo->rx_flags &= ~NETPOLL_RX_ENABLED;
 880                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 881         }
 882
 883         if (atomic_dec_and_test(&npinfo->refcnt)) {
 884                 const struct net_device_ops *ops;
 885
 886                 ops = np->dev->netdev_ops;
 887                 if (ops->ndo_netpoll_cleanup)
 888                         ops->ndo_netpoll_cleanup(np->dev);
 889
 890                 rcu_assign_pointer(np->dev->npinfo, NULL);
 891
 892                 /* avoid racing with NAPI reading npinfo */
 893                 synchronize_rcu_bh();
 894
 895                 skb_queue_purge(&npinfo->arp_tx);
 896                 skb_queue_purge(&npinfo->txq);
 897                 cancel_rearming_delayed_work(&npinfo->tx_work);
 898
 899                 /* clean after last, unfinished work */
 900                 __skb_queue_purge(&npinfo->txq);
 901                 kfree(npinfo);
 902         }
 903 }
 904 EXPORT_SYMBOL_GPL(__netpoll_cleanup);
 905
 906 void netpoll_cleanup(struct netpoll *np)
 907 {
 908         if (!np->dev)
 909                 return;
 910
 911         rtnl_lock();
 912         __netpoll_cleanup(np);
 913         rtnl_unlock();
 914
 915         dev_put(np->dev);
 916         np->dev = NULL;
 917 }
 918 EXPORT_SYMBOL(netpoll_cleanup);
 919
 920 int netpoll_trap(void)
 921 {
 922         return atomic_read(&trapped);
 923 }
 924 EXPORT_SYMBOL(netpoll_trap);
 925
 926 void netpoll_set_trap(int trap)
 927 {
 928         if (trap)
 929                 atomic_inc(&trapped);
 930         else
 931                 atomic_dec(&trapped);
 932 }
 933 EXPORT_SYMBOL(netpoll_set_trap);