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                                 dev_kfree_skb_any(skb); /* put this one back */
 220                         else
 221                                 __kfree_skb(skb);
 222                 }
 223         }
 224
 225         put_cpu_var(softnet_data);
 226 }
 227
 228 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
 229 {
 230         int count = 0;
 231         struct sk_buff *skb;
 232
 233         zap_completion_queue();
 234         refill_skbs();
 235 repeat:
 236
 237         skb = alloc_skb(len, GFP_ATOMIC);
 238         if (!skb)
 239                 skb = skb_dequeue(&skb_pool);
 240
 241         if (!skb) {
 242                 if (++count < 10) {
 243                         netpoll_poll(np);
 244                         goto repeat;
 245                 }
 246                 return NULL;
 247         }
 248
 249         atomic_set(&skb->users, 1);
 250         skb_reserve(skb, reserve);
 251         return skb;
 252 }
 253
 254 static int netpoll_owner_active(struct net_device *dev)
 255 {
 256         struct napi_struct *napi;
 257
 258         list_for_each_entry(napi, &dev->napi_list, dev_list) {
 259                 if (napi->poll_owner == smp_processor_id())
 260                         return 1;
 261         }
 262         return 0;
 263 }
 264
 265 static void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
 266 {
 267         int status = NETDEV_TX_BUSY;
 268         unsigned long tries;
 269         struct net_device *dev = np->dev;
 270         struct netpoll_info *npinfo = np->dev->npinfo;
 271
 272         if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
 273                 __kfree_skb(skb);
 274                 return;
 275         }
 276
 277         /* don't get messages out of order, and no recursion */
 278         if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
 279                 unsigned long flags;
 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(dev)) {
 286                                 if (!netif_queue_stopped(dev) &&
 287                                     !netif_subqueue_stopped(dev, skb))
 288                                         status = dev->hard_start_xmit(skb, dev);
 289                                 netif_tx_unlock(dev);
 290
 291                                 if (status == NETDEV_TX_OK)
 292                                         break;
 293
 294                         }
 295
 296                         /* tickle device maybe there is some cleanup */
 297                         netpoll_poll(np);
 298
 299                         udelay(USEC_PER_POLL);
 300                 }
 301                 local_irq_restore(flags);
 302         }
 303
 304         if (status != NETDEV_TX_OK) {
 305                 skb_queue_tail(&npinfo->txq, skb);
 306                 schedule_delayed_work(&npinfo->tx_work,0);
 307         }
 308 }
 309
 310 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
 311 {
 312         int total_len, eth_len, ip_len, udp_len;
 313         struct sk_buff *skb;
 314         struct udphdr *udph;
 315         struct iphdr *iph;
 316         struct ethhdr *eth;
 317
 318         udp_len = len + sizeof(*udph);
 319         ip_len = eth_len = udp_len + sizeof(*iph);
 320         total_len = eth_len + ETH_HLEN + NET_IP_ALIGN;
 321
 322         skb = find_skb(np, total_len, total_len - len);
 323         if (!skb)
 324                 return;
 325
 326         skb_copy_to_linear_data(skb, msg, len);
 327         skb->len += len;
 328
 329         skb_push(skb, sizeof(*udph));
 330         skb_reset_transport_header(skb);
 331         udph = udp_hdr(skb);
 332         udph->source = htons(np->local_port);
 333         udph->dest = htons(np->remote_port);
 334         udph->len = htons(udp_len);
 335         udph->check = 0;
 336         udph->check = csum_tcpudp_magic(htonl(np->local_ip),
 337                                         htonl(np->remote_ip),
 338                                         udp_len, IPPROTO_UDP,
 339                                         csum_partial((unsigned char *)udph, udp_len, 0));
 340         if (udph->check == 0)
 341                 udph->check = CSUM_MANGLED_0;
 342
 343         skb_push(skb, sizeof(*iph));
 344         skb_reset_network_header(skb);
 345         iph = ip_hdr(skb);
 346
 347         /* iph->version = 4; iph->ihl = 5; */
 348         put_unaligned(0x45, (unsigned char *)iph);
 349         iph->tos      = 0;
 350         put_unaligned(htons(ip_len), &(iph->tot_len));
 351         iph->id       = 0;
 352         iph->frag_off = 0;
 353         iph->ttl      = 64;
 354         iph->protocol = IPPROTO_UDP;
 355         iph->check    = 0;
 356         put_unaligned(htonl(np->local_ip), &(iph->saddr));
 357         put_unaligned(htonl(np->remote_ip), &(iph->daddr));
 358         iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
 359
 360         eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
 361         skb_reset_mac_header(skb);
 362         skb->protocol = eth->h_proto = htons(ETH_P_IP);
 363         memcpy(eth->h_source, np->dev->dev_addr, ETH_ALEN);
 364         memcpy(eth->h_dest, np->remote_mac, ETH_ALEN);
 365
 366         skb->dev = np->dev;
 367
 368         netpoll_send_skb(np, skb);
 369 }
 370
 371 static void arp_reply(struct sk_buff *skb)
 372 {
 373         struct netpoll_info *npinfo = skb->dev->npinfo;
 374         struct arphdr *arp;
 375         unsigned char *arp_ptr;
 376         int size, type = ARPOP_REPLY, ptype = ETH_P_ARP;
 377         __be32 sip, tip;
 378         unsigned char *sha;
 379         struct sk_buff *send_skb;
 380         struct netpoll *np = NULL;
 381
 382         if (npinfo->rx_np && npinfo->rx_np->dev == skb->dev)
 383                 np = npinfo->rx_np;
 384         if (!np)
 385                 return;
 386
 387         /* No arp on this interface */
 388         if (skb->dev->flags & IFF_NOARP)
 389                 return;
 390
 391         if (!pskb_may_pull(skb, (sizeof(struct arphdr) +
 392                                  (2 * skb->dev->addr_len) +
 393                                  (2 * sizeof(u32)))))
 394                 return;
 395
 396         skb_reset_network_header(skb);
 397         skb_reset_transport_header(skb);
 398         arp = arp_hdr(skb);
 399
 400         if ((arp->ar_hrd != htons(ARPHRD_ETHER) &&
 401              arp->ar_hrd != htons(ARPHRD_IEEE802)) ||
 402             arp->ar_pro != htons(ETH_P_IP) ||
 403             arp->ar_op != htons(ARPOP_REQUEST))
 404                 return;
 405
 406         arp_ptr = (unsigned char *)(arp+1);
 407         /* save the location of the src hw addr */
 408         sha = arp_ptr;
 409         arp_ptr += skb->dev->addr_len;
 410         memcpy(&sip, arp_ptr, 4);
 411         arp_ptr += 4;
 412         /* if we actually cared about dst hw addr, it would get copied here */
 413         arp_ptr += skb->dev->addr_len;
 414         memcpy(&tip, arp_ptr, 4);
 415
 416         /* Should we ignore arp? */
 417         if (tip != htonl(np->local_ip) ||
 418             ipv4_is_loopback(tip) || ipv4_is_multicast(tip))
 419                 return;
 420
 421         size = sizeof(struct arphdr) + 2 * (skb->dev->addr_len + 4);
 422         send_skb = find_skb(np, size + LL_RESERVED_SPACE(np->dev),
 423                             LL_RESERVED_SPACE(np->dev));
 424
 425         if (!send_skb)
 426                 return;
 427
 428         skb_reset_network_header(send_skb);
 429         arp = (struct arphdr *) skb_put(send_skb, size);
 430         send_skb->dev = skb->dev;
 431         send_skb->protocol = htons(ETH_P_ARP);
 432
 433         /* Fill the device header for the ARP frame */
 434         if (dev_hard_header(send_skb, skb->dev, ptype,
 435                             sha, np->dev->dev_addr,
 436                             send_skb->len) < 0) {
 437                 kfree_skb(send_skb);
 438                 return;
 439         }
 440
 441         /*
 442          * Fill out the arp protocol part.
 443          *
 444          * we only support ethernet device type,
 445          * which (according to RFC 1390) should always equal 1 (Ethernet).
 446          */
 447
 448         arp->ar_hrd = htons(np->dev->type);
 449         arp->ar_pro = htons(ETH_P_IP);
 450         arp->ar_hln = np->dev->addr_len;
 451         arp->ar_pln = 4;
 452         arp->ar_op = htons(type);
 453
 454         arp_ptr=(unsigned char *)(arp + 1);
 455         memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
 456         arp_ptr += np->dev->addr_len;
 457         memcpy(arp_ptr, &tip, 4);
 458         arp_ptr += 4;
 459         memcpy(arp_ptr, sha, np->dev->addr_len);
 460         arp_ptr += np->dev->addr_len;
 461         memcpy(arp_ptr, &sip, 4);
 462
 463         netpoll_send_skb(np, send_skb);
 464 }
 465
 466 int __netpoll_rx(struct sk_buff *skb)
 467 {
 468         int proto, len, ulen;
 469         struct iphdr *iph;
 470         struct udphdr *uh;
 471         struct netpoll_info *npi = skb->dev->npinfo;
 472         struct netpoll *np = npi->rx_np;
 473
 474         if (!np)
 475                 goto out;
 476         if (skb->dev->type != ARPHRD_ETHER)
 477                 goto out;
 478
 479         /* check if netpoll clients need ARP */
 480         if (skb->protocol == htons(ETH_P_ARP) &&
 481             atomic_read(&trapped)) {
 482                 skb_queue_tail(&npi->arp_tx, skb);
 483                 return 1;
 484         }
 485
 486         proto = ntohs(eth_hdr(skb)->h_proto);
 487         if (proto != ETH_P_IP)
 488                 goto out;
 489         if (skb->pkt_type == PACKET_OTHERHOST)
 490                 goto out;
 491         if (skb_shared(skb))
 492                 goto out;
 493
 494         iph = (struct iphdr *)skb->data;
 495         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 496                 goto out;
 497         if (iph->ihl < 5 || iph->version != 4)
 498                 goto out;
 499         if (!pskb_may_pull(skb, iph->ihl*4))
 500                 goto out;
 501         if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
 502                 goto out;
 503
 504         len = ntohs(iph->tot_len);
 505         if (skb->len < len || len < iph->ihl*4)
 506                 goto out;
 507
 508         /*
 509          * Our transport medium may have padded the buffer out.
 510          * Now We trim to the true length of the frame.
 511          */
 512         if (pskb_trim_rcsum(skb, len))
 513                 goto out;
 514
 515         if (iph->protocol != IPPROTO_UDP)
 516                 goto out;
 517
 518         len -= iph->ihl*4;
 519         uh = (struct udphdr *)(((char *)iph) + iph->ihl*4);
 520         ulen = ntohs(uh->len);
 521
 522         if (ulen != len)
 523                 goto out;
 524         if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr))
 525                 goto out;
 526         if (np->local_ip && np->local_ip != ntohl(iph->daddr))
 527                 goto out;
 528         if (np->remote_ip && np->remote_ip != ntohl(iph->saddr))
 529                 goto out;
 530         if (np->local_port && np->local_port != ntohs(uh->dest))
 531                 goto out;
 532
 533         np->rx_hook(np, ntohs(uh->source),
 534                     (char *)(uh+1),
 535                     ulen - sizeof(struct udphdr));
 536
 537         kfree_skb(skb);
 538         return 1;
 539
 540 out:
 541         if (atomic_read(&trapped)) {
 542                 kfree_skb(skb);
 543                 return 1;
 544         }
 545
 546         return 0;
 547 }
 548
 549 void netpoll_print_options(struct netpoll *np)
 550 {
 551         DECLARE_MAC_BUF(mac);
 552         printk(KERN_INFO "%s: local port %d\n",
 553                          np->name, np->local_port);
 554         printk(KERN_INFO "%s: local IP %d.%d.%d.%d\n",
 555                          np->name, HIPQUAD(np->local_ip));
 556         printk(KERN_INFO "%s: interface %s\n",
 557                          np->name, np->dev_name);
 558         printk(KERN_INFO "%s: remote port %d\n",
 559                          np->name, np->remote_port);
 560         printk(KERN_INFO "%s: remote IP %d.%d.%d.%d\n",
 561                          np->name, HIPQUAD(np->remote_ip));
 562         printk(KERN_INFO "%s: remote ethernet address %s\n",
 563                          np->name, print_mac(mac, np->remote_mac));
 564 }
 565
 566 int netpoll_parse_options(struct netpoll *np, char *opt)
 567 {
 568         char *cur=opt, *delim;
 569
 570         if (*cur != '@') {
 571                 if ((delim = strchr(cur, '@')) == NULL)
 572                         goto parse_failed;
 573                 *delim = 0;
 574                 np->local_port = simple_strtol(cur, NULL, 10);
 575                 cur = delim;
 576         }
 577         cur++;
 578
 579         if (*cur != '/') {
 580                 if ((delim = strchr(cur, '/')) == NULL)
 581                         goto parse_failed;
 582                 *delim = 0;
 583                 np->local_ip = ntohl(in_aton(cur));
 584                 cur = delim;
 585         }
 586         cur++;
 587
 588         if (*cur != ',') {
 589                 /* parse out dev name */
 590                 if ((delim = strchr(cur, ',')) == NULL)
 591                         goto parse_failed;
 592                 *delim = 0;
 593                 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
 594                 cur = delim;
 595         }
 596         cur++;
 597
 598         if (*cur != '@') {
 599                 /* dst port */
 600                 if ((delim = strchr(cur, '@')) == NULL)
 601                         goto parse_failed;
 602                 *delim = 0;
 603                 np->remote_port = simple_strtol(cur, NULL, 10);
 604                 cur = delim;
 605         }
 606         cur++;
 607
 608         /* dst ip */
 609         if ((delim = strchr(cur, '/')) == NULL)
 610                 goto parse_failed;
 611         *delim = 0;
 612         np->remote_ip = ntohl(in_aton(cur));
 613         cur = delim + 1;
 614
 615         if (*cur != 0) {
 616                 /* MAC address */
 617                 if ((delim = strchr(cur, ':')) == NULL)
 618                         goto parse_failed;
 619                 *delim = 0;
 620                 np->remote_mac[0] = simple_strtol(cur, NULL, 16);
 621                 cur = delim + 1;
 622                 if ((delim = strchr(cur, ':')) == NULL)
 623                         goto parse_failed;
 624                 *delim = 0;
 625                 np->remote_mac[1] = simple_strtol(cur, NULL, 16);
 626                 cur = delim + 1;
 627                 if ((delim = strchr(cur, ':')) == NULL)
 628                         goto parse_failed;
 629                 *delim = 0;
 630                 np->remote_mac[2] = simple_strtol(cur, NULL, 16);
 631                 cur = delim + 1;
 632                 if ((delim = strchr(cur, ':')) == NULL)
 633                         goto parse_failed;
 634                 *delim = 0;
 635                 np->remote_mac[3] = simple_strtol(cur, NULL, 16);
 636                 cur = delim + 1;
 637                 if ((delim = strchr(cur, ':')) == NULL)
 638                         goto parse_failed;
 639                 *delim = 0;
 640                 np->remote_mac[4] = simple_strtol(cur, NULL, 16);
 641                 cur = delim + 1;
 642                 np->remote_mac[5] = simple_strtol(cur, NULL, 16);
 643         }
 644
 645         netpoll_print_options(np);
 646
 647         return 0;
 648
 649  parse_failed:
 650         printk(KERN_INFO "%s: couldn't parse config at %s!\n",
 651                np->name, cur);
 652         return -1;
 653 }
 654
 655 int netpoll_setup(struct netpoll *np)
 656 {
 657         struct net_device *ndev = NULL;
 658         struct in_device *in_dev;
 659         struct netpoll_info *npinfo;
 660         unsigned long flags;
 661         int err;
 662
 663         if (np->dev_name)
 664                 ndev = dev_get_by_name(&init_net, np->dev_name);
 665         if (!ndev) {
 666                 printk(KERN_ERR "%s: %s doesn't exist, aborting.\n",
 667                        np->name, np->dev_name);
 668                 return -ENODEV;
 669         }
 670
 671         np->dev = ndev;
 672         if (!ndev->npinfo) {
 673                 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
 674                 if (!npinfo) {
 675                         err = -ENOMEM;
 676                         goto release;
 677                 }
 678
 679                 npinfo->rx_flags = 0;
 680                 npinfo->rx_np = NULL;
 681
 682                 spin_lock_init(&npinfo->rx_lock);
 683                 skb_queue_head_init(&npinfo->arp_tx);
 684                 skb_queue_head_init(&npinfo->txq);
 685                 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
 686
 687                 atomic_set(&npinfo->refcnt, 1);
 688         } else {
 689                 npinfo = ndev->npinfo;
 690                 atomic_inc(&npinfo->refcnt);
 691         }
 692
 693         if (!ndev->poll_controller) {
 694                 printk(KERN_ERR "%s: %s doesn't support polling, aborting.\n",
 695                        np->name, np->dev_name);
 696                 err = -ENOTSUPP;
 697                 goto release;
 698         }
 699
 700         if (!netif_running(ndev)) {
 701                 unsigned long atmost, atleast;
 702
 703                 printk(KERN_INFO "%s: device %s not up yet, forcing it\n",
 704                        np->name, np->dev_name);
 705
 706                 rtnl_lock();
 707                 err = dev_open(ndev);
 708                 rtnl_unlock();
 709
 710                 if (err) {
 711                         printk(KERN_ERR "%s: failed to open %s\n",
 712                                np->name, ndev->name);
 713                         goto release;
 714                 }
 715
 716                 atleast = jiffies + HZ/10;
 717                 atmost = jiffies + 4*HZ;
 718                 while (!netif_carrier_ok(ndev)) {
 719                         if (time_after(jiffies, atmost)) {
 720                                 printk(KERN_NOTICE
 721                                        "%s: timeout waiting for carrier\n",
 722                                        np->name);
 723                                 break;
 724                         }
 725                         cond_resched();
 726                 }
 727
 728                 /* If carrier appears to come up instantly, we don't
 729                  * trust it and pause so that we don't pump all our
 730                  * queued console messages into the bitbucket.
 731                  */
 732
 733                 if (time_before(jiffies, atleast)) {
 734                         printk(KERN_NOTICE "%s: carrier detect appears"
 735                                " untrustworthy, waiting 4 seconds\n",
 736                                np->name);
 737                         msleep(4000);
 738                 }
 739         }
 740
 741         if (!np->local_ip) {
 742                 rcu_read_lock();
 743                 in_dev = __in_dev_get_rcu(ndev);
 744
 745                 if (!in_dev || !in_dev->ifa_list) {
 746                         rcu_read_unlock();
 747                         printk(KERN_ERR "%s: no IP address for %s, aborting\n",
 748                                np->name, np->dev_name);
 749                         err = -EDESTADDRREQ;
 750                         goto release;
 751                 }
 752
 753                 np->local_ip = ntohl(in_dev->ifa_list->ifa_local);
 754                 rcu_read_unlock();
 755                 printk(KERN_INFO "%s: local IP %d.%d.%d.%d\n",
 756                        np->name, HIPQUAD(np->local_ip));
 757         }
 758
 759         if (np->rx_hook) {
 760                 spin_lock_irqsave(&npinfo->rx_lock, flags);
 761                 npinfo->rx_flags |= NETPOLL_RX_ENABLED;
 762                 npinfo->rx_np = np;
 763                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 764         }
 765
 766         /* fill up the skb queue */
 767         refill_skbs();
 768
 769         /* last thing to do is link it to the net device structure */
 770         ndev->npinfo = npinfo;
 771
 772         /* avoid racing with NAPI reading npinfo */
 773         synchronize_rcu();
 774
 775         return 0;
 776
 777  release:
 778         if (!ndev->npinfo)
 779                 kfree(npinfo);
 780         np->dev = NULL;
 781         dev_put(ndev);
 782         return err;
 783 }
 784
 785 static int __init netpoll_init(void)
 786 {
 787         skb_queue_head_init(&skb_pool);
 788         return 0;
 789 }
 790 core_initcall(netpoll_init);
 791
 792 void netpoll_cleanup(struct netpoll *np)
 793 {
 794         struct netpoll_info *npinfo;
 795         unsigned long flags;
 796
 797         if (np->dev) {
 798                 npinfo = np->dev->npinfo;
 799                 if (npinfo) {
 800                         if (npinfo->rx_np == np) {
 801                                 spin_lock_irqsave(&npinfo->rx_lock, flags);
 802                                 npinfo->rx_np = NULL;
 803                                 npinfo->rx_flags &= ~NETPOLL_RX_ENABLED;
 804                                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 805                         }
 806
 807                         if (atomic_dec_and_test(&npinfo->refcnt)) {
 808                                 skb_queue_purge(&npinfo->arp_tx);
 809                                 skb_queue_purge(&npinfo->txq);
 810                                 cancel_rearming_delayed_work(&npinfo->tx_work);
 811
 812                                 /* clean after last, unfinished work */
 813                                 __skb_queue_purge(&npinfo->txq);
 814                                 kfree(npinfo);
 815                                 np->dev->npinfo = NULL;
 816                         }
 817                 }
 818
 819                 dev_put(np->dev);
 820         }
 821
 822         np->dev = NULL;
 823 }
 824
 825 int netpoll_trap(void)
 826 {
 827         return atomic_read(&trapped);
 828 }
 829
 830 void netpoll_set_trap(int trap)
 831 {
 832         if (trap)
 833                 atomic_inc(&trapped);
 834         else
 835                 atomic_dec(&trapped);
 836 }
 837
 838 EXPORT_SYMBOL(netpoll_set_trap);
 839 EXPORT_SYMBOL(netpoll_trap);
 840 EXPORT_SYMBOL(netpoll_print_options);
 841 EXPORT_SYMBOL(netpoll_parse_options);
 842 EXPORT_SYMBOL(netpoll_setup);
 843 EXPORT_SYMBOL(netpoll_cleanup);
 844 EXPORT_SYMBOL(netpoll_send_udp);
 845 EXPORT_SYMBOL(netpoll_poll);