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