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 #define NETPOLL_RX_ENABLED  1
  43 #define NETPOLL_RX_DROP     2
  44
  45 #define MAX_SKB_SIZE \
  46                 (MAX_UDP_CHUNK + sizeof(struct udphdr) + \
  47                                 sizeof(struct iphdr) + sizeof(struct ethhdr))
  48
  49 static void zap_completion_queue(void);
  50 static void arp_reply(struct sk_buff *skb);
  51
  52 static void queue_process(struct work_struct *work)
  53 {
  54         struct netpoll_info *npinfo =
  55                 container_of(work, struct netpoll_info, tx_work.work);
  56         struct sk_buff *skb;
  57         unsigned long flags;
  58
  59         while ((skb = skb_dequeue(&npinfo->txq))) {
  60                 struct net_device *dev = skb->dev;
  61
  62                 if (!netif_device_present(dev) || !netif_running(dev)) {
  63                         __kfree_skb(skb);
  64                         continue;
  65                 }
  66
  67                 local_irq_save(flags);
  68                 netif_tx_lock(dev);
  69                 if ((netif_queue_stopped(dev) ||
  70                      netif_subqueue_stopped(dev, skb)) ||
  71                      dev->hard_start_xmit(skb, dev) != NETDEV_TX_OK) {
  72                         skb_queue_head(&npinfo->txq, skb);
  73                         netif_tx_unlock(dev);
  74                         local_irq_restore(flags);
  75
  76                         schedule_delayed_work(&npinfo->tx_work, HZ/10);
  77                         return;
  78                 }
  79                 netif_tx_unlock(dev);
  80                 local_irq_restore(flags);
  81         }
  82 }
  83
  84 static __sum16 checksum_udp(struct sk_buff *skb, struct udphdr *uh,
  85                             unsigned short ulen, __be32 saddr, __be32 daddr)
  86 {
  87         __wsum psum;
  88
  89         if (uh->check == 0 || skb_csum_unnecessary(skb))
  90                 return 0;
  91
  92         psum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0);
  93
  94         if (skb->ip_summed == CHECKSUM_COMPLETE &&
  95             !csum_fold(csum_add(psum, skb->csum)))
  96                 return 0;
  97
  98         skb->csum = psum;
  99
 100         return __skb_checksum_complete(skb);
 101 }
 102
 103 /*
 104  * Check whether delayed processing was scheduled for our NIC. If so,
 105  * we attempt to grab the poll lock and use ->poll() to pump the card.
 106  * If this fails, either we've recursed in ->poll() or it's already
 107  * running on another CPU.
 108  *
 109  * Note: we don't mask interrupts with this lock because we're using
 110  * trylock here and interrupts are already disabled in the softirq
 111  * case. Further, we test the poll_owner to avoid recursion on UP
 112  * systems where the lock doesn't exist.
 113  *
 114  * In cases where there is bi-directional communications, reading only
 115  * one message at a time can lead to packets being dropped by the
 116  * network adapter, forcing superfluous retries and possibly timeouts.
 117  * Thus, we set our budget to greater than 1.
 118  */
 119 static int poll_one_napi(struct netpoll_info *npinfo,
 120                          struct napi_struct *napi, int budget)
 121 {
 122         int work;
 123
 124         /* net_rx_action's ->poll() invocations and our's are
 125          * synchronized by this test which is only made while
 126          * holding the napi->poll_lock.
 127          */
 128         if (!test_bit(NAPI_STATE_SCHED, &napi->state))
 129                 return budget;
 130
 131         npinfo->rx_flags |= NETPOLL_RX_DROP;
 132         atomic_inc(&trapped);
 133
 134         work = napi->poll(napi, budget);
 135
 136         atomic_dec(&trapped);
 137         npinfo->rx_flags &= ~NETPOLL_RX_DROP;
 138
 139         return budget - work;
 140 }
 141
 142 static void poll_napi(struct net_device *dev)
 143 {
 144         struct napi_struct *napi;
 145         int budget = 16;
 146
 147         list_for_each_entry(napi, &dev->napi_list, dev_list) {
 148                 if (napi->poll_owner != smp_processor_id() &&
 149                     spin_trylock(&napi->poll_lock)) {
 150                         budget = poll_one_napi(dev->npinfo, napi, budget);
 151                         spin_unlock(&napi->poll_lock);
 152
 153                         if (!budget)
 154                                 break;
 155                 }
 156         }
 157 }
 158
 159 static void service_arp_queue(struct netpoll_info *npi)
 160 {
 161         if (npi) {
 162                 struct sk_buff *skb;
 163
 164                 while ((skb = skb_dequeue(&npi->arp_tx)))
 165                         arp_reply(skb);
 166         }
 167 }
 168
 169 void netpoll_poll(struct netpoll *np)
 170 {
 171         struct net_device *dev = np->dev;
 172
 173         if (!dev || !netif_running(dev) || !dev->poll_controller)
 174                 return;
 175
 176         /* Process pending work on NIC */
 177         dev->poll_controller(dev);
 178
 179         poll_napi(dev);
 180
 181         service_arp_queue(dev->npinfo);
 182
 183         zap_completion_queue();
 184 }
 185
 186 static void refill_skbs(void)
 187 {
 188         struct sk_buff *skb;
 189         unsigned long flags;
 190
 191         spin_lock_irqsave(&skb_pool.lock, flags);
 192         while (skb_pool.qlen < MAX_SKBS) {
 193                 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
 194                 if (!skb)
 195                         break;
 196
 197                 __skb_queue_tail(&skb_pool, skb);
 198         }
 199         spin_unlock_irqrestore(&skb_pool.lock, flags);
 200 }
 201
 202 static void zap_completion_queue(void)
 203 {
 204         unsigned long flags;
 205         struct softnet_data *sd = &get_cpu_var(softnet_data);
 206
 207         if (sd->completion_queue) {
 208                 struct sk_buff *clist;
 209
 210                 local_irq_save(flags);
 211                 clist = sd->completion_queue;
 212                 sd->completion_queue = NULL;
 213                 local_irq_restore(flags);
 214
 215                 while (clist != NULL) {
 216                         struct sk_buff *skb = clist;
 217                         clist = clist->next;
 218                         if (skb->destructor) {
 219                                 atomic_inc(&skb->users);
 220                                 dev_kfree_skb_any(skb); /* put this one back */
 221                         } else {
 222                                 __kfree_skb(skb);
 223                         }
 224                 }
 225         }
 226
 227         put_cpu_var(softnet_data);
 228 }
 229
 230 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
 231 {
 232         int count = 0;
 233         struct sk_buff *skb;
 234
 235         zap_completion_queue();
 236         refill_skbs();
 237 repeat:
 238
 239         skb = alloc_skb(len, GFP_ATOMIC);
 240         if (!skb)
 241                 skb = skb_dequeue(&skb_pool);
 242
 243         if (!skb) {
 244                 if (++count < 10) {
 245                         netpoll_poll(np);
 246                         goto repeat;
 247                 }
 248                 return NULL;
 249         }
 250
 251         atomic_set(&skb->users, 1);
 252         skb_reserve(skb, reserve);
 253         return skb;
 254 }
 255
 256 static int netpoll_owner_active(struct net_device *dev)
 257 {
 258         struct napi_struct *napi;
 259
 260         list_for_each_entry(napi, &dev->napi_list, dev_list) {
 261                 if (napi->poll_owner == smp_processor_id())
 262                         return 1;
 263         }
 264         return 0;
 265 }
 266
 267 static void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
 268 {
 269         int status = NETDEV_TX_BUSY;
 270         unsigned long tries;
 271         struct net_device *dev = np->dev;
 272         struct netpoll_info *npinfo = np->dev->npinfo;
 273
 274         if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
 275                 __kfree_skb(skb);
 276                 return;
 277         }
 278
 279         /* don't get messages out of order, and no recursion */
 280         if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
 281                 unsigned long flags;
 282
 283                 local_irq_save(flags);
 284                 /* try until next clock tick */
 285                 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
 286                      tries > 0; --tries) {
 287                         if (netif_tx_trylock(dev)) {
 288                                 if (!netif_queue_stopped(dev) &&
 289                                     !netif_subqueue_stopped(dev, skb))
 290                                         status = dev->hard_start_xmit(skb, dev);
 291                                 netif_tx_unlock(dev);
 292
 293                                 if (status == NETDEV_TX_OK)
 294                                         break;
 295
 296                         }
 297
 298                         /* tickle device maybe there is some cleanup */
 299                         netpoll_poll(np);
 300
 301                         udelay(USEC_PER_POLL);
 302                 }
 303                 local_irq_restore(flags);
 304         }
 305
 306         if (status != NETDEV_TX_OK) {
 307                 skb_queue_tail(&npinfo->txq, skb);
 308                 schedule_delayed_work(&npinfo->tx_work,0);
 309         }
 310 }
 311
 312 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
 313 {
 314         int total_len, eth_len, ip_len, udp_len;
 315         struct sk_buff *skb;
 316         struct udphdr *udph;
 317         struct iphdr *iph;
 318         struct ethhdr *eth;
 319
 320         udp_len = len + sizeof(*udph);
 321         ip_len = eth_len = udp_len + sizeof(*iph);
 322         total_len = eth_len + ETH_HLEN + NET_IP_ALIGN;
 323
 324         skb = find_skb(np, total_len, total_len - len);
 325         if (!skb)
 326                 return;
 327
 328         skb_copy_to_linear_data(skb, msg, len);
 329         skb->len += len;
 330
 331         skb_push(skb, sizeof(*udph));
 332         skb_reset_transport_header(skb);
 333         udph = udp_hdr(skb);
 334         udph->source = htons(np->local_port);
 335         udph->dest = htons(np->remote_port);
 336         udph->len = htons(udp_len);
 337         udph->check = 0;
 338         udph->check = csum_tcpudp_magic(htonl(np->local_ip),
 339                                         htonl(np->remote_ip),
 340                                         udp_len, IPPROTO_UDP,
 341                                         csum_partial((unsigned char *)udph, udp_len, 0));
 342         if (udph->check == 0)
 343                 udph->check = CSUM_MANGLED_0;
 344
 345         skb_push(skb, sizeof(*iph));
 346         skb_reset_network_header(skb);
 347         iph = ip_hdr(skb);
 348
 349         /* iph->version = 4; iph->ihl = 5; */
 350         put_unaligned(0x45, (unsigned char *)iph);
 351         iph->tos      = 0;
 352         put_unaligned(htons(ip_len), &(iph->tot_len));
 353         iph->id       = 0;
 354         iph->frag_off = 0;
 355         iph->ttl      = 64;
 356         iph->protocol = IPPROTO_UDP;
 357         iph->check    = 0;
 358         put_unaligned(htonl(np->local_ip), &(iph->saddr));
 359         put_unaligned(htonl(np->remote_ip), &(iph->daddr));
 360         iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
 361
 362         eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
 363         skb_reset_mac_header(skb);
 364         skb->protocol = eth->h_proto = htons(ETH_P_IP);
 365         memcpy(eth->h_source, np->dev->dev_addr, ETH_ALEN);
 366         memcpy(eth->h_dest, np->remote_mac, ETH_ALEN);
 367
 368         skb->dev = np->dev;
 369
 370         netpoll_send_skb(np, skb);
 371 }
 372
 373 static void arp_reply(struct sk_buff *skb)
 374 {
 375         struct netpoll_info *npinfo = skb->dev->npinfo;
 376         struct arphdr *arp;
 377         unsigned char *arp_ptr;
 378         int size, type = ARPOP_REPLY, ptype = ETH_P_ARP;
 379         __be32 sip, tip;
 380         unsigned char *sha;
 381         struct sk_buff *send_skb;
 382         struct netpoll *np = NULL;
 383
 384         if (npinfo->rx_np && npinfo->rx_np->dev == skb->dev)
 385                 np = npinfo->rx_np;
 386         if (!np)
 387                 return;
 388
 389         /* No arp on this interface */
 390         if (skb->dev->flags & IFF_NOARP)
 391                 return;
 392
 393         if (!pskb_may_pull(skb, (sizeof(struct arphdr) +
 394                                  (2 * skb->dev->addr_len) +
 395                                  (2 * sizeof(u32)))))
 396                 return;
 397
 398         skb_reset_network_header(skb);
 399         skb_reset_transport_header(skb);
 400         arp = arp_hdr(skb);
 401
 402         if ((arp->ar_hrd != htons(ARPHRD_ETHER) &&
 403              arp->ar_hrd != htons(ARPHRD_IEEE802)) ||
 404             arp->ar_pro != htons(ETH_P_IP) ||
 405             arp->ar_op != htons(ARPOP_REQUEST))
 406                 return;
 407
 408         arp_ptr = (unsigned char *)(arp+1);
 409         /* save the location of the src hw addr */
 410         sha = arp_ptr;
 411         arp_ptr += skb->dev->addr_len;
 412         memcpy(&sip, arp_ptr, 4);
 413         arp_ptr += 4;
 414         /* if we actually cared about dst hw addr, it would get copied here */
 415         arp_ptr += skb->dev->addr_len;
 416         memcpy(&tip, arp_ptr, 4);
 417
 418         /* Should we ignore arp? */
 419         if (tip != htonl(np->local_ip) ||
 420             ipv4_is_loopback(tip) || ipv4_is_multicast(tip))
 421                 return;
 422
 423         size = sizeof(struct arphdr) + 2 * (skb->dev->addr_len + 4);
 424         send_skb = find_skb(np, size + LL_RESERVED_SPACE(np->dev),
 425                             LL_RESERVED_SPACE(np->dev));
 426
 427         if (!send_skb)
 428                 return;
 429
 430         skb_reset_network_header(send_skb);
 431         arp = (struct arphdr *) skb_put(send_skb, size);
 432         send_skb->dev = skb->dev;
 433         send_skb->protocol = htons(ETH_P_ARP);
 434
 435         /* Fill the device header for the ARP frame */
 436         if (dev_hard_header(send_skb, skb->dev, ptype,
 437                             sha, np->dev->dev_addr,
 438                             send_skb->len) < 0) {
 439                 kfree_skb(send_skb);
 440                 return;
 441         }
 442
 443         /*
 444          * Fill out the arp protocol part.
 445          *
 446          * we only support ethernet device type,
 447          * which (according to RFC 1390) should always equal 1 (Ethernet).
 448          */
 449
 450         arp->ar_hrd = htons(np->dev->type);
 451         arp->ar_pro = htons(ETH_P_IP);
 452         arp->ar_hln = np->dev->addr_len;
 453         arp->ar_pln = 4;
 454         arp->ar_op = htons(type);
 455
 456         arp_ptr=(unsigned char *)(arp + 1);
 457         memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
 458         arp_ptr += np->dev->addr_len;
 459         memcpy(arp_ptr, &tip, 4);
 460         arp_ptr += 4;
 461         memcpy(arp_ptr, sha, np->dev->addr_len);
 462         arp_ptr += np->dev->addr_len;
 463         memcpy(arp_ptr, &sip, 4);
 464
 465         netpoll_send_skb(np, send_skb);
 466 }
 467
 468 int __netpoll_rx(struct sk_buff *skb)
 469 {
 470         int proto, len, ulen;
 471         struct iphdr *iph;
 472         struct udphdr *uh;
 473         struct netpoll_info *npi = skb->dev->npinfo;
 474         struct netpoll *np = npi->rx_np;
 475
 476         if (!np)
 477                 goto out;
 478         if (skb->dev->type != ARPHRD_ETHER)
 479                 goto out;
 480
 481         /* check if netpoll clients need ARP */
 482         if (skb->protocol == htons(ETH_P_ARP) &&
 483             atomic_read(&trapped)) {
 484                 skb_queue_tail(&npi->arp_tx, skb);
 485                 return 1;
 486         }
 487
 488         proto = ntohs(eth_hdr(skb)->h_proto);
 489         if (proto != ETH_P_IP)
 490                 goto out;
 491         if (skb->pkt_type == PACKET_OTHERHOST)
 492                 goto out;
 493         if (skb_shared(skb))
 494                 goto out;
 495
 496         iph = (struct iphdr *)skb->data;
 497         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 498                 goto out;
 499         if (iph->ihl < 5 || iph->version != 4)
 500                 goto out;
 501         if (!pskb_may_pull(skb, iph->ihl*4))
 502                 goto out;
 503         if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
 504                 goto out;
 505
 506         len = ntohs(iph->tot_len);
 507         if (skb->len < len || len < iph->ihl*4)
 508                 goto out;
 509
 510         /*
 511          * Our transport medium may have padded the buffer out.
 512          * Now We trim to the true length of the frame.
 513          */
 514         if (pskb_trim_rcsum(skb, len))
 515                 goto out;
 516
 517         if (iph->protocol != IPPROTO_UDP)
 518                 goto out;
 519
 520         len -= iph->ihl*4;
 521         uh = (struct udphdr *)(((char *)iph) + iph->ihl*4);
 522         ulen = ntohs(uh->len);
 523
 524         if (ulen != len)
 525                 goto out;
 526         if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr))
 527                 goto out;
 528         if (np->local_ip && np->local_ip != ntohl(iph->daddr))
 529                 goto out;
 530         if (np->remote_ip && np->remote_ip != ntohl(iph->saddr))
 531                 goto out;
 532         if (np->local_port && np->local_port != ntohs(uh->dest))
 533                 goto out;
 534
 535         np->rx_hook(np, ntohs(uh->source),
 536                     (char *)(uh+1),
 537                     ulen - sizeof(struct udphdr));
 538
 539         kfree_skb(skb);
 540         return 1;
 541
 542 out:
 543         if (atomic_read(&trapped)) {
 544                 kfree_skb(skb);
 545                 return 1;
 546         }
 547
 548         return 0;
 549 }
 550
 551 void netpoll_print_options(struct netpoll *np)
 552 {
 553         DECLARE_MAC_BUF(mac);
 554         printk(KERN_INFO "%s: local port %d\n",
 555                          np->name, np->local_port);
 556         printk(KERN_INFO "%s: local IP %d.%d.%d.%d\n",
 557                          np->name, HIPQUAD(np->local_ip));
 558         printk(KERN_INFO "%s: interface %s\n",
 559                          np->name, np->dev_name);
 560         printk(KERN_INFO "%s: remote port %d\n",
 561                          np->name, np->remote_port);
 562         printk(KERN_INFO "%s: remote IP %d.%d.%d.%d\n",
 563                          np->name, HIPQUAD(np->remote_ip));
 564         printk(KERN_INFO "%s: remote ethernet address %s\n",
 565                          np->name, print_mac(mac, np->remote_mac));
 566 }
 567
 568 int netpoll_parse_options(struct netpoll *np, char *opt)
 569 {
 570         char *cur=opt, *delim;
 571
 572         if (*cur != '@') {
 573                 if ((delim = strchr(cur, '@')) == NULL)
 574                         goto parse_failed;
 575                 *delim = 0;
 576                 np->local_port = simple_strtol(cur, NULL, 10);
 577                 cur = delim;
 578         }
 579         cur++;
 580
 581         if (*cur != '/') {
 582                 if ((delim = strchr(cur, '/')) == NULL)
 583                         goto parse_failed;
 584                 *delim = 0;
 585                 np->local_ip = ntohl(in_aton(cur));
 586                 cur = delim;
 587         }
 588         cur++;
 589
 590         if (*cur != ',') {
 591                 /* parse out dev name */
 592                 if ((delim = strchr(cur, ',')) == NULL)
 593                         goto parse_failed;
 594                 *delim = 0;
 595                 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
 596                 cur = delim;
 597         }
 598         cur++;
 599
 600         if (*cur != '@') {
 601                 /* dst port */
 602                 if ((delim = strchr(cur, '@')) == NULL)
 603                         goto parse_failed;
 604                 *delim = 0;
 605                 np->remote_port = simple_strtol(cur, NULL, 10);
 606                 cur = delim;
 607         }
 608         cur++;
 609
 610         /* dst ip */
 611         if ((delim = strchr(cur, '/')) == NULL)
 612                 goto parse_failed;
 613         *delim = 0;
 614         np->remote_ip = ntohl(in_aton(cur));
 615         cur = delim + 1;
 616
 617         if (*cur != 0) {
 618                 /* MAC address */
 619                 if ((delim = strchr(cur, ':')) == NULL)
 620                         goto parse_failed;
 621                 *delim = 0;
 622                 np->remote_mac[0] = 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[1] = 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[2] = 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[3] = simple_strtol(cur, NULL, 16);
 638                 cur = delim + 1;
 639                 if ((delim = strchr(cur, ':')) == NULL)
 640                         goto parse_failed;
 641                 *delim = 0;
 642                 np->remote_mac[4] = simple_strtol(cur, NULL, 16);
 643                 cur = delim + 1;
 644                 np->remote_mac[5] = simple_strtol(cur, NULL, 16);
 645         }
 646
 647         netpoll_print_options(np);
 648
 649         return 0;
 650
 651  parse_failed:
 652         printk(KERN_INFO "%s: couldn't parse config at %s!\n",
 653                np->name, cur);
 654         return -1;
 655 }
 656
 657 int netpoll_setup(struct netpoll *np)
 658 {
 659         struct net_device *ndev = NULL;
 660         struct in_device *in_dev;
 661         struct netpoll_info *npinfo;
 662         unsigned long flags;
 663         int err;
 664
 665         if (np->dev_name)
 666                 ndev = dev_get_by_name(&init_net, np->dev_name);
 667         if (!ndev) {
 668                 printk(KERN_ERR "%s: %s doesn't exist, aborting.\n",
 669                        np->name, np->dev_name);
 670                 return -ENODEV;
 671         }
 672
 673         np->dev = ndev;
 674         if (!ndev->npinfo) {
 675                 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
 676                 if (!npinfo) {
 677                         err = -ENOMEM;
 678                         goto release;
 679                 }
 680
 681                 npinfo->rx_flags = 0;
 682                 npinfo->rx_np = NULL;
 683
 684                 spin_lock_init(&npinfo->rx_lock);
 685                 skb_queue_head_init(&npinfo->arp_tx);
 686                 skb_queue_head_init(&npinfo->txq);
 687                 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
 688
 689                 atomic_set(&npinfo->refcnt, 1);
 690         } else {
 691                 npinfo = ndev->npinfo;
 692                 atomic_inc(&npinfo->refcnt);
 693         }
 694
 695         if (!ndev->poll_controller) {
 696                 printk(KERN_ERR "%s: %s doesn't support polling, aborting.\n",
 697                        np->name, np->dev_name);
 698                 err = -ENOTSUPP;
 699                 goto release;
 700         }
 701
 702         if (!netif_running(ndev)) {
 703                 unsigned long atmost, atleast;
 704
 705                 printk(KERN_INFO "%s: device %s not up yet, forcing it\n",
 706                        np->name, np->dev_name);
 707
 708                 rtnl_lock();
 709                 err = dev_open(ndev);
 710                 rtnl_unlock();
 711
 712                 if (err) {
 713                         printk(KERN_ERR "%s: failed to open %s\n",
 714                                np->name, ndev->name);
 715                         goto release;
 716                 }
 717
 718                 atleast = jiffies + HZ/10;
 719                 atmost = jiffies + 4*HZ;
 720                 while (!netif_carrier_ok(ndev)) {
 721                         if (time_after(jiffies, atmost)) {
 722                                 printk(KERN_NOTICE
 723                                        "%s: timeout waiting for carrier\n",
 724                                        np->name);
 725                                 break;
 726                         }
 727                         cond_resched();
 728                 }
 729
 730                 /* If carrier appears to come up instantly, we don't
 731                  * trust it and pause so that we don't pump all our
 732                  * queued console messages into the bitbucket.
 733                  */
 734
 735                 if (time_before(jiffies, atleast)) {
 736                         printk(KERN_NOTICE "%s: carrier detect appears"
 737                                " untrustworthy, waiting 4 seconds\n",
 738                                np->name);
 739                         msleep(4000);
 740                 }
 741         }
 742
 743         if (!np->local_ip) {
 744                 rcu_read_lock();
 745                 in_dev = __in_dev_get_rcu(ndev);
 746
 747                 if (!in_dev || !in_dev->ifa_list) {
 748                         rcu_read_unlock();
 749                         printk(KERN_ERR "%s: no IP address for %s, aborting\n",
 750                                np->name, np->dev_name);
 751                         err = -EDESTADDRREQ;
 752                         goto release;
 753                 }
 754
 755                 np->local_ip = ntohl(in_dev->ifa_list->ifa_local);
 756                 rcu_read_unlock();
 757                 printk(KERN_INFO "%s: local IP %d.%d.%d.%d\n",
 758                        np->name, HIPQUAD(np->local_ip));
 759         }
 760
 761         if (np->rx_hook) {
 762                 spin_lock_irqsave(&npinfo->rx_lock, flags);
 763                 npinfo->rx_flags |= NETPOLL_RX_ENABLED;
 764                 npinfo->rx_np = np;
 765                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 766         }
 767
 768         /* fill up the skb queue */
 769         refill_skbs();
 770
 771         /* last thing to do is link it to the net device structure */
 772         ndev->npinfo = npinfo;
 773
 774         /* avoid racing with NAPI reading npinfo */
 775         synchronize_rcu();
 776
 777         return 0;
 778
 779  release:
 780         if (!ndev->npinfo)
 781                 kfree(npinfo);
 782         np->dev = NULL;
 783         dev_put(ndev);
 784         return err;
 785 }
 786
 787 static int __init netpoll_init(void)
 788 {
 789         skb_queue_head_init(&skb_pool);
 790         return 0;
 791 }
 792 core_initcall(netpoll_init);
 793
 794 void netpoll_cleanup(struct netpoll *np)
 795 {
 796         struct netpoll_info *npinfo;
 797         unsigned long flags;
 798
 799         if (np->dev) {
 800                 npinfo = np->dev->npinfo;
 801                 if (npinfo) {
 802                         if (npinfo->rx_np == np) {
 803                                 spin_lock_irqsave(&npinfo->rx_lock, flags);
 804                                 npinfo->rx_np = NULL;
 805                                 npinfo->rx_flags &= ~NETPOLL_RX_ENABLED;
 806                                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 807                         }
 808
 809                         if (atomic_dec_and_test(&npinfo->refcnt)) {
 810                                 skb_queue_purge(&npinfo->arp_tx);
 811                                 skb_queue_purge(&npinfo->txq);
 812                                 cancel_rearming_delayed_work(&npinfo->tx_work);
 813
 814                                 /* clean after last, unfinished work */
 815                                 __skb_queue_purge(&npinfo->txq);
 816                                 kfree(npinfo);
 817                                 np->dev->npinfo = NULL;
 818                         }
 819                 }
 820
 821                 dev_put(np->dev);
 822         }
 823
 824         np->dev = NULL;
 825 }
 826
 827 int netpoll_trap(void)
 828 {
 829         return atomic_read(&trapped);
 830 }
 831
 832 void netpoll_set_trap(int trap)
 833 {
 834         if (trap)
 835                 atomic_inc(&trapped);
 836         else
 837                 atomic_dec(&trapped);
 838 }
 839
 840 EXPORT_SYMBOL(netpoll_set_trap);
 841 EXPORT_SYMBOL(netpoll_trap);
 842 EXPORT_SYMBOL(netpoll_print_options);
 843 EXPORT_SYMBOL(netpoll_parse_options);
 844 EXPORT_SYMBOL(netpoll_setup);
 845 EXPORT_SYMBOL(netpoll_cleanup);
 846 EXPORT_SYMBOL(netpoll_send_udp);
 847 EXPORT_SYMBOL(netpoll_poll);